fs/btrfs/inode.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/pagemap.h>
  25 #include <linux/highmem.h>
  26 #include <linux/time.h>
  27 #include <linux/init.h>
  28 #include <linux/string.h>
  29 #include <linux/backing-dev.h>
  30 #include <linux/mpage.h>
  31 #include <linux/swap.h>
  32 #include <linux/writeback.h>
  33 #include <linux/statfs.h>
  34 #include <linux/compat.h>
  35 #include <linux/aio.h>
  36 #include <linux/bit_spinlock.h>
  37 #include <linux/xattr.h>
  38 #include <linux/posix_acl.h>
  39 #include <linux/falloc.h>
  40 #include <linux/slab.h>
  41 #include <linux/ratelimit.h>
  42 #include <linux/mount.h>
  43 #include <linux/btrfs.h>
  44 #include <linux/blkdev.h>
  45 #include "compat.h"
  46 #include "ctree.h"
  47 #include "disk-io.h"
  48 #include "transaction.h"
  49 #include "btrfs_inode.h"
  50 #include "print-tree.h"
  51 #include "ordered-data.h"
  52 #include "xattr.h"
  53 #include "tree-log.h"
  54 #include "volumes.h"
  55 #include "compression.h"
  56 #include "locking.h"
  57 #include "free-space-cache.h"
  58 #include "inode-map.h"
  59 #include "backref.h"
  60
  61 struct btrfs_iget_args {
  62         u64 ino;
  63         struct btrfs_root *root;
  64 };
  65
  66 static const struct inode_operations btrfs_dir_inode_operations;
  67 static const struct inode_operations btrfs_symlink_inode_operations;
  68 static const struct inode_operations btrfs_dir_ro_inode_operations;
  69 static const struct inode_operations btrfs_special_inode_operations;
  70 static const struct inode_operations btrfs_file_inode_operations;
  71 static const struct address_space_operations btrfs_aops;
  72 static const struct address_space_operations btrfs_symlink_aops;
  73 static const struct file_operations btrfs_dir_file_operations;
  74 static struct extent_io_ops btrfs_extent_io_ops;
  75
  76 static struct kmem_cache *btrfs_inode_cachep;
  77 static struct kmem_cache *btrfs_delalloc_work_cachep;
  78 struct kmem_cache *btrfs_trans_handle_cachep;
  79 struct kmem_cache *btrfs_transaction_cachep;
  80 struct kmem_cache *btrfs_path_cachep;
  81 struct kmem_cache *btrfs_free_space_cachep;
  82
  83 #define S_SHIFT 12
  84 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  85         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  86         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  87         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  88         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  89         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  90         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  91         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  92 };
  93
  94 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
  95 static int btrfs_truncate(struct inode *inode);
  96 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
  97 static noinline int cow_file_range(struct inode *inode,
  98                                    struct page *locked_page,
  99                                    u64 start, u64 end, int *page_started,
 100                                    unsigned long *nr_written, int unlock);
 101 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
 102                                            u64 len, u64 orig_start,
 103                                            u64 block_start, u64 block_len,
 104                                            u64 orig_block_len, u64 ram_bytes,
 105                                            int type);
 106
 107 static int btrfs_dirty_inode(struct inode *inode);
 108
 109 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
 110                                      struct inode *inode,  struct inode *dir,
 111                                      const struct qstr *qstr)
 112 {
 113         int err;
 114
 115         err = btrfs_init_acl(trans, inode, dir);
 116         if (!err)
 117                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 118         return err;
 119 }
 120
 121 /*
 122  * this does all the hard work for inserting an inline extent into
 123  * the btree.  The caller should have done a btrfs_drop_extents so that
 124  * no overlapping inline items exist in the btree
 125  */
 126 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 127                                 struct btrfs_root *root, struct inode *inode,
 128                                 u64 start, size_t size, size_t compressed_size,
 129                                 int compress_type,
 130                                 struct page **compressed_pages)
 131 {
 132         struct btrfs_key key;
 133         struct btrfs_path *path;
 134         struct extent_buffer *leaf;
 135         struct page *page = NULL;
 136         char *kaddr;
 137         unsigned long ptr;
 138         struct btrfs_file_extent_item *ei;
 139         int err = 0;
 140         int ret;
 141         size_t cur_size = size;
 142         size_t datasize;
 143         unsigned long offset;
 144
 145         if (compressed_size && compressed_pages)
 146                 cur_size = compressed_size;
 147
 148         path = btrfs_alloc_path();
 149         if (!path)
 150                 return -ENOMEM;
 151
 152         path->leave_spinning = 1;
 153
 154         key.objectid = btrfs_ino(inode);
 155         key.offset = start;
 156         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 157         datasize = btrfs_file_extent_calc_inline_size(cur_size);
 158
 159         inode_add_bytes(inode, size);
 160         ret = btrfs_insert_empty_item(trans, root, path, &key,
 161                                       datasize);
 162         if (ret) {
 163                 err = ret;
 164                 goto fail;
 165         }
 166         leaf = path->nodes[0];
 167         ei = btrfs_item_ptr(leaf, path->slots[0],
 168                             struct btrfs_file_extent_item);
 169         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 170         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 171         btrfs_set_file_extent_encryption(leaf, ei, 0);
 172         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 173         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 174         ptr = btrfs_file_extent_inline_start(ei);
 175
 176         if (compress_type != BTRFS_COMPRESS_NONE) {
 177                 struct page *cpage;
 178                 int i = 0;
 179                 while (compressed_size > 0) {
 180                         cpage = compressed_pages[i];
 181                         cur_size = min_t(unsigned long, compressed_size,
 182                                        PAGE_CACHE_SIZE);
 183
 184                         kaddr = kmap_atomic(cpage);
 185                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 186                         kunmap_atomic(kaddr);
 187
 188                         i++;
 189                         ptr += cur_size;
 190                         compressed_size -= cur_size;
 191                 }
 192                 btrfs_set_file_extent_compression(leaf, ei,
 193                                                   compress_type);
 194         } else {
 195                 page = find_get_page(inode->i_mapping,
 196                                      start >> PAGE_CACHE_SHIFT);
 197                 btrfs_set_file_extent_compression(leaf, ei, 0);
 198                 kaddr = kmap_atomic(page);
 199                 offset = start & (PAGE_CACHE_SIZE - 1);
 200                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 201                 kunmap_atomic(kaddr);
 202                 page_cache_release(page);
 203         }
 204         btrfs_mark_buffer_dirty(leaf);
 205         btrfs_free_path(path);
 206
 207         /*
 208          * we're an inline extent, so nobody can
 209          * extend the file past i_size without locking
 210          * a page we already have locked.
 211          *
 212          * We must do any isize and inode updates
 213          * before we unlock the pages.  Otherwise we
 214          * could end up racing with unlink.
 215          */
 216         BTRFS_I(inode)->disk_i_size = inode->i_size;
 217         ret = btrfs_update_inode(trans, root, inode);
 218
 219         return ret;
 220 fail:
 221         btrfs_free_path(path);
 222         return err;
 223 }
 224
 225
 226 /*
 227  * conditionally insert an inline extent into the file.  This
 228  * does the checks required to make sure the data is small enough
 229  * to fit as an inline extent.
