fs/btrfs/inode.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/pagemap.h>
  25 #include <linux/highmem.h>
  26 #include <linux/time.h>
  27 #include <linux/init.h>
  28 #include <linux/string.h>
  29 #include <linux/backing-dev.h>
  30 #include <linux/mpage.h>
  31 #include <linux/swap.h>
  32 #include <linux/writeback.h>
  33 #include <linux/statfs.h>
  34 #include <linux/compat.h>
  35 #include <linux/aio.h>
  36 #include <linux/bit_spinlock.h>
  37 #include <linux/xattr.h>
  38 #include <linux/posix_acl.h>
  39 #include <linux/falloc.h>
  40 #include <linux/slab.h>
  41 #include <linux/ratelimit.h>
  42 #include <linux/mount.h>
  43 #include <linux/btrfs.h>
  44 #include <linux/blkdev.h>
  45 #include <linux/posix_acl_xattr.h>
  46 #include "ctree.h"
  47 #include "disk-io.h"
  48 #include "transaction.h"
  49 #include "btrfs_inode.h"
  50 #include "print-tree.h"
  51 #include "ordered-data.h"
  52 #include "xattr.h"
  53 #include "tree-log.h"
  54 #include "volumes.h"
  55 #include "compression.h"
  56 #include "locking.h"
  57 #include "free-space-cache.h"
  58 #include "inode-map.h"
  59 #include "backref.h"
  60 #include "hash.h"
  61 #include "props.h"
  62
  63 struct btrfs_iget_args {
  64         struct btrfs_key *location;
  65         struct btrfs_root *root;
  66 };
  67
  68 static const struct inode_operations btrfs_dir_inode_operations;
  69 static const struct inode_operations btrfs_symlink_inode_operations;
  70 static const struct inode_operations btrfs_dir_ro_inode_operations;
  71 static const struct inode_operations btrfs_special_inode_operations;
  72 static const struct inode_operations btrfs_file_inode_operations;
  73 static const struct address_space_operations btrfs_aops;
  74 static const struct address_space_operations btrfs_symlink_aops;
  75 static const struct file_operations btrfs_dir_file_operations;
  76 static struct extent_io_ops btrfs_extent_io_ops;
  77
  78 static struct kmem_cache *btrfs_inode_cachep;
  79 static struct kmem_cache *btrfs_delalloc_work_cachep;
  80 struct kmem_cache *btrfs_trans_handle_cachep;
  81 struct kmem_cache *btrfs_transaction_cachep;
  82 struct kmem_cache *btrfs_path_cachep;
  83 struct kmem_cache *btrfs_free_space_cachep;
  84
  85 #define S_SHIFT 12
  86 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  87         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  88         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  89         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  90         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  91         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  92         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  93         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  94 };
  95
  96 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
  97 static int btrfs_truncate(struct inode *inode);
  98 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
  99 static noinline int cow_file_range(struct inode *inode,
 100                                    struct page *locked_page,
 101                                    u64 start, u64 end, int *page_started,
 102                                    unsigned long *nr_written, int unlock);
 103 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
 104                                            u64 len, u64 orig_start,
 105                                            u64 block_start, u64 block_len,
 106                                            u64 orig_block_len, u64 ram_bytes,
 107                                            int type);
 108
 109 static int btrfs_dirty_inode(struct inode *inode);
 110
 111 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
 112                                      struct inode *inode,  struct inode *dir,
 113                                      const struct qstr *qstr)
 114 {
 115         int err;
 116
 117         err = btrfs_init_acl(trans, inode, dir);
 118         if (!err)
 119                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 120         return err;
 121 }
 122
 123 /*
 124  * this does all the hard work for inserting an inline extent into
 125  * the btree.  The caller should have done a btrfs_drop_extents so that
 126  * no overlapping inline items exist in the btree
 127  */
 128 static int insert_inline_extent(struct btrfs_trans_handle *trans,
 129                                 struct btrfs_path *path, int extent_inserted,
 130                                 struct btrfs_root *root, struct inode *inode,
 131                                 u64 start, size_t size, size_t compressed_size,
 132                                 int compress_type,
 133                                 struct page **compressed_pages)
 134 {
 135         struct extent_buffer *leaf;
 136         struct page *page = NULL;
 137         char *kaddr;
 138         unsigned long ptr;
 139         struct btrfs_file_extent_item *ei;
 140         int err = 0;
 141         int ret;
 142         size_t cur_size = size;
 143         unsigned long offset;
 144
 145         if (compressed_size && compressed_pages)
 146                 cur_size = compressed_size;
 147
 148         inode_add_bytes(inode, size);
 149
 150         if (!extent_inserted) {
 151                 struct btrfs_key key;
 152                 size_t datasize;
 153
 154                 key.objectid = btrfs_ino(inode);
 155                 key.offset = start;
 156                 key.type = BTRFS_EXTENT_DATA_KEY;
 157
 158                 datasize = btrfs_file_extent_calc_inline_size(cur_size);
 159                 path->leave_spinning = 1;
 160                 ret = btrfs_insert_empty_item(trans, root, path, &key,
 161                                               datasize);
 162                 if (ret) {
 163                         err = ret;
 164                         goto fail;
 165                 }
 166         }
 167         leaf = path->nodes[0];
 168         ei = btrfs_item_ptr(leaf, path->slots[0],
 169                             struct btrfs_file_extent_item);
 170         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 171         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 172         btrfs_set_file_extent_encryption(leaf, ei, 0);
 173         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 174         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 175         ptr = btrfs_file_extent_inline_start(ei);
 176
 177         if (compress_type != BTRFS_COMPRESS_NONE) {
 178                 struct page *cpage;
 179                 int i = 0;
 180                 while (compressed_size > 0) {
 181                         cpage = compressed_pages[i];
 182                         cur_size = min_t(unsigned long, compressed_size,
 183                                        PAGE_CACHE_SIZE);
 184
 185                         kaddr = kmap_atomic(cpage);
 186                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 187                         kunmap_atomic(kaddr);
 188
 189                         i++;
 190                         ptr += cur_size;
 191                         compressed_size -= cur_size;
 192                 }
 193                 btrfs_set_file_extent_compression(leaf, ei,
 194                                                   compress_type);
 195         } else {
 196                 page = find_get_page(inode->i_mapping,
 197                                      start >> PAGE_CACHE_SHIFT);
 198                 btrfs_set_file_extent_compression(leaf, ei, 0);
 199                 kaddr = kmap_atomic(page);
 200                 offset = start & (PAGE_CACHE_SIZE - 1);
 201                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 202                 kunmap_atomic(kaddr);
 203                 page_cache_release(page);
 204         }
 205         btrfs_mark_buffer_dirty(leaf);
 206         btrfs_release_path(path);
 207
 208         /*
 209          * we're an inline extent, so nobody can
 210          * extend the file past i_size without locking
 211          * a page we already have locked.
 212          *
 213          * We must do any isize and inode updates
 214          * before we unlock the pages.  Otherwise we
 215          * could end up racing with unlink.
 216          */
 217         BTRFS_I(inode)->disk_i_size = inode->i_size;
 218         ret = btrfs_update_inode(trans, root, inode);
 219
 220         return ret;
 221 fail:
 222         return err;
 223 }
 224
 225
 226 /*
 227  * conditionally insert an inline extent into the file.  This
 228  * does the checks required to make sure the data is small enough
 229  * to fit as an inline extent.
 230  */
 231 static noinline int cow_file_range_inline(struct btrfs_root *root,
 232                                           struct inode *inode, u64 start,
 233                                           u64 end, size_t compressed_size,
 234                                           int compress_type,
 235                                           struct page **compressed_pages)
 236 {
 237         struct btrfs_trans_handle *trans;
 238         u64 isize = i_size_read(inode);
 239         u64 actual_end = min(end + 1, isize);
 240         u64 inline_len = actual_end - start;
 241         u64 aligned_end = ALIGN(end, root->sectorsize);
 242         u64 data_len = inline_len;
 243         int ret;
 244         struct btrfs_path *path;
 245         int extent_inserted = 0;
 246         u32 extent_item_size;
 247
 248         if (compressed_size)
 249                 data_len = compressed_size;
 250
 251         if (start > 0 ||
 252             actual_end > PAGE_CACHE_SIZE ||
 253             data_len > BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 254             (!compressed_size &&
 255             (actual_end & (root->sectorsize - 1)) == 0) ||
 256             end + 1 < isize ||
 257             data_len > root->fs_info->max_inline) {
 258                 return 1;
 259         }
 260
 261         path = btrfs_alloc_path();
 262         if (!path)
 263                 return -ENOMEM;
 264
 265         trans = btrfs_join_transaction(root);
 266         if (IS_ERR(trans)) {
 267                 btrfs_free_path(path);
 268                 return PTR_ERR(trans);
 269         }
 270         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 271
 272         if (compressed_size && compressed_pages)
 273                 extent_item_size = btrfs_file_extent_calc_inline_size(
 274                    compressed_size);
 275         else
 276                 extent_item_size = btrfs_file_extent_calc_inline_size(
 277                     inline_len);
 278
 279         ret = __btrfs_drop_extents(trans, root, inode, path,
 280                                    start, aligned_end, NULL,
 281                                    1, 1, extent_item_size, &extent_inserted);
 282         if (ret) {
 283                 btrfs_abort_transaction(trans, root, ret);
 284                 goto out;
 285         }
 286
 287         if (isize > actual_end)
 288                 inline_len = min_t(u64, isize, actual_end);
 289         ret = insert_inline_extent(trans, path, extent_inserted,
 290                                    root, inode, start,
 291                                    inline_len, compressed_size,
 292                                    compress_type, compressed_pages);
 293         if (ret && ret != -ENOSPC) {
 294                 btrfs_abort_transaction(trans, root, ret);
 295                 goto out;
 296         } else if (ret == -ENOSPC) {
 297                 ret = 1;
 298                 goto out;
 299         }
 300
 301         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
 302         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 303         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 304 out:
 305         btrfs_free_path(path);
 306         btrfs_end_transaction(trans, root);
 307         return ret;
 308 }
 309
 310 struct async_extent {
 311         u64 start;
 312         u64 ram_size;
 313         u64 compressed_size;
 314         struct page **pages;
 315         unsigned long nr_pages;
 316         int compress_type;
 317         struct list_head list;
 318 };
 319
 320 struct async_cow {
 321         struct inode *inode;
 322         struct btrfs_root *root;
 323         struct page *locked_page;
 324         u64 start;
 325         u64 end;
 326         struct list_head extents;
 327         struct btrfs_work work;
 328 };
 329
 330 static noinline int add_async_extent(struct async_cow *cow,
 331                                      u64 start, u64 ram_size,
 332                                      u64 compressed_size,
 333                                      struct page **pages,
 334                                      unsigned long nr_pages,
 335                                      int compress_type)
 336 {
 337         struct async_extent *async_extent;
 338
 339         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 340         BUG_ON(!async_extent); /* -ENOMEM */
 341         async_extent->start = start;
 342         async_extent->ram_size = ram_size;
 343         async_extent->compressed_size = compressed_size;
 344         async_extent->pages = pages;
 345         async_extent->nr_pages = nr_pages;
 346         async_extent->compress_type = compress_type;
 347         list_add_tail(&async_extent->list, &cow->extents);
 348         return 0;
 349 }
 350
 351 static inline int inode_need_compress(struct inode *inode)
 352 {
 353         struct btrfs_root *root = BTRFS_I(inode)->root;
 354
 355         /* force compress */
 356         if (btrfs_test_opt(root, FORCE_COMPRESS))
 357                 return 1;
 358         /* bad compression ratios */
 359         if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
 360                 return 0;
 361         if (btrfs_test_opt(root, COMPRESS) ||
 362             BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
 363             BTRFS_I(inode)->force_compress)
 364                 return 1;
 365         return 0;
 366 }
 367
 368 /*
 369  * we create compressed extents in two phases.  The first
 370  * phase compresses a range of pages that have already been
 371  * locked (both pages and state bits are locked).
 372  *
 373  * This is done inside an ordered work queue, and the compression
 374  * is spread across many cpus.  The actual IO submission is step
 375  * two, and the ordered work queue takes care of making sure that
 376  * happens in the same order things were put onto the queue by
 377  * writepages and friends.
 378  *
 379  * If this code finds it can't get good compression, it puts an
 380  * entry onto the work queue to write the uncompressed bytes.  This
 381  * makes sure that both compressed inodes and uncompressed inodes
 382  * are written in the same order that the flusher thread sent them
 383  * down.
 384  */
 385 static noinline void compress_file_range(struct inode *inode,
 386                                         struct page *locked_page,
 387                                         u64 start, u64 end,
 388                                         struct async_cow *async_cow,
 389                                         int *num_added)
 390 {
 391         struct btrfs_root *root = BTRFS_I(inode)->root;
 392         u64 num_bytes;
 393         u64 blocksize = root->sectorsize;
 394         u64 actual_end;
 395         u64 isize = i_size_read(inode);
 396         int ret = 0;
 397         struct page **pages = NULL;
 398         unsigned long nr_pages;
 399         unsigned long nr_pages_ret = 0;
 400         unsigned long total_compressed = 0;
 401         unsigned long total_in = 0;
 402         unsigned long max_compressed = 128 * 1024;
 403         unsigned long max_uncompressed = 128 * 1024;
 404         int i;
 405         int will_compress;
 406         int compress_type = root->fs_info->compress_type;
 407         int redirty = 0;
 408
 409         /* if this is a small write inside eof, kick off a defrag */
 410         if ((end - start + 1) < 16 * 1024 &&
 411             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 412                 btrfs_add_inode_defrag(NULL, inode);
 413
 414         actual_end = min_t(u64, isize, end + 1);
 415 again:
 416         will_compress = 0;
 417         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 418         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 419
 420         /*
 421          * we don't want to send crud past the end of i_size through
 422          * compression, that's just a waste of CPU time.  So, if the
 423          * end of the file is before the start of our current
 424          * requested range of bytes, we bail out to the uncompressed
 425          * cleanup code that can deal with all of this.
 426          *
 427          * It isn't really the fastest way to fix things, but this is a
 428          * very uncommon corner.
 429          */
 430         if (actual_end <= start)
 431                 goto cleanup_and_bail_uncompressed;
 432
 433         total_compressed = actual_end - start;
 434
 435         /*
 436          * skip compression for a small file range(<=blocksize) that
 437          * isn't an inline extent, since it dosen't save disk space at all.
 438          */
 439         if (total_compressed <= blocksize &&
 440            (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 441                 goto cleanup_and_bail_uncompressed;
 442
 443         /* we want to make sure that amount of ram required to uncompress
 444          * an extent is reasonable, so we limit the total size in ram
 445          * of a compressed extent to 128k.  This is a crucial number
 446          * because it also controls how easily we can spread reads across
 447          * cpus for decompression.
 448          *
 449          * We also want to make sure the amount of IO required to do
 450          * a random read is reasonably small, so we limit the size of
 451          * a compressed extent to 128k.
 452          */
 453         total_compressed = min(total_compressed, max_uncompressed);
 454         num_bytes = ALIGN(end - start + 1, blocksize);
 455         num_bytes = max(blocksize,  num_bytes);
 456         total_in = 0;
 457         ret = 0;
 458
 459         /*
 460          * we do compression for mount -o compress and when the
 461          * inode has not been flagged as nocompress.  This flag can
 462          * change at any time if we discover bad compression ratios.
 463          */
 464         if (inode_need_compress(inode)) {
 465                 WARN_ON(pages);
 466                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 467                 if (!pages) {
 468                         /* just bail out to the uncompressed code */
 469                         goto cont;
 470                 }
 471
 472                 if (BTRFS_I(inode)->force_compress)
 473                         compress_type = BTRFS_I(inode)->force_compress;
 474
 475                 /*
 476                  * we need to call clear_page_dirty_for_io on each
 477                  * page in the range.  Otherwise applications with the file
 478                  * mmap'd can wander in and change the page contents while
 479                  * we are compressing them.
 480                  *
 481                  * If the compression fails for any reason, we set the pages
 482                  * dirty again later on.
 483                  */
 484                 extent_range_clear_dirty_for_io(inode, start, end);
 485                 redirty = 1;
 486                 ret = btrfs_compress_pages(compress_type,
 487                                            inode->i_mapping, start,
 488                                            total_compressed, pages,
 489                                            nr_pages, &nr_pages_ret,
 490                                            &total_in,
 491                                            &total_compressed,
 492                                            max_compressed);
 493
 494                 if (!ret) {
 495                         unsigned long offset = total_compressed &
 496                                 (PAGE_CACHE_SIZE - 1);
 497                         struct page *page = pages[nr_pages_ret - 1];
 498                         char *kaddr;
 499
 500                         /* zero the tail end of the last page, we might be
 501                          * sending it down to disk
 502                          */
 503                         if (offset) {
 504                                 kaddr = kmap_atomic(page);
 505                                 memset(kaddr + offset, 0,
 506                                        PAGE_CACHE_SIZE - offset);
 507                                 kunmap_atomic(kaddr);
 508                         }
 509                         will_compress = 1;
 510                 }
 511         }
 512 cont:
 513         if (start == 0) {
 514                 /* lets try to make an inline extent */
 515                 if (ret || total_in < (actual_end - start)) {
 516                         /* we didn't compress the entire range, try
 517                          * to make an uncompressed inline extent.
 518                          */
 519                         ret = cow_file_range_inline(root, inode, start, end,
 520                                                     0, 0, NULL);
 521                 } else {
 522                         /* try making a compressed inline extent */
 523                         ret = cow_file_range_inline(root, inode, start, end,
 524                                                     total_compressed,
 525                                                     compress_type, pages);
 526                 }
 527                 if (ret <= 0) {
 528                         unsigned long clear_flags = EXTENT_DELALLOC |
 529                                 EXTENT_DEFRAG;
 530                         unsigned long page_error_op;
 531
 532                         clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
 533                         page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
 534
 535                         /*
 536                          * inline extent creation worked or returned error,
 537                          * we don't need to create any more async work items.
 538                          * Unlock and free up our temp pages.
 539                          */
 540                         extent_clear_unlock_delalloc(inode, start, end, NULL,
 541                                                      clear_flags, PAGE_UNLOCK |
 542                                                      PAGE_CLEAR_DIRTY |
 543                                                      PAGE_SET_WRITEBACK |
 544                                                      page_error_op |
 545                                                      PAGE_END_WRITEBACK);
 546                         goto free_pages_out;
 547                 }
 548         }
 549
 550         if (will_compress) {
 551                 /*
 552                  * we aren't doing an inline extent round the compressed size
 553                  * up to a block size boundary so the allocator does sane
 554                  * things
 555                  */
 556                 total_compressed = ALIGN(total_compressed, blocksize);
 557
 558                 /*
 559                  * one last check to make sure the compression is really a
 560                  * win, compare the page count read with the blocks on disk
 561                  */
 562                 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
 563                 if (total_compressed >= total_in) {
 564                         will_compress = 0;
 565                 } else {
 566                         num_bytes = total_in;
 567                 }
 568         }
 569         if (!will_compress && pages) {
 570                 /*
 571                  * the compression code ran but failed to make things smaller,
 572                  * free any pages it allocated and our page pointer array
 573                  */
 574                 for (i = 0; i < nr_pages_ret; i++) {
 575                         WARN_ON(pages[i]->mapping);
 576                         page_cache_release(pages[i]);
 577                 }
 578                 kfree(pages);
 579                 pages = NULL;
 580                 total_compressed = 0;
 581                 nr_pages_ret = 0;
 582
 583                 /* flag the file so we don't compress in the future */
 584                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 585                     !(BTRFS_I(inode)->force_compress)) {
 586                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 587                 }
 588         }
 589         if (will_compress) {
 590                 *num_added += 1;
 591
 592                 /* the async work queues will take care of doing actual
 593                  * allocation on disk for these compressed pages,
 594                  * and will submit them to the elevator.
 595                  */
 596                 add_async_extent(async_cow, start, num_bytes,
 597                                  total_compressed, pages, nr_pages_ret,
 598                                  compress_type);
 599
 600                 if (start + num_bytes < end) {
 601                         start += num_bytes;
 602                         pages = NULL;
 603                         cond_resched();
 604                         goto again;
 605                 }
 606         } else {
 607 cleanup_and_bail_uncompressed:
 608                 /*
 609                  * No compression, but we still need to write the pages in
 610                  * the file we've been given so far.  redirty the locked
 611                  * page if it corresponds to our extent and set things up
 612                  * for the async work queue to run cow_file_range to do
 613                  * the normal delalloc dance
 614                  */
 615                 if (page_offset(locked_page) >= start &&
 616                     page_offset(locked_page) <= end) {
 617                         __set_page_dirty_nobuffers(locked_page);
 618                         /* unlocked later on in the async handlers */
 619                 }
 620                 if (redirty)
 621                         extent_range_redirty_for_io(inode, start, end);
 622                 add_async_extent(async_cow, start, end - start + 1,
 623                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 624                 *num_added += 1;
 625         }
 626
 627         return;
 628
 629 free_pages_out:
 630         for (i = 0; i < nr_pages_ret; i++) {
 631                 WARN_ON(pages[i]->mapping);
 632                 page_cache_release(pages[i]);
 633         }
 634         kfree(pages);
 635 }
 636
 637 static void free_async_extent_pages(struct async_extent *async_extent)
 638 {
 639         int i;
 640
 641         if (!async_extent->pages)
 642                 return;
 643
 644         for (i = 0; i < async_extent->nr_pages; i++) {
 645                 WARN_ON(async_extent->pages[i]->mapping);
 646                 page_cache_release(async_extent->pages[i]);
 647         }
 648         kfree(async_extent->pages);
 649         async_extent->nr_pages = 0;
 650         async_extent->pages = NULL;
 651 }
 652
 653 /*
 654  * phase two of compressed writeback.  This is the ordered portion
 655  * of the code, which only gets called in the order the work was
 656  * queued.  We walk all the async extents created by compress_file_range
 657  * and send them down to the disk.
 658  */
 659 static noinline void submit_compressed_extents(struct inode *inode,
 660                                               struct async_cow *async_cow)
 661 {
 662         struct async_extent *async_extent;
 663         u64 alloc_hint = 0;
 664         struct btrfs_key ins;
 665         struct extent_map *em;
 666         struct btrfs_root *root = BTRFS_I(inode)->root;
 667         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 668         struct extent_io_tree *io_tree;
 669         int ret = 0;
 670
 671 again:
 672         while (!list_empty(&async_cow->extents)) {
 673                 async_extent = list_entry(async_cow->extents.next,
 674                                           struct async_extent, list);
 675                 list_del(&async_extent->list);
 676
 677                 io_tree = &BTRFS_I(inode)->io_tree;
 678
 679 retry:
 680                 /* did the compression code fall back to uncompressed IO? */
 681                 if (!async_extent->pages) {
 682                         int page_started = 0;
 683                         unsigned long nr_written = 0;
 684
 685                         lock_extent(io_tree, async_extent->start,
 686                                          async_extent->start +
 687                                          async_extent->ram_size - 1);
 688
 689                         /* allocate blocks */
 690                         ret = cow_file_range(inode, async_cow->locked_page,
 691                                              async_extent->start,
 692                                              async_extent->start +
 693                                              async_extent->ram_size - 1,
 694                                              &page_started, &nr_written, 0);
 695
 696                         /* JDM XXX */
 697
 698                         /*
 699                          * if page_started, cow_file_range inserted an
 700                          * inline extent and took care of all the unlocking
 701                          * and IO for us.  Otherwise, we need to submit
 702                          * all those pages down to the drive.
 703                          */
 704                         if (!page_started && !ret)
 705                                 extent_write_locked_range(io_tree,
 706                                                   inode, async_extent->start,
 707                                                   async_extent->start +
 708                                                   async_extent->ram_size - 1,
 709                                                   btrfs_get_extent,
 710                                                   WB_SYNC_ALL);
 711                         else if (ret)
 712                                 unlock_page(async_cow->locked_page);
 713                         kfree(async_extent);
 714                         cond_resched();
 715                         continue;
 716                 }
 717
 718                 lock_extent(io_tree, async_extent->start,
 719                             async_extent->start + async_extent->ram_size - 1);
 720
 721                 ret = btrfs_reserve_extent(root,
 722                                            async_extent->compressed_size,
 723                                            async_extent->compressed_size,
 724                                            0, alloc_hint, &ins, 1, 1);
 725                 if (ret) {
 726                         free_async_extent_pages(async_extent);
 727
 728                         if (ret == -ENOSPC) {
 729                                 unlock_extent(io_tree, async_extent->start,
 730                                               async_extent->start +
 731                                               async_extent->ram_size - 1);
 732
 733                                 /*
 734                                  * we need to redirty the pages if we decide to
 735                                  * fallback to uncompressed IO, otherwise we
 736                                  * will not submit these pages down to lower
 737                                  * layers.
 738                                  */
 739                                 extent_range_redirty_for_io(inode,
 740                                                 async_extent->start,
 741                                                 async_extent->start +
 742                                                 async_extent->ram_size - 1);
 743
 744                                 goto retry;
 745                         }
 746                         goto out_free;
 747                 }
 748
 749                 /*
 750                  * here we're doing allocation and writeback of the
 751                  * compressed pages
 752                  */
 753                 btrfs_drop_extent_cache(inode, async_extent->start,
 754                                         async_extent->start +
 755                                         async_extent->ram_size - 1, 0);
 756
 757                 em = alloc_extent_map();
 758                 if (!em) {
 759                         ret = -ENOMEM;
 760                         goto out_free_reserve;
 761                 }
 762                 em->start = async_extent->start;
 763                 em->len = async_extent->ram_size;
 764                 em->orig_start = em->start;
 765                 em->mod_start = em->start;
 766                 em->mod_len = em->len;
 767
 768                 em->block_start = ins.objectid;
 769                 em->block_len = ins.offset;
 770                 em->orig_block_len = ins.offset;
 771                 em->ram_bytes = async_extent->ram_size;
 772                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 773                 em->compress_type = async_extent->compress_type;
 774                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 775                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 776                 em->generation = -1;
 777
 778                 while (1) {
 779                         write_lock(&em_tree->lock);
 780                         ret = add_extent_mapping(em_tree, em, 1);
 781                         write_unlock(&em_tree->lock);
 782                         if (ret != -EEXIST) {
 783                                 free_extent_map(em);
 784                                 break;
 785                         }
 786                         btrfs_drop_extent_cache(inode, async_extent->start,
 787                                                 async_extent->start +
 788                                                 async_extent->ram_size - 1, 0);
 789                 }
 790
 791                 if (ret)
 792                         goto out_free_reserve;
 793
 794                 ret = btrfs_add_ordered_extent_compress(inode,
 795                                                 async_extent->start,
 796                                                 ins.objectid,
 797                                                 async_extent->ram_size,
 798                                                 ins.offset,
 799                                                 BTRFS_ORDERED_COMPRESSED,
 800                                                 async_extent->compress_type);
 801                 if (ret) {
 802                         btrfs_drop_extent_cache(inode, async_extent->start,
 803                                                 async_extent->start +
 804                                                 async_extent->ram_size - 1, 0);
 805                         goto out_free_reserve;
 806                 }
 807
 808                 /*
 809                  * clear dirty, set writeback and unlock the pages.
 810                  */
 811                 extent_clear_unlock_delalloc(inode, async_extent->start,
 812                                 async_extent->start +
 813                                 async_extent->ram_size - 1,
 814                                 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
 815                                 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
 816                                 PAGE_SET_WRITEBACK);
 817                 ret = btrfs_submit_compressed_write(inode,
 818                                     async_extent->start,
 819                                     async_extent->ram_size,
 820                                     ins.objectid,
 821                                     ins.offset, async_extent->pages,
 822                                     async_extent->nr_pages);
 823                 if (ret) {
 824                         struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
 825                         struct page *p = async_extent->pages[0];
 826                         const u64 start = async_extent->start;
 827                         const u64 end = start + async_extent->ram_size - 1;
 828
 829                         p->mapping = inode->i_mapping;
 830                         tree->ops->writepage_end_io_hook(p, start, end,
 831                                                          NULL, 0);
 832                         p->mapping = NULL;
 833                         extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
 834                                                      PAGE_END_WRITEBACK |
 835                                                      PAGE_SET_ERROR);
 836                         free_async_extent_pages(async_extent);
 837                 }
 838                 alloc_hint = ins.objectid + ins.offset;
 839                 kfree(async_extent);
 840                 cond_resched();
 841         }
 842         return;
 843 out_free_reserve:
 844         btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
 845 out_free:
 846         extent_clear_unlock_delalloc(inode, async_extent->start,
 847                                      async_extent->start +
 848                                      async_extent->ram_size - 1,
 849                                      NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
 850                                      EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
 851                                      PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
 852                                      PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
 853                                      PAGE_SET_ERROR);
 854         free_async_extent_pages(async_extent);
 855         kfree(async_extent);
 856         goto again;
 857 }
 858
 859 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 860                                       u64 num_bytes)
 861 {
 862         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 863         struct extent_map *em;
 864         u64 alloc_hint = 0;
 865
 866         read_lock(&em_tree->lock);
 867         em = search_extent_mapping(em_tree, start, num_bytes);
 868         if (em) {
 869                 /*
 870                  * if block start isn't an actual block number then find the
 871                  * first block in this inode and use that as a hint.  If that
 872                  * block is also bogus then just don't worry about it.
 873                  */
 874                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 875                         free_extent_map(em);
 876                         em = search_extent_mapping(em_tree, 0, 0);
 877                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 878                                 alloc_hint = em->block_start;
 879                         if (em)
 880                                 free_extent_map(em);
 881                 } else {
 882                         alloc_hint = em->block_start;
 883                         free_extent_map(em);
 884                 }
 885         }
 886         read_unlock(&em_tree->lock);
 887
 888         return alloc_hint;
 889 }
 890
 891 /*
 892  * when extent_io.c finds a delayed allocation range in the file,
 893  * the call backs end up in this code.  The basic idea is to
 894  * allocate extents on disk for the range, and create ordered data structs
 895  * in ram to track those extents.
 896  *
 897  * locked_page is the page that writepage had locked already.  We use
 898  * it to make sure we don't do extra locks or unlocks.
 899  *
 900  * *page_started is set to one if we unlock locked_page and do everything
 901  * required to start IO on it.  It may be clean and already done with
 902  * IO when we return.
 903  */
 904 static noinline int cow_file_range(struct inode *inode,
 905                                    struct page *locked_page,
 906                                    u64 start, u64 end, int *page_started,
 907                                    unsigned long *nr_written,
 908                                    int unlock)
 909 {
 910         struct btrfs_root *root = BTRFS_I(inode)->root;
 911         u64 alloc_hint = 0;
 912         u64 num_bytes;
 913         unsigned long ram_size;
 914         u64 disk_num_bytes;
 915         u64 cur_alloc_size;
 916         u64 blocksize = root->sectorsize;
 917         struct btrfs_key ins;
 918         struct extent_map *em;
 919         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 920         int ret = 0;
 921
 922         if (btrfs_is_free_space_inode(inode)) {
 923                 WARN_ON_ONCE(1);
 924                 ret = -EINVAL;
 925                 goto out_unlock;
 926         }
 927
 928         num_bytes = ALIGN(end - start + 1, blocksize);
 929         num_bytes = max(blocksize,  num_bytes);
 930         disk_num_bytes = num_bytes;
 931
 932         /* if this is a small write inside eof, kick off defrag */
 933         if (num_bytes < 64 * 1024 &&
 934             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 935                 btrfs_add_inode_defrag(NULL, inode);
 936
 937         if (start == 0) {
 938                 /* lets try to make an inline extent */
 939                 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
 940                                             NULL);
 941                 if (ret == 0) {
 942                         extent_clear_unlock_delalloc(inode, start, end, NULL,
 943                                      EXTENT_LOCKED | EXTENT_DELALLOC |
 944                                      EXTENT_DEFRAG, PAGE_UNLOCK |
 945                                      PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
 946                                      PAGE_END_WRITEBACK);
 947
 948                         *nr_written = *nr_written +
 949                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 950                         *page_started = 1;
 951                         goto out;
 952                 } else if (ret < 0) {
 953                         goto out_unlock;
 954                 }
 955         }
 956
 957         BUG_ON(disk_num_bytes >
 958                btrfs_super_total_bytes(root->fs_info->super_copy));
 959
 960         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 961         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 962
 963         while (disk_num_bytes > 0) {
 964                 unsigned long op;
 965
 966                 cur_alloc_size = disk_num_bytes;
 967                 ret = btrfs_reserve_extent(root, cur_alloc_size,
 968                                            root->sectorsize, 0, alloc_hint,
 969                                            &ins, 1, 1);
 970                 if (ret < 0)
 971                         goto out_unlock;
 972
 973                 em = alloc_extent_map();
 974                 if (!em) {
 975                         ret = -ENOMEM;
 976                         goto out_reserve;
 977                 }
 978                 em->start = start;
 979                 em->orig_start = em->start;
 980                 ram_size = ins.offset;
 981                 em->len = ins.offset;
 982                 em->mod_start = em->start;
 983                 em->mod_len = em->len;
 984
 985                 em->block_start = ins.objectid;
 986                 em->block_len = ins.offset;
 987                 em->orig_block_len = ins.offset;
 988                 em->ram_bytes = ram_size;
 989                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 990                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 991                 em->generation = -1;
 992
 993                 while (1) {
 994                         write_lock(&em_tree->lock);
 995                         ret = add_extent_mapping(em_tree, em, 1);
 996                         write_unlock(&em_tree->lock);
 997                         if (ret != -EEXIST) {
 998                                 free_extent_map(em);
 999                                 break;
1000                         }
1001                         btrfs_drop_extent_cache(inode, start,
1002                                                 start + ram_size - 1, 0);
1003                 }
1004                 if (ret)
1005                         goto out_reserve;
1006
1007                 cur_alloc_size = ins.offset;
1008                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1009                                                ram_size, cur_alloc_size, 0);
1010                 if (ret)
1011                         goto out_drop_extent_cache;
1012
1013                 if (root->root_key.objectid ==
1014                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1015                         ret = btrfs_reloc_clone_csums(inode, start,
1016                                                       cur_alloc_size);
1017                         if (ret)
1018                                 goto out_drop_extent_cache;
1019                 }
1020
1021                 if (disk_num_bytes < cur_alloc_size)
1022                         break;
1023
1024                 /* we're not doing compressed IO, don't unlock the first
1025                  * page (which the caller expects to stay locked), don't
1026                  * clear any dirty bits and don't set any writeback bits
1027                  *
1028                  * Do set the Private2 bit so we know this page was properly
1029                  * setup for writepage
1030                  */
1031                 op = unlock ? PAGE_UNLOCK : 0;
1032                 op |= PAGE_SET_PRIVATE2;
1033
1034                 extent_clear_unlock_delalloc(inode, start,
1035                                              start + ram_size - 1, locked_page,
1036                                              EXTENT_LOCKED | EXTENT_DELALLOC,
1037                                              op);
1038                 disk_num_bytes -= cur_alloc_size;
1039                 num_bytes -= cur_alloc_size;
1040                 alloc_hint = ins.objectid + ins.offset;
1041                 start += cur_alloc_size;
1042         }
1043 out:
1044         return ret;
1045
1046 out_drop_extent_cache:
1047         btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
1048 out_reserve:
1049         btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
1050 out_unlock:
1051         extent_clear_unlock_delalloc(inode, start, end, locked_page,
1052                                      EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
1053                                      EXTENT_DELALLOC | EXTENT_DEFRAG,
1054                                      PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1055                                      PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
1056         goto out;
1057 }
1058
1059 /*
1060  * work queue call back to started compression on a file and pages
1061  */
1062 static noinline void async_cow_start(struct btrfs_work *work)
1063 {
1064         struct async_cow *async_cow;
1065         int num_added = 0;
1066         async_cow = container_of(work, struct async_cow, work);
1067
1068         compress_file_range(async_cow->inode, async_cow->locked_page,
1069                             async_cow->start, async_cow->end, async_cow,
1070                             &num_added);
1071         if (num_added == 0) {
1072                 btrfs_add_delayed_iput(async_cow->inode);
1073                 async_cow->inode = NULL;
1074         }
1075 }
1076
1077 /*
1078  * work queue call back to submit previously compressed pages
1079  */
1080 static noinline void async_cow_submit(struct btrfs_work *work)
1081 {
1082         struct async_cow *async_cow;
1083         struct btrfs_root *root;
1084         unsigned long nr_pages;
1085
1086         async_cow = container_of(work, struct async_cow, work);
1087
1088         root = async_cow->root;
1089         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1090                 PAGE_CACHE_SHIFT;
1091
1092         if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1093             5 * 1024 * 1024 &&
1094             waitqueue_active(&root->fs_info->async_submit_wait))
1095                 wake_up(&root->fs_info->async_submit_wait);
1096
1097         if (async_cow->inode)
1098                 submit_compressed_extents(async_cow->inode, async_cow);
1099 }
1100
1101 static noinline void async_cow_free(struct btrfs_work *work)
1102 {
1103         struct async_cow *async_cow;
1104         async_cow = container_of(work, struct async_cow, work);
1105         if (async_cow->inode)
1106                 btrfs_add_delayed_iput(async_cow->inode);
1107         kfree(async_cow);
1108 }
1109
1110 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1111                                 u64 start, u64 end, int *page_started,
1112                                 unsigned long *nr_written)
1113 {
1114         struct async_cow *async_cow;
1115         struct btrfs_root *root = BTRFS_I(inode)->root;
1116         unsigned long nr_pages;
1117         u64 cur_end;
1118         int limit = 10 * 1024 * 1024;
1119
1120         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1121                          1, 0, NULL, GFP_NOFS);
1122         while (start < end) {
1123                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1124                 BUG_ON(!async_cow); /* -ENOMEM */
1125                 async_cow->inode = igrab(inode);
1126                 async_cow->root = root;
1127                 async_cow->locked_page = locked_page;
1128                 async_cow->start = start;
1129
1130                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1131                     !btrfs_test_opt(root, FORCE_COMPRESS))
1132                         cur_end = end;
1133                 else
1134                         cur_end = min(end, start + 512 * 1024 - 1);
1135
1136                 async_cow->end = cur_end;
1137                 INIT_LIST_HEAD(&async_cow->extents);
1138
1139                 btrfs_init_work(&async_cow->work,
1140                                 btrfs_delalloc_helper,
1141                                 async_cow_start, async_cow_submit,
1142                                 async_cow_free);
1143
1144                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1145                         PAGE_CACHE_SHIFT;
1146                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1147
1148                 btrfs_queue_work(root->fs_info->delalloc_workers,
1149                                  &async_cow->work);
1150
1151                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1152                         wait_event(root->fs_info->async_submit_wait,
1153                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1154                             limit));
1155                 }
1156
1157                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1158                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1159                         wait_event(root->fs_info->async_submit_wait,
1160                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1161                            0));
1162                 }
1163
1164                 *nr_written += nr_pages;
1165                 start = cur_end + 1;
1166         }
1167         *page_started = 1;
1168         return 0;
1169 }
1170
1171 static noinline int csum_exist_in_range(struct btrfs_root *root,
1172                                         u64 bytenr, u64 num_bytes)
1173 {
1174         int ret;
1175         struct btrfs_ordered_sum *sums;
1176         LIST_HEAD(list);
1177
1178         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1179                                        bytenr + num_bytes - 1, &list, 0);
1180         if (ret == 0 && list_empty(&list))
1181                 return 0;
1182
1183         while (!list_empty(&list)) {
1184                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1185                 list_del(&sums->list);
1186                 kfree(sums);
1187         }
1188         return 1;
1189 }
1190
1191 /*
1192  * when nowcow writeback call back.  This checks for snapshots or COW copies
1193  * of the extents that exist in the file, and COWs the file as required.
