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/bit_spinlock.h>
  36 #include <linux/xattr.h>
  37 #include <linux/posix_acl.h>
  38 #include <linux/falloc.h>
  39 #include <linux/slab.h>
  40 #include <linux/ratelimit.h>
  41 #include <linux/mount.h>
  42 #include "compat.h"
  43 #include "ctree.h"
  44 #include "disk-io.h"
  45 #include "transaction.h"
  46 #include "btrfs_inode.h"
  47 #include "ioctl.h"
  48 #include "print-tree.h"
  49 #include "ordered-data.h"
  50 #include "xattr.h"
  51 #include "tree-log.h"
  52 #include "volumes.h"
  53 #include "compression.h"
  54 #include "locking.h"
  55 #include "free-space-cache.h"
  56 #include "inode-map.h"
  57
  58 struct btrfs_iget_args {
  59         u64 ino;
  60         struct btrfs_root *root;
  61 };
  62
  63 static const struct inode_operations btrfs_dir_inode_operations;
  64 static const struct inode_operations btrfs_symlink_inode_operations;
  65 static const struct inode_operations btrfs_dir_ro_inode_operations;
  66 static const struct inode_operations btrfs_special_inode_operations;
  67 static const struct inode_operations btrfs_file_inode_operations;
  68 static const struct address_space_operations btrfs_aops;
  69 static const struct address_space_operations btrfs_symlink_aops;
  70 static const struct file_operations btrfs_dir_file_operations;
  71 static struct extent_io_ops btrfs_extent_io_ops;
  72
  73 static struct kmem_cache *btrfs_inode_cachep;
  74 struct kmem_cache *btrfs_trans_handle_cachep;
  75 struct kmem_cache *btrfs_transaction_cachep;
  76 struct kmem_cache *btrfs_path_cachep;
  77 struct kmem_cache *btrfs_free_space_cachep;
  78
  79 #define S_SHIFT 12
  80 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  81         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  82         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  83         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  84         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  85         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  86         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  87         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  88 };
  89
  90 static int btrfs_setsize(struct inode *inode, loff_t newsize);
  91 static int btrfs_truncate(struct inode *inode);
  92 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
  93 static noinline int cow_file_range(struct inode *inode,
  94                                    struct page *locked_page,
  95                                    u64 start, u64 end, int *page_started,
  96                                    unsigned long *nr_written, int unlock);
  97 static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
  98                                 struct btrfs_root *root, struct inode *inode);
  99
 100 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
 101                                      struct inode *inode,  struct inode *dir,
 102                                      const struct qstr *qstr)
 103 {
 104         int err;
 105
 106         err = btrfs_init_acl(trans, inode, dir);
 107         if (!err)
 108                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 109         return err;
 110 }
 111
 112 /*
 113  * this does all the hard work for inserting an inline extent into
 114  * the btree.  The caller should have done a btrfs_drop_extents so that
 115  * no overlapping inline items exist in the btree
 116  */
 117 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 118                                 struct btrfs_root *root, struct inode *inode,
 119                                 u64 start, size_t size, size_t compressed_size,
 120                                 int compress_type,
 121                                 struct page **compressed_pages)
 122 {
 123         struct btrfs_key key;
 124         struct btrfs_path *path;
 125         struct extent_buffer *leaf;
 126         struct page *page = NULL;
 127         char *kaddr;
 128         unsigned long ptr;
 129         struct btrfs_file_extent_item *ei;
 130         int err = 0;
 131         int ret;
 132         size_t cur_size = size;
 133         size_t datasize;
 134         unsigned long offset;
 135
 136         if (compressed_size && compressed_pages)
 137                 cur_size = compressed_size;
 138
 139         path = btrfs_alloc_path();
 140         if (!path)
 141                 return -ENOMEM;
 142
 143         path->leave_spinning = 1;
 144
 145         key.objectid = btrfs_ino(inode);
 146         key.offset = start;
 147         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 148         datasize = btrfs_file_extent_calc_inline_size(cur_size);
 149
 150         inode_add_bytes(inode, size);
 151         ret = btrfs_insert_empty_item(trans, root, path, &key,
 152                                       datasize);
 153         if (ret) {
 154                 err = ret;
 155                 goto fail;
 156         }
 157         leaf = path->nodes[0];
 158         ei = btrfs_item_ptr(leaf, path->slots[0],
 159                             struct btrfs_file_extent_item);
 160         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 161         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 162         btrfs_set_file_extent_encryption(leaf, ei, 0);
 163         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 164         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 165         ptr = btrfs_file_extent_inline_start(ei);
 166
 167         if (compress_type != BTRFS_COMPRESS_NONE) {
 168                 struct page *cpage;
 169                 int i = 0;
 170                 while (compressed_size > 0) {
 171                         cpage = compressed_pages[i];
 172                         cur_size = min_t(unsigned long, compressed_size,
 173                                        PAGE_CACHE_SIZE);
 174
 175                         kaddr = kmap_atomic(cpage);
 176                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 177                         kunmap_atomic(kaddr);
 178
 179                         i++;
 180                         ptr += cur_size;
 181                         compressed_size -= cur_size;
 182                 }
 183                 btrfs_set_file_extent_compression(leaf, ei,
 184                                                   compress_type);
 185         } else {
 186                 page = find_get_page(inode->i_mapping,
 187                                      start >> PAGE_CACHE_SHIFT);
 188                 btrfs_set_file_extent_compression(leaf, ei, 0);
 189                 kaddr = kmap_atomic(page);
 190                 offset = start & (PAGE_CACHE_SIZE - 1);
 191                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 192                 kunmap_atomic(kaddr);
 193                 page_cache_release(page);
 194         }
 195         btrfs_mark_buffer_dirty(leaf);
 196         btrfs_free_path(path);
 197
 198         /*
 199          * we're an inline extent, so nobody can
 200          * extend the file past i_size without locking
 201          * a page we already have locked.
 202          *
 203          * We must do any isize and inode updates
 204          * before we unlock the pages.  Otherwise we
 205          * could end up racing with unlink.
 206          */
 207         BTRFS_I(inode)->disk_i_size = inode->i_size;
 208         ret = btrfs_update_inode(trans, root, inode);
 209
 210         return ret;
 211 fail:
 212         btrfs_free_path(path);
 213         return err;
 214 }
 215
 216
 217 /*
 218  * conditionally insert an inline extent into the file.  This
 219  * does the checks required to make sure the data is small enough
 220  * to fit as an inline extent.
 221  */
 222 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 223                                  struct btrfs_root *root,
 224                                  struct inode *inode, u64 start, u64 end,
 225                                  size_t compressed_size, int compress_type,
 226                                  struct page **compressed_pages)
 227 {
 228         u64 isize = i_size_read(inode);
 229         u64 actual_end = min(end + 1, isize);
 230         u64 inline_len = actual_end - start;
 231         u64 aligned_end = (end + root->sectorsize - 1) &
 232                         ~((u64)root->sectorsize - 1);
 233         u64 hint_byte;
 234         u64 data_len = inline_len;
 235         int ret;
 236
 237         if (compressed_size)
 238                 data_len = compressed_size;
 239
 240         if (start > 0 ||
 241             actual_end >= PAGE_CACHE_SIZE ||
 242             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 243             (!compressed_size &&
 244             (actual_end & (root->sectorsize - 1)) == 0) ||
 245             end + 1 < isize ||
 246             data_len > root->fs_info->max_inline) {
 247                 return 1;
 248         }
 249
 250         ret = btrfs_drop_extents(trans, inode, start, aligned_end,
 251                                  &hint_byte, 1);
 252         if (ret)
 253                 return ret;
 254
 255         if (isize > actual_end)
 256                 inline_len = min_t(u64, isize, actual_end);
 257         ret = insert_inline_extent(trans, root, inode, start,
 258                                    inline_len, compressed_size,
 259                                    compress_type, compressed_pages);
 260         if (ret && ret != -ENOSPC) {
 261                 btrfs_abort_transaction(trans, root, ret);
 262                 return ret;
 263         } else if (ret == -ENOSPC) {
 264                 return 1;
 265         }
 266
 267         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 268         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 269         return 0;
 270 }
 271
 272 struct async_extent {
 273         u64 start;
 274         u64 ram_size;
 275         u64 compressed_size;
 276         struct page **pages;
 277         unsigned long nr_pages;
 278         int compress_type;
 279         struct list_head list;
 280 };
 281
 282 struct async_cow {
 283         struct inode *inode;
 284         struct btrfs_root *root;
 285         struct page *locked_page;
 286         u64 start;
 287         u64 end;
 288         struct list_head extents;
 289         struct btrfs_work work;
 290 };
 291
 292 static noinline int add_async_extent(struct async_cow *cow,
 293                                      u64 start, u64 ram_size,
 294                                      u64 compressed_size,
 295                                      struct page **pages,
 296                                      unsigned long nr_pages,
 297                                      int compress_type)
 298 {
 299         struct async_extent *async_extent;
 300
 301         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 302         BUG_ON(!async_extent); /* -ENOMEM */
 303         async_extent->start = start;
 304         async_extent->ram_size = ram_size;
 305         async_extent->compressed_size = compressed_size;
 306         async_extent->pages = pages;
 307         async_extent->nr_pages = nr_pages;
 308         async_extent->compress_type = compress_type;
 309         list_add_tail(&async_extent->list, &cow->extents);
 310         return 0;
 311 }
 312
 313 /*
 314  * we create compressed extents in two phases.  The first
 315  * phase compresses a range of pages that have already been
 316  * locked (both pages and state bits are locked).
 317  *
 318  * This is done inside an ordered work queue, and the compression
 319  * is spread across many cpus.  The actual IO submission is step
 320  * two, and the ordered work queue takes care of making sure that
 321  * happens in the same order things were put onto the queue by
 322  * writepages and friends.
 323  *
 324  * If this code finds it can't get good compression, it puts an
 325  * entry onto the work queue to write the uncompressed bytes.  This
 326  * makes sure that both compressed inodes and uncompressed inodes
 327  * are written in the same order that pdflush sent them down.
 328  */
 329 static noinline int compress_file_range(struct inode *inode,
 330                                         struct page *locked_page,
 331                                         u64 start, u64 end,
 332                                         struct async_cow *async_cow,
 333                                         int *num_added)
 334 {
 335         struct btrfs_root *root = BTRFS_I(inode)->root;
 336         struct btrfs_trans_handle *trans;
 337         u64 num_bytes;
 338         u64 blocksize = root->sectorsize;
 339         u64 actual_end;
 340         u64 isize = i_size_read(inode);
 341         int ret = 0;
 342         struct page **pages = NULL;
 343         unsigned long nr_pages;
 344         unsigned long nr_pages_ret = 0;
 345         unsigned long total_compressed = 0;
 346         unsigned long total_in = 0;
 347         unsigned long max_compressed = 128 * 1024;
 348         unsigned long max_uncompressed = 128 * 1024;
 349         int i;
 350         int will_compress;
 351         int compress_type = root->fs_info->compress_type;
 352
 353         /* if this is a small write inside eof, kick off a defrag */
 354         if ((end - start + 1) < 16 * 1024 &&
 355             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 356                 btrfs_add_inode_defrag(NULL, inode);
 357
 358         actual_end = min_t(u64, isize, end + 1);
 359 again:
 360         will_compress = 0;
 361         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 362         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 363
 364         /*
 365          * we don't want to send crud past the end of i_size through
 366          * compression, that's just a waste of CPU time.  So, if the
 367          * end of the file is before the start of our current
 368          * requested range of bytes, we bail out to the uncompressed
 369          * cleanup code that can deal with all of this.
 370          *
 371          * It isn't really the fastest way to fix things, but this is a
 372          * very uncommon corner.
 373          */
 374         if (actual_end <= start)
 375                 goto cleanup_and_bail_uncompressed;
 376
 377         total_compressed = actual_end - start;
 378
 379         /* we want to make sure that amount of ram required to uncompress
 380          * an extent is reasonable, so we limit the total size in ram
 381          * of a compressed extent to 128k.  This is a crucial number
 382          * because it also controls how easily we can spread reads across
 383          * cpus for decompression.
 384          *
 385          * We also want to make sure the amount of IO required to do
 386          * a random read is reasonably small, so we limit the size of
 387          * a compressed extent to 128k.
 388          */
 389         total_compressed = min(total_compressed, max_uncompressed);
 390         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 391         num_bytes = max(blocksize,  num_bytes);
 392         total_in = 0;
 393         ret = 0;
 394
 395         /*
 396          * we do compression for mount -o compress and when the
 397          * inode has not been flagged as nocompress.  This flag can
 398          * change at any time if we discover bad compression ratios.
 399          */
 400         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 401             (btrfs_test_opt(root, COMPRESS) ||
 402              (BTRFS_I(inode)->force_compress) ||
 403              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 404                 WARN_ON(pages);
 405                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 406                 if (!pages) {
 407                         /* just bail out to the uncompressed code */
 408                         goto cont;
 409                 }
 410
 411                 if (BTRFS_I(inode)->force_compress)
 412                         compress_type = BTRFS_I(inode)->force_compress;
 413
 414                 ret = btrfs_compress_pages(compress_type,
 415                                            inode->i_mapping, start,
 416                                            total_compressed, pages,
 417                                            nr_pages, &nr_pages_ret,
 418                                            &total_in,
 419                                            &total_compressed,
 420                                            max_compressed);
 421
 422                 if (!ret) {
 423                         unsigned long offset = total_compressed &
 424                                 (PAGE_CACHE_SIZE - 1);
 425                         struct page *page = pages[nr_pages_ret - 1];
 426                         char *kaddr;
 427
 428                         /* zero the tail end of the last page, we might be
 429                          * sending it down to disk
 430                          */
 431                         if (offset) {
 432                                 kaddr = kmap_atomic(page);
 433                                 memset(kaddr + offset, 0,
 434                                        PAGE_CACHE_SIZE - offset);
 435                                 kunmap_atomic(kaddr);
 436                         }
 437                         will_compress = 1;
 438                 }
 439         }
 440 cont:
 441         if (start == 0) {
 442                 trans = btrfs_join_transaction(root);
 443                 if (IS_ERR(trans)) {
 444                         ret = PTR_ERR(trans);
 445                         trans = NULL;
 446                         goto cleanup_and_out;
 447                 }
 448                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 449
 450                 /* lets try to make an inline extent */
 451                 if (ret || total_in < (actual_end - start)) {
 452                         /* we didn't compress the entire range, try
 453                          * to make an uncompressed inline extent.
 454                          */
 455                         ret = cow_file_range_inline(trans, root, inode,
 456                                                     start, end, 0, 0, NULL);
 457                 } else {
 458                         /* try making a compressed inline extent */
 459                         ret = cow_file_range_inline(trans, root, inode,
 460                                                     start, end,
 461                                                     total_compressed,
 462                                                     compress_type, pages);
 463                 }
 464                 if (ret <= 0) {
 465                         /*
 466                          * inline extent creation worked or returned error,
 467                          * we don't need to create any more async work items.
 468                          * Unlock and free up our temp pages.
 469                          */
 470                         extent_clear_unlock_delalloc(inode,
 471                              &BTRFS_I(inode)->io_tree,
 472                              start, end, NULL,
 473                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
 474                              EXTENT_CLEAR_DELALLOC |
 475                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
 476
 477                         btrfs_end_transaction(trans, root);
 478                         goto free_pages_out;
 479                 }
 480                 btrfs_end_transaction(trans, root);
 481         }
 482
 483         if (will_compress) {
 484                 /*
 485                  * we aren't doing an inline extent round the compressed size
 486                  * up to a block size boundary so the allocator does sane
 487                  * things
 488                  */
 489                 total_compressed = (total_compressed + blocksize - 1) &
 490                         ~(blocksize - 1);
 491
 492                 /*
 493                  * one last check to make sure the compression is really a
 494                  * win, compare the page count read with the blocks on disk
 495                  */
 496                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
 497                         ~(PAGE_CACHE_SIZE - 1);
 498                 if (total_compressed >= total_in) {
 499                         will_compress = 0;
 500                 } else {
 501                         num_bytes = total_in;
 502                 }
 503         }
 504         if (!will_compress && pages) {
 505                 /*
 506                  * the compression code ran but failed to make things smaller,
 507                  * free any pages it allocated and our page pointer array
 508                  */
 509                 for (i = 0; i < nr_pages_ret; i++) {
 510                         WARN_ON(pages[i]->mapping);
 511                         page_cache_release(pages[i]);
 512                 }
 513                 kfree(pages);
 514                 pages = NULL;
 515                 total_compressed = 0;
 516                 nr_pages_ret = 0;
 517
 518                 /* flag the file so we don't compress in the future */
 519                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 520                     !(BTRFS_I(inode)->force_compress)) {
 521                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 522                 }
 523         }
 524         if (will_compress) {
 525                 *num_added += 1;
 526
 527                 /* the async work queues will take care of doing actual
 528                  * allocation on disk for these compressed pages,
 529                  * and will submit them to the elevator.
 530                  */
 531                 add_async_extent(async_cow, start, num_bytes,
 532                                  total_compressed, pages, nr_pages_ret,
 533                                  compress_type);
 534
 535                 if (start + num_bytes < end) {
 536                         start += num_bytes;
 537                         pages = NULL;
 538                         cond_resched();
 539                         goto again;
 540                 }
 541         } else {
 542 cleanup_and_bail_uncompressed:
 543                 /*
 544                  * No compression, but we still need to write the pages in
 545                  * the file we've been given so far.  redirty the locked
 546                  * page if it corresponds to our extent and set things up
 547                  * for the async work queue to run cow_file_range to do
 548                  * the normal delalloc dance
 549                  */
 550                 if (page_offset(locked_page) >= start &&
 551                     page_offset(locked_page) <= end) {
 552                         __set_page_dirty_nobuffers(locked_page);
 553                         /* unlocked later on in the async handlers */
 554                 }
 555                 add_async_extent(async_cow, start, end - start + 1,
 556                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 557                 *num_added += 1;
 558         }
 559
 560 out:
 561         return ret;
 562
 563 free_pages_out:
 564         for (i = 0; i < nr_pages_ret; i++) {
 565                 WARN_ON(pages[i]->mapping);
 566                 page_cache_release(pages[i]);
 567         }
 568         kfree(pages);
 569
 570         goto out;
 571
 572 cleanup_and_out:
 573         extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 574                                      start, end, NULL,
 575                                      EXTENT_CLEAR_UNLOCK_PAGE |
 576                                      EXTENT_CLEAR_DIRTY |
 577                                      EXTENT_CLEAR_DELALLOC |
 578                                      EXTENT_SET_WRITEBACK |
 579                                      EXTENT_END_WRITEBACK);
 580         if (!trans || IS_ERR(trans))
 581                 btrfs_error(root->fs_info, ret, "Failed to join transaction");
 582         else
 583                 btrfs_abort_transaction(trans, root, ret);
 584         goto free_pages_out;
 585 }
 586
 587 /*
 588  * phase two of compressed writeback.  This is the ordered portion
 589  * of the code, which only gets called in the order the work was
 590  * queued.  We walk all the async extents created by compress_file_range
 591  * and send them down to the disk.
 592  */
 593 static noinline int submit_compressed_extents(struct inode *inode,
 594                                               struct async_cow *async_cow)
 595 {
 596         struct async_extent *async_extent;
 597         u64 alloc_hint = 0;
 598         struct btrfs_trans_handle *trans;
 599         struct btrfs_key ins;
 600         struct extent_map *em;
 601         struct btrfs_root *root = BTRFS_I(inode)->root;
 602         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 603         struct extent_io_tree *io_tree;
 604         int ret = 0;
 605
 606         if (list_empty(&async_cow->extents))
 607                 return 0;
 608
 609
 610         while (!list_empty(&async_cow->extents)) {
 611                 async_extent = list_entry(async_cow->extents.next,
 612                                           struct async_extent, list);
 613                 list_del(&async_extent->list);
 614
 615                 io_tree = &BTRFS_I(inode)->io_tree;
 616
 617 retry:
 618                 /* did the compression code fall back to uncompressed IO? */
 619                 if (!async_extent->pages) {
 620                         int page_started = 0;
 621                         unsigned long nr_written = 0;
 622
 623                         lock_extent(io_tree, async_extent->start,
 624                                          async_extent->start +
 625                                          async_extent->ram_size - 1);
 626
 627                         /* allocate blocks */
 628                         ret = cow_file_range(inode, async_cow->locked_page,
 629                                              async_extent->start,
 630                                              async_extent->start +
 631                                              async_extent->ram_size - 1,
 632                                              &page_started, &nr_written, 0);
 633
 634                         /* JDM XXX */
 635
 636                         /*
 637                          * if page_started, cow_file_range inserted an
 638                          * inline extent and took care of all the unlocking
 639                          * and IO for us.  Otherwise, we need to submit
 640                          * all those pages down to the drive.
 641                          */
 642                         if (!page_started && !ret)
 643                                 extent_write_locked_range(io_tree,
 644                                                   inode, async_extent->start,
 645                                                   async_extent->start +
 646                                                   async_extent->ram_size - 1,
 647                                                   btrfs_get_extent,
 648                                                   WB_SYNC_ALL);
 649                         kfree(async_extent);
 650                         cond_resched();
 651                         continue;
 652                 }
 653
 654                 lock_extent(io_tree, async_extent->start,
 655                             async_extent->start + async_extent->ram_size - 1);
 656
 657                 trans = btrfs_join_transaction(root);
 658                 if (IS_ERR(trans)) {
 659                         ret = PTR_ERR(trans);
 660                 } else {
 661                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 662                         ret = btrfs_reserve_extent(trans, root,
 663                                            async_extent->compressed_size,
 664                                            async_extent->compressed_size,
 665                                            0, alloc_hint, &ins, 1);
 666                         if (ret)
 667                                 btrfs_abort_transaction(trans, root, ret);
 668                         btrfs_end_transaction(trans, root);
 669                 }
 670
 671                 if (ret) {
 672                         int i;
 673                         for (i = 0; i < async_extent->nr_pages; i++) {
 674                                 WARN_ON(async_extent->pages[i]->mapping);
 675                                 page_cache_release(async_extent->pages[i]);
 676                         }
 677                         kfree(async_extent->pages);
 678                         async_extent->nr_pages = 0;
 679                         async_extent->pages = NULL;
 680                         unlock_extent(io_tree, async_extent->start,
 681                                       async_extent->start +
 682                                       async_extent->ram_size - 1);
 683                         if (ret == -ENOSPC)
 684                                 goto retry;
 685                         goto out_free; /* JDM: Requeue? */
 686                 }
 687
 688                 /*
 689                  * here we're doing allocation and writeback of the
 690                  * compressed pages
 691                  */
 692                 btrfs_drop_extent_cache(inode, async_extent->start,
 693                                         async_extent->start +
 694                                         async_extent->ram_size - 1, 0);
 695
 696                 em = alloc_extent_map();
 697                 BUG_ON(!em); /* -ENOMEM */
 698                 em->start = async_extent->start;
 699                 em->len = async_extent->ram_size;
 700                 em->orig_start = em->start;
 701
 702                 em->block_start = ins.objectid;
 703                 em->block_len = ins.offset;
 704                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 705                 em->compress_type = async_extent->compress_type;
 706                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 707                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 708
 709                 while (1) {
 710                         write_lock(&em_tree->lock);
 711                         ret = add_extent_mapping(em_tree, em);
 712                         write_unlock(&em_tree->lock);
 713                         if (ret != -EEXIST) {
 714                                 free_extent_map(em);
 715                                 break;
 716                         }
 717                         btrfs_drop_extent_cache(inode, async_extent->start,
 718                                                 async_extent->start +
 719                                                 async_extent->ram_size - 1, 0);
 720                 }
 721
 722                 ret = btrfs_add_ordered_extent_compress(inode,
 723                                                 async_extent->start,
 724                                                 ins.objectid,
 725                                                 async_extent->ram_size,
 726                                                 ins.offset,
 727                                                 BTRFS_ORDERED_COMPRESSED,
 728                                                 async_extent->compress_type);
 729                 BUG_ON(ret); /* -ENOMEM */
 730
 731                 /*
 732                  * clear dirty, set writeback and unlock the pages.
 733                  */
 734                 extent_clear_unlock_delalloc(inode,
 735                                 &BTRFS_I(inode)->io_tree,
 736                                 async_extent->start,
 737                                 async_extent->start +
 738                                 async_extent->ram_size - 1,
 739                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 740                                 EXTENT_CLEAR_UNLOCK |
 741                                 EXTENT_CLEAR_DELALLOC |
 742                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
 743
 744                 ret = btrfs_submit_compressed_write(inode,
 745                                     async_extent->start,
 746                                     async_extent->ram_size,
 747                                     ins.objectid,
 748                                     ins.offset, async_extent->pages,
 749                                     async_extent->nr_pages);
 750
 751                 BUG_ON(ret); /* -ENOMEM */
 752                 alloc_hint = ins.objectid + ins.offset;
 753                 kfree(async_extent);
 754                 cond_resched();
 755         }
 756         ret = 0;
 757 out:
 758         return ret;
 759 out_free:
 760         kfree(async_extent);
 761         goto out;
 762 }
 763
 764 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 765                                       u64 num_bytes)
 766 {
 767         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 768         struct extent_map *em;
 769         u64 alloc_hint = 0;
 770
 771         read_lock(&em_tree->lock);
 772         em = search_extent_mapping(em_tree, start, num_bytes);
 773         if (em) {
 774                 /*
 775                  * if block start isn't an actual block number then find the
 776                  * first block in this inode and use that as a hint.  If that
 777                  * block is also bogus then just don't worry about it.
 778                  */
 779                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 780                         free_extent_map(em);
 781                         em = search_extent_mapping(em_tree, 0, 0);
 782                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 783                                 alloc_hint = em->block_start;
 784                         if (em)
 785                                 free_extent_map(em);
 786                 } else {
 787                         alloc_hint = em->block_start;
 788                         free_extent_map(em);
 789                 }
 790         }
 791         read_unlock(&em_tree->lock);
 792
 793         return alloc_hint;
 794 }
 795
 796 /*
 797  * when extent_io.c finds a delayed allocation range in the file,
 798  * the call backs end up in this code.  The basic idea is to
 799  * allocate extents on disk for the range, and create ordered data structs
 800  * in ram to track those extents.
 801  *
 802  * locked_page is the page that writepage had locked already.  We use
 803  * it to make sure we don't do extra locks or unlocks.
 804  *
 805  * *page_started is set to one if we unlock locked_page and do everything
 806  * required to start IO on it.  It may be clean and already done with
 807  * IO when we return.