 230  */
 231 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 232                                  struct btrfs_root *root,
 233                                  struct inode *inode, u64 start, u64 end,
 234                                  size_t compressed_size, int compress_type,
 235                                  struct page **compressed_pages)
 236 {
 237         u64 isize = i_size_read(inode);
 238         u64 actual_end = min(end + 1, isize);
 239         u64 inline_len = actual_end - start;
 240         u64 aligned_end = ALIGN(end, root->sectorsize);
 241         u64 data_len = inline_len;
 242         int ret;
 243
 244         if (compressed_size)
 245                 data_len = compressed_size;
 246
 247         if (start > 0 ||
 248             actual_end >= PAGE_CACHE_SIZE ||
 249             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 250             (!compressed_size &&
 251             (actual_end & (root->sectorsize - 1)) == 0) ||
 252             end + 1 < isize ||
 253             data_len > root->fs_info->max_inline) {
 254                 return 1;
 255         }
 256
 257         ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1);
 258         if (ret)
 259                 return ret;
 260
 261         if (isize > actual_end)
 262                 inline_len = min_t(u64, isize, actual_end);
 263         ret = insert_inline_extent(trans, root, inode, start,
 264                                    inline_len, compressed_size,
 265                                    compress_type, compressed_pages);
 266         if (ret && ret != -ENOSPC) {
 267                 btrfs_abort_transaction(trans, root, ret);
 268                 return ret;
 269         } else if (ret == -ENOSPC) {
 270                 return 1;
 271         }
 272
 273         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
 274         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 275         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 276         return 0;
 277 }
 278
 279 struct async_extent {
 280         u64 start;
 281         u64 ram_size;
 282         u64 compressed_size;
 283         struct page **pages;
 284         unsigned long nr_pages;
 285         int compress_type;
 286         struct list_head list;
 287 };
 288
 289 struct async_cow {
 290         struct inode *inode;
 291         struct btrfs_root *root;
 292         struct page *locked_page;
 293         u64 start;
 294         u64 end;
 295         struct list_head extents;
 296         struct btrfs_work work;
 297 };
 298
 299 static noinline int add_async_extent(struct async_cow *cow,
 300                                      u64 start, u64 ram_size,
 301                                      u64 compressed_size,
 302                                      struct page **pages,
 303                                      unsigned long nr_pages,
 304                                      int compress_type)
 305 {
 306         struct async_extent *async_extent;
 307
 308         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 309         BUG_ON(!async_extent); /* -ENOMEM */
 310         async_extent->start = start;
 311         async_extent->ram_size = ram_size;
 312         async_extent->compressed_size = compressed_size;
 313         async_extent->pages = pages;
 314         async_extent->nr_pages = nr_pages;
 315         async_extent->compress_type = compress_type;
 316         list_add_tail(&async_extent->list, &cow->extents);
 317         return 0;
 318 }
 319
 320 /*
 321  * we create compressed extents in two phases.  The first
 322  * phase compresses a range of pages that have already been
 323  * locked (both pages and state bits are locked).
 324  *
 325  * This is done inside an ordered work queue, and the compression
 326  * is spread across many cpus.  The actual IO submission is step
 327  * two, and the ordered work queue takes care of making sure that
 328  * happens in the same order things were put onto the queue by
 329  * writepages and friends.
 330  *
 331  * If this code finds it can't get good compression, it puts an
 332  * entry onto the work queue to write the uncompressed bytes.  This
 333  * makes sure that both compressed inodes and uncompressed inodes
 334  * are written in the same order that the flusher thread sent them
 335  * down.
 336  */
 337 static noinline int compress_file_range(struct inode *inode,
 338                                         struct page *locked_page,
 339                                         u64 start, u64 end,
 340                                         struct async_cow *async_cow,
 341                                         int *num_added)
 342 {
 343         struct btrfs_root *root = BTRFS_I(inode)->root;
 344         struct btrfs_trans_handle *trans;
 345         u64 num_bytes;
 346         u64 blocksize = root->sectorsize;
 347         u64 actual_end;
 348         u64 isize = i_size_read(inode);
 349         int ret = 0;
 350         struct page **pages = NULL;
 351         unsigned long nr_pages;
 352         unsigned long nr_pages_ret = 0;
 353         unsigned long total_compressed = 0;
 354         unsigned long total_in = 0;
 355         unsigned long max_compressed = 128 * 1024;
 356         unsigned long max_uncompressed = 128 * 1024;
 357         int i;
 358         int will_compress;
 359         int compress_type = root->fs_info->compress_type;
 360         int redirty = 0;
 361
 362         /* if this is a small write inside eof, kick off a defrag */
 363         if ((end - start + 1) < 16 * 1024 &&
 364             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 365                 btrfs_add_inode_defrag(NULL, inode);
 366
 367         actual_end = min_t(u64, isize, end + 1);
 368 again:
 369         will_compress = 0;
 370         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 371         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 372
 373         /*
 374          * we don't want to send crud past the end of i_size through
 375          * compression, that's just a waste of CPU time.  So, if the
 376          * end of the file is before the start of our current
 377          * requested range of bytes, we bail out to the uncompressed
 378          * cleanup code that can deal with all of this.
 379          *
 380          * It isn't really the fastest way to fix things, but this is a
 381          * very uncommon corner.
 382          */
 383         if (actual_end <= start)
 384                 goto cleanup_and_bail_uncompressed;
 385
 386         total_compressed = actual_end - start;
 387
 388         /* we want to make sure that amount of ram required to uncompress
 389          * an extent is reasonable, so we limit the total size in ram
 390          * of a compressed extent to 128k.  This is a crucial number
 391          * because it also controls how easily we can spread reads across
 392          * cpus for decompression.
 393          *
 394          * We also want to make sure the amount of IO required to do
 395          * a random read is reasonably small, so we limit the size of
 396          * a compressed extent to 128k.
 397          */
 398         total_compressed = min(total_compressed, max_uncompressed);
 399         num_bytes = ALIGN(end - start + 1, blocksize);
 400         num_bytes = max(blocksize,  num_bytes);
 401         total_in = 0;
 402         ret = 0;
 403
 404         /*
 405          * we do compression for mount -o compress and when the
 406          * inode has not been flagged as nocompress.  This flag can
 407          * change at any time if we discover bad compression ratios.
 408          */
 409         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 410             (btrfs_test_opt(root, COMPRESS) ||
 411              (BTRFS_I(inode)->force_compress) ||
 412              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 413                 WARN_ON(pages);
 414                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 415                 if (!pages) {
 416                         /* just bail out to the uncompressed code */
 417                         goto cont;
 418                 }
 419
 420                 if (BTRFS_I(inode)->force_compress)
 421                         compress_type = BTRFS_I(inode)->force_compress;
 422
 423                 /*
 424                  * we need to call clear_page_dirty_for_io on each
 425                  * page in the range.  Otherwise applications with the file
 426                  * mmap'd can wander in and change the page contents while
 427                  * we are compressing them.
 428                  *
 429                  * If the compression fails for any reason, we set the pages
 430                  * dirty again later on.
 431                  */
 432                 extent_range_clear_dirty_for_io(inode, start, end);
 433                 redirty = 1;
 434                 ret = btrfs_compress_pages(compress_type,
 435                                            inode->i_mapping, start,
 436                                            total_compressed, pages,
 437                                            nr_pages, &nr_pages_ret,
 438                                            &total_in,
 439                                            &total_compressed,
 440                                            max_compressed);
 441
 442                 if (!ret) {
 443                         unsigned long offset = total_compressed &
 444                                 (PAGE_CACHE_SIZE - 1);
 445                         struct page *page = pages[nr_pages_ret - 1];
 446                         char *kaddr;
 447
 448                         /* zero the tail end of the last page, we might be
 449                          * sending it down to disk
 450                          */
 451                         if (offset) {
 452                                 kaddr = kmap_atomic(page);
 453                                 memset(kaddr + offset, 0,
 454                                        PAGE_CACHE_SIZE - offset);
 455                                 kunmap_atomic(kaddr);
 456                         }
 457                         will_compress = 1;
 458                 }
 459         }
 460 cont:
 461         if (start == 0) {
 462                 trans = btrfs_join_transaction(root);
 463                 if (IS_ERR(trans)) {
 464                         ret = PTR_ERR(trans);
 465                         trans = NULL;
 466                         goto cleanup_and_out;
 467                 }
 468                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 469
 470                 /* lets try to make an inline extent */
 471                 if (ret || total_in < (actual_end - start)) {
 472                         /* we didn't compress the entire range, try
 473                          * to make an uncompressed inline extent.
 474                          */
 475                         ret = cow_file_range_inline(trans, root, inode,
 476                                                     start, end, 0, 0, NULL);
 477                 } else {
 478                         /* try making a compressed inline extent */
 479                         ret = cow_file_range_inline(trans, root, inode,
 480                                                     start, end,
 481                                                     total_compressed,
 482                                                     compress_type, pages);
 483                 }
 484                 if (ret <= 0) {
 485                         /*
 486                          * inline extent creation worked or returned error,
 487                          * we don't need to create any more async work items.
 488                          * Unlock and free up our temp pages.