1194  *
1195  * If no cow copies or snapshots exist, we write directly to the existing
1196  * blocks on disk
1197  */
1198 static noinline int run_delalloc_nocow(struct inode *inode,
1199                                        struct page *locked_page,
1200                               u64 start, u64 end, int *page_started, int force,
1201                               unsigned long *nr_written)
1202 {
1203         struct btrfs_root *root = BTRFS_I(inode)->root;
1204         struct btrfs_trans_handle *trans;
1205         struct extent_buffer *leaf;
1206         struct btrfs_path *path;
1207         struct btrfs_file_extent_item *fi;
1208         struct btrfs_key found_key;
1209         u64 cow_start;
1210         u64 cur_offset;
1211         u64 extent_end;
1212         u64 extent_offset;
1213         u64 disk_bytenr;
1214         u64 num_bytes;
1215         u64 disk_num_bytes;
1216         u64 ram_bytes;
1217         int extent_type;
1218         int ret, err;
1219         int type;
1220         int nocow;
1221         int check_prev = 1;
1222         bool nolock;
1223         u64 ino = btrfs_ino(inode);
1224
1225         path = btrfs_alloc_path();
1226         if (!path) {
1227                 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1228                                              EXTENT_LOCKED | EXTENT_DELALLOC |
1229                                              EXTENT_DO_ACCOUNTING |
1230                                              EXTENT_DEFRAG, PAGE_UNLOCK |
1231                                              PAGE_CLEAR_DIRTY |
1232                                              PAGE_SET_WRITEBACK |
1233                                              PAGE_END_WRITEBACK);
1234                 return -ENOMEM;
1235         }
1236
1237         nolock = btrfs_is_free_space_inode(inode);
1238
1239         if (nolock)
1240                 trans = btrfs_join_transaction_nolock(root);
1241         else
1242                 trans = btrfs_join_transaction(root);
1243
1244         if (IS_ERR(trans)) {
1245                 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1246                                              EXTENT_LOCKED | EXTENT_DELALLOC |
1247                                              EXTENT_DO_ACCOUNTING |
1248                                              EXTENT_DEFRAG, PAGE_UNLOCK |
1249                                              PAGE_CLEAR_DIRTY |
1250                                              PAGE_SET_WRITEBACK |
1251                                              PAGE_END_WRITEBACK);
1252                 btrfs_free_path(path);
1253                 return PTR_ERR(trans);
1254         }
1255
1256         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1257
1258         cow_start = (u64)-1;
1259         cur_offset = start;
1260         while (1) {
1261                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1262                                                cur_offset, 0);
1263                 if (ret < 0)
1264                         goto error;
1265                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1266                         leaf = path->nodes[0];
1267                         btrfs_item_key_to_cpu(leaf, &found_key,
1268                                               path->slots[0] - 1);
1269                         if (found_key.objectid == ino &&
1270                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1271                                 path->slots[0]--;
1272                 }
1273                 check_prev = 0;
1274 next_slot:
1275                 leaf = path->nodes[0];
1276                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1277                         ret = btrfs_next_leaf(root, path);
1278                         if (ret < 0)
1279                                 goto error;
1280                         if (ret > 0)
1281                                 break;
1282                         leaf = path->nodes[0];
1283                 }
1284
1285                 nocow = 0;
1286                 disk_bytenr = 0;
1287                 num_bytes = 0;
1288                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1289
1290                 if (found_key.objectid > ino ||
1291                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1292                     found_key.offset > end)
1293                         break;
1294
1295                 if (found_key.offset > cur_offset) {
1296                         extent_end = found_key.offset;
1297                         extent_type = 0;
1298                         goto out_check;
1299                 }
1300
1301                 fi = btrfs_item_ptr(leaf, path->slots[0],
1302                                     struct btrfs_file_extent_item);
1303                 extent_type = btrfs_file_extent_type(leaf, fi);
1304
1305                 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1306                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1307                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1308                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1309                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1310                         extent_end = found_key.offset +
1311                                 btrfs_file_extent_num_bytes(leaf, fi);
1312                         disk_num_bytes =
1313                                 btrfs_file_extent_disk_num_bytes(leaf, fi);
1314                         if (extent_end <= start) {
1315                                 path->slots[0]++;
1316                                 goto next_slot;
1317                         }
1318                         if (disk_bytenr == 0)
1319                                 goto out_check;
1320                         if (btrfs_file_extent_compression(leaf, fi) ||
1321                             btrfs_file_extent_encryption(leaf, fi) ||
1322                             btrfs_file_extent_other_encoding(leaf, fi))
1323                                 goto out_check;
1324                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1325                                 goto out_check;
1326                         if (btrfs_extent_readonly(root, disk_bytenr))
1327                                 goto out_check;
1328                         if (btrfs_cross_ref_exist(trans, root, ino,
1329                                                   found_key.offset -
1330                                                   extent_offset, disk_bytenr))
1331                                 goto out_check;
1332                         disk_bytenr += extent_offset;
1333                         disk_bytenr += cur_offset - found_key.offset;
1334                         num_bytes = min(end + 1, extent_end) - cur_offset;
1335                         /*
1336                          * if there are pending snapshots for this root,
1337                          * we fall into common COW way.
1338                          */
1339                         if (!nolock) {
1340                                 err = btrfs_start_write_no_snapshoting(root);
1341                                 if (!err)
1342                                         goto out_check;
1343                         }
1344                         /*
1345                          * force cow if csum exists in the range.
1346                          * this ensure that csum for a given extent are
1347                          * either valid or do not exist.
1348                          */
1349                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1350                                 goto out_check;
1351                         nocow = 1;
1352                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1353                         extent_end = found_key.offset +
1354                                 btrfs_file_extent_inline_len(leaf,
1355                                                      path->slots[0], fi);
1356                         extent_end = ALIGN(extent_end, root->sectorsize);
1357                 } else {
1358                         BUG_ON(1);
1359                 }
1360 out_check:
1361                 if (extent_end <= start) {
1362                         path->slots[0]++;
1363                         if (!nolock && nocow)
1364                                 btrfs_end_write_no_snapshoting(root);
1365                         goto next_slot;
1366                 }
1367                 if (!nocow) {
1368                         if (cow_start == (u64)-1)
1369                                 cow_start = cur_offset;
1370                         cur_offset = extent_end;
1371                         if (cur_offset > end)
1372                                 break;
1373                         path->slots[0]++;
1374                         goto next_slot;
1375                 }
1376
1377                 btrfs_release_path(path);
1378                 if (cow_start != (u64)-1) {
1379                         ret = cow_file_range(inode, locked_page,
1380                                              cow_start, found_key.offset - 1,
1381                                              page_started, nr_written, 1);
1382                         if (ret) {
1383                                 if (!nolock && nocow)
1384                                         btrfs_end_write_no_snapshoting(root);
1385                                 goto error;
1386                         }
1387                         cow_start = (u64)-1;
1388                 }
1389
1390                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1391                         struct extent_map *em;
1392                         struct extent_map_tree *em_tree;
1393                         em_tree = &BTRFS_I(inode)->extent_tree;
1394                         em = alloc_extent_map();
1395                         BUG_ON(!em); /* -ENOMEM */
1396                         em->start = cur_offset;
1397                         em->orig_start = found_key.offset - extent_offset;
1398                         em->len = num_bytes;
1399                         em->block_len = num_bytes;
1400                         em->block_start = disk_bytenr;
1401                         em->orig_block_len = disk_num_bytes;
1402                         em->ram_bytes = ram_bytes;
1403                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1404                         em->mod_start = em->start;
1405                         em->mod_len = em->len;
1406                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1407                         set_bit(EXTENT_FLAG_FILLING, &em->flags);
1408                         em->generation = -1;
1409                         while (1) {
1410                                 write_lock(&em_tree->lock);
1411                                 ret = add_extent_mapping(em_tree, em, 1);
1412                                 write_unlock(&em_tree->lock);
1413                                 if (ret != -EEXIST) {
1414                                         free_extent_map(em);
1415                                         break;
1416                                 }
1417                                 btrfs_drop_extent_cache(inode, em->start,
1418                                                 em->start + em->len - 1, 0);
1419                         }
1420                         type = BTRFS_ORDERED_PREALLOC;
1421                 } else {
1422                         type = BTRFS_ORDERED_NOCOW;
1423                 }
1424
1425                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1426                                                num_bytes, num_bytes, type);
1427                 BUG_ON(ret); /* -ENOMEM */
1428
1429                 if (root->root_key.objectid ==
1430                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1431                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1432                                                       num_bytes);
1433                         if (ret) {
1434                                 if (!nolock && nocow)
1435                                         btrfs_end_write_no_snapshoting(root);
1436                                 goto error;
1437                         }
1438                 }
1439
1440                 extent_clear_unlock_delalloc(inode, cur_offset,
1441                                              cur_offset + num_bytes - 1,
1442                                              locked_page, EXTENT_LOCKED |
1443                                              EXTENT_DELALLOC, PAGE_UNLOCK |
1444                                              PAGE_SET_PRIVATE2);
1445                 if (!nolock && nocow)
1446                         btrfs_end_write_no_snapshoting(root);
1447                 cur_offset = extent_end;
1448                 if (cur_offset > end)
1449                         break;
1450         }
1451         btrfs_release_path(path);
1452
1453         if (cur_offset <= end && cow_start == (u64)-1) {
1454                 cow_start = cur_offset;
1455                 cur_offset = end;
1456         }
1457
1458         if (cow_start != (u64)-1) {
1459                 ret = cow_file_range(inode, locked_page, cow_start, end,
1460                                      page_started, nr_written, 1);
1461                 if (ret)
1462                         goto error;
1463         }
1464
1465 error:
1466         err = btrfs_end_transaction(trans, root);
1467         if (!ret)
1468                 ret = err;
1469
1470         if (ret && cur_offset < end)
1471                 extent_clear_unlock_delalloc(inode, cur_offset, end,
1472                                              locked_page, EXTENT_LOCKED |
1473                                              EXTENT_DELALLOC | EXTENT_DEFRAG |
1474                                              EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1475                                              PAGE_CLEAR_DIRTY |
1476                                              PAGE_SET_WRITEBACK |
1477                                              PAGE_END_WRITEBACK);
1478         btrfs_free_path(path);
1479         return ret;
1480 }
1481
1482 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1483 {
1484
1485         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1486             !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1487                 return 0;
1488
1489         /*
1490          * @defrag_bytes is a hint value, no spinlock held here,
1491          * if is not zero, it means the file is defragging.
1492          * Force cow if given extent needs to be defragged.
1493          */
1494         if (BTRFS_I(inode)->defrag_bytes &&
1495             test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1496                            EXTENT_DEFRAG, 0, NULL))
1497                 return 1;
1498
1499         return 0;
1500 }
1501
1502 /*
1503  * extent_io.c call back to do delayed allocation processing
1504  */
1505 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1506                               u64 start, u64 end, int *page_started,
1507                               unsigned long *nr_written)
1508 {
1509         int ret;
1510         int force_cow = need_force_cow(inode, start, end);
1511
1512         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1513                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1514                                          page_started, 1, nr_written);
1515         } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1516                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1517                                          page_started, 0, nr_written);
1518         } else if (!inode_need_compress(inode)) {
1519                 ret = cow_file_range(inode, locked_page, start, end,
1520                                       page_started, nr_written, 1);
1521         } else {
1522                 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1523                         &BTRFS_I(inode)->runtime_flags);
1524                 ret = cow_file_range_async(inode, locked_page, start, end,
1525                                            page_started, nr_written);
1526         }
1527         return ret;
1528 }
1529
1530 static void btrfs_split_extent_hook(struct inode *inode,
1531                                     struct extent_state *orig, u64 split)
1532 {
1533         /* not delalloc, ignore it */
1534         if (!(orig->state & EXTENT_DELALLOC))
1535                 return;
1536
1537         spin_lock(&BTRFS_I(inode)->lock);
1538         BTRFS_I(inode)->outstanding_extents++;
1539         spin_unlock(&BTRFS_I(inode)->lock);
1540 }
1541
1542 /*
1543  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1544  * extents so we can keep track of new extents that are just merged onto old
1545  * extents, such as when we are doing sequential writes, so we can properly
1546  * account for the metadata space we'll need.
1547  */
1548 static void btrfs_merge_extent_hook(struct inode *inode,
1549                                     struct extent_state *new,
1550                                     struct extent_state *other)
1551 {
1552         /* not delalloc, ignore it */
1553         if (!(other->state & EXTENT_DELALLOC))
1554                 return;
1555
1556         spin_lock(&BTRFS_I(inode)->lock);
1557         BTRFS_I(inode)->outstanding_extents--;
1558         spin_unlock(&BTRFS_I(inode)->lock);
1559 }
1560
1561 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1562                                       struct inode *inode)
1563 {
1564         spin_lock(&root->delalloc_lock);
1565         if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1566                 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1567                               &root->delalloc_inodes);
1568                 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1569                         &BTRFS_I(inode)->runtime_flags);
1570                 root->nr_delalloc_inodes++;
1571                 if (root->nr_delalloc_inodes == 1) {
1572                         spin_lock(&root->fs_info->delalloc_root_lock);
1573                         BUG_ON(!list_empty(&root->delalloc_root));
1574                         list_add_tail(&root->delalloc_root,
1575                                       &root->fs_info->delalloc_roots);
1576                         spin_unlock(&root->fs_info->delalloc_root_lock);
1577                 }
1578         }
1579         spin_unlock(&root->delalloc_lock);
1580 }
1581
1582 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1583                                      struct inode *inode)
1584 {
1585         spin_lock(&root->delalloc_lock);
1586         if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1587                 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1588                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1589                           &BTRFS_I(inode)->runtime_flags);
1590                 root->nr_delalloc_inodes--;
1591                 if (!root->nr_delalloc_inodes) {
1592                         spin_lock(&root->fs_info->delalloc_root_lock);
1593                         BUG_ON(list_empty(&root->delalloc_root));
1594                         list_del_init(&root->delalloc_root);
1595                         spin_unlock(&root->fs_info->delalloc_root_lock);
1596                 }
1597         }
1598         spin_unlock(&root->delalloc_lock);
1599 }
1600
1601 /*
1602  * extent_io.c set_bit_hook, used to track delayed allocation
1603  * bytes in this file, and to maintain the list of inodes that
1604  * have pending delalloc work to be done.
1605  */
1606 static void btrfs_set_bit_hook(struct inode *inode,
1607                                struct extent_state *state, unsigned long *bits)
1608 {
1609
1610         if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1611                 WARN_ON(1);
1612         /*
1613          * set_bit and clear bit hooks normally require _irqsave/restore
1614          * but in this case, we are only testing for the DELALLOC
1615          * bit, which is only set or cleared with irqs on
1616          */
1617         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1618                 struct btrfs_root *root = BTRFS_I(inode)->root;
1619                 u64 len = state->end + 1 - state->start;
1620                 bool do_list = !btrfs_is_free_space_inode(inode);
1621
1622                 if (*bits & EXTENT_FIRST_DELALLOC) {
1623                         *bits &= ~EXTENT_FIRST_DELALLOC;
1624                 } else {
1625                         spin_lock(&BTRFS_I(inode)->lock);
1626                         BTRFS_I(inode)->outstanding_extents++;
1627                         spin_unlock(&BTRFS_I(inode)->lock);
1628                 }
1629
1630                 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1631                                      root->fs_info->delalloc_batch);
1632                 spin_lock(&BTRFS_I(inode)->lock);
1633                 BTRFS_I(inode)->delalloc_bytes += len;
1634                 if (*bits & EXTENT_DEFRAG)
1635                         BTRFS_I(inode)->defrag_bytes += len;
1636                 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1637                                          &BTRFS_I(inode)->runtime_flags))
1638                         btrfs_add_delalloc_inodes(root, inode);
1639                 spin_unlock(&BTRFS_I(inode)->lock);
1640         }
1641 }
1642
1643 /*
1644  * extent_io.c clear_bit_hook, see set_bit_hook for why
1645  */
1646 static void btrfs_clear_bit_hook(struct inode *inode,
1647                                  struct extent_state *state,
1648                                  unsigned long *bits)
1649 {
1650         u64 len = state->end + 1 - state->start;
1651
1652         spin_lock(&BTRFS_I(inode)->lock);
1653         if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG))
1654                 BTRFS_I(inode)->defrag_bytes -= len;
1655         spin_unlock(&BTRFS_I(inode)->lock);
1656
1657         /*
1658          * set_bit and clear bit hooks normally require _irqsave/restore
1659          * but in this case, we are only testing for the DELALLOC
1660          * bit, which is only set or cleared with irqs on
1661          */
1662         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1663                 struct btrfs_root *root = BTRFS_I(inode)->root;
1664                 bool do_list = !btrfs_is_free_space_inode(inode);
1665
1666                 if (*bits & EXTENT_FIRST_DELALLOC) {
1667                         *bits &= ~EXTENT_FIRST_DELALLOC;
1668                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1669                         spin_lock(&BTRFS_I(inode)->lock);
1670                         BTRFS_I(inode)->outstanding_extents--;
1671                         spin_unlock(&BTRFS_I(inode)->lock);
1672                 }
1673
1674                 /*
1675                  * We don't reserve metadata space for space cache inodes so we
1676                  * don't need to call dellalloc_release_metadata if there is an
1677                  * error.
1678                  */
1679                 if (*bits & EXTENT_DO_ACCOUNTING &&
1680                     root != root->fs_info->tree_root)
1681                         btrfs_delalloc_release_metadata(inode, len);
1682
1683                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1684                     && do_list && !(state->state & EXTENT_NORESERVE))
1685                         btrfs_free_reserved_data_space(inode, len);
1686
1687                 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1688                                      root->fs_info->delalloc_batch);
1689                 spin_lock(&BTRFS_I(inode)->lock);
1690                 BTRFS_I(inode)->delalloc_bytes -= len;
1691                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1692                     test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1693                              &BTRFS_I(inode)->runtime_flags))
1694                         btrfs_del_delalloc_inode(root, inode);
1695                 spin_unlock(&BTRFS_I(inode)->lock);
1696         }
1697 }
1698
1699 /*
1700  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1701  * we don't create bios that span stripes or chunks
1702  */
1703 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1704                          size_t size, struct bio *bio,
1705                          unsigned long bio_flags)
1706 {
1707         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1708         u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1709         u64 length = 0;
1710         u64 map_length;
1711         int ret;
1712
1713         if (bio_flags & EXTENT_BIO_COMPRESSED)
1714                 return 0;
1715
1716         length = bio->bi_iter.bi_size;
1717         map_length = length;
1718         ret = btrfs_map_block(root->fs_info, rw, logical,
1719                               &map_length, NULL, 0);
1720         /* Will always return 0 with map_multi == NULL */
1721         BUG_ON(ret < 0);
1722         if (map_length < length + size)
1723                 return 1;
1724         return 0;
1725 }
1726
1727 /*
1728  * in order to insert checksums into the metadata in large chunks,
1729  * we wait until bio submission time.   All the pages in the bio are
1730  * checksummed and sums are attached onto the ordered extent record.
1731  *
1732  * At IO completion time the cums attached on the ordered extent record
1733  * are inserted into the btree
1734  */
1735 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1736                                     struct bio *bio, int mirror_num,
1737                                     unsigned long bio_flags,
1738                                     u64 bio_offset)
1739 {
1740         struct btrfs_root *root = BTRFS_I(inode)->root;
1741         int ret = 0;
1742
1743         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1744         BUG_ON(ret); /* -ENOMEM */
1745         return 0;
1746 }
1747
1748 /*
1749  * in order to insert checksums into the metadata in large chunks,
1750  * we wait until bio submission time.   All the pages in the bio are
1751  * checksummed and sums are attached onto the ordered extent record.
1752  *
1753  * At IO completion time the cums attached on the ordered extent record
1754  * are inserted into the btree
1755  */
1756 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1757                           int mirror_num, unsigned long bio_flags,
1758                           u64 bio_offset)
1759 {
1760         struct btrfs_root *root = BTRFS_I(inode)->root;
1761         int ret;
1762
1763         ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1764         if (ret)
1765                 bio_endio(bio, ret);
1766         return ret;
1767 }
1768
1769 /*
1770  * extent_io.c submission hook. This does the right thing for csum calculation
1771  * on write, or reading the csums from the tree before a read
1772  */
1773 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1774                           int mirror_num, unsigned long bio_flags,
1775                           u64 bio_offset)
1776 {
1777         struct btrfs_root *root = BTRFS_I(inode)->root;
1778         int ret = 0;
1779         int skip_sum;
1780         int metadata = 0;
1781         int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1782
1783         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1784
1785         if (btrfs_is_free_space_inode(inode))
1786                 metadata = 2;
1787
1788         if (!(rw & REQ_WRITE)) {
1789                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1790                 if (ret)
1791                         goto out;
1792
1793                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1794                         ret = btrfs_submit_compressed_read(inode, bio,
1795                                                            mirror_num,
1796                                                            bio_flags);
1797                         goto out;
1798                 } else if (!skip_sum) {
1799                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1800                         if (ret)
1801                                 goto out;
1802                 }
1803                 goto mapit;
1804         } else if (async && !skip_sum) {
1805                 /* csum items have already been cloned */
1806                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1807                         goto mapit;
1808                 /* we're doing a write, do the async checksumming */
1809                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1810                                    inode, rw, bio, mirror_num,
1811                                    bio_flags, bio_offset,
1812                                    __btrfs_submit_bio_start,
1813                                    __btrfs_submit_bio_done);
1814                 goto out;
1815         } else if (!skip_sum) {
1816                 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1817                 if (ret)
1818                         goto out;
1819         }
1820
1821 mapit:
1822         ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1823
1824 out:
1825         if (ret < 0)
1826                 bio_endio(bio, ret);
1827         return ret;
1828 }
1829
1830 /*
1831  * given a list of ordered sums record them in the inode.  This happens
1832  * at IO completion time based on sums calculated at bio submission time.
1833  */
1834 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1835                              struct inode *inode, u64 file_offset,
1836                              struct list_head *list)
1837 {
1838         struct btrfs_ordered_sum *sum;
1839
1840         list_for_each_entry(sum, list, list) {
1841                 trans->adding_csums = 1;
1842                 btrfs_csum_file_blocks(trans,
1843                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1844                 trans->adding_csums = 0;
1845         }
1846         return 0;
1847 }
1848
1849 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1850                               struct extent_state **cached_state)
1851 {
1852         WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1853         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1854                                    cached_state, GFP_NOFS);
1855 }
1856
1857 /* see btrfs_writepage_start_hook for details on why this is required */
1858 struct btrfs_writepage_fixup {
1859         struct page *page;
1860         struct btrfs_work work;
1861 };
1862
1863 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1864 {
1865         struct btrfs_writepage_fixup *fixup;
1866         struct btrfs_ordered_extent *ordered;
1867         struct extent_state *cached_state = NULL;
1868         struct page *page;
1869         struct inode *inode;
1870         u64 page_start;
1871         u64 page_end;
1872         int ret;
1873
1874         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1875         page = fixup->page;
1876 again:
1877         lock_page(page);
1878         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1879                 ClearPageChecked(page);
1880                 goto out_page;
1881         }
1882
1883         inode = page->mapping->host;
1884         page_start = page_offset(page);
1885         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1886
1887         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1888                          &cached_state);
1889
1890         /* already ordered? We're done */
1891         if (PagePrivate2(page))
1892                 goto out;
1893
1894         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1895         if (ordered) {
1896                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1897                                      page_end, &cached_state, GFP_NOFS);
1898                 unlock_page(page);
1899                 btrfs_start_ordered_extent(inode, ordered, 1);
1900                 btrfs_put_ordered_extent(ordered);
1901                 goto again;
1902         }
1903
1904         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1905         if (ret) {
1906                 mapping_set_error(page->mapping, ret);
1907                 end_extent_writepage(page, ret, page_start, page_end);
1908                 ClearPageChecked(page);
1909                 goto out;
1910          }
1911
1912         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1913         ClearPageChecked(page);
1914         set_page_dirty(page);
1915 out:
1916         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1917                              &cached_state, GFP_NOFS);
1918 out_page:
1919         unlock_page(page);
1920         page_cache_release(page);
1921         kfree(fixup);
1922 }
1923
1924 /*
1925  * There are a few paths in the higher layers of the kernel that directly
1926  * set the page dirty bit without asking the filesystem if it is a
1927  * good idea.  This causes problems because we want to make sure COW
1928  * properly happens and the data=ordered rules are followed.
1929  *
1930  * In our case any range that doesn't have the ORDERED bit set
1931  * hasn't been properly setup for IO.  We kick off an async process
1932  * to fix it up.  The async helper will wait for ordered extents, set
1933  * the delalloc bit and make it safe to write the page.
1934  */
1935 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1936 {
1937         struct inode *inode = page->mapping->host;
1938         struct btrfs_writepage_fixup *fixup;
1939         struct btrfs_root *root = BTRFS_I(inode)->root;
1940
1941         /* this page is properly in the ordered list */
1942         if (TestClearPagePrivate2(page))
1943                 return 0;
1944
1945         if (PageChecked(page))
1946                 return -EAGAIN;
1947
1948         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1949         if (!fixup)
1950                 return -EAGAIN;
1951
1952         SetPageChecked(page);
1953         page_cache_get(page);
1954         btrfs_init_work(&fixup->work, btrfs_fixup_helper,
1955                         btrfs_writepage_fixup_worker, NULL, NULL);
1956         fixup->page = page;
1957         btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work);
1958         return -EBUSY;
1959 }
1960
1961 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1962                                        struct inode *inode, u64 file_pos,
1963                                        u64 disk_bytenr, u64 disk_num_bytes,
1964                                        u64 num_bytes, u64 ram_bytes,
1965                                        u8 compression, u8 encryption,
1966                                        u16 other_encoding, int extent_type)
1967 {
1968         struct btrfs_root *root = BTRFS_I(inode)->root;
1969         struct btrfs_file_extent_item *fi;
1970         struct btrfs_path *path;
1971         struct extent_buffer *leaf;
1972         struct btrfs_key ins;
1973         int extent_inserted = 0;
1974         int ret;
1975
1976         path = btrfs_alloc_path();
1977         if (!path)
1978                 return -ENOMEM;
1979
1980         /*
1981          * we may be replacing one extent in the tree with another.
1982          * The new extent is pinned in the extent map, and we don't want
1983          * to drop it from the cache until it is completely in the btree.
1984          *
1985          * So, tell btrfs_drop_extents to leave this extent in the cache.
1986          * the caller is expected to unpin it and allow it to be merged
1987          * with the others.
1988          */
1989         ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
1990                                    file_pos + num_bytes, NULL, 0,
1991                                    1, sizeof(*fi), &extent_inserted);
1992         if (ret)
1993                 goto out;
1994
1995         if (!extent_inserted) {
1996                 ins.objectid = btrfs_ino(inode);
1997                 ins.offset = file_pos;
1998                 ins.type = BTRFS_EXTENT_DATA_KEY;
1999
2000                 path->leave_spinning = 1;
2001                 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2002                                               sizeof(*fi));
2003                 if (ret)
2004                         goto out;
2005         }
2006         leaf = path->nodes[0];
2007         fi = btrfs_item_ptr(leaf, path->slots[0],
2008                             struct btrfs_file_extent_item);
2009         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2010         btrfs_set_file_extent_type(leaf, fi, extent_type);
2011         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2012         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2013         btrfs_set_file_extent_offset(leaf, fi, 0);
2014         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2015         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2016         btrfs_set_file_extent_compression(leaf, fi, compression);
2017         btrfs_set_file_extent_encryption(leaf, fi, encryption);
2018         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2019
2020         btrfs_mark_buffer_dirty(leaf);
2021         btrfs_release_path(path);
2022
2023         inode_add_bytes(inode, num_bytes);
2024
2025         ins.objectid = disk_bytenr;
2026         ins.offset = disk_num_bytes;
2027         ins.type = BTRFS_EXTENT_ITEM_KEY;
2028         ret = btrfs_alloc_reserved_file_extent(trans, root,
2029                                         root->root_key.objectid,
2030                                         btrfs_ino(inode), file_pos, &ins);
2031 out:
2032         btrfs_free_path(path);
2033
2034         return ret;
2035 }
2036
2037 /* snapshot-aware defrag */
2038 struct sa_defrag_extent_backref {
2039         struct rb_node node;
2040         struct old_sa_defrag_extent *old;
2041         u64 root_id;
2042         u64 inum;
2043         u64 file_pos;
2044         u64 extent_offset;
2045         u64 num_bytes;
2046         u64 generation;
2047 };
2048
2049 struct old_sa_defrag_extent {
2050         struct list_head list;
2051         struct new_sa_defrag_extent *new;
2052
2053         u64 extent_offset;
2054         u64 bytenr;
2055         u64 offset;
2056         u64 len;
2057         int count;
2058 };
2059
2060 struct new_sa_defrag_extent {
2061         struct rb_root root;
2062         struct list_head head;
2063         struct btrfs_path *path;
2064         struct inode *inode;
2065         u64 file_pos;
2066         u64 len;
2067         u64 bytenr;
2068         u64 disk_len;
2069         u8 compress_type;
2070 };
2071
2072 static int backref_comp(struct sa_defrag_extent_backref *b1,
2073                         struct sa_defrag_extent_backref *b2)
2074 {
2075         if (b1->root_id < b2->root_id)
2076                 return -1;
2077         else if (b1->root_id > b2->root_id)
2078                 return 1;
2079
2080         if (b1->inum < b2->inum)
2081                 return -1;
2082         else if (b1->inum > b2->inum)
2083                 return 1;
2084
2085         if (b1->file_pos < b2->file_pos)
2086                 return -1;
2087         else if (b1->file_pos > b2->file_pos)
2088                 return 1;
2089
2090         /*
2091          * [------------------------------] ===> (a range of space)
2092          *     |<--->|   |<---->| =============> (fs/file tree A)
2093          * |<---------------------------->| ===> (fs/file tree B)
2094          *
2095          * A range of space can refer to two file extents in one tree while
2096          * refer to only one file extent in another tree.
2097          *
2098          * So we may process a disk offset more than one time(two extents in A)
2099          * and locate at the same extent(one extent in B), then insert two same
2100          * backrefs(both refer to the extent in B).
2101          */
2102         return 0;
2103 }
2104
2105 static void backref_insert(struct rb_root *root,
2106                            struct sa_defrag_extent_backref *backref)
2107 {
2108         struct rb_node **p = &root->rb_node;
2109         struct rb_node *parent = NULL;
2110         struct sa_defrag_extent_backref *entry;
2111         int ret;
2112
2113         while (*p) {
2114                 parent = *p;
2115                 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2116
2117                 ret = backref_comp(backref, entry);
2118                 if (ret < 0)
2119                         p = &(*p)->rb_left;
2120                 else
2121                         p = &(*p)->rb_right;
2122         }
2123
2124         rb_link_node(&backref->node, parent, p);
2125         rb_insert_color(&backref->node, root);
2126 }
2127
2128 /*
2129  * Note the backref might has changed, and in this case we just return 0.
2130  */
2131 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2132                                        void *ctx)
2133 {
2134         struct btrfs_file_extent_item *extent;
2135         struct btrfs_fs_info *fs_info;
2136         struct old_sa_defrag_extent *old = ctx;
2137         struct new_sa_defrag_extent *new = old->new;
2138         struct btrfs_path *path = new->path;
2139         struct btrfs_key key;
2140         struct btrfs_root *root;
2141         struct sa_defrag_extent_backref *backref;
2142         struct extent_buffer *leaf;
2143         struct inode *inode = new->inode;
2144         int slot;
2145         int ret;
2146         u64 extent_offset;
2147         u64 num_bytes;
2148
2149         if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2150             inum == btrfs_ino(inode))
2151                 return 0;
2152
2153         key.objectid = root_id;
2154         key.type = BTRFS_ROOT_ITEM_KEY;
2155         key.offset = (u64)-1;
2156
2157         fs_info = BTRFS_I(inode)->root->fs_info;
2158         root = btrfs_read_fs_root_no_name(fs_info, &key);
2159         if (IS_ERR(root)) {
2160                 if (PTR_ERR(root) == -ENOENT)
2161                         return 0;
2162                 WARN_ON(1);
2163                 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2164                          inum, offset, root_id);
2165                 return PTR_ERR(root);
2166         }
2167
2168         key.objectid = inum;
2169         key.type = BTRFS_EXTENT_DATA_KEY;
2170         if (offset > (u64)-1 << 32)
2171                 key.offset = 0;
2172         else
2173                 key.offset = offset;
2174
2175         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2176         if (WARN_ON(ret < 0))
2177                 return ret;
2178         ret = 0;
2179
2180         while (1) {
2181                 cond_resched();
2182
2183                 leaf = path->nodes[0];
2184                 slot = path->slots[0];
2185
2186                 if (slot >= btrfs_header_nritems(leaf)) {
2187                         ret = btrfs_next_leaf(root, path);
2188                         if (ret < 0) {
2189                                 goto out;
2190                         } else if (ret > 0) {
2191                                 ret = 0;
2192                                 goto out;
2193                         }
2194                         continue;
2195                 }
2196
2197                 path->slots[0]++;
2198
2199                 btrfs_item_key_to_cpu(leaf, &key, slot);
2200
2201                 if (key.objectid > inum)
2202                         goto out;
2203
2204                 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2205                         continue;
2206
2207                 extent = btrfs_item_ptr(leaf, slot,
2208                                         struct btrfs_file_extent_item);
2209
2210                 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2211                         continue;
2212
2213                 /*
2214                  * 'offset' refers to the exact key.offset,
2215                  * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2216                  * (key.offset - extent_offset).
2217                  */
2218                 if (key.offset != offset)
2219                         continue;
2220
2221                 extent_offset = btrfs_file_extent_offset(leaf, extent);
2222                 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2223
2224                 if (extent_offset >= old->extent_offset + old->offset +
2225                     old->len || extent_offset + num_bytes <=
2226                     old->extent_offset + old->offset)
2227                         continue;
2228                 break;
2229         }
2230
2231         backref = kmalloc(sizeof(*backref), GFP_NOFS);
2232         if (!backref) {
2233                 ret = -ENOENT;
2234                 goto out;
2235         }
2236
2237         backref->root_id = root_id;
2238         backref->inum = inum;
2239         backref->file_pos = offset;
2240         backref->num_bytes = num_bytes;
2241         backref->extent_offset = extent_offset;
2242         backref->generation = btrfs_file_extent_generation(leaf, extent);
2243         backref->old = old;
2244         backref_insert(&new->root, backref);
2245         old->count++;
2246 out:
2247         btrfs_release_path(path);
2248         WARN_ON(ret);
2249         return ret;
2250 }
2251
2252 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2253                                    struct new_sa_defrag_extent *new)
2254 {
2255         struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2256         struct old_sa_defrag_extent *old, *tmp;
2257         int ret;
2258
2259         new->path = path;
2260
2261         list_for_each_entry_safe(old, tmp, &new->head, list) {
2262                 ret = iterate_inodes_from_logical(old->bytenr +
2263                                                   old->extent_offset, fs_info,
2264                                                   path, record_one_backref,
2265                                                   old);
2266                 if (ret < 0 && ret != -ENOENT)
2267                         return false;
2268
2269                 /* no backref to be processed for this extent */
2270                 if (!old->count) {
2271                         list_del(&old->list);
2272                         kfree(old);
2273                 }
2274         }
2275
2276         if (list_empty(&new->head))
2277                 return false;
2278
2279         return true;
2280 }
2281
2282 static int relink_is_mergable(struct extent_buffer *leaf,
2283                               struct btrfs_file_extent_item *fi,
2284                               struct new_sa_defrag_extent *new)
2285 {
2286         if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2287                 return 0;
2288
2289         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2290                 return 0;
2291
2292         if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2293                 return 0;
2294
2295         if (btrfs_file_extent_encryption(leaf, fi) ||
2296             btrfs_file_extent_other_encoding(leaf, fi))
2297                 return 0;
2298
2299         return 1;
2300 }
2301
2302 /*
2303  * Note the backref might has changed, and in this case we just return 0.