 808  */
 809 static noinline int cow_file_range(struct inode *inode,
 810                                    struct page *locked_page,
 811                                    u64 start, u64 end, int *page_started,
 812                                    unsigned long *nr_written,
 813                                    int unlock)
 814 {
 815         struct btrfs_root *root = BTRFS_I(inode)->root;
 816         struct btrfs_trans_handle *trans;
 817         u64 alloc_hint = 0;
 818         u64 num_bytes;
 819         unsigned long ram_size;
 820         u64 disk_num_bytes;
 821         u64 cur_alloc_size;
 822         u64 blocksize = root->sectorsize;
 823         struct btrfs_key ins;
 824         struct extent_map *em;
 825         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 826         int ret = 0;
 827
 828         BUG_ON(btrfs_is_free_space_inode(root, inode));
 829         trans = btrfs_join_transaction(root);
 830         if (IS_ERR(trans)) {
 831                 extent_clear_unlock_delalloc(inode,
 832                              &BTRFS_I(inode)->io_tree,
 833                              start, end, NULL,
 834                              EXTENT_CLEAR_UNLOCK_PAGE |
 835                              EXTENT_CLEAR_UNLOCK |
 836                              EXTENT_CLEAR_DELALLOC |
 837                              EXTENT_CLEAR_DIRTY |
 838                              EXTENT_SET_WRITEBACK |
 839                              EXTENT_END_WRITEBACK);
 840                 return PTR_ERR(trans);
 841         }
 842         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 843
 844         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 845         num_bytes = max(blocksize,  num_bytes);
 846         disk_num_bytes = num_bytes;
 847         ret = 0;
 848
 849         /* if this is a small write inside eof, kick off defrag */
 850         if (num_bytes < 64 * 1024 &&
 851             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 852                 btrfs_add_inode_defrag(trans, inode);
 853
 854         if (start == 0) {
 855                 /* lets try to make an inline extent */
 856                 ret = cow_file_range_inline(trans, root, inode,
 857                                             start, end, 0, 0, NULL);
 858                 if (ret == 0) {
 859                         extent_clear_unlock_delalloc(inode,
 860                                      &BTRFS_I(inode)->io_tree,
 861                                      start, end, NULL,
 862                                      EXTENT_CLEAR_UNLOCK_PAGE |
 863                                      EXTENT_CLEAR_UNLOCK |
 864                                      EXTENT_CLEAR_DELALLOC |
 865                                      EXTENT_CLEAR_DIRTY |
 866                                      EXTENT_SET_WRITEBACK |
 867                                      EXTENT_END_WRITEBACK);
 868
 869                         *nr_written = *nr_written +
 870                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 871                         *page_started = 1;
 872                         goto out;
 873                 } else if (ret < 0) {
 874                         btrfs_abort_transaction(trans, root, ret);
 875                         goto out_unlock;
 876                 }
 877         }
 878
 879         BUG_ON(disk_num_bytes >
 880                btrfs_super_total_bytes(root->fs_info->super_copy));
 881
 882         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 883         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 884
 885         while (disk_num_bytes > 0) {
 886                 unsigned long op;
 887
 888                 cur_alloc_size = disk_num_bytes;
 889                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 890                                            root->sectorsize, 0, alloc_hint,
 891                                            &ins, 1);
 892                 if (ret < 0) {
 893                         btrfs_abort_transaction(trans, root, ret);
 894                         goto out_unlock;
 895                 }
 896
 897                 em = alloc_extent_map();
 898                 BUG_ON(!em); /* -ENOMEM */
 899                 em->start = start;
 900                 em->orig_start = em->start;
 901                 ram_size = ins.offset;
 902                 em->len = ins.offset;
 903
 904                 em->block_start = ins.objectid;
 905                 em->block_len = ins.offset;
 906                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 907                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 908
 909                 while (1) {
 910                         write_lock(&em_tree->lock);
 911                         ret = add_extent_mapping(em_tree, em);
 912                         write_unlock(&em_tree->lock);
 913                         if (ret != -EEXIST) {
 914                                 free_extent_map(em);
 915                                 break;
 916                         }
 917                         btrfs_drop_extent_cache(inode, start,
 918                                                 start + ram_size - 1, 0);
 919                 }
 920
 921                 cur_alloc_size = ins.offset;
 922                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 923                                                ram_size, cur_alloc_size, 0);
 924                 BUG_ON(ret); /* -ENOMEM */
 925
 926                 if (root->root_key.objectid ==
 927                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 928                         ret = btrfs_reloc_clone_csums(inode, start,
 929                                                       cur_alloc_size);
 930                         if (ret) {
 931                                 btrfs_abort_transaction(trans, root, ret);
 932                                 goto out_unlock;
 933                         }
 934                 }
 935
 936                 if (disk_num_bytes < cur_alloc_size)
 937                         break;
 938
 939                 /* we're not doing compressed IO, don't unlock the first
 940                  * page (which the caller expects to stay locked), don't
 941                  * clear any dirty bits and don't set any writeback bits
 942                  *
 943                  * Do set the Private2 bit so we know this page was properly
 944                  * setup for writepage
 945                  */
 946                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
 947                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
 948                         EXTENT_SET_PRIVATE2;
 949
 950                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 951                                              start, start + ram_size - 1,
 952                                              locked_page, op);
 953                 disk_num_bytes -= cur_alloc_size;
 954                 num_bytes -= cur_alloc_size;
 955                 alloc_hint = ins.objectid + ins.offset;
 956                 start += cur_alloc_size;
 957         }
 958         ret = 0;
 959 out:
 960         btrfs_end_transaction(trans, root);
 961
 962         return ret;
 963 out_unlock:
 964         extent_clear_unlock_delalloc(inode,
 965                      &BTRFS_I(inode)->io_tree,
 966                      start, end, NULL,
 967                      EXTENT_CLEAR_UNLOCK_PAGE |
 968                      EXTENT_CLEAR_UNLOCK |
 969                      EXTENT_CLEAR_DELALLOC |
 970                      EXTENT_CLEAR_DIRTY |
 971                      EXTENT_SET_WRITEBACK |
 972                      EXTENT_END_WRITEBACK);
 973
 974         goto out;
 975 }
 976
 977 /*
 978  * work queue call back to started compression on a file and pages
 979  */
 980 static noinline void async_cow_start(struct btrfs_work *work)
 981 {
 982         struct async_cow *async_cow;
 983         int num_added = 0;
 984         async_cow = container_of(work, struct async_cow, work);
 985
 986         compress_file_range(async_cow->inode, async_cow->locked_page,
 987                             async_cow->start, async_cow->end, async_cow,
 988                             &num_added);
 989         if (num_added == 0)
 990                 async_cow->inode = NULL;
 991 }
 992
 993 /*
 994  * work queue call back to submit previously compressed pages
 995  */
 996 static noinline void async_cow_submit(struct btrfs_work *work)
 997 {
 998         struct async_cow *async_cow;
 999         struct btrfs_root *root;
1000         unsigned long nr_pages;
1001
1002         async_cow = container_of(work, struct async_cow, work);
1003
1004         root = async_cow->root;
1005         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1006                 PAGE_CACHE_SHIFT;
1007
1008         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
1009
1010         if (atomic_read(&root->fs_info->async_delalloc_pages) <
1011             5 * 1042 * 1024 &&
1012             waitqueue_active(&root->fs_info->async_submit_wait))
1013                 wake_up(&root->fs_info->async_submit_wait);
1014
1015         if (async_cow->inode)
1016                 submit_compressed_extents(async_cow->inode, async_cow);
1017 }
1018
1019 static noinline void async_cow_free(struct btrfs_work *work)
1020 {
1021         struct async_cow *async_cow;
1022         async_cow = container_of(work, struct async_cow, work);
1023         kfree(async_cow);
1024 }
1025
1026 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1027                                 u64 start, u64 end, int *page_started,
1028                                 unsigned long *nr_written)
1029 {
1030         struct async_cow *async_cow;
1031         struct btrfs_root *root = BTRFS_I(inode)->root;
1032         unsigned long nr_pages;
1033         u64 cur_end;
1034         int limit = 10 * 1024 * 1042;
1035
1036         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1037                          1, 0, NULL, GFP_NOFS);
1038         while (start < end) {
1039                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1040                 BUG_ON(!async_cow); /* -ENOMEM */
1041                 async_cow->inode = inode;
1042                 async_cow->root = root;
1043                 async_cow->locked_page = locked_page;
1044                 async_cow->start = start;
1045
1046                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1047                         cur_end = end;
1048                 else
1049                         cur_end = min(end, start + 512 * 1024 - 1);
1050
1051                 async_cow->end = cur_end;
1052                 INIT_LIST_HEAD(&async_cow->extents);
1053
1054                 async_cow->work.func = async_cow_start;
1055                 async_cow->work.ordered_func = async_cow_submit;
1056                 async_cow->work.ordered_free = async_cow_free;
1057                 async_cow->work.flags = 0;
1058
1059                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1060                         PAGE_CACHE_SHIFT;
1061                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1062
1063                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1064                                    &async_cow->work);
1065
1066                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1067                         wait_event(root->fs_info->async_submit_wait,
1068                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1069                             limit));
1070                 }
1071
1072                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1073                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1074                         wait_event(root->fs_info->async_submit_wait,
1075                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1076                            0));
1077                 }
1078
1079                 *nr_written += nr_pages;
1080                 start = cur_end + 1;
1081         }
1082         *page_started = 1;
1083         return 0;
1084 }
1085
1086 static noinline int csum_exist_in_range(struct btrfs_root *root,
1087                                         u64 bytenr, u64 num_bytes)
1088 {
1089         int ret;
1090         struct btrfs_ordered_sum *sums;
1091         LIST_HEAD(list);
1092
1093         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1094                                        bytenr + num_bytes - 1, &list, 0);
1095         if (ret == 0 && list_empty(&list))
1096                 return 0;
1097
1098         while (!list_empty(&list)) {
1099                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1100                 list_del(&sums->list);
1101                 kfree(sums);
1102         }
1103         return 1;
1104 }
1105
1106 /*
1107  * when nowcow writeback call back.  This checks for snapshots or COW copies
1108  * of the extents that exist in the file, and COWs the file as required.
1109  *
1110  * If no cow copies or snapshots exist, we write directly to the existing
1111  * blocks on disk
1112  */
1113 static noinline int run_delalloc_nocow(struct inode *inode,
1114                                        struct page *locked_page,
1115                               u64 start, u64 end, int *page_started, int force,
1116                               unsigned long *nr_written)
1117 {
1118         struct btrfs_root *root = BTRFS_I(inode)->root;
1119         struct btrfs_trans_handle *trans;
1120         struct extent_buffer *leaf;
1121         struct btrfs_path *path;
1122         struct btrfs_file_extent_item *fi;
1123         struct btrfs_key found_key;
1124         u64 cow_start;
1125         u64 cur_offset;
1126         u64 extent_end;
1127         u64 extent_offset;
1128         u64 disk_bytenr;
1129         u64 num_bytes;
1130         int extent_type;
1131         int ret, err;
1132         int type;
1133         int nocow;
1134         int check_prev = 1;
1135         bool nolock;
1136         u64 ino = btrfs_ino(inode);
1137
1138         path = btrfs_alloc_path();
1139         if (!path)
1140                 return -ENOMEM;
1141
1142         nolock = btrfs_is_free_space_inode(root, inode);
1143
1144         if (nolock)
1145                 trans = btrfs_join_transaction_nolock(root);
1146         else
1147                 trans = btrfs_join_transaction(root);
1148
1149         if (IS_ERR(trans)) {
1150                 btrfs_free_path(path);
1151                 return PTR_ERR(trans);
1152         }
1153
1154         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1155
1156         cow_start = (u64)-1;
1157         cur_offset = start;
1158         while (1) {
1159                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1160                                                cur_offset, 0);
1161                 if (ret < 0) {
1162                         btrfs_abort_transaction(trans, root, ret);
1163                         goto error;
1164                 }
1165                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1166                         leaf = path->nodes[0];
1167                         btrfs_item_key_to_cpu(leaf, &found_key,
1168                                               path->slots[0] - 1);
1169                         if (found_key.objectid == ino &&
1170                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1171                                 path->slots[0]--;
1172                 }
1173                 check_prev = 0;
1174 next_slot:
1175                 leaf = path->nodes[0];
1176                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1177                         ret = btrfs_next_leaf(root, path);
1178                         if (ret < 0) {
1179                                 btrfs_abort_transaction(trans, root, ret);
1180                                 goto error;
1181                         }
1182                         if (ret > 0)
1183                                 break;
1184                         leaf = path->nodes[0];
1185                 }
1186
1187                 nocow = 0;
1188                 disk_bytenr = 0;
1189                 num_bytes = 0;
1190                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1191
1192                 if (found_key.objectid > ino ||
1193                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1194                     found_key.offset > end)
1195                         break;
1196
1197                 if (found_key.offset > cur_offset) {
1198                         extent_end = found_key.offset;
1199                         extent_type = 0;
1200                         goto out_check;
1201                 }
1202
1203                 fi = btrfs_item_ptr(leaf, path->slots[0],
1204                                     struct btrfs_file_extent_item);
1205                 extent_type = btrfs_file_extent_type(leaf, fi);
1206
1207                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1208                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1209                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1210                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1211                         extent_end = found_key.offset +
1212                                 btrfs_file_extent_num_bytes(leaf, fi);
1213                         if (extent_end <= start) {
1214                                 path->slots[0]++;
1215                                 goto next_slot;
1216                         }
1217                         if (disk_bytenr == 0)
1218                                 goto out_check;
1219                         if (btrfs_file_extent_compression(leaf, fi) ||
1220                             btrfs_file_extent_encryption(leaf, fi) ||
1221                             btrfs_file_extent_other_encoding(leaf, fi))
1222                                 goto out_check;
1223                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1224                                 goto out_check;
1225                         if (btrfs_extent_readonly(root, disk_bytenr))
1226                                 goto out_check;
1227                         if (btrfs_cross_ref_exist(trans, root, ino,
1228                                                   found_key.offset -
1229                                                   extent_offset, disk_bytenr))
1230                                 goto out_check;
1231                         disk_bytenr += extent_offset;
1232                         disk_bytenr += cur_offset - found_key.offset;
1233                         num_bytes = min(end + 1, extent_end) - cur_offset;
1234                         /*
1235                          * force cow if csum exists in the range.
1236                          * this ensure that csum for a given extent are
1237                          * either valid or do not exist.
1238                          */
1239                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1240                                 goto out_check;
1241                         nocow = 1;
1242                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1243                         extent_end = found_key.offset +
1244                                 btrfs_file_extent_inline_len(leaf, fi);
1245                         extent_end = ALIGN(extent_end, root->sectorsize);
1246                 } else {
1247                         BUG_ON(1);
1248                 }
1249 out_check:
1250                 if (extent_end <= start) {
1251                         path->slots[0]++;
1252                         goto next_slot;
1253                 }
1254                 if (!nocow) {
1255                         if (cow_start == (u64)-1)
1256                                 cow_start = cur_offset;
1257                         cur_offset = extent_end;
1258                         if (cur_offset > end)
1259                                 break;
1260                         path->slots[0]++;
1261                         goto next_slot;
1262                 }
1263
1264                 btrfs_release_path(path);
1265                 if (cow_start != (u64)-1) {
1266                         ret = cow_file_range(inode, locked_page, cow_start,
1267                                         found_key.offset - 1, page_started,
1268                                         nr_written, 1);
1269                         if (ret) {
1270                                 btrfs_abort_transaction(trans, root, ret);
1271                                 goto error;
1272                         }
1273                         cow_start = (u64)-1;
1274                 }
1275
1276                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1277                         struct extent_map *em;
1278                         struct extent_map_tree *em_tree;
1279                         em_tree = &BTRFS_I(inode)->extent_tree;
1280                         em = alloc_extent_map();
1281                         BUG_ON(!em); /* -ENOMEM */
1282                         em->start = cur_offset;
1283                         em->orig_start = em->start;
1284                         em->len = num_bytes;
1285                         em->block_len = num_bytes;
1286                         em->block_start = disk_bytenr;
1287                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1288                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1289                         while (1) {
1290                                 write_lock(&em_tree->lock);
1291                                 ret = add_extent_mapping(em_tree, em);
1292                                 write_unlock(&em_tree->lock);
1293                                 if (ret != -EEXIST) {
1294                                         free_extent_map(em);
1295                                         break;
1296                                 }
1297                                 btrfs_drop_extent_cache(inode, em->start,
1298                                                 em->start + em->len - 1, 0);
1299                         }
1300                         type = BTRFS_ORDERED_PREALLOC;
1301                 } else {
1302                         type = BTRFS_ORDERED_NOCOW;
1303                 }
1304
1305                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1306                                                num_bytes, num_bytes, type);
1307                 BUG_ON(ret); /* -ENOMEM */
1308
1309                 if (root->root_key.objectid ==
1310                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1311                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1312                                                       num_bytes);
1313                         if (ret) {
1314                                 btrfs_abort_transaction(trans, root, ret);
1315                                 goto error;
1316                         }
1317                 }
1318
1319                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1320                                 cur_offset, cur_offset + num_bytes - 1,
1321                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1322                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1323                                 EXTENT_SET_PRIVATE2);
1324                 cur_offset = extent_end;
1325                 if (cur_offset > end)
1326                         break;
1327         }
1328         btrfs_release_path(path);
1329
1330         if (cur_offset <= end && cow_start == (u64)-1)
1331                 cow_start = cur_offset;
1332         if (cow_start != (u64)-1) {
1333                 ret = cow_file_range(inode, locked_page, cow_start, end,
1334                                      page_started, nr_written, 1);
1335                 if (ret) {
1336                         btrfs_abort_transaction(trans, root, ret);
1337                         goto error;
1338                 }
1339         }
1340
1341 error:
1342         if (nolock) {
1343                 err = btrfs_end_transaction_nolock(trans, root);
1344         } else {
1345                 err = btrfs_end_transaction(trans, root);
1346         }
1347         if (!ret)
1348                 ret = err;
1349
1350         btrfs_free_path(path);
1351         return ret;
1352 }
1353
1354 /*
1355  * extent_io.c call back to do delayed allocation processing
1356  */
1357 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1358                               u64 start, u64 end, int *page_started,
1359                               unsigned long *nr_written)
1360 {
1361         int ret;
1362         struct btrfs_root *root = BTRFS_I(inode)->root;
1363
1364         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1365                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1366                                          page_started, 1, nr_written);
1367         else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1368                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1369                                          page_started, 0, nr_written);
1370         else if (!btrfs_test_opt(root, COMPRESS) &&
1371                  !(BTRFS_I(inode)->force_compress) &&
1372                  !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1373                 ret = cow_file_range(inode, locked_page, start, end,
1374                                       page_started, nr_written, 1);
1375         else
1376                 ret = cow_file_range_async(inode, locked_page, start, end,
1377                                            page_started, nr_written);
1378         return ret;
1379 }
1380
1381 static void btrfs_split_extent_hook(struct inode *inode,
1382                                     struct extent_state *orig, u64 split)
1383 {
1384         /* not delalloc, ignore it */
1385         if (!(orig->state & EXTENT_DELALLOC))
1386                 return;
1387
1388         spin_lock(&BTRFS_I(inode)->lock);
1389         BTRFS_I(inode)->outstanding_extents++;
1390         spin_unlock(&BTRFS_I(inode)->lock);
1391 }
1392
1393 /*
1394  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1395  * extents so we can keep track of new extents that are just merged onto old
1396  * extents, such as when we are doing sequential writes, so we can properly
1397  * account for the metadata space we'll need.
1398  */
1399 static void btrfs_merge_extent_hook(struct inode *inode,
1400                                     struct extent_state *new,
1401                                     struct extent_state *other)
1402 {
1403         /* not delalloc, ignore it */
1404         if (!(other->state & EXTENT_DELALLOC))
1405                 return;
1406
1407         spin_lock(&BTRFS_I(inode)->lock);
1408         BTRFS_I(inode)->outstanding_extents--;
1409         spin_unlock(&BTRFS_I(inode)->lock);
1410 }
1411
1412 /*
1413  * extent_io.c set_bit_hook, used to track delayed allocation
1414  * bytes in this file, and to maintain the list of inodes that
1415  * have pending delalloc work to be done.
1416  */
1417 static void btrfs_set_bit_hook(struct inode *inode,
1418                                struct extent_state *state, int *bits)
1419 {
1420
1421         /*
1422          * set_bit and clear bit hooks normally require _irqsave/restore
1423          * but in this case, we are only testing for the DELALLOC
1424          * bit, which is only set or cleared with irqs on
1425          */
1426         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1427                 struct btrfs_root *root = BTRFS_I(inode)->root;
1428                 u64 len = state->end + 1 - state->start;
1429                 bool do_list = !btrfs_is_free_space_inode(root, inode);
1430
1431                 if (*bits & EXTENT_FIRST_DELALLOC) {
1432                         *bits &= ~EXTENT_FIRST_DELALLOC;
1433                 } else {
1434                         spin_lock(&BTRFS_I(inode)->lock);
1435                         BTRFS_I(inode)->outstanding_extents++;
1436                         spin_unlock(&BTRFS_I(inode)->lock);
1437                 }
1438
1439                 spin_lock(&root->fs_info->delalloc_lock);
1440                 BTRFS_I(inode)->delalloc_bytes += len;
1441                 root->fs_info->delalloc_bytes += len;
1442                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1443                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1444                                       &root->fs_info->delalloc_inodes);
1445                 }
1446                 spin_unlock(&root->fs_info->delalloc_lock);
1447         }
1448 }
1449
1450 /*
1451  * extent_io.c clear_bit_hook, see set_bit_hook for why
1452  */
1453 static void btrfs_clear_bit_hook(struct inode *inode,
1454                                  struct extent_state *state, int *bits)
1455 {
1456         /*
1457          * set_bit and clear bit hooks normally require _irqsave/restore
1458          * but in this case, we are only testing for the DELALLOC
1459          * bit, which is only set or cleared with irqs on
1460          */
1461         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1462                 struct btrfs_root *root = BTRFS_I(inode)->root;
1463                 u64 len = state->end + 1 - state->start;
1464                 bool do_list = !btrfs_is_free_space_inode(root, inode);
1465
1466                 if (*bits & EXTENT_FIRST_DELALLOC) {
1467                         *bits &= ~EXTENT_FIRST_DELALLOC;
1468                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1469                         spin_lock(&BTRFS_I(inode)->lock);
1470                         BTRFS_I(inode)->outstanding_extents--;
1471                         spin_unlock(&BTRFS_I(inode)->lock);
1472                 }
1473
1474                 if (*bits & EXTENT_DO_ACCOUNTING)
1475                         btrfs_delalloc_release_metadata(inode, len);
1476
1477                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1478                     && do_list)
1479                         btrfs_free_reserved_data_space(inode, len);
1480
1481                 spin_lock(&root->fs_info->delalloc_lock);
1482                 root->fs_info->delalloc_bytes -= len;
1483                 BTRFS_I(inode)->delalloc_bytes -= len;
1484
1485                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1486                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1487                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1488                 }
1489                 spin_unlock(&root->fs_info->delalloc_lock);
1490         }
1491 }
1492
1493 /*
1494  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1495  * we don't create bios that span stripes or chunks
1496  */
1497 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1498                          size_t size, struct bio *bio,
1499                          unsigned long bio_flags)
1500 {
1501         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1502         struct btrfs_mapping_tree *map_tree;
1503         u64 logical = (u64)bio->bi_sector << 9;
1504         u64 length = 0;
1505         u64 map_length;
1506         int ret;
1507
1508         if (bio_flags & EXTENT_BIO_COMPRESSED)
1509                 return 0;
1510
1511         length = bio->bi_size;
1512         map_tree = &root->fs_info->mapping_tree;
1513         map_length = length;
1514         ret = btrfs_map_block(map_tree, READ, logical,
1515                               &map_length, NULL, 0);
1516         /* Will always return 0 or 1 with map_multi == NULL */
1517         BUG_ON(ret < 0);
1518         if (map_length < length + size)
1519                 return 1;
1520         return 0;
1521 }
1522
1523 /*
1524  * in order to insert checksums into the metadata in large chunks,
1525  * we wait until bio submission time.   All the pages in the bio are
1526  * checksummed and sums are attached onto the ordered extent record.
1527  *
1528  * At IO completion time the cums attached on the ordered extent record
1529  * are inserted into the btree
1530  */
1531 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1532                                     struct bio *bio, int mirror_num,
1533                                     unsigned long bio_flags,
1534                                     u64 bio_offset)
1535 {
1536         struct btrfs_root *root = BTRFS_I(inode)->root;
1537         int ret = 0;
1538
1539         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1540         BUG_ON(ret); /* -ENOMEM */
1541         return 0;
1542 }
1543
1544 /*
1545  * in order to insert checksums into the metadata in large chunks,
1546  * we wait until bio submission time.   All the pages in the bio are
1547  * checksummed and sums are attached onto the ordered extent record.
1548  *
1549  * At IO completion time the cums attached on the ordered extent record
1550  * are inserted into the btree
1551  */
1552 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1553                           int mirror_num, unsigned long bio_flags,
1554                           u64 bio_offset)
1555 {
1556         struct btrfs_root *root = BTRFS_I(inode)->root;
1557         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1558 }
1559
1560 /*
1561  * extent_io.c submission hook. This does the right thing for csum calculation
1562  * on write, or reading the csums from the tree before a read
1563  */
1564 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1565                           int mirror_num, unsigned long bio_flags,
1566                           u64 bio_offset)
1567 {
1568         struct btrfs_root *root = BTRFS_I(inode)->root;
1569         int ret = 0;
1570         int skip_sum;
1571         int metadata = 0;
1572
1573         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1574
1575         if (btrfs_is_free_space_inode(root, inode))
1576                 metadata = 2;
1577
1578         if (!(rw & REQ_WRITE)) {
1579                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1580                 if (ret)
1581                         return ret;
1582
1583                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1584                         return btrfs_submit_compressed_read(inode, bio,
1585                                                     mirror_num, bio_flags);
1586                 } else if (!skip_sum) {
1587                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1588                         if (ret)
1589                                 return ret;
1590                 }
1591                 goto mapit;
1592         } else if (!skip_sum) {
1593                 /* csum items have already been cloned */
1594                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1595                         goto mapit;
1596                 /* we're doing a write, do the async checksumming */
1597                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1598                                    inode, rw, bio, mirror_num,
1599                                    bio_flags, bio_offset,
1600                                    __btrfs_submit_bio_start,
1601                                    __btrfs_submit_bio_done);
1602         }
1603
1604 mapit:
1605         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1606 }
1607
1608 /*
1609  * given a list of ordered sums record them in the inode.  This happens
1610  * at IO completion time based on sums calculated at bio submission time.
1611  */
1612 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1613                              struct inode *inode, u64 file_offset,
1614                              struct list_head *list)
1615 {
1616         struct btrfs_ordered_sum *sum;
1617
1618         list_for_each_entry(sum, list, list) {
1619                 btrfs_csum_file_blocks(trans,
1620                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1621         }
1622         return 0;
1623 }
1624
1625 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1626                               struct extent_state **cached_state)
1627 {
1628         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1629                 WARN_ON(1);
1630         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1631                                    cached_state, GFP_NOFS);
1632 }
1633
1634 /* see btrfs_writepage_start_hook for details on why this is required */
1635 struct btrfs_writepage_fixup {
1636         struct page *page;
1637         struct btrfs_work work;
1638 };
1639
1640 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1641 {
1642         struct btrfs_writepage_fixup *fixup;
1643         struct btrfs_ordered_extent *ordered;
1644         struct extent_state *cached_state = NULL;
1645         struct page *page;
1646         struct inode *inode;
1647         u64 page_start;
1648         u64 page_end;
1649         int ret;
1650
1651         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1652         page = fixup->page;
1653 again:
1654         lock_page(page);
1655         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1656                 ClearPageChecked(page);
1657                 goto out_page;
1658         }
1659
1660         inode = page->mapping->host;
1661         page_start = page_offset(page);
1662         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1663
1664         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1665                          &cached_state);
1666
1667         /* already ordered? We're done */
1668         if (PagePrivate2(page))
1669                 goto out;
1670
1671         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1672         if (ordered) {
1673                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1674                                      page_end, &cached_state, GFP_NOFS);
1675                 unlock_page(page);
1676                 btrfs_start_ordered_extent(inode, ordered, 1);
1677                 btrfs_put_ordered_extent(ordered);
1678                 goto again;
1679         }
1680
1681         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1682         if (ret) {
1683                 mapping_set_error(page->mapping, ret);
1684                 end_extent_writepage(page, ret, page_start, page_end);
1685                 ClearPageChecked(page);
1686                 goto out;
1687          }
1688
1689         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1690         ClearPageChecked(page);
1691         set_page_dirty(page);
1692 out:
1693         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1694                              &cached_state, GFP_NOFS);
1695 out_page:
1696         unlock_page(page);
1697         page_cache_release(page);
1698         kfree(fixup);
1699 }
1700
1701 /*
1702  * There are a few paths in the higher layers of the kernel that directly
1703  * set the page dirty bit without asking the filesystem if it is a
1704  * good idea.  This causes problems because we want to make sure COW
1705  * properly happens and the data=ordered rules are followed.
1706  *
1707  * In our case any range that doesn't have the ORDERED bit set
1708  * hasn't been properly setup for IO.  We kick off an async process
1709  * to fix it up.  The async helper will wait for ordered extents, set
1710  * the delalloc bit and make it safe to write the page.
1711  */
1712 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1713 {
1714         struct inode *inode = page->mapping->host;
1715         struct btrfs_writepage_fixup *fixup;
1716         struct btrfs_root *root = BTRFS_I(inode)->root;
1717
1718         /* this page is properly in the ordered list */
1719         if (TestClearPagePrivate2(page))
1720                 return 0;
1721
1722         if (PageChecked(page))
1723                 return -EAGAIN;
1724
1725         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1726         if (!fixup)
1727                 return -EAGAIN;
1728
1729         SetPageChecked(page);
1730         page_cache_get(page);
1731         fixup->work.func = btrfs_writepage_fixup_worker;
1732         fixup->page = page;
1733         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1734         return -EBUSY;
1735 }
1736
1737 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1738                                        struct inode *inode, u64 file_pos,
1739                                        u64 disk_bytenr, u64 disk_num_bytes,
1740                                        u64 num_bytes, u64 ram_bytes,
1741                                        u8 compression, u8 encryption,
1742                                        u16 other_encoding, int extent_type)
1743 {
1744         struct btrfs_root *root = BTRFS_I(inode)->root;
1745         struct btrfs_file_extent_item *fi;
1746         struct btrfs_path *path;
1747         struct extent_buffer *leaf;
1748         struct btrfs_key ins;
1749         u64 hint;
1750         int ret;
1751
1752         path = btrfs_alloc_path();
1753         if (!path)
1754                 return -ENOMEM;
1755
1756         path->leave_spinning = 1;
1757
1758         /*
1759          * we may be replacing one extent in the tree with another.
1760          * The new extent is pinned in the extent map, and we don't want
1761          * to drop it from the cache until it is completely in the btree.
1762          *
1763          * So, tell btrfs_drop_extents to leave this extent in the cache.
1764          * the caller is expected to unpin it and allow it to be merged
1765          * with the others.
1766          */
1767         ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1768                                  &hint, 0);
1769         if (ret)
1770                 goto out;
1771
1772         ins.objectid = btrfs_ino(inode);
1773         ins.offset = file_pos;
1774         ins.type = BTRFS_EXTENT_DATA_KEY;
1775         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1776         if (ret)
1777                 goto out;
1778         leaf = path->nodes[0];
1779         fi = btrfs_item_ptr(leaf, path->slots[0],
1780                             struct btrfs_file_extent_item);
1781         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1782         btrfs_set_file_extent_type(leaf, fi, extent_type);
1783         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1784         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1785         btrfs_set_file_extent_offset(leaf, fi, 0);
1786         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1787         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1788         btrfs_set_file_extent_compression(leaf, fi, compression);
1789         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1790         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1791
1792         btrfs_unlock_up_safe(path, 1);
1793         btrfs_set_lock_blocking(leaf);
1794
1795         btrfs_mark_buffer_dirty(leaf);
1796
1797         inode_add_bytes(inode, num_bytes);
1798
1799         ins.objectid = disk_bytenr;
1800         ins.offset = disk_num_bytes;
1801         ins.type = BTRFS_EXTENT_ITEM_KEY;
1802         ret = btrfs_alloc_reserved_file_extent(trans, root,
1803                                         root->root_key.objectid,
1804                                         btrfs_ino(inode), file_pos, &ins);
1805 out:
1806         btrfs_free_path(path);
1807
1808         return ret;
1809 }
1810
1811 /*
1812  * helper function for btrfs_finish_ordered_io, this
1813  * just reads in some of the csum leaves to prime them into ram
1814  * before we start the transaction.  It limits the amount of btree
1815  * reads required while inside the transaction.
1816  */
1817 /* as ordered data IO finishes, this gets called so we can finish
1818  * an ordered extent if the range of bytes in the file it covers are
1819  * fully written.