 489                          */
 490                         extent_clear_unlock_delalloc(inode,
 491                              &BTRFS_I(inode)->io_tree,
 492                              start, end, NULL,
 493                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
 494                              EXTENT_CLEAR_DELALLOC |
 495                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
 496
 497                         btrfs_end_transaction(trans, root);
 498                         goto free_pages_out;
 499                 }
 500                 btrfs_end_transaction(trans, root);
 501         }
 502
 503         if (will_compress) {
 504                 /*
 505                  * we aren't doing an inline extent round the compressed size
 506                  * up to a block size boundary so the allocator does sane
 507                  * things
 508                  */
 509                 total_compressed = ALIGN(total_compressed, blocksize);
 510
 511                 /*
 512                  * one last check to make sure the compression is really a
 513                  * win, compare the page count read with the blocks on disk
 514                  */
 515                 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
 516                 if (total_compressed >= total_in) {
 517                         will_compress = 0;
 518                 } else {
 519                         num_bytes = total_in;
 520                 }
 521         }
 522         if (!will_compress && pages) {
 523                 /*
 524                  * the compression code ran but failed to make things smaller,
 525                  * free any pages it allocated and our page pointer array
 526                  */
 527                 for (i = 0; i < nr_pages_ret; i++) {
 528                         WARN_ON(pages[i]->mapping);
 529                         page_cache_release(pages[i]);
 530                 }
 531                 kfree(pages);
 532                 pages = NULL;
 533                 total_compressed = 0;
 534                 nr_pages_ret = 0;
 535
 536                 /* flag the file so we don't compress in the future */
 537                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 538                     !(BTRFS_I(inode)->force_compress)) {
 539                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 540                 }
 541         }
 542         if (will_compress) {
 543                 *num_added += 1;
 544
 545                 /* the async work queues will take care of doing actual
 546                  * allocation on disk for these compressed pages,
 547                  * and will submit them to the elevator.
 548                  */
 549                 add_async_extent(async_cow, start, num_bytes,
 550                                  total_compressed, pages, nr_pages_ret,
 551                                  compress_type);
 552
 553                 if (start + num_bytes < end) {
 554                         start += num_bytes;
 555                         pages = NULL;
 556                         cond_resched();
 557                         goto again;
 558                 }
 559         } else {
 560 cleanup_and_bail_uncompressed:
 561                 /*
 562                  * No compression, but we still need to write the pages in
 563                  * the file we've been given so far.  redirty the locked
 564                  * page if it corresponds to our extent and set things up
 565                  * for the async work queue to run cow_file_range to do
 566                  * the normal delalloc dance
 567                  */
 568                 if (page_offset(locked_page) >= start &&
 569                     page_offset(locked_page) <= end) {
 570                         __set_page_dirty_nobuffers(locked_page);
 571                         /* unlocked later on in the async handlers */
 572                 }
 573                 if (redirty)
 574                         extent_range_redirty_for_io(inode, start, end);
 575                 add_async_extent(async_cow, start, end - start + 1,
 576                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 577                 *num_added += 1;
 578         }
 579
 580 out:
 581         return ret;
 582
 583 free_pages_out:
 584         for (i = 0; i < nr_pages_ret; i++) {
 585                 WARN_ON(pages[i]->mapping);
 586                 page_cache_release(pages[i]);
 587         }
 588         kfree(pages);
 589
 590         goto out;
 591
 592 cleanup_and_out:
 593         extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 594                                      start, end, NULL,
 595                                      EXTENT_CLEAR_UNLOCK_PAGE |
 596                                      EXTENT_CLEAR_DIRTY |
 597                                      EXTENT_CLEAR_DELALLOC |
 598                                      EXTENT_SET_WRITEBACK |
 599                                      EXTENT_END_WRITEBACK);
 600         if (!trans || IS_ERR(trans))
 601                 btrfs_error(root->fs_info, ret, "Failed to join transaction");
 602         else
 603                 btrfs_abort_transaction(trans, root, ret);
 604         goto free_pages_out;
 605 }
 606
 607 /*
 608  * phase two of compressed writeback.  This is the ordered portion
 609  * of the code, which only gets called in the order the work was
 610  * queued.  We walk all the async extents created by compress_file_range
 611  * and send them down to the disk.
 612  */
 613 static noinline int submit_compressed_extents(struct inode *inode,
 614                                               struct async_cow *async_cow)
 615 {
 616         struct async_extent *async_extent;
 617         u64 alloc_hint = 0;
 618         struct btrfs_trans_handle *trans;
 619         struct btrfs_key ins;
 620         struct extent_map *em;
 621         struct btrfs_root *root = BTRFS_I(inode)->root;
 622         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 623         struct extent_io_tree *io_tree;
 624         int ret = 0;
 625
 626         if (list_empty(&async_cow->extents))
 627                 return 0;
 628
 629 again:
 630         while (!list_empty(&async_cow->extents)) {
 631                 async_extent = list_entry(async_cow->extents.next,
 632                                           struct async_extent, list);
 633                 list_del(&async_extent->list);
 634
 635                 io_tree = &BTRFS_I(inode)->io_tree;
 636
 637 retry:
 638                 /* did the compression code fall back to uncompressed IO? */
 639                 if (!async_extent->pages) {
 640                         int page_started = 0;
 641                         unsigned long nr_written = 0;
 642
 643                         lock_extent(io_tree, async_extent->start,
 644                                          async_extent->start +
 645                                          async_extent->ram_size - 1);
 646
 647                         /* allocate blocks */
 648                         ret = cow_file_range(inode, async_cow->locked_page,
 649                                              async_extent->start,
 650                                              async_extent->start +
 651                                              async_extent->ram_size - 1,
 652                                              &page_started, &nr_written, 0);
 653
 654                         /* JDM XXX */
 655
 656                         /*
 657                          * if page_started, cow_file_range inserted an
 658                          * inline extent and took care of all the unlocking
 659                          * and IO for us.  Otherwise, we need to submit
 660                          * all those pages down to the drive.
 661                          */
 662                         if (!page_started && !ret)
 663                                 extent_write_locked_range(io_tree,
 664                                                   inode, async_extent->start,
 665                                                   async_extent->start +
 666                                                   async_extent->ram_size - 1,
 667                                                   btrfs_get_extent,
 668                                                   WB_SYNC_ALL);
 669                         else if (ret)
 670                                 unlock_page(async_cow->locked_page);
 671                         kfree(async_extent);
 672                         cond_resched();
 673                         continue;
 674                 }
 675
 676                 lock_extent(io_tree, async_extent->start,
 677                             async_extent->start + async_extent->ram_size - 1);
 678
 679                 trans = btrfs_join_transaction(root);
 680                 if (IS_ERR(trans)) {
 681                         ret = PTR_ERR(trans);
 682                 } else {
 683                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 684                         ret = btrfs_reserve_extent(trans, root,
 685                                            async_extent->compressed_size,
 686                                            async_extent->compressed_size,
 687                                            0, alloc_hint, &ins, 1);
 688                         if (ret && ret != -ENOSPC)
 689                                 btrfs_abort_transaction(trans, root, ret);
 690                         btrfs_end_transaction(trans, root);
 691                 }
 692
 693                 if (ret) {
 694                         int i;
 695
 696                         for (i = 0; i < async_extent->nr_pages; i++) {
 697                                 WARN_ON(async_extent->pages[i]->mapping);
 698                                 page_cache_release(async_extent->pages[i]);
 699                         }
 700                         kfree(async_extent->pages);
 701                         async_extent->nr_pages = 0;
 702                         async_extent->pages = NULL;
 703
 704                         if (ret == -ENOSPC)
 705                                 goto retry;
 706                         goto out_free;
 707                 }
 708
 709                 /*
 710                  * here we're doing allocation and writeback of the
 711                  * compressed pages
 712                  */
 713                 btrfs_drop_extent_cache(inode, async_extent->start,
 714                                         async_extent->start +
 715                                         async_extent->ram_size - 1, 0);
 716
 717                 em = alloc_extent_map();
 718                 if (!em) {
 719                         ret = -ENOMEM;
 720                         goto out_free_reserve;
 721                 }
 722                 em->start = async_extent->start;
 723                 em->len = async_extent->ram_size;
 724                 em->orig_start = em->start;
 725                 em->mod_start = em->start;
 726                 em->mod_len = em->len;
 727
 728                 em->block_start = ins.objectid;
 729                 em->block_len = ins.offset;
 730                 em->orig_block_len = ins.offset;
 731                 em->ram_bytes = async_extent->ram_size;
 732                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 733                 em->compress_type = async_extent->compress_type;
 734                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 735                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 736                 em->generation = -1;
 737
 738                 while (1) {
 739                         write_lock(&em_tree->lock);
 740                         ret = add_extent_mapping(em_tree, em, 1);
 741                         write_unlock(&em_tree->lock);
 742                         if (ret != -EEXIST) {
 743                                 free_extent_map(em);
 744                                 break;
 745                         }
 746                         btrfs_drop_extent_cache(inode, async_extent->start,
 747                                                 async_extent->start +
 748                                                 async_extent->ram_size - 1, 0);
 749                 }
 750
 751                 if (ret)
 752                         goto out_free_reserve;
 753
 754                 ret = btrfs_add_ordered_extent_compress(inode,
 755                                                 async_extent->start,
 756                                                 ins.objectid,
 757                                                 async_extent->ram_size,
 758                                                 ins.offset,
 759                                                 BTRFS_ORDERED_COMPRESSED,
 760                                                 async_extent->compress_type);
 761                 if (ret)
 762                         goto out_free_reserve;
 763
 764                 /*
 765                  * clear dirty, set writeback and unlock the pages.