2304  */
2305 static noinline int relink_extent_backref(struct btrfs_path *path,
2306                                  struct sa_defrag_extent_backref *prev,
2307                                  struct sa_defrag_extent_backref *backref)
2308 {
2309         struct btrfs_file_extent_item *extent;
2310         struct btrfs_file_extent_item *item;
2311         struct btrfs_ordered_extent *ordered;
2312         struct btrfs_trans_handle *trans;
2313         struct btrfs_fs_info *fs_info;
2314         struct btrfs_root *root;
2315         struct btrfs_key key;
2316         struct extent_buffer *leaf;
2317         struct old_sa_defrag_extent *old = backref->old;
2318         struct new_sa_defrag_extent *new = old->new;
2319         struct inode *src_inode = new->inode;
2320         struct inode *inode;
2321         struct extent_state *cached = NULL;
2322         int ret = 0;
2323         u64 start;
2324         u64 len;
2325         u64 lock_start;
2326         u64 lock_end;
2327         bool merge = false;
2328         int index;
2329
2330         if (prev && prev->root_id == backref->root_id &&
2331             prev->inum == backref->inum &&
2332             prev->file_pos + prev->num_bytes == backref->file_pos)
2333                 merge = true;
2334
2335         /* step 1: get root */
2336         key.objectid = backref->root_id;
2337         key.type = BTRFS_ROOT_ITEM_KEY;
2338         key.offset = (u64)-1;
2339
2340         fs_info = BTRFS_I(src_inode)->root->fs_info;
2341         index = srcu_read_lock(&fs_info->subvol_srcu);
2342
2343         root = btrfs_read_fs_root_no_name(fs_info, &key);
2344         if (IS_ERR(root)) {
2345                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2346                 if (PTR_ERR(root) == -ENOENT)
2347                         return 0;
2348                 return PTR_ERR(root);
2349         }
2350
2351         if (btrfs_root_readonly(root)) {
2352                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2353                 return 0;
2354         }
2355
2356         /* step 2: get inode */
2357         key.objectid = backref->inum;
2358         key.type = BTRFS_INODE_ITEM_KEY;
2359         key.offset = 0;
2360
2361         inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2362         if (IS_ERR(inode)) {
2363                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2364                 return 0;
2365         }
2366
2367         srcu_read_unlock(&fs_info->subvol_srcu, index);
2368
2369         /* step 3: relink backref */
2370         lock_start = backref->file_pos;
2371         lock_end = backref->file_pos + backref->num_bytes - 1;
2372         lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2373                          0, &cached);
2374
2375         ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2376         if (ordered) {
2377                 btrfs_put_ordered_extent(ordered);
2378                 goto out_unlock;
2379         }
2380
2381         trans = btrfs_join_transaction(root);
2382         if (IS_ERR(trans)) {
2383                 ret = PTR_ERR(trans);
2384                 goto out_unlock;
2385         }
2386
2387         key.objectid = backref->inum;
2388         key.type = BTRFS_EXTENT_DATA_KEY;
2389         key.offset = backref->file_pos;
2390
2391         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2392         if (ret < 0) {
2393                 goto out_free_path;
2394         } else if (ret > 0) {
2395                 ret = 0;
2396                 goto out_free_path;
2397         }
2398
2399         extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2400                                 struct btrfs_file_extent_item);
2401
2402         if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2403             backref->generation)
2404                 goto out_free_path;
2405
2406         btrfs_release_path(path);
2407
2408         start = backref->file_pos;
2409         if (backref->extent_offset < old->extent_offset + old->offset)
2410                 start += old->extent_offset + old->offset -
2411                          backref->extent_offset;
2412
2413         len = min(backref->extent_offset + backref->num_bytes,
2414                   old->extent_offset + old->offset + old->len);
2415         len -= max(backref->extent_offset, old->extent_offset + old->offset);
2416
2417         ret = btrfs_drop_extents(trans, root, inode, start,
2418                                  start + len, 1);
2419         if (ret)
2420                 goto out_free_path;
2421 again:
2422         key.objectid = btrfs_ino(inode);
2423         key.type = BTRFS_EXTENT_DATA_KEY;
2424         key.offset = start;
2425
2426         path->leave_spinning = 1;
2427         if (merge) {
2428                 struct btrfs_file_extent_item *fi;
2429                 u64 extent_len;
2430                 struct btrfs_key found_key;
2431
2432                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2433                 if (ret < 0)
2434                         goto out_free_path;
2435
2436                 path->slots[0]--;
2437                 leaf = path->nodes[0];
2438                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2439
2440                 fi = btrfs_item_ptr(leaf, path->slots[0],
2441                                     struct btrfs_file_extent_item);
2442                 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2443
2444                 if (extent_len + found_key.offset == start &&
2445                     relink_is_mergable(leaf, fi, new)) {
2446                         btrfs_set_file_extent_num_bytes(leaf, fi,
2447                                                         extent_len + len);
2448                         btrfs_mark_buffer_dirty(leaf);
2449                         inode_add_bytes(inode, len);
2450
2451                         ret = 1;
2452                         goto out_free_path;
2453                 } else {
2454                         merge = false;
2455                         btrfs_release_path(path);
2456                         goto again;
2457                 }
2458         }
2459
2460         ret = btrfs_insert_empty_item(trans, root, path, &key,
2461                                         sizeof(*extent));
2462         if (ret) {
2463                 btrfs_abort_transaction(trans, root, ret);
2464                 goto out_free_path;
2465         }
2466
2467         leaf = path->nodes[0];
2468         item = btrfs_item_ptr(leaf, path->slots[0],
2469                                 struct btrfs_file_extent_item);
2470         btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2471         btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2472         btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2473         btrfs_set_file_extent_num_bytes(leaf, item, len);
2474         btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2475         btrfs_set_file_extent_generation(leaf, item, trans->transid);
2476         btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2477         btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2478         btrfs_set_file_extent_encryption(leaf, item, 0);
2479         btrfs_set_file_extent_other_encoding(leaf, item, 0);
2480
2481         btrfs_mark_buffer_dirty(leaf);
2482         inode_add_bytes(inode, len);
2483         btrfs_release_path(path);
2484
2485         ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2486                         new->disk_len, 0,
2487                         backref->root_id, backref->inum,
2488                         new->file_pos, 0);      /* start - extent_offset */
2489         if (ret) {
2490                 btrfs_abort_transaction(trans, root, ret);
2491                 goto out_free_path;
2492         }
2493
2494         ret = 1;
2495 out_free_path:
2496         btrfs_release_path(path);
2497         path->leave_spinning = 0;
2498         btrfs_end_transaction(trans, root);
2499 out_unlock:
2500         unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2501                              &cached, GFP_NOFS);
2502         iput(inode);
2503         return ret;
2504 }
2505
2506 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2507 {
2508         struct old_sa_defrag_extent *old, *tmp;
2509
2510         if (!new)
2511                 return;
2512
2513         list_for_each_entry_safe(old, tmp, &new->head, list) {
2514                 list_del(&old->list);
2515                 kfree(old);
2516         }
2517         kfree(new);
2518 }
2519
2520 static void relink_file_extents(struct new_sa_defrag_extent *new)
2521 {
2522         struct btrfs_path *path;
2523         struct sa_defrag_extent_backref *backref;
2524         struct sa_defrag_extent_backref *prev = NULL;
2525         struct inode *inode;
2526         struct btrfs_root *root;
2527         struct rb_node *node;
2528         int ret;
2529
2530         inode = new->inode;
2531         root = BTRFS_I(inode)->root;
2532
2533         path = btrfs_alloc_path();
2534         if (!path)
2535                 return;
2536
2537         if (!record_extent_backrefs(path, new)) {
2538                 btrfs_free_path(path);
2539                 goto out;
2540         }
2541         btrfs_release_path(path);
2542
2543         while (1) {
2544                 node = rb_first(&new->root);
2545                 if (!node)
2546                         break;
2547                 rb_erase(node, &new->root);
2548
2549                 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2550
2551                 ret = relink_extent_backref(path, prev, backref);
2552                 WARN_ON(ret < 0);
2553
2554                 kfree(prev);
2555
2556                 if (ret == 1)
2557                         prev = backref;
2558                 else
2559                         prev = NULL;
2560                 cond_resched();
2561         }
2562         kfree(prev);
2563
2564         btrfs_free_path(path);
2565 out:
2566         free_sa_defrag_extent(new);
2567
2568         atomic_dec(&root->fs_info->defrag_running);
2569         wake_up(&root->fs_info->transaction_wait);
2570 }
2571
2572 static struct new_sa_defrag_extent *
2573 record_old_file_extents(struct inode *inode,
2574                         struct btrfs_ordered_extent *ordered)
2575 {
2576         struct btrfs_root *root = BTRFS_I(inode)->root;
2577         struct btrfs_path *path;
2578         struct btrfs_key key;
2579         struct old_sa_defrag_extent *old;
2580         struct new_sa_defrag_extent *new;
2581         int ret;
2582
2583         new = kmalloc(sizeof(*new), GFP_NOFS);
2584         if (!new)
2585                 return NULL;
2586
2587         new->inode = inode;
2588         new->file_pos = ordered->file_offset;
2589         new->len = ordered->len;
2590         new->bytenr = ordered->start;
2591         new->disk_len = ordered->disk_len;
2592         new->compress_type = ordered->compress_type;
2593         new->root = RB_ROOT;
2594         INIT_LIST_HEAD(&new->head);
2595
2596         path = btrfs_alloc_path();
2597         if (!path)
2598                 goto out_kfree;
2599
2600         key.objectid = btrfs_ino(inode);
2601         key.type = BTRFS_EXTENT_DATA_KEY;
2602         key.offset = new->file_pos;
2603
2604         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2605         if (ret < 0)
2606                 goto out_free_path;
2607         if (ret > 0 && path->slots[0] > 0)
2608                 path->slots[0]--;
2609
2610         /* find out all the old extents for the file range */
2611         while (1) {
2612                 struct btrfs_file_extent_item *extent;
2613                 struct extent_buffer *l;
2614                 int slot;
2615                 u64 num_bytes;
2616                 u64 offset;
2617                 u64 end;
2618                 u64 disk_bytenr;
2619                 u64 extent_offset;
2620
2621                 l = path->nodes[0];
2622                 slot = path->slots[0];
2623
2624                 if (slot >= btrfs_header_nritems(l)) {
2625                         ret = btrfs_next_leaf(root, path);
2626                         if (ret < 0)
2627                                 goto out_free_path;
2628                         else if (ret > 0)
2629                                 break;
2630                         continue;
2631                 }
2632
2633                 btrfs_item_key_to_cpu(l, &key, slot);
2634
2635                 if (key.objectid != btrfs_ino(inode))
2636                         break;
2637                 if (key.type != BTRFS_EXTENT_DATA_KEY)
2638                         break;
2639                 if (key.offset >= new->file_pos + new->len)
2640                         break;
2641
2642                 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2643
2644                 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2645                 if (key.offset + num_bytes < new->file_pos)
2646                         goto next;
2647
2648                 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2649                 if (!disk_bytenr)
2650                         goto next;
2651
2652                 extent_offset = btrfs_file_extent_offset(l, extent);
2653
2654                 old = kmalloc(sizeof(*old), GFP_NOFS);
2655                 if (!old)
2656                         goto out_free_path;
2657
2658                 offset = max(new->file_pos, key.offset);
2659                 end = min(new->file_pos + new->len, key.offset + num_bytes);
2660
2661                 old->bytenr = disk_bytenr;
2662                 old->extent_offset = extent_offset;
2663                 old->offset = offset - key.offset;
2664                 old->len = end - offset;
2665                 old->new = new;
2666                 old->count = 0;
2667                 list_add_tail(&old->list, &new->head);
2668 next:
2669                 path->slots[0]++;
2670                 cond_resched();
2671         }
2672
2673         btrfs_free_path(path);
2674         atomic_inc(&root->fs_info->defrag_running);
2675
2676         return new;
2677
2678 out_free_path:
2679         btrfs_free_path(path);
2680 out_kfree:
2681         free_sa_defrag_extent(new);
2682         return NULL;
2683 }
2684
2685 static void btrfs_release_delalloc_bytes(struct btrfs_root *root,
2686                                          u64 start, u64 len)
2687 {
2688         struct btrfs_block_group_cache *cache;
2689
2690         cache = btrfs_lookup_block_group(root->fs_info, start);
2691         ASSERT(cache);
2692
2693         spin_lock(&cache->lock);
2694         cache->delalloc_bytes -= len;
2695         spin_unlock(&cache->lock);
2696
2697         btrfs_put_block_group(cache);
2698 }
2699
2700 /* as ordered data IO finishes, this gets called so we can finish
2701  * an ordered extent if the range of bytes in the file it covers are
2702  * fully written.
2703  */
2704 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2705 {
2706         struct inode *inode = ordered_extent->inode;
2707         struct btrfs_root *root = BTRFS_I(inode)->root;
2708         struct btrfs_trans_handle *trans = NULL;
2709         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2710         struct extent_state *cached_state = NULL;
2711         struct new_sa_defrag_extent *new = NULL;
2712         int compress_type = 0;
2713         int ret = 0;
2714         u64 logical_len = ordered_extent->len;
2715         bool nolock;
2716         bool truncated = false;
2717
2718         nolock = btrfs_is_free_space_inode(inode);
2719
2720         if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2721                 ret = -EIO;
2722                 goto out;
2723         }
2724
2725         btrfs_free_io_failure_record(inode, ordered_extent->file_offset,
2726                                      ordered_extent->file_offset +
2727                                      ordered_extent->len - 1);
2728
2729         if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2730                 truncated = true;
2731                 logical_len = ordered_extent->truncated_len;
2732                 /* Truncated the entire extent, don't bother adding */
2733                 if (!logical_len)
2734                         goto out;
2735         }
2736
2737         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2738                 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2739                 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2740                 if (nolock)
2741                         trans = btrfs_join_transaction_nolock(root);
2742                 else
2743                         trans = btrfs_join_transaction(root);
2744                 if (IS_ERR(trans)) {
2745                         ret = PTR_ERR(trans);
2746                         trans = NULL;
2747                         goto out;
2748                 }
2749                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2750                 ret = btrfs_update_inode_fallback(trans, root, inode);
2751                 if (ret) /* -ENOMEM or corruption */
2752                         btrfs_abort_transaction(trans, root, ret);
2753                 goto out;
2754         }
2755
2756         lock_extent_bits(io_tree, ordered_extent->file_offset,
2757                          ordered_extent->file_offset + ordered_extent->len - 1,
2758                          0, &cached_state);
2759
2760         ret = test_range_bit(io_tree, ordered_extent->file_offset,
2761                         ordered_extent->file_offset + ordered_extent->len - 1,
2762                         EXTENT_DEFRAG, 1, cached_state);
2763         if (ret) {
2764                 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2765                 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2766                         /* the inode is shared */
2767                         new = record_old_file_extents(inode, ordered_extent);
2768
2769                 clear_extent_bit(io_tree, ordered_extent->file_offset,
2770                         ordered_extent->file_offset + ordered_extent->len - 1,
2771                         EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2772         }
2773
2774         if (nolock)
2775                 trans = btrfs_join_transaction_nolock(root);
2776         else
2777                 trans = btrfs_join_transaction(root);
2778         if (IS_ERR(trans)) {
2779                 ret = PTR_ERR(trans);
2780                 trans = NULL;
2781                 goto out_unlock;
2782         }
2783
2784         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2785
2786         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2787                 compress_type = ordered_extent->compress_type;
2788         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2789                 BUG_ON(compress_type);
2790                 ret = btrfs_mark_extent_written(trans, inode,
2791                                                 ordered_extent->file_offset,
2792                                                 ordered_extent->file_offset +
2793                                                 logical_len);
2794         } else {
2795                 BUG_ON(root == root->fs_info->tree_root);
2796                 ret = insert_reserved_file_extent(trans, inode,
2797                                                 ordered_extent->file_offset,
2798                                                 ordered_extent->start,
2799                                                 ordered_extent->disk_len,
2800                                                 logical_len, logical_len,
2801                                                 compress_type, 0, 0,
2802                                                 BTRFS_FILE_EXTENT_REG);
2803                 if (!ret)
2804                         btrfs_release_delalloc_bytes(root,
2805                                                      ordered_extent->start,
2806                                                      ordered_extent->disk_len);
2807         }
2808         unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2809                            ordered_extent->file_offset, ordered_extent->len,
2810                            trans->transid);
2811         if (ret < 0) {
2812                 btrfs_abort_transaction(trans, root, ret);
2813                 goto out_unlock;
2814         }
2815
2816         add_pending_csums(trans, inode, ordered_extent->file_offset,
2817                           &ordered_extent->list);
2818
2819         btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2820         ret = btrfs_update_inode_fallback(trans, root, inode);
2821         if (ret) { /* -ENOMEM or corruption */
2822                 btrfs_abort_transaction(trans, root, ret);
2823                 goto out_unlock;
2824         }
2825         ret = 0;
2826 out_unlock:
2827         unlock_extent_cached(io_tree, ordered_extent->file_offset,
2828                              ordered_extent->file_offset +
2829                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
2830 out:
2831         if (root != root->fs_info->tree_root)
2832                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2833         if (trans)
2834                 btrfs_end_transaction(trans, root);
2835
2836         if (ret || truncated) {
2837                 u64 start, end;
2838
2839                 if (truncated)
2840                         start = ordered_extent->file_offset + logical_len;
2841                 else
2842                         start = ordered_extent->file_offset;
2843                 end = ordered_extent->file_offset + ordered_extent->len - 1;
2844                 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2845
2846                 /* Drop the cache for the part of the extent we didn't write. */
2847                 btrfs_drop_extent_cache(inode, start, end, 0);
2848
2849                 /*
2850                  * If the ordered extent had an IOERR or something else went
2851                  * wrong we need to return the space for this ordered extent
2852                  * back to the allocator.  We only free the extent in the
2853                  * truncated case if we didn't write out the extent at all.
2854                  */
2855                 if ((ret || !logical_len) &&
2856                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2857                     !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2858                         btrfs_free_reserved_extent(root, ordered_extent->start,
2859                                                    ordered_extent->disk_len, 1);
2860         }
2861
2862
2863         /*
2864          * This needs to be done to make sure anybody waiting knows we are done
2865          * updating everything for this ordered extent.
2866          */
2867         btrfs_remove_ordered_extent(inode, ordered_extent);
2868
2869         /* for snapshot-aware defrag */
2870         if (new) {
2871                 if (ret) {
2872                         free_sa_defrag_extent(new);
2873                         atomic_dec(&root->fs_info->defrag_running);
2874                 } else {
2875                         relink_file_extents(new);
2876                 }
2877         }
2878
2879         /* once for us */
2880         btrfs_put_ordered_extent(ordered_extent);
2881         /* once for the tree */
2882         btrfs_put_ordered_extent(ordered_extent);
2883
2884         return ret;
2885 }
2886
2887 static void finish_ordered_fn(struct btrfs_work *work)
2888 {
2889         struct btrfs_ordered_extent *ordered_extent;
2890         ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2891         btrfs_finish_ordered_io(ordered_extent);
2892 }
2893
2894 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2895                                 struct extent_state *state, int uptodate)
2896 {
2897         struct inode *inode = page->mapping->host;
2898         struct btrfs_root *root = BTRFS_I(inode)->root;
2899         struct btrfs_ordered_extent *ordered_extent = NULL;
2900         struct btrfs_workqueue *wq;
2901         btrfs_work_func_t func;
2902
2903         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2904
2905         ClearPagePrivate2(page);
2906         if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2907                                             end - start + 1, uptodate))
2908                 return 0;
2909
2910         if (btrfs_is_free_space_inode(inode)) {
2911                 wq = root->fs_info->endio_freespace_worker;
2912                 func = btrfs_freespace_write_helper;
2913         } else {
2914                 wq = root->fs_info->endio_write_workers;
2915                 func = btrfs_endio_write_helper;
2916         }
2917
2918         btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
2919                         NULL);
2920         btrfs_queue_work(wq, &ordered_extent->work);
2921
2922         return 0;
2923 }
2924
2925 static int __readpage_endio_check(struct inode *inode,
2926                                   struct btrfs_io_bio *io_bio,
2927                                   int icsum, struct page *page,
2928                                   int pgoff, u64 start, size_t len)
2929 {
2930         char *kaddr;
2931         u32 csum_expected;
2932         u32 csum = ~(u32)0;
2933         static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
2934                                       DEFAULT_RATELIMIT_BURST);
2935
2936         csum_expected = *(((u32 *)io_bio->csum) + icsum);
2937
2938         kaddr = kmap_atomic(page);
2939         csum = btrfs_csum_data(kaddr + pgoff, csum,  len);
2940         btrfs_csum_final(csum, (char *)&csum);
2941         if (csum != csum_expected)
2942                 goto zeroit;
2943
2944         kunmap_atomic(kaddr);
2945         return 0;
2946 zeroit:
2947         if (__ratelimit(&_rs))
2948                 btrfs_warn(BTRFS_I(inode)->root->fs_info,
2949                            "csum failed ino %llu off %llu csum %u expected csum %u",
2950                            btrfs_ino(inode), start, csum, csum_expected);
2951         memset(kaddr + pgoff, 1, len);
2952         flush_dcache_page(page);
2953         kunmap_atomic(kaddr);
2954         if (csum_expected == 0)
2955                 return 0;
2956         return -EIO;
2957 }
2958
2959 /*
2960  * when reads are done, we need to check csums to verify the data is correct
2961  * if there's a match, we allow the bio to finish.  If not, the code in
2962  * extent_io.c will try to find good copies for us.
2963  */
2964 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
2965                                       u64 phy_offset, struct page *page,
2966                                       u64 start, u64 end, int mirror)
2967 {
2968         size_t offset = start - page_offset(page);
2969         struct inode *inode = page->mapping->host;
2970         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2971         struct btrfs_root *root = BTRFS_I(inode)->root;
2972
2973         if (PageChecked(page)) {
2974                 ClearPageChecked(page);
2975                 return 0;
2976         }
2977
2978         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2979                 return 0;
2980
2981         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2982             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2983                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2984                                   GFP_NOFS);
2985                 return 0;
2986         }
2987
2988         phy_offset >>= inode->i_sb->s_blocksize_bits;
2989         return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
2990                                       start, (size_t)(end - start + 1));
2991 }
2992
2993 struct delayed_iput {
2994         struct list_head list;
2995         struct inode *inode;
2996 };
2997
2998 /* JDM: If this is fs-wide, why can't we add a pointer to
2999  * btrfs_inode instead and avoid the allocation? */
3000 void btrfs_add_delayed_iput(struct inode *inode)
3001 {
3002         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
3003         struct delayed_iput *delayed;
3004
3005         if (atomic_add_unless(&inode->i_count, -1, 1))
3006                 return;
3007
3008         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
3009         delayed->inode = inode;
3010
3011         spin_lock(&fs_info->delayed_iput_lock);
3012         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
3013         spin_unlock(&fs_info->delayed_iput_lock);
3014 }
3015
3016 void btrfs_run_delayed_iputs(struct btrfs_root *root)
3017 {
3018         LIST_HEAD(list);
3019         struct btrfs_fs_info *fs_info = root->fs_info;
3020         struct delayed_iput *delayed;
3021         int empty;
3022
3023         spin_lock(&fs_info->delayed_iput_lock);
3024         empty = list_empty(&fs_info->delayed_iputs);
3025         spin_unlock(&fs_info->delayed_iput_lock);
3026         if (empty)
3027                 return;
3028
3029         spin_lock(&fs_info->delayed_iput_lock);
3030         list_splice_init(&fs_info->delayed_iputs, &list);
3031         spin_unlock(&fs_info->delayed_iput_lock);
3032
3033         while (!list_empty(&list)) {
3034                 delayed = list_entry(list.next, struct delayed_iput, list);
3035                 list_del(&delayed->list);
3036                 iput(delayed->inode);
3037                 kfree(delayed);
3038         }
3039 }
3040
3041 /*
3042  * This is called in transaction commit time. If there are no orphan
3043  * files in the subvolume, it removes orphan item and frees block_rsv
3044  * structure.
3045  */
3046 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3047                               struct btrfs_root *root)
3048 {
3049         struct btrfs_block_rsv *block_rsv;
3050         int ret;
3051
3052         if (atomic_read(&root->orphan_inodes) ||
3053             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3054                 return;
3055
3056         spin_lock(&root->orphan_lock);
3057         if (atomic_read(&root->orphan_inodes)) {
3058                 spin_unlock(&root->orphan_lock);
3059                 return;
3060         }
3061
3062         if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3063                 spin_unlock(&root->orphan_lock);
3064                 return;
3065         }
3066
3067         block_rsv = root->orphan_block_rsv;
3068         root->orphan_block_rsv = NULL;
3069         spin_unlock(&root->orphan_lock);
3070
3071         if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3072             btrfs_root_refs(&root->root_item) > 0) {
3073                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
3074                                             root->root_key.objectid);
3075                 if (ret)
3076                         btrfs_abort_transaction(trans, root, ret);
3077                 else
3078                         clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3079                                   &root->state);
3080         }
3081
3082         if (block_rsv) {
3083                 WARN_ON(block_rsv->size > 0);
3084                 btrfs_free_block_rsv(root, block_rsv);
3085         }
3086 }
3087
3088 /*
3089  * This creates an orphan entry for the given inode in case something goes
3090  * wrong in the middle of an unlink/truncate.
3091  *
3092  * NOTE: caller of this function should reserve 5 units of metadata for
3093  *       this function.
3094  */
3095 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
3096 {
3097         struct btrfs_root *root = BTRFS_I(inode)->root;
3098         struct btrfs_block_rsv *block_rsv = NULL;
3099         int reserve = 0;
3100         int insert = 0;
3101         int ret;
3102
3103         if (!root->orphan_block_rsv) {
3104                 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3105                 if (!block_rsv)
3106                         return -ENOMEM;
3107         }
3108
3109         spin_lock(&root->orphan_lock);
3110         if (!root->orphan_block_rsv) {
3111                 root->orphan_block_rsv = block_rsv;
3112         } else if (block_rsv) {
3113                 btrfs_free_block_rsv(root, block_rsv);
3114                 block_rsv = NULL;
3115         }
3116
3117         if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3118                               &BTRFS_I(inode)->runtime_flags)) {
3119 #if 0
3120                 /*
3121                  * For proper ENOSPC handling, we should do orphan
3122                  * cleanup when mounting. But this introduces backward
3123                  * compatibility issue.
3124                  */
3125                 if (!xchg(&root->orphan_item_inserted, 1))
3126                         insert = 2;
3127                 else
3128                         insert = 1;
3129 #endif
3130                 insert = 1;
3131                 atomic_inc(&root->orphan_inodes);
3132         }
3133
3134         if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3135                               &BTRFS_I(inode)->runtime_flags))
3136                 reserve = 1;
3137         spin_unlock(&root->orphan_lock);
3138
3139         /* grab metadata reservation from transaction handle */
3140         if (reserve) {
3141                 ret = btrfs_orphan_reserve_metadata(trans, inode);
3142                 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
3143         }
3144
3145         /* insert an orphan item to track this unlinked/truncated file */
3146         if (insert >= 1) {
3147                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3148                 if (ret) {
3149                         atomic_dec(&root->orphan_inodes);
3150                         if (reserve) {
3151                                 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3152                                           &BTRFS_I(inode)->runtime_flags);
3153                                 btrfs_orphan_release_metadata(inode);
3154                         }
3155                         if (ret != -EEXIST) {
3156                                 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3157                                           &BTRFS_I(inode)->runtime_flags);
3158                                 btrfs_abort_transaction(trans, root, ret);
3159                                 return ret;
3160                         }
3161                 }
3162                 ret = 0;
3163         }
3164
3165         /* insert an orphan item to track subvolume contains orphan files */
3166         if (insert >= 2) {
3167                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3168                                                root->root_key.objectid);
3169                 if (ret && ret != -EEXIST) {
3170                         btrfs_abort_transaction(trans, root, ret);
3171                         return ret;
3172                 }
3173         }
3174         return 0;
3175 }
3176
3177 /*
3178  * We have done the truncate/delete so we can go ahead and remove the orphan
3179  * item for this particular inode.
3180  */
3181 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3182                             struct inode *inode)
3183 {
3184         struct btrfs_root *root = BTRFS_I(inode)->root;
3185         int delete_item = 0;
3186         int release_rsv = 0;
3187         int ret = 0;
3188
3189         spin_lock(&root->orphan_lock);
3190         if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3191                                &BTRFS_I(inode)->runtime_flags))
3192                 delete_item = 1;
3193
3194         if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3195                                &BTRFS_I(inode)->runtime_flags))
3196                 release_rsv = 1;
3197         spin_unlock(&root->orphan_lock);
3198
3199         if (delete_item) {
3200                 atomic_dec(&root->orphan_inodes);
3201                 if (trans)
3202                         ret = btrfs_del_orphan_item(trans, root,
3203                                                     btrfs_ino(inode));
3204         }
3205
3206         if (release_rsv)
3207                 btrfs_orphan_release_metadata(inode);
3208
3209         return ret;
3210 }
3211
3212 /*
3213  * this cleans up any orphans that may be left on the list from the last use
3214  * of this root.
3215  */
3216 int btrfs_orphan_cleanup(struct btrfs_root *root)
3217 {
3218         struct btrfs_path *path;
3219         struct extent_buffer *leaf;
3220         struct btrfs_key key, found_key;
3221         struct btrfs_trans_handle *trans;
3222         struct inode *inode;
3223         u64 last_objectid = 0;
3224         int ret = 0, nr_unlink = 0, nr_truncate = 0;
3225
3226         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3227                 return 0;
3228
3229         path = btrfs_alloc_path();
3230         if (!path) {
3231                 ret = -ENOMEM;
3232                 goto out;
3233         }
3234         path->reada = -1;
3235
3236         key.objectid = BTRFS_ORPHAN_OBJECTID;
3237         key.type = BTRFS_ORPHAN_ITEM_KEY;
3238         key.offset = (u64)-1;
3239
3240         while (1) {
3241                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3242                 if (ret < 0)
3243                         goto out;
3244
3245                 /*
3246                  * if ret == 0 means we found what we were searching for, which
3247                  * is weird, but possible, so only screw with path if we didn't
3248                  * find the key and see if we have stuff that matches
3249                  */
3250                 if (ret > 0) {
3251                         ret = 0;
3252                         if (path->slots[0] == 0)
3253                                 break;
3254                         path->slots[0]--;
3255                 }
3256
3257                 /* pull out the item */
3258                 leaf = path->nodes[0];
3259                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3260
3261                 /* make sure the item matches what we want */
3262                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3263                         break;
3264                 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3265                         break;
3266
3267                 /* release the path since we're done with it */
3268                 btrfs_release_path(path);
3269
3270                 /*
3271                  * this is where we are basically btrfs_lookup, without the
3272                  * crossing root thing.  we store the inode number in the
3273                  * offset of the orphan item.
3274                  */
3275
3276                 if (found_key.offset == last_objectid) {
3277                         btrfs_err(root->fs_info,
3278                                 "Error removing orphan entry, stopping orphan cleanup");
3279                         ret = -EINVAL;
3280                         goto out;
3281                 }
3282
3283                 last_objectid = found_key.offset;
3284
3285                 found_key.objectid = found_key.offset;
3286                 found_key.type = BTRFS_INODE_ITEM_KEY;
3287                 found_key.offset = 0;
3288                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3289                 ret = PTR_ERR_OR_ZERO(inode);
3290                 if (ret && ret != -ESTALE)
3291                         goto out;
3292
3293                 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3294                         struct btrfs_root *dead_root;
3295                         struct btrfs_fs_info *fs_info = root->fs_info;
3296                         int is_dead_root = 0;
3297
3298                         /*
3299                          * this is an orphan in the tree root. Currently these
3300                          * could come from 2 sources:
3301                          *  a) a snapshot deletion in progress
3302                          *  b) a free space cache inode
3303                          * We need to distinguish those two, as the snapshot
3304                          * orphan must not get deleted.
3305                          * find_dead_roots already ran before us, so if this
3306                          * is a snapshot deletion, we should find the root
3307                          * in the dead_roots list
3308                          */
3309                         spin_lock(&fs_info->trans_lock);
3310                         list_for_each_entry(dead_root, &fs_info->dead_roots,
3311                                             root_list) {
3312                                 if (dead_root->root_key.objectid ==
3313                                     found_key.objectid) {
3314                                         is_dead_root = 1;
3315                                         break;
3316                                 }
3317                         }
3318                         spin_unlock(&fs_info->trans_lock);
3319                         if (is_dead_root) {
3320                                 /* prevent this orphan from being found again */
3321                                 key.offset = found_key.objectid - 1;
3322                                 continue;
3323                         }
3324                 }
3325                 /*
3326                  * Inode is already gone but the orphan item is still there,
3327                  * kill the orphan item.
3328                  */
3329                 if (ret == -ESTALE) {
3330                         trans = btrfs_start_transaction(root, 1);
3331                         if (IS_ERR(trans)) {
3332                                 ret = PTR_ERR(trans);
3333                                 goto out;
3334                         }
3335                         btrfs_debug(root->fs_info, "auto deleting %Lu",
3336                                 found_key.objectid);
3337                         ret = btrfs_del_orphan_item(trans, root,
3338                                                     found_key.objectid);
3339                         btrfs_end_transaction(trans, root);
3340                         if (ret)
3341                                 goto out;
3342                         continue;
3343                 }
3344
3345                 /*
3346                  * add this inode to the orphan list so btrfs_orphan_del does
3347                  * the proper thing when we hit it
3348                  */
3349                 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3350                         &BTRFS_I(inode)->runtime_flags);
3351                 atomic_inc(&root->orphan_inodes);
3352
3353                 /* if we have links, this was a truncate, lets do that */
3354                 if (inode->i_nlink) {
3355                         if (WARN_ON(!S_ISREG(inode->i_mode))) {
3356                                 iput(inode);
3357                                 continue;
3358                         }
3359                         nr_truncate++;
3360
3361                         /* 1 for the orphan item deletion. */
3362                         trans = btrfs_start_transaction(root, 1);
3363                         if (IS_ERR(trans)) {
3364                                 iput(inode);
3365                                 ret = PTR_ERR(trans);
3366                                 goto out;
3367                         }
3368                         ret = btrfs_orphan_add(trans, inode);
3369                         btrfs_end_transaction(trans, root);
3370                         if (ret) {
3371                                 iput(inode);
3372                                 goto out;
3373                         }
3374
3375                         ret = btrfs_truncate(inode);
3376                         if (ret)
3377                                 btrfs_orphan_del(NULL, inode);
3378                 } else {
3379                         nr_unlink++;
3380                 }
3381
3382                 /* this will do delete_inode and everything for us */
3383                 iput(inode);
3384                 if (ret)
3385                         goto out;
3386         }
3387         /* release the path since we're done with it */
3388         btrfs_release_path(path);
3389
3390         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3391
3392         if (root->orphan_block_rsv)
3393                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3394                                         (u64)-1);
3395
3396         if (root->orphan_block_rsv ||
3397             test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3398                 trans = btrfs_join_transaction(root);
3399                 if (!IS_ERR(trans))
3400                         btrfs_end_transaction(trans, root);
3401         }
3402
3403         if (nr_unlink)
3404                 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3405         if (nr_truncate)
3406                 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3407
3408 out:
3409         if (ret)
3410                 btrfs_err(root->fs_info,
3411                         "could not do orphan cleanup %d", ret);
3412         btrfs_free_path(path);
3413         return ret;
3414 }
3415
3416 /*
3417  * very simple check to peek ahead in the leaf looking for xattrs.  If we
3418  * don't find any xattrs, we know there can't be any acls.
3419  *
3420  * slot is the slot the inode is in, objectid is the objectid of the inode
3421  */
3422 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3423                                           int slot, u64 objectid,
3424                                           int *first_xattr_slot)
3425 {
3426         u32 nritems = btrfs_header_nritems(leaf);
3427         struct btrfs_key found_key;
3428         static u64 xattr_access = 0;
3429         static u64 xattr_default = 0;
3430         int scanned = 0;
3431
3432         if (!xattr_access) {
3433                 xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS,
3434                                         strlen(POSIX_ACL_XATTR_ACCESS));
3435                 xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT,
3436                                         strlen(POSIX_ACL_XATTR_DEFAULT));
3437         }
3438
3439         slot++;
3440         *first_xattr_slot = -1;
3441         while (slot < nritems) {
3442                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3443
3444                 /* we found a different objectid, there must not be acls */
3445                 if (found_key.objectid != objectid)
3446                         return 0;
3447
3448                 /* we found an xattr, assume we've got an acl */
3449                 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3450                         if (*first_xattr_slot == -1)
3451                                 *first_xattr_slot = slot;
3452                         if (found_key.offset == xattr_access ||
3453                             found_key.offset == xattr_default)
3454                                 return 1;
3455                 }
3456
3457                 /*
3458                  * we found a key greater than an xattr key, there can't
3459                  * be any acls later on
3460                  */
3461                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3462                         return 0;
3463
3464                 slot++;
3465                 scanned++;
3466
3467                 /*
3468                  * it goes inode, inode backrefs, xattrs, extents,
3469                  * so if there are a ton of hard links to an inode there can
3470                  * be a lot of backrefs.  Don't waste time searching too hard,
3471                  * this is just an optimization
3472                  */
3473                 if (scanned >= 8)
3474                         break;
3475         }
3476         /* we hit the end of the leaf before we found an xattr or
3477          * something larger than an xattr.  We have to assume the inode
3478          * has acls
3479          */
3480         if (*first_xattr_slot == -1)
3481                 *first_xattr_slot = slot;
3482         return 1;
3483 }
3484
3485 /*
3486  * read an inode from the btree into the in-memory inode
3487  */
3488 static void btrfs_read_locked_inode(struct inode *inode)
3489 {
3490         struct btrfs_path *path;
3491         struct extent_buffer *leaf;
3492         struct btrfs_inode_item *inode_item;
3493         struct btrfs_timespec *tspec;
3494         struct btrfs_root *root = BTRFS_I(inode)->root;
3495         struct btrfs_key location;
3496         unsigned long ptr;
3497         int maybe_acls;
3498         u32 rdev;
3499         int ret;
3500         bool filled = false;
3501         int first_xattr_slot;
3502
3503         ret = btrfs_fill_inode(inode, &rdev);
3504         if (!ret)
3505                 filled = true;
3506
3507         path = btrfs_alloc_path();
3508         if (!path)
3509                 goto make_bad;
3510
3511         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3512
3513         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3514         if (ret)
3515                 goto make_bad;
3516
3517         leaf = path->nodes[0];
3518
3519         if (filled)
3520                 goto cache_index;
3521
3522         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3523                                     struct btrfs_inode_item);
3524         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3525         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3526         i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3527         i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3528         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3529
3530         tspec = btrfs_inode_atime(inode_item);
3531         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3532         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3533
3534         tspec = btrfs_inode_mtime(inode_item);
3535         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3536         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3537
3538         tspec = btrfs_inode_ctime(inode_item);
3539         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3540         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3541
3542         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3543         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3544         BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3545
3546         /*
3547          * If we were modified in the current generation and evicted from memory
3548          * and then re-read we need to do a full sync since we don't have any
3549          * idea about which extents were modified before we were evicted from
3550          * cache.