1820  */
1821 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
1822 {
1823         struct inode *inode = ordered_extent->inode;
1824         struct btrfs_root *root = BTRFS_I(inode)->root;
1825         struct btrfs_trans_handle *trans = NULL;
1826         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1827         struct extent_state *cached_state = NULL;
1828         int compress_type = 0;
1829         int ret;
1830         bool nolock;
1831
1832         nolock = btrfs_is_free_space_inode(root, inode);
1833
1834         if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
1835                 ret = -EIO;
1836                 goto out;
1837         }
1838
1839         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1840                 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
1841                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1842                 if (!ret) {
1843                         if (nolock)
1844                                 trans = btrfs_join_transaction_nolock(root);
1845                         else
1846                                 trans = btrfs_join_transaction(root);
1847                         if (IS_ERR(trans))
1848                                 return PTR_ERR(trans);
1849                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1850                         ret = btrfs_update_inode_fallback(trans, root, inode);
1851                         if (ret) /* -ENOMEM or corruption */
1852                                 btrfs_abort_transaction(trans, root, ret);
1853                 }
1854                 goto out;
1855         }
1856
1857         lock_extent_bits(io_tree, ordered_extent->file_offset,
1858                          ordered_extent->file_offset + ordered_extent->len - 1,
1859                          0, &cached_state);
1860
1861         if (nolock)
1862                 trans = btrfs_join_transaction_nolock(root);
1863         else
1864                 trans = btrfs_join_transaction(root);
1865         if (IS_ERR(trans)) {
1866                 ret = PTR_ERR(trans);
1867                 trans = NULL;
1868                 goto out_unlock;
1869         }
1870         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1871
1872         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1873                 compress_type = ordered_extent->compress_type;
1874         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1875                 BUG_ON(compress_type);
1876                 ret = btrfs_mark_extent_written(trans, inode,
1877                                                 ordered_extent->file_offset,
1878                                                 ordered_extent->file_offset +
1879                                                 ordered_extent->len);
1880         } else {
1881                 BUG_ON(root == root->fs_info->tree_root);
1882                 ret = insert_reserved_file_extent(trans, inode,
1883                                                 ordered_extent->file_offset,
1884                                                 ordered_extent->start,
1885                                                 ordered_extent->disk_len,
1886                                                 ordered_extent->len,
1887                                                 ordered_extent->len,
1888                                                 compress_type, 0, 0,
1889                                                 BTRFS_FILE_EXTENT_REG);
1890                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1891                                    ordered_extent->file_offset,
1892                                    ordered_extent->len);
1893         }
1894
1895         if (ret < 0) {
1896                 btrfs_abort_transaction(trans, root, ret);
1897                 goto out_unlock;
1898         }
1899
1900         add_pending_csums(trans, inode, ordered_extent->file_offset,
1901                           &ordered_extent->list);
1902
1903         ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1904         if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1905                 ret = btrfs_update_inode_fallback(trans, root, inode);
1906                 if (ret) { /* -ENOMEM or corruption */
1907                         btrfs_abort_transaction(trans, root, ret);
1908                         goto out_unlock;
1909                 }
1910         }
1911         ret = 0;
1912 out_unlock:
1913         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1914                              ordered_extent->file_offset +
1915                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1916 out:
1917         if (root != root->fs_info->tree_root)
1918                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1919         if (trans) {
1920                 if (nolock)
1921                         btrfs_end_transaction_nolock(trans, root);
1922                 else
1923                         btrfs_end_transaction(trans, root);
1924         }
1925
1926         if (ret)
1927                 clear_extent_uptodate(io_tree, ordered_extent->file_offset,
1928                                       ordered_extent->file_offset +
1929                                       ordered_extent->len - 1, NULL, GFP_NOFS);
1930
1931         /*
1932          * This needs to be dont to make sure anybody waiting knows we are done
1933          * upating everything for this ordered extent.
1934          */
1935         btrfs_remove_ordered_extent(inode, ordered_extent);
1936
1937         /* once for us */
1938         btrfs_put_ordered_extent(ordered_extent);
1939         /* once for the tree */
1940         btrfs_put_ordered_extent(ordered_extent);
1941
1942         return ret;
1943 }
1944
1945 static void finish_ordered_fn(struct btrfs_work *work)
1946 {
1947         struct btrfs_ordered_extent *ordered_extent;
1948         ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
1949         btrfs_finish_ordered_io(ordered_extent);
1950 }
1951
1952 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1953                                 struct extent_state *state, int uptodate)
1954 {
1955         struct inode *inode = page->mapping->host;
1956         struct btrfs_root *root = BTRFS_I(inode)->root;
1957         struct btrfs_ordered_extent *ordered_extent = NULL;
1958         struct btrfs_workers *workers;
1959
1960         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1961
1962         ClearPagePrivate2(page);
1963         if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1964                                             end - start + 1, uptodate))
1965                 return 0;
1966
1967         ordered_extent->work.func = finish_ordered_fn;
1968         ordered_extent->work.flags = 0;
1969
1970         if (btrfs_is_free_space_inode(root, inode))
1971                 workers = &root->fs_info->endio_freespace_worker;
1972         else
1973                 workers = &root->fs_info->endio_write_workers;
1974         btrfs_queue_worker(workers, &ordered_extent->work);
1975
1976         return 0;
1977 }
1978
1979 /*
1980  * when reads are done, we need to check csums to verify the data is correct
1981  * if there's a match, we allow the bio to finish.  If not, the code in
1982  * extent_io.c will try to find good copies for us.
1983  */
1984 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1985                                struct extent_state *state, int mirror)
1986 {
1987         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1988         struct inode *inode = page->mapping->host;
1989         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1990         char *kaddr;
1991         u64 private = ~(u32)0;
1992         int ret;
1993         struct btrfs_root *root = BTRFS_I(inode)->root;
1994         u32 csum = ~(u32)0;
1995
1996         if (PageChecked(page)) {
1997                 ClearPageChecked(page);
1998                 goto good;
1999         }
2000
2001         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2002                 goto good;
2003
2004         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2005             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2006                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2007                                   GFP_NOFS);
2008                 return 0;
2009         }
2010
2011         if (state && state->start == start) {
2012                 private = state->private;
2013                 ret = 0;
2014         } else {
2015                 ret = get_state_private(io_tree, start, &private);
2016         }
2017         kaddr = kmap_atomic(page);
2018         if (ret)
2019                 goto zeroit;
2020
2021         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
2022         btrfs_csum_final(csum, (char *)&csum);
2023         if (csum != private)
2024                 goto zeroit;
2025
2026         kunmap_atomic(kaddr);
2027 good:
2028         return 0;
2029
2030 zeroit:
2031         printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
2032                        "private %llu\n",
2033                        (unsigned long long)btrfs_ino(page->mapping->host),
2034                        (unsigned long long)start, csum,
2035                        (unsigned long long)private);
2036         memset(kaddr + offset, 1, end - start + 1);
2037         flush_dcache_page(page);
2038         kunmap_atomic(kaddr);
2039         if (private == 0)
2040                 return 0;
2041         return -EIO;
2042 }
2043
2044 struct delayed_iput {
2045         struct list_head list;
2046         struct inode *inode;
2047 };
2048
2049 /* JDM: If this is fs-wide, why can't we add a pointer to
2050  * btrfs_inode instead and avoid the allocation? */
2051 void btrfs_add_delayed_iput(struct inode *inode)
2052 {
2053         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2054         struct delayed_iput *delayed;
2055
2056         if (atomic_add_unless(&inode->i_count, -1, 1))
2057                 return;
2058
2059         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2060         delayed->inode = inode;
2061
2062         spin_lock(&fs_info->delayed_iput_lock);
2063         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2064         spin_unlock(&fs_info->delayed_iput_lock);
2065 }
2066
2067 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2068 {
2069         LIST_HEAD(list);
2070         struct btrfs_fs_info *fs_info = root->fs_info;
2071         struct delayed_iput *delayed;
2072         int empty;
2073
2074         spin_lock(&fs_info->delayed_iput_lock);
2075         empty = list_empty(&fs_info->delayed_iputs);
2076         spin_unlock(&fs_info->delayed_iput_lock);
2077         if (empty)
2078                 return;
2079
2080         down_read(&root->fs_info->cleanup_work_sem);
2081         spin_lock(&fs_info->delayed_iput_lock);
2082         list_splice_init(&fs_info->delayed_iputs, &list);
2083         spin_unlock(&fs_info->delayed_iput_lock);
2084
2085         while (!list_empty(&list)) {
2086                 delayed = list_entry(list.next, struct delayed_iput, list);
2087                 list_del(&delayed->list);
2088                 iput(delayed->inode);
2089                 kfree(delayed);
2090         }
2091         up_read(&root->fs_info->cleanup_work_sem);
2092 }
2093
2094 enum btrfs_orphan_cleanup_state {
2095         ORPHAN_CLEANUP_STARTED  = 1,
2096         ORPHAN_CLEANUP_DONE     = 2,
2097 };
2098
2099 /*
2100  * This is called in transaction commit time. If there are no orphan
2101  * files in the subvolume, it removes orphan item and frees block_rsv
2102  * structure.
2103  */
2104 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2105                               struct btrfs_root *root)
2106 {
2107         struct btrfs_block_rsv *block_rsv;
2108         int ret;
2109
2110         if (atomic_read(&root->orphan_inodes) ||
2111             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2112                 return;
2113
2114         spin_lock(&root->orphan_lock);
2115         if (atomic_read(&root->orphan_inodes)) {
2116                 spin_unlock(&root->orphan_lock);
2117                 return;
2118         }
2119
2120         if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2121                 spin_unlock(&root->orphan_lock);
2122                 return;
2123         }
2124
2125         block_rsv = root->orphan_block_rsv;
2126         root->orphan_block_rsv = NULL;
2127         spin_unlock(&root->orphan_lock);
2128
2129         if (root->orphan_item_inserted &&
2130             btrfs_root_refs(&root->root_item) > 0) {
2131                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2132                                             root->root_key.objectid);
2133                 BUG_ON(ret);
2134                 root->orphan_item_inserted = 0;
2135         }
2136
2137         if (block_rsv) {
2138                 WARN_ON(block_rsv->size > 0);
2139                 btrfs_free_block_rsv(root, block_rsv);
2140         }
2141 }
2142
2143 /*
2144  * This creates an orphan entry for the given inode in case something goes
2145  * wrong in the middle of an unlink/truncate.
2146  *
2147  * NOTE: caller of this function should reserve 5 units of metadata for
2148  *       this function.
2149  */
2150 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2151 {
2152         struct btrfs_root *root = BTRFS_I(inode)->root;
2153         struct btrfs_block_rsv *block_rsv = NULL;
2154         int reserve = 0;
2155         int insert = 0;
2156         int ret;
2157
2158         if (!root->orphan_block_rsv) {
2159                 block_rsv = btrfs_alloc_block_rsv(root);
2160                 if (!block_rsv)
2161                         return -ENOMEM;
2162         }
2163
2164         spin_lock(&root->orphan_lock);
2165         if (!root->orphan_block_rsv) {
2166                 root->orphan_block_rsv = block_rsv;
2167         } else if (block_rsv) {
2168                 btrfs_free_block_rsv(root, block_rsv);
2169                 block_rsv = NULL;
2170         }
2171
2172         if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2173                               &BTRFS_I(inode)->runtime_flags)) {
2174 #if 0
2175                 /*
2176                  * For proper ENOSPC handling, we should do orphan
2177                  * cleanup when mounting. But this introduces backward
2178                  * compatibility issue.
2179                  */
2180                 if (!xchg(&root->orphan_item_inserted, 1))
2181                         insert = 2;
2182                 else
2183                         insert = 1;
2184 #endif
2185                 insert = 1;
2186                 atomic_dec(&root->orphan_inodes);
2187         }
2188
2189         if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2190                               &BTRFS_I(inode)->runtime_flags))
2191                 reserve = 1;
2192         spin_unlock(&root->orphan_lock);
2193
2194         /* grab metadata reservation from transaction handle */
2195         if (reserve) {
2196                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2197                 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
2198         }
2199
2200         /* insert an orphan item to track this unlinked/truncated file */
2201         if (insert >= 1) {
2202                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2203                 if (ret && ret != -EEXIST) {
2204                         clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2205                                   &BTRFS_I(inode)->runtime_flags);
2206                         btrfs_abort_transaction(trans, root, ret);
2207                         return ret;
2208                 }
2209                 ret = 0;
2210         }
2211
2212         /* insert an orphan item to track subvolume contains orphan files */
2213         if (insert >= 2) {
2214                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2215                                                root->root_key.objectid);
2216                 if (ret && ret != -EEXIST) {
2217                         btrfs_abort_transaction(trans, root, ret);
2218                         return ret;
2219                 }
2220         }
2221         return 0;
2222 }
2223
2224 /*
2225  * We have done the truncate/delete so we can go ahead and remove the orphan
2226  * item for this particular inode.
2227  */
2228 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2229 {
2230         struct btrfs_root *root = BTRFS_I(inode)->root;
2231         int delete_item = 0;
2232         int release_rsv = 0;
2233         int ret = 0;
2234
2235         spin_lock(&root->orphan_lock);
2236         if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2237                                &BTRFS_I(inode)->runtime_flags))
2238                 delete_item = 1;
2239
2240         if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2241                                &BTRFS_I(inode)->runtime_flags))
2242                 release_rsv = 1;
2243         spin_unlock(&root->orphan_lock);
2244
2245         if (trans && delete_item) {
2246                 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2247                 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2248         }
2249
2250         if (release_rsv) {
2251                 btrfs_orphan_release_metadata(inode);
2252                 atomic_dec(&root->orphan_inodes);
2253         }
2254
2255         return 0;
2256 }
2257
2258 /*
2259  * this cleans up any orphans that may be left on the list from the last use
2260  * of this root.
2261  */
2262 int btrfs_orphan_cleanup(struct btrfs_root *root)
2263 {
2264         struct btrfs_path *path;
2265         struct extent_buffer *leaf;
2266         struct btrfs_key key, found_key;
2267         struct btrfs_trans_handle *trans;
2268         struct inode *inode;
2269         u64 last_objectid = 0;
2270         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2271
2272         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2273                 return 0;
2274
2275         path = btrfs_alloc_path();
2276         if (!path) {
2277                 ret = -ENOMEM;
2278                 goto out;
2279         }
2280         path->reada = -1;
2281
2282         key.objectid = BTRFS_ORPHAN_OBJECTID;
2283         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2284         key.offset = (u64)-1;
2285
2286         while (1) {
2287                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2288                 if (ret < 0)
2289                         goto out;
2290
2291                 /*
2292                  * if ret == 0 means we found what we were searching for, which
2293                  * is weird, but possible, so only screw with path if we didn't
2294                  * find the key and see if we have stuff that matches
2295                  */
2296                 if (ret > 0) {
2297                         ret = 0;
2298                         if (path->slots[0] == 0)
2299                                 break;
2300                         path->slots[0]--;
2301                 }
2302
2303                 /* pull out the item */
2304                 leaf = path->nodes[0];
2305                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2306
2307                 /* make sure the item matches what we want */
2308                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2309                         break;
2310                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2311                         break;
2312
2313                 /* release the path since we're done with it */
2314                 btrfs_release_path(path);
2315
2316                 /*
2317                  * this is where we are basically btrfs_lookup, without the
2318                  * crossing root thing.  we store the inode number in the
2319                  * offset of the orphan item.
2320                  */
2321
2322                 if (found_key.offset == last_objectid) {
2323                         printk(KERN_ERR "btrfs: Error removing orphan entry, "
2324                                "stopping orphan cleanup\n");
2325                         ret = -EINVAL;
2326                         goto out;
2327                 }
2328
2329                 last_objectid = found_key.offset;
2330
2331                 found_key.objectid = found_key.offset;
2332                 found_key.type = BTRFS_INODE_ITEM_KEY;
2333                 found_key.offset = 0;
2334                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2335                 ret = PTR_RET(inode);
2336                 if (ret && ret != -ESTALE)
2337                         goto out;
2338
2339                 if (ret == -ESTALE && root == root->fs_info->tree_root) {
2340                         struct btrfs_root *dead_root;
2341                         struct btrfs_fs_info *fs_info = root->fs_info;
2342                         int is_dead_root = 0;
2343
2344                         /*
2345                          * this is an orphan in the tree root. Currently these
2346                          * could come from 2 sources:
2347                          *  a) a snapshot deletion in progress
2348                          *  b) a free space cache inode
2349                          * We need to distinguish those two, as the snapshot
2350                          * orphan must not get deleted.
2351                          * find_dead_roots already ran before us, so if this
2352                          * is a snapshot deletion, we should find the root
2353                          * in the dead_roots list
2354                          */
2355                         spin_lock(&fs_info->trans_lock);
2356                         list_for_each_entry(dead_root, &fs_info->dead_roots,
2357                                             root_list) {
2358                                 if (dead_root->root_key.objectid ==
2359                                     found_key.objectid) {
2360                                         is_dead_root = 1;
2361                                         break;
2362                                 }
2363                         }
2364                         spin_unlock(&fs_info->trans_lock);
2365                         if (is_dead_root) {
2366                                 /* prevent this orphan from being found again */
2367                                 key.offset = found_key.objectid - 1;
2368                                 continue;
2369                         }
2370                 }
2371                 /*
2372                  * Inode is already gone but the orphan item is still there,
2373                  * kill the orphan item.
2374                  */
2375                 if (ret == -ESTALE) {
2376                         trans = btrfs_start_transaction(root, 1);
2377                         if (IS_ERR(trans)) {
2378                                 ret = PTR_ERR(trans);
2379                                 goto out;
2380                         }
2381                         printk(KERN_ERR "auto deleting %Lu\n",
2382                                found_key.objectid);
2383                         ret = btrfs_del_orphan_item(trans, root,
2384                                                     found_key.objectid);
2385                         BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2386                         btrfs_end_transaction(trans, root);
2387                         continue;
2388                 }
2389
2390                 /*
2391                  * add this inode to the orphan list so btrfs_orphan_del does
2392                  * the proper thing when we hit it
2393                  */
2394                 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2395                         &BTRFS_I(inode)->runtime_flags);
2396
2397                 /* if we have links, this was a truncate, lets do that */
2398                 if (inode->i_nlink) {
2399                         if (!S_ISREG(inode->i_mode)) {
2400                                 WARN_ON(1);
2401                                 iput(inode);
2402                                 continue;
2403                         }
2404                         nr_truncate++;
2405                         ret = btrfs_truncate(inode);
2406                 } else {
2407                         nr_unlink++;
2408                 }
2409
2410                 /* this will do delete_inode and everything for us */
2411                 iput(inode);
2412                 if (ret)
2413                         goto out;
2414         }
2415         /* release the path since we're done with it */
2416         btrfs_release_path(path);
2417
2418         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2419
2420         if (root->orphan_block_rsv)
2421                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2422                                         (u64)-1);
2423
2424         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2425                 trans = btrfs_join_transaction(root);
2426                 if (!IS_ERR(trans))
2427                         btrfs_end_transaction(trans, root);
2428         }
2429
2430         if (nr_unlink)
2431                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2432         if (nr_truncate)
2433                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2434
2435 out:
2436         if (ret)
2437                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2438         btrfs_free_path(path);
2439         return ret;
2440 }
2441
2442 /*
2443  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2444  * don't find any xattrs, we know there can't be any acls.
2445  *
2446  * slot is the slot the inode is in, objectid is the objectid of the inode
2447  */
2448 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2449                                           int slot, u64 objectid)
2450 {
2451         u32 nritems = btrfs_header_nritems(leaf);
2452         struct btrfs_key found_key;
2453         int scanned = 0;
2454
2455         slot++;
2456         while (slot < nritems) {
2457                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2458
2459                 /* we found a different objectid, there must not be acls */
2460                 if (found_key.objectid != objectid)
2461                         return 0;
2462
2463                 /* we found an xattr, assume we've got an acl */
2464                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2465                         return 1;
2466
2467                 /*
2468                  * we found a key greater than an xattr key, there can't
2469                  * be any acls later on
2470                  */
2471                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2472                         return 0;
2473
2474                 slot++;
2475                 scanned++;
2476
2477                 /*
2478                  * it goes inode, inode backrefs, xattrs, extents,
2479                  * so if there are a ton of hard links to an inode there can
2480                  * be a lot of backrefs.  Don't waste time searching too hard,
2481                  * this is just an optimization
2482                  */
2483                 if (scanned >= 8)
2484                         break;
2485         }
2486         /* we hit the end of the leaf before we found an xattr or
2487          * something larger than an xattr.  We have to assume the inode
2488          * has acls
2489          */
2490         return 1;
2491 }
2492
2493 /*
2494  * read an inode from the btree into the in-memory inode
2495  */
2496 static void btrfs_read_locked_inode(struct inode *inode)
2497 {
2498         struct btrfs_path *path;
2499         struct extent_buffer *leaf;
2500         struct btrfs_inode_item *inode_item;
2501         struct btrfs_timespec *tspec;
2502         struct btrfs_root *root = BTRFS_I(inode)->root;
2503         struct btrfs_key location;
2504         int maybe_acls;
2505         u32 rdev;
2506         int ret;
2507         bool filled = false;
2508
2509         ret = btrfs_fill_inode(inode, &rdev);
2510         if (!ret)
2511                 filled = true;
2512
2513         path = btrfs_alloc_path();
2514         if (!path)
2515                 goto make_bad;
2516
2517         path->leave_spinning = 1;
2518         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2519
2520         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2521         if (ret)
2522                 goto make_bad;
2523
2524         leaf = path->nodes[0];
2525
2526         if (filled)
2527                 goto cache_acl;
2528
2529         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2530                                     struct btrfs_inode_item);
2531         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2532         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2533         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2534         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2535         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2536
2537         tspec = btrfs_inode_atime(inode_item);
2538         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2539         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2540
2541         tspec = btrfs_inode_mtime(inode_item);
2542         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2543         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2544
2545         tspec = btrfs_inode_ctime(inode_item);
2546         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2547         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2548
2549         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2550         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2551         inode->i_version = btrfs_inode_sequence(leaf, inode_item);
2552         inode->i_generation = BTRFS_I(inode)->generation;
2553         inode->i_rdev = 0;
2554         rdev = btrfs_inode_rdev(leaf, inode_item);
2555
2556         BTRFS_I(inode)->index_cnt = (u64)-1;
2557         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2558 cache_acl:
2559         /*
2560          * try to precache a NULL acl entry for files that don't have
2561          * any xattrs or acls
2562          */
2563         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2564                                            btrfs_ino(inode));
2565         if (!maybe_acls)
2566                 cache_no_acl(inode);
2567
2568         btrfs_free_path(path);
2569
2570         switch (inode->i_mode & S_IFMT) {
2571         case S_IFREG:
2572                 inode->i_mapping->a_ops = &btrfs_aops;
2573                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2574                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2575                 inode->i_fop = &btrfs_file_operations;
2576                 inode->i_op = &btrfs_file_inode_operations;
2577                 break;
2578         case S_IFDIR:
2579                 inode->i_fop = &btrfs_dir_file_operations;
2580                 if (root == root->fs_info->tree_root)
2581                         inode->i_op = &btrfs_dir_ro_inode_operations;
2582                 else
2583                         inode->i_op = &btrfs_dir_inode_operations;
2584                 break;
2585         case S_IFLNK:
2586                 inode->i_op = &btrfs_symlink_inode_operations;
2587                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2588                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2589                 break;
2590         default:
2591                 inode->i_op = &btrfs_special_inode_operations;
2592                 init_special_inode(inode, inode->i_mode, rdev);
2593                 break;
2594         }
2595
2596         btrfs_update_iflags(inode);
2597         return;
2598
2599 make_bad:
2600         btrfs_free_path(path);
2601         make_bad_inode(inode);
2602 }
2603
2604 /*
2605  * given a leaf and an inode, copy the inode fields into the leaf
2606  */
2607 static void fill_inode_item(struct btrfs_trans_handle *trans,
2608                             struct extent_buffer *leaf,
2609                             struct btrfs_inode_item *item,
2610                             struct inode *inode)
2611 {
2612         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2613         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2614         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2615         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2616         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2617
2618         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2619                                inode->i_atime.tv_sec);
2620         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2621                                 inode->i_atime.tv_nsec);
2622
2623         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2624                                inode->i_mtime.tv_sec);
2625         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2626                                 inode->i_mtime.tv_nsec);
2627
2628         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2629                                inode->i_ctime.tv_sec);
2630         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2631                                 inode->i_ctime.tv_nsec);
2632
2633         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2634         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2635         btrfs_set_inode_sequence(leaf, item, inode->i_version);
2636         btrfs_set_inode_transid(leaf, item, trans->transid);
2637         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2638         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2639         btrfs_set_inode_block_group(leaf, item, 0);
2640 }
2641
2642 /*
2643  * copy everything in the in-memory inode into the btree.
2644  */
2645 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
2646                                 struct btrfs_root *root, struct inode *inode)
2647 {
2648         struct btrfs_inode_item *inode_item;
2649         struct btrfs_path *path;
2650         struct extent_buffer *leaf;
2651         int ret;
2652
2653         path = btrfs_alloc_path();
2654         if (!path)
2655                 return -ENOMEM;
2656
2657         path->leave_spinning = 1;
2658         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2659                                  1);
2660         if (ret) {
2661                 if (ret > 0)
2662                         ret = -ENOENT;
2663                 goto failed;
2664         }
2665
2666         btrfs_unlock_up_safe(path, 1);
2667         leaf = path->nodes[0];
2668         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2669                                     struct btrfs_inode_item);
2670
2671         fill_inode_item(trans, leaf, inode_item, inode);
2672         btrfs_mark_buffer_dirty(leaf);
2673         btrfs_set_inode_last_trans(trans, inode);
2674         ret = 0;
2675 failed:
2676         btrfs_free_path(path);
2677         return ret;
2678 }
2679
2680 /*
2681  * copy everything in the in-memory inode into the btree.
2682  */
2683 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2684                                 struct btrfs_root *root, struct inode *inode)
2685 {
2686         int ret;
2687
2688         /*
2689          * If the inode is a free space inode, we can deadlock during commit
2690          * if we put it into the delayed code.
2691          *
2692          * The data relocation inode should also be directly updated
2693          * without delay
2694          */
2695         if (!btrfs_is_free_space_inode(root, inode)
2696             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2697                 ret = btrfs_delayed_update_inode(trans, root, inode);
2698                 if (!ret)
2699                         btrfs_set_inode_last_trans(trans, inode);
2700                 return ret;
2701         }
2702
2703         return btrfs_update_inode_item(trans, root, inode);
2704 }
2705
2706 static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2707                                 struct btrfs_root *root, struct inode *inode)
2708 {
2709         int ret;
2710
2711         ret = btrfs_update_inode(trans, root, inode);
2712         if (ret == -ENOSPC)
2713                 return btrfs_update_inode_item(trans, root, inode);
2714         return ret;
2715 }
2716
2717 /*
2718  * unlink helper that gets used here in inode.c and in the tree logging
2719  * recovery code.  It remove a link in a directory with a given name, and
2720  * also drops the back refs in the inode to the directory
2721  */
2722 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2723                                 struct btrfs_root *root,
2724                                 struct inode *dir, struct inode *inode,
2725                                 const char *name, int name_len)
2726 {
2727         struct btrfs_path *path;
2728         int ret = 0;
2729         struct extent_buffer *leaf;
2730         struct btrfs_dir_item *di;
2731         struct btrfs_key key;
2732         u64 index;
2733         u64 ino = btrfs_ino(inode);
2734         u64 dir_ino = btrfs_ino(dir);
2735
2736         path = btrfs_alloc_path();
2737         if (!path) {
2738                 ret = -ENOMEM;
2739                 goto out;
2740         }
2741
2742         path->leave_spinning = 1;
2743         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2744                                     name, name_len, -1);
2745         if (IS_ERR(di)) {
2746                 ret = PTR_ERR(di);
2747                 goto err;
2748         }
2749         if (!di) {
2750                 ret = -ENOENT;
2751                 goto err;
2752         }
2753         leaf = path->nodes[0];
2754         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2755         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2756         if (ret)
2757                 goto err;
2758         btrfs_release_path(path);
2759
2760         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2761                                   dir_ino, &index);
2762         if (ret) {
2763                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2764                        "inode %llu parent %llu\n", name_len, name,
2765                        (unsigned long long)ino, (unsigned long long)dir_ino);
2766                 btrfs_abort_transaction(trans, root, ret);
2767                 goto err;
2768         }
2769
2770         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2771         if (ret) {
2772                 btrfs_abort_transaction(trans, root, ret);
2773                 goto err;
2774         }
2775
2776         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2777                                          inode, dir_ino);
2778         if (ret != 0 && ret != -ENOENT) {
2779                 btrfs_abort_transaction(trans, root, ret);
2780                 goto err;
2781         }
2782
2783         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2784                                            dir, index);
2785         if (ret == -ENOENT)
2786                 ret = 0;
2787 err:
2788         btrfs_free_path(path);
2789         if (ret)
2790                 goto out;
2791
2792         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2793         inode_inc_iversion(inode);
2794         inode_inc_iversion(dir);
2795         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2796         btrfs_update_inode(trans, root, dir);
2797 out:
2798         return ret;
2799 }
2800
2801 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2802                        struct btrfs_root *root,
2803                        struct inode *dir, struct inode *inode,
2804                        const char *name, int name_len)
2805 {
2806         int ret;
2807         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2808         if (!ret) {
2809                 btrfs_drop_nlink(inode);
2810                 ret = btrfs_update_inode(trans, root, inode);
2811         }
2812         return ret;
2813 }
2814
2815
2816 /* helper to check if there is any shared block in the path */
2817 static int check_path_shared(struct btrfs_root *root,
2818                              struct btrfs_path *path)
2819 {
2820         struct extent_buffer *eb;
2821         int level;
2822         u64 refs = 1;
2823
2824         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2825                 int ret;
2826
2827                 if (!path->nodes[level])
2828                         break;
2829                 eb = path->nodes[level];
2830                 if (!btrfs_block_can_be_shared(root, eb))
2831                         continue;
2832                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2833                                                &refs, NULL);
2834                 if (refs > 1)
2835                         return 1;
2836         }
2837         return 0;
2838 }
2839
2840 /*
2841  * helper to start transaction for unlink and rmdir.
2842  *
2843  * unlink and rmdir are special in btrfs, they do not always free space.
2844  * so in enospc case, we should make sure they will free space before
2845  * allowing them to use the global metadata reservation.