 766                  */
 767                 extent_clear_unlock_delalloc(inode,
 768                                 &BTRFS_I(inode)->io_tree,
 769                                 async_extent->start,
 770                                 async_extent->start +
 771                                 async_extent->ram_size - 1,
 772                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 773                                 EXTENT_CLEAR_UNLOCK |
 774                                 EXTENT_CLEAR_DELALLOC |
 775                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
 776
 777                 ret = btrfs_submit_compressed_write(inode,
 778                                     async_extent->start,
 779                                     async_extent->ram_size,
 780                                     ins.objectid,
 781                                     ins.offset, async_extent->pages,
 782                                     async_extent->nr_pages);
 783                 alloc_hint = ins.objectid + ins.offset;
 784                 kfree(async_extent);
 785                 if (ret)
 786                         goto out;
 787                 cond_resched();
 788         }
 789         ret = 0;
 790 out:
 791         return ret;
 792 out_free_reserve:
 793         btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
 794 out_free:
 795         extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 796                                      async_extent->start,
 797                                      async_extent->start +
 798                                      async_extent->ram_size - 1,
 799                                      NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 800                                      EXTENT_CLEAR_UNLOCK |
 801                                      EXTENT_CLEAR_DELALLOC |
 802                                      EXTENT_CLEAR_DIRTY |
 803                                      EXTENT_SET_WRITEBACK |
 804                                      EXTENT_END_WRITEBACK);
 805         kfree(async_extent);
 806         goto again;
 807 }
 808
 809 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 810                                       u64 num_bytes)
 811 {
 812         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 813         struct extent_map *em;
 814         u64 alloc_hint = 0;
 815
 816         read_lock(&em_tree->lock);
 817         em = search_extent_mapping(em_tree, start, num_bytes);
 818         if (em) {
 819                 /*
 820                  * if block start isn't an actual block number then find the
 821                  * first block in this inode and use that as a hint.  If that
 822                  * block is also bogus then just don't worry about it.
 823                  */
 824                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 825                         free_extent_map(em);
 826                         em = search_extent_mapping(em_tree, 0, 0);
 827                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 828                                 alloc_hint = em->block_start;
 829                         if (em)
 830                                 free_extent_map(em);
 831                 } else {
 832                         alloc_hint = em->block_start;
 833                         free_extent_map(em);
 834                 }
 835         }
 836         read_unlock(&em_tree->lock);
 837
 838         return alloc_hint;
 839 }
 840
 841 /*
 842  * when extent_io.c finds a delayed allocation range in the file,
 843  * the call backs end up in this code.  The basic idea is to
 844  * allocate extents on disk for the range, and create ordered data structs
 845  * in ram to track those extents.
 846  *
 847  * locked_page is the page that writepage had locked already.  We use
 848  * it to make sure we don't do extra locks or unlocks.
 849  *
 850  * *page_started is set to one if we unlock locked_page and do everything
 851  * required to start IO on it.  It may be clean and already done with
 852  * IO when we return.
 853  */
 854 static noinline int __cow_file_range(struct btrfs_trans_handle *trans,
 855                                      struct inode *inode,
 856                                      struct btrfs_root *root,
 857                                      struct page *locked_page,
 858                                      u64 start, u64 end, int *page_started,
 859                                      unsigned long *nr_written,
 860                                      int unlock)
 861 {
 862         u64 alloc_hint = 0;
 863         u64 num_bytes;
 864         unsigned long ram_size;
 865         u64 disk_num_bytes;
 866         u64 cur_alloc_size;
 867         u64 blocksize = root->sectorsize;
 868         struct btrfs_key ins;
 869         struct extent_map *em;
 870         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 871         int ret = 0;
 872
 873         BUG_ON(btrfs_is_free_space_inode(inode));
 874
 875         num_bytes = ALIGN(end - start + 1, blocksize);
 876         num_bytes = max(blocksize,  num_bytes);
 877         disk_num_bytes = num_bytes;
 878
 879         /* if this is a small write inside eof, kick off defrag */
 880         if (num_bytes < 64 * 1024 &&
 881             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 882                 btrfs_add_inode_defrag(trans, inode);
 883
 884         if (start == 0) {
 885                 /* lets try to make an inline extent */
 886                 ret = cow_file_range_inline(trans, root, inode,
 887                                             start, end, 0, 0, NULL);
 888                 if (ret == 0) {
 889                         extent_clear_unlock_delalloc(inode,
 890                                      &BTRFS_I(inode)->io_tree,
 891                                      start, end, NULL,
 892                                      EXTENT_CLEAR_UNLOCK_PAGE |
 893                                      EXTENT_CLEAR_UNLOCK |
 894                                      EXTENT_CLEAR_DELALLOC |
 895                                      EXTENT_CLEAR_DIRTY |
 896                                      EXTENT_SET_WRITEBACK |
 897                                      EXTENT_END_WRITEBACK);
 898
 899                         *nr_written = *nr_written +
 900                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 901                         *page_started = 1;
 902                         goto out;
 903                 } else if (ret < 0) {
 904                         btrfs_abort_transaction(trans, root, ret);
 905                         goto out_unlock;
 906                 }
 907         }
 908
 909         BUG_ON(disk_num_bytes >
 910                btrfs_super_total_bytes(root->fs_info->super_copy));
 911
 912         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 913         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 914
 915         while (disk_num_bytes > 0) {
 916                 unsigned long op;
 917
 918                 cur_alloc_size = disk_num_bytes;
 919                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 920                                            root->sectorsize, 0, alloc_hint,
 921                                            &ins, 1);
 922                 if (ret < 0) {
 923                         btrfs_abort_transaction(trans, root, ret);
 924                         goto out_unlock;
 925                 }
 926
 927                 em = alloc_extent_map();
 928                 if (!em) {
 929                         ret = -ENOMEM;
 930                         goto out_reserve;
 931                 }
 932                 em->start = start;
 933                 em->orig_start = em->start;
 934                 ram_size = ins.offset;
 935                 em->len = ins.offset;
 936                 em->mod_start = em->start;
 937                 em->mod_len = em->len;
 938
 939                 em->block_start = ins.objectid;
 940                 em->block_len = ins.offset;
 941                 em->orig_block_len = ins.offset;
 942                 em->ram_bytes = ram_size;
 943                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 944                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 945                 em->generation = -1;
 946
 947                 while (1) {
 948                         write_lock(&em_tree->lock);
 949                         ret = add_extent_mapping(em_tree, em, 1);
 950                         write_unlock(&em_tree->lock);
 951                         if (ret != -EEXIST) {
 952                                 free_extent_map(em);
 953                                 break;
 954                         }
 955                         btrfs_drop_extent_cache(inode, start,
 956                                                 start + ram_size - 1, 0);
 957                 }
 958                 if (ret)
 959                         goto out_reserve;
 960
 961                 cur_alloc_size = ins.offset;
 962                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 963                                                ram_size, cur_alloc_size, 0);