3551          */
3552         if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3553                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3554                         &BTRFS_I(inode)->runtime_flags);
3555
3556         inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3557         inode->i_generation = BTRFS_I(inode)->generation;
3558         inode->i_rdev = 0;
3559         rdev = btrfs_inode_rdev(leaf, inode_item);
3560
3561         BTRFS_I(inode)->index_cnt = (u64)-1;
3562         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3563
3564 cache_index:
3565         path->slots[0]++;
3566         if (inode->i_nlink != 1 ||
3567             path->slots[0] >= btrfs_header_nritems(leaf))
3568                 goto cache_acl;
3569
3570         btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3571         if (location.objectid != btrfs_ino(inode))
3572                 goto cache_acl;
3573
3574         ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3575         if (location.type == BTRFS_INODE_REF_KEY) {
3576                 struct btrfs_inode_ref *ref;
3577
3578                 ref = (struct btrfs_inode_ref *)ptr;
3579                 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3580         } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3581                 struct btrfs_inode_extref *extref;
3582
3583                 extref = (struct btrfs_inode_extref *)ptr;
3584                 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3585                                                                      extref);
3586         }
3587 cache_acl:
3588         /*
3589          * try to precache a NULL acl entry for files that don't have
3590          * any xattrs or acls
3591          */
3592         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3593                                            btrfs_ino(inode), &first_xattr_slot);
3594         if (first_xattr_slot != -1) {
3595                 path->slots[0] = first_xattr_slot;
3596                 ret = btrfs_load_inode_props(inode, path);
3597                 if (ret)
3598                         btrfs_err(root->fs_info,
3599                                   "error loading props for ino %llu (root %llu): %d",
3600                                   btrfs_ino(inode),
3601                                   root->root_key.objectid, ret);
3602         }
3603         btrfs_free_path(path);
3604
3605         if (!maybe_acls)
3606                 cache_no_acl(inode);
3607
3608         switch (inode->i_mode & S_IFMT) {
3609         case S_IFREG:
3610                 inode->i_mapping->a_ops = &btrfs_aops;
3611                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3612                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3613                 inode->i_fop = &btrfs_file_operations;
3614                 inode->i_op = &btrfs_file_inode_operations;
3615                 break;
3616         case S_IFDIR:
3617                 inode->i_fop = &btrfs_dir_file_operations;
3618                 if (root == root->fs_info->tree_root)
3619                         inode->i_op = &btrfs_dir_ro_inode_operations;
3620                 else
3621                         inode->i_op = &btrfs_dir_inode_operations;
3622                 break;
3623         case S_IFLNK:
3624                 inode->i_op = &btrfs_symlink_inode_operations;
3625                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3626                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3627                 break;
3628         default:
3629                 inode->i_op = &btrfs_special_inode_operations;
3630                 init_special_inode(inode, inode->i_mode, rdev);
3631                 break;
3632         }
3633
3634         btrfs_update_iflags(inode);
3635         return;
3636
3637 make_bad:
3638         btrfs_free_path(path);
3639         make_bad_inode(inode);
3640 }
3641
3642 /*
3643  * given a leaf and an inode, copy the inode fields into the leaf
3644  */
3645 static void fill_inode_item(struct btrfs_trans_handle *trans,
3646                             struct extent_buffer *leaf,
3647                             struct btrfs_inode_item *item,
3648                             struct inode *inode)
3649 {
3650         struct btrfs_map_token token;
3651
3652         btrfs_init_map_token(&token);
3653
3654         btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3655         btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3656         btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3657                                    &token);
3658         btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3659         btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3660
3661         btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3662                                      inode->i_atime.tv_sec, &token);
3663         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3664                                       inode->i_atime.tv_nsec, &token);
3665
3666         btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3667                                      inode->i_mtime.tv_sec, &token);
3668         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3669                                       inode->i_mtime.tv_nsec, &token);
3670
3671         btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3672                                      inode->i_ctime.tv_sec, &token);
3673         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3674                                       inode->i_ctime.tv_nsec, &token);
3675
3676         btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3677                                      &token);
3678         btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3679                                          &token);
3680         btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3681         btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3682         btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3683         btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3684         btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3685 }
3686
3687 /*
3688  * copy everything in the in-memory inode into the btree.
3689  */
3690 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3691                                 struct btrfs_root *root, struct inode *inode)
3692 {
3693         struct btrfs_inode_item *inode_item;
3694         struct btrfs_path *path;
3695         struct extent_buffer *leaf;
3696         int ret;
3697
3698         path = btrfs_alloc_path();
3699         if (!path)
3700                 return -ENOMEM;
3701
3702         path->leave_spinning = 1;
3703         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3704                                  1);
3705         if (ret) {
3706                 if (ret > 0)
3707                         ret = -ENOENT;
3708                 goto failed;
3709         }
3710
3711         leaf = path->nodes[0];
3712         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3713                                     struct btrfs_inode_item);
3714
3715         fill_inode_item(trans, leaf, inode_item, inode);
3716         btrfs_mark_buffer_dirty(leaf);
3717         btrfs_set_inode_last_trans(trans, inode);
3718         ret = 0;
3719 failed:
3720         btrfs_free_path(path);
3721         return ret;
3722 }
3723
3724 /*
3725  * copy everything in the in-memory inode into the btree.
3726  */
3727 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3728                                 struct btrfs_root *root, struct inode *inode)
3729 {
3730         int ret;
3731
3732         /*
3733          * If the inode is a free space inode, we can deadlock during commit
3734          * if we put it into the delayed code.
3735          *
3736          * The data relocation inode should also be directly updated
3737          * without delay
3738          */
3739         if (!btrfs_is_free_space_inode(inode)
3740             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3741             && !root->fs_info->log_root_recovering) {
3742                 btrfs_update_root_times(trans, root);
3743
3744                 ret = btrfs_delayed_update_inode(trans, root, inode);
3745                 if (!ret)
3746                         btrfs_set_inode_last_trans(trans, inode);
3747                 return ret;
3748         }
3749
3750         return btrfs_update_inode_item(trans, root, inode);
3751 }
3752
3753 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3754                                          struct btrfs_root *root,
3755                                          struct inode *inode)
3756 {
3757         int ret;
3758
3759         ret = btrfs_update_inode(trans, root, inode);
3760         if (ret == -ENOSPC)
3761                 return btrfs_update_inode_item(trans, root, inode);
3762         return ret;
3763 }
3764
3765 /*
3766  * unlink helper that gets used here in inode.c and in the tree logging
3767  * recovery code.  It remove a link in a directory with a given name, and
3768  * also drops the back refs in the inode to the directory
3769  */
3770 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3771                                 struct btrfs_root *root,
3772                                 struct inode *dir, struct inode *inode,
3773                                 const char *name, int name_len)
3774 {
3775         struct btrfs_path *path;
3776         int ret = 0;
3777         struct extent_buffer *leaf;
3778         struct btrfs_dir_item *di;
3779         struct btrfs_key key;
3780         u64 index;
3781         u64 ino = btrfs_ino(inode);
3782         u64 dir_ino = btrfs_ino(dir);
3783
3784         path = btrfs_alloc_path();
3785         if (!path) {
3786                 ret = -ENOMEM;
3787                 goto out;
3788         }
3789
3790         path->leave_spinning = 1;
3791         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3792                                     name, name_len, -1);
3793         if (IS_ERR(di)) {
3794                 ret = PTR_ERR(di);
3795                 goto err;
3796         }
3797         if (!di) {
3798                 ret = -ENOENT;
3799                 goto err;
3800         }
3801         leaf = path->nodes[0];
3802         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3803         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3804         if (ret)
3805                 goto err;
3806         btrfs_release_path(path);
3807
3808         /*
3809          * If we don't have dir index, we have to get it by looking up
3810          * the inode ref, since we get the inode ref, remove it directly,
3811          * it is unnecessary to do delayed deletion.
3812          *
3813          * But if we have dir index, needn't search inode ref to get it.
3814          * Since the inode ref is close to the inode item, it is better
3815          * that we delay to delete it, and just do this deletion when
3816          * we update the inode item.
3817          */
3818         if (BTRFS_I(inode)->dir_index) {
3819                 ret = btrfs_delayed_delete_inode_ref(inode);
3820                 if (!ret) {
3821                         index = BTRFS_I(inode)->dir_index;
3822                         goto skip_backref;
3823                 }
3824         }
3825
3826         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3827                                   dir_ino, &index);
3828         if (ret) {
3829                 btrfs_info(root->fs_info,
3830                         "failed to delete reference to %.*s, inode %llu parent %llu",
3831                         name_len, name, ino, dir_ino);
3832                 btrfs_abort_transaction(trans, root, ret);
3833                 goto err;
3834         }
3835 skip_backref:
3836         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3837         if (ret) {
3838                 btrfs_abort_transaction(trans, root, ret);
3839                 goto err;
3840         }
3841
3842         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3843                                          inode, dir_ino);
3844         if (ret != 0 && ret != -ENOENT) {
3845                 btrfs_abort_transaction(trans, root, ret);
3846                 goto err;
3847         }
3848
3849         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3850                                            dir, index);
3851         if (ret == -ENOENT)
3852                 ret = 0;
3853         else if (ret)
3854                 btrfs_abort_transaction(trans, root, ret);
3855 err:
3856         btrfs_free_path(path);
3857         if (ret)
3858                 goto out;
3859
3860         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3861         inode_inc_iversion(inode);
3862         inode_inc_iversion(dir);
3863         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3864         ret = btrfs_update_inode(trans, root, dir);
3865 out:
3866         return ret;
3867 }
3868
3869 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3870                        struct btrfs_root *root,
3871                        struct inode *dir, struct inode *inode,
3872                        const char *name, int name_len)
3873 {
3874         int ret;
3875         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3876         if (!ret) {
3877                 drop_nlink(inode);
3878                 ret = btrfs_update_inode(trans, root, inode);
3879         }
3880         return ret;
3881 }
3882
3883 /*
3884  * helper to start transaction for unlink and rmdir.
3885  *
3886  * unlink and rmdir are special in btrfs, they do not always free space, so
3887  * if we cannot make our reservations the normal way try and see if there is
3888  * plenty of slack room in the global reserve to migrate, otherwise we cannot
3889  * allow the unlink to occur.
3890  */
3891 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
3892 {
3893         struct btrfs_trans_handle *trans;
3894         struct btrfs_root *root = BTRFS_I(dir)->root;
3895         int ret;
3896
3897         /*
3898          * 1 for the possible orphan item
3899          * 1 for the dir item
3900          * 1 for the dir index
3901          * 1 for the inode ref
3902          * 1 for the inode
3903          */
3904         trans = btrfs_start_transaction(root, 5);
3905         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3906                 return trans;
3907
3908         if (PTR_ERR(trans) == -ENOSPC) {
3909                 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
3910
3911                 trans = btrfs_start_transaction(root, 0);
3912                 if (IS_ERR(trans))
3913                         return trans;
3914                 ret = btrfs_cond_migrate_bytes(root->fs_info,
3915                                                &root->fs_info->trans_block_rsv,
3916                                                num_bytes, 5);
3917                 if (ret) {
3918                         btrfs_end_transaction(trans, root);
3919                         return ERR_PTR(ret);
3920                 }
3921                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3922                 trans->bytes_reserved = num_bytes;
3923         }
3924         return trans;
3925 }
3926
3927 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3928 {
3929         struct btrfs_root *root = BTRFS_I(dir)->root;
3930         struct btrfs_trans_handle *trans;
3931         struct inode *inode = dentry->d_inode;
3932         int ret;
3933
3934         trans = __unlink_start_trans(dir);
3935         if (IS_ERR(trans))
3936                 return PTR_ERR(trans);
3937
3938         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3939
3940         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3941                                  dentry->d_name.name, dentry->d_name.len);
3942         if (ret)
3943                 goto out;
3944
3945         if (inode->i_nlink == 0) {
3946                 ret = btrfs_orphan_add(trans, inode);
3947                 if (ret)
3948                         goto out;
3949         }
3950
3951 out:
3952         btrfs_end_transaction(trans, root);
3953         btrfs_btree_balance_dirty(root);
3954         return ret;
3955 }
3956
3957 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3958                         struct btrfs_root *root,
3959                         struct inode *dir, u64 objectid,
3960                         const char *name, int name_len)
3961 {
3962         struct btrfs_path *path;
3963         struct extent_buffer *leaf;
3964         struct btrfs_dir_item *di;
3965         struct btrfs_key key;
3966         u64 index;
3967         int ret;
3968         u64 dir_ino = btrfs_ino(dir);
3969
3970         path = btrfs_alloc_path();
3971         if (!path)
3972                 return -ENOMEM;
3973
3974         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3975                                    name, name_len, -1);
3976         if (IS_ERR_OR_NULL(di)) {
3977                 if (!di)
3978                         ret = -ENOENT;
3979                 else
3980                         ret = PTR_ERR(di);
3981                 goto out;
3982         }
3983
3984         leaf = path->nodes[0];
3985         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3986         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3987         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3988         if (ret) {
3989                 btrfs_abort_transaction(trans, root, ret);
3990                 goto out;
3991         }
3992         btrfs_release_path(path);
3993
3994         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3995                                  objectid, root->root_key.objectid,
3996                                  dir_ino, &index, name, name_len);
3997         if (ret < 0) {
3998                 if (ret != -ENOENT) {
3999                         btrfs_abort_transaction(trans, root, ret);
4000                         goto out;
4001                 }
4002                 di = btrfs_search_dir_index_item(root, path, dir_ino,
4003                                                  name, name_len);
4004                 if (IS_ERR_OR_NULL(di)) {
4005                         if (!di)
4006                                 ret = -ENOENT;
4007                         else
4008                                 ret = PTR_ERR(di);
4009                         btrfs_abort_transaction(trans, root, ret);
4010                         goto out;
4011                 }
4012
4013                 leaf = path->nodes[0];
4014                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4015                 btrfs_release_path(path);
4016                 index = key.offset;
4017         }
4018         btrfs_release_path(path);
4019
4020         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
4021         if (ret) {
4022                 btrfs_abort_transaction(trans, root, ret);
4023                 goto out;
4024         }
4025
4026         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4027         inode_inc_iversion(dir);
4028         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
4029         ret = btrfs_update_inode_fallback(trans, root, dir);
4030         if (ret)
4031                 btrfs_abort_transaction(trans, root, ret);
4032 out:
4033         btrfs_free_path(path);
4034         return ret;
4035 }
4036
4037 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4038 {
4039         struct inode *inode = dentry->d_inode;
4040         int err = 0;
4041         struct btrfs_root *root = BTRFS_I(dir)->root;
4042         struct btrfs_trans_handle *trans;
4043
4044         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4045                 return -ENOTEMPTY;
4046         if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
4047                 return -EPERM;
4048
4049         trans = __unlink_start_trans(dir);
4050         if (IS_ERR(trans))
4051                 return PTR_ERR(trans);
4052
4053         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4054                 err = btrfs_unlink_subvol(trans, root, dir,
4055                                           BTRFS_I(inode)->location.objectid,
4056                                           dentry->d_name.name,
4057                                           dentry->d_name.len);
4058                 goto out;
4059         }
4060
4061         err = btrfs_orphan_add(trans, inode);
4062         if (err)
4063                 goto out;
4064
4065         /* now the directory is empty */
4066         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
4067                                  dentry->d_name.name, dentry->d_name.len);
4068         if (!err)
4069                 btrfs_i_size_write(inode, 0);
4070 out:
4071         btrfs_end_transaction(trans, root);
4072         btrfs_btree_balance_dirty(root);
4073
4074         return err;
4075 }
4076
4077 /*
4078  * this can truncate away extent items, csum items and directory items.
4079  * It starts at a high offset and removes keys until it can't find
4080  * any higher than new_size
4081  *
4082  * csum items that cross the new i_size are truncated to the new size
4083  * as well.
4084  *
4085  * min_type is the minimum key type to truncate down to.  If set to 0, this
4086  * will kill all the items on this inode, including the INODE_ITEM_KEY.
4087  */
4088 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4089                                struct btrfs_root *root,
4090                                struct inode *inode,
4091                                u64 new_size, u32 min_type)
4092 {
4093         struct btrfs_path *path;
4094         struct extent_buffer *leaf;
4095         struct btrfs_file_extent_item *fi;
4096         struct btrfs_key key;
4097         struct btrfs_key found_key;
4098         u64 extent_start = 0;
4099         u64 extent_num_bytes = 0;
4100         u64 extent_offset = 0;
4101         u64 item_end = 0;
4102         u64 last_size = (u64)-1;
4103         u32 found_type = (u8)-1;
4104         int found_extent;
4105         int del_item;
4106         int pending_del_nr = 0;
4107         int pending_del_slot = 0;
4108         int extent_type = -1;
4109         int ret;
4110         int err = 0;
4111         u64 ino = btrfs_ino(inode);
4112
4113         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4114
4115         path = btrfs_alloc_path();
4116         if (!path)
4117                 return -ENOMEM;
4118         path->reada = -1;
4119
4120         /*
4121          * We want to drop from the next block forward in case this new size is
4122          * not block aligned since we will be keeping the last block of the
4123          * extent just the way it is.
4124          */
4125         if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4126             root == root->fs_info->tree_root)
4127                 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4128                                         root->sectorsize), (u64)-1, 0);
4129
4130         /*
4131          * This function is also used to drop the items in the log tree before
4132          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4133          * it is used to drop the loged items. So we shouldn't kill the delayed
4134          * items.
4135          */
4136         if (min_type == 0 && root == BTRFS_I(inode)->root)
4137                 btrfs_kill_delayed_inode_items(inode);
4138
4139         key.objectid = ino;
4140         key.offset = (u64)-1;
4141         key.type = (u8)-1;
4142
4143 search_again:
4144         path->leave_spinning = 1;
4145         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4146         if (ret < 0) {
4147                 err = ret;
4148                 goto out;
4149         }
4150
4151         if (ret > 0) {
4152                 /* there are no items in the tree for us to truncate, we're
4153                  * done
4154                  */
4155                 if (path->slots[0] == 0)
4156                         goto out;
4157                 path->slots[0]--;
4158         }
4159
4160         while (1) {
4161                 fi = NULL;
4162                 leaf = path->nodes[0];
4163                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4164                 found_type = found_key.type;
4165
4166                 if (found_key.objectid != ino)
4167                         break;
4168
4169                 if (found_type < min_type)
4170                         break;
4171
4172                 item_end = found_key.offset;
4173                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4174                         fi = btrfs_item_ptr(leaf, path->slots[0],
4175                                             struct btrfs_file_extent_item);
4176                         extent_type = btrfs_file_extent_type(leaf, fi);
4177                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4178                                 item_end +=
4179                                     btrfs_file_extent_num_bytes(leaf, fi);
4180                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4181                                 item_end += btrfs_file_extent_inline_len(leaf,
4182                                                          path->slots[0], fi);
4183                         }
4184                         item_end--;
4185                 }
4186                 if (found_type > min_type) {
4187                         del_item = 1;
4188                 } else {
4189                         if (item_end < new_size)
4190                                 break;
4191                         if (found_key.offset >= new_size)
4192                                 del_item = 1;
4193                         else
4194                                 del_item = 0;
4195                 }
4196                 found_extent = 0;
4197                 /* FIXME, shrink the extent if the ref count is only 1 */
4198                 if (found_type != BTRFS_EXTENT_DATA_KEY)
4199                         goto delete;
4200
4201                 if (del_item)
4202                         last_size = found_key.offset;
4203                 else
4204                         last_size = new_size;
4205
4206                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4207                         u64 num_dec;
4208                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4209                         if (!del_item) {
4210                                 u64 orig_num_bytes =
4211                                         btrfs_file_extent_num_bytes(leaf, fi);
4212                                 extent_num_bytes = ALIGN(new_size -
4213                                                 found_key.offset,
4214                                                 root->sectorsize);
4215                                 btrfs_set_file_extent_num_bytes(leaf, fi,
4216                                                          extent_num_bytes);
4217                                 num_dec = (orig_num_bytes -
4218                                            extent_num_bytes);
4219                                 if (test_bit(BTRFS_ROOT_REF_COWS,
4220                                              &root->state) &&
4221                                     extent_start != 0)
4222                                         inode_sub_bytes(inode, num_dec);
4223                                 btrfs_mark_buffer_dirty(leaf);
4224                         } else {
4225                                 extent_num_bytes =
4226                                         btrfs_file_extent_disk_num_bytes(leaf,
4227                                                                          fi);
4228                                 extent_offset = found_key.offset -
4229                                         btrfs_file_extent_offset(leaf, fi);
4230
4231                                 /* FIXME blocksize != 4096 */
4232                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4233                                 if (extent_start != 0) {
4234                                         found_extent = 1;
4235                                         if (test_bit(BTRFS_ROOT_REF_COWS,
4236                                                      &root->state))
4237                                                 inode_sub_bytes(inode, num_dec);
4238                                 }
4239                         }
4240                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4241                         /*
4242                          * we can't truncate inline items that have had
4243                          * special encodings
4244                          */
4245                         if (!del_item &&
4246                             btrfs_file_extent_compression(leaf, fi) == 0 &&
4247                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
4248                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4249                                 u32 size = new_size - found_key.offset;
4250
4251                                 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4252                                         inode_sub_bytes(inode, item_end + 1 -
4253                                                         new_size);
4254
4255                                 /*
4256                                  * update the ram bytes to properly reflect
4257                                  * the new size of our item
4258                                  */
4259                                 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4260                                 size =
4261                                     btrfs_file_extent_calc_inline_size(size);
4262                                 btrfs_truncate_item(root, path, size, 1);
4263                         } else if (test_bit(BTRFS_ROOT_REF_COWS,
4264                                             &root->state)) {
4265                                 inode_sub_bytes(inode, item_end + 1 -
4266                                                 found_key.offset);
4267                         }
4268                 }
4269 delete:
4270                 if (del_item) {
4271                         if (!pending_del_nr) {
4272                                 /* no pending yet, add ourselves */
4273                                 pending_del_slot = path->slots[0];
4274                                 pending_del_nr = 1;
4275                         } else if (pending_del_nr &&
4276                                    path->slots[0] + 1 == pending_del_slot) {
4277                                 /* hop on the pending chunk */
4278                                 pending_del_nr++;
4279                                 pending_del_slot = path->slots[0];
4280                         } else {
4281                                 BUG();
4282                         }
4283                 } else {
4284                         break;
4285                 }
4286                 if (found_extent &&
4287                     (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4288                      root == root->fs_info->tree_root)) {
4289                         btrfs_set_path_blocking(path);
4290                         ret = btrfs_free_extent(trans, root, extent_start,
4291                                                 extent_num_bytes, 0,
4292                                                 btrfs_header_owner(leaf),
4293                                                 ino, extent_offset, 0);
4294                         BUG_ON(ret);
4295                 }
4296
4297                 if (found_type == BTRFS_INODE_ITEM_KEY)
4298                         break;
4299
4300                 if (path->slots[0] == 0 ||
4301                     path->slots[0] != pending_del_slot) {
4302                         if (pending_del_nr) {
4303                                 ret = btrfs_del_items(trans, root, path,
4304                                                 pending_del_slot,
4305                                                 pending_del_nr);
4306                                 if (ret) {
4307                                         btrfs_abort_transaction(trans,
4308                                                                 root, ret);
4309                                         goto error;
4310                                 }
4311                                 pending_del_nr = 0;
4312                         }
4313                         btrfs_release_path(path);
4314                         goto search_again;
4315                 } else {
4316                         path->slots[0]--;
4317                 }
4318         }
4319 out:
4320         if (pending_del_nr) {
4321                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4322                                       pending_del_nr);
4323                 if (ret)
4324                         btrfs_abort_transaction(trans, root, ret);
4325         }
4326 error:
4327         if (last_size != (u64)-1 &&
4328             root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
4329                 btrfs_ordered_update_i_size(inode, last_size, NULL);
4330         btrfs_free_path(path);
4331         return err;
4332 }
4333
4334 /*
4335  * btrfs_truncate_page - read, zero a chunk and write a page
4336  * @inode - inode that we're zeroing
4337  * @from - the offset to start zeroing
4338  * @len - the length to zero, 0 to zero the entire range respective to the
4339  *      offset
4340  * @front - zero up to the offset instead of from the offset on
4341  *
4342  * This will find the page for the "from" offset and cow the page and zero the
4343  * part we want to zero.  This is used with truncate and hole punching.
4344  */
4345 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4346                         int front)
4347 {
4348         struct address_space *mapping = inode->i_mapping;
4349         struct btrfs_root *root = BTRFS_I(inode)->root;
4350         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4351         struct btrfs_ordered_extent *ordered;
4352         struct extent_state *cached_state = NULL;
4353         char *kaddr;
4354         u32 blocksize = root->sectorsize;
4355         pgoff_t index = from >> PAGE_CACHE_SHIFT;
4356         unsigned offset = from & (PAGE_CACHE_SIZE-1);
4357         struct page *page;
4358         gfp_t mask = btrfs_alloc_write_mask(mapping);
4359         int ret = 0;
4360         u64 page_start;
4361         u64 page_end;
4362
4363         if ((offset & (blocksize - 1)) == 0 &&
4364             (!len || ((len & (blocksize - 1)) == 0)))
4365                 goto out;
4366         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
4367         if (ret)
4368                 goto out;
4369
4370 again:
4371         page = find_or_create_page(mapping, index, mask);
4372         if (!page) {
4373                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4374                 ret = -ENOMEM;
4375                 goto out;
4376         }
4377
4378         page_start = page_offset(page);
4379         page_end = page_start + PAGE_CACHE_SIZE - 1;
4380
4381         if (!PageUptodate(page)) {
4382                 ret = btrfs_readpage(NULL, page);
4383                 lock_page(page);
4384                 if (page->mapping != mapping) {
4385                         unlock_page(page);
4386                         page_cache_release(page);
4387                         goto again;
4388                 }
4389                 if (!PageUptodate(page)) {
4390                         ret = -EIO;
4391                         goto out_unlock;
4392                 }
4393         }
4394         wait_on_page_writeback(page);
4395
4396         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
4397         set_page_extent_mapped(page);
4398
4399         ordered = btrfs_lookup_ordered_extent(inode, page_start);
4400         if (ordered) {
4401                 unlock_extent_cached(io_tree, page_start, page_end,
4402                                      &cached_state, GFP_NOFS);
4403                 unlock_page(page);
4404                 page_cache_release(page);
4405                 btrfs_start_ordered_extent(inode, ordered, 1);
4406                 btrfs_put_ordered_extent(ordered);
4407                 goto again;
4408         }
4409
4410         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
4411                           EXTENT_DIRTY | EXTENT_DELALLOC |
4412                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4413                           0, 0, &cached_state, GFP_NOFS);
4414
4415         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4416                                         &cached_state);
4417         if (ret) {
4418                 unlock_extent_cached(io_tree, page_start, page_end,
4419                                      &cached_state, GFP_NOFS);
4420                 goto out_unlock;
4421         }
4422
4423         if (offset != PAGE_CACHE_SIZE) {
4424                 if (!len)
4425                         len = PAGE_CACHE_SIZE - offset;
4426                 kaddr = kmap(page);
4427                 if (front)
4428                         memset(kaddr, 0, offset);
4429                 else
4430                         memset(kaddr + offset, 0, len);
4431                 flush_dcache_page(page);
4432                 kunmap(page);
4433         }
4434         ClearPageChecked(page);
4435         set_page_dirty(page);
4436         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4437                              GFP_NOFS);
4438
4439 out_unlock:
4440         if (ret)
4441                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4442         unlock_page(page);
4443         page_cache_release(page);
4444 out:
4445         return ret;
4446 }
4447
4448 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4449                              u64 offset, u64 len)
4450 {
4451         struct btrfs_trans_handle *trans;
4452         int ret;
4453
4454         /*
4455          * Still need to make sure the inode looks like it's been updated so
4456          * that any holes get logged if we fsync.
4457          */
4458         if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4459                 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4460                 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4461                 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4462                 return 0;
4463         }
4464
4465         /*
4466          * 1 - for the one we're dropping
4467          * 1 - for the one we're adding
4468          * 1 - for updating the inode.
4469          */
4470         trans = btrfs_start_transaction(root, 3);
4471         if (IS_ERR(trans))
4472                 return PTR_ERR(trans);
4473
4474         ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4475         if (ret) {
4476                 btrfs_abort_transaction(trans, root, ret);
4477                 btrfs_end_transaction(trans, root);
4478                 return ret;
4479         }
4480
4481         ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4482                                        0, 0, len, 0, len, 0, 0, 0);
4483         if (ret)
4484                 btrfs_abort_transaction(trans, root, ret);
4485         else
4486                 btrfs_update_inode(trans, root, inode);
4487         btrfs_end_transaction(trans, root);
4488         return ret;
4489 }
4490
4491 /*
4492  * This function puts in dummy file extents for the area we're creating a hole
4493  * for.  So if we are truncating this file to a larger size we need to insert
4494  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4495  * the range between oldsize and size
4496  */
4497 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4498 {
4499         struct btrfs_root *root = BTRFS_I(inode)->root;
4500         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4501         struct extent_map *em = NULL;
4502         struct extent_state *cached_state = NULL;
4503         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4504         u64 hole_start = ALIGN(oldsize, root->sectorsize);
4505         u64 block_end = ALIGN(size, root->sectorsize);
4506         u64 last_byte;
4507         u64 cur_offset;
4508         u64 hole_size;
4509         int err = 0;
4510
4511         /*
4512          * If our size started in the middle of a page we need to zero out the
4513          * rest of the page before we expand the i_size, otherwise we could
4514          * expose stale data.
4515          */
4516         err = btrfs_truncate_page(inode, oldsize, 0, 0);
4517         if (err)
4518                 return err;
4519
4520         if (size <= hole_start)
4521                 return 0;
4522
4523         while (1) {
4524                 struct btrfs_ordered_extent *ordered;
4525
4526                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
4527                                  &cached_state);
4528                 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4529                                                      block_end - hole_start);
4530                 if (!ordered)
4531                         break;
4532                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4533                                      &cached_state, GFP_NOFS);
4534                 btrfs_start_ordered_extent(inode, ordered, 1);
4535                 btrfs_put_ordered_extent(ordered);
4536         }
4537
4538         cur_offset = hole_start;
4539         while (1) {
4540                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4541                                 block_end - cur_offset, 0);
4542                 if (IS_ERR(em)) {
4543                         err = PTR_ERR(em);
4544                         em = NULL;
4545                         break;
4546                 }
4547                 last_byte = min(extent_map_end(em), block_end);
4548                 last_byte = ALIGN(last_byte , root->sectorsize);
4549                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4550                         struct extent_map *hole_em;
4551                         hole_size = last_byte - cur_offset;
4552
4553                         err = maybe_insert_hole(root, inode, cur_offset,
4554                                                 hole_size);
4555                         if (err)
4556                                 break;
4557                         btrfs_drop_extent_cache(inode, cur_offset,
4558                                                 cur_offset + hole_size - 1, 0);
4559                         hole_em = alloc_extent_map();
4560                         if (!hole_em) {
4561                                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4562                                         &BTRFS_I(inode)->runtime_flags);
4563                                 goto next;
4564                         }
4565                         hole_em->start = cur_offset;
4566                         hole_em->len = hole_size;
4567                         hole_em->orig_start = cur_offset;
4568
4569                         hole_em->block_start = EXTENT_MAP_HOLE;
4570                         hole_em->block_len = 0;
4571                         hole_em->orig_block_len = 0;
4572                         hole_em->ram_bytes = hole_size;
4573                         hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4574                         hole_em->compress_type = BTRFS_COMPRESS_NONE;
4575                         hole_em->generation = root->fs_info->generation;
4576
4577                         while (1) {
4578                                 write_lock(&em_tree->lock);
4579                                 err = add_extent_mapping(em_tree, hole_em, 1);
4580                                 write_unlock(&em_tree->lock);
4581                                 if (err != -EEXIST)
4582                                         break;
4583                                 btrfs_drop_extent_cache(inode, cur_offset,
4584                                                         cur_offset +
4585                                                         hole_size - 1, 0);
4586                         }
4587                         free_extent_map(hole_em);
4588                 }
4589 next:
4590                 free_extent_map(em);
4591                 em = NULL;
4592                 cur_offset = last_byte;
4593                 if (cur_offset >= block_end)
4594                         break;
4595         }
4596         free_extent_map(em);
4597         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4598                              GFP_NOFS);
4599         return err;
4600 }
4601
4602 static int wait_snapshoting_atomic_t(atomic_t *a)
4603 {
4604         schedule();
4605         return 0;
4606 }
4607
4608 static void wait_for_snapshot_creation(struct btrfs_root *root)
4609 {
4610         while (true) {
4611                 int ret;
4612
4613                 ret = btrfs_start_write_no_snapshoting(root);
4614                 if (ret)
4615                         break;
4616                 wait_on_atomic_t(&root->will_be_snapshoted,
4617                                  wait_snapshoting_atomic_t,
4618                                  TASK_UNINTERRUPTIBLE);
4619         }
4620 }
4621
4622 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4623 {
4624         struct btrfs_root *root = BTRFS_I(inode)->root;
4625         struct btrfs_trans_handle *trans;
4626         loff_t oldsize = i_size_read(inode);
4627         loff_t newsize = attr->ia_size;
4628         int mask = attr->ia_valid;
4629         int ret;
4630
4631         /*
4632          * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4633          * special case where we need to update the times despite not having
4634          * these flags set.  For all other operations the VFS set these flags
4635          * explicitly if it wants a timestamp update.
4636          */
4637         if (newsize != oldsize) {
4638                 inode_inc_iversion(inode);
4639                 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4640                         inode->i_ctime = inode->i_mtime =
4641                                 current_fs_time(inode->i_sb);
4642         }
4643
4644         if (newsize > oldsize) {
4645                 truncate_pagecache(inode, newsize);
4646                 /*
4647                  * Don't do an expanding truncate while snapshoting is ongoing.
4648                  * This is to ensure the snapshot captures a fully consistent
4649                  * state of this file - if the snapshot captures this expanding
4650                  * truncation, it must capture all writes that happened before
4651                  * this truncation.
4652                  */
4653                 wait_for_snapshot_creation(root);
4654                 ret = btrfs_cont_expand(inode, oldsize, newsize);
4655                 if (ret) {
4656                         btrfs_end_write_no_snapshoting(root);
4657                         return ret;
4658                 }
4659
4660                 trans = btrfs_start_transaction(root, 1);
4661                 if (IS_ERR(trans)) {
4662                         btrfs_end_write_no_snapshoting(root);
4663                         return PTR_ERR(trans);
4664                 }
4665
4666                 i_size_write(inode, newsize);
4667                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4668                 ret = btrfs_update_inode(trans, root, inode);
4669                 btrfs_end_write_no_snapshoting(root);
4670                 btrfs_end_transaction(trans, root);
4671         } else {
4672
4673                 /*
4674                  * We're truncating a file that used to have good data down to
4675                  * zero. Make sure it gets into the ordered flush list so that
4676                  * any new writes get down to disk quickly.
4677                  */
4678                 if (newsize == 0)
4679                         set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4680                                 &BTRFS_I(inode)->runtime_flags);
4681
4682                 /*
4683                  * 1 for the orphan item we're going to add
4684                  * 1 for the orphan item deletion.
4685                  */
4686                 trans = btrfs_start_transaction(root, 2);
4687                 if (IS_ERR(trans))
4688                         return PTR_ERR(trans);
4689
4690                 /*
4691                  * We need to do this in case we fail at _any_ point during the
4692                  * actual truncate.  Once we do the truncate_setsize we could
4693                  * invalidate pages which forces any outstanding ordered io to
4694                  * be instantly completed which will give us extents that need
4695                  * to be truncated.  If we fail to get an orphan inode down we
4696                  * could have left over extents that were never meant to live,
4697                  * so we need to garuntee from this point on that everything
4698                  * will be consistent.
4699                  */
4700                 ret = btrfs_orphan_add(trans, inode);
4701                 btrfs_end_transaction(trans, root);
4702                 if (ret)
4703                         return ret;
4704
4705                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4706                 truncate_setsize(inode, newsize);
4707
4708                 /* Disable nonlocked read DIO to avoid the end less truncate */
4709                 btrfs_inode_block_unlocked_dio(inode);
4710                 inode_dio_wait(inode);
4711                 btrfs_inode_resume_unlocked_dio(inode);
4712
4713                 ret = btrfs_truncate(inode);
4714                 if (ret && inode->i_nlink) {
4715                         int err;
4716
4717                         /*
4718                          * failed to truncate, disk_i_size is only adjusted down
4719                          * as we remove extents, so it should represent the true
4720                          * size of the inode, so reset the in memory size and
4721                          * delete our orphan entry.
4722                          */
4723                         trans = btrfs_join_transaction(root);
4724                         if (IS_ERR(trans)) {
4725                                 btrfs_orphan_del(NULL, inode);
4726                                 return ret;
4727                         }
4728                         i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4729                         err = btrfs_orphan_del(trans, inode);
4730                         if (err)
4731                                 btrfs_abort_transaction(trans, root, err);
4732                         btrfs_end_transaction(trans, root);
4733                 }
4734         }
4735
4736         return ret;
4737 }
4738
4739 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4740 {
4741         struct inode *inode = dentry->d_inode;
4742         struct btrfs_root *root = BTRFS_I(inode)->root;
4743         int err;
4744
4745         if (btrfs_root_readonly(root))
4746                 return -EROFS;
4747
4748         err = inode_change_ok(inode, attr);
4749         if (err)
4750                 return err;
4751
4752         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
4753                 err = btrfs_setsize(inode, attr);
4754                 if (err)
4755                         return err;
4756         }
4757
4758         if (attr->ia_valid) {
4759                 setattr_copy(inode, attr);
4760                 inode_inc_iversion(inode);
4761                 err = btrfs_dirty_inode(inode);
4762
4763                 if (!err && attr->ia_valid & ATTR_MODE)
4764                         err = posix_acl_chmod(inode, inode->i_mode);
4765         }
4766
4767         return err;
4768 }
4769
4770 /*
4771  * While truncating the inode pages during eviction, we get the VFS calling
4772  * btrfs_invalidatepage() against each page of the inode. This is slow because
4773  * the calls to btrfs_invalidatepage() result in a huge amount of calls to
4774  * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
4775  * extent_state structures over and over, wasting lots of time.
4776  *
4777  * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
4778  * those expensive operations on a per page basis and do only the ordered io
4779  * finishing, while we release here the extent_map and extent_state structures,
4780  * without the excessive merging and splitting.