2846  */
2847 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2848                                                        struct dentry *dentry)
2849 {
2850         struct btrfs_trans_handle *trans;
2851         struct btrfs_root *root = BTRFS_I(dir)->root;
2852         struct btrfs_path *path;
2853         struct btrfs_inode_ref *ref;
2854         struct btrfs_dir_item *di;
2855         struct inode *inode = dentry->d_inode;
2856         u64 index;
2857         int check_link = 1;
2858         int err = -ENOSPC;
2859         int ret;
2860         u64 ino = btrfs_ino(inode);
2861         u64 dir_ino = btrfs_ino(dir);
2862
2863         /*
2864          * 1 for the possible orphan item
2865          * 1 for the dir item
2866          * 1 for the dir index
2867          * 1 for the inode ref
2868          * 1 for the inode ref in the tree log
2869          * 2 for the dir entries in the log
2870          * 1 for the inode
2871          */
2872         trans = btrfs_start_transaction(root, 8);
2873         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2874                 return trans;
2875
2876         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2877                 return ERR_PTR(-ENOSPC);
2878
2879         /* check if there is someone else holds reference */
2880         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2881                 return ERR_PTR(-ENOSPC);
2882
2883         if (atomic_read(&inode->i_count) > 2)
2884                 return ERR_PTR(-ENOSPC);
2885
2886         if (xchg(&root->fs_info->enospc_unlink, 1))
2887                 return ERR_PTR(-ENOSPC);
2888
2889         path = btrfs_alloc_path();
2890         if (!path) {
2891                 root->fs_info->enospc_unlink = 0;
2892                 return ERR_PTR(-ENOMEM);
2893         }
2894
2895         /* 1 for the orphan item */
2896         trans = btrfs_start_transaction(root, 1);
2897         if (IS_ERR(trans)) {
2898                 btrfs_free_path(path);
2899                 root->fs_info->enospc_unlink = 0;
2900                 return trans;
2901         }
2902
2903         path->skip_locking = 1;
2904         path->search_commit_root = 1;
2905
2906         ret = btrfs_lookup_inode(trans, root, path,
2907                                 &BTRFS_I(dir)->location, 0);
2908         if (ret < 0) {
2909                 err = ret;
2910                 goto out;
2911         }
2912         if (ret == 0) {
2913                 if (check_path_shared(root, path))
2914                         goto out;
2915         } else {
2916                 check_link = 0;
2917         }
2918         btrfs_release_path(path);
2919
2920         ret = btrfs_lookup_inode(trans, root, path,
2921                                 &BTRFS_I(inode)->location, 0);
2922         if (ret < 0) {
2923                 err = ret;
2924                 goto out;
2925         }
2926         if (ret == 0) {
2927                 if (check_path_shared(root, path))
2928                         goto out;
2929         } else {
2930                 check_link = 0;
2931         }
2932         btrfs_release_path(path);
2933
2934         if (ret == 0 && S_ISREG(inode->i_mode)) {
2935                 ret = btrfs_lookup_file_extent(trans, root, path,
2936                                                ino, (u64)-1, 0);
2937                 if (ret < 0) {
2938                         err = ret;
2939                         goto out;
2940                 }
2941                 BUG_ON(ret == 0); /* Corruption */
2942                 if (check_path_shared(root, path))
2943                         goto out;
2944                 btrfs_release_path(path);
2945         }
2946
2947         if (!check_link) {
2948                 err = 0;
2949                 goto out;
2950         }
2951
2952         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2953                                 dentry->d_name.name, dentry->d_name.len, 0);
2954         if (IS_ERR(di)) {
2955                 err = PTR_ERR(di);
2956                 goto out;
2957         }
2958         if (di) {
2959                 if (check_path_shared(root, path))
2960                         goto out;
2961         } else {
2962                 err = 0;
2963                 goto out;
2964         }
2965         btrfs_release_path(path);
2966
2967         ref = btrfs_lookup_inode_ref(trans, root, path,
2968                                 dentry->d_name.name, dentry->d_name.len,
2969                                 ino, dir_ino, 0);
2970         if (IS_ERR(ref)) {
2971                 err = PTR_ERR(ref);
2972                 goto out;
2973         }
2974         BUG_ON(!ref); /* Logic error */
2975         if (check_path_shared(root, path))
2976                 goto out;
2977         index = btrfs_inode_ref_index(path->nodes[0], ref);
2978         btrfs_release_path(path);
2979
2980         /*
2981          * This is a commit root search, if we can lookup inode item and other
2982          * relative items in the commit root, it means the transaction of
2983          * dir/file creation has been committed, and the dir index item that we
2984          * delay to insert has also been inserted into the commit root. So
2985          * we needn't worry about the delayed insertion of the dir index item
2986          * here.
2987          */
2988         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
2989                                 dentry->d_name.name, dentry->d_name.len, 0);
2990         if (IS_ERR(di)) {
2991                 err = PTR_ERR(di);
2992                 goto out;
2993         }
2994         BUG_ON(ret == -ENOENT);
2995         if (check_path_shared(root, path))
2996                 goto out;
2997
2998         err = 0;
2999 out:
3000         btrfs_free_path(path);
3001         /* Migrate the orphan reservation over */
3002         if (!err)
3003                 err = btrfs_block_rsv_migrate(trans->block_rsv,
3004                                 &root->fs_info->global_block_rsv,
3005                                 trans->bytes_reserved);
3006
3007         if (err) {
3008                 btrfs_end_transaction(trans, root);
3009                 root->fs_info->enospc_unlink = 0;
3010                 return ERR_PTR(err);
3011         }
3012
3013         trans->block_rsv = &root->fs_info->global_block_rsv;
3014         return trans;
3015 }
3016
3017 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
3018                                struct btrfs_root *root)
3019 {
3020         if (trans->block_rsv == &root->fs_info->global_block_rsv) {
3021                 btrfs_block_rsv_release(root, trans->block_rsv,
3022                                         trans->bytes_reserved);
3023                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3024                 BUG_ON(!root->fs_info->enospc_unlink);
3025                 root->fs_info->enospc_unlink = 0;
3026         }
3027         btrfs_end_transaction(trans, root);
3028 }
3029
3030 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3031 {
3032         struct btrfs_root *root = BTRFS_I(dir)->root;
3033         struct btrfs_trans_handle *trans;
3034         struct inode *inode = dentry->d_inode;
3035         int ret;
3036         unsigned long nr = 0;
3037
3038         trans = __unlink_start_trans(dir, dentry);
3039         if (IS_ERR(trans))
3040                 return PTR_ERR(trans);
3041
3042         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3043
3044         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3045                                  dentry->d_name.name, dentry->d_name.len);
3046         if (ret)
3047                 goto out;
3048
3049         if (inode->i_nlink == 0) {
3050                 ret = btrfs_orphan_add(trans, inode);
3051                 if (ret)
3052                         goto out;
3053         }
3054
3055 out:
3056         nr = trans->blocks_used;
3057         __unlink_end_trans(trans, root);
3058         btrfs_btree_balance_dirty(root, nr);
3059         return ret;
3060 }
3061
3062 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3063                         struct btrfs_root *root,
3064                         struct inode *dir, u64 objectid,
3065                         const char *name, int name_len)
3066 {
3067         struct btrfs_path *path;
3068         struct extent_buffer *leaf;
3069         struct btrfs_dir_item *di;
3070         struct btrfs_key key;
3071         u64 index;
3072         int ret;
3073         u64 dir_ino = btrfs_ino(dir);
3074
3075         path = btrfs_alloc_path();
3076         if (!path)
3077                 return -ENOMEM;
3078
3079         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3080                                    name, name_len, -1);
3081         if (IS_ERR_OR_NULL(di)) {
3082                 if (!di)
3083                         ret = -ENOENT;
3084                 else
3085                         ret = PTR_ERR(di);
3086                 goto out;
3087         }
3088
3089         leaf = path->nodes[0];
3090         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3091         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3092         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3093         if (ret) {
3094                 btrfs_abort_transaction(trans, root, ret);
3095                 goto out;
3096         }
3097         btrfs_release_path(path);
3098
3099         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3100                                  objectid, root->root_key.objectid,
3101                                  dir_ino, &index, name, name_len);
3102         if (ret < 0) {
3103                 if (ret != -ENOENT) {
3104                         btrfs_abort_transaction(trans, root, ret);
3105                         goto out;
3106                 }
3107                 di = btrfs_search_dir_index_item(root, path, dir_ino,
3108                                                  name, name_len);
3109                 if (IS_ERR_OR_NULL(di)) {
3110                         if (!di)
3111                                 ret = -ENOENT;
3112                         else
3113                                 ret = PTR_ERR(di);
3114                         btrfs_abort_transaction(trans, root, ret);
3115                         goto out;
3116                 }
3117
3118                 leaf = path->nodes[0];
3119                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3120                 btrfs_release_path(path);
3121                 index = key.offset;
3122         }
3123         btrfs_release_path(path);
3124
3125         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3126         if (ret) {
3127                 btrfs_abort_transaction(trans, root, ret);
3128                 goto out;
3129         }
3130
3131         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3132         inode_inc_iversion(dir);
3133         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3134         ret = btrfs_update_inode(trans, root, dir);
3135         if (ret)
3136                 btrfs_abort_transaction(trans, root, ret);
3137 out:
3138         btrfs_free_path(path);
3139         return ret;
3140 }
3141
3142 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3143 {
3144         struct inode *inode = dentry->d_inode;
3145         int err = 0;
3146         struct btrfs_root *root = BTRFS_I(dir)->root;
3147         struct btrfs_trans_handle *trans;
3148         unsigned long nr = 0;
3149
3150         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3151             btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3152                 return -ENOTEMPTY;
3153
3154         trans = __unlink_start_trans(dir, dentry);
3155         if (IS_ERR(trans))
3156                 return PTR_ERR(trans);
3157
3158         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3159                 err = btrfs_unlink_subvol(trans, root, dir,
3160                                           BTRFS_I(inode)->location.objectid,
3161                                           dentry->d_name.name,
3162                                           dentry->d_name.len);
3163                 goto out;
3164         }
3165
3166         err = btrfs_orphan_add(trans, inode);
3167         if (err)
3168                 goto out;
3169
3170         /* now the directory is empty */
3171         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3172                                  dentry->d_name.name, dentry->d_name.len);
3173         if (!err)
3174                 btrfs_i_size_write(inode, 0);
3175 out:
3176         nr = trans->blocks_used;
3177         __unlink_end_trans(trans, root);
3178         btrfs_btree_balance_dirty(root, nr);
3179
3180         return err;
3181 }
3182
3183 /*
3184  * this can truncate away extent items, csum items and directory items.
3185  * It starts at a high offset and removes keys until it can't find
3186  * any higher than new_size
3187  *
3188  * csum items that cross the new i_size are truncated to the new size
3189  * as well.
3190  *
3191  * min_type is the minimum key type to truncate down to.  If set to 0, this
3192  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3193  */
3194 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3195                                struct btrfs_root *root,
3196                                struct inode *inode,
3197                                u64 new_size, u32 min_type)
3198 {
3199         struct btrfs_path *path;
3200         struct extent_buffer *leaf;
3201         struct btrfs_file_extent_item *fi;
3202         struct btrfs_key key;
3203         struct btrfs_key found_key;
3204         u64 extent_start = 0;
3205         u64 extent_num_bytes = 0;
3206         u64 extent_offset = 0;
3207         u64 item_end = 0;
3208         u64 mask = root->sectorsize - 1;
3209         u32 found_type = (u8)-1;
3210         int found_extent;
3211         int del_item;
3212         int pending_del_nr = 0;
3213         int pending_del_slot = 0;
3214         int extent_type = -1;
3215         int ret;
3216         int err = 0;
3217         u64 ino = btrfs_ino(inode);
3218
3219         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3220
3221         path = btrfs_alloc_path();
3222         if (!path)
3223                 return -ENOMEM;
3224         path->reada = -1;
3225
3226         if (root->ref_cows || root == root->fs_info->tree_root)
3227                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3228
3229         /*
3230          * This function is also used to drop the items in the log tree before
3231          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3232          * it is used to drop the loged items. So we shouldn't kill the delayed
3233          * items.
3234          */
3235         if (min_type == 0 && root == BTRFS_I(inode)->root)
3236                 btrfs_kill_delayed_inode_items(inode);
3237
3238         key.objectid = ino;
3239         key.offset = (u64)-1;
3240         key.type = (u8)-1;
3241
3242 search_again:
3243         path->leave_spinning = 1;
3244         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3245         if (ret < 0) {
3246                 err = ret;
3247                 goto out;
3248         }
3249
3250         if (ret > 0) {
3251                 /* there are no items in the tree for us to truncate, we're
3252                  * done
3253                  */
3254                 if (path->slots[0] == 0)
3255                         goto out;
3256                 path->slots[0]--;
3257         }
3258
3259         while (1) {
3260                 fi = NULL;
3261                 leaf = path->nodes[0];
3262                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3263                 found_type = btrfs_key_type(&found_key);
3264
3265                 if (found_key.objectid != ino)
3266                         break;
3267
3268                 if (found_type < min_type)
3269                         break;
3270
3271                 item_end = found_key.offset;
3272                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3273                         fi = btrfs_item_ptr(leaf, path->slots[0],
3274                                             struct btrfs_file_extent_item);
3275                         extent_type = btrfs_file_extent_type(leaf, fi);
3276                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3277                                 item_end +=
3278                                     btrfs_file_extent_num_bytes(leaf, fi);
3279                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3280                                 item_end += btrfs_file_extent_inline_len(leaf,
3281                                                                          fi);
3282                         }
3283                         item_end--;
3284                 }
3285                 if (found_type > min_type) {
3286                         del_item = 1;
3287                 } else {
3288                         if (item_end < new_size)
3289                                 break;
3290                         if (found_key.offset >= new_size)
3291                                 del_item = 1;
3292                         else
3293                                 del_item = 0;
3294                 }
3295                 found_extent = 0;
3296                 /* FIXME, shrink the extent if the ref count is only 1 */
3297                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3298                         goto delete;
3299
3300                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3301                         u64 num_dec;
3302                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3303                         if (!del_item) {
3304                                 u64 orig_num_bytes =
3305                                         btrfs_file_extent_num_bytes(leaf, fi);
3306                                 extent_num_bytes = new_size -
3307                                         found_key.offset + root->sectorsize - 1;
3308                                 extent_num_bytes = extent_num_bytes &
3309                                         ~((u64)root->sectorsize - 1);
3310                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3311                                                          extent_num_bytes);
3312                                 num_dec = (orig_num_bytes -
3313                                            extent_num_bytes);
3314                                 if (root->ref_cows && extent_start != 0)
3315                                         inode_sub_bytes(inode, num_dec);
3316                                 btrfs_mark_buffer_dirty(leaf);
3317                         } else {
3318                                 extent_num_bytes =
3319                                         btrfs_file_extent_disk_num_bytes(leaf,
3320                                                                          fi);
3321                                 extent_offset = found_key.offset -
3322                                         btrfs_file_extent_offset(leaf, fi);
3323
3324                                 /* FIXME blocksize != 4096 */
3325                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3326                                 if (extent_start != 0) {
3327                                         found_extent = 1;
3328                                         if (root->ref_cows)
3329                                                 inode_sub_bytes(inode, num_dec);
3330                                 }
3331                         }
3332                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3333                         /*
3334                          * we can't truncate inline items that have had
3335                          * special encodings
3336                          */
3337                         if (!del_item &&
3338                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3339                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3340                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3341                                 u32 size = new_size - found_key.offset;
3342
3343                                 if (root->ref_cows) {
3344                                         inode_sub_bytes(inode, item_end + 1 -
3345                                                         new_size);
3346                                 }
3347                                 size =
3348                                     btrfs_file_extent_calc_inline_size(size);
3349                                 btrfs_truncate_item(trans, root, path,
3350                                                     size, 1);
3351                         } else if (root->ref_cows) {
3352                                 inode_sub_bytes(inode, item_end + 1 -
3353                                                 found_key.offset);
3354                         }
3355                 }
3356 delete:
3357                 if (del_item) {
3358                         if (!pending_del_nr) {
3359                                 /* no pending yet, add ourselves */
3360                                 pending_del_slot = path->slots[0];
3361                                 pending_del_nr = 1;
3362                         } else if (pending_del_nr &&
3363                                    path->slots[0] + 1 == pending_del_slot) {
3364                                 /* hop on the pending chunk */
3365                                 pending_del_nr++;
3366                                 pending_del_slot = path->slots[0];
3367                         } else {
3368                                 BUG();
3369                         }
3370                 } else {
3371                         break;
3372                 }
3373                 if (found_extent && (root->ref_cows ||
3374                                      root == root->fs_info->tree_root)) {
3375                         btrfs_set_path_blocking(path);
3376                         ret = btrfs_free_extent(trans, root, extent_start,
3377                                                 extent_num_bytes, 0,
3378                                                 btrfs_header_owner(leaf),
3379                                                 ino, extent_offset, 0);
3380                         BUG_ON(ret);
3381                 }
3382
3383                 if (found_type == BTRFS_INODE_ITEM_KEY)
3384                         break;
3385
3386                 if (path->slots[0] == 0 ||
3387                     path->slots[0] != pending_del_slot) {
3388                         if (root->ref_cows &&
3389                             BTRFS_I(inode)->location.objectid !=
3390                                                 BTRFS_FREE_INO_OBJECTID) {
3391                                 err = -EAGAIN;
3392                                 goto out;
3393                         }
3394                         if (pending_del_nr) {
3395                                 ret = btrfs_del_items(trans, root, path,
3396                                                 pending_del_slot,
3397                                                 pending_del_nr);
3398                                 if (ret) {
3399                                         btrfs_abort_transaction(trans,
3400                                                                 root, ret);
3401                                         goto error;
3402                                 }
3403                                 pending_del_nr = 0;
3404                         }
3405                         btrfs_release_path(path);
3406                         goto search_again;
3407                 } else {
3408                         path->slots[0]--;
3409                 }
3410         }
3411 out:
3412         if (pending_del_nr) {
3413                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3414                                       pending_del_nr);
3415                 if (ret)
3416                         btrfs_abort_transaction(trans, root, ret);
3417         }
3418 error:
3419         btrfs_free_path(path);
3420         return err;
3421 }
3422
3423 /*
3424  * taken from block_truncate_page, but does cow as it zeros out
3425  * any bytes left in the last page in the file.
3426  */
3427 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3428 {
3429         struct inode *inode = mapping->host;
3430         struct btrfs_root *root = BTRFS_I(inode)->root;
3431         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3432         struct btrfs_ordered_extent *ordered;
3433         struct extent_state *cached_state = NULL;
3434         char *kaddr;
3435         u32 blocksize = root->sectorsize;
3436         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3437         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3438         struct page *page;
3439         gfp_t mask = btrfs_alloc_write_mask(mapping);
3440         int ret = 0;
3441         u64 page_start;
3442         u64 page_end;
3443
3444         if ((offset & (blocksize - 1)) == 0)
3445                 goto out;
3446         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3447         if (ret)
3448                 goto out;
3449
3450         ret = -ENOMEM;
3451 again:
3452         page = find_or_create_page(mapping, index, mask);
3453         if (!page) {
3454                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3455                 goto out;
3456         }
3457
3458         page_start = page_offset(page);
3459         page_end = page_start + PAGE_CACHE_SIZE - 1;
3460
3461         if (!PageUptodate(page)) {
3462                 ret = btrfs_readpage(NULL, page);
3463                 lock_page(page);
3464                 if (page->mapping != mapping) {
3465                         unlock_page(page);
3466                         page_cache_release(page);
3467                         goto again;
3468                 }
3469                 if (!PageUptodate(page)) {
3470                         ret = -EIO;
3471                         goto out_unlock;
3472                 }
3473         }
3474         wait_on_page_writeback(page);
3475
3476         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
3477         set_page_extent_mapped(page);
3478
3479         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3480         if (ordered) {
3481                 unlock_extent_cached(io_tree, page_start, page_end,
3482                                      &cached_state, GFP_NOFS);
3483                 unlock_page(page);
3484                 page_cache_release(page);
3485                 btrfs_start_ordered_extent(inode, ordered, 1);
3486                 btrfs_put_ordered_extent(ordered);
3487                 goto again;
3488         }
3489
3490         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3491                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3492                           0, 0, &cached_state, GFP_NOFS);
3493
3494         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3495                                         &cached_state);
3496         if (ret) {
3497                 unlock_extent_cached(io_tree, page_start, page_end,
3498                                      &cached_state, GFP_NOFS);
3499                 goto out_unlock;
3500         }
3501
3502         ret = 0;
3503         if (offset != PAGE_CACHE_SIZE) {
3504                 kaddr = kmap(page);
3505                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3506                 flush_dcache_page(page);
3507                 kunmap(page);
3508         }
3509         ClearPageChecked(page);
3510         set_page_dirty(page);
3511         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3512                              GFP_NOFS);
3513
3514 out_unlock:
3515         if (ret)
3516                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3517         unlock_page(page);
3518         page_cache_release(page);
3519 out:
3520         return ret;
3521 }
3522
3523 /*
3524  * This function puts in dummy file extents for the area we're creating a hole
3525  * for.  So if we are truncating this file to a larger size we need to insert
3526  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3527  * the range between oldsize and size
3528  */
3529 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3530 {
3531         struct btrfs_trans_handle *trans;
3532         struct btrfs_root *root = BTRFS_I(inode)->root;
3533         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3534         struct extent_map *em = NULL;
3535         struct extent_state *cached_state = NULL;
3536         u64 mask = root->sectorsize - 1;
3537         u64 hole_start = (oldsize + mask) & ~mask;
3538         u64 block_end = (size + mask) & ~mask;
3539         u64 last_byte;
3540         u64 cur_offset;
3541         u64 hole_size;
3542         int err = 0;
3543
3544         if (size <= hole_start)
3545                 return 0;
3546
3547         while (1) {
3548                 struct btrfs_ordered_extent *ordered;
3549                 btrfs_wait_ordered_range(inode, hole_start,
3550                                          block_end - hole_start);
3551                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3552                                  &cached_state);
3553                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3554                 if (!ordered)
3555                         break;
3556                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3557                                      &cached_state, GFP_NOFS);
3558                 btrfs_put_ordered_extent(ordered);
3559         }
3560
3561         cur_offset = hole_start;
3562         while (1) {
3563                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3564                                 block_end - cur_offset, 0);
3565                 if (IS_ERR(em)) {
3566                         err = PTR_ERR(em);
3567                         break;
3568                 }
3569                 last_byte = min(extent_map_end(em), block_end);
3570                 last_byte = (last_byte + mask) & ~mask;
3571                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3572                         u64 hint_byte = 0;
3573                         hole_size = last_byte - cur_offset;
3574
3575                         trans = btrfs_start_transaction(root, 3);
3576                         if (IS_ERR(trans)) {
3577                                 err = PTR_ERR(trans);
3578                                 break;
3579                         }
3580
3581                         err = btrfs_drop_extents(trans, inode, cur_offset,
3582                                                  cur_offset + hole_size,
3583                                                  &hint_byte, 1);
3584                         if (err) {
3585                                 btrfs_abort_transaction(trans, root, err);
3586                                 btrfs_end_transaction(trans, root);
3587                                 break;
3588                         }
3589
3590                         err = btrfs_insert_file_extent(trans, root,
3591                                         btrfs_ino(inode), cur_offset, 0,
3592                                         0, hole_size, 0, hole_size,
3593                                         0, 0, 0);
3594                         if (err) {
3595                                 btrfs_abort_transaction(trans, root, err);
3596                                 btrfs_end_transaction(trans, root);
3597                                 break;
3598                         }
3599
3600                         btrfs_drop_extent_cache(inode, hole_start,
3601                                         last_byte - 1, 0);
3602
3603                         btrfs_update_inode(trans, root, inode);
3604                         btrfs_end_transaction(trans, root);
3605                 }
3606                 free_extent_map(em);
3607                 em = NULL;
3608                 cur_offset = last_byte;
3609                 if (cur_offset >= block_end)
3610                         break;
3611         }
3612
3613         free_extent_map(em);
3614         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3615                              GFP_NOFS);
3616         return err;
3617 }
3618
3619 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3620 {
3621         struct btrfs_root *root = BTRFS_I(inode)->root;
3622         struct btrfs_trans_handle *trans;
3623         loff_t oldsize = i_size_read(inode);
3624         int ret;
3625
3626         if (newsize == oldsize)
3627                 return 0;
3628
3629         if (newsize > oldsize) {
3630                 truncate_pagecache(inode, oldsize, newsize);
3631                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3632                 if (ret)
3633                         return ret;
3634
3635                 trans = btrfs_start_transaction(root, 1);
3636                 if (IS_ERR(trans))
3637                         return PTR_ERR(trans);
3638
3639                 i_size_write(inode, newsize);
3640                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3641                 ret = btrfs_update_inode(trans, root, inode);
3642                 btrfs_end_transaction(trans, root);
3643         } else {
3644
3645                 /*
3646                  * We're truncating a file that used to have good data down to
3647                  * zero. Make sure it gets into the ordered flush list so that
3648                  * any new writes get down to disk quickly.
3649                  */
3650                 if (newsize == 0)
3651                         set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
3652                                 &BTRFS_I(inode)->runtime_flags);
3653
3654                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3655                 truncate_setsize(inode, newsize);
3656                 ret = btrfs_truncate(inode);
3657         }
3658
3659         return ret;
3660 }
3661
3662 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3663 {
3664         struct inode *inode = dentry->d_inode;
3665         struct btrfs_root *root = BTRFS_I(inode)->root;
3666         int err;
3667
3668         if (btrfs_root_readonly(root))
3669                 return -EROFS;
3670
3671         err = inode_change_ok(inode, attr);
3672         if (err)
3673                 return err;
3674
3675         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3676                 err = btrfs_setsize(inode, attr->ia_size);
3677                 if (err)
3678                         return err;
3679         }
3680
3681         if (attr->ia_valid) {
3682                 setattr_copy(inode, attr);
3683                 inode_inc_iversion(inode);
3684                 err = btrfs_dirty_inode(inode);
3685
3686                 if (!err && attr->ia_valid & ATTR_MODE)
3687                         err = btrfs_acl_chmod(inode);
3688         }
3689
3690         return err;
3691 }
3692
3693 void btrfs_evict_inode(struct inode *inode)
3694 {
3695         struct btrfs_trans_handle *trans;
3696         struct btrfs_root *root = BTRFS_I(inode)->root;
3697         struct btrfs_block_rsv *rsv, *global_rsv;
3698         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
3699         unsigned long nr;
3700         int ret;
3701
3702         trace_btrfs_inode_evict(inode);
3703
3704         truncate_inode_pages(&inode->i_data, 0);
3705         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3706                                btrfs_is_free_space_inode(root, inode)))
3707                 goto no_delete;
3708
3709         if (is_bad_inode(inode)) {
3710                 btrfs_orphan_del(NULL, inode);
3711                 goto no_delete;
3712         }
3713         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3714         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3715
3716         if (root->fs_info->log_root_recovering) {
3717                 BUG_ON(!test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3718                                  &BTRFS_I(inode)->runtime_flags));
3719                 goto no_delete;
3720         }
3721
3722         if (inode->i_nlink > 0) {
3723                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3724                 goto no_delete;
3725         }
3726
3727         rsv = btrfs_alloc_block_rsv(root);
3728         if (!rsv) {
3729                 btrfs_orphan_del(NULL, inode);
3730                 goto no_delete;
3731         }
3732         rsv->size = min_size;
3733         global_rsv = &root->fs_info->global_block_rsv;
3734
3735         btrfs_i_size_write(inode, 0);
3736
3737         /*
3738          * This is a bit simpler than btrfs_truncate since
3739          *
3740          * 1) We've already reserved our space for our orphan item in the
3741          *    unlink.
3742          * 2) We're going to delete the inode item, so we don't need to update
3743          *    it at all.
3744          *
3745          * So we just need to reserve some slack space in case we add bytes when
3746          * doing the truncate.
3747          */
3748         while (1) {
3749                 ret = btrfs_block_rsv_refill_noflush(root, rsv, min_size);
3750
3751                 /*
3752                  * Try and steal from the global reserve since we will
3753                  * likely not use this space anyway, we want to try as
3754                  * hard as possible to get this to work.
3755                  */
3756                 if (ret)
3757                         ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
3758
3759                 if (ret) {
3760                         printk(KERN_WARNING "Could not get space for a "
3761                                "delete, will truncate on mount %d\n", ret);
3762                         btrfs_orphan_del(NULL, inode);
3763                         btrfs_free_block_rsv(root, rsv);
3764                         goto no_delete;
3765                 }
3766
3767                 trans = btrfs_start_transaction(root, 0);
3768                 if (IS_ERR(trans)) {
3769                         btrfs_orphan_del(NULL, inode);
3770                         btrfs_free_block_rsv(root, rsv);
3771                         goto no_delete;
3772                 }
3773
3774                 trans->block_rsv = rsv;
3775
3776                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3777                 if (ret != -EAGAIN)
3778                         break;
3779
3780                 nr = trans->blocks_used;
3781                 btrfs_end_transaction(trans, root);
3782                 trans = NULL;
3783                 btrfs_btree_balance_dirty(root, nr);
3784         }
3785
3786         btrfs_free_block_rsv(root, rsv);
3787
3788         if (ret == 0) {
3789                 trans->block_rsv = root->orphan_block_rsv;
3790                 ret = btrfs_orphan_del(trans, inode);
3791                 BUG_ON(ret);
3792         }
3793
3794         trans->block_rsv = &root->fs_info->trans_block_rsv;
3795         if (!(root == root->fs_info->tree_root ||
3796               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3797                 btrfs_return_ino(root, btrfs_ino(inode));
3798
3799         nr = trans->blocks_used;
3800         btrfs_end_transaction(trans, root);
3801         btrfs_btree_balance_dirty(root, nr);
3802 no_delete:
3803         clear_inode(inode);
3804         return;
3805 }
3806
3807 /*
3808  * this returns the key found in the dir entry in the location pointer.
3809  * If no dir entries were found, location->objectid is 0.
3810  */
3811 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3812                                struct btrfs_key *location)
3813 {
3814         const char *name = dentry->d_name.name;
3815         int namelen = dentry->d_name.len;
3816         struct btrfs_dir_item *di;
3817         struct btrfs_path *path;
3818         struct btrfs_root *root = BTRFS_I(dir)->root;
3819         int ret = 0;
3820
3821         path = btrfs_alloc_path();
3822         if (!path)
3823                 return -ENOMEM;
3824
3825         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3826                                     namelen, 0);
3827         if (IS_ERR(di))
3828                 ret = PTR_ERR(di);
3829
3830         if (IS_ERR_OR_NULL(di))
3831                 goto out_err;
3832
3833         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3834 out:
3835         btrfs_free_path(path);
3836         return ret;
3837 out_err:
3838         location->objectid = 0;
3839         goto out;
3840 }
3841
3842 /*
3843  * when we hit a tree root in a directory, the btrfs part of the inode
3844  * needs to be changed to reflect the root directory of the tree root.  This
3845  * is kind of like crossing a mount point.