 964                 if (ret)
 965                         goto out_reserve;
 966
 967                 if (root->root_key.objectid ==
 968                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 969                         ret = btrfs_reloc_clone_csums(inode, start,
 970                                                       cur_alloc_size);
 971                         if (ret) {
 972                                 btrfs_abort_transaction(trans, root, ret);
 973                                 goto out_reserve;
 974                         }
 975                 }
 976
 977                 if (disk_num_bytes < cur_alloc_size)
 978                         break;
 979
 980                 /* we're not doing compressed IO, don't unlock the first
 981                  * page (which the caller expects to stay locked), don't
 982                  * clear any dirty bits and don't set any writeback bits
 983                  *
 984                  * Do set the Private2 bit so we know this page was properly
 985                  * setup for writepage
 986                  */
 987                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
 988                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
 989                         EXTENT_SET_PRIVATE2;
 990
 991                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 992                                              start, start + ram_size - 1,
 993                                              locked_page, op);
 994                 disk_num_bytes -= cur_alloc_size;
 995                 num_bytes -= cur_alloc_size;
 996                 alloc_hint = ins.objectid + ins.offset;
 997                 start += cur_alloc_size;
 998         }
 999 out:
1000         return ret;
1001
1002 out_reserve:
1003         btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
1004 out_unlock:
1005         extent_clear_unlock_delalloc(inode,
1006                      &BTRFS_I(inode)->io_tree,
1007                      start, end, locked_page,
1008                      EXTENT_CLEAR_UNLOCK_PAGE |
1009                      EXTENT_CLEAR_UNLOCK |
1010                      EXTENT_CLEAR_DELALLOC |
1011                      EXTENT_CLEAR_DIRTY |
1012                      EXTENT_SET_WRITEBACK |
1013                      EXTENT_END_WRITEBACK);
1014
1015         goto out;
1016 }
1017
1018 static noinline int cow_file_range(struct inode *inode,
1019                                    struct page *locked_page,
1020                                    u64 start, u64 end, int *page_started,
1021                                    unsigned long *nr_written,
1022                                    int unlock)
1023 {
1024         struct btrfs_trans_handle *trans;
1025         struct btrfs_root *root = BTRFS_I(inode)->root;
1026         int ret;
1027
1028         trans = btrfs_join_transaction(root);
1029         if (IS_ERR(trans)) {
1030                 extent_clear_unlock_delalloc(inode,
1031                              &BTRFS_I(inode)->io_tree,
1032                              start, end, locked_page,
1033                              EXTENT_CLEAR_UNLOCK_PAGE |
1034                              EXTENT_CLEAR_UNLOCK |
1035                              EXTENT_CLEAR_DELALLOC |
1036                              EXTENT_CLEAR_DIRTY |
1037                              EXTENT_SET_WRITEBACK |
1038                              EXTENT_END_WRITEBACK);
1039                 return PTR_ERR(trans);
1040         }
1041         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1042
1043         ret = __cow_file_range(trans, inode, root, locked_page, start, end,
1044                                page_started, nr_written, unlock);
1045
1046         btrfs_end_transaction(trans, root);
1047
1048         return ret;
1049 }
1050
1051 /*
1052  * work queue call back to started compression on a file and pages
1053  */
1054 static noinline void async_cow_start(struct btrfs_work *work)
1055 {
1056         struct async_cow *async_cow;
1057         int num_added = 0;
1058         async_cow = container_of(work, struct async_cow, work);
1059
1060         compress_file_range(async_cow->inode, async_cow->locked_page,
1061                             async_cow->start, async_cow->end, async_cow,
1062                             &num_added);
1063         if (num_added == 0) {
1064                 btrfs_add_delayed_iput(async_cow->inode);
1065                 async_cow->inode = NULL;
1066         }
1067 }
1068
1069 /*
1070  * work queue call back to submit previously compressed pages
1071  */
1072 static noinline void async_cow_submit(struct btrfs_work *work)
1073 {
1074         struct async_cow *async_cow;
1075         struct btrfs_root *root;
1076         unsigned long nr_pages;
1077
1078         async_cow = container_of(work, struct async_cow, work);
1079
1080         root = async_cow->root;
1081         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1082                 PAGE_CACHE_SHIFT;
1083
1084         if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1085             5 * 1024 * 1024 &&
1086             waitqueue_active(&root->fs_info->async_submit_wait))
1087                 wake_up(&root->fs_info->async_submit_wait);
1088
1089         if (async_cow->inode)
1090                 submit_compressed_extents(async_cow->inode, async_cow);
1091 }
1092
1093 static noinline void async_cow_free(struct btrfs_work *work)
1094 {
1095         struct async_cow *async_cow;
1096         async_cow = container_of(work, struct async_cow, work);
1097         if (async_cow->inode)
1098                 btrfs_add_delayed_iput(async_cow->inode);
1099         kfree(async_cow);
1100 }
1101
1102 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1103                                 u64 start, u64 end, int *page_started,
1104                                 unsigned long *nr_written)
1105 {
1106         struct async_cow *async_cow;
1107         struct btrfs_root *root = BTRFS_I(inode)->root;
1108         unsigned long nr_pages;
1109         u64 cur_end;
1110         int limit = 10 * 1024 * 1024;
1111
1112         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1113                          1, 0, NULL, GFP_NOFS);
1114         while (start < end) {
1115                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1116                 BUG_ON(!async_cow); /* -ENOMEM */
1117                 async_cow->inode = igrab(inode);
1118                 async_cow->root = root;
1119                 async_cow->locked_page = locked_page;
1120                 async_cow->start = start;
1121
1122                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1123                         cur_end = end;
1124                 else
1125                         cur_end = min(end, start + 512 * 1024 - 1);
1126
1127                 async_cow->end = cur_end;
1128                 INIT_LIST_HEAD(&async_cow->extents);
1129
1130                 async_cow->work.func = async_cow_start;
1131                 async_cow->work.ordered_func = async_cow_submit;
1132                 async_cow->work.ordered_free = async_cow_free;
1133                 async_cow->work.flags = 0;
1134
1135                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1136                         PAGE_CACHE_SHIFT;
1137                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1138
1139                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1140                                    &async_cow->work);
1141
1142                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1143                         wait_event(root->fs_info->async_submit_wait,
1144                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1145                             limit));
1146                 }
1147
1148                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1149                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1150                         wait_event(root->fs_info->async_submit_wait,
1151                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1152                            0));
1153                 }
1154
1155                 *nr_written += nr_pages;
1156                 start = cur_end + 1;
1157         }
1158         *page_started = 1;
1159         return 0;
1160 }
1161
1162 static noinline int csum_exist_in_range(struct btrfs_root *root,
1163                                         u64 bytenr, u64 num_bytes)
1164 {
1165         int ret;
1166         struct btrfs_ordered_sum *sums;
1167         LIST_HEAD(list);
1168
1169         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1170                                        bytenr + num_bytes - 1, &list, 0);
1171         if (ret == 0 && list_empty(&list))
1172                 return 0;
1173
1174         while (!list_empty(&list)) {
1175                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1176                 list_del(&sums->list);
1177                 kfree(sums);
1178         }
1179         return 1;
1180 }
1181
1182 /*
1183  * when nowcow writeback call back.  This checks for snapshots or COW copies
1184  * of the extents that exist in the file, and COWs the file as required.