4781  */
4782 static void evict_inode_truncate_pages(struct inode *inode)
4783 {
4784         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4785         struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
4786         struct rb_node *node;
4787
4788         ASSERT(inode->i_state & I_FREEING);
4789         truncate_inode_pages_final(&inode->i_data);
4790
4791         write_lock(&map_tree->lock);
4792         while (!RB_EMPTY_ROOT(&map_tree->map)) {
4793                 struct extent_map *em;
4794
4795                 node = rb_first(&map_tree->map);
4796                 em = rb_entry(node, struct extent_map, rb_node);
4797                 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
4798                 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
4799                 remove_extent_mapping(map_tree, em);
4800                 free_extent_map(em);
4801                 if (need_resched()) {
4802                         write_unlock(&map_tree->lock);
4803                         cond_resched();
4804                         write_lock(&map_tree->lock);
4805                 }
4806         }
4807         write_unlock(&map_tree->lock);
4808
4809         spin_lock(&io_tree->lock);
4810         while (!RB_EMPTY_ROOT(&io_tree->state)) {
4811                 struct extent_state *state;
4812                 struct extent_state *cached_state = NULL;
4813
4814                 node = rb_first(&io_tree->state);
4815                 state = rb_entry(node, struct extent_state, rb_node);
4816                 atomic_inc(&state->refs);
4817                 spin_unlock(&io_tree->lock);
4818
4819                 lock_extent_bits(io_tree, state->start, state->end,
4820                                  0, &cached_state);
4821                 clear_extent_bit(io_tree, state->start, state->end,
4822                                  EXTENT_LOCKED | EXTENT_DIRTY |
4823                                  EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
4824                                  EXTENT_DEFRAG, 1, 1,
4825                                  &cached_state, GFP_NOFS);
4826                 free_extent_state(state);
4827
4828                 cond_resched();
4829                 spin_lock(&io_tree->lock);
4830         }
4831         spin_unlock(&io_tree->lock);
4832 }
4833
4834 void btrfs_evict_inode(struct inode *inode)
4835 {
4836         struct btrfs_trans_handle *trans;
4837         struct btrfs_root *root = BTRFS_I(inode)->root;
4838         struct btrfs_block_rsv *rsv, *global_rsv;
4839         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
4840         int ret;
4841
4842         trace_btrfs_inode_evict(inode);
4843
4844         evict_inode_truncate_pages(inode);
4845
4846         if (inode->i_nlink &&
4847             ((btrfs_root_refs(&root->root_item) != 0 &&
4848               root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
4849              btrfs_is_free_space_inode(inode)))
4850                 goto no_delete;
4851
4852         if (is_bad_inode(inode)) {
4853                 btrfs_orphan_del(NULL, inode);
4854                 goto no_delete;
4855         }
4856         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4857         btrfs_wait_ordered_range(inode, 0, (u64)-1);
4858
4859         btrfs_free_io_failure_record(inode, 0, (u64)-1);
4860
4861         if (root->fs_info->log_root_recovering) {
4862                 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
4863                                  &BTRFS_I(inode)->runtime_flags));
4864                 goto no_delete;
4865         }
4866
4867         if (inode->i_nlink > 0) {
4868                 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
4869                        root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
4870                 goto no_delete;
4871         }
4872
4873         ret = btrfs_commit_inode_delayed_inode(inode);
4874         if (ret) {
4875                 btrfs_orphan_del(NULL, inode);
4876                 goto no_delete;
4877         }
4878
4879         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4880         if (!rsv) {
4881                 btrfs_orphan_del(NULL, inode);
4882                 goto no_delete;
4883         }
4884         rsv->size = min_size;
4885         rsv->failfast = 1;
4886         global_rsv = &root->fs_info->global_block_rsv;
4887
4888         btrfs_i_size_write(inode, 0);
4889
4890         /*
4891          * This is a bit simpler than btrfs_truncate since we've already
4892          * reserved our space for our orphan item in the unlink, so we just
4893          * need to reserve some slack space in case we add bytes and update
4894          * inode item when doing the truncate.
4895          */
4896         while (1) {
4897                 ret = btrfs_block_rsv_refill(root, rsv, min_size,
4898                                              BTRFS_RESERVE_FLUSH_LIMIT);
4899
4900                 /*
4901                  * Try and steal from the global reserve since we will
4902                  * likely not use this space anyway, we want to try as
4903                  * hard as possible to get this to work.
4904                  */
4905                 if (ret)
4906                         ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
4907
4908                 if (ret) {
4909                         btrfs_warn(root->fs_info,
4910                                 "Could not get space for a delete, will truncate on mount %d",
4911                                 ret);
4912                         btrfs_orphan_del(NULL, inode);
4913                         btrfs_free_block_rsv(root, rsv);
4914                         goto no_delete;
4915                 }
4916
4917                 trans = btrfs_join_transaction(root);
4918                 if (IS_ERR(trans)) {
4919                         btrfs_orphan_del(NULL, inode);
4920                         btrfs_free_block_rsv(root, rsv);
4921                         goto no_delete;
4922                 }
4923
4924                 trans->block_rsv = rsv;
4925
4926                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
4927                 if (ret != -ENOSPC)
4928                         break;
4929
4930                 trans->block_rsv = &root->fs_info->trans_block_rsv;
4931                 btrfs_end_transaction(trans, root);
4932                 trans = NULL;
4933                 btrfs_btree_balance_dirty(root);
4934         }
4935
4936         btrfs_free_block_rsv(root, rsv);
4937
4938         /*
4939          * Errors here aren't a big deal, it just means we leave orphan items
4940          * in the tree.  They will be cleaned up on the next mount.
4941          */
4942         if (ret == 0) {
4943                 trans->block_rsv = root->orphan_block_rsv;
4944                 btrfs_orphan_del(trans, inode);
4945         } else {
4946                 btrfs_orphan_del(NULL, inode);
4947         }
4948
4949         trans->block_rsv = &root->fs_info->trans_block_rsv;
4950         if (!(root == root->fs_info->tree_root ||
4951               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
4952                 btrfs_return_ino(root, btrfs_ino(inode));
4953
4954         btrfs_end_transaction(trans, root);
4955         btrfs_btree_balance_dirty(root);
4956 no_delete:
4957         btrfs_remove_delayed_node(inode);
4958         clear_inode(inode);
4959         return;
4960 }
4961
4962 /*
4963  * this returns the key found in the dir entry in the location pointer.
4964  * If no dir entries were found, location->objectid is 0.
4965  */
4966 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4967                                struct btrfs_key *location)
4968 {
4969         const char *name = dentry->d_name.name;
4970         int namelen = dentry->d_name.len;
4971         struct btrfs_dir_item *di;
4972         struct btrfs_path *path;
4973         struct btrfs_root *root = BTRFS_I(dir)->root;
4974         int ret = 0;
4975
4976         path = btrfs_alloc_path();
4977         if (!path)
4978                 return -ENOMEM;
4979
4980         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
4981                                     namelen, 0);
4982         if (IS_ERR(di))
4983                 ret = PTR_ERR(di);
4984
4985         if (IS_ERR_OR_NULL(di))
4986                 goto out_err;
4987
4988         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
4989 out:
4990         btrfs_free_path(path);
4991         return ret;
4992 out_err:
4993         location->objectid = 0;
4994         goto out;
4995 }
4996
4997 /*
4998  * when we hit a tree root in a directory, the btrfs part of the inode
4999  * needs to be changed to reflect the root directory of the tree root.  This
5000  * is kind of like crossing a mount point.
5001  */
5002 static int fixup_tree_root_location(struct btrfs_root *root,
5003                                     struct inode *dir,
5004                                     struct dentry *dentry,
5005                                     struct btrfs_key *location,
5006                                     struct btrfs_root **sub_root)
5007 {
5008         struct btrfs_path *path;
5009         struct btrfs_root *new_root;
5010         struct btrfs_root_ref *ref;
5011         struct extent_buffer *leaf;
5012         struct btrfs_key key;
5013         int ret;
5014         int err = 0;
5015
5016         path = btrfs_alloc_path();
5017         if (!path) {
5018                 err = -ENOMEM;
5019                 goto out;
5020         }
5021
5022         err = -ENOENT;
5023         key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5024         key.type = BTRFS_ROOT_REF_KEY;
5025         key.offset = location->objectid;
5026
5027         ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, path,
5028                                 0, 0);
5029         if (ret) {
5030                 if (ret < 0)
5031                         err = ret;
5032                 goto out;
5033         }
5034
5035         leaf = path->nodes[0];
5036         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5037         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5038             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5039                 goto out;
5040
5041         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5042                                    (unsigned long)(ref + 1),
5043                                    dentry->d_name.len);
5044         if (ret)
5045                 goto out;
5046
5047         btrfs_release_path(path);
5048
5049         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
5050         if (IS_ERR(new_root)) {
5051                 err = PTR_ERR(new_root);
5052                 goto out;
5053         }
5054
5055         *sub_root = new_root;
5056         location->objectid = btrfs_root_dirid(&new_root->root_item);
5057         location->type = BTRFS_INODE_ITEM_KEY;
5058         location->offset = 0;
5059         err = 0;
5060 out:
5061         btrfs_free_path(path);
5062         return err;
5063 }
5064
5065 static void inode_tree_add(struct inode *inode)
5066 {
5067         struct btrfs_root *root = BTRFS_I(inode)->root;
5068         struct btrfs_inode *entry;
5069         struct rb_node **p;
5070         struct rb_node *parent;
5071         struct rb_node *new = &BTRFS_I(inode)->rb_node;
5072         u64 ino = btrfs_ino(inode);
5073
5074         if (inode_unhashed(inode))
5075                 return;
5076         parent = NULL;
5077         spin_lock(&root->inode_lock);
5078         p = &root->inode_tree.rb_node;
5079         while (*p) {
5080                 parent = *p;
5081                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5082
5083                 if (ino < btrfs_ino(&entry->vfs_inode))
5084                         p = &parent->rb_left;
5085                 else if (ino > btrfs_ino(&entry->vfs_inode))
5086                         p = &parent->rb_right;
5087                 else {
5088                         WARN_ON(!(entry->vfs_inode.i_state &
5089                                   (I_WILL_FREE | I_FREEING)));
5090                         rb_replace_node(parent, new, &root->inode_tree);
5091                         RB_CLEAR_NODE(parent);
5092                         spin_unlock(&root->inode_lock);
5093                         return;
5094                 }
5095         }
5096         rb_link_node(new, parent, p);
5097         rb_insert_color(new, &root->inode_tree);
5098         spin_unlock(&root->inode_lock);
5099 }
5100
5101 static void inode_tree_del(struct inode *inode)
5102 {
5103         struct btrfs_root *root = BTRFS_I(inode)->root;
5104         int empty = 0;
5105
5106         spin_lock(&root->inode_lock);
5107         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5108                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5109                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5110                 empty = RB_EMPTY_ROOT(&root->inode_tree);
5111         }
5112         spin_unlock(&root->inode_lock);
5113
5114         if (empty && btrfs_root_refs(&root->root_item) == 0) {
5115                 synchronize_srcu(&root->fs_info->subvol_srcu);
5116                 spin_lock(&root->inode_lock);
5117                 empty = RB_EMPTY_ROOT(&root->inode_tree);
5118                 spin_unlock(&root->inode_lock);
5119                 if (empty)
5120                         btrfs_add_dead_root(root);
5121         }
5122 }
5123
5124 void btrfs_invalidate_inodes(struct btrfs_root *root)
5125 {
5126         struct rb_node *node;
5127         struct rb_node *prev;
5128         struct btrfs_inode *entry;
5129         struct inode *inode;
5130         u64 objectid = 0;
5131
5132         if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
5133                 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5134
5135         spin_lock(&root->inode_lock);
5136 again:
5137         node = root->inode_tree.rb_node;
5138         prev = NULL;
5139         while (node) {
5140                 prev = node;
5141                 entry = rb_entry(node, struct btrfs_inode, rb_node);
5142
5143                 if (objectid < btrfs_ino(&entry->vfs_inode))
5144                         node = node->rb_left;
5145                 else if (objectid > btrfs_ino(&entry->vfs_inode))
5146                         node = node->rb_right;
5147                 else
5148                         break;
5149         }
5150         if (!node) {
5151                 while (prev) {
5152                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
5153                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
5154                                 node = prev;
5155                                 break;
5156                         }
5157                         prev = rb_next(prev);
5158                 }
5159         }
5160         while (node) {
5161                 entry = rb_entry(node, struct btrfs_inode, rb_node);
5162                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
5163                 inode = igrab(&entry->vfs_inode);
5164                 if (inode) {
5165                         spin_unlock(&root->inode_lock);
5166                         if (atomic_read(&inode->i_count) > 1)
5167                                 d_prune_aliases(inode);
5168                         /*
5169                          * btrfs_drop_inode will have it removed from
5170                          * the inode cache when its usage count
5171                          * hits zero.
5172                          */
5173                         iput(inode);
5174                         cond_resched();
5175                         spin_lock(&root->inode_lock);
5176                         goto again;
5177                 }
5178
5179                 if (cond_resched_lock(&root->inode_lock))
5180                         goto again;
5181
5182                 node = rb_next(node);
5183         }
5184         spin_unlock(&root->inode_lock);
5185 }
5186
5187 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5188 {
5189         struct btrfs_iget_args *args = p;
5190         inode->i_ino = args->location->objectid;
5191         memcpy(&BTRFS_I(inode)->location, args->location,
5192                sizeof(*args->location));
5193         BTRFS_I(inode)->root = args->root;
5194         return 0;
5195 }
5196
5197 static int btrfs_find_actor(struct inode *inode, void *opaque)
5198 {
5199         struct btrfs_iget_args *args = opaque;
5200         return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5201                 args->root == BTRFS_I(inode)->root;
5202 }
5203
5204 static struct inode *btrfs_iget_locked(struct super_block *s,
5205                                        struct btrfs_key *location,
5206                                        struct btrfs_root *root)
5207 {
5208         struct inode *inode;
5209         struct btrfs_iget_args args;
5210         unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5211
5212         args.location = location;
5213         args.root = root;
5214
5215         inode = iget5_locked(s, hashval, btrfs_find_actor,
5216                              btrfs_init_locked_inode,
5217                              (void *)&args);
5218         return inode;
5219 }
5220
5221 /* Get an inode object given its location and corresponding root.
5222  * Returns in *is_new if the inode was read from disk
5223  */
5224 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5225                          struct btrfs_root *root, int *new)
5226 {
5227         struct inode *inode;
5228
5229         inode = btrfs_iget_locked(s, location, root);
5230         if (!inode)
5231                 return ERR_PTR(-ENOMEM);
5232
5233         if (inode->i_state & I_NEW) {
5234                 btrfs_read_locked_inode(inode);
5235                 if (!is_bad_inode(inode)) {
5236                         inode_tree_add(inode);
5237                         unlock_new_inode(inode);
5238                         if (new)
5239                                 *new = 1;
5240                 } else {
5241                         unlock_new_inode(inode);
5242                         iput(inode);
5243                         inode = ERR_PTR(-ESTALE);
5244                 }
5245         }
5246
5247         return inode;
5248 }
5249
5250 static struct inode *new_simple_dir(struct super_block *s,
5251                                     struct btrfs_key *key,
5252                                     struct btrfs_root *root)
5253 {
5254         struct inode *inode = new_inode(s);
5255
5256         if (!inode)
5257                 return ERR_PTR(-ENOMEM);
5258
5259         BTRFS_I(inode)->root = root;
5260         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5261         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5262
5263         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5264         inode->i_op = &btrfs_dir_ro_inode_operations;
5265         inode->i_fop = &simple_dir_operations;
5266         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5267         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5268
5269         return inode;
5270 }
5271
5272 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5273 {
5274         struct inode *inode;
5275         struct btrfs_root *root = BTRFS_I(dir)->root;
5276         struct btrfs_root *sub_root = root;
5277         struct btrfs_key location;
5278         int index;
5279         int ret = 0;
5280
5281         if (dentry->d_name.len > BTRFS_NAME_LEN)
5282                 return ERR_PTR(-ENAMETOOLONG);
5283
5284         ret = btrfs_inode_by_name(dir, dentry, &location);
5285         if (ret < 0)
5286                 return ERR_PTR(ret);
5287
5288         if (location.objectid == 0)
5289                 return ERR_PTR(-ENOENT);
5290
5291         if (location.type == BTRFS_INODE_ITEM_KEY) {
5292                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5293                 return inode;
5294         }
5295
5296         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5297
5298         index = srcu_read_lock(&root->fs_info->subvol_srcu);
5299         ret = fixup_tree_root_location(root, dir, dentry,
5300                                        &location, &sub_root);
5301         if (ret < 0) {
5302                 if (ret != -ENOENT)
5303                         inode = ERR_PTR(ret);
5304                 else
5305                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
5306         } else {
5307                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5308         }
5309         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5310
5311         if (!IS_ERR(inode) && root != sub_root) {
5312                 down_read(&root->fs_info->cleanup_work_sem);
5313                 if (!(inode->i_sb->s_flags & MS_RDONLY))
5314                         ret = btrfs_orphan_cleanup(sub_root);
5315                 up_read(&root->fs_info->cleanup_work_sem);
5316                 if (ret) {
5317                         iput(inode);
5318                         inode = ERR_PTR(ret);
5319                 }
5320         }
5321
5322         return inode;
5323 }
5324
5325 static int btrfs_dentry_delete(const struct dentry *dentry)
5326 {
5327         struct btrfs_root *root;
5328         struct inode *inode = dentry->d_inode;
5329
5330         if (!inode && !IS_ROOT(dentry))
5331                 inode = dentry->d_parent->d_inode;
5332
5333         if (inode) {
5334                 root = BTRFS_I(inode)->root;
5335                 if (btrfs_root_refs(&root->root_item) == 0)
5336                         return 1;
5337
5338                 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5339                         return 1;
5340         }
5341         return 0;
5342 }
5343
5344 static void btrfs_dentry_release(struct dentry *dentry)
5345 {
5346         kfree(dentry->d_fsdata);
5347 }
5348
5349 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5350                                    unsigned int flags)
5351 {
5352         struct inode *inode;
5353
5354         inode = btrfs_lookup_dentry(dir, dentry);
5355         if (IS_ERR(inode)) {
5356                 if (PTR_ERR(inode) == -ENOENT)
5357                         inode = NULL;
5358                 else
5359                         return ERR_CAST(inode);
5360         }
5361
5362         return d_splice_alias(inode, dentry);
5363 }
5364
5365 unsigned char btrfs_filetype_table[] = {
5366         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5367 };
5368
5369 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5370 {
5371         struct inode *inode = file_inode(file);
5372         struct btrfs_root *root = BTRFS_I(inode)->root;
5373         struct btrfs_item *item;
5374         struct btrfs_dir_item *di;
5375         struct btrfs_key key;
5376         struct btrfs_key found_key;
5377         struct btrfs_path *path;
5378         struct list_head ins_list;
5379         struct list_head del_list;
5380         int ret;
5381         struct extent_buffer *leaf;
5382         int slot;
5383         unsigned char d_type;
5384         int over = 0;
5385         u32 di_cur;
5386         u32 di_total;
5387         u32 di_len;
5388         int key_type = BTRFS_DIR_INDEX_KEY;
5389         char tmp_name[32];
5390         char *name_ptr;
5391         int name_len;
5392         int is_curr = 0;        /* ctx->pos points to the current index? */
5393
5394         /* FIXME, use a real flag for deciding about the key type */
5395         if (root->fs_info->tree_root == root)
5396                 key_type = BTRFS_DIR_ITEM_KEY;
5397
5398         if (!dir_emit_dots(file, ctx))
5399                 return 0;
5400
5401         path = btrfs_alloc_path();
5402         if (!path)
5403                 return -ENOMEM;
5404
5405         path->reada = 1;
5406
5407         if (key_type == BTRFS_DIR_INDEX_KEY) {
5408                 INIT_LIST_HEAD(&ins_list);
5409                 INIT_LIST_HEAD(&del_list);
5410                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5411         }
5412
5413         key.type = key_type;
5414         key.offset = ctx->pos;
5415         key.objectid = btrfs_ino(inode);
5416
5417         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5418         if (ret < 0)
5419                 goto err;
5420
5421         while (1) {
5422                 leaf = path->nodes[0];
5423                 slot = path->slots[0];
5424                 if (slot >= btrfs_header_nritems(leaf)) {
5425                         ret = btrfs_next_leaf(root, path);
5426                         if (ret < 0)
5427                                 goto err;
5428                         else if (ret > 0)
5429                                 break;
5430                         continue;
5431                 }
5432
5433                 item = btrfs_item_nr(slot);
5434                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5435
5436                 if (found_key.objectid != key.objectid)
5437                         break;
5438                 if (found_key.type != key_type)
5439                         break;
5440                 if (found_key.offset < ctx->pos)
5441                         goto next;
5442                 if (key_type == BTRFS_DIR_INDEX_KEY &&
5443                     btrfs_should_delete_dir_index(&del_list,
5444                                                   found_key.offset))
5445                         goto next;
5446
5447                 ctx->pos = found_key.offset;
5448                 is_curr = 1;
5449
5450                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5451                 di_cur = 0;
5452                 di_total = btrfs_item_size(leaf, item);
5453
5454                 while (di_cur < di_total) {
5455                         struct btrfs_key location;
5456
5457                         if (verify_dir_item(root, leaf, di))
5458                                 break;
5459
5460                         name_len = btrfs_dir_name_len(leaf, di);
5461                         if (name_len <= sizeof(tmp_name)) {
5462                                 name_ptr = tmp_name;
5463                         } else {
5464                                 name_ptr = kmalloc(name_len, GFP_NOFS);
5465                                 if (!name_ptr) {
5466                                         ret = -ENOMEM;
5467                                         goto err;
5468                                 }
5469                         }
5470                         read_extent_buffer(leaf, name_ptr,
5471                                            (unsigned long)(di + 1), name_len);
5472
5473                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5474                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
5475
5476
5477                         /* is this a reference to our own snapshot? If so
5478                          * skip it.
5479                          *
5480                          * In contrast to old kernels, we insert the snapshot's
5481                          * dir item and dir index after it has been created, so
5482                          * we won't find a reference to our own snapshot. We
5483                          * still keep the following code for backward
5484                          * compatibility.
5485                          */
5486                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
5487                             location.objectid == root->root_key.objectid) {
5488                                 over = 0;
5489                                 goto skip;
5490                         }
5491                         over = !dir_emit(ctx, name_ptr, name_len,
5492                                        location.objectid, d_type);
5493
5494 skip:
5495                         if (name_ptr != tmp_name)
5496                                 kfree(name_ptr);
5497
5498                         if (over)
5499                                 goto nopos;
5500                         di_len = btrfs_dir_name_len(leaf, di) +
5501                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
5502                         di_cur += di_len;
5503                         di = (struct btrfs_dir_item *)((char *)di + di_len);
5504                 }
5505 next:
5506                 path->slots[0]++;
5507         }
5508
5509         if (key_type == BTRFS_DIR_INDEX_KEY) {
5510                 if (is_curr)
5511                         ctx->pos++;
5512                 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5513                 if (ret)
5514                         goto nopos;
5515         }
5516
5517         /* Reached end of directory/root. Bump pos past the last item. */
5518         ctx->pos++;
5519
5520         /*
5521          * Stop new entries from being returned after we return the last
5522          * entry.
5523          *
5524          * New directory entries are assigned a strictly increasing
5525          * offset.  This means that new entries created during readdir
5526          * are *guaranteed* to be seen in the future by that readdir.
5527          * This has broken buggy programs which operate on names as
5528          * they're returned by readdir.  Until we re-use freed offsets
5529          * we have this hack to stop new entries from being returned
5530          * under the assumption that they'll never reach this huge
5531          * offset.
5532          *
5533          * This is being careful not to overflow 32bit loff_t unless the
5534          * last entry requires it because doing so has broken 32bit apps
5535          * in the past.
5536          */
5537         if (key_type == BTRFS_DIR_INDEX_KEY) {
5538                 if (ctx->pos >= INT_MAX)
5539                         ctx->pos = LLONG_MAX;
5540                 else
5541                         ctx->pos = INT_MAX;
5542         }
5543 nopos:
5544         ret = 0;
5545 err:
5546         if (key_type == BTRFS_DIR_INDEX_KEY)
5547                 btrfs_put_delayed_items(&ins_list, &del_list);
5548         btrfs_free_path(path);
5549         return ret;
5550 }
5551
5552 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5553 {
5554         struct btrfs_root *root = BTRFS_I(inode)->root;
5555         struct btrfs_trans_handle *trans;
5556         int ret = 0;
5557         bool nolock = false;
5558
5559         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5560                 return 0;
5561
5562         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5563                 nolock = true;
5564
5565         if (wbc->sync_mode == WB_SYNC_ALL) {
5566                 if (nolock)
5567                         trans = btrfs_join_transaction_nolock(root);
5568                 else
5569                         trans = btrfs_join_transaction(root);
5570                 if (IS_ERR(trans))
5571                         return PTR_ERR(trans);
5572                 ret = btrfs_commit_transaction(trans, root);
5573         }
5574         return ret;
5575 }
5576
5577 /*
5578  * This is somewhat expensive, updating the tree every time the
5579  * inode changes.  But, it is most likely to find the inode in cache.
5580  * FIXME, needs more benchmarking...there are no reasons other than performance
5581  * to keep or drop this code.
5582  */
5583 static int btrfs_dirty_inode(struct inode *inode)
5584 {
5585         struct btrfs_root *root = BTRFS_I(inode)->root;
5586         struct btrfs_trans_handle *trans;
5587         int ret;
5588
5589         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5590                 return 0;
5591
5592         trans = btrfs_join_transaction(root);
5593         if (IS_ERR(trans))
5594                 return PTR_ERR(trans);
5595
5596         ret = btrfs_update_inode(trans, root, inode);
5597         if (ret && ret == -ENOSPC) {
5598                 /* whoops, lets try again with the full transaction */
5599                 btrfs_end_transaction(trans, root);
5600                 trans = btrfs_start_transaction(root, 1);
5601                 if (IS_ERR(trans))
5602                         return PTR_ERR(trans);
5603
5604                 ret = btrfs_update_inode(trans, root, inode);
5605         }
5606         btrfs_end_transaction(trans, root);
5607         if (BTRFS_I(inode)->delayed_node)
5608                 btrfs_balance_delayed_items(root);
5609
5610         return ret;
5611 }
5612
5613 /*
5614  * This is a copy of file_update_time.  We need this so we can return error on
5615  * ENOSPC for updating the inode in the case of file write and mmap writes.
5616  */
5617 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5618                              int flags)
5619 {
5620         struct btrfs_root *root = BTRFS_I(inode)->root;
5621
5622         if (btrfs_root_readonly(root))
5623                 return -EROFS;
5624
5625         if (flags & S_VERSION)
5626                 inode_inc_iversion(inode);
5627         if (flags & S_CTIME)
5628                 inode->i_ctime = *now;
5629         if (flags & S_MTIME)
5630                 inode->i_mtime = *now;
5631         if (flags & S_ATIME)
5632                 inode->i_atime = *now;
5633         return btrfs_dirty_inode(inode);
5634 }
5635
5636 /*
5637  * find the highest existing sequence number in a directory
5638  * and then set the in-memory index_cnt variable to reflect
5639  * free sequence numbers
5640  */
5641 static int btrfs_set_inode_index_count(struct inode *inode)
5642 {
5643         struct btrfs_root *root = BTRFS_I(inode)->root;
5644         struct btrfs_key key, found_key;
5645         struct btrfs_path *path;
5646         struct extent_buffer *leaf;
5647         int ret;
5648
5649         key.objectid = btrfs_ino(inode);
5650         key.type = BTRFS_DIR_INDEX_KEY;
5651         key.offset = (u64)-1;
5652
5653         path = btrfs_alloc_path();
5654         if (!path)
5655                 return -ENOMEM;
5656
5657         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5658         if (ret < 0)
5659                 goto out;
5660         /* FIXME: we should be able to handle this */
5661         if (ret == 0)
5662                 goto out;
5663         ret = 0;
5664
5665         /*
5666          * MAGIC NUMBER EXPLANATION:
5667          * since we search a directory based on f_pos we have to start at 2
5668          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5669          * else has to start at 2
5670          */
5671         if (path->slots[0] == 0) {
5672                 BTRFS_I(inode)->index_cnt = 2;
5673                 goto out;
5674         }
5675
5676         path->slots[0]--;
5677
5678         leaf = path->nodes[0];
5679         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5680
5681         if (found_key.objectid != btrfs_ino(inode) ||
5682             found_key.type != BTRFS_DIR_INDEX_KEY) {
5683                 BTRFS_I(inode)->index_cnt = 2;
5684                 goto out;
5685         }
5686
5687         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5688 out:
5689         btrfs_free_path(path);
5690         return ret;
5691 }
5692
5693 /*
5694  * helper to find a free sequence number in a given directory.  This current
5695  * code is very simple, later versions will do smarter things in the btree
5696  */
5697 int btrfs_set_inode_index(struct inode *dir, u64 *index)
5698 {
5699         int ret = 0;
5700
5701         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
5702                 ret = btrfs_inode_delayed_dir_index_count(dir);
5703                 if (ret) {
5704                         ret = btrfs_set_inode_index_count(dir);
5705                         if (ret)
5706                                 return ret;
5707                 }
5708         }
5709
5710         *index = BTRFS_I(dir)->index_cnt;
5711         BTRFS_I(dir)->index_cnt++;
5712
5713         return ret;
5714 }
5715
5716 static int btrfs_insert_inode_locked(struct inode *inode)
5717 {
5718         struct btrfs_iget_args args;
5719         args.location = &BTRFS_I(inode)->location;
5720         args.root = BTRFS_I(inode)->root;
5721
5722         return insert_inode_locked4(inode,
5723                    btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
5724                    btrfs_find_actor, &args);
5725 }
5726
5727 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5728                                      struct btrfs_root *root,
5729                                      struct inode *dir,
5730                                      const char *name, int name_len,
5731                                      u64 ref_objectid, u64 objectid,
5732                                      umode_t mode, u64 *index)
5733 {
5734         struct inode *inode;
5735         struct btrfs_inode_item *inode_item;
5736         struct btrfs_key *location;
5737         struct btrfs_path *path;
5738         struct btrfs_inode_ref *ref;
5739         struct btrfs_key key[2];
5740         u32 sizes[2];
5741         int nitems = name ? 2 : 1;
5742         unsigned long ptr;
5743         int ret;
5744
5745         path = btrfs_alloc_path();
5746         if (!path)
5747                 return ERR_PTR(-ENOMEM);
5748
5749         inode = new_inode(root->fs_info->sb);
5750         if (!inode) {
5751                 btrfs_free_path(path);
5752                 return ERR_PTR(-ENOMEM);
5753         }
5754
5755         /*
5756          * O_TMPFILE, set link count to 0, so that after this point,
5757          * we fill in an inode item with the correct link count.
5758          */
5759         if (!name)
5760                 set_nlink(inode, 0);
5761
5762         /*
5763          * we have to initialize this early, so we can reclaim the inode
5764          * number if we fail afterwards in this function.
5765          */
5766         inode->i_ino = objectid;
5767
5768         if (dir && name) {
5769                 trace_btrfs_inode_request(dir);
5770
5771                 ret = btrfs_set_inode_index(dir, index);
5772                 if (ret) {
5773                         btrfs_free_path(path);
5774                         iput(inode);
5775                         return ERR_PTR(ret);
5776                 }
5777         } else if (dir) {
5778                 *index = 0;
5779         }
5780         /*
5781          * index_cnt is ignored for everything but a dir,
5782          * btrfs_get_inode_index_count has an explanation for the magic
5783          * number
5784          */
5785         BTRFS_I(inode)->index_cnt = 2;
5786         BTRFS_I(inode)->dir_index = *index;
5787         BTRFS_I(inode)->root = root;
5788         BTRFS_I(inode)->generation = trans->transid;
5789         inode->i_generation = BTRFS_I(inode)->generation;
5790
5791         /*
5792          * We could have gotten an inode number from somebody who was fsynced
5793          * and then removed in this same transaction, so let's just set full
5794          * sync since it will be a full sync anyway and this will blow away the
5795          * old info in the log.
5796          */
5797         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5798
5799         key[0].objectid = objectid;
5800         key[0].type = BTRFS_INODE_ITEM_KEY;
5801         key[0].offset = 0;
5802
5803         sizes[0] = sizeof(struct btrfs_inode_item);
5804
5805         if (name) {
5806                 /*
5807                  * Start new inodes with an inode_ref. This is slightly more
5808                  * efficient for small numbers of hard links since they will
5809                  * be packed into one item. Extended refs will kick in if we
5810                  * add more hard links than can fit in the ref item.
5811                  */
5812                 key[1].objectid = objectid;
5813                 key[1].type = BTRFS_INODE_REF_KEY;
5814                 key[1].offset = ref_objectid;
5815
5816                 sizes[1] = name_len + sizeof(*ref);
5817         }
5818
5819         location = &BTRFS_I(inode)->location;
5820         location->objectid = objectid;
5821         location->offset = 0;
5822         location->type = BTRFS_INODE_ITEM_KEY;
5823
5824         ret = btrfs_insert_inode_locked(inode);
5825         if (ret < 0)
5826                 goto fail;
5827
5828         path->leave_spinning = 1;
5829         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
5830         if (ret != 0)
5831                 goto fail_unlock;
5832
5833         inode_init_owner(inode, dir, mode);
5834         inode_set_bytes(inode, 0);
5835         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5836         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5837                                   struct btrfs_inode_item);
5838         memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
5839                              sizeof(*inode_item));
5840         fill_inode_item(trans, path->nodes[0], inode_item, inode);
5841
5842         if (name) {
5843                 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
5844                                      struct btrfs_inode_ref);
5845                 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
5846                 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
5847                 ptr = (unsigned long)(ref + 1);
5848                 write_extent_buffer(path->nodes[0], name, ptr, name_len);
5849         }
5850
5851         btrfs_mark_buffer_dirty(path->nodes[0]);
5852         btrfs_free_path(path);
5853
5854         btrfs_inherit_iflags(inode, dir);
5855
5856         if (S_ISREG(mode)) {
5857                 if (btrfs_test_opt(root, NODATASUM))
5858                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
5859                 if (btrfs_test_opt(root, NODATACOW))
5860                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
5861                                 BTRFS_INODE_NODATASUM;
5862         }
5863
5864         inode_tree_add(inode);
5865
5866         trace_btrfs_inode_new(inode);
5867         btrfs_set_inode_last_trans(trans, inode);
5868
5869         btrfs_update_root_times(trans, root);
5870
5871         ret = btrfs_inode_inherit_props(trans, inode, dir);
5872         if (ret)
5873                 btrfs_err(root->fs_info,
5874                           "error inheriting props for ino %llu (root %llu): %d",
5875                           btrfs_ino(inode), root->root_key.objectid, ret);
5876
5877         return inode;
5878
5879 fail_unlock:
5880         unlock_new_inode(inode);
5881 fail:
5882         if (dir && name)
5883                 BTRFS_I(dir)->index_cnt--;
5884         btrfs_free_path(path);
5885         iput(inode);
5886         return ERR_PTR(ret);
5887 }
5888
5889 static inline u8 btrfs_inode_type(struct inode *inode)
5890 {
5891         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
5892 }
5893
5894 /*
5895  * utility function to add 'inode' into 'parent_inode' with
5896  * a give name and a given sequence number.
5897  * if 'add_backref' is true, also insert a backref from the
5898  * inode to the parent directory.
5899  */
5900 int btrfs_add_link(struct btrfs_trans_handle *trans,
5901                    struct inode *parent_inode, struct inode *inode,
5902                    const char *name, int name_len, int add_backref, u64 index)
5903 {
5904         int ret = 0;
5905         struct btrfs_key key;
5906         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
5907         u64 ino = btrfs_ino(inode);
5908         u64 parent_ino = btrfs_ino(parent_inode);
5909
5910         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5911                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
5912         } else {
5913                 key.objectid = ino;
5914                 key.type = BTRFS_INODE_ITEM_KEY;
5915                 key.offset = 0;
5916         }
5917
5918         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5919                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
5920                                          key.objectid, root->root_key.objectid,
5921                                          parent_ino, index, name, name_len);
5922         } else if (add_backref) {
5923                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
5924                                              parent_ino, index);
5925         }
5926
5927         /* Nothing to clean up yet */
5928         if (ret)
5929                 return ret;
5930
5931         ret = btrfs_insert_dir_item(trans, root, name, name_len,
5932                                     parent_inode, &key,
5933                                     btrfs_inode_type(inode), index);
5934         if (ret == -EEXIST || ret == -EOVERFLOW)
5935                 goto fail_dir_item;
5936         else if (ret) {
5937                 btrfs_abort_transaction(trans, root, ret);
5938                 return ret;
5939         }
5940
5941         btrfs_i_size_write(parent_inode, parent_inode->i_size +
5942                            name_len * 2);
5943         inode_inc_iversion(parent_inode);
5944         parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
5945         ret = btrfs_update_inode(trans, root, parent_inode);
5946         if (ret)
5947                 btrfs_abort_transaction(trans, root, ret);
5948         return ret;
5949
5950 fail_dir_item:
5951         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5952                 u64 local_index;
5953                 int err;
5954                 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
5955                                  key.objectid, root->root_key.objectid,
5956                                  parent_ino, &local_index, name, name_len);
5957
5958         } else if (add_backref) {
5959                 u64 local_index;
5960                 int err;
5961
5962                 err = btrfs_del_inode_ref(trans, root, name, name_len,
5963                                           ino, parent_ino, &local_index);
5964         }
5965         return ret;
5966 }
5967
5968 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
5969                             struct inode *dir, struct dentry *dentry,
5970                             struct inode *inode, int backref, u64 index)
5971 {
5972         int err = btrfs_add_link(trans, dir, inode,
5973                                  dentry->d_name.name, dentry->d_name.len,
5974                                  backref, index);
5975         if (err > 0)
5976                 err = -EEXIST;
5977         return err;
5978 }
5979
5980 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
5981                         umode_t mode, dev_t rdev)
5982 {
5983         struct btrfs_trans_handle *trans;
5984         struct btrfs_root *root = BTRFS_I(dir)->root;
5985         struct inode *inode = NULL;
5986         int err;
5987         int drop_inode = 0;
5988         u64 objectid;
5989         u64 index = 0;
5990
5991         if (!new_valid_dev(rdev))
5992                 return -EINVAL;
5993
5994         /*
5995          * 2 for inode item and ref
5996          * 2 for dir items
5997          * 1 for xattr if selinux is on
5998          */
5999         trans = btrfs_start_transaction(root, 5);
6000         if (IS_ERR(trans))
6001                 return PTR_ERR(trans);
6002
6003         err = btrfs_find_free_ino(root, &objectid);
6004         if (err)
6005                 goto out_unlock;
6006
6007         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6008                                 dentry->d_name.len, btrfs_ino(dir), objectid,
6009                                 mode, &index);
6010         if (IS_ERR(inode)) {
6011                 err = PTR_ERR(inode);
6012                 goto out_unlock;
6013         }
6014
6015         /*
6016         * If the active LSM wants to access the inode during
6017         * d_instantiate it needs these. Smack checks to see
6018         * if the filesystem supports xattrs by looking at the
6019         * ops vector.
6020         */
6021         inode->i_op = &btrfs_special_inode_operations;
6022         init_special_inode(inode, inode->i_mode, rdev);
6023
6024         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6025         if (err)
6026                 goto out_unlock_inode;
6027
6028         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6029         if (err) {
6030                 goto out_unlock_inode;
6031         } else {
6032                 btrfs_update_inode(trans, root, inode);
6033                 unlock_new_inode(inode);
6034                 d_instantiate(dentry, inode);
6035         }
6036
6037 out_unlock:
6038         btrfs_end_transaction(trans, root);
6039         btrfs_balance_delayed_items(root);
6040         btrfs_btree_balance_dirty(root);
6041         if (drop_inode) {
6042                 inode_dec_link_count(inode);
6043                 iput(inode);
6044         }
6045         return err;
6046
6047 out_unlock_inode:
6048         drop_inode = 1;
6049         unlock_new_inode(inode);
6050         goto out_unlock;
6051
6052 }
6053
6054 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6055                         umode_t mode, bool excl)
6056 {
6057         struct btrfs_trans_handle *trans;
6058         struct btrfs_root *root = BTRFS_I(dir)->root;
6059         struct inode *inode = NULL;
6060         int drop_inode_on_err = 0;
6061         int err;
6062         u64 objectid;
6063         u64 index = 0;
6064
6065         /*
6066          * 2 for inode item and ref
6067          * 2 for dir items
6068          * 1 for xattr if selinux is on
6069          */
6070         trans = btrfs_start_transaction(root, 5);
6071         if (IS_ERR(trans))
6072                 return PTR_ERR(trans);
6073
6074         err = btrfs_find_free_ino(root, &objectid);
6075         if (err)
6076                 goto out_unlock;
6077
6078         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6079                                 dentry->d_name.len, btrfs_ino(dir), objectid,
6080                                 mode, &index);
6081         if (IS_ERR(inode)) {
6082                 err = PTR_ERR(inode);
6083                 goto out_unlock;
6084         }
6085         drop_inode_on_err = 1;
6086         /*
6087         * If the active LSM wants to access the inode during
6088         * d_instantiate it needs these. Smack checks to see
6089         * if the filesystem supports xattrs by looking at the
6090         * ops vector.