3846  */
3847 static int fixup_tree_root_location(struct btrfs_root *root,
3848                                     struct inode *dir,
3849                                     struct dentry *dentry,
3850                                     struct btrfs_key *location,
3851                                     struct btrfs_root **sub_root)
3852 {
3853         struct btrfs_path *path;
3854         struct btrfs_root *new_root;
3855         struct btrfs_root_ref *ref;
3856         struct extent_buffer *leaf;
3857         int ret;
3858         int err = 0;
3859
3860         path = btrfs_alloc_path();
3861         if (!path) {
3862                 err = -ENOMEM;
3863                 goto out;
3864         }
3865
3866         err = -ENOENT;
3867         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3868                                   BTRFS_I(dir)->root->root_key.objectid,
3869                                   location->objectid);
3870         if (ret) {
3871                 if (ret < 0)
3872                         err = ret;
3873                 goto out;
3874         }
3875
3876         leaf = path->nodes[0];
3877         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3878         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3879             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3880                 goto out;
3881
3882         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3883                                    (unsigned long)(ref + 1),
3884                                    dentry->d_name.len);
3885         if (ret)
3886                 goto out;
3887
3888         btrfs_release_path(path);
3889
3890         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3891         if (IS_ERR(new_root)) {
3892                 err = PTR_ERR(new_root);
3893                 goto out;
3894         }
3895
3896         if (btrfs_root_refs(&new_root->root_item) == 0) {
3897                 err = -ENOENT;
3898                 goto out;
3899         }
3900
3901         *sub_root = new_root;
3902         location->objectid = btrfs_root_dirid(&new_root->root_item);
3903         location->type = BTRFS_INODE_ITEM_KEY;
3904         location->offset = 0;
3905         err = 0;
3906 out:
3907         btrfs_free_path(path);
3908         return err;
3909 }
3910
3911 static void inode_tree_add(struct inode *inode)
3912 {
3913         struct btrfs_root *root = BTRFS_I(inode)->root;
3914         struct btrfs_inode *entry;
3915         struct rb_node **p;
3916         struct rb_node *parent;
3917         u64 ino = btrfs_ino(inode);
3918 again:
3919         p = &root->inode_tree.rb_node;
3920         parent = NULL;
3921
3922         if (inode_unhashed(inode))
3923                 return;
3924
3925         spin_lock(&root->inode_lock);
3926         while (*p) {
3927                 parent = *p;
3928                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3929
3930                 if (ino < btrfs_ino(&entry->vfs_inode))
3931                         p = &parent->rb_left;
3932                 else if (ino > btrfs_ino(&entry->vfs_inode))
3933                         p = &parent->rb_right;
3934                 else {
3935                         WARN_ON(!(entry->vfs_inode.i_state &
3936                                   (I_WILL_FREE | I_FREEING)));
3937                         rb_erase(parent, &root->inode_tree);
3938                         RB_CLEAR_NODE(parent);
3939                         spin_unlock(&root->inode_lock);
3940                         goto again;
3941                 }
3942         }
3943         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3944         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3945         spin_unlock(&root->inode_lock);
3946 }
3947
3948 static void inode_tree_del(struct inode *inode)
3949 {
3950         struct btrfs_root *root = BTRFS_I(inode)->root;
3951         int empty = 0;
3952
3953         spin_lock(&root->inode_lock);
3954         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3955                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3956                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3957                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3958         }
3959         spin_unlock(&root->inode_lock);
3960
3961         /*
3962          * Free space cache has inodes in the tree root, but the tree root has a
3963          * root_refs of 0, so this could end up dropping the tree root as a
3964          * snapshot, so we need the extra !root->fs_info->tree_root check to
3965          * make sure we don't drop it.
3966          */
3967         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3968             root != root->fs_info->tree_root) {
3969                 synchronize_srcu(&root->fs_info->subvol_srcu);
3970                 spin_lock(&root->inode_lock);
3971                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3972                 spin_unlock(&root->inode_lock);
3973                 if (empty)
3974                         btrfs_add_dead_root(root);
3975         }
3976 }
3977
3978 void btrfs_invalidate_inodes(struct btrfs_root *root)
3979 {
3980         struct rb_node *node;
3981         struct rb_node *prev;
3982         struct btrfs_inode *entry;
3983         struct inode *inode;
3984         u64 objectid = 0;
3985
3986         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3987
3988         spin_lock(&root->inode_lock);
3989 again:
3990         node = root->inode_tree.rb_node;
3991         prev = NULL;
3992         while (node) {
3993                 prev = node;
3994                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3995
3996                 if (objectid < btrfs_ino(&entry->vfs_inode))
3997                         node = node->rb_left;
3998                 else if (objectid > btrfs_ino(&entry->vfs_inode))
3999                         node = node->rb_right;
4000                 else
4001                         break;
4002         }
4003         if (!node) {
4004                 while (prev) {
4005                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
4006                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4007                                 node = prev;
4008                                 break;
4009                         }
4010                         prev = rb_next(prev);
4011                 }
4012         }
4013         while (node) {
4014                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4015                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4016                 inode = igrab(&entry->vfs_inode);
4017                 if (inode) {
4018                         spin_unlock(&root->inode_lock);
4019                         if (atomic_read(&inode->i_count) > 1)
4020                                 d_prune_aliases(inode);
4021                         /*
4022                          * btrfs_drop_inode will have it removed from
4023                          * the inode cache when its usage count
4024                          * hits zero.
4025                          */
4026                         iput(inode);
4027                         cond_resched();
4028                         spin_lock(&root->inode_lock);
4029                         goto again;
4030                 }
4031
4032                 if (cond_resched_lock(&root->inode_lock))
4033                         goto again;
4034
4035                 node = rb_next(node);
4036         }
4037         spin_unlock(&root->inode_lock);
4038 }
4039
4040 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4041 {
4042         struct btrfs_iget_args *args = p;
4043         inode->i_ino = args->ino;
4044         BTRFS_I(inode)->root = args->root;
4045         btrfs_set_inode_space_info(args->root, inode);
4046         return 0;
4047 }
4048
4049 static int btrfs_find_actor(struct inode *inode, void *opaque)
4050 {
4051         struct btrfs_iget_args *args = opaque;
4052         return args->ino == btrfs_ino(inode) &&
4053                 args->root == BTRFS_I(inode)->root;
4054 }
4055
4056 static struct inode *btrfs_iget_locked(struct super_block *s,
4057                                        u64 objectid,
4058                                        struct btrfs_root *root)
4059 {
4060         struct inode *inode;
4061         struct btrfs_iget_args args;
4062         args.ino = objectid;
4063         args.root = root;
4064
4065         inode = iget5_locked(s, objectid, btrfs_find_actor,
4066                              btrfs_init_locked_inode,
4067                              (void *)&args);
4068         return inode;
4069 }
4070
4071 /* Get an inode object given its location and corresponding root.
4072  * Returns in *is_new if the inode was read from disk
4073  */
4074 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4075                          struct btrfs_root *root, int *new)
4076 {
4077         struct inode *inode;
4078
4079         inode = btrfs_iget_locked(s, location->objectid, root);
4080         if (!inode)
4081                 return ERR_PTR(-ENOMEM);
4082
4083         if (inode->i_state & I_NEW) {
4084                 BTRFS_I(inode)->root = root;
4085                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4086                 btrfs_read_locked_inode(inode);
4087                 if (!is_bad_inode(inode)) {
4088                         inode_tree_add(inode);
4089                         unlock_new_inode(inode);
4090                         if (new)
4091                                 *new = 1;
4092                 } else {
4093                         unlock_new_inode(inode);
4094                         iput(inode);
4095                         inode = ERR_PTR(-ESTALE);
4096                 }
4097         }
4098
4099         return inode;
4100 }
4101
4102 static struct inode *new_simple_dir(struct super_block *s,
4103                                     struct btrfs_key *key,
4104                                     struct btrfs_root *root)
4105 {
4106         struct inode *inode = new_inode(s);
4107
4108         if (!inode)
4109                 return ERR_PTR(-ENOMEM);
4110
4111         BTRFS_I(inode)->root = root;
4112         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4113         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4114
4115         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4116         inode->i_op = &btrfs_dir_ro_inode_operations;
4117         inode->i_fop = &simple_dir_operations;
4118         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4119         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4120
4121         return inode;
4122 }
4123
4124 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4125 {
4126         struct inode *inode;
4127         struct btrfs_root *root = BTRFS_I(dir)->root;
4128         struct btrfs_root *sub_root = root;
4129         struct btrfs_key location;
4130         int index;
4131         int ret = 0;
4132
4133         if (dentry->d_name.len > BTRFS_NAME_LEN)
4134                 return ERR_PTR(-ENAMETOOLONG);
4135
4136         if (unlikely(d_need_lookup(dentry))) {
4137                 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
4138                 kfree(dentry->d_fsdata);
4139                 dentry->d_fsdata = NULL;
4140                 /* This thing is hashed, drop it for now */
4141                 d_drop(dentry);
4142         } else {
4143                 ret = btrfs_inode_by_name(dir, dentry, &location);
4144         }
4145
4146         if (ret < 0)
4147                 return ERR_PTR(ret);
4148
4149         if (location.objectid == 0)
4150                 return NULL;
4151
4152         if (location.type == BTRFS_INODE_ITEM_KEY) {
4153                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4154                 return inode;
4155         }
4156
4157         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4158
4159         index = srcu_read_lock(&root->fs_info->subvol_srcu);
4160         ret = fixup_tree_root_location(root, dir, dentry,
4161                                        &location, &sub_root);
4162         if (ret < 0) {
4163                 if (ret != -ENOENT)
4164                         inode = ERR_PTR(ret);
4165                 else
4166                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
4167         } else {
4168                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4169         }
4170         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4171
4172         if (!IS_ERR(inode) && root != sub_root) {
4173                 down_read(&root->fs_info->cleanup_work_sem);
4174                 if (!(inode->i_sb->s_flags & MS_RDONLY))
4175                         ret = btrfs_orphan_cleanup(sub_root);
4176                 up_read(&root->fs_info->cleanup_work_sem);
4177                 if (ret)
4178                         inode = ERR_PTR(ret);
4179         }
4180
4181         return inode;
4182 }
4183
4184 static int btrfs_dentry_delete(const struct dentry *dentry)
4185 {
4186         struct btrfs_root *root;
4187         struct inode *inode = dentry->d_inode;
4188
4189         if (!inode && !IS_ROOT(dentry))
4190                 inode = dentry->d_parent->d_inode;
4191
4192         if (inode) {
4193                 root = BTRFS_I(inode)->root;
4194                 if (btrfs_root_refs(&root->root_item) == 0)
4195                         return 1;
4196
4197                 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
4198                         return 1;
4199         }
4200         return 0;
4201 }
4202
4203 static void btrfs_dentry_release(struct dentry *dentry)
4204 {
4205         if (dentry->d_fsdata)
4206                 kfree(dentry->d_fsdata);
4207 }
4208
4209 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4210                                    struct nameidata *nd)
4211 {
4212         struct dentry *ret;
4213
4214         ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4215         if (unlikely(d_need_lookup(dentry))) {
4216                 spin_lock(&dentry->d_lock);
4217                 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4218                 spin_unlock(&dentry->d_lock);
4219         }
4220         return ret;
4221 }
4222
4223 unsigned char btrfs_filetype_table[] = {
4224         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4225 };
4226
4227 static int btrfs_real_readdir(struct file *filp, void *dirent,
4228                               filldir_t filldir)
4229 {
4230         struct inode *inode = filp->f_dentry->d_inode;
4231         struct btrfs_root *root = BTRFS_I(inode)->root;
4232         struct btrfs_item *item;
4233         struct btrfs_dir_item *di;
4234         struct btrfs_key key;
4235         struct btrfs_key found_key;
4236         struct btrfs_path *path;
4237         struct list_head ins_list;
4238         struct list_head del_list;
4239         int ret;
4240         struct extent_buffer *leaf;
4241         int slot;
4242         unsigned char d_type;
4243         int over = 0;
4244         u32 di_cur;
4245         u32 di_total;
4246         u32 di_len;
4247         int key_type = BTRFS_DIR_INDEX_KEY;
4248         char tmp_name[32];
4249         char *name_ptr;
4250         int name_len;
4251         int is_curr = 0;        /* filp->f_pos points to the current index? */
4252
4253         /* FIXME, use a real flag for deciding about the key type */
4254         if (root->fs_info->tree_root == root)
4255                 key_type = BTRFS_DIR_ITEM_KEY;
4256
4257         /* special case for "." */
4258         if (filp->f_pos == 0) {
4259                 over = filldir(dirent, ".", 1,
4260                                filp->f_pos, btrfs_ino(inode), DT_DIR);
4261                 if (over)
4262                         return 0;
4263                 filp->f_pos = 1;
4264         }
4265         /* special case for .., just use the back ref */
4266         if (filp->f_pos == 1) {
4267                 u64 pino = parent_ino(filp->f_path.dentry);
4268                 over = filldir(dirent, "..", 2,
4269                                filp->f_pos, pino, DT_DIR);
4270                 if (over)
4271                         return 0;
4272                 filp->f_pos = 2;
4273         }
4274         path = btrfs_alloc_path();
4275         if (!path)
4276                 return -ENOMEM;
4277
4278         path->reada = 1;
4279
4280         if (key_type == BTRFS_DIR_INDEX_KEY) {
4281                 INIT_LIST_HEAD(&ins_list);
4282                 INIT_LIST_HEAD(&del_list);
4283                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4284         }
4285
4286         btrfs_set_key_type(&key, key_type);
4287         key.offset = filp->f_pos;
4288         key.objectid = btrfs_ino(inode);
4289
4290         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4291         if (ret < 0)
4292                 goto err;
4293
4294         while (1) {
4295                 leaf = path->nodes[0];
4296                 slot = path->slots[0];
4297                 if (slot >= btrfs_header_nritems(leaf)) {
4298                         ret = btrfs_next_leaf(root, path);
4299                         if (ret < 0)
4300                                 goto err;
4301                         else if (ret > 0)
4302                                 break;
4303                         continue;
4304                 }
4305
4306                 item = btrfs_item_nr(leaf, slot);
4307                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4308
4309                 if (found_key.objectid != key.objectid)
4310                         break;
4311                 if (btrfs_key_type(&found_key) != key_type)
4312                         break;
4313                 if (found_key.offset < filp->f_pos)
4314                         goto next;
4315                 if (key_type == BTRFS_DIR_INDEX_KEY &&
4316                     btrfs_should_delete_dir_index(&del_list,
4317                                                   found_key.offset))
4318                         goto next;
4319
4320                 filp->f_pos = found_key.offset;
4321                 is_curr = 1;
4322
4323                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4324                 di_cur = 0;
4325                 di_total = btrfs_item_size(leaf, item);
4326
4327                 while (di_cur < di_total) {
4328                         struct btrfs_key location;
4329
4330                         if (verify_dir_item(root, leaf, di))
4331                                 break;
4332
4333                         name_len = btrfs_dir_name_len(leaf, di);
4334                         if (name_len <= sizeof(tmp_name)) {
4335                                 name_ptr = tmp_name;
4336                         } else {
4337                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4338                                 if (!name_ptr) {
4339                                         ret = -ENOMEM;
4340                                         goto err;
4341                                 }
4342                         }
4343                         read_extent_buffer(leaf, name_ptr,
4344                                            (unsigned long)(di + 1), name_len);
4345
4346                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4347                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4348
4349
4350                         /* is this a reference to our own snapshot? If so
4351                          * skip it.
4352                          *
4353                          * In contrast to old kernels, we insert the snapshot's
4354                          * dir item and dir index after it has been created, so
4355                          * we won't find a reference to our own snapshot. We
4356                          * still keep the following code for backward
4357                          * compatibility.
4358                          */
4359                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4360                             location.objectid == root->root_key.objectid) {
4361                                 over = 0;
4362                                 goto skip;
4363                         }
4364                         over = filldir(dirent, name_ptr, name_len,
4365                                        found_key.offset, location.objectid,
4366                                        d_type);
4367
4368 skip:
4369                         if (name_ptr != tmp_name)
4370                                 kfree(name_ptr);
4371
4372                         if (over)
4373                                 goto nopos;
4374                         di_len = btrfs_dir_name_len(leaf, di) +
4375                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4376                         di_cur += di_len;
4377                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4378                 }
4379 next:
4380                 path->slots[0]++;
4381         }
4382
4383         if (key_type == BTRFS_DIR_INDEX_KEY) {
4384                 if (is_curr)
4385                         filp->f_pos++;
4386                 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4387                                                       &ins_list);
4388                 if (ret)
4389                         goto nopos;
4390         }
4391
4392         /* Reached end of directory/root. Bump pos past the last item. */
4393         if (key_type == BTRFS_DIR_INDEX_KEY)
4394                 /*
4395                  * 32-bit glibc will use getdents64, but then strtol -
4396                  * so the last number we can serve is this.
4397                  */
4398                 filp->f_pos = 0x7fffffff;
4399         else
4400                 filp->f_pos++;
4401 nopos:
4402         ret = 0;
4403 err:
4404         if (key_type == BTRFS_DIR_INDEX_KEY)
4405                 btrfs_put_delayed_items(&ins_list, &del_list);
4406         btrfs_free_path(path);
4407         return ret;
4408 }
4409
4410 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4411 {
4412         struct btrfs_root *root = BTRFS_I(inode)->root;
4413         struct btrfs_trans_handle *trans;
4414         int ret = 0;
4415         bool nolock = false;
4416
4417         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4418                 return 0;
4419
4420         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(root, inode))
4421                 nolock = true;
4422
4423         if (wbc->sync_mode == WB_SYNC_ALL) {
4424                 if (nolock)
4425                         trans = btrfs_join_transaction_nolock(root);
4426                 else
4427                         trans = btrfs_join_transaction(root);
4428                 if (IS_ERR(trans))
4429                         return PTR_ERR(trans);
4430                 if (nolock)
4431                         ret = btrfs_end_transaction_nolock(trans, root);
4432                 else
4433                         ret = btrfs_commit_transaction(trans, root);
4434         }
4435         return ret;
4436 }
4437
4438 /*
4439  * This is somewhat expensive, updating the tree every time the
4440  * inode changes.  But, it is most likely to find the inode in cache.
4441  * FIXME, needs more benchmarking...there are no reasons other than performance
4442  * to keep or drop this code.
4443  */
4444 int btrfs_dirty_inode(struct inode *inode)
4445 {
4446         struct btrfs_root *root = BTRFS_I(inode)->root;
4447         struct btrfs_trans_handle *trans;
4448         int ret;
4449
4450         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4451                 return 0;
4452
4453         trans = btrfs_join_transaction(root);
4454         if (IS_ERR(trans))
4455                 return PTR_ERR(trans);
4456
4457         ret = btrfs_update_inode(trans, root, inode);
4458         if (ret && ret == -ENOSPC) {
4459                 /* whoops, lets try again with the full transaction */
4460                 btrfs_end_transaction(trans, root);
4461                 trans = btrfs_start_transaction(root, 1);
4462                 if (IS_ERR(trans))
4463                         return PTR_ERR(trans);
4464
4465                 ret = btrfs_update_inode(trans, root, inode);
4466         }
4467         btrfs_end_transaction(trans, root);
4468         if (BTRFS_I(inode)->delayed_node)
4469                 btrfs_balance_delayed_items(root);
4470
4471         return ret;
4472 }
4473
4474 /*
4475  * This is a copy of file_update_time.  We need this so we can return error on
4476  * ENOSPC for updating the inode in the case of file write and mmap writes.
4477  */
4478 int btrfs_update_time(struct file *file)
4479 {
4480         struct inode *inode = file->f_path.dentry->d_inode;
4481         struct timespec now;
4482         int ret;
4483         enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
4484
4485         /* First try to exhaust all avenues to not sync */
4486         if (IS_NOCMTIME(inode))
4487                 return 0;
4488
4489         now = current_fs_time(inode->i_sb);
4490         if (!timespec_equal(&inode->i_mtime, &now))
4491                 sync_it = S_MTIME;
4492
4493         if (!timespec_equal(&inode->i_ctime, &now))
4494                 sync_it |= S_CTIME;
4495
4496         if (IS_I_VERSION(inode))
4497                 sync_it |= S_VERSION;
4498
4499         if (!sync_it)
4500                 return 0;
4501
4502         /* Finally allowed to write? Takes lock. */
4503         if (mnt_want_write_file(file))
4504                 return 0;
4505
4506         /* Only change inode inside the lock region */
4507         if (sync_it & S_VERSION)
4508                 inode_inc_iversion(inode);
4509         if (sync_it & S_CTIME)
4510                 inode->i_ctime = now;
4511         if (sync_it & S_MTIME)
4512                 inode->i_mtime = now;
4513         ret = btrfs_dirty_inode(inode);
4514         if (!ret)
4515                 mark_inode_dirty_sync(inode);
4516         mnt_drop_write(file->f_path.mnt);
4517         return ret;
4518 }
4519
4520 /*
4521  * find the highest existing sequence number in a directory
4522  * and then set the in-memory index_cnt variable to reflect
4523  * free sequence numbers
4524  */
4525 static int btrfs_set_inode_index_count(struct inode *inode)
4526 {
4527         struct btrfs_root *root = BTRFS_I(inode)->root;
4528         struct btrfs_key key, found_key;
4529         struct btrfs_path *path;
4530         struct extent_buffer *leaf;
4531         int ret;
4532
4533         key.objectid = btrfs_ino(inode);
4534         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4535         key.offset = (u64)-1;
4536
4537         path = btrfs_alloc_path();
4538         if (!path)
4539                 return -ENOMEM;
4540
4541         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4542         if (ret < 0)
4543                 goto out;
4544         /* FIXME: we should be able to handle this */
4545         if (ret == 0)
4546                 goto out;
4547         ret = 0;
4548
4549         /*
4550          * MAGIC NUMBER EXPLANATION:
4551          * since we search a directory based on f_pos we have to start at 2
4552          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4553          * else has to start at 2
4554          */
4555         if (path->slots[0] == 0) {
4556                 BTRFS_I(inode)->index_cnt = 2;
4557                 goto out;
4558         }
4559
4560         path->slots[0]--;
4561
4562         leaf = path->nodes[0];
4563         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4564
4565         if (found_key.objectid != btrfs_ino(inode) ||
4566             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4567                 BTRFS_I(inode)->index_cnt = 2;
4568                 goto out;
4569         }
4570
4571         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4572 out:
4573         btrfs_free_path(path);
4574         return ret;
4575 }
4576
4577 /*
4578  * helper to find a free sequence number in a given directory.  This current
4579  * code is very simple, later versions will do smarter things in the btree
4580  */
4581 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4582 {
4583         int ret = 0;
4584
4585         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4586                 ret = btrfs_inode_delayed_dir_index_count(dir);
4587                 if (ret) {
4588                         ret = btrfs_set_inode_index_count(dir);
4589                         if (ret)
4590                                 return ret;
4591                 }
4592         }
4593
4594         *index = BTRFS_I(dir)->index_cnt;
4595         BTRFS_I(dir)->index_cnt++;
4596
4597         return ret;
4598 }
4599
4600 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4601                                      struct btrfs_root *root,
4602                                      struct inode *dir,
4603                                      const char *name, int name_len,
4604                                      u64 ref_objectid, u64 objectid,
4605                                      umode_t mode, u64 *index)
4606 {
4607         struct inode *inode;
4608         struct btrfs_inode_item *inode_item;
4609         struct btrfs_key *location;
4610         struct btrfs_path *path;
4611         struct btrfs_inode_ref *ref;
4612         struct btrfs_key key[2];
4613         u32 sizes[2];
4614         unsigned long ptr;
4615         int ret;
4616         int owner;
4617
4618         path = btrfs_alloc_path();
4619         if (!path)
4620                 return ERR_PTR(-ENOMEM);
4621
4622         inode = new_inode(root->fs_info->sb);
4623         if (!inode) {
4624                 btrfs_free_path(path);
4625                 return ERR_PTR(-ENOMEM);
4626         }
4627
4628         /*
4629          * we have to initialize this early, so we can reclaim the inode
4630          * number if we fail afterwards in this function.
4631          */
4632         inode->i_ino = objectid;
4633
4634         if (dir) {
4635                 trace_btrfs_inode_request(dir);
4636
4637                 ret = btrfs_set_inode_index(dir, index);
4638                 if (ret) {
4639                         btrfs_free_path(path);
4640                         iput(inode);
4641                         return ERR_PTR(ret);
4642                 }
4643         }
4644         /*
4645          * index_cnt is ignored for everything but a dir,
4646          * btrfs_get_inode_index_count has an explanation for the magic
4647          * number
4648          */
4649         BTRFS_I(inode)->index_cnt = 2;
4650         BTRFS_I(inode)->root = root;
4651         BTRFS_I(inode)->generation = trans->transid;
4652         inode->i_generation = BTRFS_I(inode)->generation;
4653         btrfs_set_inode_space_info(root, inode);
4654
4655         if (S_ISDIR(mode))
4656                 owner = 0;
4657         else
4658                 owner = 1;
4659
4660         key[0].objectid = objectid;
4661         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4662         key[0].offset = 0;
4663
4664         key[1].objectid = objectid;
4665         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4666         key[1].offset = ref_objectid;
4667
4668         sizes[0] = sizeof(struct btrfs_inode_item);
4669         sizes[1] = name_len + sizeof(*ref);
4670
4671         path->leave_spinning = 1;
4672         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4673         if (ret != 0)
4674                 goto fail;
4675
4676         inode_init_owner(inode, dir, mode);
4677         inode_set_bytes(inode, 0);
4678         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4679         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4680                                   struct btrfs_inode_item);
4681         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4682
4683         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4684                              struct btrfs_inode_ref);
4685         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4686         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4687         ptr = (unsigned long)(ref + 1);
4688         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4689
4690         btrfs_mark_buffer_dirty(path->nodes[0]);
4691         btrfs_free_path(path);
4692
4693         location = &BTRFS_I(inode)->location;
4694         location->objectid = objectid;
4695         location->offset = 0;
4696         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4697
4698         btrfs_inherit_iflags(inode, dir);
4699
4700         if (S_ISREG(mode)) {
4701                 if (btrfs_test_opt(root, NODATASUM))
4702                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4703                 if (btrfs_test_opt(root, NODATACOW) ||
4704                     (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4705                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4706         }
4707
4708         insert_inode_hash(inode);
4709         inode_tree_add(inode);
4710
4711         trace_btrfs_inode_new(inode);
4712         btrfs_set_inode_last_trans(trans, inode);
4713
4714         return inode;
4715 fail:
4716         if (dir)
4717                 BTRFS_I(dir)->index_cnt--;
4718         btrfs_free_path(path);
4719         iput(inode);
4720         return ERR_PTR(ret);
4721 }
4722
4723 static inline u8 btrfs_inode_type(struct inode *inode)
4724 {
4725         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4726 }
4727
4728 /*
4729  * utility function to add 'inode' into 'parent_inode' with
4730  * a give name and a given sequence number.
4731  * if 'add_backref' is true, also insert a backref from the
4732  * inode to the parent directory.
4733  */
4734 int btrfs_add_link(struct btrfs_trans_handle *trans,
4735                    struct inode *parent_inode, struct inode *inode,
4736                    const char *name, int name_len, int add_backref, u64 index)
4737 {
4738         int ret = 0;
4739         struct btrfs_key key;
4740         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4741         u64 ino = btrfs_ino(inode);
4742         u64 parent_ino = btrfs_ino(parent_inode);
4743
4744         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4745                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4746         } else {
4747                 key.objectid = ino;
4748                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4749                 key.offset = 0;
4750         }
4751
4752         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4753                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4754                                          key.objectid, root->root_key.objectid,
4755                                          parent_ino, index, name, name_len);
4756         } else if (add_backref) {
4757                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4758                                              parent_ino, index);
4759         }
4760
4761         /* Nothing to clean up yet */
4762         if (ret)
4763                 return ret;
4764
4765         ret = btrfs_insert_dir_item(trans, root, name, name_len,
4766                                     parent_inode, &key,
4767                                     btrfs_inode_type(inode), index);
4768         if (ret == -EEXIST)
4769                 goto fail_dir_item;
4770         else if (ret) {
4771                 btrfs_abort_transaction(trans, root, ret);
4772                 return ret;
4773         }
4774
4775         btrfs_i_size_write(parent_inode, parent_inode->i_size +
4776                            name_len * 2);
4777         inode_inc_iversion(parent_inode);
4778         parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4779         ret = btrfs_update_inode(trans, root, parent_inode);
4780         if (ret)
4781                 btrfs_abort_transaction(trans, root, ret);
4782         return ret;
4783
4784 fail_dir_item:
4785         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4786                 u64 local_index;
4787                 int err;
4788                 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4789                                  key.objectid, root->root_key.objectid,
4790                                  parent_ino, &local_index, name, name_len);
4791
4792         } else if (add_backref) {
4793                 u64 local_index;
4794                 int err;
4795
4796                 err = btrfs_del_inode_ref(trans, root, name, name_len,
4797                                           ino, parent_ino, &local_index);
4798         }
4799         return ret;
4800 }
4801
4802 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4803                             struct inode *dir, struct dentry *dentry,
4804                             struct inode *inode, int backref, u64 index)
4805 {
4806         int err = btrfs_add_link(trans, dir, inode,
4807                                  dentry->d_name.name, dentry->d_name.len,
4808                                  backref, index);
4809         if (err > 0)
4810                 err = -EEXIST;
4811         return err;
4812 }
4813
4814 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4815                         umode_t mode, dev_t rdev)
4816 {
4817         struct btrfs_trans_handle *trans;
4818         struct btrfs_root *root = BTRFS_I(dir)->root;
4819         struct inode *inode = NULL;
4820         int err;
4821         int drop_inode = 0;
4822         u64 objectid;
4823         unsigned long nr = 0;
4824         u64 index = 0;
4825
4826         if (!new_valid_dev(rdev))
4827                 return -EINVAL;
4828
4829         /*
4830          * 2 for inode item and ref
4831          * 2 for dir items
4832          * 1 for xattr if selinux is on
4833          */
4834         trans = btrfs_start_transaction(root, 5);
4835         if (IS_ERR(trans))
4836                 return PTR_ERR(trans);
4837
4838         err = btrfs_find_free_ino(root, &objectid);
4839         if (err)
4840                 goto out_unlock;
4841
4842         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4843                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4844                                 mode, &index);
4845         if (IS_ERR(inode)) {
4846                 err = PTR_ERR(inode);
4847                 goto out_unlock;
4848         }
4849
4850         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4851         if (err) {
4852                 drop_inode = 1;
4853                 goto out_unlock;
4854         }
4855
4856         /*
4857         * If the active LSM wants to access the inode during
4858         * d_instantiate it needs these. Smack checks to see
4859         * if the filesystem supports xattrs by looking at the
4860         * ops vector.