1185  *
1186  * If no cow copies or snapshots exist, we write directly to the existing
1187  * blocks on disk
1188  */
1189 static noinline int run_delalloc_nocow(struct inode *inode,
1190                                        struct page *locked_page,
1191                               u64 start, u64 end, int *page_started, int force,
1192                               unsigned long *nr_written)
1193 {
1194         struct btrfs_root *root = BTRFS_I(inode)->root;
1195         struct btrfs_trans_handle *trans;
1196         struct extent_buffer *leaf;
1197         struct btrfs_path *path;
1198         struct btrfs_file_extent_item *fi;
1199         struct btrfs_key found_key;
1200         u64 cow_start;
1201         u64 cur_offset;
1202         u64 extent_end;
1203         u64 extent_offset;
1204         u64 disk_bytenr;
1205         u64 num_bytes;
1206         u64 disk_num_bytes;
1207         u64 ram_bytes;
1208         int extent_type;
1209         int ret, err;
1210         int type;
1211         int nocow;
1212         int check_prev = 1;
1213         bool nolock;
1214         u64 ino = btrfs_ino(inode);
1215
1216         path = btrfs_alloc_path();
1217         if (!path) {
1218                 extent_clear_unlock_delalloc(inode,
1219                              &BTRFS_I(inode)->io_tree,
1220                              start, end, locked_page,
1221                              EXTENT_CLEAR_UNLOCK_PAGE |
1222                              EXTENT_CLEAR_UNLOCK |
1223                              EXTENT_CLEAR_DELALLOC |
1224                              EXTENT_CLEAR_DIRTY |
1225                              EXTENT_SET_WRITEBACK |
1226                              EXTENT_END_WRITEBACK);
1227                 return -ENOMEM;
1228         }
1229
1230         nolock = btrfs_is_free_space_inode(inode);
1231
1232         if (nolock)
1233                 trans = btrfs_join_transaction_nolock(root);
1234         else
1235                 trans = btrfs_join_transaction(root);
1236
1237         if (IS_ERR(trans)) {
1238                 extent_clear_unlock_delalloc(inode,
1239                              &BTRFS_I(inode)->io_tree,
1240                              start, end, locked_page,
1241                              EXTENT_CLEAR_UNLOCK_PAGE |
1242                              EXTENT_CLEAR_UNLOCK |
1243                              EXTENT_CLEAR_DELALLOC |
1244                              EXTENT_CLEAR_DIRTY |
1245                              EXTENT_SET_WRITEBACK |
1246                              EXTENT_END_WRITEBACK);
1247                 btrfs_free_path(path);
1248                 return PTR_ERR(trans);
1249         }
1250
1251         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1252
1253         cow_start = (u64)-1;
1254         cur_offset = start;
1255         while (1) {
1256                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1257                                                cur_offset, 0);
1258                 if (ret < 0) {
1259                         btrfs_abort_transaction(trans, root, ret);
1260                         goto error;
1261                 }
1262                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1263                         leaf = path->nodes[0];
1264                         btrfs_item_key_to_cpu(leaf, &found_key,
1265                                               path->slots[0] - 1);
1266                         if (found_key.objectid == ino &&
1267                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1268                                 path->slots[0]--;
1269                 }
1270                 check_prev = 0;
1271 next_slot:
1272                 leaf = path->nodes[0];
1273                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1274                         ret = btrfs_next_leaf(root, path);
1275                         if (ret < 0) {
1276                                 btrfs_abort_transaction(trans, root, ret);
1277                                 goto error;
1278                         }
1279                         if (ret > 0)
1280                                 break;
1281                         leaf = path->nodes[0];
1282                 }
1283
1284                 nocow = 0;
1285                 disk_bytenr = 0;
1286                 num_bytes = 0;
1287                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1288
1289                 if (found_key.objectid > ino ||
1290                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1291                     found_key.offset > end)
1292                         break;
1293
1294                 if (found_key.offset > cur_offset) {
1295                         extent_end = found_key.offset;
1296                         extent_type = 0;
1297                         goto out_check;
1298                 }
1299
1300                 fi = btrfs_item_ptr(leaf, path->slots[0],
1301                                     struct btrfs_file_extent_item);
1302                 extent_type = btrfs_file_extent_type(leaf, fi);
1303
1304                 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1305                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1306                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1307                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1308                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1309                         extent_end = found_key.offset +
1310                                 btrfs_file_extent_num_bytes(leaf, fi);
1311                         disk_num_bytes =
1312                                 btrfs_file_extent_disk_num_bytes(leaf, fi);
1313                         if (extent_end <= start) {
1314                                 path->slots[0]++;
1315                                 goto next_slot;
1316                         }
1317                         if (disk_bytenr == 0)
1318                                 goto out_check;
1319                         if (btrfs_file_extent_compression(leaf, fi) ||
1320                             btrfs_file_extent_encryption(leaf, fi) ||
1321                             btrfs_file_extent_other_encoding(leaf, fi))
1322                                 goto out_check;
1323                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1324                                 goto out_check;
1325                         if (btrfs_extent_readonly(root, disk_bytenr))
1326                                 goto out_check;
1327                         if (btrfs_cross_ref_exist(trans, root, ino,
1328                                                   found_key.offset -
1329                                                   extent_offset, disk_bytenr))
1330                                 goto out_check;
1331                         disk_bytenr += extent_offset;
1332                         disk_bytenr += cur_offset - found_key.offset;
1333                         num_bytes = min(end + 1, extent_end) - cur_offset;
1334                         /*
1335                          * force cow if csum exists in the range.
1336                          * this ensure that csum for a given extent are
1337                          * either valid or do not exist.
1338                          */
1339                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1340                                 goto out_check;
1341                         nocow = 1;
1342                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1343                         extent_end = found_key.offset +
1344                                 btrfs_file_extent_inline_len(leaf, fi);
1345                         extent_end = ALIGN(extent_end, root->sectorsize);
1346                 } else {
1347                         BUG_ON(1);
1348                 }
1349 out_check:
1350                 if (extent_end <= start) {
1351                         path->slots[0]++;
1352                         goto next_slot;
1353                 }
1354                 if (!nocow) {
1355                         if (cow_start == (u64)-1)
1356                                 cow_start = cur_offset;
1357                         cur_offset = extent_end;
1358                         if (cur_offset > end)
1359                                 break;
1360                         path->slots[0]++;
1361                         goto next_slot;
1362                 }
1363
1364                 btrfs_release_path(path);
1365                 if (cow_start != (u64)-1) {
1366                         ret = __cow_file_range(trans, inode, root, locked_page,
1367                                                cow_start, found_key.offset - 1,
1368                                                page_started, nr_written, 1);
1369                         if (ret) {
1370                                 btrfs_abort_transaction(trans, root, ret);
1371                                 goto error;
1372                         }
1373                         cow_start = (u64)-1;
1374                 }
1375
1376                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1377                         struct extent_map *em;
1378                         struct extent_map_tree *em_tree;
1379                         em_tree = &BTRFS_I(inode)->extent_tree;
1380                         em = alloc_extent_map();
1381                         BUG_ON(!em); /* -ENOMEM */
1382                         em->start = cur_offset;
1383                         em->orig_start = found_key.offset - extent_offset;
1384                         em->len = num_bytes;
1385                         em->block_len = num_bytes;
1386                         em->block_start = disk_bytenr;
1387                         em->orig_block_len = disk_num_bytes;
1388                         em->ram_bytes = ram_bytes;
1389                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1390                         em->mod_start = em->start;
1391                         em->mod_len = em->len;
1392                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1393                         set_bit(EXTENT_FLAG_FILLING, &em->flags);
1394                         em->generation = -1;
1395                         while (1) {
1396                                 write_lock(&em_tree->lock);
1397                                 ret = add_extent_mapping(em_tree, em, 1);
1398                                 write_unlock(&em_tree->lock);
1399                                 if (ret != -EEXIST) {
1400                                         free_extent_map(em);
1401                                         break;
1402                                 }
1403                                 btrfs_drop_extent_cache(inode, em->start,
1404                                                 em->start + em->len - 1, 0);
1405                         }
1406                         type = BTRFS_ORDERED_PREALLOC;
1407                 } else {
1408                         type = BTRFS_ORDERED_NOCOW;
1409                 }
1410
1411                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1412                                                num_bytes, num_bytes, type);
1413                 BUG_ON(ret); /* -ENOMEM */
1414
1415                 if (root->root_key.objectid ==
1416                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1417                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1418                                                       num_bytes);
1419                         if (ret) {
1420                                 btrfs_abort_transaction(trans, root, ret);
1421                                 goto error;
1422                         }
1423                 }
1424
1425                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1426                                 cur_offset, cur_offset + num_bytes - 1,
1427                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1428                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1429                                 EXTENT_SET_PRIVATE2);
1430                 cur_offset = extent_end;
1431                 if (cur_offset > end)
1432                         break;
1433         }
1434         btrfs_release_path(path);
1435
1436         if (cur_offset <= end && cow_start == (u64)-1) {
1437                 cow_start = cur_offset;
1438                 cur_offset = end;
1439         }
1440
1441         if (cow_start != (u64)-1) {
1442                 ret = __cow_file_range(trans, inode, root, locked_page,
1443                                        cow_start, end,
1444                                        page_started, nr_written, 1);
1445                 if (ret) {
1446                         btrfs_abort_transaction(trans, root, ret);
1447                         goto error;
1448                 }
1449         }
1450
1451 error:
1452         err = btrfs_end_transaction(trans, root);
1453         if (!ret)
1454                 ret = err;
1455
1456         if (ret && cur_offset < end)
1457                 extent_clear_unlock_delalloc(inode,
1458                              &BTRFS_I(inode)->io_tree,
1459                              cur_offset, end, locked_page,
1460                              EXTENT_CLEAR_UNLOCK_PAGE |
1461                              EXTENT_CLEAR_UNLOCK |
1462                              EXTENT_CLEAR_DELALLOC |
1463                              EXTENT_CLEAR_DIRTY |
1464                              EXTENT_SET_WRITEBACK |
1465                              EXTENT_END_WRITEBACK);
1466
1467         btrfs_free_path(path);
1468         return ret;
1469 }
1470
1471 /*
1472  * extent_io.c call back to do delayed allocation processing
1473  */
1474 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1475                               u64 start, u64 end, int *page_started,
1476                               unsigned long *nr_written)
1477 {
1478         int ret;
1479         struct btrfs_root *root = BTRFS_I(inode)->root;
1480
1481         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1482                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1483                                          page_started, 1, nr_written);
1484         } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1485                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1486                                          page_started, 0, nr_written);
1487         } else if (!btrfs_test_opt(root, COMPRESS) &&
1488                    !(BTRFS_I(inode)->force_compress) &&
1489                    !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1490                 ret = cow_file_range(inode, locked_page, start, end,
1491                                       page_started, nr_written, 1);
1492         } else {
1493                 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1494                         &BTRFS_I(inode)->runtime_flags);
1495                 ret = cow_file_range_async(inode, locked_page, start, end,
1496                                            page_started, nr_written);
1497         }
1498         return ret;
1499 }
1500
1501 static void btrfs_split_extent_hook(struct inode *inode,
1502                                     struct extent_state *orig, u64 split)
1503 {
1504         /* not delalloc, ignore it */
1505         if (!(orig->state & EXTENT_DELALLOC))
1506                 return;
1507
1508         spin_lock(&BTRFS_I(inode)->lock);
1509         BTRFS_I(inode)->outstanding_extents++;
1510         spin_unlock(&BTRFS_I(inode)->lock);
1511 }
1512
1513 /*
1514  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1515  * extents so we can keep track of new extents that are just merged onto old
1516  * extents, such as when we are doing sequential writes, so we can properly
1517  * account for the metadata space we'll need.
