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