4861         */
4862
4863         inode->i_op = &btrfs_special_inode_operations;
4864         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4865         if (err)
4866                 drop_inode = 1;
4867         else {
4868                 init_special_inode(inode, inode->i_mode, rdev);
4869                 btrfs_update_inode(trans, root, inode);
4870                 d_instantiate(dentry, inode);
4871         }
4872 out_unlock:
4873         nr = trans->blocks_used;
4874         btrfs_end_transaction(trans, root);
4875         btrfs_btree_balance_dirty(root, nr);
4876         if (drop_inode) {
4877                 inode_dec_link_count(inode);
4878                 iput(inode);
4879         }
4880         return err;
4881 }
4882
4883 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4884                         umode_t mode, struct nameidata *nd)
4885 {
4886         struct btrfs_trans_handle *trans;
4887         struct btrfs_root *root = BTRFS_I(dir)->root;
4888         struct inode *inode = NULL;
4889         int drop_inode = 0;
4890         int err;
4891         unsigned long nr = 0;
4892         u64 objectid;
4893         u64 index = 0;
4894
4895         /*
4896          * 2 for inode item and ref
4897          * 2 for dir items
4898          * 1 for xattr if selinux is on
4899          */
4900         trans = btrfs_start_transaction(root, 5);
4901         if (IS_ERR(trans))
4902                 return PTR_ERR(trans);
4903
4904         err = btrfs_find_free_ino(root, &objectid);
4905         if (err)
4906                 goto out_unlock;
4907
4908         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4909                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4910                                 mode, &index);
4911         if (IS_ERR(inode)) {
4912                 err = PTR_ERR(inode);
4913                 goto out_unlock;
4914         }
4915
4916         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4917         if (err) {
4918                 drop_inode = 1;
4919                 goto out_unlock;
4920         }
4921
4922         /*
4923         * If the active LSM wants to access the inode during
4924         * d_instantiate it needs these. Smack checks to see
4925         * if the filesystem supports xattrs by looking at the
4926         * ops vector.
4927         */
4928         inode->i_fop = &btrfs_file_operations;
4929         inode->i_op = &btrfs_file_inode_operations;
4930
4931         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4932         if (err)
4933                 drop_inode = 1;
4934         else {
4935                 inode->i_mapping->a_ops = &btrfs_aops;
4936                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4937                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4938                 d_instantiate(dentry, inode);
4939         }
4940 out_unlock:
4941         nr = trans->blocks_used;
4942         btrfs_end_transaction(trans, root);
4943         if (drop_inode) {
4944                 inode_dec_link_count(inode);
4945                 iput(inode);
4946         }
4947         btrfs_btree_balance_dirty(root, nr);
4948         return err;
4949 }
4950
4951 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4952                       struct dentry *dentry)
4953 {
4954         struct btrfs_trans_handle *trans;
4955         struct btrfs_root *root = BTRFS_I(dir)->root;
4956         struct inode *inode = old_dentry->d_inode;
4957         u64 index;
4958         unsigned long nr = 0;
4959         int err;
4960         int drop_inode = 0;
4961
4962         /* do not allow sys_link's with other subvols of the same device */
4963         if (root->objectid != BTRFS_I(inode)->root->objectid)
4964                 return -EXDEV;
4965
4966         if (inode->i_nlink == ~0U)
4967                 return -EMLINK;
4968
4969         err = btrfs_set_inode_index(dir, &index);
4970         if (err)
4971                 goto fail;
4972
4973         /*
4974          * 2 items for inode and inode ref
4975          * 2 items for dir items
4976          * 1 item for parent inode
4977          */
4978         trans = btrfs_start_transaction(root, 5);
4979         if (IS_ERR(trans)) {
4980                 err = PTR_ERR(trans);
4981                 goto fail;
4982         }
4983
4984         btrfs_inc_nlink(inode);
4985         inode_inc_iversion(inode);
4986         inode->i_ctime = CURRENT_TIME;
4987         ihold(inode);
4988
4989         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4990
4991         if (err) {
4992                 drop_inode = 1;
4993         } else {
4994                 struct dentry *parent = dentry->d_parent;
4995                 err = btrfs_update_inode(trans, root, inode);
4996                 if (err)
4997                         goto fail;
4998                 d_instantiate(dentry, inode);
4999                 btrfs_log_new_name(trans, inode, NULL, parent);
5000         }
5001
5002         nr = trans->blocks_used;
5003         btrfs_end_transaction(trans, root);
5004 fail:
5005         if (drop_inode) {
5006                 inode_dec_link_count(inode);
5007                 iput(inode);
5008         }
5009         btrfs_btree_balance_dirty(root, nr);
5010         return err;
5011 }
5012
5013 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5014 {
5015         struct inode *inode = NULL;
5016         struct btrfs_trans_handle *trans;
5017         struct btrfs_root *root = BTRFS_I(dir)->root;
5018         int err = 0;
5019         int drop_on_err = 0;
5020         u64 objectid = 0;
5021         u64 index = 0;
5022         unsigned long nr = 1;
5023
5024         /*
5025          * 2 items for inode and ref
5026          * 2 items for dir items
5027          * 1 for xattr if selinux is on
5028          */
5029         trans = btrfs_start_transaction(root, 5);
5030         if (IS_ERR(trans))
5031                 return PTR_ERR(trans);
5032
5033         err = btrfs_find_free_ino(root, &objectid);
5034         if (err)
5035                 goto out_fail;
5036
5037         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5038                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5039                                 S_IFDIR | mode, &index);
5040         if (IS_ERR(inode)) {
5041                 err = PTR_ERR(inode);
5042                 goto out_fail;
5043         }
5044
5045         drop_on_err = 1;
5046
5047         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5048         if (err)
5049                 goto out_fail;
5050
5051         inode->i_op = &btrfs_dir_inode_operations;
5052         inode->i_fop = &btrfs_dir_file_operations;
5053
5054         btrfs_i_size_write(inode, 0);
5055         err = btrfs_update_inode(trans, root, inode);
5056         if (err)
5057                 goto out_fail;
5058
5059         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5060                              dentry->d_name.len, 0, index);
5061         if (err)
5062                 goto out_fail;
5063
5064         d_instantiate(dentry, inode);
5065         drop_on_err = 0;
5066
5067 out_fail:
5068         nr = trans->blocks_used;
5069         btrfs_end_transaction(trans, root);
5070         if (drop_on_err)
5071                 iput(inode);
5072         btrfs_btree_balance_dirty(root, nr);
5073         return err;
5074 }
5075
5076 /* helper for btfs_get_extent.  Given an existing extent in the tree,
5077  * and an extent that you want to insert, deal with overlap and insert
5078  * the new extent into the tree.
5079  */
5080 static int merge_extent_mapping(struct extent_map_tree *em_tree,
5081                                 struct extent_map *existing,
5082                                 struct extent_map *em,
5083                                 u64 map_start, u64 map_len)
5084 {
5085         u64 start_diff;
5086
5087         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5088         start_diff = map_start - em->start;
5089         em->start = map_start;
5090         em->len = map_len;
5091         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5092             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
5093                 em->block_start += start_diff;
5094                 em->block_len -= start_diff;
5095         }
5096         return add_extent_mapping(em_tree, em);
5097 }
5098
5099 static noinline int uncompress_inline(struct btrfs_path *path,
5100                                       struct inode *inode, struct page *page,
5101                                       size_t pg_offset, u64 extent_offset,
5102                                       struct btrfs_file_extent_item *item)
5103 {
5104         int ret;
5105         struct extent_buffer *leaf = path->nodes[0];
5106         char *tmp;
5107         size_t max_size;
5108         unsigned long inline_size;
5109         unsigned long ptr;
5110         int compress_type;
5111
5112         WARN_ON(pg_offset != 0);
5113         compress_type = btrfs_file_extent_compression(leaf, item);
5114         max_size = btrfs_file_extent_ram_bytes(leaf, item);
5115         inline_size = btrfs_file_extent_inline_item_len(leaf,
5116                                         btrfs_item_nr(leaf, path->slots[0]));
5117         tmp = kmalloc(inline_size, GFP_NOFS);
5118         if (!tmp)
5119                 return -ENOMEM;
5120         ptr = btrfs_file_extent_inline_start(item);
5121
5122         read_extent_buffer(leaf, tmp, ptr, inline_size);
5123
5124         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5125         ret = btrfs_decompress(compress_type, tmp, page,
5126                                extent_offset, inline_size, max_size);
5127         if (ret) {
5128                 char *kaddr = kmap_atomic(page);
5129                 unsigned long copy_size = min_t(u64,
5130                                   PAGE_CACHE_SIZE - pg_offset,
5131                                   max_size - extent_offset);
5132                 memset(kaddr + pg_offset, 0, copy_size);
5133                 kunmap_atomic(kaddr);
5134         }
5135         kfree(tmp);
5136         return 0;
5137 }
5138
5139 /*
5140  * a bit scary, this does extent mapping from logical file offset to the disk.
5141  * the ugly parts come from merging extents from the disk with the in-ram
5142  * representation.  This gets more complex because of the data=ordered code,
5143  * where the in-ram extents might be locked pending data=ordered completion.
5144  *
5145  * This also copies inline extents directly into the page.
5146  */
5147
5148 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5149                                     size_t pg_offset, u64 start, u64 len,
5150                                     int create)
5151 {
5152         int ret;
5153         int err = 0;
5154         u64 bytenr;
5155         u64 extent_start = 0;
5156         u64 extent_end = 0;
5157         u64 objectid = btrfs_ino(inode);
5158         u32 found_type;
5159         struct btrfs_path *path = NULL;
5160         struct btrfs_root *root = BTRFS_I(inode)->root;
5161         struct btrfs_file_extent_item *item;
5162         struct extent_buffer *leaf;
5163         struct btrfs_key found_key;
5164         struct extent_map *em = NULL;
5165         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5166         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5167         struct btrfs_trans_handle *trans = NULL;
5168         int compress_type;
5169
5170 again:
5171         read_lock(&em_tree->lock);
5172         em = lookup_extent_mapping(em_tree, start, len);
5173         if (em)
5174                 em->bdev = root->fs_info->fs_devices->latest_bdev;
5175         read_unlock(&em_tree->lock);
5176
5177         if (em) {
5178                 if (em->start > start || em->start + em->len <= start)
5179                         free_extent_map(em);
5180                 else if (em->block_start == EXTENT_MAP_INLINE && page)
5181                         free_extent_map(em);
5182                 else
5183                         goto out;
5184         }
5185         em = alloc_extent_map();
5186         if (!em) {
5187                 err = -ENOMEM;
5188                 goto out;
5189         }
5190         em->bdev = root->fs_info->fs_devices->latest_bdev;
5191         em->start = EXTENT_MAP_HOLE;
5192         em->orig_start = EXTENT_MAP_HOLE;
5193         em->len = (u64)-1;
5194         em->block_len = (u64)-1;
5195
5196         if (!path) {
5197                 path = btrfs_alloc_path();
5198                 if (!path) {
5199                         err = -ENOMEM;
5200                         goto out;
5201                 }
5202                 /*
5203                  * Chances are we'll be called again, so go ahead and do
5204                  * readahead
5205                  */
5206                 path->reada = 1;
5207         }
5208
5209         ret = btrfs_lookup_file_extent(trans, root, path,
5210                                        objectid, start, trans != NULL);
5211         if (ret < 0) {
5212                 err = ret;
5213                 goto out;
5214         }
5215
5216         if (ret != 0) {
5217                 if (path->slots[0] == 0)
5218                         goto not_found;
5219                 path->slots[0]--;
5220         }
5221
5222         leaf = path->nodes[0];
5223         item = btrfs_item_ptr(leaf, path->slots[0],
5224                               struct btrfs_file_extent_item);
5225         /* are we inside the extent that was found? */
5226         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5227         found_type = btrfs_key_type(&found_key);
5228         if (found_key.objectid != objectid ||
5229             found_type != BTRFS_EXTENT_DATA_KEY) {
5230                 goto not_found;
5231         }
5232
5233         found_type = btrfs_file_extent_type(leaf, item);
5234         extent_start = found_key.offset;
5235         compress_type = btrfs_file_extent_compression(leaf, item);
5236         if (found_type == BTRFS_FILE_EXTENT_REG ||
5237             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5238                 extent_end = extent_start +
5239                        btrfs_file_extent_num_bytes(leaf, item);
5240         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5241                 size_t size;
5242                 size = btrfs_file_extent_inline_len(leaf, item);
5243                 extent_end = (extent_start + size + root->sectorsize - 1) &
5244                         ~((u64)root->sectorsize - 1);
5245         }
5246
5247         if (start >= extent_end) {
5248                 path->slots[0]++;
5249                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5250                         ret = btrfs_next_leaf(root, path);
5251                         if (ret < 0) {
5252                                 err = ret;
5253                                 goto out;
5254                         }
5255                         if (ret > 0)
5256                                 goto not_found;
5257                         leaf = path->nodes[0];
5258                 }
5259                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5260                 if (found_key.objectid != objectid ||
5261                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5262                         goto not_found;
5263                 if (start + len <= found_key.offset)
5264                         goto not_found;
5265                 em->start = start;
5266                 em->len = found_key.offset - start;
5267                 goto not_found_em;
5268         }
5269
5270         if (found_type == BTRFS_FILE_EXTENT_REG ||
5271             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5272                 em->start = extent_start;
5273                 em->len = extent_end - extent_start;
5274                 em->orig_start = extent_start -
5275                                  btrfs_file_extent_offset(leaf, item);
5276                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5277                 if (bytenr == 0) {
5278                         em->block_start = EXTENT_MAP_HOLE;
5279                         goto insert;
5280                 }
5281                 if (compress_type != BTRFS_COMPRESS_NONE) {
5282                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5283                         em->compress_type = compress_type;
5284                         em->block_start = bytenr;
5285                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5286                                                                          item);
5287                 } else {
5288                         bytenr += btrfs_file_extent_offset(leaf, item);
5289                         em->block_start = bytenr;
5290                         em->block_len = em->len;
5291                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5292                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5293                 }
5294                 goto insert;
5295         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5296                 unsigned long ptr;
5297                 char *map;
5298                 size_t size;
5299                 size_t extent_offset;
5300                 size_t copy_size;
5301
5302                 em->block_start = EXTENT_MAP_INLINE;
5303                 if (!page || create) {
5304                         em->start = extent_start;
5305                         em->len = extent_end - extent_start;
5306                         goto out;
5307                 }
5308
5309                 size = btrfs_file_extent_inline_len(leaf, item);
5310                 extent_offset = page_offset(page) + pg_offset - extent_start;
5311                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5312                                 size - extent_offset);
5313                 em->start = extent_start + extent_offset;
5314                 em->len = (copy_size + root->sectorsize - 1) &
5315                         ~((u64)root->sectorsize - 1);
5316                 em->orig_start = EXTENT_MAP_INLINE;
5317                 if (compress_type) {
5318                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5319                         em->compress_type = compress_type;
5320                 }
5321                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5322                 if (create == 0 && !PageUptodate(page)) {
5323                         if (btrfs_file_extent_compression(leaf, item) !=
5324                             BTRFS_COMPRESS_NONE) {
5325                                 ret = uncompress_inline(path, inode, page,
5326                                                         pg_offset,
5327                                                         extent_offset, item);
5328                                 BUG_ON(ret); /* -ENOMEM */
5329                         } else {
5330                                 map = kmap(page);
5331                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5332                                                    copy_size);
5333                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5334                                         memset(map + pg_offset + copy_size, 0,
5335                                                PAGE_CACHE_SIZE - pg_offset -
5336                                                copy_size);
5337                                 }
5338                                 kunmap(page);
5339                         }
5340                         flush_dcache_page(page);
5341                 } else if (create && PageUptodate(page)) {
5342                         BUG();
5343                         if (!trans) {
5344                                 kunmap(page);
5345                                 free_extent_map(em);
5346                                 em = NULL;
5347
5348                                 btrfs_release_path(path);
5349                                 trans = btrfs_join_transaction(root);
5350
5351                                 if (IS_ERR(trans))
5352                                         return ERR_CAST(trans);
5353                                 goto again;
5354                         }
5355                         map = kmap(page);
5356                         write_extent_buffer(leaf, map + pg_offset, ptr,
5357                                             copy_size);
5358                         kunmap(page);
5359                         btrfs_mark_buffer_dirty(leaf);
5360                 }
5361                 set_extent_uptodate(io_tree, em->start,
5362                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
5363                 goto insert;
5364         } else {
5365                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5366                 WARN_ON(1);
5367         }
5368 not_found:
5369         em->start = start;
5370         em->len = len;
5371 not_found_em:
5372         em->block_start = EXTENT_MAP_HOLE;
5373         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5374 insert:
5375         btrfs_release_path(path);
5376         if (em->start > start || extent_map_end(em) <= start) {
5377                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5378                        "[%llu %llu]\n", (unsigned long long)em->start,
5379                        (unsigned long long)em->len,
5380                        (unsigned long long)start,
5381                        (unsigned long long)len);
5382                 err = -EIO;
5383                 goto out;
5384         }
5385
5386         err = 0;
5387         write_lock(&em_tree->lock);
5388         ret = add_extent_mapping(em_tree, em);
5389         /* it is possible that someone inserted the extent into the tree
5390          * while we had the lock dropped.  It is also possible that
5391          * an overlapping map exists in the tree
5392          */
5393         if (ret == -EEXIST) {
5394                 struct extent_map *existing;
5395
5396                 ret = 0;
5397
5398                 existing = lookup_extent_mapping(em_tree, start, len);
5399                 if (existing && (existing->start > start ||
5400                     existing->start + existing->len <= start)) {
5401                         free_extent_map(existing);
5402                         existing = NULL;
5403                 }
5404                 if (!existing) {
5405                         existing = lookup_extent_mapping(em_tree, em->start,
5406                                                          em->len);
5407                         if (existing) {
5408                                 err = merge_extent_mapping(em_tree, existing,
5409                                                            em, start,
5410                                                            root->sectorsize);
5411                                 free_extent_map(existing);
5412                                 if (err) {
5413                                         free_extent_map(em);
5414                                         em = NULL;
5415                                 }
5416                         } else {
5417                                 err = -EIO;
5418                                 free_extent_map(em);
5419                                 em = NULL;
5420                         }
5421                 } else {
5422                         free_extent_map(em);
5423                         em = existing;
5424                         err = 0;
5425                 }
5426         }
5427         write_unlock(&em_tree->lock);
5428 out:
5429
5430         trace_btrfs_get_extent(root, em);
5431
5432         if (path)
5433                 btrfs_free_path(path);
5434         if (trans) {
5435                 ret = btrfs_end_transaction(trans, root);
5436                 if (!err)
5437                         err = ret;
5438         }
5439         if (err) {
5440                 free_extent_map(em);
5441                 return ERR_PTR(err);
5442         }
5443         BUG_ON(!em); /* Error is always set */
5444         return em;
5445 }
5446
5447 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5448                                            size_t pg_offset, u64 start, u64 len,
5449                                            int create)
5450 {
5451         struct extent_map *em;
5452         struct extent_map *hole_em = NULL;
5453         u64 range_start = start;
5454         u64 end;
5455         u64 found;
5456         u64 found_end;
5457         int err = 0;
5458
5459         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5460         if (IS_ERR(em))
5461                 return em;
5462         if (em) {
5463                 /*
5464                  * if our em maps to a hole, there might
5465                  * actually be delalloc bytes behind it
5466                  */
5467                 if (em->block_start != EXTENT_MAP_HOLE)
5468                         return em;
5469                 else
5470                         hole_em = em;
5471         }
5472
5473         /* check to see if we've wrapped (len == -1 or similar) */
5474         end = start + len;
5475         if (end < start)
5476                 end = (u64)-1;
5477         else
5478                 end -= 1;
5479
5480         em = NULL;
5481
5482         /* ok, we didn't find anything, lets look for delalloc */
5483         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5484                                  end, len, EXTENT_DELALLOC, 1);
5485         found_end = range_start + found;
5486         if (found_end < range_start)
5487                 found_end = (u64)-1;
5488
5489         /*
5490          * we didn't find anything useful, return
5491          * the original results from get_extent()
5492          */
5493         if (range_start > end || found_end <= start) {
5494                 em = hole_em;
5495                 hole_em = NULL;
5496                 goto out;
5497         }
5498
5499         /* adjust the range_start to make sure it doesn't
5500          * go backwards from the start they passed in
5501          */
5502         range_start = max(start,range_start);
5503         found = found_end - range_start;
5504
5505         if (found > 0) {
5506                 u64 hole_start = start;
5507                 u64 hole_len = len;
5508
5509                 em = alloc_extent_map();
5510                 if (!em) {
5511                         err = -ENOMEM;
5512                         goto out;
5513                 }
5514                 /*
5515                  * when btrfs_get_extent can't find anything it
5516                  * returns one huge hole
5517                  *
5518                  * make sure what it found really fits our range, and
5519                  * adjust to make sure it is based on the start from
5520                  * the caller
5521                  */
5522                 if (hole_em) {
5523                         u64 calc_end = extent_map_end(hole_em);
5524
5525                         if (calc_end <= start || (hole_em->start > end)) {
5526                                 free_extent_map(hole_em);
5527                                 hole_em = NULL;
5528                         } else {
5529                                 hole_start = max(hole_em->start, start);
5530                                 hole_len = calc_end - hole_start;
5531                         }
5532                 }
5533                 em->bdev = NULL;
5534                 if (hole_em && range_start > hole_start) {
5535                         /* our hole starts before our delalloc, so we
5536                          * have to return just the parts of the hole
5537                          * that go until  the delalloc starts
5538                          */
5539                         em->len = min(hole_len,
5540                                       range_start - hole_start);
5541                         em->start = hole_start;
5542                         em->orig_start = hole_start;
5543                         /*
5544                          * don't adjust block start at all,
5545                          * it is fixed at EXTENT_MAP_HOLE
5546                          */
5547                         em->block_start = hole_em->block_start;
5548                         em->block_len = hole_len;
5549                 } else {
5550                         em->start = range_start;
5551                         em->len = found;
5552                         em->orig_start = range_start;
5553                         em->block_start = EXTENT_MAP_DELALLOC;
5554                         em->block_len = found;
5555                 }
5556         } else if (hole_em) {
5557                 return hole_em;
5558         }
5559 out:
5560
5561         free_extent_map(hole_em);
5562         if (err) {
5563                 free_extent_map(em);
5564                 return ERR_PTR(err);
5565         }
5566         return em;
5567 }
5568
5569 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5570                                                   struct extent_map *em,
5571                                                   u64 start, u64 len)
5572 {
5573         struct btrfs_root *root = BTRFS_I(inode)->root;
5574         struct btrfs_trans_handle *trans;
5575         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5576         struct btrfs_key ins;
5577         u64 alloc_hint;
5578         int ret;
5579         bool insert = false;
5580
5581         /*
5582          * Ok if the extent map we looked up is a hole and is for the exact
5583          * range we want, there is no reason to allocate a new one, however if
5584          * it is not right then we need to free this one and drop the cache for
5585          * our range.
5586          */
5587         if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5588             em->len != len) {
5589                 free_extent_map(em);
5590                 em = NULL;
5591                 insert = true;
5592                 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5593         }
5594
5595         trans = btrfs_join_transaction(root);
5596         if (IS_ERR(trans))
5597                 return ERR_CAST(trans);
5598
5599         if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5600                 btrfs_add_inode_defrag(trans, inode);
5601
5602         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5603
5604         alloc_hint = get_extent_allocation_hint(inode, start, len);
5605         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5606                                    alloc_hint, &ins, 1);
5607         if (ret) {
5608                 em = ERR_PTR(ret);
5609                 goto out;
5610         }
5611
5612         if (!em) {
5613                 em = alloc_extent_map();
5614                 if (!em) {
5615                         em = ERR_PTR(-ENOMEM);
5616                         goto out;
5617                 }
5618         }
5619
5620         em->start = start;
5621         em->orig_start = em->start;
5622         em->len = ins.offset;
5623
5624         em->block_start = ins.objectid;
5625         em->block_len = ins.offset;
5626         em->bdev = root->fs_info->fs_devices->latest_bdev;
5627
5628         /*
5629          * We need to do this because if we're using the original em we searched
5630          * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5631          */
5632         em->flags = 0;
5633         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5634
5635         while (insert) {
5636                 write_lock(&em_tree->lock);
5637                 ret = add_extent_mapping(em_tree, em);
5638                 write_unlock(&em_tree->lock);
5639                 if (ret != -EEXIST)
5640                         break;
5641                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5642         }
5643
5644         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5645                                            ins.offset, ins.offset, 0);
5646         if (ret) {
5647                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5648                 em = ERR_PTR(ret);
5649         }
5650 out:
5651         btrfs_end_transaction(trans, root);
5652         return em;
5653 }
5654
5655 /*
5656  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5657  * block must be cow'd
5658  */
5659 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5660                                       struct inode *inode, u64 offset, u64 len)
5661 {
5662         struct btrfs_path *path;
5663         int ret;
5664         struct extent_buffer *leaf;
5665         struct btrfs_root *root = BTRFS_I(inode)->root;
5666         struct btrfs_file_extent_item *fi;
5667         struct btrfs_key key;
5668         u64 disk_bytenr;
5669         u64 backref_offset;
5670         u64 extent_end;
5671         u64 num_bytes;
5672         int slot;
5673         int found_type;
5674
5675         path = btrfs_alloc_path();
5676         if (!path)
5677                 return -ENOMEM;
5678
5679         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5680                                        offset, 0);
5681         if (ret < 0)
5682                 goto out;
5683
5684         slot = path->slots[0];
5685         if (ret == 1) {
5686                 if (slot == 0) {
5687                         /* can't find the item, must cow */
5688                         ret = 0;
5689                         goto out;
5690                 }
5691                 slot--;
5692         }
5693         ret = 0;
5694         leaf = path->nodes[0];
5695         btrfs_item_key_to_cpu(leaf, &key, slot);
5696         if (key.objectid != btrfs_ino(inode) ||
5697             key.type != BTRFS_EXTENT_DATA_KEY) {
5698                 /* not our file or wrong item type, must cow */
5699                 goto out;
5700         }
5701
5702         if (key.offset > offset) {
5703                 /* Wrong offset, must cow */
5704                 goto out;
5705         }
5706
5707         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5708         found_type = btrfs_file_extent_type(leaf, fi);
5709         if (found_type != BTRFS_FILE_EXTENT_REG &&
5710             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5711                 /* not a regular extent, must cow */
5712                 goto out;
5713         }
5714         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5715         backref_offset = btrfs_file_extent_offset(leaf, fi);
5716
5717         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5718         if (extent_end < offset + len) {
5719                 /* extent doesn't include our full range, must cow */
5720                 goto out;
5721         }
5722
5723         if (btrfs_extent_readonly(root, disk_bytenr))
5724                 goto out;
5725
5726         /*
5727          * look for other files referencing this extent, if we
5728          * find any we must cow
5729          */
5730         if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5731                                   key.offset - backref_offset, disk_bytenr))
5732                 goto out;
5733
5734         /*
5735          * adjust disk_bytenr and num_bytes to cover just the bytes
5736          * in this extent we are about to write.  If there
5737          * are any csums in that range we have to cow in order
5738          * to keep the csums correct
5739          */
5740         disk_bytenr += backref_offset;
5741         disk_bytenr += offset - key.offset;
5742         num_bytes = min(offset + len, extent_end) - offset;
5743         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5744                                 goto out;
5745         /*
5746          * all of the above have passed, it is safe to overwrite this extent
5747          * without cow
5748          */
5749         ret = 1;
5750 out:
5751         btrfs_free_path(path);
5752         return ret;
5753 }
5754
5755 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5756                                    struct buffer_head *bh_result, int create)
5757 {
5758         struct extent_map *em;
5759         struct btrfs_root *root = BTRFS_I(inode)->root;
5760         u64 start = iblock << inode->i_blkbits;
5761         u64 len = bh_result->b_size;
5762         struct btrfs_trans_handle *trans;
5763
5764         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5765         if (IS_ERR(em))
5766                 return PTR_ERR(em);
5767
5768         /*
5769          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5770          * io.  INLINE is special, and we could probably kludge it in here, but
5771          * it's still buffered so for safety lets just fall back to the generic
5772          * buffered path.
5773          *
5774          * For COMPRESSED we _have_ to read the entire extent in so we can
5775          * decompress it, so there will be buffering required no matter what we
5776          * do, so go ahead and fallback to buffered.
5777          *
5778          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5779          * to buffered IO.  Don't blame me, this is the price we pay for using
5780          * the generic code.
5781          */
5782         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5783             em->block_start == EXTENT_MAP_INLINE) {
5784                 free_extent_map(em);
5785                 return -ENOTBLK;
5786         }
5787
5788         /* Just a good old fashioned hole, return */
5789         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5790                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5791                 free_extent_map(em);
5792                 /* DIO will do one hole at a time, so just unlock a sector */
5793                 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5794                               start + root->sectorsize - 1);
5795                 return 0;
5796         }
5797
5798         /*
5799          * We don't allocate a new extent in the following cases
5800          *
5801          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
5802          * existing extent.
5803          * 2) The extent is marked as PREALLOC.  We're good to go here and can
5804          * just use the extent.