1518  */
1519 static void btrfs_merge_extent_hook(struct inode *inode,
1520                                     struct extent_state *new,
1521                                     struct extent_state *other)
1522 {
1523         /* not delalloc, ignore it */
1524         if (!(other->state & EXTENT_DELALLOC))
1525                 return;
1526
1527         spin_lock(&BTRFS_I(inode)->lock);
1528         BTRFS_I(inode)->outstanding_extents--;
1529         spin_unlock(&BTRFS_I(inode)->lock);
1530 }
1531
1532 /*
1533  * extent_io.c set_bit_hook, used to track delayed allocation
1534  * bytes in this file, and to maintain the list of inodes that
1535  * have pending delalloc work to be done.
1536  */
1537 static void btrfs_set_bit_hook(struct inode *inode,
1538                                struct extent_state *state, unsigned long *bits)
1539 {
1540
1541         /*
1542          * set_bit and clear bit hooks normally require _irqsave/restore
1543          * but in this case, we are only testing for the DELALLOC
1544          * bit, which is only set or cleared with irqs on
1545          */
1546         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1547                 struct btrfs_root *root = BTRFS_I(inode)->root;
1548                 u64 len = state->end + 1 - state->start;
1549                 bool do_list = !btrfs_is_free_space_inode(inode);
1550
1551                 if (*bits & EXTENT_FIRST_DELALLOC) {
1552                         *bits &= ~EXTENT_FIRST_DELALLOC;
1553                 } else {
1554                         spin_lock(&BTRFS_I(inode)->lock);
1555                         BTRFS_I(inode)->outstanding_extents++;
1556                         spin_unlock(&BTRFS_I(inode)->lock);
1557                 }
1558
1559                 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1560                                      root->fs_info->delalloc_batch);
1561                 spin_lock(&BTRFS_I(inode)->lock);
1562                 BTRFS_I(inode)->delalloc_bytes += len;
1563                 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1564                                          &BTRFS_I(inode)->runtime_flags)) {
1565                         spin_lock(&root->fs_info->delalloc_lock);
1566                         if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1567                                 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1568                                               &root->fs_info->delalloc_inodes);
1569                                 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1570                                         &BTRFS_I(inode)->runtime_flags);
1571                         }
1572                         spin_unlock(&root->fs_info->delalloc_lock);
1573                 }
1574                 spin_unlock(&BTRFS_I(inode)->lock);
1575         }
1576 }
1577
1578 /*
1579  * extent_io.c clear_bit_hook, see set_bit_hook for why
1580  */
1581 static void btrfs_clear_bit_hook(struct inode *inode,
1582                                  struct extent_state *state,
1583                                  unsigned long *bits)
1584 {
1585         /*
1586          * set_bit and clear bit hooks normally require _irqsave/restore
1587          * but in this case, we are only testing for the DELALLOC
1588          * bit, which is only set or cleared with irqs on
1589          */
1590         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1591                 struct btrfs_root *root = BTRFS_I(inode)->root;
1592                 u64 len = state->end + 1 - state->start;
1593                 bool do_list = !btrfs_is_free_space_inode(inode);
1594
1595                 if (*bits & EXTENT_FIRST_DELALLOC) {
1596                         *bits &= ~EXTENT_FIRST_DELALLOC;
1597                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1598                         spin_lock(&BTRFS_I(inode)->lock);
1599                         BTRFS_I(inode)->outstanding_extents--;
1600                         spin_unlock(&BTRFS_I(inode)->lock);
1601                 }
1602
1603                 if (*bits & EXTENT_DO_ACCOUNTING)
1604                         btrfs_delalloc_release_metadata(inode, len);
1605
1606                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1607                     && do_list)
1608                         btrfs_free_reserved_data_space(inode, len);
1609
1610                 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1611                                      root->fs_info->delalloc_batch);
1612                 spin_lock(&BTRFS_I(inode)->lock);
1613                 BTRFS_I(inode)->delalloc_bytes -= len;
1614                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1615                     test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1616                              &BTRFS_I(inode)->runtime_flags)) {
1617                         spin_lock(&root->fs_info->delalloc_lock);
1618                         if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1619                                 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1620                                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1621                                           &BTRFS_I(inode)->runtime_flags);
1622                         }
1623                         spin_unlock(&root->fs_info->delalloc_lock);
1624                 }
1625                 spin_unlock(&BTRFS_I(inode)->lock);
1626         }
1627 }
1628
1629 /*
1630  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1631  * we don't create bios that span stripes or chunks
1632  */
1633 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1634                          size_t size, struct bio *bio,
1635                          unsigned long bio_flags)
1636 {
1637         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1638         u64 logical = (u64)bio->bi_sector << 9;
1639         u64 length = 0;
1640         u64 map_length;
1641         int ret;
1642
1643         if (bio_flags & EXTENT_BIO_COMPRESSED)
1644                 return 0;
1645
1646         length = bio->bi_size;
1647         map_length = length;
1648         ret = btrfs_map_block(root->fs_info, rw, logical,
1649                               &map_length, NULL, 0);
1650         /* Will always return 0 with map_multi == NULL */
1651         BUG_ON(ret < 0);
1652         if (map_length < length + size)
1653                 return 1;
1654         return 0;
1655 }
1656
1657 /*
1658  * in order to insert checksums into the metadata in large chunks,
1659  * we wait until bio submission time.   All the pages in the bio are
1660  * checksummed and sums are attached onto the ordered extent record.
1661  *
1662  * At IO completion time the cums attached on the ordered extent record
1663  * are inserted into the btree
1664  */
1665 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1666                                     struct bio *bio, int mirror_num,
1667                                     unsigned long bio_flags,
1668                                     u64 bio_offset)
1669 {
1670         struct btrfs_root *root = BTRFS_I(inode)->root;
1671         int ret = 0;
1672
1673         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1674         BUG_ON(ret); /* -ENOMEM */
1675         return 0;
1676 }
1677
1678 /*
1679  * in order to insert checksums into the metadata in large chunks,
1680  * we wait until bio submission time.   All the pages in the bio are
1681  * checksummed and sums are attached onto the ordered extent record.