6091         */
6092         inode->i_fop = &btrfs_file_operations;
6093         inode->i_op = &btrfs_file_inode_operations;
6094         inode->i_mapping->a_ops = &btrfs_aops;
6095         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
6096
6097         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6098         if (err)
6099                 goto out_unlock_inode;
6100
6101         err = btrfs_update_inode(trans, root, inode);
6102         if (err)
6103                 goto out_unlock_inode;
6104
6105         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6106         if (err)
6107                 goto out_unlock_inode;
6108
6109         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6110         unlock_new_inode(inode);
6111         d_instantiate(dentry, inode);
6112
6113 out_unlock:
6114         btrfs_end_transaction(trans, root);
6115         if (err && drop_inode_on_err) {
6116                 inode_dec_link_count(inode);
6117                 iput(inode);
6118         }
6119         btrfs_balance_delayed_items(root);
6120         btrfs_btree_balance_dirty(root);
6121         return err;
6122
6123 out_unlock_inode:
6124         unlock_new_inode(inode);
6125         goto out_unlock;
6126
6127 }
6128
6129 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6130                       struct dentry *dentry)
6131 {
6132         struct btrfs_trans_handle *trans;
6133         struct btrfs_root *root = BTRFS_I(dir)->root;
6134         struct inode *inode = old_dentry->d_inode;
6135         u64 index;
6136         int err;
6137         int drop_inode = 0;
6138
6139         /* do not allow sys_link's with other subvols of the same device */
6140         if (root->objectid != BTRFS_I(inode)->root->objectid)
6141                 return -EXDEV;
6142
6143         if (inode->i_nlink >= BTRFS_LINK_MAX)
6144                 return -EMLINK;
6145
6146         err = btrfs_set_inode_index(dir, &index);
6147         if (err)
6148                 goto fail;
6149
6150         /*
6151          * 2 items for inode and inode ref
6152          * 2 items for dir items
6153          * 1 item for parent inode
6154          */
6155         trans = btrfs_start_transaction(root, 5);
6156         if (IS_ERR(trans)) {
6157                 err = PTR_ERR(trans);
6158                 goto fail;
6159         }
6160
6161         /* There are several dir indexes for this inode, clear the cache. */
6162         BTRFS_I(inode)->dir_index = 0ULL;
6163         inc_nlink(inode);
6164         inode_inc_iversion(inode);
6165         inode->i_ctime = CURRENT_TIME;
6166         ihold(inode);
6167         set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6168
6169         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
6170
6171         if (err) {
6172                 drop_inode = 1;
6173         } else {
6174                 struct dentry *parent = dentry->d_parent;
6175                 err = btrfs_update_inode(trans, root, inode);
6176                 if (err)
6177                         goto fail;
6178                 if (inode->i_nlink == 1) {
6179                         /*
6180                          * If new hard link count is 1, it's a file created
6181                          * with open(2) O_TMPFILE flag.
6182                          */
6183                         err = btrfs_orphan_del(trans, inode);
6184                         if (err)
6185                                 goto fail;
6186                 }
6187                 d_instantiate(dentry, inode);
6188                 btrfs_log_new_name(trans, inode, NULL, parent);
6189         }
6190
6191         btrfs_end_transaction(trans, root);
6192         btrfs_balance_delayed_items(root);
6193 fail:
6194         if (drop_inode) {
6195                 inode_dec_link_count(inode);
6196                 iput(inode);
6197         }
6198         btrfs_btree_balance_dirty(root);
6199         return err;
6200 }
6201
6202 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6203 {
6204         struct inode *inode = NULL;
6205         struct btrfs_trans_handle *trans;
6206         struct btrfs_root *root = BTRFS_I(dir)->root;
6207         int err = 0;
6208         int drop_on_err = 0;
6209         u64 objectid = 0;
6210         u64 index = 0;
6211
6212         /*
6213          * 2 items for inode and ref
6214          * 2 items for dir items
6215          * 1 for xattr if selinux is on
6216          */
6217         trans = btrfs_start_transaction(root, 5);
6218         if (IS_ERR(trans))
6219                 return PTR_ERR(trans);
6220
6221         err = btrfs_find_free_ino(root, &objectid);
6222         if (err)
6223                 goto out_fail;
6224
6225         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6226                                 dentry->d_name.len, btrfs_ino(dir), objectid,
6227                                 S_IFDIR | mode, &index);
6228         if (IS_ERR(inode)) {
6229                 err = PTR_ERR(inode);
6230                 goto out_fail;
6231         }
6232
6233         drop_on_err = 1;
6234         /* these must be set before we unlock the inode */
6235         inode->i_op = &btrfs_dir_inode_operations;
6236         inode->i_fop = &btrfs_dir_file_operations;
6237
6238         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6239         if (err)
6240                 goto out_fail_inode;
6241
6242         btrfs_i_size_write(inode, 0);
6243         err = btrfs_update_inode(trans, root, inode);
6244         if (err)
6245                 goto out_fail_inode;
6246
6247         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
6248                              dentry->d_name.len, 0, index);
6249         if (err)
6250                 goto out_fail_inode;
6251
6252         d_instantiate(dentry, inode);
6253         /*
6254          * mkdir is special.  We're unlocking after we call d_instantiate
6255          * to avoid a race with nfsd calling d_instantiate.
6256          */
6257         unlock_new_inode(inode);
6258         drop_on_err = 0;
6259
6260 out_fail:
6261         btrfs_end_transaction(trans, root);
6262         if (drop_on_err) {
6263                 inode_dec_link_count(inode);
6264                 iput(inode);
6265         }
6266         btrfs_balance_delayed_items(root);
6267         btrfs_btree_balance_dirty(root);
6268         return err;
6269
6270 out_fail_inode:
6271         unlock_new_inode(inode);
6272         goto out_fail;
6273 }
6274
6275 /* Find next extent map of a given extent map, caller needs to ensure locks */
6276 static struct extent_map *next_extent_map(struct extent_map *em)
6277 {
6278         struct rb_node *next;
6279
6280         next = rb_next(&em->rb_node);
6281         if (!next)
6282                 return NULL;
6283         return container_of(next, struct extent_map, rb_node);
6284 }
6285
6286 static struct extent_map *prev_extent_map(struct extent_map *em)
6287 {
6288         struct rb_node *prev;
6289
6290         prev = rb_prev(&em->rb_node);
6291         if (!prev)
6292                 return NULL;
6293         return container_of(prev, struct extent_map, rb_node);
6294 }
6295
6296 /* helper for btfs_get_extent.  Given an existing extent in the tree,
6297  * the existing extent is the nearest extent to map_start,
6298  * and an extent that you want to insert, deal with overlap and insert
6299  * the best fitted new extent into the tree.
6300  */
6301 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6302                                 struct extent_map *existing,
6303                                 struct extent_map *em,
6304                                 u64 map_start)
6305 {
6306         struct extent_map *prev;
6307         struct extent_map *next;
6308         u64 start;
6309         u64 end;
6310         u64 start_diff;
6311
6312         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6313
6314         if (existing->start > map_start) {
6315                 next = existing;
6316                 prev = prev_extent_map(next);
6317         } else {
6318                 prev = existing;
6319                 next = next_extent_map(prev);
6320         }
6321
6322         start = prev ? extent_map_end(prev) : em->start;
6323         start = max_t(u64, start, em->start);
6324         end = next ? next->start : extent_map_end(em);
6325         end = min_t(u64, end, extent_map_end(em));
6326         start_diff = start - em->start;
6327         em->start = start;
6328         em->len = end - start;
6329         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6330             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6331                 em->block_start += start_diff;
6332                 em->block_len -= start_diff;
6333         }
6334         return add_extent_mapping(em_tree, em, 0);
6335 }
6336
6337 static noinline int uncompress_inline(struct btrfs_path *path,
6338                                       struct inode *inode, struct page *page,
6339                                       size_t pg_offset, u64 extent_offset,
6340                                       struct btrfs_file_extent_item *item)
6341 {
6342         int ret;
6343         struct extent_buffer *leaf = path->nodes[0];
6344         char *tmp;
6345         size_t max_size;
6346         unsigned long inline_size;
6347         unsigned long ptr;
6348         int compress_type;
6349
6350         WARN_ON(pg_offset != 0);
6351         compress_type = btrfs_file_extent_compression(leaf, item);
6352         max_size = btrfs_file_extent_ram_bytes(leaf, item);
6353         inline_size = btrfs_file_extent_inline_item_len(leaf,
6354                                         btrfs_item_nr(path->slots[0]));
6355         tmp = kmalloc(inline_size, GFP_NOFS);
6356         if (!tmp)
6357                 return -ENOMEM;
6358         ptr = btrfs_file_extent_inline_start(item);
6359
6360         read_extent_buffer(leaf, tmp, ptr, inline_size);
6361
6362         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
6363         ret = btrfs_decompress(compress_type, tmp, page,
6364                                extent_offset, inline_size, max_size);
6365         kfree(tmp);
6366         return ret;
6367 }
6368
6369 /*
6370  * a bit scary, this does extent mapping from logical file offset to the disk.
6371  * the ugly parts come from merging extents from the disk with the in-ram
6372  * representation.  This gets more complex because of the data=ordered code,
6373  * where the in-ram extents might be locked pending data=ordered completion.
6374  *
6375  * This also copies inline extents directly into the page.
6376  */
6377
6378 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6379                                     size_t pg_offset, u64 start, u64 len,
6380                                     int create)
6381 {
6382         int ret;
6383         int err = 0;
6384         u64 extent_start = 0;
6385         u64 extent_end = 0;
6386         u64 objectid = btrfs_ino(inode);
6387         u32 found_type;
6388         struct btrfs_path *path = NULL;
6389         struct btrfs_root *root = BTRFS_I(inode)->root;
6390         struct btrfs_file_extent_item *item;
6391         struct extent_buffer *leaf;
6392         struct btrfs_key found_key;
6393         struct extent_map *em = NULL;
6394         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6395         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6396         struct btrfs_trans_handle *trans = NULL;
6397         const bool new_inline = !page || create;
6398
6399 again:
6400         read_lock(&em_tree->lock);
6401         em = lookup_extent_mapping(em_tree, start, len);
6402         if (em)
6403                 em->bdev = root->fs_info->fs_devices->latest_bdev;
6404         read_unlock(&em_tree->lock);
6405
6406         if (em) {
6407                 if (em->start > start || em->start + em->len <= start)
6408                         free_extent_map(em);
6409                 else if (em->block_start == EXTENT_MAP_INLINE && page)
6410                         free_extent_map(em);
6411                 else
6412                         goto out;
6413         }
6414         em = alloc_extent_map();
6415         if (!em) {
6416                 err = -ENOMEM;
6417                 goto out;
6418         }
6419         em->bdev = root->fs_info->fs_devices->latest_bdev;
6420         em->start = EXTENT_MAP_HOLE;
6421         em->orig_start = EXTENT_MAP_HOLE;
6422         em->len = (u64)-1;
6423         em->block_len = (u64)-1;
6424
6425         if (!path) {
6426                 path = btrfs_alloc_path();
6427                 if (!path) {
6428                         err = -ENOMEM;
6429                         goto out;
6430                 }
6431                 /*
6432                  * Chances are we'll be called again, so go ahead and do
6433                  * readahead
6434                  */
6435                 path->reada = 1;
6436         }
6437
6438         ret = btrfs_lookup_file_extent(trans, root, path,
6439                                        objectid, start, trans != NULL);
6440         if (ret < 0) {
6441                 err = ret;
6442                 goto out;
6443         }
6444
6445         if (ret != 0) {
6446                 if (path->slots[0] == 0)
6447                         goto not_found;
6448                 path->slots[0]--;
6449         }
6450
6451         leaf = path->nodes[0];
6452         item = btrfs_item_ptr(leaf, path->slots[0],
6453                               struct btrfs_file_extent_item);
6454         /* are we inside the extent that was found? */
6455         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6456         found_type = found_key.type;
6457         if (found_key.objectid != objectid ||
6458             found_type != BTRFS_EXTENT_DATA_KEY) {
6459                 /*
6460                  * If we backup past the first extent we want to move forward
6461                  * and see if there is an extent in front of us, otherwise we'll
6462                  * say there is a hole for our whole search range which can
6463                  * cause problems.
6464                  */
6465                 extent_end = start;
6466                 goto next;
6467         }
6468
6469         found_type = btrfs_file_extent_type(leaf, item);
6470         extent_start = found_key.offset;
6471         if (found_type == BTRFS_FILE_EXTENT_REG ||
6472             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6473                 extent_end = extent_start +
6474                        btrfs_file_extent_num_bytes(leaf, item);
6475         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6476                 size_t size;
6477                 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6478                 extent_end = ALIGN(extent_start + size, root->sectorsize);
6479         }
6480 next:
6481         if (start >= extent_end) {
6482                 path->slots[0]++;
6483                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6484                         ret = btrfs_next_leaf(root, path);
6485                         if (ret < 0) {
6486                                 err = ret;
6487                                 goto out;
6488                         }
6489                         if (ret > 0)
6490                                 goto not_found;
6491                         leaf = path->nodes[0];
6492                 }
6493                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6494                 if (found_key.objectid != objectid ||
6495                     found_key.type != BTRFS_EXTENT_DATA_KEY)
6496                         goto not_found;
6497                 if (start + len <= found_key.offset)
6498                         goto not_found;
6499                 if (start > found_key.offset)
6500                         goto next;
6501                 em->start = start;
6502                 em->orig_start = start;
6503                 em->len = found_key.offset - start;
6504                 goto not_found_em;
6505         }
6506
6507         btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em);
6508
6509         if (found_type == BTRFS_FILE_EXTENT_REG ||
6510             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6511                 goto insert;
6512         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6513                 unsigned long ptr;
6514                 char *map;
6515                 size_t size;
6516                 size_t extent_offset;
6517                 size_t copy_size;
6518
6519                 if (new_inline)
6520                         goto out;
6521
6522                 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6523                 extent_offset = page_offset(page) + pg_offset - extent_start;
6524                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6525                                 size - extent_offset);
6526                 em->start = extent_start + extent_offset;
6527                 em->len = ALIGN(copy_size, root->sectorsize);
6528                 em->orig_block_len = em->len;
6529                 em->orig_start = em->start;
6530                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6531                 if (create == 0 && !PageUptodate(page)) {
6532                         if (btrfs_file_extent_compression(leaf, item) !=
6533                             BTRFS_COMPRESS_NONE) {
6534                                 ret = uncompress_inline(path, inode, page,
6535                                                         pg_offset,
6536                                                         extent_offset, item);
6537                                 if (ret) {
6538                                         err = ret;
6539                                         goto out;
6540                                 }
6541                         } else {
6542                                 map = kmap(page);
6543                                 read_extent_buffer(leaf, map + pg_offset, ptr,
6544                                                    copy_size);
6545                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6546                                         memset(map + pg_offset + copy_size, 0,
6547                                                PAGE_CACHE_SIZE - pg_offset -
6548                                                copy_size);
6549                                 }
6550                                 kunmap(page);
6551                         }
6552                         flush_dcache_page(page);
6553                 } else if (create && PageUptodate(page)) {
6554                         BUG();
6555                         if (!trans) {
6556                                 kunmap(page);
6557                                 free_extent_map(em);
6558                                 em = NULL;
6559
6560                                 btrfs_release_path(path);
6561                                 trans = btrfs_join_transaction(root);
6562
6563                                 if (IS_ERR(trans))
6564                                         return ERR_CAST(trans);
6565                                 goto again;
6566                         }
6567                         map = kmap(page);
6568                         write_extent_buffer(leaf, map + pg_offset, ptr,
6569                                             copy_size);
6570                         kunmap(page);
6571                         btrfs_mark_buffer_dirty(leaf);
6572                 }
6573                 set_extent_uptodate(io_tree, em->start,
6574                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
6575                 goto insert;
6576         }
6577 not_found:
6578         em->start = start;
6579         em->orig_start = start;
6580         em->len = len;
6581 not_found_em:
6582         em->block_start = EXTENT_MAP_HOLE;
6583         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6584 insert:
6585         btrfs_release_path(path);
6586         if (em->start > start || extent_map_end(em) <= start) {
6587                 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6588                         em->start, em->len, start, len);
6589                 err = -EIO;
6590                 goto out;
6591         }
6592
6593         err = 0;
6594         write_lock(&em_tree->lock);
6595         ret = add_extent_mapping(em_tree, em, 0);
6596         /* it is possible that someone inserted the extent into the tree
6597          * while we had the lock dropped.  It is also possible that
6598          * an overlapping map exists in the tree
6599          */
6600         if (ret == -EEXIST) {
6601                 struct extent_map *existing;
6602
6603                 ret = 0;
6604
6605                 existing = search_extent_mapping(em_tree, start, len);
6606                 /*
6607                  * existing will always be non-NULL, since there must be
6608                  * extent causing the -EEXIST.
6609                  */
6610                 if (start >= extent_map_end(existing) ||
6611                     start <= existing->start) {
6612                         /*
6613                          * The existing extent map is the one nearest to
6614                          * the [start, start + len) range which overlaps
6615                          */
6616                         err = merge_extent_mapping(em_tree, existing,
6617                                                    em, start);
6618                         free_extent_map(existing);
6619                         if (err) {
6620                                 free_extent_map(em);
6621                                 em = NULL;
6622                         }
6623                 } else {
6624                         free_extent_map(em);
6625                         em = existing;
6626                         err = 0;
6627                 }
6628         }
6629         write_unlock(&em_tree->lock);
6630 out:
6631
6632         trace_btrfs_get_extent(root, em);
6633
6634         if (path)
6635                 btrfs_free_path(path);
6636         if (trans) {
6637                 ret = btrfs_end_transaction(trans, root);
6638                 if (!err)
6639                         err = ret;
6640         }
6641         if (err) {
6642                 free_extent_map(em);
6643                 return ERR_PTR(err);
6644         }
6645         BUG_ON(!em); /* Error is always set */
6646         return em;
6647 }
6648
6649 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6650                                            size_t pg_offset, u64 start, u64 len,
6651                                            int create)
6652 {
6653         struct extent_map *em;
6654         struct extent_map *hole_em = NULL;
6655         u64 range_start = start;
6656         u64 end;
6657         u64 found;
6658         u64 found_end;
6659         int err = 0;
6660
6661         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6662         if (IS_ERR(em))
6663                 return em;
6664         if (em) {
6665                 /*
6666                  * if our em maps to
6667                  * -  a hole or
6668                  * -  a pre-alloc extent,
6669                  * there might actually be delalloc bytes behind it.
6670                  */
6671                 if (em->block_start != EXTENT_MAP_HOLE &&
6672                     !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6673                         return em;
6674                 else
6675                         hole_em = em;
6676         }
6677
6678         /* check to see if we've wrapped (len == -1 or similar) */
6679         end = start + len;
6680         if (end < start)
6681                 end = (u64)-1;
6682         else
6683                 end -= 1;
6684
6685         em = NULL;
6686
6687         /* ok, we didn't find anything, lets look for delalloc */
6688         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6689                                  end, len, EXTENT_DELALLOC, 1);
6690         found_end = range_start + found;
6691         if (found_end < range_start)
6692                 found_end = (u64)-1;
6693
6694         /*
6695          * we didn't find anything useful, return
6696          * the original results from get_extent()
6697          */
6698         if (range_start > end || found_end <= start) {
6699                 em = hole_em;
6700                 hole_em = NULL;
6701                 goto out;
6702         }
6703
6704         /* adjust the range_start to make sure it doesn't
6705          * go backwards from the start they passed in
6706          */
6707         range_start = max(start, range_start);
6708         found = found_end - range_start;
6709
6710         if (found > 0) {
6711                 u64 hole_start = start;
6712                 u64 hole_len = len;
6713
6714                 em = alloc_extent_map();
6715                 if (!em) {
6716                         err = -ENOMEM;
6717                         goto out;
6718                 }
6719                 /*
6720                  * when btrfs_get_extent can't find anything it
6721                  * returns one huge hole
6722                  *
6723                  * make sure what it found really fits our range, and
6724                  * adjust to make sure it is based on the start from
6725                  * the caller
6726                  */
6727                 if (hole_em) {
6728                         u64 calc_end = extent_map_end(hole_em);
6729
6730                         if (calc_end <= start || (hole_em->start > end)) {
6731                                 free_extent_map(hole_em);
6732                                 hole_em = NULL;
6733                         } else {
6734                                 hole_start = max(hole_em->start, start);
6735                                 hole_len = calc_end - hole_start;
6736                         }
6737                 }
6738                 em->bdev = NULL;
6739                 if (hole_em && range_start > hole_start) {
6740                         /* our hole starts before our delalloc, so we
6741                          * have to return just the parts of the hole
6742                          * that go until  the delalloc starts
6743                          */
6744                         em->len = min(hole_len,
6745                                       range_start - hole_start);
6746                         em->start = hole_start;
6747                         em->orig_start = hole_start;
6748                         /*
6749                          * don't adjust block start at all,
6750                          * it is fixed at EXTENT_MAP_HOLE
6751                          */
6752                         em->block_start = hole_em->block_start;
6753                         em->block_len = hole_len;
6754                         if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6755                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6756                 } else {
6757                         em->start = range_start;
6758                         em->len = found;
6759                         em->orig_start = range_start;
6760                         em->block_start = EXTENT_MAP_DELALLOC;
6761                         em->block_len = found;
6762                 }
6763         } else if (hole_em) {
6764                 return hole_em;
6765         }
6766 out:
6767
6768         free_extent_map(hole_em);
6769         if (err) {
6770                 free_extent_map(em);
6771                 return ERR_PTR(err);
6772         }
6773         return em;
6774 }
6775
6776 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6777                                                   u64 start, u64 len)
6778 {
6779         struct btrfs_root *root = BTRFS_I(inode)->root;
6780         struct extent_map *em;
6781         struct btrfs_key ins;
6782         u64 alloc_hint;
6783         int ret;
6784
6785         alloc_hint = get_extent_allocation_hint(inode, start, len);
6786         ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
6787                                    alloc_hint, &ins, 1, 1);
6788         if (ret)
6789                 return ERR_PTR(ret);
6790
6791         em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
6792                               ins.offset, ins.offset, ins.offset, 0);
6793         if (IS_ERR(em)) {
6794                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
6795                 return em;
6796         }
6797
6798         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
6799                                            ins.offset, ins.offset, 0);
6800         if (ret) {
6801                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
6802                 free_extent_map(em);
6803                 return ERR_PTR(ret);
6804         }
6805
6806         return em;
6807 }
6808
6809 /*
6810  * returns 1 when the nocow is safe, < 1 on error, 0 if the
6811  * block must be cow'd
6812  */
6813 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
6814                               u64 *orig_start, u64 *orig_block_len,
6815                               u64 *ram_bytes)
6816 {
6817         struct btrfs_trans_handle *trans;
6818         struct btrfs_path *path;
6819         int ret;
6820         struct extent_buffer *leaf;
6821         struct btrfs_root *root = BTRFS_I(inode)->root;
6822         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6823         struct btrfs_file_extent_item *fi;
6824         struct btrfs_key key;
6825         u64 disk_bytenr;
6826         u64 backref_offset;
6827         u64 extent_end;
6828         u64 num_bytes;
6829         int slot;
6830         int found_type;
6831         bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
6832
6833         path = btrfs_alloc_path();
6834         if (!path)
6835                 return -ENOMEM;
6836
6837         ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
6838                                        offset, 0);
6839         if (ret < 0)
6840                 goto out;
6841
6842         slot = path->slots[0];
6843         if (ret == 1) {
6844                 if (slot == 0) {
6845                         /* can't find the item, must cow */
6846                         ret = 0;
6847                         goto out;
6848                 }
6849                 slot--;
6850         }
6851         ret = 0;
6852         leaf = path->nodes[0];
6853         btrfs_item_key_to_cpu(leaf, &key, slot);
6854         if (key.objectid != btrfs_ino(inode) ||
6855             key.type != BTRFS_EXTENT_DATA_KEY) {
6856                 /* not our file or wrong item type, must cow */
6857                 goto out;
6858         }
6859
6860         if (key.offset > offset) {
6861                 /* Wrong offset, must cow */
6862                 goto out;
6863         }
6864
6865         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6866         found_type = btrfs_file_extent_type(leaf, fi);
6867         if (found_type != BTRFS_FILE_EXTENT_REG &&
6868             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6869                 /* not a regular extent, must cow */
6870                 goto out;
6871         }
6872
6873         if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
6874                 goto out;
6875
6876         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
6877         if (extent_end <= offset)
6878                 goto out;
6879
6880         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6881         if (disk_bytenr == 0)
6882                 goto out;
6883
6884         if (btrfs_file_extent_compression(leaf, fi) ||
6885             btrfs_file_extent_encryption(leaf, fi) ||
6886             btrfs_file_extent_other_encoding(leaf, fi))
6887                 goto out;
6888
6889         backref_offset = btrfs_file_extent_offset(leaf, fi);
6890
6891         if (orig_start) {
6892                 *orig_start = key.offset - backref_offset;
6893                 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
6894                 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6895         }
6896
6897         if (btrfs_extent_readonly(root, disk_bytenr))
6898                 goto out;
6899
6900         num_bytes = min(offset + *len, extent_end) - offset;
6901         if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6902                 u64 range_end;
6903
6904                 range_end = round_up(offset + num_bytes, root->sectorsize) - 1;
6905                 ret = test_range_bit(io_tree, offset, range_end,
6906                                      EXTENT_DELALLOC, 0, NULL);
6907                 if (ret) {
6908                         ret = -EAGAIN;
6909                         goto out;
6910                 }
6911         }
6912
6913         btrfs_release_path(path);
6914
6915         /*
6916          * look for other files referencing this extent, if we
6917          * find any we must cow
6918          */
6919         trans = btrfs_join_transaction(root);
6920         if (IS_ERR(trans)) {
6921                 ret = 0;
6922                 goto out;
6923         }
6924
6925         ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
6926                                     key.offset - backref_offset, disk_bytenr);
6927         btrfs_end_transaction(trans, root);
6928         if (ret) {
6929                 ret = 0;
6930                 goto out;
6931         }
6932
6933         /*
6934          * adjust disk_bytenr and num_bytes to cover just the bytes
6935          * in this extent we are about to write.  If there
6936          * are any csums in that range we have to cow in order
6937          * to keep the csums correct
6938          */
6939         disk_bytenr += backref_offset;
6940         disk_bytenr += offset - key.offset;
6941         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
6942                                 goto out;
6943         /*
6944          * all of the above have passed, it is safe to overwrite this extent
6945          * without cow
6946          */
6947         *len = num_bytes;
6948         ret = 1;
6949 out:
6950         btrfs_free_path(path);
6951         return ret;
6952 }
6953
6954 bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
6955 {
6956         struct radix_tree_root *root = &inode->i_mapping->page_tree;
6957         int found = false;
6958         void **pagep = NULL;
6959         struct page *page = NULL;
6960         int start_idx;
6961         int end_idx;
6962
6963         start_idx = start >> PAGE_CACHE_SHIFT;
6964
6965         /*
6966          * end is the last byte in the last page.  end == start is legal
6967          */
6968         end_idx = end >> PAGE_CACHE_SHIFT;
6969
6970         rcu_read_lock();
6971
6972         /* Most of the code in this while loop is lifted from
6973          * find_get_page.  It's been modified to begin searching from a
6974          * page and return just the first page found in that range.  If the
6975          * found idx is less than or equal to the end idx then we know that
6976          * a page exists.  If no pages are found or if those pages are
6977          * outside of the range then we're fine (yay!) */
6978         while (page == NULL &&
6979                radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
6980                 page = radix_tree_deref_slot(pagep);
6981                 if (unlikely(!page))
6982                         break;
6983
6984                 if (radix_tree_exception(page)) {
6985                         if (radix_tree_deref_retry(page)) {
6986                                 page = NULL;
6987                                 continue;
6988                         }
6989                         /*
6990                          * Otherwise, shmem/tmpfs must be storing a swap entry
6991                          * here as an exceptional entry: so return it without
6992                          * attempting to raise page count.
6993                          */
6994                         page = NULL;
6995                         break; /* TODO: Is this relevant for this use case? */
6996                 }
6997
6998                 if (!page_cache_get_speculative(page)) {
6999                         page = NULL;
7000                         continue;
7001                 }
7002
7003                 /*
7004                  * Has the page moved?
7005                  * This is part of the lockless pagecache protocol. See
7006                  * include/linux/pagemap.h for details.
7007                  */
7008                 if (unlikely(page != *pagep)) {
7009                         page_cache_release(page);
7010                         page = NULL;
7011                 }
7012         }
7013
7014         if (page) {
7015                 if (page->index <= end_idx)
7016                         found = true;
7017                 page_cache_release(page);
7018         }
7019
7020         rcu_read_unlock();
7021         return found;
7022 }
7023
7024 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7025                               struct extent_state **cached_state, int writing)
7026 {
7027         struct btrfs_ordered_extent *ordered;
7028         int ret = 0;
7029
7030         while (1) {
7031                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7032                                  0, cached_state);
7033                 /*
7034                  * We're concerned with the entire range that we're going to be
7035                  * doing DIO to, so we need to make sure theres no ordered
7036                  * extents in this range.
7037                  */
7038                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
7039                                                      lockend - lockstart + 1);
7040
7041                 /*
7042                  * We need to make sure there are no buffered pages in this
7043                  * range either, we could have raced between the invalidate in
7044                  * generic_file_direct_write and locking the extent.  The
7045                  * invalidate needs to happen so that reads after a write do not
7046                  * get stale data.
7047                  */
7048                 if (!ordered &&
7049                     (!writing ||
7050                      !btrfs_page_exists_in_range(inode, lockstart, lockend)))
7051                         break;
7052
7053                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7054                                      cached_state, GFP_NOFS);
7055
7056                 if (ordered) {
7057                         btrfs_start_ordered_extent(inode, ordered, 1);
7058                         btrfs_put_ordered_extent(ordered);
7059                 } else {
7060                         /* Screw you mmap */
7061                         ret = btrfs_fdatawrite_range(inode, lockstart, lockend);
7062                         if (ret)
7063                                 break;
7064                         ret = filemap_fdatawait_range(inode->i_mapping,
7065                                                       lockstart,
7066                                                       lockend);
7067                         if (ret)
7068                                 break;
7069
7070                         /*
7071                          * If we found a page that couldn't be invalidated just
7072                          * fall back to buffered.
7073                          */
7074                         ret = invalidate_inode_pages2_range(inode->i_mapping,
7075                                         lockstart >> PAGE_CACHE_SHIFT,
7076                                         lockend >> PAGE_CACHE_SHIFT);
7077                         if (ret)
7078                                 break;
7079                 }
7080
7081                 cond_resched();
7082         }
7083
7084         return ret;
7085 }
7086
7087 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
7088                                            u64 len, u64 orig_start,
7089                                            u64 block_start, u64 block_len,
7090                                            u64 orig_block_len, u64 ram_bytes,
7091                                            int type)
7092 {
7093         struct extent_map_tree *em_tree;
7094         struct extent_map *em;
7095         struct btrfs_root *root = BTRFS_I(inode)->root;
7096         int ret;
7097
7098         em_tree = &BTRFS_I(inode)->extent_tree;
7099         em = alloc_extent_map();
7100         if (!em)
7101                 return ERR_PTR(-ENOMEM);
7102
7103         em->start = start;
7104         em->orig_start = orig_start;
7105         em->mod_start = start;
7106         em->mod_len = len;
7107         em->len = len;
7108         em->block_len = block_len;
7109         em->block_start = block_start;
7110         em->bdev = root->fs_info->fs_devices->latest_bdev;
7111         em->orig_block_len = orig_block_len;
7112         em->ram_bytes = ram_bytes;
7113         em->generation = -1;
7114         set_bit(EXTENT_FLAG_PINNED, &em->flags);
7115         if (type == BTRFS_ORDERED_PREALLOC)
7116                 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7117
7118         do {
7119                 btrfs_drop_extent_cache(inode, em->start,
7120                                 em->start + em->len - 1, 0);
7121                 write_lock(&em_tree->lock);
7122                 ret = add_extent_mapping(em_tree, em, 1);
7123                 write_unlock(&em_tree->lock);
7124         } while (ret == -EEXIST);
7125
7126         if (ret) {
7127                 free_extent_map(em);
7128                 return ERR_PTR(ret);
7129         }
7130
7131         return em;
7132 }
7133
7134
7135 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7136                                    struct buffer_head *bh_result, int create)
7137 {
7138         struct extent_map *em;
7139         struct btrfs_root *root = BTRFS_I(inode)->root;
7140         struct extent_state *cached_state = NULL;
7141         u64 start = iblock << inode->i_blkbits;
7142         u64 lockstart, lockend;
7143         u64 len = bh_result->b_size;
7144         int unlock_bits = EXTENT_LOCKED;
7145         int ret = 0;
7146
7147         if (create)
7148                 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
7149         else
7150                 len = min_t(u64, len, root->sectorsize);
7151
7152         lockstart = start;
7153         lockend = start + len - 1;
7154
7155         /*
7156          * If this errors out it's because we couldn't invalidate pagecache for
7157          * this range and we need to fallback to buffered.
7158          */
7159         if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
7160                 return -ENOTBLK;
7161
7162         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
7163         if (IS_ERR(em)) {
7164                 ret = PTR_ERR(em);
7165                 goto unlock_err;
7166         }
7167
7168         /*
7169          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7170          * io.  INLINE is special, and we could probably kludge it in here, but
7171          * it's still buffered so for safety lets just fall back to the generic
7172          * buffered path.
7173          *
7174          * For COMPRESSED we _have_ to read the entire extent in so we can
7175          * decompress it, so there will be buffering required no matter what we
7176          * do, so go ahead and fallback to buffered.
7177          *
7178          * We return -ENOTBLK because thats what makes DIO go ahead and go back
7179          * to buffered IO.  Don't blame me, this is the price we pay for using
7180          * the generic code.
7181          */
7182         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7183             em->block_start == EXTENT_MAP_INLINE) {
7184                 free_extent_map(em);
7185                 ret = -ENOTBLK;
7186                 goto unlock_err;
7187         }
7188
7189         /* Just a good old fashioned hole, return */
7190         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7191                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7192                 free_extent_map(em);
7193                 goto unlock_err;
7194         }
7195
7196         /*
7197          * We don't allocate a new extent in the following cases
7198          *
7199          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
7200          * existing extent.
7201          * 2) The extent is marked as PREALLOC.  We're good to go here and can
7202          * just use the extent.
7203          *
7204          */
7205         if (!create) {
7206                 len = min(len, em->len - (start - em->start));
7207                 lockstart = start + len;
7208                 goto unlock;
7209         }
7210
7211         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7212             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7213              em->block_start != EXTENT_MAP_HOLE)) {
7214                 int type;
7215                 int ret;
7216                 u64 block_start, orig_start, orig_block_len, ram_bytes;
7217
7218                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7219                         type = BTRFS_ORDERED_PREALLOC;
7220                 else
7221                         type = BTRFS_ORDERED_NOCOW;
7222                 len = min(len, em->len - (start - em->start));
7223                 block_start = em->block_start + (start - em->start);
7224
7225                 if (can_nocow_extent(inode, start, &len, &orig_start,
7226                                      &orig_block_len, &ram_bytes) == 1) {
7227                         if (type == BTRFS_ORDERED_PREALLOC) {
7228                                 free_extent_map(em);
7229                                 em = create_pinned_em(inode, start, len,
7230                                                        orig_start,
7231                                                        block_start, len,
7232                                                        orig_block_len,
7233                                                        ram_bytes, type);
7234                                 if (IS_ERR(em)) {
7235                                         ret = PTR_ERR(em);
7236                                         goto unlock_err;
7237                                 }
7238                         }
7239
7240                         ret = btrfs_add_ordered_extent_dio(inode, start,
7241                                            block_start, len, len, type);
7242                         if (ret) {
7243                                 free_extent_map(em);
7244                                 goto unlock_err;
7245                         }
7246                         goto unlock;
7247                 }
7248         }
7249
7250         /*
7251          * this will cow the extent, reset the len in case we changed
7252          * it above
7253          */
7254         len = bh_result->b_size;
7255         free_extent_map(em);
7256         em = btrfs_new_extent_direct(inode, start, len);
7257         if (IS_ERR(em)) {
7258                 ret = PTR_ERR(em);
7259                 goto unlock_err;
7260         }
7261         len = min(len, em->len - (start - em->start));
7262 unlock:
7263         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7264                 inode->i_blkbits;
7265         bh_result->b_size = len;
7266         bh_result->b_bdev = em->bdev;
7267         set_buffer_mapped(bh_result);
7268         if (create) {
7269                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7270                         set_buffer_new(bh_result);
7271
7272                 /*
7273                  * Need to update the i_size under the extent lock so buffered
7274                  * readers will get the updated i_size when we unlock.
7275                  */
7276                 if (start + len > i_size_read(inode))
7277                         i_size_write(inode, start + len);
7278
7279                 spin_lock(&BTRFS_I(inode)->lock);
7280                 BTRFS_I(inode)->outstanding_extents++;
7281                 spin_unlock(&BTRFS_I(inode)->lock);
7282
7283                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7284                                      lockstart + len - 1, EXTENT_DELALLOC, NULL,
7285                                      &cached_state, GFP_NOFS);
7286                 BUG_ON(ret);
7287         }
7288
7289         /*
7290          * In the case of write we need to clear and unlock the entire range,
7291          * in the case of read we need to unlock only the end area that we
7292          * aren't using if there is any left over space.