5805          *
5806          */
5807         if (!create) {
5808                 len = em->len - (start - em->start);
5809                 goto map;
5810         }
5811
5812         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5813             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5814              em->block_start != EXTENT_MAP_HOLE)) {
5815                 int type;
5816                 int ret;
5817                 u64 block_start;
5818
5819                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5820                         type = BTRFS_ORDERED_PREALLOC;
5821                 else
5822                         type = BTRFS_ORDERED_NOCOW;
5823                 len = min(len, em->len - (start - em->start));
5824                 block_start = em->block_start + (start - em->start);
5825
5826                 /*
5827                  * we're not going to log anything, but we do need
5828                  * to make sure the current transaction stays open
5829                  * while we look for nocow cross refs
5830                  */
5831                 trans = btrfs_join_transaction(root);
5832                 if (IS_ERR(trans))
5833                         goto must_cow;
5834
5835                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5836                         ret = btrfs_add_ordered_extent_dio(inode, start,
5837                                            block_start, len, len, type);
5838                         btrfs_end_transaction(trans, root);
5839                         if (ret) {
5840                                 free_extent_map(em);
5841                                 return ret;
5842                         }
5843                         goto unlock;
5844                 }
5845                 btrfs_end_transaction(trans, root);
5846         }
5847 must_cow:
5848         /*
5849          * this will cow the extent, reset the len in case we changed
5850          * it above
5851          */
5852         len = bh_result->b_size;
5853         em = btrfs_new_extent_direct(inode, em, start, len);
5854         if (IS_ERR(em))
5855                 return PTR_ERR(em);
5856         len = min(len, em->len - (start - em->start));
5857 unlock:
5858         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5859                           EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5860                           0, NULL, GFP_NOFS);
5861 map:
5862         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5863                 inode->i_blkbits;
5864         bh_result->b_size = len;
5865         bh_result->b_bdev = em->bdev;
5866         set_buffer_mapped(bh_result);
5867         if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5868                 set_buffer_new(bh_result);
5869
5870         free_extent_map(em);
5871
5872         return 0;
5873 }
5874
5875 struct btrfs_dio_private {
5876         struct inode *inode;
5877         u64 logical_offset;
5878         u64 disk_bytenr;
5879         u64 bytes;
5880         u32 *csums;
5881         void *private;
5882
5883         /* number of bios pending for this dio */
5884         atomic_t pending_bios;
5885
5886         /* IO errors */
5887         int errors;
5888
5889         struct bio *orig_bio;
5890 };
5891
5892 static void btrfs_endio_direct_read(struct bio *bio, int err)
5893 {
5894         struct btrfs_dio_private *dip = bio->bi_private;
5895         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5896         struct bio_vec *bvec = bio->bi_io_vec;
5897         struct inode *inode = dip->inode;
5898         struct btrfs_root *root = BTRFS_I(inode)->root;
5899         u64 start;
5900         u32 *private = dip->csums;
5901
5902         start = dip->logical_offset;
5903         do {
5904                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5905                         struct page *page = bvec->bv_page;
5906                         char *kaddr;
5907                         u32 csum = ~(u32)0;
5908                         unsigned long flags;
5909
5910                         local_irq_save(flags);
5911                         kaddr = kmap_atomic(page);
5912                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5913                                                csum, bvec->bv_len);
5914                         btrfs_csum_final(csum, (char *)&csum);
5915                         kunmap_atomic(kaddr);
5916                         local_irq_restore(flags);
5917
5918                         flush_dcache_page(bvec->bv_page);
5919                         if (csum != *private) {
5920                                 printk(KERN_ERR "btrfs csum failed ino %llu off"
5921                                       " %llu csum %u private %u\n",
5922                                       (unsigned long long)btrfs_ino(inode),
5923                                       (unsigned long long)start,
5924                                       csum, *private);
5925                                 err = -EIO;
5926                         }
5927                 }
5928
5929                 start += bvec->bv_len;
5930                 private++;
5931                 bvec++;
5932         } while (bvec <= bvec_end);
5933
5934         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5935                       dip->logical_offset + dip->bytes - 1);
5936         bio->bi_private = dip->private;
5937
5938         kfree(dip->csums);
5939         kfree(dip);
5940
5941         /* If we had a csum failure make sure to clear the uptodate flag */
5942         if (err)
5943                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5944         dio_end_io(bio, err);
5945 }
5946
5947 static void btrfs_endio_direct_write(struct bio *bio, int err)
5948 {
5949         struct btrfs_dio_private *dip = bio->bi_private;
5950         struct inode *inode = dip->inode;
5951         struct btrfs_root *root = BTRFS_I(inode)->root;
5952         struct btrfs_ordered_extent *ordered = NULL;
5953         u64 ordered_offset = dip->logical_offset;
5954         u64 ordered_bytes = dip->bytes;
5955         int ret;
5956
5957         if (err)
5958                 goto out_done;
5959 again:
5960         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5961                                                    &ordered_offset,
5962                                                    ordered_bytes, !err);
5963         if (!ret)
5964                 goto out_test;
5965
5966         ordered->work.func = finish_ordered_fn;
5967         ordered->work.flags = 0;
5968         btrfs_queue_worker(&root->fs_info->endio_write_workers,
5969                            &ordered->work);
5970 out_test:
5971         /*
5972          * our bio might span multiple ordered extents.  If we haven't
5973          * completed the accounting for the whole dio, go back and try again
5974          */
5975         if (ordered_offset < dip->logical_offset + dip->bytes) {
5976                 ordered_bytes = dip->logical_offset + dip->bytes -
5977                         ordered_offset;
5978                 ordered = NULL;
5979                 goto again;
5980         }
5981 out_done:
5982         bio->bi_private = dip->private;
5983
5984         kfree(dip);
5985
5986         /* If we had an error make sure to clear the uptodate flag */
5987         if (err)
5988                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5989         dio_end_io(bio, err);
5990 }
5991
5992 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5993                                     struct bio *bio, int mirror_num,
5994                                     unsigned long bio_flags, u64 offset)
5995 {
5996         int ret;
5997         struct btrfs_root *root = BTRFS_I(inode)->root;
5998         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5999         BUG_ON(ret); /* -ENOMEM */
6000         return 0;
6001 }
6002
6003 static void btrfs_end_dio_bio(struct bio *bio, int err)
6004 {
6005         struct btrfs_dio_private *dip = bio->bi_private;
6006
6007         if (err) {
6008                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
6009                       "sector %#Lx len %u err no %d\n",
6010                       (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
6011                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
6012                 dip->errors = 1;
6013
6014                 /*
6015                  * before atomic variable goto zero, we must make sure
6016                  * dip->errors is perceived to be set.
6017                  */
6018                 smp_mb__before_atomic_dec();
6019         }
6020
6021         /* if there are more bios still pending for this dio, just exit */
6022         if (!atomic_dec_and_test(&dip->pending_bios))
6023                 goto out;
6024
6025         if (dip->errors)
6026                 bio_io_error(dip->orig_bio);
6027         else {
6028                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
6029                 bio_endio(dip->orig_bio, 0);
6030         }
6031 out:
6032         bio_put(bio);
6033 }
6034
6035 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
6036                                        u64 first_sector, gfp_t gfp_flags)
6037 {
6038         int nr_vecs = bio_get_nr_vecs(bdev);
6039         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
6040 }
6041
6042 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
6043                                          int rw, u64 file_offset, int skip_sum,
6044                                          u32 *csums, int async_submit)
6045 {
6046         int write = rw & REQ_WRITE;
6047         struct btrfs_root *root = BTRFS_I(inode)->root;
6048         int ret;
6049
6050         bio_get(bio);
6051
6052         if (!write) {
6053                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
6054                 if (ret)
6055                         goto err;
6056         }
6057
6058         if (skip_sum)
6059                 goto map;
6060
6061         if (write && async_submit) {
6062                 ret = btrfs_wq_submit_bio(root->fs_info,
6063                                    inode, rw, bio, 0, 0,
6064                                    file_offset,
6065                                    __btrfs_submit_bio_start_direct_io,
6066                                    __btrfs_submit_bio_done);
6067                 goto err;
6068         } else if (write) {
6069                 /*
6070                  * If we aren't doing async submit, calculate the csum of the
6071                  * bio now.
6072                  */
6073                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
6074                 if (ret)
6075                         goto err;
6076         } else if (!skip_sum) {
6077                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
6078                                           file_offset, csums);
6079                 if (ret)
6080                         goto err;
6081         }
6082
6083 map:
6084         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
6085 err:
6086         bio_put(bio);
6087         return ret;
6088 }
6089
6090 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
6091                                     int skip_sum)
6092 {
6093         struct inode *inode = dip->inode;
6094         struct btrfs_root *root = BTRFS_I(inode)->root;
6095         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6096         struct bio *bio;
6097         struct bio *orig_bio = dip->orig_bio;
6098         struct bio_vec *bvec = orig_bio->bi_io_vec;
6099         u64 start_sector = orig_bio->bi_sector;
6100         u64 file_offset = dip->logical_offset;
6101         u64 submit_len = 0;
6102         u64 map_length;
6103         int nr_pages = 0;
6104         u32 *csums = dip->csums;
6105         int ret = 0;
6106         int async_submit = 0;
6107         int write = rw & REQ_WRITE;
6108
6109         map_length = orig_bio->bi_size;
6110         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6111                               &map_length, NULL, 0);
6112         if (ret) {
6113                 bio_put(orig_bio);
6114                 return -EIO;
6115         }
6116
6117         if (map_length >= orig_bio->bi_size) {
6118                 bio = orig_bio;
6119                 goto submit;
6120         }
6121
6122         async_submit = 1;
6123         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
6124         if (!bio)
6125                 return -ENOMEM;
6126         bio->bi_private = dip;
6127         bio->bi_end_io = btrfs_end_dio_bio;
6128         atomic_inc(&dip->pending_bios);
6129
6130         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6131                 if (unlikely(map_length < submit_len + bvec->bv_len ||
6132                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6133                                  bvec->bv_offset) < bvec->bv_len)) {
6134                         /*
6135                          * inc the count before we submit the bio so
6136                          * we know the end IO handler won't happen before
6137                          * we inc the count. Otherwise, the dip might get freed
6138                          * before we're done setting it up
6139                          */
6140                         atomic_inc(&dip->pending_bios);
6141                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
6142                                                      file_offset, skip_sum,
6143                                                      csums, async_submit);
6144                         if (ret) {
6145                                 bio_put(bio);
6146                                 atomic_dec(&dip->pending_bios);
6147                                 goto out_err;
6148                         }
6149
6150                         /* Write's use the ordered csums */
6151                         if (!write && !skip_sum)
6152                                 csums = csums + nr_pages;
6153                         start_sector += submit_len >> 9;
6154                         file_offset += submit_len;
6155
6156                         submit_len = 0;
6157                         nr_pages = 0;
6158
6159                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6160                                                   start_sector, GFP_NOFS);
6161                         if (!bio)
6162                                 goto out_err;
6163                         bio->bi_private = dip;
6164                         bio->bi_end_io = btrfs_end_dio_bio;
6165
6166                         map_length = orig_bio->bi_size;
6167                         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6168                                               &map_length, NULL, 0);
6169                         if (ret) {
6170                                 bio_put(bio);
6171                                 goto out_err;
6172                         }
6173                 } else {
6174                         submit_len += bvec->bv_len;
6175                         nr_pages ++;
6176                         bvec++;
6177                 }
6178         }
6179
6180 submit:
6181         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6182                                      csums, async_submit);
6183         if (!ret)
6184                 return 0;
6185
6186         bio_put(bio);
6187 out_err:
6188         dip->errors = 1;
6189         /*
6190          * before atomic variable goto zero, we must
6191          * make sure dip->errors is perceived to be set.
6192          */
6193         smp_mb__before_atomic_dec();
6194         if (atomic_dec_and_test(&dip->pending_bios))
6195                 bio_io_error(dip->orig_bio);
6196
6197         /* bio_end_io() will handle error, so we needn't return it */
6198         return 0;
6199 }
6200
6201 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6202                                 loff_t file_offset)
6203 {
6204         struct btrfs_root *root = BTRFS_I(inode)->root;
6205         struct btrfs_dio_private *dip;
6206         struct bio_vec *bvec = bio->bi_io_vec;
6207         int skip_sum;
6208         int write = rw & REQ_WRITE;
6209         int ret = 0;
6210
6211         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6212
6213         dip = kmalloc(sizeof(*dip), GFP_NOFS);
6214         if (!dip) {
6215                 ret = -ENOMEM;
6216                 goto free_ordered;
6217         }
6218         dip->csums = NULL;
6219
6220         /* Write's use the ordered csum stuff, so we don't need dip->csums */
6221         if (!write && !skip_sum) {
6222                 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6223                 if (!dip->csums) {
6224                         kfree(dip);
6225                         ret = -ENOMEM;
6226                         goto free_ordered;
6227                 }
6228         }
6229
6230         dip->private = bio->bi_private;
6231         dip->inode = inode;
6232         dip->logical_offset = file_offset;
6233
6234         dip->bytes = 0;
6235         do {
6236                 dip->bytes += bvec->bv_len;
6237                 bvec++;
6238         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6239
6240         dip->disk_bytenr = (u64)bio->bi_sector << 9;
6241         bio->bi_private = dip;
6242         dip->errors = 0;
6243         dip->orig_bio = bio;
6244         atomic_set(&dip->pending_bios, 0);
6245
6246         if (write)
6247                 bio->bi_end_io = btrfs_endio_direct_write;
6248         else
6249                 bio->bi_end_io = btrfs_endio_direct_read;
6250
6251         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6252         if (!ret)
6253                 return;
6254 free_ordered:
6255         /*
6256          * If this is a write, we need to clean up the reserved space and kill
6257          * the ordered extent.
6258          */
6259         if (write) {
6260                 struct btrfs_ordered_extent *ordered;
6261                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6262                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6263                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6264                         btrfs_free_reserved_extent(root, ordered->start,
6265                                                    ordered->disk_len);
6266                 btrfs_put_ordered_extent(ordered);
6267                 btrfs_put_ordered_extent(ordered);
6268         }
6269         bio_endio(bio, ret);
6270 }
6271
6272 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6273                         const struct iovec *iov, loff_t offset,
6274                         unsigned long nr_segs)
6275 {
6276         int seg;
6277         int i;
6278         size_t size;
6279         unsigned long addr;
6280         unsigned blocksize_mask = root->sectorsize - 1;
6281         ssize_t retval = -EINVAL;
6282         loff_t end = offset;
6283
6284         if (offset & blocksize_mask)
6285                 goto out;
6286
6287         /* Check the memory alignment.  Blocks cannot straddle pages */
6288         for (seg = 0; seg < nr_segs; seg++) {
6289                 addr = (unsigned long)iov[seg].iov_base;
6290                 size = iov[seg].iov_len;
6291                 end += size;
6292                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6293                         goto out;
6294
6295                 /* If this is a write we don't need to check anymore */
6296                 if (rw & WRITE)
6297                         continue;
6298
6299                 /*
6300                  * Check to make sure we don't have duplicate iov_base's in this
6301                  * iovec, if so return EINVAL, otherwise we'll get csum errors
6302                  * when reading back.
6303                  */
6304                 for (i = seg + 1; i < nr_segs; i++) {
6305                         if (iov[seg].iov_base == iov[i].iov_base)
6306                                 goto out;
6307                 }
6308         }
6309         retval = 0;
6310 out:
6311         return retval;
6312 }
6313 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6314                         const struct iovec *iov, loff_t offset,
6315                         unsigned long nr_segs)
6316 {
6317         struct file *file = iocb->ki_filp;
6318         struct inode *inode = file->f_mapping->host;
6319         struct btrfs_ordered_extent *ordered;
6320         struct extent_state *cached_state = NULL;
6321         u64 lockstart, lockend;
6322         ssize_t ret;
6323         int writing = rw & WRITE;
6324         int write_bits = 0;
6325         size_t count = iov_length(iov, nr_segs);
6326
6327         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6328                             offset, nr_segs)) {
6329                 return 0;
6330         }
6331
6332         lockstart = offset;
6333         lockend = offset + count - 1;
6334
6335         if (writing) {
6336                 ret = btrfs_delalloc_reserve_space(inode, count);
6337                 if (ret)
6338                         goto out;
6339         }
6340
6341         while (1) {
6342                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6343                                  0, &cached_state);
6344                 /*
6345                  * We're concerned with the entire range that we're going to be
6346                  * doing DIO to, so we need to make sure theres no ordered
6347                  * extents in this range.
6348                  */
6349                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6350                                                      lockend - lockstart + 1);
6351                 if (!ordered)
6352                         break;
6353                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6354                                      &cached_state, GFP_NOFS);
6355                 btrfs_start_ordered_extent(inode, ordered, 1);
6356                 btrfs_put_ordered_extent(ordered);
6357                 cond_resched();
6358         }
6359
6360         /*
6361          * we don't use btrfs_set_extent_delalloc because we don't want
6362          * the dirty or uptodate bits
6363          */
6364         if (writing) {
6365                 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6366                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6367                                      EXTENT_DELALLOC, NULL, &cached_state,
6368                                      GFP_NOFS);
6369                 if (ret) {
6370                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6371                                          lockend, EXTENT_LOCKED | write_bits,
6372                                          1, 0, &cached_state, GFP_NOFS);
6373                         goto out;
6374                 }
6375         }
6376
6377         free_extent_state(cached_state);
6378         cached_state = NULL;
6379
6380         ret = __blockdev_direct_IO(rw, iocb, inode,
6381                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6382                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6383                    btrfs_submit_direct, 0);
6384
6385         if (ret < 0 && ret != -EIOCBQUEUED) {
6386                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6387                               offset + iov_length(iov, nr_segs) - 1,
6388                               EXTENT_LOCKED | write_bits, 1, 0,
6389                               &cached_state, GFP_NOFS);
6390         } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6391                 /*
6392                  * We're falling back to buffered, unlock the section we didn't
6393                  * do IO on.
6394                  */
6395                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6396                               offset + iov_length(iov, nr_segs) - 1,
6397                               EXTENT_LOCKED | write_bits, 1, 0,
6398                               &cached_state, GFP_NOFS);
6399         }
6400 out:
6401         free_extent_state(cached_state);
6402         return ret;
6403 }
6404
6405 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6406                 __u64 start, __u64 len)
6407 {
6408         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6409 }
6410
6411 int btrfs_readpage(struct file *file, struct page *page)
6412 {
6413         struct extent_io_tree *tree;
6414         tree = &BTRFS_I(page->mapping->host)->io_tree;
6415         return extent_read_full_page(tree, page, btrfs_get_extent, 0);
6416 }
6417
6418 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6419 {
6420         struct extent_io_tree *tree;
6421
6422
6423         if (current->flags & PF_MEMALLOC) {
6424                 redirty_page_for_writepage(wbc, page);
6425                 unlock_page(page);
6426                 return 0;
6427         }
6428         tree = &BTRFS_I(page->mapping->host)->io_tree;
6429         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6430 }
6431
6432 int btrfs_writepages(struct address_space *mapping,
6433                      struct writeback_control *wbc)
6434 {
6435         struct extent_io_tree *tree;
6436
6437         tree = &BTRFS_I(mapping->host)->io_tree;
6438         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6439 }
6440
6441 static int
6442 btrfs_readpages(struct file *file, struct address_space *mapping,
6443                 struct list_head *pages, unsigned nr_pages)
6444 {
6445         struct extent_io_tree *tree;
6446         tree = &BTRFS_I(mapping->host)->io_tree;
6447         return extent_readpages(tree, mapping, pages, nr_pages,
6448                                 btrfs_get_extent);
6449 }
6450 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6451 {
6452         struct extent_io_tree *tree;
6453         struct extent_map_tree *map;
6454         int ret;
6455
6456         tree = &BTRFS_I(page->mapping->host)->io_tree;
6457         map = &BTRFS_I(page->mapping->host)->extent_tree;
6458         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6459         if (ret == 1) {
6460                 ClearPagePrivate(page);
6461                 set_page_private(page, 0);
6462                 page_cache_release(page);
6463         }
6464         return ret;
6465 }
6466
6467 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6468 {
6469         if (PageWriteback(page) || PageDirty(page))
6470                 return 0;
6471         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6472 }
6473
6474 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6475 {
6476         struct inode *inode = page->mapping->host;
6477         struct extent_io_tree *tree;
6478         struct btrfs_ordered_extent *ordered;
6479         struct extent_state *cached_state = NULL;
6480         u64 page_start = page_offset(page);
6481         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6482
6483         /*
6484          * we have the page locked, so new writeback can't start,
6485          * and the dirty bit won't be cleared while we are here.
6486          *
6487          * Wait for IO on this page so that we can safely clear
6488          * the PagePrivate2 bit and do ordered accounting
6489          */
6490         wait_on_page_writeback(page);
6491
6492         tree = &BTRFS_I(inode)->io_tree;
6493         if (offset) {
6494                 btrfs_releasepage(page, GFP_NOFS);
6495                 return;
6496         }
6497         lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6498         ordered = btrfs_lookup_ordered_extent(inode,
6499                                            page_offset(page));
6500         if (ordered) {
6501                 /*
6502                  * IO on this page will never be started, so we need
6503                  * to account for any ordered extents now
6504                  */
6505                 clear_extent_bit(tree, page_start, page_end,
6506                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6507                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6508                                  &cached_state, GFP_NOFS);
6509                 /*
6510                  * whoever cleared the private bit is responsible
6511                  * for the finish_ordered_io
6512                  */
6513                 if (TestClearPagePrivate2(page) &&
6514                     btrfs_dec_test_ordered_pending(inode, &ordered, page_start,
6515                                                    PAGE_CACHE_SIZE, 1)) {
6516                         btrfs_finish_ordered_io(ordered);
6517                 }
6518                 btrfs_put_ordered_extent(ordered);
6519                 cached_state = NULL;
6520                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6521         }
6522         clear_extent_bit(tree, page_start, page_end,
6523                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6524                  EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6525         __btrfs_releasepage(page, GFP_NOFS);
6526
6527         ClearPageChecked(page);
6528         if (PagePrivate(page)) {
6529                 ClearPagePrivate(page);
6530                 set_page_private(page, 0);
6531                 page_cache_release(page);
6532         }
6533 }
6534
6535 /*
6536  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6537  * called from a page fault handler when a page is first dirtied. Hence we must
6538  * be careful to check for EOF conditions here. We set the page up correctly
6539  * for a written page which means we get ENOSPC checking when writing into
6540  * holes and correct delalloc and unwritten extent mapping on filesystems that
6541  * support these features.
6542  *
6543  * We are not allowed to take the i_mutex here so we have to play games to
6544  * protect against truncate races as the page could now be beyond EOF.  Because
6545  * vmtruncate() writes the inode size before removing pages, once we have the
6546  * page lock we can determine safely if the page is beyond EOF. If it is not
6547  * beyond EOF, then the page is guaranteed safe against truncation until we
6548  * unlock the page.
6549  */
6550 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6551 {
6552         struct page *page = vmf->page;
6553         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6554         struct btrfs_root *root = BTRFS_I(inode)->root;
6555         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6556         struct btrfs_ordered_extent *ordered;
6557         struct extent_state *cached_state = NULL;
6558         char *kaddr;
6559         unsigned long zero_start;
6560         loff_t size;
6561         int ret;
6562         int reserved = 0;
6563         u64 page_start;
6564         u64 page_end;
6565
6566         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6567         if (!ret) {
6568                 ret = btrfs_update_time(vma->vm_file);
6569                 reserved = 1;
6570         }
6571         if (ret) {
6572                 if (ret == -ENOMEM)
6573                         ret = VM_FAULT_OOM;
6574                 else /* -ENOSPC, -EIO, etc */
6575                         ret = VM_FAULT_SIGBUS;
6576                 if (reserved)
6577                         goto out;
6578                 goto out_noreserve;
6579         }
6580
6581         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6582 again:
6583         lock_page(page);
6584         size = i_size_read(inode);
6585         page_start = page_offset(page);
6586         page_end = page_start + PAGE_CACHE_SIZE - 1;
6587
6588         if ((page->mapping != inode->i_mapping) ||
6589             (page_start >= size)) {
6590                 /* page got truncated out from underneath us */
6591                 goto out_unlock;
6592         }
6593         wait_on_page_writeback(page);
6594
6595         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
6596         set_page_extent_mapped(page);
6597
6598         /*
6599          * we can't set the delalloc bits if there are pending ordered
6600          * extents.  Drop our locks and wait for them to finish
6601          */
6602         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6603         if (ordered) {
6604                 unlock_extent_cached(io_tree, page_start, page_end,
6605                                      &cached_state, GFP_NOFS);
6606                 unlock_page(page);
6607                 btrfs_start_ordered_extent(inode, ordered, 1);
6608                 btrfs_put_ordered_extent(ordered);
6609                 goto again;
6610         }
6611
6612         /*
6613          * XXX - page_mkwrite gets called every time the page is dirtied, even
6614          * if it was already dirty, so for space accounting reasons we need to
6615          * clear any delalloc bits for the range we are fixing to save.  There
6616          * is probably a better way to do this, but for now keep consistent with
6617          * prepare_pages in the normal write path.
6618          */
6619         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6620                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6621                           0, 0, &cached_state, GFP_NOFS);
6622
6623         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6624                                         &cached_state);
6625         if (ret) {
6626                 unlock_extent_cached(io_tree, page_start, page_end,
6627                                      &cached_state, GFP_NOFS);
6628                 ret = VM_FAULT_SIGBUS;
6629                 goto out_unlock;
6630         }
6631         ret = 0;
6632
6633         /* page is wholly or partially inside EOF */
6634         if (page_start + PAGE_CACHE_SIZE > size)
6635                 zero_start = size & ~PAGE_CACHE_MASK;
6636         else
6637                 zero_start = PAGE_CACHE_SIZE;
6638
6639         if (zero_start != PAGE_CACHE_SIZE) {
6640                 kaddr = kmap(page);
6641                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6642                 flush_dcache_page(page);
6643                 kunmap(page);
6644         }
6645         ClearPageChecked(page);
6646         set_page_dirty(page);
6647         SetPageUptodate(page);
6648
6649         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6650         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6651
6652         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6653
6654 out_unlock:
6655         if (!ret)
6656                 return VM_FAULT_LOCKED;
6657         unlock_page(page);
6658 out:
6659         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6660 out_noreserve:
6661         return ret;
6662 }
6663
6664 static int btrfs_truncate(struct inode *inode)
6665 {
6666         struct btrfs_root *root = BTRFS_I(inode)->root;
6667         struct btrfs_block_rsv *rsv;
6668         int ret;
6669         int err = 0;
6670         struct btrfs_trans_handle *trans;
6671         unsigned long nr;
6672         u64 mask = root->sectorsize - 1;
6673         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
6674
6675         ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6676         if (ret)
6677                 return ret;
6678
6679         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6680         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6681
6682         /*
6683          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
6684          * 3 things going on here
6685          *
6686          * 1) We need to reserve space for our orphan item and the space to
6687          * delete our orphan item.  Lord knows we don't want to have a dangling
6688          * orphan item because we didn't reserve space to remove it.
6689          *
6690          * 2) We need to reserve space to update our inode.
6691          *
6692          * 3) We need to have something to cache all the space that is going to
6693          * be free'd up by the truncate operation, but also have some slack
6694          * space reserved in case it uses space during the truncate (thank you
6695          * very much snapshotting).
6696          *
6697          * And we need these to all be seperate.  The fact is we can use alot of
6698          * space doing the truncate, and we have no earthly idea how much space
6699          * we will use, so we need the truncate reservation to be seperate so it
6700          * doesn't end up using space reserved for updating the inode or
6701          * removing the orphan item.  We also need to be able to stop the
6702          * transaction and start a new one, which means we need to be able to
6703          * update the inode several times, and we have no idea of knowing how
6704          * many times that will be, so we can't just reserve 1 item for the
6705          * entirety of the opration, so that has to be done seperately as well.
6706          * Then there is the orphan item, which does indeed need to be held on
6707          * to for the whole operation, and we need nobody to touch this reserved
6708          * space except the orphan code.
6709          *
6710          * So that leaves us with
6711          *
6712          * 1) root->orphan_block_rsv - for the orphan deletion.
6713          * 2) rsv - for the truncate reservation, which we will steal from the
6714          * transaction reservation.
6715          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6716          * updating the inode.
6717          */
6718         rsv = btrfs_alloc_block_rsv(root);
6719         if (!rsv)
6720                 return -ENOMEM;
6721         rsv->size = min_size;
6722
6723         /*
6724          * 1 for the truncate slack space
6725          * 1 for the orphan item we're going to add
6726          * 1 for the orphan item deletion
6727          * 1 for updating the inode.
6728          */
6729         trans = btrfs_start_transaction(root, 4);
6730         if (IS_ERR(trans)) {
6731                 err = PTR_ERR(trans);
6732                 goto out;
6733         }
6734
6735         /* Migrate the slack space for the truncate to our reserve */
6736         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6737                                       min_size);
6738         BUG_ON(ret);
6739
6740         ret = btrfs_orphan_add(trans, inode);
6741         if (ret) {
6742                 btrfs_end_transaction(trans, root);
6743                 goto out;
6744         }
6745
6746         /*
6747          * setattr is responsible for setting the ordered_data_close flag,
6748          * but that is only tested during the last file release.  That
6749          * could happen well after the next commit, leaving a great big
6750          * window where new writes may get lost if someone chooses to write
6751          * to this file after truncating to zero
6752          *
6753          * The inode doesn't have any dirty data here, and so if we commit
6754          * this is a noop.  If someone immediately starts writing to the inode
6755          * it is very likely we'll catch some of their writes in this
6756          * transaction, and the commit will find this file on the ordered
6757          * data list with good things to send down.
6758          *
6759          * This is a best effort solution, there is still a window where
6760          * using truncate to replace the contents of the file will
6761          * end up with a zero length file after a crash.
6762          */
6763         if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
6764                                            &BTRFS_I(inode)->runtime_flags))
6765                 btrfs_add_ordered_operation(trans, root, inode);
6766
6767         while (1) {
6768                 ret = btrfs_block_rsv_refill(root, rsv, min_size);
6769                 if (ret) {
6770                         /*
6771                          * This can only happen with the original transaction we
6772                          * started above, every other time we shouldn't have a
6773                          * transaction started yet.
6774                          */
6775                         if (ret == -EAGAIN)
6776                                 goto end_trans;
6777                         err = ret;
6778                         break;
6779                 }
6780
6781                 if (!trans) {
6782                         /* Just need the 1 for updating the inode */
6783                         trans = btrfs_start_transaction(root, 1);
6784                         if (IS_ERR(trans)) {
6785                                 ret = err = PTR_ERR(trans);
6786                                 trans = NULL;
6787                                 break;
6788                         }
6789                 }
6790
6791                 trans->block_rsv = rsv;
6792
6793                 ret = btrfs_truncate_inode_items(trans, root, inode,
6794                                                  inode->i_size,
6795                                                  BTRFS_EXTENT_DATA_KEY);
6796                 if (ret != -EAGAIN) {
6797                         err = ret;
6798                         break;
6799                 }
6800
6801                 trans->block_rsv = &root->fs_info->trans_block_rsv;
6802                 ret = btrfs_update_inode(trans, root, inode);
6803                 if (ret) {
6804                         err = ret;
6805                         break;
6806                 }
6807 end_trans:
6808                 nr = trans->blocks_used;
6809                 btrfs_end_transaction(trans, root);
6810                 trans = NULL;
6811                 btrfs_btree_balance_dirty(root, nr);
6812         }
6813
6814         if (ret == 0 && inode->i_nlink > 0) {
6815                 trans->block_rsv = root->orphan_block_rsv;
6816                 ret = btrfs_orphan_del(trans, inode);
6817                 if (ret)
6818                         err = ret;
6819         } else if (ret && inode->i_nlink > 0) {
6820                 /*
6821                  * Failed to do the truncate, remove us from the in memory
6822                  * orphan list.
6823                  */
6824                 ret = btrfs_orphan_del(NULL, inode);
6825         }
6826
6827         if (trans) {
6828                 trans->block_rsv = &root->fs_info->trans_block_rsv;
6829                 ret = btrfs_update_inode(trans, root, inode);
6830                 if (ret && !err)
6831                         err = ret;
6832
6833                 nr = trans->blocks_used;
6834                 ret = btrfs_end_transaction(trans, root);
6835                 btrfs_btree_balance_dirty(root, nr);
6836         }
6837
6838 out:
6839         btrfs_free_block_rsv(root, rsv);
6840
6841         if (ret && !err)
6842                 err = ret;
6843
6844         return err;
6845 }
6846
6847 /*
6848  * create a new subvolume directory/inode (helper for the ioctl).
6849  */
6850 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6851                              struct btrfs_root *new_root, u64 new_dirid)
6852 {
6853         struct inode *inode;
6854         int err;
6855         u64 index = 0;
6856
6857         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
6858                                 new_dirid, new_dirid,
6859                                 S_IFDIR | (~current_umask() & S_IRWXUGO),
6860                                 &index);
6861         if (IS_ERR(inode))
6862                 return PTR_ERR(inode);
6863         inode->i_op = &btrfs_dir_inode_operations;
6864         inode->i_fop = &btrfs_dir_file_operations;
6865
6866         set_nlink(inode, 1);
6867         btrfs_i_size_write(inode, 0);
6868
6869         err = btrfs_update_inode(trans, new_root, inode);
6870
6871         iput(inode);
6872         return err;
6873 }
6874
6875 struct inode *btrfs_alloc_inode(struct super_block *sb)
6876 {
6877         struct btrfs_inode *ei;
6878         struct inode *inode;
6879
6880         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6881         if (!ei)
6882                 return NULL;
6883
6884         ei->root = NULL;
6885         ei->space_info = NULL;
6886         ei->generation = 0;
6887         ei->last_trans = 0;
6888         ei->last_sub_trans = 0;
6889         ei->logged_trans = 0;
6890         ei->delalloc_bytes = 0;
6891         ei->disk_i_size = 0;
6892         ei->flags = 0;
6893         ei->csum_bytes = 0;
6894         ei->index_cnt = (u64)-1;
6895         ei->last_unlink_trans = 0;
6896
6897         spin_lock_init(&ei->lock);
6898         ei->outstanding_extents = 0;
6899         ei->reserved_extents = 0;
6900
6901         ei->runtime_flags = 0;
6902         ei->force_compress = BTRFS_COMPRESS_NONE;
6903
6904         ei->delayed_node = NULL;
6905
6906         inode = &ei->vfs_inode;
6907         extent_map_tree_init(&ei->extent_tree);
6908         extent_io_tree_init(&ei->io_tree, &inode->i_data);
6909         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
6910         ei->io_tree.track_uptodate = 1;
6911         ei->io_failure_tree.track_uptodate = 1;
6912         mutex_init(&ei->log_mutex);
6913         mutex_init(&ei->delalloc_mutex);
6914         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6915         INIT_LIST_HEAD(&ei->delalloc_inodes);
6916         INIT_LIST_HEAD(&ei->ordered_operations);
6917         RB_CLEAR_NODE(&ei->rb_node);
6918
6919         return inode;
6920 }
6921
6922 static void btrfs_i_callback(struct rcu_head *head)
6923 {
6924         struct inode *inode = container_of(head, struct inode, i_rcu);
6925         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6926 }
6927
6928 void btrfs_destroy_inode(struct inode *inode)
6929 {
6930         struct btrfs_ordered_extent *ordered;
6931         struct btrfs_root *root = BTRFS_I(inode)->root;
6932
6933         WARN_ON(!list_empty(&inode->i_dentry));
6934         WARN_ON(inode->i_data.nrpages);
6935         WARN_ON(BTRFS_I(inode)->outstanding_extents);
6936         WARN_ON(BTRFS_I(inode)->reserved_extents);
6937         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
6938         WARN_ON(BTRFS_I(inode)->csum_bytes);
6939
6940         /*
6941          * This can happen where we create an inode, but somebody else also
6942          * created the same inode and we need to destroy the one we already
6943          * created.
6944          */
6945         if (!root)
6946                 goto free;
6947
6948         /*
6949          * Make sure we're properly removed from the ordered operation
6950          * lists.
6951          */
6952         smp_mb();
6953         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6954                 spin_lock(&root->fs_info->ordered_extent_lock);
6955                 list_del_init(&BTRFS_I(inode)->ordered_operations);
6956                 spin_unlock(&root->fs_info->ordered_extent_lock);
6957         }
6958
6959         if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
6960                      &BTRFS_I(inode)->runtime_flags)) {
6961                 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
6962                        (unsigned long long)btrfs_ino(inode));
6963                 atomic_dec(&root->orphan_inodes);
6964         }
6965
6966         while (1) {
6967                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6968                 if (!ordered)
6969                         break;
6970                 else {
6971                         printk(KERN_ERR "btrfs found ordered "
6972                                "extent %llu %llu on inode cleanup\n",
6973                                (unsigned long long)ordered->file_offset,
6974                                (unsigned long long)ordered->len);
6975                         btrfs_remove_ordered_extent(inode, ordered);
6976                         btrfs_put_ordered_extent(ordered);
6977                         btrfs_put_ordered_extent(ordered);
6978                 }
6979         }
6980         inode_tree_del(inode);
6981         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6982 free:
6983         btrfs_remove_delayed_node(inode);
6984         call_rcu(&inode->i_rcu, btrfs_i_callback);
6985 }
6986
6987 int btrfs_drop_inode(struct inode *inode)
6988 {
6989         struct btrfs_root *root = BTRFS_I(inode)->root;
6990
6991         if (btrfs_root_refs(&root->root_item) == 0 &&
6992             !btrfs_is_free_space_inode(root, inode))
6993                 return 1;
6994         else
6995                 return generic_drop_inode(inode);
6996 }
6997
6998 static void init_once(void *foo)
6999 {
7000         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
7001
7002         inode_init_once(&ei->vfs_inode);
7003 }
7004
7005 void btrfs_destroy_cachep(void)
7006 {
7007         if (btrfs_inode_cachep)
7008                 kmem_cache_destroy(btrfs_inode_cachep);
7009         if (btrfs_trans_handle_cachep)
7010                 kmem_cache_destroy(btrfs_trans_handle_cachep);
7011         if (btrfs_transaction_cachep)
7012                 kmem_cache_destroy(btrfs_transaction_cachep);
7013         if (btrfs_path_cachep)
7014                 kmem_cache_destroy(btrfs_path_cachep);
7015         if (btrfs_free_space_cachep)
7016                 kmem_cache_destroy(btrfs_free_space_cachep);
7017 }
7018
7019 int btrfs_init_cachep(void)
7020 {
7021         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
7022                         sizeof(struct btrfs_inode), 0,
7023                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
7024         if (!btrfs_inode_cachep)
7025                 goto fail;
7026
7027         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
7028                         sizeof(struct btrfs_trans_handle), 0,
7029                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7030         if (!btrfs_trans_handle_cachep)
7031                 goto fail;
7032
7033         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
7034                         sizeof(struct btrfs_transaction), 0,
7035                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7036         if (!btrfs_transaction_cachep)
7037                 goto fail;
7038
7039         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
7040                         sizeof(struct btrfs_path), 0,
7041                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7042         if (!btrfs_path_cachep)
7043                 goto fail;
7044
7045         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
7046                         sizeof(struct btrfs_free_space), 0,
7047                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7048         if (!btrfs_free_space_cachep)
7049                 goto fail;
7050
7051         return 0;
7052 fail:
7053         btrfs_destroy_cachep();
7054         return -ENOMEM;
7055 }
7056
7057 static int btrfs_getattr(struct vfsmount *mnt,
7058                          struct dentry *dentry, struct kstat *stat)
7059 {
7060         struct inode *inode = dentry->d_inode;
7061         u32 blocksize = inode->i_sb->s_blocksize;
7062
7063         generic_fillattr(inode, stat);
7064         stat->dev = BTRFS_I(inode)->root->anon_dev;
7065         stat->blksize = PAGE_CACHE_SIZE;
7066         stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
7067                 ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9;
7068         return 0;
7069 }
7070
7071 /*
7072  * If a file is moved, it will inherit the cow and compression flags of the new
7073  * directory.
7074  */
7075 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
7076 {
7077         struct btrfs_inode *b_dir = BTRFS_I(dir);
7078         struct btrfs_inode *b_inode = BTRFS_I(inode);
7079
7080         if (b_dir->flags & BTRFS_INODE_NODATACOW)
7081                 b_inode->flags |= BTRFS_INODE_NODATACOW;
7082         else
7083                 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
7084
7085         if (b_dir->flags & BTRFS_INODE_COMPRESS)
7086                 b_inode->flags |= BTRFS_INODE_COMPRESS;
7087         else
7088                 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
7089 }
7090
7091 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
7092                            struct inode *new_dir, struct dentry *new_dentry)
7093 {
7094         struct btrfs_trans_handle *trans;
7095         struct btrfs_root *root = BTRFS_I(old_dir)->root;
7096         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
7097         struct inode *new_inode = new_dentry->d_inode;
7098         struct inode *old_inode = old_dentry->d_inode;
7099         struct timespec ctime = CURRENT_TIME;
7100         u64 index = 0;
7101         u64 root_objectid;
7102         int ret;
7103         u64 old_ino = btrfs_ino(old_inode);
7104
7105         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
7106                 return -EPERM;
7107
7108         /* we only allow rename subvolume link between subvolumes */
7109         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
7110                 return -EXDEV;
7111
7112         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
7113             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
7114                 return -ENOTEMPTY;
7115
7116         if (S_ISDIR(old_inode->i_mode) && new_inode &&
7117             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
7118                 return -ENOTEMPTY;
7119         /*
7120          * we're using rename to replace one file with another.
7121          * and the replacement file is large.  Start IO on it now so
7122          * we don't add too much work to the end of the transaction
7123          */
7124         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
7125             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
7126                 filemap_flush(old_inode->i_mapping);
7127
7128         /* close the racy window with snapshot create/destroy ioctl */
7129         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7130                 down_read(&root->fs_info->subvol_sem);
7131         /*
7132          * We want to reserve the absolute worst case amount of items.  So if
7133          * both inodes are subvols and we need to unlink them then that would
7134          * require 4 item modifications, but if they are both normal inodes it
7135          * would require 5 item modifications, so we'll assume their normal
7136          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7137          * should cover the worst case number of items we'll modify.
7138          */
7139         trans = btrfs_start_transaction(root, 20);
7140         if (IS_ERR(trans)) {
7141                 ret = PTR_ERR(trans);
7142                 goto out_notrans;
7143         }
7144
7145         if (dest != root)
7146                 btrfs_record_root_in_trans(trans, dest);
7147
7148         ret = btrfs_set_inode_index(new_dir, &index);
7149         if (ret)
7150                 goto out_fail;
7151
7152         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7153                 /* force full log commit if subvolume involved. */
7154                 root->fs_info->last_trans_log_full_commit = trans->transid;
7155         } else {
7156                 ret = btrfs_insert_inode_ref(trans, dest,
7157                                              new_dentry->d_name.name,
7158                                              new_dentry->d_name.len,
7159                                              old_ino,
7160                                              btrfs_ino(new_dir), index);
7161                 if (ret)
7162                         goto out_fail;
7163                 /*
7164                  * this is an ugly little race, but the rename is required
7165                  * to make sure that if we crash, the inode is either at the
7166                  * old name or the new one.  pinning the log transaction lets
7167                  * us make sure we don't allow a log commit to come in after
7168                  * we unlink the name but before we add the new name back in.
7169                  */
7170                 btrfs_pin_log_trans(root);
7171         }
7172         /*
7173          * make sure the inode gets flushed if it is replacing
7174          * something.
7175          */
7176         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7177                 btrfs_add_ordered_operation(trans, root, old_inode);
7178
7179         inode_inc_iversion(old_dir);
7180         inode_inc_iversion(new_dir);
7181         inode_inc_iversion(old_inode);
7182         old_dir->i_ctime = old_dir->i_mtime = ctime;
7183         new_dir->i_ctime = new_dir->i_mtime = ctime;
7184         old_inode->i_ctime = ctime;
7185
7186         if (old_dentry->d_parent != new_dentry->d_parent)
7187                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7188
7189         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7190                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7191                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7192                                         old_dentry->d_name.name,
7193                                         old_dentry->d_name.len);
7194         } else {
7195                 ret = __btrfs_unlink_inode(trans, root, old_dir,
7196                                         old_dentry->d_inode,
7197                                         old_dentry->d_name.name,
7198                                         old_dentry->d_name.len);
7199                 if (!ret)
7200                         ret = btrfs_update_inode(trans, root, old_inode);
7201         }
7202         if (ret) {
7203                 btrfs_abort_transaction(trans, root, ret);
7204                 goto out_fail;
7205         }
7206
7207         if (new_inode) {
7208                 inode_inc_iversion(new_inode);
7209                 new_inode->i_ctime = CURRENT_TIME;
7210                 if (unlikely(btrfs_ino(new_inode) ==
7211                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7212                         root_objectid = BTRFS_I(new_inode)->location.objectid;
7213                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
7214                                                 root_objectid,
7215                                                 new_dentry->d_name.name,
7216                                                 new_dentry->d_name.len);
7217                         BUG_ON(new_inode->i_nlink == 0);
7218                 } else {
7219                         ret = btrfs_unlink_inode(trans, dest, new_dir,
7220                                                  new_dentry->d_inode,
7221                                                  new_dentry->d_name.name,
7222                                                  new_dentry->d_name.len);
7223                 }
7224                 if (!ret && new_inode->i_nlink == 0) {
7225                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7226                         BUG_ON(ret);
7227                 }
7228                 if (ret) {
7229                         btrfs_abort_transaction(trans, root, ret);
7230                         goto out_fail;
7231                 }
7232         }
7233
7234         fixup_inode_flags(new_dir, old_inode);
7235
7236         ret = btrfs_add_link(trans, new_dir, old_inode,
7237                              new_dentry->d_name.name,
7238                              new_dentry->d_name.len, 0, index);
7239         if (ret) {
7240                 btrfs_abort_transaction(trans, root, ret);
7241                 goto out_fail;
7242         }
7243
7244         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7245                 struct dentry *parent = new_dentry->d_parent;
7246                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7247                 btrfs_end_log_trans(root);
7248         }
7249 out_fail:
7250         btrfs_end_transaction(trans, root);
7251 out_notrans:
7252         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7253                 up_read(&root->fs_info->subvol_sem);
7254
7255         return ret;
7256 }
7257
7258 /*
7259  * some fairly slow code that needs optimization. This walks the list
7260  * of all the inodes with pending delalloc and forces them to disk.
7261  */
7262 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7263 {
7264         struct list_head *head = &root->fs_info->delalloc_inodes;
7265         struct btrfs_inode *binode;
7266         struct inode *inode;
7267
7268         if (root->fs_info->sb->s_flags & MS_RDONLY)
7269                 return -EROFS;
7270
7271         spin_lock(&root->fs_info->delalloc_lock);
7272         while (!list_empty(head)) {
7273                 binode = list_entry(head->next, struct btrfs_inode,
7274                                     delalloc_inodes);
7275                 inode = igrab(&binode->vfs_inode);
7276                 if (!inode)
7277                         list_del_init(&binode->delalloc_inodes);
7278                 spin_unlock(&root->fs_info->delalloc_lock);
7279                 if (inode) {
7280                         filemap_flush(inode->i_mapping);
7281                         if (delay_iput)
7282                                 btrfs_add_delayed_iput(inode);
7283                         else
7284                                 iput(inode);
7285                 }
7286                 cond_resched();
7287                 spin_lock(&root->fs_info->delalloc_lock);
7288         }
7289         spin_unlock(&root->fs_info->delalloc_lock);
7290
7291         /* the filemap_flush will queue IO into the worker threads, but
7292          * we have to make sure the IO is actually started and that
7293          * ordered extents get created before we return
7294          */
7295         atomic_inc(&root->fs_info->async_submit_draining);
7296         while (atomic_read(&root->fs_info->nr_async_submits) ||
7297               atomic_read(&root->fs_info->async_delalloc_pages)) {
7298                 wait_event(root->fs_info->async_submit_wait,
7299                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7300                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7301         }
7302         atomic_dec(&root->fs_info->async_submit_draining);
7303         return 0;
7304 }
7305
7306 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7307                          const char *symname)
7308 {
7309         struct btrfs_trans_handle *trans;
7310         struct btrfs_root *root = BTRFS_I(dir)->root;
7311         struct btrfs_path *path;
7312         struct btrfs_key key;
7313         struct inode *inode = NULL;
7314         int err;
7315         int drop_inode = 0;
7316         u64 objectid;
7317         u64 index = 0 ;
7318         int name_len;
7319         int datasize;
7320         unsigned long ptr;
7321         struct btrfs_file_extent_item *ei;
7322         struct extent_buffer *leaf;
7323         unsigned long nr = 0;
7324
7325         name_len = strlen(symname) + 1;
7326         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7327                 return -ENAMETOOLONG;
7328
7329         /*
7330          * 2 items for inode item and ref
7331          * 2 items for dir items
7332          * 1 item for xattr if selinux is on
7333          */
7334         trans = btrfs_start_transaction(root, 5);
7335         if (IS_ERR(trans))
7336                 return PTR_ERR(trans);
7337
7338         err = btrfs_find_free_ino(root, &objectid);
7339         if (err)
7340                 goto out_unlock;
7341
7342         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7343                                 dentry->d_name.len, btrfs_ino(dir), objectid,
7344                                 S_IFLNK|S_IRWXUGO, &index);
7345         if (IS_ERR(inode)) {
7346                 err = PTR_ERR(inode);
7347                 goto out_unlock;
7348         }
7349
7350         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7351         if (err) {
7352                 drop_inode = 1;
7353                 goto out_unlock;
7354         }
7355
7356         /*
7357         * If the active LSM wants to access the inode during
7358         * d_instantiate it needs these. Smack checks to see
7359         * if the filesystem supports xattrs by looking at the
7360         * ops vector.
7361         */
7362         inode->i_fop = &btrfs_file_operations;
7363         inode->i_op = &btrfs_file_inode_operations;
7364
7365         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7366         if (err)
7367                 drop_inode = 1;
7368         else {
7369                 inode->i_mapping->a_ops = &btrfs_aops;
7370                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7371                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7372         }
7373         if (drop_inode)
7374                 goto out_unlock;
7375
7376         path = btrfs_alloc_path();
7377         if (!path) {
7378                 err = -ENOMEM;
7379                 drop_inode = 1;
7380                 goto out_unlock;
7381         }
7382         key.objectid = btrfs_ino(inode);
7383         key.offset = 0;
7384         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7385         datasize = btrfs_file_extent_calc_inline_size(name_len);
7386         err = btrfs_insert_empty_item(trans, root, path, &key,
7387                                       datasize);
7388         if (err) {
7389                 drop_inode = 1;
7390                 btrfs_free_path(path);
7391                 goto out_unlock;
7392         }
7393         leaf = path->nodes[0];
7394         ei = btrfs_item_ptr(leaf, path->slots[0],
7395                             struct btrfs_file_extent_item);
7396         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7397         btrfs_set_file_extent_type(leaf, ei,
7398                                    BTRFS_FILE_EXTENT_INLINE);
7399         btrfs_set_file_extent_encryption(leaf, ei, 0);
7400         btrfs_set_file_extent_compression(leaf, ei, 0);
7401         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7402         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7403
7404         ptr = btrfs_file_extent_inline_start(ei);
7405         write_extent_buffer(leaf, symname, ptr, name_len);
7406         btrfs_mark_buffer_dirty(leaf);
7407         btrfs_free_path(path);
7408
7409         inode->i_op = &btrfs_symlink_inode_operations;
7410         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7411         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7412         inode_set_bytes(inode, name_len);
7413         btrfs_i_size_write(inode, name_len - 1);
7414         err = btrfs_update_inode(trans, root, inode);
7415         if (err)
7416                 drop_inode = 1;
7417
7418 out_unlock:
7419         if (!err)
7420                 d_instantiate(dentry, inode);
7421         nr = trans->blocks_used;
7422         btrfs_end_transaction(trans, root);
7423         if (drop_inode) {
7424                 inode_dec_link_count(inode);
7425                 iput(inode);
7426         }
7427         btrfs_btree_balance_dirty(root, nr);
7428         return err;
7429 }
7430
7431 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7432                                        u64 start, u64 num_bytes, u64 min_size,
7433                                        loff_t actual_len, u64 *alloc_hint,
7434                                        struct btrfs_trans_handle *trans)
7435 {
7436         struct btrfs_root *root = BTRFS_I(inode)->root;
7437         struct btrfs_key ins;
7438         u64 cur_offset = start;
7439         u64 i_size;
7440         int ret = 0;
7441         bool own_trans = true;
7442
7443         if (trans)
7444                 own_trans = false;
7445         while (num_bytes > 0) {
7446                 if (own_trans) {
7447                         trans = btrfs_start_transaction(root, 3);
7448                         if (IS_ERR(trans)) {
7449                                 ret = PTR_ERR(trans);
7450                                 break;
7451                         }
7452                 }
7453
7454                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7455                                            0, *alloc_hint, &ins, 1);
7456                 if (ret) {
7457                         if (own_trans)
7458                                 btrfs_end_transaction(trans, root);
7459                         break;
7460                 }
7461
7462                 ret = insert_reserved_file_extent(trans, inode,
7463                                                   cur_offset, ins.objectid,
7464                                                   ins.offset, ins.offset,
7465                                                   ins.offset, 0, 0, 0,
7466                                                   BTRFS_FILE_EXTENT_PREALLOC);
7467                 if (ret) {
7468                         btrfs_abort_transaction(trans, root, ret);
7469                         if (own_trans)
7470                                 btrfs_end_transaction(trans, root);
7471                         break;
7472                 }
7473                 btrfs_drop_extent_cache(inode, cur_offset,
7474                                         cur_offset + ins.offset -1, 0);
7475
7476                 num_bytes -= ins.offset;
7477                 cur_offset += ins.offset;
7478                 *alloc_hint = ins.objectid + ins.offset;
7479
7480                 inode_inc_iversion(inode);
7481                 inode->i_ctime = CURRENT_TIME;
7482                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7483                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7484                     (actual_len > inode->i_size) &&
7485                     (cur_offset > inode->i_size)) {
7486                         if (cur_offset > actual_len)
7487                                 i_size = actual_len;
7488                         else
7489                                 i_size = cur_offset;
7490                         i_size_write(inode, i_size);
7491                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7492                 }
7493
7494                 ret = btrfs_update_inode(trans, root, inode);
7495
7496                 if (ret) {
7497                         btrfs_abort_transaction(trans, root, ret);
7498                         if (own_trans)
7499                                 btrfs_end_transaction(trans, root);
7500                         break;
7501                 }
7502
7503                 if (own_trans)
7504                         btrfs_end_transaction(trans, root);
7505         }
7506         return ret;
7507 }
7508
7509 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7510                               u64 start, u64 num_bytes, u64 min_size,
7511                               loff_t actual_len, u64 *alloc_hint)
7512 {
7513         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7514                                            min_size, actual_len, alloc_hint,
7515                                            NULL);
7516 }
7517
7518 int btrfs_prealloc_file_range_trans(struct inode *inode,
7519                                     struct btrfs_trans_handle *trans, int mode,
7520                                     u64 start, u64 num_bytes, u64 min_size,
7521                                     loff_t actual_len, u64 *alloc_hint)
7522 {
7523         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7524                                            min_size, actual_len, alloc_hint, trans);
7525 }
7526
7527 static int btrfs_set_page_dirty(struct page *page)
7528 {
7529         return __set_page_dirty_nobuffers(page);
7530 }
7531
7532 static int btrfs_permission(struct inode *inode, int mask)
7533 {
7534         struct btrfs_root *root = BTRFS_I(inode)->root;
7535         umode_t mode = inode->i_mode;
7536
7537         if (mask & MAY_WRITE &&
7538             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7539                 if (btrfs_root_readonly(root))
7540                         return -EROFS;
7541                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7542                         return -EACCES;
7543         }
7544         return generic_permission(inode, mask);
7545 }
7546
7547 static const struct inode_operations btrfs_dir_inode_operations = {
7548         .getattr        = btrfs_getattr,
7549         .lookup         = btrfs_lookup,
7550         .create         = btrfs_create,
7551         .unlink         = btrfs_unlink,
7552         .link           = btrfs_link,
7553         .mkdir          = btrfs_mkdir,
7554         .rmdir          = btrfs_rmdir,
7555         .rename         = btrfs_rename,
7556         .symlink        = btrfs_symlink,
7557         .setattr        = btrfs_setattr,
7558         .mknod          = btrfs_mknod,
7559         .setxattr       = btrfs_setxattr,
7560         .getxattr       = btrfs_getxattr,
7561         .listxattr      = btrfs_listxattr,
7562         .removexattr    = btrfs_removexattr,
7563         .permission     = btrfs_permission,
7564         .get_acl        = btrfs_get_acl,
7565 };
7566 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7567         .lookup         = btrfs_lookup,
7568         .permission     = btrfs_permission,
7569         .get_acl        = btrfs_get_acl,
7570 };
7571
7572 static const struct file_operations btrfs_dir_file_operations = {
7573         .llseek         = generic_file_llseek,
7574         .read           = generic_read_dir,
7575         .readdir        = btrfs_real_readdir,
7576         .unlocked_ioctl = btrfs_ioctl,
7577 #ifdef CONFIG_COMPAT
7578         .compat_ioctl   = btrfs_ioctl,
7579 #endif
7580         .release        = btrfs_release_file,
7581         .fsync          = btrfs_sync_file,
7582 };
7583
7584 static struct extent_io_ops btrfs_extent_io_ops = {
7585         .fill_delalloc = run_delalloc_range,
7586         .submit_bio_hook = btrfs_submit_bio_hook,
7587         .merge_bio_hook = btrfs_merge_bio_hook,
7588         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7589         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7590         .writepage_start_hook = btrfs_writepage_start_hook,
7591         .set_bit_hook = btrfs_set_bit_hook,
7592         .clear_bit_hook = btrfs_clear_bit_hook,
7593         .merge_extent_hook = btrfs_merge_extent_hook,
7594         .split_extent_hook = btrfs_split_extent_hook,
7595 };
7596
7597 /*
7598  * btrfs doesn't support the bmap operation because swapfiles
7599  * use bmap to make a mapping of extents in the file.  They assume
7600  * these extents won't change over the life of the file and they
7601  * use the bmap result to do IO directly to the drive.
7602  *
7603  * the btrfs bmap call would return logical addresses that aren't
7604  * suitable for IO and they also will change frequently as COW
7605  * operations happen.  So, swapfile + btrfs == corruption.
7606  *
7607  * For now we're avoiding this by dropping bmap.
7608  */
7609 static const struct address_space_operations btrfs_aops = {
7610         .readpage       = btrfs_readpage,
7611         .writepage      = btrfs_writepage,
7612         .writepages     = btrfs_writepages,
7613         .readpages      = btrfs_readpages,
7614         .direct_IO      = btrfs_direct_IO,
7615         .invalidatepage = btrfs_invalidatepage,
7616         .releasepage    = btrfs_releasepage,
7617         .set_page_dirty = btrfs_set_page_dirty,
7618         .error_remove_page = generic_error_remove_page,
7619 };
7620
7621 static const struct address_space_operations btrfs_symlink_aops = {
7622         .readpage       = btrfs_readpage,
7623         .writepage      = btrfs_writepage,
7624         .invalidatepage = btrfs_invalidatepage,
7625         .releasepage    = btrfs_releasepage,
7626 };
7627
7628 static const struct inode_operations btrfs_file_inode_operations = {
7629         .getattr        = btrfs_getattr,
7630         .setattr        = btrfs_setattr,
7631         .setxattr       = btrfs_setxattr,
7632         .getxattr       = btrfs_getxattr,
7633         .listxattr      = btrfs_listxattr,
7634         .removexattr    = btrfs_removexattr,
7635         .permission     = btrfs_permission,
7636         .fiemap         = btrfs_fiemap,
7637         .get_acl        = btrfs_get_acl,
7638 };
7639 static const struct inode_operations btrfs_special_inode_operations = {
7640         .getattr        = btrfs_getattr,
7641         .setattr        = btrfs_setattr,
7642         .permission     = btrfs_permission,
7643         .setxattr       = btrfs_setxattr,
7644         .getxattr       = btrfs_getxattr,
7645         .listxattr      = btrfs_listxattr,
7646         .removexattr    = btrfs_removexattr,
7647         .get_acl        = btrfs_get_acl,
7648 };
7649 static const struct inode_operations btrfs_symlink_inode_operations = {
7650         .readlink       = generic_readlink,
7651         .follow_link    = page_follow_link_light,
7652         .put_link       = page_put_link,
7653         .getattr        = btrfs_getattr,
7654         .setattr        = btrfs_setattr,
7655         .permission     = btrfs_permission,
7656         .setxattr       = btrfs_setxattr,
7657         .getxattr       = btrfs_getxattr,
7658         .listxattr      = btrfs_listxattr,
7659         .removexattr    = btrfs_removexattr,
7660         .get_acl        = btrfs_get_acl,
7661 };
7662
7663 const struct dentry_operations btrfs_dentry_operations = {
7664         .d_delete       = btrfs_dentry_delete,
7665         .d_release      = btrfs_dentry_release,
7666 };