7293          */
7294         if (lockstart < lockend) {
7295                 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7296                                  lockend, unlock_bits, 1, 0,
7297                                  &cached_state, GFP_NOFS);
7298         } else {
7299                 free_extent_state(cached_state);
7300         }
7301
7302         free_extent_map(em);
7303
7304         return 0;
7305
7306 unlock_err:
7307         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7308                          unlock_bits, 1, 0, &cached_state, GFP_NOFS);
7309         return ret;
7310 }
7311
7312 static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7313                                         int rw, int mirror_num)
7314 {
7315         struct btrfs_root *root = BTRFS_I(inode)->root;
7316         int ret;
7317
7318         BUG_ON(rw & REQ_WRITE);
7319
7320         bio_get(bio);
7321
7322         ret = btrfs_bio_wq_end_io(root->fs_info, bio,
7323                                   BTRFS_WQ_ENDIO_DIO_REPAIR);
7324         if (ret)
7325                 goto err;
7326
7327         ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
7328 err:
7329         bio_put(bio);
7330         return ret;
7331 }
7332
7333 static int btrfs_check_dio_repairable(struct inode *inode,
7334                                       struct bio *failed_bio,
7335                                       struct io_failure_record *failrec,
7336                                       int failed_mirror)
7337 {
7338         int num_copies;
7339
7340         num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
7341                                       failrec->logical, failrec->len);
7342         if (num_copies == 1) {
7343                 /*
7344                  * we only have a single copy of the data, so don't bother with
7345                  * all the retry and error correction code that follows. no
7346                  * matter what the error is, it is very likely to persist.
7347                  */
7348                 pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
7349                          num_copies, failrec->this_mirror, failed_mirror);
7350                 return 0;
7351         }
7352
7353         failrec->failed_mirror = failed_mirror;
7354         failrec->this_mirror++;
7355         if (failrec->this_mirror == failed_mirror)
7356                 failrec->this_mirror++;
7357
7358         if (failrec->this_mirror > num_copies) {
7359                 pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
7360                          num_copies, failrec->this_mirror, failed_mirror);
7361                 return 0;
7362         }
7363
7364         return 1;
7365 }
7366
7367 static int dio_read_error(struct inode *inode, struct bio *failed_bio,
7368                           struct page *page, u64 start, u64 end,
7369                           int failed_mirror, bio_end_io_t *repair_endio,
7370                           void *repair_arg)
7371 {
7372         struct io_failure_record *failrec;
7373         struct bio *bio;
7374         int isector;
7375         int read_mode;
7376         int ret;
7377
7378         BUG_ON(failed_bio->bi_rw & REQ_WRITE);
7379
7380         ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7381         if (ret)
7382                 return ret;
7383
7384         ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7385                                          failed_mirror);
7386         if (!ret) {
7387                 free_io_failure(inode, failrec);
7388                 return -EIO;
7389         }
7390
7391         if (failed_bio->bi_vcnt > 1)
7392                 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
7393         else
7394                 read_mode = READ_SYNC;
7395
7396         isector = start - btrfs_io_bio(failed_bio)->logical;
7397         isector >>= inode->i_sb->s_blocksize_bits;
7398         bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7399                                       0, isector, repair_endio, repair_arg);
7400         if (!bio) {
7401                 free_io_failure(inode, failrec);
7402                 return -EIO;
7403         }
7404
7405         btrfs_debug(BTRFS_I(inode)->root->fs_info,
7406                     "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
7407                     read_mode, failrec->this_mirror, failrec->in_validation);
7408
7409         ret = submit_dio_repair_bio(inode, bio, read_mode,
7410                                     failrec->this_mirror);
7411         if (ret) {
7412                 free_io_failure(inode, failrec);
7413                 bio_put(bio);
7414         }
7415
7416         return ret;
7417 }
7418
7419 struct btrfs_retry_complete {
7420         struct completion done;
7421         struct inode *inode;
7422         u64 start;
7423         int uptodate;
7424 };
7425
7426 static void btrfs_retry_endio_nocsum(struct bio *bio, int err)
7427 {
7428         struct btrfs_retry_complete *done = bio->bi_private;
7429         struct bio_vec *bvec;
7430         int i;
7431
7432         if (err)
7433                 goto end;
7434
7435         done->uptodate = 1;
7436         bio_for_each_segment_all(bvec, bio, i)
7437                 clean_io_failure(done->inode, done->start, bvec->bv_page, 0);
7438 end:
7439         complete(&done->done);
7440         bio_put(bio);
7441 }
7442
7443 static int __btrfs_correct_data_nocsum(struct inode *inode,
7444                                        struct btrfs_io_bio *io_bio)
7445 {
7446         struct bio_vec *bvec;
7447         struct btrfs_retry_complete done;
7448         u64 start;
7449         int i;
7450         int ret;
7451
7452         start = io_bio->logical;
7453         done.inode = inode;
7454
7455         bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7456 try_again:
7457                 done.uptodate = 0;
7458                 done.start = start;
7459                 init_completion(&done.done);
7460
7461                 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start,
7462                                      start + bvec->bv_len - 1,
7463                                      io_bio->mirror_num,
7464                                      btrfs_retry_endio_nocsum, &done);
7465                 if (ret)
7466                         return ret;
7467
7468                 wait_for_completion(&done.done);
7469
7470                 if (!done.uptodate) {
7471                         /* We might have another mirror, so try again */
7472                         goto try_again;
7473                 }
7474
7475                 start += bvec->bv_len;
7476         }
7477
7478         return 0;
7479 }
7480
7481 static void btrfs_retry_endio(struct bio *bio, int err)
7482 {
7483         struct btrfs_retry_complete *done = bio->bi_private;
7484         struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7485         struct bio_vec *bvec;
7486         int uptodate;
7487         int ret;
7488         int i;
7489
7490         if (err)
7491                 goto end;
7492
7493         uptodate = 1;
7494         bio_for_each_segment_all(bvec, bio, i) {
7495                 ret = __readpage_endio_check(done->inode, io_bio, i,
7496                                              bvec->bv_page, 0,
7497                                              done->start, bvec->bv_len);
7498                 if (!ret)
7499                         clean_io_failure(done->inode, done->start,
7500                                          bvec->bv_page, 0);
7501                 else
7502                         uptodate = 0;
7503         }
7504
7505         done->uptodate = uptodate;
7506 end:
7507         complete(&done->done);
7508         bio_put(bio);
7509 }
7510
7511 static int __btrfs_subio_endio_read(struct inode *inode,
7512                                     struct btrfs_io_bio *io_bio, int err)
7513 {
7514         struct bio_vec *bvec;
7515         struct btrfs_retry_complete done;
7516         u64 start;
7517         u64 offset = 0;
7518         int i;
7519         int ret;
7520
7521         err = 0;
7522         start = io_bio->logical;
7523         done.inode = inode;
7524
7525         bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7526                 ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page,
7527                                              0, start, bvec->bv_len);
7528                 if (likely(!ret))
7529                         goto next;
7530 try_again:
7531                 done.uptodate = 0;
7532                 done.start = start;
7533                 init_completion(&done.done);
7534
7535                 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start,
7536                                      start + bvec->bv_len - 1,
7537                                      io_bio->mirror_num,
7538                                      btrfs_retry_endio, &done);
7539                 if (ret) {
7540                         err = ret;
7541                         goto next;
7542                 }
7543
7544                 wait_for_completion(&done.done);
7545
7546                 if (!done.uptodate) {
7547                         /* We might have another mirror, so try again */
7548                         goto try_again;
7549                 }
7550 next:
7551                 offset += bvec->bv_len;
7552                 start += bvec->bv_len;
7553         }
7554
7555         return err;
7556 }
7557
7558 static int btrfs_subio_endio_read(struct inode *inode,
7559                                   struct btrfs_io_bio *io_bio, int err)
7560 {
7561         bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7562
7563         if (skip_csum) {
7564                 if (unlikely(err))
7565                         return __btrfs_correct_data_nocsum(inode, io_bio);
7566                 else
7567                         return 0;
7568         } else {
7569                 return __btrfs_subio_endio_read(inode, io_bio, err);
7570         }
7571 }
7572
7573 static void btrfs_endio_direct_read(struct bio *bio, int err)
7574 {
7575         struct btrfs_dio_private *dip = bio->bi_private;
7576         struct inode *inode = dip->inode;
7577         struct bio *dio_bio;
7578         struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7579
7580         if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
7581                 err = btrfs_subio_endio_read(inode, io_bio, err);
7582
7583         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
7584                       dip->logical_offset + dip->bytes - 1);
7585         dio_bio = dip->dio_bio;
7586
7587         kfree(dip);
7588
7589         /* If we had a csum failure make sure to clear the uptodate flag */
7590         if (err)
7591                 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7592         dio_end_io(dio_bio, err);
7593
7594         if (io_bio->end_io)
7595                 io_bio->end_io(io_bio, err);
7596         bio_put(bio);
7597 }
7598
7599 static void btrfs_endio_direct_write(struct bio *bio, int err)
7600 {
7601         struct btrfs_dio_private *dip = bio->bi_private;
7602         struct inode *inode = dip->inode;
7603         struct btrfs_root *root = BTRFS_I(inode)->root;
7604         struct btrfs_ordered_extent *ordered = NULL;
7605         u64 ordered_offset = dip->logical_offset;
7606         u64 ordered_bytes = dip->bytes;
7607         struct bio *dio_bio;
7608         int ret;
7609
7610         if (err)
7611                 goto out_done;
7612 again:
7613         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
7614                                                    &ordered_offset,
7615                                                    ordered_bytes, !err);
7616         if (!ret)
7617                 goto out_test;
7618
7619         btrfs_init_work(&ordered->work, btrfs_endio_write_helper,
7620                         finish_ordered_fn, NULL, NULL);
7621         btrfs_queue_work(root->fs_info->endio_write_workers,
7622                          &ordered->work);
7623 out_test:
7624         /*
7625          * our bio might span multiple ordered extents.  If we haven't
7626          * completed the accounting for the whole dio, go back and try again
7627          */
7628         if (ordered_offset < dip->logical_offset + dip->bytes) {
7629                 ordered_bytes = dip->logical_offset + dip->bytes -
7630                         ordered_offset;
7631                 ordered = NULL;
7632                 goto again;
7633         }
7634 out_done:
7635         dio_bio = dip->dio_bio;
7636
7637         kfree(dip);
7638
7639         /* If we had an error make sure to clear the uptodate flag */
7640         if (err)
7641                 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7642         dio_end_io(dio_bio, err);
7643         bio_put(bio);
7644 }
7645
7646 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
7647                                     struct bio *bio, int mirror_num,
7648                                     unsigned long bio_flags, u64 offset)
7649 {
7650         int ret;
7651         struct btrfs_root *root = BTRFS_I(inode)->root;
7652         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
7653         BUG_ON(ret); /* -ENOMEM */
7654         return 0;
7655 }
7656
7657 static void btrfs_end_dio_bio(struct bio *bio, int err)
7658 {
7659         struct btrfs_dio_private *dip = bio->bi_private;
7660
7661         if (err)
7662                 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
7663                            "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
7664                            btrfs_ino(dip->inode), bio->bi_rw,
7665                            (unsigned long long)bio->bi_iter.bi_sector,
7666                            bio->bi_iter.bi_size, err);
7667
7668         if (dip->subio_endio)
7669                 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
7670
7671         if (err) {
7672                 dip->errors = 1;
7673
7674                 /*
7675                  * before atomic variable goto zero, we must make sure
7676                  * dip->errors is perceived to be set.
7677                  */
7678                 smp_mb__before_atomic();
7679         }
7680
7681         /* if there are more bios still pending for this dio, just exit */
7682         if (!atomic_dec_and_test(&dip->pending_bios))
7683                 goto out;
7684
7685         if (dip->errors) {
7686                 bio_io_error(dip->orig_bio);
7687         } else {
7688                 set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags);
7689                 bio_endio(dip->orig_bio, 0);
7690         }
7691 out:
7692         bio_put(bio);
7693 }
7694
7695 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
7696                                        u64 first_sector, gfp_t gfp_flags)
7697 {
7698         int nr_vecs = bio_get_nr_vecs(bdev);
7699         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
7700 }
7701
7702 static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root *root,
7703                                                  struct inode *inode,
7704                                                  struct btrfs_dio_private *dip,
7705                                                  struct bio *bio,
7706                                                  u64 file_offset)
7707 {
7708         struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7709         struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
7710         int ret;
7711
7712         /*
7713          * We load all the csum data we need when we submit
7714          * the first bio to reduce the csum tree search and
7715          * contention.
7716          */
7717         if (dip->logical_offset == file_offset) {
7718                 ret = btrfs_lookup_bio_sums_dio(root, inode, dip->orig_bio,
7719                                                 file_offset);
7720                 if (ret)
7721                         return ret;
7722         }
7723
7724         if (bio == dip->orig_bio)
7725                 return 0;
7726
7727         file_offset -= dip->logical_offset;
7728         file_offset >>= inode->i_sb->s_blocksize_bits;
7729         io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
7730
7731         return 0;
7732 }
7733
7734 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
7735                                          int rw, u64 file_offset, int skip_sum,
7736                                          int async_submit)
7737 {
7738         struct btrfs_dio_private *dip = bio->bi_private;
7739         int write = rw & REQ_WRITE;
7740         struct btrfs_root *root = BTRFS_I(inode)->root;
7741         int ret;
7742
7743         if (async_submit)
7744                 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7745
7746         bio_get(bio);
7747
7748         if (!write) {
7749                 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
7750                                 BTRFS_WQ_ENDIO_DATA);
7751                 if (ret)
7752                         goto err;
7753         }
7754
7755         if (skip_sum)
7756                 goto map;
7757
7758         if (write && async_submit) {
7759                 ret = btrfs_wq_submit_bio(root->fs_info,
7760                                    inode, rw, bio, 0, 0,
7761                                    file_offset,
7762                                    __btrfs_submit_bio_start_direct_io,
7763                                    __btrfs_submit_bio_done);
7764                 goto err;
7765         } else if (write) {
7766                 /*
7767                  * If we aren't doing async submit, calculate the csum of the
7768                  * bio now.
7769                  */
7770                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
7771                 if (ret)
7772                         goto err;
7773         } else {
7774                 ret = btrfs_lookup_and_bind_dio_csum(root, inode, dip, bio,
7775                                                      file_offset);
7776                 if (ret)
7777                         goto err;
7778         }
7779 map:
7780         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
7781 err:
7782         bio_put(bio);
7783         return ret;
7784 }
7785
7786 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
7787                                     int skip_sum)
7788 {
7789         struct inode *inode = dip->inode;
7790         struct btrfs_root *root = BTRFS_I(inode)->root;
7791         struct bio *bio;
7792         struct bio *orig_bio = dip->orig_bio;
7793         struct bio_vec *bvec = orig_bio->bi_io_vec;
7794         u64 start_sector = orig_bio->bi_iter.bi_sector;
7795         u64 file_offset = dip->logical_offset;
7796         u64 submit_len = 0;
7797         u64 map_length;
7798         int nr_pages = 0;
7799         int ret;
7800         int async_submit = 0;
7801
7802         map_length = orig_bio->bi_iter.bi_size;
7803         ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
7804                               &map_length, NULL, 0);
7805         if (ret)
7806                 return -EIO;
7807
7808         if (map_length >= orig_bio->bi_iter.bi_size) {
7809                 bio = orig_bio;
7810                 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
7811                 goto submit;
7812         }
7813
7814         /* async crcs make it difficult to collect full stripe writes. */
7815         if (btrfs_get_alloc_profile(root, 1) &
7816             (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6))
7817                 async_submit = 0;
7818         else
7819                 async_submit = 1;
7820
7821         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
7822         if (!bio)
7823                 return -ENOMEM;
7824
7825         bio->bi_private = dip;
7826         bio->bi_end_io = btrfs_end_dio_bio;
7827         btrfs_io_bio(bio)->logical = file_offset;
7828         atomic_inc(&dip->pending_bios);
7829
7830         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
7831                 if (map_length < submit_len + bvec->bv_len ||
7832                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
7833                                  bvec->bv_offset) < bvec->bv_len) {
7834                         /*
7835                          * inc the count before we submit the bio so
7836                          * we know the end IO handler won't happen before
7837                          * we inc the count. Otherwise, the dip might get freed
7838                          * before we're done setting it up
7839                          */
7840                         atomic_inc(&dip->pending_bios);
7841                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
7842                                                      file_offset, skip_sum,
7843                                                      async_submit);
7844                         if (ret) {
7845                                 bio_put(bio);
7846                                 atomic_dec(&dip->pending_bios);
7847                                 goto out_err;
7848                         }
7849
7850                         start_sector += submit_len >> 9;
7851                         file_offset += submit_len;
7852
7853                         submit_len = 0;
7854                         nr_pages = 0;
7855
7856                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
7857                                                   start_sector, GFP_NOFS);
7858                         if (!bio)
7859                                 goto out_err;
7860                         bio->bi_private = dip;
7861                         bio->bi_end_io = btrfs_end_dio_bio;
7862                         btrfs_io_bio(bio)->logical = file_offset;
7863
7864                         map_length = orig_bio->bi_iter.bi_size;
7865                         ret = btrfs_map_block(root->fs_info, rw,
7866                                               start_sector << 9,
7867                                               &map_length, NULL, 0);
7868                         if (ret) {
7869                                 bio_put(bio);
7870                                 goto out_err;
7871                         }
7872                 } else {
7873                         submit_len += bvec->bv_len;
7874                         nr_pages++;
7875                         bvec++;
7876                 }
7877         }
7878
7879 submit:
7880         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
7881                                      async_submit);
7882         if (!ret)
7883                 return 0;
7884
7885         bio_put(bio);
7886 out_err:
7887         dip->errors = 1;
7888         /*
7889          * before atomic variable goto zero, we must
7890          * make sure dip->errors is perceived to be set.
7891          */
7892         smp_mb__before_atomic();
7893         if (atomic_dec_and_test(&dip->pending_bios))
7894                 bio_io_error(dip->orig_bio);
7895
7896         /* bio_end_io() will handle error, so we needn't return it */
7897         return 0;
7898 }
7899
7900 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
7901                                 struct inode *inode, loff_t file_offset)
7902 {
7903         struct btrfs_root *root = BTRFS_I(inode)->root;
7904         struct btrfs_dio_private *dip;
7905         struct bio *io_bio;
7906         struct btrfs_io_bio *btrfs_bio;
7907         int skip_sum;
7908         int write = rw & REQ_WRITE;
7909         int ret = 0;
7910
7911         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7912
7913         io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
7914         if (!io_bio) {
7915                 ret = -ENOMEM;
7916                 goto free_ordered;
7917         }
7918
7919         dip = kzalloc(sizeof(*dip), GFP_NOFS);
7920         if (!dip) {
7921                 ret = -ENOMEM;
7922                 goto free_io_bio;
7923         }
7924
7925         dip->private = dio_bio->bi_private;
7926         dip->inode = inode;
7927         dip->logical_offset = file_offset;
7928         dip->bytes = dio_bio->bi_iter.bi_size;
7929         dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
7930         io_bio->bi_private = dip;
7931         dip->orig_bio = io_bio;
7932         dip->dio_bio = dio_bio;
7933         atomic_set(&dip->pending_bios, 0);
7934         btrfs_bio = btrfs_io_bio(io_bio);
7935         btrfs_bio->logical = file_offset;
7936
7937         if (write) {
7938                 io_bio->bi_end_io = btrfs_endio_direct_write;
7939         } else {
7940                 io_bio->bi_end_io = btrfs_endio_direct_read;
7941                 dip->subio_endio = btrfs_subio_endio_read;
7942         }
7943
7944         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
7945         if (!ret)
7946                 return;
7947
7948         if (btrfs_bio->end_io)
7949                 btrfs_bio->end_io(btrfs_bio, ret);
7950 free_io_bio:
7951         bio_put(io_bio);
7952
7953 free_ordered:
7954         /*
7955          * If this is a write, we need to clean up the reserved space and kill
7956          * the ordered extent.
7957          */
7958         if (write) {
7959                 struct btrfs_ordered_extent *ordered;
7960                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
7961                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
7962                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
7963                         btrfs_free_reserved_extent(root, ordered->start,
7964                                                    ordered->disk_len, 1);
7965                 btrfs_put_ordered_extent(ordered);
7966                 btrfs_put_ordered_extent(ordered);
7967         }
7968         bio_endio(dio_bio, ret);
7969 }
7970
7971 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
7972                         const struct iov_iter *iter, loff_t offset)
7973 {
7974         int seg;
7975         int i;
7976         unsigned blocksize_mask = root->sectorsize - 1;
7977         ssize_t retval = -EINVAL;
7978
7979         if (offset & blocksize_mask)
7980                 goto out;
7981
7982         if (iov_iter_alignment(iter) & blocksize_mask)
7983                 goto out;
7984
7985         /* If this is a write we don't need to check anymore */
7986         if (rw & WRITE)
7987                 return 0;
7988         /*
7989          * Check to make sure we don't have duplicate iov_base's in this
7990          * iovec, if so return EINVAL, otherwise we'll get csum errors
7991          * when reading back.
7992          */
7993         for (seg = 0; seg < iter->nr_segs; seg++) {
7994                 for (i = seg + 1; i < iter->nr_segs; i++) {
7995                         if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
7996                                 goto out;
7997                 }
7998         }
7999         retval = 0;
8000 out:
8001         return retval;
8002 }
8003
8004 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
8005                         struct iov_iter *iter, loff_t offset)
8006 {
8007         struct file *file = iocb->ki_filp;
8008         struct inode *inode = file->f_mapping->host;
8009         size_t count = 0;
8010         int flags = 0;
8011         bool wakeup = true;
8012         bool relock = false;
8013         ssize_t ret;
8014
8015         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iter, offset))
8016                 return 0;
8017
8018         atomic_inc(&inode->i_dio_count);
8019         smp_mb__after_atomic();
8020
8021         /*
8022          * The generic stuff only does filemap_write_and_wait_range, which
8023          * isn't enough if we've written compressed pages to this area, so
8024          * we need to flush the dirty pages again to make absolutely sure
8025          * that any outstanding dirty pages are on disk.
8026          */
8027         count = iov_iter_count(iter);
8028         if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8029                      &BTRFS_I(inode)->runtime_flags))
8030                 filemap_fdatawrite_range(inode->i_mapping, offset,
8031                                          offset + count - 1);
8032
8033         if (rw & WRITE) {
8034                 /*
8035                  * If the write DIO is beyond the EOF, we need update
8036                  * the isize, but it is protected by i_mutex. So we can
8037                  * not unlock the i_mutex at this case.
8038                  */
8039                 if (offset + count <= inode->i_size) {
8040                         mutex_unlock(&inode->i_mutex);
8041                         relock = true;
8042                 }
8043                 ret = btrfs_delalloc_reserve_space(inode, count);
8044                 if (ret)
8045                         goto out;
8046         } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8047                                      &BTRFS_I(inode)->runtime_flags)) {
8048                 inode_dio_done(inode);
8049                 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8050                 wakeup = false;
8051         }
8052
8053         ret = __blockdev_direct_IO(rw, iocb, inode,
8054                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
8055                         iter, offset, btrfs_get_blocks_direct, NULL,
8056                         btrfs_submit_direct, flags);
8057         if (rw & WRITE) {
8058                 if (ret < 0 && ret != -EIOCBQUEUED)
8059                         btrfs_delalloc_release_space(inode, count);
8060                 else if (ret >= 0 && (size_t)ret < count)
8061                         btrfs_delalloc_release_space(inode,
8062                                                      count - (size_t)ret);
8063                 else
8064                         btrfs_delalloc_release_metadata(inode, 0);
8065         }
8066 out:
8067         if (wakeup)
8068                 inode_dio_done(inode);
8069         if (relock)
8070                 mutex_lock(&inode->i_mutex);
8071
8072         return ret;
8073 }
8074
8075 #define BTRFS_FIEMAP_FLAGS      (FIEMAP_FLAG_SYNC)
8076
8077 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8078                 __u64 start, __u64 len)
8079 {
8080         int     ret;
8081
8082         ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8083         if (ret)
8084                 return ret;
8085
8086         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
8087 }
8088
8089 int btrfs_readpage(struct file *file, struct page *page)
8090 {
8091         struct extent_io_tree *tree;
8092         tree = &BTRFS_I(page->mapping->host)->io_tree;
8093         return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8094 }
8095
8096 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8097 {
8098         struct extent_io_tree *tree;
8099
8100
8101         if (current->flags & PF_MEMALLOC) {
8102                 redirty_page_for_writepage(wbc, page);
8103                 unlock_page(page);
8104                 return 0;
8105         }
8106         tree = &BTRFS_I(page->mapping->host)->io_tree;
8107         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
8108 }
8109
8110 static int btrfs_writepages(struct address_space *mapping,
8111                             struct writeback_control *wbc)
8112 {
8113         struct extent_io_tree *tree;
8114
8115         tree = &BTRFS_I(mapping->host)->io_tree;
8116         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8117 }
8118
8119 static int
8120 btrfs_readpages(struct file *file, struct address_space *mapping,
8121                 struct list_head *pages, unsigned nr_pages)
8122 {
8123         struct extent_io_tree *tree;
8124         tree = &BTRFS_I(mapping->host)->io_tree;
8125         return extent_readpages(tree, mapping, pages, nr_pages,
8126                                 btrfs_get_extent);
8127 }
8128 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8129 {
8130         struct extent_io_tree *tree;
8131         struct extent_map_tree *map;
8132         int ret;
8133
8134         tree = &BTRFS_I(page->mapping->host)->io_tree;
8135         map = &BTRFS_I(page->mapping->host)->extent_tree;
8136         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
8137         if (ret == 1) {
8138                 ClearPagePrivate(page);
8139                 set_page_private(page, 0);
8140                 page_cache_release(page);
8141         }
8142         return ret;
8143 }
8144
8145 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8146 {
8147         if (PageWriteback(page) || PageDirty(page))
8148                 return 0;
8149         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
8150 }
8151
8152 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8153                                  unsigned int length)
8154 {
8155         struct inode *inode = page->mapping->host;
8156         struct extent_io_tree *tree;
8157         struct btrfs_ordered_extent *ordered;
8158         struct extent_state *cached_state = NULL;
8159         u64 page_start = page_offset(page);
8160         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
8161         int inode_evicting = inode->i_state & I_FREEING;
8162
8163         /*
8164          * we have the page locked, so new writeback can't start,
8165          * and the dirty bit won't be cleared while we are here.
8166          *
8167          * Wait for IO on this page so that we can safely clear
8168          * the PagePrivate2 bit and do ordered accounting
8169          */
8170         wait_on_page_writeback(page);
8171
8172         tree = &BTRFS_I(inode)->io_tree;
8173         if (offset) {
8174                 btrfs_releasepage(page, GFP_NOFS);
8175                 return;
8176         }
8177
8178         if (!inode_evicting)
8179                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
8180         ordered = btrfs_lookup_ordered_extent(inode, page_start);
8181         if (ordered) {
8182                 /*
8183                  * IO on this page will never be started, so we need
8184                  * to account for any ordered extents now
8185                  */
8186                 if (!inode_evicting)
8187                         clear_extent_bit(tree, page_start, page_end,
8188                                          EXTENT_DIRTY | EXTENT_DELALLOC |
8189                                          EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8190                                          EXTENT_DEFRAG, 1, 0, &cached_state,
8191                                          GFP_NOFS);
8192                 /*
8193                  * whoever cleared the private bit is responsible
8194                  * for the finish_ordered_io
8195                  */
8196                 if (TestClearPagePrivate2(page)) {
8197                         struct btrfs_ordered_inode_tree *tree;
8198                         u64 new_len;
8199
8200                         tree = &BTRFS_I(inode)->ordered_tree;
8201
8202                         spin_lock_irq(&tree->lock);
8203                         set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8204                         new_len = page_start - ordered->file_offset;
8205                         if (new_len < ordered->truncated_len)
8206                                 ordered->truncated_len = new_len;
8207                         spin_unlock_irq(&tree->lock);
8208
8209                         if (btrfs_dec_test_ordered_pending(inode, &ordered,
8210                                                            page_start,
8211                                                            PAGE_CACHE_SIZE, 1))
8212                                 btrfs_finish_ordered_io(ordered);
8213                 }
8214                 btrfs_put_ordered_extent(ordered);
8215                 if (!inode_evicting) {
8216                         cached_state = NULL;
8217                         lock_extent_bits(tree, page_start, page_end, 0,
8218                                          &cached_state);
8219                 }
8220         }
8221
8222         if (!inode_evicting) {
8223                 clear_extent_bit(tree, page_start, page_end,
8224                                  EXTENT_LOCKED | EXTENT_DIRTY |
8225                                  EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8226                                  EXTENT_DEFRAG, 1, 1,
8227                                  &cached_state, GFP_NOFS);
8228
8229                 __btrfs_releasepage(page, GFP_NOFS);
8230         }
8231
8232         ClearPageChecked(page);
8233         if (PagePrivate(page)) {
8234                 ClearPagePrivate(page);
8235                 set_page_private(page, 0);
8236                 page_cache_release(page);
8237         }
8238 }
8239
8240 /*
8241  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8242  * called from a page fault handler when a page is first dirtied. Hence we must
8243  * be careful to check for EOF conditions here. We set the page up correctly
8244  * for a written page which means we get ENOSPC checking when writing into
8245  * holes and correct delalloc and unwritten extent mapping on filesystems that
8246  * support these features.
8247  *
8248  * We are not allowed to take the i_mutex here so we have to play games to
8249  * protect against truncate races as the page could now be beyond EOF.  Because
8250  * vmtruncate() writes the inode size before removing pages, once we have the
8251  * page lock we can determine safely if the page is beyond EOF. If it is not
8252  * beyond EOF, then the page is guaranteed safe against truncation until we
8253  * unlock the page.
8254  */
8255 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
8256 {
8257         struct page *page = vmf->page;
8258         struct inode *inode = file_inode(vma->vm_file);
8259         struct btrfs_root *root = BTRFS_I(inode)->root;
8260         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8261         struct btrfs_ordered_extent *ordered;
8262         struct extent_state *cached_state = NULL;
8263         char *kaddr;
8264         unsigned long zero_start;
8265         loff_t size;
8266         int ret;
8267         int reserved = 0;
8268         u64 page_start;
8269         u64 page_end;
8270
8271         sb_start_pagefault(inode->i_sb);
8272         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
8273         if (!ret) {
8274                 ret = file_update_time(vma->vm_file);
8275                 reserved = 1;
8276         }
8277         if (ret) {
8278                 if (ret == -ENOMEM)
8279                         ret = VM_FAULT_OOM;
8280                 else /* -ENOSPC, -EIO, etc */
8281                         ret = VM_FAULT_SIGBUS;
8282                 if (reserved)
8283                         goto out;
8284                 goto out_noreserve;
8285         }
8286
8287         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8288 again:
8289         lock_page(page);
8290         size = i_size_read(inode);
8291         page_start = page_offset(page);
8292         page_end = page_start + PAGE_CACHE_SIZE - 1;
8293
8294         if ((page->mapping != inode->i_mapping) ||
8295             (page_start >= size)) {
8296                 /* page got truncated out from underneath us */
8297                 goto out_unlock;
8298         }
8299         wait_on_page_writeback(page);
8300
8301         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
8302         set_page_extent_mapped(page);
8303
8304         /*
8305          * we can't set the delalloc bits if there are pending ordered
8306          * extents.  Drop our locks and wait for them to finish
8307          */
8308         ordered = btrfs_lookup_ordered_extent(inode, page_start);
8309         if (ordered) {
8310                 unlock_extent_cached(io_tree, page_start, page_end,
8311                                      &cached_state, GFP_NOFS);
8312                 unlock_page(page);
8313                 btrfs_start_ordered_extent(inode, ordered, 1);
8314                 btrfs_put_ordered_extent(ordered);
8315                 goto again;
8316         }
8317
8318         /*
8319          * XXX - page_mkwrite gets called every time the page is dirtied, even
8320          * if it was already dirty, so for space accounting reasons we need to
8321          * clear any delalloc bits for the range we are fixing to save.  There
8322          * is probably a better way to do this, but for now keep consistent with
8323          * prepare_pages in the normal write path.
8324          */
8325         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
8326                           EXTENT_DIRTY | EXTENT_DELALLOC |
8327                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
8328                           0, 0, &cached_state, GFP_NOFS);
8329
8330         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
8331                                         &cached_state);
8332         if (ret) {
8333                 unlock_extent_cached(io_tree, page_start, page_end,
8334                                      &cached_state, GFP_NOFS);
8335                 ret = VM_FAULT_SIGBUS;
8336                 goto out_unlock;
8337         }
8338         ret = 0;
8339
8340         /* page is wholly or partially inside EOF */
8341         if (page_start + PAGE_CACHE_SIZE > size)
8342                 zero_start = size & ~PAGE_CACHE_MASK;
8343         else
8344                 zero_start = PAGE_CACHE_SIZE;
8345
8346         if (zero_start != PAGE_CACHE_SIZE) {
8347                 kaddr = kmap(page);
8348                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
8349                 flush_dcache_page(page);
8350                 kunmap(page);
8351         }
8352         ClearPageChecked(page);
8353         set_page_dirty(page);
8354         SetPageUptodate(page);
8355
8356         BTRFS_I(inode)->last_trans = root->fs_info->generation;
8357         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
8358         BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
8359
8360         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
8361
8362 out_unlock:
8363         if (!ret) {
8364                 sb_end_pagefault(inode->i_sb);
8365                 return VM_FAULT_LOCKED;
8366         }
8367         unlock_page(page);
8368 out:
8369         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
8370 out_noreserve:
8371         sb_end_pagefault(inode->i_sb);
8372         return ret;
8373 }
8374
8375 static int btrfs_truncate(struct inode *inode)
8376 {
8377         struct btrfs_root *root = BTRFS_I(inode)->root;
8378         struct btrfs_block_rsv *rsv;
8379         int ret = 0;
8380         int err = 0;
8381         struct btrfs_trans_handle *trans;
8382         u64 mask = root->sectorsize - 1;
8383         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
8384
8385         ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8386                                        (u64)-1);
8387         if (ret)
8388                 return ret;
8389
8390         /*
8391          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
8392          * 3 things going on here
8393          *
8394          * 1) We need to reserve space for our orphan item and the space to
8395          * delete our orphan item.  Lord knows we don't want to have a dangling
8396          * orphan item because we didn't reserve space to remove it.
8397          *
8398          * 2) We need to reserve space to update our inode.
8399          *
8400          * 3) We need to have something to cache all the space that is going to
8401          * be free'd up by the truncate operation, but also have some slack
8402          * space reserved in case it uses space during the truncate (thank you
8403          * very much snapshotting).
8404          *
8405          * And we need these to all be seperate.  The fact is we can use alot of
8406          * space doing the truncate, and we have no earthly idea how much space
8407          * we will use, so we need the truncate reservation to be seperate so it
8408          * doesn't end up using space reserved for updating the inode or
8409          * removing the orphan item.  We also need to be able to stop the
8410          * transaction and start a new one, which means we need to be able to
8411          * update the inode several times, and we have no idea of knowing how
8412          * many times that will be, so we can't just reserve 1 item for the
8413          * entirety of the opration, so that has to be done seperately as well.
8414          * Then there is the orphan item, which does indeed need to be held on
8415          * to for the whole operation, and we need nobody to touch this reserved
8416          * space except the orphan code.
8417          *
8418          * So that leaves us with
8419          *
8420          * 1) root->orphan_block_rsv - for the orphan deletion.
8421          * 2) rsv - for the truncate reservation, which we will steal from the
8422          * transaction reservation.
8423          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
8424          * updating the inode.
8425          */
8426         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
8427         if (!rsv)
8428                 return -ENOMEM;
8429         rsv->size = min_size;
8430         rsv->failfast = 1;
8431
8432         /*
8433          * 1 for the truncate slack space
8434          * 1 for updating the inode.
8435          */
8436         trans = btrfs_start_transaction(root, 2);
8437         if (IS_ERR(trans)) {
8438                 err = PTR_ERR(trans);
8439                 goto out;
8440         }
8441
8442         /* Migrate the slack space for the truncate to our reserve */
8443         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
8444                                       min_size);
8445         BUG_ON(ret);
8446
8447         /*
8448          * So if we truncate and then write and fsync we normally would just
8449          * write the extents that changed, which is a problem if we need to
8450          * first truncate that entire inode.  So set this flag so we write out
8451          * all of the extents in the inode to the sync log so we're completely
8452          * safe.
8453          */
8454         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
8455         trans->block_rsv = rsv;
8456
8457         while (1) {
8458                 ret = btrfs_truncate_inode_items(trans, root, inode,
8459                                                  inode->i_size,
8460                                                  BTRFS_EXTENT_DATA_KEY);
8461                 if (ret != -ENOSPC) {
8462                         err = ret;
8463                         break;
8464                 }
8465
8466                 trans->block_rsv = &root->fs_info->trans_block_rsv;
8467                 ret = btrfs_update_inode(trans, root, inode);
8468                 if (ret) {
8469                         err = ret;
8470                         break;
8471                 }
8472
8473                 btrfs_end_transaction(trans, root);
8474                 btrfs_btree_balance_dirty(root);
8475
8476                 trans = btrfs_start_transaction(root, 2);
8477                 if (IS_ERR(trans)) {
8478                         ret = err = PTR_ERR(trans);
8479                         trans = NULL;
8480                         break;
8481                 }
8482
8483                 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
8484                                               rsv, min_size);
8485                 BUG_ON(ret);    /* shouldn't happen */
8486                 trans->block_rsv = rsv;
8487         }
8488
8489         if (ret == 0 && inode->i_nlink > 0) {
8490                 trans->block_rsv = root->orphan_block_rsv;
8491                 ret = btrfs_orphan_del(trans, inode);
8492                 if (ret)
8493                         err = ret;
8494         }
8495
8496         if (trans) {
8497                 trans->block_rsv = &root->fs_info->trans_block_rsv;
8498                 ret = btrfs_update_inode(trans, root, inode);
8499                 if (ret && !err)
8500                         err = ret;
8501
8502                 ret = btrfs_end_transaction(trans, root);
8503                 btrfs_btree_balance_dirty(root);
8504         }
8505
8506 out:
8507         btrfs_free_block_rsv(root, rsv);
8508
8509         if (ret && !err)
8510                 err = ret;
8511
8512         return err;
8513 }
8514
8515 /*
8516  * create a new subvolume directory/inode (helper for the ioctl).
8517  */
8518 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
8519                              struct btrfs_root *new_root,
8520                              struct btrfs_root *parent_root,
8521                              u64 new_dirid)
8522 {
8523         struct inode *inode;
8524         int err;
8525         u64 index = 0;
8526
8527         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
8528                                 new_dirid, new_dirid,
8529                                 S_IFDIR | (~current_umask() & S_IRWXUGO),
8530                                 &index);
8531         if (IS_ERR(inode))
8532                 return PTR_ERR(inode);
8533         inode->i_op = &btrfs_dir_inode_operations;
8534         inode->i_fop = &btrfs_dir_file_operations;
8535
8536         set_nlink(inode, 1);
8537         btrfs_i_size_write(inode, 0);
8538         unlock_new_inode(inode);
8539
8540         err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
8541         if (err)
8542                 btrfs_err(new_root->fs_info,
8543                           "error inheriting subvolume %llu properties: %d",
8544                           new_root->root_key.objectid, err);
8545
8546         err = btrfs_update_inode(trans, new_root, inode);
8547
8548         iput(inode);
8549         return err;
8550 }
8551
8552 struct inode *btrfs_alloc_inode(struct super_block *sb)
8553 {
8554         struct btrfs_inode *ei;
8555         struct inode *inode;
8556
8557         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
8558         if (!ei)
8559                 return NULL;
8560
8561         ei->root = NULL;
8562         ei->generation = 0;
8563         ei->last_trans = 0;
8564         ei->last_sub_trans = 0;
8565         ei->logged_trans = 0;
8566         ei->delalloc_bytes = 0;
8567         ei->defrag_bytes = 0;
8568         ei->disk_i_size = 0;
8569         ei->flags = 0;
8570         ei->csum_bytes = 0;
8571         ei->index_cnt = (u64)-1;
8572         ei->dir_index = 0;
8573         ei->last_unlink_trans = 0;
8574         ei->last_log_commit = 0;
8575
8576         spin_lock_init(&ei->lock);
8577         ei->outstanding_extents = 0;
8578         ei->reserved_extents = 0;
8579
8580         ei->runtime_flags = 0;
8581         ei->force_compress = BTRFS_COMPRESS_NONE;
8582
8583         ei->delayed_node = NULL;
8584
8585         inode = &ei->vfs_inode;
8586         extent_map_tree_init(&ei->extent_tree);
8587         extent_io_tree_init(&ei->io_tree, &inode->i_data);
8588         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
8589         ei->io_tree.track_uptodate = 1;
8590         ei->io_failure_tree.track_uptodate = 1;
8591         atomic_set(&ei->sync_writers, 0);
8592         mutex_init(&ei->log_mutex);
8593         mutex_init(&ei->delalloc_mutex);
8594         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8595         INIT_LIST_HEAD(&ei->delalloc_inodes);
8596         RB_CLEAR_NODE(&ei->rb_node);
8597
8598         return inode;
8599 }
8600
8601 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8602 void btrfs_test_destroy_inode(struct inode *inode)
8603 {
8604         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8605         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8606 }
8607 #endif
8608
8609 static void btrfs_i_callback(struct rcu_head *head)
8610 {
8611         struct inode *inode = container_of(head, struct inode, i_rcu);
8612         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8613 }
8614
8615 void btrfs_destroy_inode(struct inode *inode)
8616 {
8617         struct btrfs_ordered_extent *ordered;
8618         struct btrfs_root *root = BTRFS_I(inode)->root;
8619
8620         WARN_ON(!hlist_empty(&inode->i_dentry));
8621         WARN_ON(inode->i_data.nrpages);
8622         WARN_ON(BTRFS_I(inode)->outstanding_extents);
8623         WARN_ON(BTRFS_I(inode)->reserved_extents);
8624         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
8625         WARN_ON(BTRFS_I(inode)->csum_bytes);
8626         WARN_ON(BTRFS_I(inode)->defrag_bytes);
8627
8628         /*
8629          * This can happen where we create an inode, but somebody else also
8630          * created the same inode and we need to destroy the one we already
8631          * created.
8632          */
8633         if (!root)
8634                 goto free;
8635
8636         if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
8637                      &BTRFS_I(inode)->runtime_flags)) {
8638                 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
8639                         btrfs_ino(inode));
8640                 atomic_dec(&root->orphan_inodes);
8641         }
8642
8643         while (1) {
8644                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8645                 if (!ordered)
8646                         break;
8647                 else {
8648                         btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
8649                                 ordered->file_offset, ordered->len);
8650                         btrfs_remove_ordered_extent(inode, ordered);
8651                         btrfs_put_ordered_extent(ordered);
8652                         btrfs_put_ordered_extent(ordered);
8653                 }
8654         }
8655         inode_tree_del(inode);
8656         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8657 free:
8658         call_rcu(&inode->i_rcu, btrfs_i_callback);
8659 }
8660
8661 int btrfs_drop_inode(struct inode *inode)
8662 {
8663         struct btrfs_root *root = BTRFS_I(inode)->root;
8664
8665         if (root == NULL)
8666                 return 1;
8667
8668         /* the snap/subvol tree is on deleting */
8669         if (btrfs_root_refs(&root->root_item) == 0)
8670                 return 1;
8671         else
8672                 return generic_drop_inode(inode);
8673 }
8674
8675 static void init_once(void *foo)
8676 {
8677         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8678
8679         inode_init_once(&ei->vfs_inode);
8680 }
8681
8682 void btrfs_destroy_cachep(void)
8683 {
8684         /*
8685          * Make sure all delayed rcu free inodes are flushed before we
8686          * destroy cache.
8687          */
8688         rcu_barrier();
8689         if (btrfs_inode_cachep)
8690                 kmem_cache_destroy(btrfs_inode_cachep);
8691         if (btrfs_trans_handle_cachep)
8692                 kmem_cache_destroy(btrfs_trans_handle_cachep);
8693         if (btrfs_transaction_cachep)
8694                 kmem_cache_destroy(btrfs_transaction_cachep);
8695         if (btrfs_path_cachep)
8696                 kmem_cache_destroy(btrfs_path_cachep);
8697         if (btrfs_free_space_cachep)
8698                 kmem_cache_destroy(btrfs_free_space_cachep);
8699         if (btrfs_delalloc_work_cachep)
8700                 kmem_cache_destroy(btrfs_delalloc_work_cachep);
8701 }
8702
8703 int btrfs_init_cachep(void)
8704 {
8705         btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8706                         sizeof(struct btrfs_inode), 0,
8707                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
8708         if (!btrfs_inode_cachep)
8709                 goto fail;
8710
8711         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8712                         sizeof(struct btrfs_trans_handle), 0,
8713                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8714         if (!btrfs_trans_handle_cachep)
8715                 goto fail;
8716
8717         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
8718                         sizeof(struct btrfs_transaction), 0,
8719                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8720         if (!btrfs_transaction_cachep)
8721                 goto fail;
8722
8723         btrfs_path_cachep = kmem_cache_create("btrfs_path",
8724                         sizeof(struct btrfs_path), 0,
8725                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8726         if (!btrfs_path_cachep)
8727                 goto fail;
8728
8729         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
8730                         sizeof(struct btrfs_free_space), 0,
8731                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8732         if (!btrfs_free_space_cachep)
8733                 goto fail;
8734
8735         btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
8736                         sizeof(struct btrfs_delalloc_work), 0,
8737                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
8738                         NULL);
8739         if (!btrfs_delalloc_work_cachep)
8740                 goto fail;
8741
8742         return 0;
8743 fail:
8744         btrfs_destroy_cachep();
8745         return -ENOMEM;
8746 }
8747
8748 static int btrfs_getattr(struct vfsmount *mnt,
8749                          struct dentry *dentry, struct kstat *stat)
8750 {
8751         u64 delalloc_bytes;
8752         struct inode *inode = dentry->d_inode;
8753         u32 blocksize = inode->i_sb->s_blocksize;
8754
8755         generic_fillattr(inode, stat);
8756         stat->dev = BTRFS_I(inode)->root->anon_dev;
8757         stat->blksize = PAGE_CACHE_SIZE;
8758
8759         spin_lock(&BTRFS_I(inode)->lock);
8760         delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
8761         spin_unlock(&BTRFS_I(inode)->lock);
8762         stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
8763                         ALIGN(delalloc_bytes, blocksize)) >> 9;
8764         return 0;
8765 }
8766
8767 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
8768                            struct inode *new_dir, struct dentry *new_dentry)
8769 {
8770         struct btrfs_trans_handle *trans;
8771         struct btrfs_root *root = BTRFS_I(old_dir)->root;
8772         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8773         struct inode *new_inode = new_dentry->d_inode;
8774         struct inode *old_inode = old_dentry->d_inode;
8775         struct timespec ctime = CURRENT_TIME;
8776         u64 index = 0;
8777         u64 root_objectid;
8778         int ret;
8779         u64 old_ino = btrfs_ino(old_inode);
8780
8781         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
8782                 return -EPERM;
8783
8784         /* we only allow rename subvolume link between subvolumes */
8785         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8786                 return -EXDEV;
8787
8788         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8789             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
8790                 return -ENOTEMPTY;
8791
8792         if (S_ISDIR(old_inode->i_mode) && new_inode &&
8793             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8794                 return -ENOTEMPTY;
8795
8796
8797         /* check for collisions, even if the  name isn't there */
8798         ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
8799                              new_dentry->d_name.name,
8800                              new_dentry->d_name.len);
8801
8802         if (ret) {
8803                 if (ret == -EEXIST) {
8804                         /* we shouldn't get
8805                          * eexist without a new_inode */
8806                         if (WARN_ON(!new_inode)) {
8807                                 return ret;
8808                         }
8809                 } else {
8810                         /* maybe -EOVERFLOW */
8811                         return ret;
8812                 }
8813         }
8814         ret = 0;
8815
8816         /*
8817          * we're using rename to replace one file with another.  Start IO on it
8818          * now so  we don't add too much work to the end of the transaction
8819          */
8820         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
8821                 filemap_flush(old_inode->i_mapping);
8822
8823         /* close the racy window with snapshot create/destroy ioctl */
8824         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8825                 down_read(&root->fs_info->subvol_sem);
8826         /*
8827          * We want to reserve the absolute worst case amount of items.  So if
8828          * both inodes are subvols and we need to unlink them then that would
8829          * require 4 item modifications, but if they are both normal inodes it
8830          * would require 5 item modifications, so we'll assume their normal
8831          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8832          * should cover the worst case number of items we'll modify.
8833          */
8834         trans = btrfs_start_transaction(root, 11);
8835         if (IS_ERR(trans)) {
8836                 ret = PTR_ERR(trans);
8837                 goto out_notrans;
8838         }
8839
8840         if (dest != root)
8841                 btrfs_record_root_in_trans(trans, dest);
8842
8843         ret = btrfs_set_inode_index(new_dir, &index);
8844         if (ret)
8845                 goto out_fail;
8846
8847         BTRFS_I(old_inode)->dir_index = 0ULL;
8848         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8849                 /* force full log commit if subvolume involved. */
8850                 btrfs_set_log_full_commit(root->fs_info, trans);
8851         } else {
8852                 ret = btrfs_insert_inode_ref(trans, dest,
8853                                              new_dentry->d_name.name,
8854                                              new_dentry->d_name.len,
8855                                              old_ino,
8856                                              btrfs_ino(new_dir), index);
8857                 if (ret)
8858                         goto out_fail;
8859                 /*
8860                  * this is an ugly little race, but the rename is required
8861                  * to make sure that if we crash, the inode is either at the
8862                  * old name or the new one.  pinning the log transaction lets
8863                  * us make sure we don't allow a log commit to come in after
8864                  * we unlink the name but before we add the new name back in.
8865                  */
8866                 btrfs_pin_log_trans(root);
8867         }
8868
8869         inode_inc_iversion(old_dir);
8870         inode_inc_iversion(new_dir);
8871         inode_inc_iversion(old_inode);
8872         old_dir->i_ctime = old_dir->i_mtime = ctime;
8873         new_dir->i_ctime = new_dir->i_mtime = ctime;
8874         old_inode->i_ctime = ctime;
8875
8876         if (old_dentry->d_parent != new_dentry->d_parent)
8877                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
8878
8879         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8880                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
8881                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
8882                                         old_dentry->d_name.name,
8883                                         old_dentry->d_name.len);
8884         } else {
8885                 ret = __btrfs_unlink_inode(trans, root, old_dir,
8886                                         old_dentry->d_inode,
8887                                         old_dentry->d_name.name,
8888                                         old_dentry->d_name.len);
8889                 if (!ret)
8890                         ret = btrfs_update_inode(trans, root, old_inode);
8891         }
8892         if (ret) {
8893                 btrfs_abort_transaction(trans, root, ret);
8894                 goto out_fail;
8895         }
8896
8897         if (new_inode) {
8898                 inode_inc_iversion(new_inode);
8899                 new_inode->i_ctime = CURRENT_TIME;
8900                 if (unlikely(btrfs_ino(new_inode) ==
8901                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
8902                         root_objectid = BTRFS_I(new_inode)->location.objectid;
8903                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
8904                                                 root_objectid,
8905                                                 new_dentry->d_name.name,
8906                                                 new_dentry->d_name.len);
8907                         BUG_ON(new_inode->i_nlink == 0);
8908                 } else {
8909                         ret = btrfs_unlink_inode(trans, dest, new_dir,
8910                                                  new_dentry->d_inode,
8911                                                  new_dentry->d_name.name,
8912                                                  new_dentry->d_name.len);
8913                 }
8914                 if (!ret && new_inode->i_nlink == 0)
8915                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
8916                 if (ret) {
8917                         btrfs_abort_transaction(trans, root, ret);
8918                         goto out_fail;
8919                 }
8920         }
8921
8922         ret = btrfs_add_link(trans, new_dir, old_inode,
8923                              new_dentry->d_name.name,
8924                              new_dentry->d_name.len, 0, index);
8925         if (ret) {
8926                 btrfs_abort_transaction(trans, root, ret);
8927                 goto out_fail;
8928         }
8929
8930         if (old_inode->i_nlink == 1)
8931                 BTRFS_I(old_inode)->dir_index = index;
8932
8933         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
8934                 struct dentry *parent = new_dentry->d_parent;
8935                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
8936                 btrfs_end_log_trans(root);
8937         }
8938 out_fail:
8939         btrfs_end_transaction(trans, root);
8940 out_notrans:
8941         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8942                 up_read(&root->fs_info->subvol_sem);
8943
8944         return ret;
8945 }
8946
8947 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
8948                          struct inode *new_dir, struct dentry *new_dentry,
8949                          unsigned int flags)
8950 {
8951         if (flags & ~RENAME_NOREPLACE)
8952                 return -EINVAL;
8953
8954         return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry);
8955 }
8956
8957 static void btrfs_run_delalloc_work(struct btrfs_work *work)
8958 {
8959         struct btrfs_delalloc_work *delalloc_work;
8960         struct inode *inode;
8961
8962         delalloc_work = container_of(work, struct btrfs_delalloc_work,
8963                                      work);
8964         inode = delalloc_work->inode;
8965         if (delalloc_work->wait) {
8966                 btrfs_wait_ordered_range(inode, 0, (u64)-1);
8967         } else {
8968                 filemap_flush(inode->i_mapping);
8969                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8970                              &BTRFS_I(inode)->runtime_flags))
8971                         filemap_flush(inode->i_mapping);
8972         }
8973
8974         if (delalloc_work->delay_iput)
8975                 btrfs_add_delayed_iput(inode);
8976         else
8977                 iput(inode);
8978         complete(&delalloc_work->completion);
8979 }
8980
8981 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
8982                                                     int wait, int delay_iput)
8983 {
8984         struct btrfs_delalloc_work *work;
8985
8986         work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
8987         if (!work)
8988                 return NULL;
8989
8990         init_completion(&work->completion);
8991         INIT_LIST_HEAD(&work->list);
8992         work->inode = inode;
8993         work->wait = wait;
8994         work->delay_iput = delay_iput;
8995         WARN_ON_ONCE(!inode);
8996         btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
8997                         btrfs_run_delalloc_work, NULL, NULL);
8998
8999         return work;
9000 }
9001
9002 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
9003 {
9004         wait_for_completion(&work->completion);
9005         kmem_cache_free(btrfs_delalloc_work_cachep, work);
9006 }
9007
9008 /*
9009  * some fairly slow code that needs optimization. This walks the list
9010  * of all the inodes with pending delalloc and forces them to disk.
9011  */
9012 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
9013                                    int nr)
9014 {
9015         struct btrfs_inode *binode;
9016         struct inode *inode;
9017         struct btrfs_delalloc_work *work, *next;
9018         struct list_head works;
9019         struct list_head splice;
9020         int ret = 0;
9021
9022         INIT_LIST_HEAD(&works);
9023         INIT_LIST_HEAD(&splice);
9024
9025         mutex_lock(&root->delalloc_mutex);
9026         spin_lock(&root->delalloc_lock);
9027         list_splice_init(&root->delalloc_inodes, &splice);
9028         while (!list_empty(&splice)) {
9029                 binode = list_entry(splice.next, struct btrfs_inode,
9030                                     delalloc_inodes);
9031
9032                 list_move_tail(&binode->delalloc_inodes,
9033                                &root->delalloc_inodes);
9034                 inode = igrab(&binode->vfs_inode);
9035                 if (!inode) {
9036                         cond_resched_lock(&root->delalloc_lock);
9037                         continue;
9038                 }
9039                 spin_unlock(&root->delalloc_lock);
9040
9041                 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
9042                 if (!work) {
9043                         if (delay_iput)
9044                                 btrfs_add_delayed_iput(inode);
9045                         else
9046                                 iput(inode);
9047                         ret = -ENOMEM;
9048                         goto out;
9049                 }
9050                 list_add_tail(&work->list, &works);
9051                 btrfs_queue_work(root->fs_info->flush_workers,
9052                                  &work->work);
9053                 ret++;
9054                 if (nr != -1 && ret >= nr)
9055                         goto out;
9056                 cond_resched();
9057                 spin_lock(&root->delalloc_lock);
9058         }
9059         spin_unlock(&root->delalloc_lock);
9060
9061 out:
9062         list_for_each_entry_safe(work, next, &works, list) {
9063                 list_del_init(&work->list);
9064                 btrfs_wait_and_free_delalloc_work(work);
9065         }
9066
9067         if (!list_empty_careful(&splice)) {
9068                 spin_lock(&root->delalloc_lock);
9069                 list_splice_tail(&splice, &root->delalloc_inodes);
9070                 spin_unlock(&root->delalloc_lock);
9071         }
9072         mutex_unlock(&root->delalloc_mutex);
9073         return ret;
9074 }
9075
9076 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
9077 {
9078         int ret;
9079
9080         if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
9081                 return -EROFS;
9082
9083         ret = __start_delalloc_inodes(root, delay_iput, -1);
9084         if (ret > 0)
9085                 ret = 0;
9086         /*
9087          * the filemap_flush will queue IO into the worker threads, but
9088          * we have to make sure the IO is actually started and that
9089          * ordered extents get created before we return
9090          */
9091         atomic_inc(&root->fs_info->async_submit_draining);
9092         while (atomic_read(&root->fs_info->nr_async_submits) ||
9093               atomic_read(&root->fs_info->async_delalloc_pages)) {
9094                 wait_event(root->fs_info->async_submit_wait,
9095                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
9096                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
9097         }
9098         atomic_dec(&root->fs_info->async_submit_draining);
9099         return ret;
9100 }
9101
9102 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
9103                                int nr)
9104 {
9105         struct btrfs_root *root;
9106         struct list_head splice;
9107         int ret;
9108
9109         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9110                 return -EROFS;
9111
9112         INIT_LIST_HEAD(&splice);
9113
9114         mutex_lock(&fs_info->delalloc_root_mutex);
9115         spin_lock(&fs_info->delalloc_root_lock);
9116         list_splice_init(&fs_info->delalloc_roots, &splice);
9117         while (!list_empty(&splice) && nr) {
9118                 root = list_first_entry(&splice, struct btrfs_root,
9119                                         delalloc_root);
9120                 root = btrfs_grab_fs_root(root);
9121                 BUG_ON(!root);
9122                 list_move_tail(&root->delalloc_root,
9123                                &fs_info->delalloc_roots);
9124                 spin_unlock(&fs_info->delalloc_root_lock);
9125
9126                 ret = __start_delalloc_inodes(root, delay_iput, nr);
9127                 btrfs_put_fs_root(root);
9128                 if (ret < 0)
9129                         goto out;
9130
9131                 if (nr != -1) {
9132                         nr -= ret;
9133                         WARN_ON(nr < 0);
9134                 }
9135                 spin_lock(&fs_info->delalloc_root_lock);
9136         }
9137         spin_unlock(&fs_info->delalloc_root_lock);
9138
9139         ret = 0;
9140         atomic_inc(&fs_info->async_submit_draining);
9141         while (atomic_read(&fs_info->nr_async_submits) ||
9142               atomic_read(&fs_info->async_delalloc_pages)) {
9143                 wait_event(fs_info->async_submit_wait,
9144                    (atomic_read(&fs_info->nr_async_submits) == 0 &&
9145                     atomic_read(&fs_info->async_delalloc_pages) == 0));
9146         }
9147         atomic_dec(&fs_info->async_submit_draining);
9148 out:
9149         if (!list_empty_careful(&splice)) {
9150                 spin_lock(&fs_info->delalloc_root_lock);
9151                 list_splice_tail(&splice, &fs_info->delalloc_roots);
9152                 spin_unlock(&fs_info->delalloc_root_lock);
9153         }
9154         mutex_unlock(&fs_info->delalloc_root_mutex);
9155         return ret;
9156 }
9157
9158 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
9159                          const char *symname)
9160 {
9161         struct btrfs_trans_handle *trans;
9162         struct btrfs_root *root = BTRFS_I(dir)->root;
9163         struct btrfs_path *path;
9164         struct btrfs_key key;
9165         struct inode *inode = NULL;
9166         int err;
9167         int drop_inode = 0;
9168         u64 objectid;
9169         u64 index = 0;
9170         int name_len;
9171         int datasize;
9172         unsigned long ptr;
9173         struct btrfs_file_extent_item *ei;
9174         struct extent_buffer *leaf;
9175
9176         name_len = strlen(symname);
9177         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
9178                 return -ENAMETOOLONG;
9179
9180         /*
9181          * 2 items for inode item and ref
9182          * 2 items for dir items
9183          * 1 item for xattr if selinux is on
9184          */
9185         trans = btrfs_start_transaction(root, 5);
9186         if (IS_ERR(trans))
9187                 return PTR_ERR(trans);
9188
9189         err = btrfs_find_free_ino(root, &objectid);
9190         if (err)
9191                 goto out_unlock;
9192
9193         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9194                                 dentry->d_name.len, btrfs_ino(dir), objectid,
9195                                 S_IFLNK|S_IRWXUGO, &index);
9196         if (IS_ERR(inode)) {
9197                 err = PTR_ERR(inode);
9198                 goto out_unlock;
9199         }
9200
9201         /*
9202         * If the active LSM wants to access the inode during
9203         * d_instantiate it needs these. Smack checks to see
9204         * if the filesystem supports xattrs by looking at the
9205         * ops vector.
9206         */
9207         inode->i_fop = &btrfs_file_operations;
9208         inode->i_op = &btrfs_file_inode_operations;
9209         inode->i_mapping->a_ops = &btrfs_aops;
9210         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
9211         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9212
9213         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
9214         if (err)
9215                 goto out_unlock_inode;
9216
9217         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
9218         if (err)
9219                 goto out_unlock_inode;
9220
9221         path = btrfs_alloc_path();
9222         if (!path) {
9223                 err = -ENOMEM;
9224                 goto out_unlock_inode;
9225         }
9226         key.objectid = btrfs_ino(inode);
9227         key.offset = 0;
9228         key.type = BTRFS_EXTENT_DATA_KEY;
9229         datasize = btrfs_file_extent_calc_inline_size(name_len);
9230         err = btrfs_insert_empty_item(trans, root, path, &key,
9231                                       datasize);
9232         if (err) {
9233                 btrfs_free_path(path);
9234                 goto out_unlock_inode;
9235         }
9236         leaf = path->nodes[0];
9237         ei = btrfs_item_ptr(leaf, path->slots[0],
9238                             struct btrfs_file_extent_item);
9239         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9240         btrfs_set_file_extent_type(leaf, ei,
9241                                    BTRFS_FILE_EXTENT_INLINE);
9242         btrfs_set_file_extent_encryption(leaf, ei, 0);
9243         btrfs_set_file_extent_compression(leaf, ei, 0);
9244         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9245         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9246
9247         ptr = btrfs_file_extent_inline_start(ei);
9248         write_extent_buffer(leaf, symname, ptr, name_len);
9249         btrfs_mark_buffer_dirty(leaf);
9250         btrfs_free_path(path);
9251
9252         inode->i_op = &btrfs_symlink_inode_operations;
9253         inode->i_mapping->a_ops = &btrfs_symlink_aops;
9254         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
9255         inode_set_bytes(inode, name_len);
9256         btrfs_i_size_write(inode, name_len);
9257         err = btrfs_update_inode(trans, root, inode);
9258         if (err) {
9259                 drop_inode = 1;
9260                 goto out_unlock_inode;
9261         }
9262
9263         unlock_new_inode(inode);
9264         d_instantiate(dentry, inode);
9265
9266 out_unlock:
9267         btrfs_end_transaction(trans, root);
9268         if (drop_inode) {
9269                 inode_dec_link_count(inode);
9270                 iput(inode);
9271         }
9272         btrfs_btree_balance_dirty(root);
9273         return err;
9274
9275 out_unlock_inode:
9276         drop_inode = 1;
9277         unlock_new_inode(inode);
9278         goto out_unlock;
9279 }
9280
9281 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9282                                        u64 start, u64 num_bytes, u64 min_size,
9283                                        loff_t actual_len, u64 *alloc_hint,
9284                                        struct btrfs_trans_handle *trans)
9285 {
9286         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9287         struct extent_map *em;
9288         struct btrfs_root *root = BTRFS_I(inode)->root;
9289         struct btrfs_key ins;
9290         u64 cur_offset = start;
9291         u64 i_size;
9292         u64 cur_bytes;
9293         int ret = 0;
9294         bool own_trans = true;
9295
9296         if (trans)
9297                 own_trans = false;
9298         while (num_bytes > 0) {
9299                 if (own_trans) {
9300                         trans = btrfs_start_transaction(root, 3);
9301                         if (IS_ERR(trans)) {
9302                                 ret = PTR_ERR(trans);
9303                                 break;
9304                         }
9305                 }
9306
9307                 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
9308                 cur_bytes = max(cur_bytes, min_size);
9309                 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
9310                                            *alloc_hint, &ins, 1, 0);
9311                 if (ret) {
9312                         if (own_trans)
9313                                 btrfs_end_transaction(trans, root);
9314                         break;
9315                 }
9316
9317                 ret = insert_reserved_file_extent(trans, inode,
9318                                                   cur_offset, ins.objectid,
9319                                                   ins.offset, ins.offset,
9320                                                   ins.offset, 0, 0, 0,
9321                                                   BTRFS_FILE_EXTENT_PREALLOC);
9322                 if (ret) {
9323                         btrfs_free_reserved_extent(root, ins.objectid,
9324                                                    ins.offset, 0);
9325                         btrfs_abort_transaction(trans, root, ret);
9326                         if (own_trans)
9327                                 btrfs_end_transaction(trans, root);
9328                         break;
9329                 }
9330                 btrfs_drop_extent_cache(inode, cur_offset,
9331                                         cur_offset + ins.offset -1, 0);
9332
9333                 em = alloc_extent_map();
9334                 if (!em) {
9335                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
9336                                 &BTRFS_I(inode)->runtime_flags);
9337                         goto next;
9338                 }
9339
9340                 em->start = cur_offset;
9341                 em->orig_start = cur_offset;
9342                 em->len = ins.offset;
9343                 em->block_start = ins.objectid;
9344                 em->block_len = ins.offset;
9345                 em->orig_block_len = ins.offset;
9346                 em->ram_bytes = ins.offset;
9347                 em->bdev = root->fs_info->fs_devices->latest_bdev;
9348                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9349                 em->generation = trans->transid;
9350
9351                 while (1) {
9352                         write_lock(&em_tree->lock);
9353                         ret = add_extent_mapping(em_tree, em, 1);
9354                         write_unlock(&em_tree->lock);
9355                         if (ret != -EEXIST)
9356                                 break;
9357                         btrfs_drop_extent_cache(inode, cur_offset,
9358                                                 cur_offset + ins.offset - 1,
9359                                                 0);
9360                 }
9361                 free_extent_map(em);
9362 next:
9363                 num_bytes -= ins.offset;
9364                 cur_offset += ins.offset;
9365                 *alloc_hint = ins.objectid + ins.offset;
9366
9367                 inode_inc_iversion(inode);
9368                 inode->i_ctime = CURRENT_TIME;
9369                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
9370                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
9371                     (actual_len > inode->i_size) &&
9372                     (cur_offset > inode->i_size)) {
9373                         if (cur_offset > actual_len)
9374                                 i_size = actual_len;
9375                         else
9376                                 i_size = cur_offset;
9377                         i_size_write(inode, i_size);
9378                         btrfs_ordered_update_i_size(inode, i_size, NULL);
9379                 }
9380
9381                 ret = btrfs_update_inode(trans, root, inode);
9382
9383                 if (ret) {
9384                         btrfs_abort_transaction(trans, root, ret);
9385                         if (own_trans)
9386                                 btrfs_end_transaction(trans, root);
9387                         break;
9388                 }
9389
9390                 if (own_trans)
9391                         btrfs_end_transaction(trans, root);
9392         }
9393         return ret;
9394 }
9395
9396 int btrfs_prealloc_file_range(struct inode *inode, int mode,
9397                               u64 start, u64 num_bytes, u64 min_size,
9398                               loff_t actual_len, u64 *alloc_hint)
9399 {
9400         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9401                                            min_size, actual_len, alloc_hint,
9402                                            NULL);
9403 }
9404
9405 int btrfs_prealloc_file_range_trans(struct inode *inode,
9406                                     struct btrfs_trans_handle *trans, int mode,
9407                                     u64 start, u64 num_bytes, u64 min_size,
9408                                     loff_t actual_len, u64 *alloc_hint)
9409 {
9410         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9411                                            min_size, actual_len, alloc_hint, trans);
9412 }
9413
9414 static int btrfs_set_page_dirty(struct page *page)
9415 {
9416         return __set_page_dirty_nobuffers(page);
9417 }
9418
9419 static int btrfs_permission(struct inode *inode, int mask)
9420 {
9421         struct btrfs_root *root = BTRFS_I(inode)->root;
9422         umode_t mode = inode->i_mode;
9423
9424         if (mask & MAY_WRITE &&
9425             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9426                 if (btrfs_root_readonly(root))
9427                         return -EROFS;
9428                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9429                         return -EACCES;
9430         }
9431         return generic_permission(inode, mask);
9432 }
9433
9434 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
9435 {
9436         struct btrfs_trans_handle *trans;
9437         struct btrfs_root *root = BTRFS_I(dir)->root;
9438         struct inode *inode = NULL;
9439         u64 objectid;
9440         u64 index;
9441         int ret = 0;
9442
9443         /*
9444          * 5 units required for adding orphan entry
9445          */
9446         trans = btrfs_start_transaction(root, 5);
9447         if (IS_ERR(trans))
9448                 return PTR_ERR(trans);
9449
9450         ret = btrfs_find_free_ino(root, &objectid);
9451         if (ret)
9452                 goto out;
9453
9454         inode = btrfs_new_inode(trans, root, dir, NULL, 0,
9455                                 btrfs_ino(dir), objectid, mode, &index);
9456         if (IS_ERR(inode)) {
9457                 ret = PTR_ERR(inode);
9458                 inode = NULL;
9459                 goto out;
9460         }
9461
9462         inode->i_fop = &btrfs_file_operations;
9463         inode->i_op = &btrfs_file_inode_operations;
9464
9465         inode->i_mapping->a_ops = &btrfs_aops;
9466         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
9467         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9468
9469         ret = btrfs_init_inode_security(trans, inode, dir, NULL);
9470         if (ret)
9471                 goto out_inode;
9472
9473         ret = btrfs_update_inode(trans, root, inode);
9474         if (ret)
9475                 goto out_inode;
9476         ret = btrfs_orphan_add(trans, inode);
9477         if (ret)
9478                 goto out_inode;
9479
9480         /*
9481          * We set number of links to 0 in btrfs_new_inode(), and here we set
9482          * it to 1 because d_tmpfile() will issue a warning if the count is 0,
9483          * through:
9484          *
9485          *    d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9486          */
9487         set_nlink(inode, 1);
9488         unlock_new_inode(inode);
9489         d_tmpfile(dentry, inode);
9490         mark_inode_dirty(inode);
9491
9492 out:
9493         btrfs_end_transaction(trans, root);
9494         if (ret)
9495                 iput(inode);
9496         btrfs_balance_delayed_items(root);
9497         btrfs_btree_balance_dirty(root);
9498         return ret;
9499
9500 out_inode:
9501         unlock_new_inode(inode);
9502         goto out;
9503
9504 }
9505
9506 /* Inspired by filemap_check_errors() */
9507 int btrfs_inode_check_errors(struct inode *inode)
9508 {
9509         int ret = 0;
9510
9511         if (test_bit(AS_ENOSPC, &inode->i_mapping->flags) &&
9512             test_and_clear_bit(AS_ENOSPC, &inode->i_mapping->flags))
9513                 ret = -ENOSPC;
9514         if (test_bit(AS_EIO, &inode->i_mapping->flags) &&
9515             test_and_clear_bit(AS_EIO, &inode->i_mapping->flags))
9516                 ret = -EIO;
9517
9518         return ret;
9519 }
9520
9521 static const struct inode_operations btrfs_dir_inode_operations = {
9522         .getattr        = btrfs_getattr,
9523         .lookup         = btrfs_lookup,
9524         .create         = btrfs_create,
9525         .unlink         = btrfs_unlink,
9526         .link           = btrfs_link,
9527         .mkdir          = btrfs_mkdir,
9528         .rmdir          = btrfs_rmdir,
9529         .rename2        = btrfs_rename2,
9530         .symlink        = btrfs_symlink,
9531         .setattr        = btrfs_setattr,
9532         .mknod          = btrfs_mknod,
9533         .setxattr       = btrfs_setxattr,
9534         .getxattr       = btrfs_getxattr,
9535         .listxattr      = btrfs_listxattr,
9536         .removexattr    = btrfs_removexattr,
9537         .permission     = btrfs_permission,
9538         .get_acl        = btrfs_get_acl,
9539         .set_acl        = btrfs_set_acl,
9540         .update_time    = btrfs_update_time,
9541         .tmpfile        = btrfs_tmpfile,
9542 };
9543 static const struct inode_operations btrfs_dir_ro_inode_operations = {
9544         .lookup         = btrfs_lookup,
9545         .permission     = btrfs_permission,
9546         .get_acl        = btrfs_get_acl,
9547         .set_acl        = btrfs_set_acl,
9548         .update_time    = btrfs_update_time,
9549 };
9550
9551 static const struct file_operations btrfs_dir_file_operations = {
9552         .llseek         = generic_file_llseek,
9553         .read           = generic_read_dir,
9554         .iterate        = btrfs_real_readdir,
9555         .unlocked_ioctl = btrfs_ioctl,
9556 #ifdef CONFIG_COMPAT
9557         .compat_ioctl   = btrfs_ioctl,
9558 #endif
9559         .release        = btrfs_release_file,
9560         .fsync          = btrfs_sync_file,
9561 };
9562
9563 static struct extent_io_ops btrfs_extent_io_ops = {
9564         .fill_delalloc = run_delalloc_range,
9565         .submit_bio_hook = btrfs_submit_bio_hook,
9566         .merge_bio_hook = btrfs_merge_bio_hook,
9567         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
9568         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
9569         .writepage_start_hook = btrfs_writepage_start_hook,
9570         .set_bit_hook = btrfs_set_bit_hook,
9571         .clear_bit_hook = btrfs_clear_bit_hook,
9572         .merge_extent_hook = btrfs_merge_extent_hook,
9573         .split_extent_hook = btrfs_split_extent_hook,
9574 };
9575
9576 /*
9577  * btrfs doesn't support the bmap operation because swapfiles
9578  * use bmap to make a mapping of extents in the file.  They assume
9579  * these extents won't change over the life of the file and they
9580  * use the bmap result to do IO directly to the drive.
9581  *
9582  * the btrfs bmap call would return logical addresses that aren't
9583  * suitable for IO and they also will change frequently as COW
9584  * operations happen.  So, swapfile + btrfs == corruption.
9585  *
9586  * For now we're avoiding this by dropping bmap.
9587  */
9588 static const struct address_space_operations btrfs_aops = {
9589         .readpage       = btrfs_readpage,
9590         .writepage      = btrfs_writepage,
9591         .writepages     = btrfs_writepages,
9592         .readpages      = btrfs_readpages,
9593         .direct_IO      = btrfs_direct_IO,
9594         .invalidatepage = btrfs_invalidatepage,
9595         .releasepage    = btrfs_releasepage,
9596         .set_page_dirty = btrfs_set_page_dirty,
9597         .error_remove_page = generic_error_remove_page,
9598 };
9599
9600 static const struct address_space_operations btrfs_symlink_aops = {
9601         .readpage       = btrfs_readpage,
9602         .writepage      = btrfs_writepage,
9603         .invalidatepage = btrfs_invalidatepage,
9604         .releasepage    = btrfs_releasepage,
9605 };
9606
9607 static const struct inode_operations btrfs_file_inode_operations = {
9608         .getattr        = btrfs_getattr,
9609         .setattr        = btrfs_setattr,
9610         .setxattr       = btrfs_setxattr,
9611         .getxattr       = btrfs_getxattr,
9612         .listxattr      = btrfs_listxattr,
9613         .removexattr    = btrfs_removexattr,
9614         .permission     = btrfs_permission,
9615         .fiemap         = btrfs_fiemap,
9616         .get_acl        = btrfs_get_acl,
9617         .set_acl        = btrfs_set_acl,
9618         .update_time    = btrfs_update_time,
9619 };
9620 static const struct inode_operations btrfs_special_inode_operations = {
9621         .getattr        = btrfs_getattr,
9622         .setattr        = btrfs_setattr,
9623         .permission     = btrfs_permission,
9624         .setxattr       = btrfs_setxattr,
9625         .getxattr       = btrfs_getxattr,
9626         .listxattr      = btrfs_listxattr,
9627         .removexattr    = btrfs_removexattr,
9628         .get_acl        = btrfs_get_acl,
9629         .set_acl        = btrfs_set_acl,
9630         .update_time    = btrfs_update_time,
9631 };
9632 static const struct inode_operations btrfs_symlink_inode_operations = {
9633         .readlink       = generic_readlink,
9634         .follow_link    = page_follow_link_light,
9635         .put_link       = page_put_link,
9636         .getattr        = btrfs_getattr,
9637         .setattr        = btrfs_setattr,
9638         .permission     = btrfs_permission,
9639         .setxattr       = btrfs_setxattr,
9640         .getxattr       = btrfs_getxattr,
9641         .listxattr      = btrfs_listxattr,
9642         .removexattr    = btrfs_removexattr,
9643         .update_time    = btrfs_update_time,
9644 };
9645
9646 const struct dentry_operations btrfs_dentry_operations = {
9647         .d_delete       = btrfs_dentry_delete,
9648         .d_release      = btrfs_dentry_release,
9649 };