1682  *
1683  * At IO completion time the cums attached on the ordered extent record
1684  * are inserted into the btree
1685  */
1686 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1687                           int mirror_num, unsigned long bio_flags,
1688                           u64 bio_offset)
1689 {
1690         struct btrfs_root *root = BTRFS_I(inode)->root;
1691         int ret;
1692
1693         ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1694         if (ret)
1695                 bio_endio(bio, ret);
1696         return ret;
1697 }
1698
1699 /*
1700  * extent_io.c submission hook. This does the right thing for csum calculation
1701  * on write, or reading the csums from the tree before a read
1702  */
1703 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1704                           int mirror_num, unsigned long bio_flags,
1705                           u64 bio_offset)
1706 {
1707         struct btrfs_root *root = BTRFS_I(inode)->root;
1708         int ret = 0;
1709         int skip_sum;
1710         int metadata = 0;
1711         int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1712
1713         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1714
1715         if (btrfs_is_free_space_inode(inode))
1716                 metadata = 2;
1717
1718         if (!(rw & REQ_WRITE)) {
1719                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1720                 if (ret)
1721                         goto out;
1722
1723                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1724                         ret = btrfs_submit_compressed_read(inode, bio,
1725                                                            mirror_num,
1726                                                            bio_flags);
1727                         goto out;
1728                 } else if (!skip_sum) {
1729                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1730                         if (ret)
1731                                 goto out;
1732                 }
1733                 goto mapit;
1734         } else if (async && !skip_sum) {
1735                 /* csum items have already been cloned */
1736                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1737                         goto mapit;
1738                 /* we're doing a write, do the async checksumming */
1739                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1740                                    inode, rw, bio, mirror_num,
1741                                    bio_flags, bio_offset,
1742                                    __btrfs_submit_bio_start,
1743                                    __btrfs_submit_bio_done);
1744                 goto out;
1745         } else if (!skip_sum) {
1746                 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1747                 if (ret)
1748                         goto out;
1749         }
1750
1751 mapit:
1752         ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1753
1754 out:
1755         if (ret < 0)
1756                 bio_endio(bio, ret);
1757         return ret;
1758 }
1759
1760 /*
1761  * given a list of ordered sums record them in the inode.  This happens
1762  * at IO completion time based on sums calculated at bio submission time.
1763  */
1764 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1765                              struct inode *inode, u64 file_offset,
1766                              struct list_head *list)
1767 {
1768         struct btrfs_ordered_sum *sum;
1769
1770         list_for_each_entry(sum, list, list) {
1771                 trans->adding_csums = 1;
1772                 btrfs_csum_file_blocks(trans,
1773                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1774                 trans->adding_csums = 0;
1775         }
1776         return 0;
1777 }
1778
1779 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1780                               struct extent_state **cached_state)
1781 {
1782         WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1783         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1784                                    cached_state, GFP_NOFS);
1785 }
1786
1787 /* see btrfs_writepage_start_hook for details on why this is required */
1788 struct btrfs_writepage_fixup {
1789         struct page *page;
1790         struct btrfs_work work;
1791 };
1792
1793 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1794 {
1795         struct btrfs_writepage_fixup *fixup;
1796         struct btrfs_ordered_extent *ordered;
1797         struct extent_state *cached_state = NULL;
1798         struct page *page;
1799         struct inode *inode;
1800         u64 page_start;
1801         u64 page_end;
1802         int ret;
1803
1804         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1805         page = fixup->page;
1806 again:
1807         lock_page(page);
1808         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1809                 ClearPageChecked(page);
1810                 goto out_page;
1811         }
1812
1813         inode = page->mapping->host;
1814         page_start = page_offset(page);
1815         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1816
1817         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1818                          &cached_state);
1819
1820         /* already ordered? We're done */
1821         if (PagePrivate2(page))
1822                 goto out;
1823
1824         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1825         if (ordered) {
1826                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1827                                      page_end, &cached_state, GFP_NOFS);
1828                 unlock_page(page);
1829                 btrfs_start_ordered_extent(inode, ordered, 1);
1830                 btrfs_put_ordered_extent(ordered);
1831                 goto again;
1832         }
1833
1834         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1835         if (ret) {
1836                 mapping_set_error(page->mapping, ret);
1837                 end_extent_writepage(page, ret, page_start, page_end);
1838                 ClearPageChecked(page);
1839                 goto out;
1840          }
1841
1842         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1843         ClearPageChecked(page);
1844         set_page_dirty(page);
1845 out:
1846         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1847                              &cached_state, GFP_NOFS);
1848 out_page:
1849         unlock_page(page);
1850         page_cache_release(page);
1851         kfree(fixup);
1852 }
1853
1854 /*
1855  * There are a few paths in the higher layers of the kernel that directly
1856  * set the page dirty bit without asking the filesystem if it is a
1857  * good idea.  This causes problems because we want to make sure COW
1858  * properly happens and the data=ordered rules are followed.
1859  *
1860  * In our case any range that doesn't have the ORDERED bit set
1861  * hasn't been properly setup for IO.  We kick off an async process
1862  * to fix it up.  The async helper will wait for ordered extents, set
1863  * the delalloc bit and make it safe to write the page.
1864  */
1865 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1866 {
1867         struct inode *inode = page->mapping->host;
1868         struct btrfs_writepage_fixup *fixup;
1869         struct btrfs_root *root = BTRFS_I(inode)->root;
1870
1871         /* this page is properly in the ordered list */
1872         if (TestClearPagePrivate2(page))
1873                 return 0;
1874
1875         if (PageChecked(page))
1876                 return -EAGAIN;
1877
1878         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1879         if (!fixup)
1880                 return -EAGAIN;
1881
1882         SetPageChecked(page);
1883         page_cache_get(page);
1884         fixup->work.func = btrfs_writepage_fixup_worker;
1885         fixup->page = page;
1886         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1887         return -EBUSY;
1888 }
1889
1890 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1891                                        struct inode *inode, u64 file_pos,
1892                                        u64 disk_bytenr, u64 disk_num_bytes,
1893                                        u64 num_bytes, u64 ram_bytes,
1894                                        u8 compression, u8 encryption,
1895                                        u16 other_encoding, int extent_type)
1896 {
1897         struct btrfs_root *root = BTRFS_I(inode)->root;
1898         struct btrfs_file_extent_item *fi;
1899         struct btrfs_path *path;
1900         struct extent_buffer *leaf;
1901         struct btrfs_key ins;
1902         int ret;
1903
1904         path = btrfs_alloc_path();
1905         if (!path)
1906                 return -ENOMEM;
1907
1908         path->leave_spinning = 1;
1909
1910         /*
1911          * we may be replacing one extent in the tree with another.
1912          * The new extent is pinned in the extent map, and we don't want
1913          * to drop it from the cache until it is completely in the btree.
1914          *
1915          * So, tell btrfs_drop_extents to leave this extent in the cache.
1916          * the caller is expected to unpin it and allow it to be merged
1917          * with the others.
1918          */
1919         ret = btrfs_drop_extents(trans, root, inode, file_pos,
1920                                  file_pos + num_bytes, 0);
1921         if (ret)
1922                 goto out;
1923
1924         ins.objectid = btrfs_ino(inode);
1925         ins.offset = file_pos;
1926         ins.type = BTRFS_EXTENT_DATA_KEY;
1927         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1928         if (ret)
1929                 goto out;
1930         leaf = path->nodes[0];
1931         fi = btrfs_item_ptr(leaf, path->slots[0],
1932                             struct btrfs_file_extent_item);
1933         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1934         btrfs_set_file_extent_type(leaf, fi, extent_type);
1935         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1936         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1937         btrfs_set_file_extent_offset(leaf, fi, 0);
1938         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1939         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1940         btrfs_set_file_extent_compression(leaf, fi, compression);
1941         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1942         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1943
1944         btrfs_mark_buffer_dirty(leaf);
1945         btrfs_release_path(path);
1946
1947         inode_add_bytes(inode, num_bytes);
1948
1949         ins.objectid = disk_bytenr;
1950         ins.offset = disk_num_bytes;
1951         ins.type = BTRFS_EXTENT_ITEM_KEY;
1952         ret = btrfs_alloc_reserved_file_extent(trans, root,
1953                                         root->root_key.objectid,
1954                                         btrfs_ino(inode), file_pos, &ins);
1955 out:
1956         btrfs_free_path(path);
1957
1958         return ret;
1959 }
1960
1961 /* snapshot-aware defrag */
1962 struct sa_defrag_extent_backref {
1963         struct rb_node node;
1964         struct old_sa_defrag_extent *old;
1965         u64 root_id;
1966         u64 inum;
1967         u64 file_pos;
1968         u64 extent_offset;
1969         u64 num_bytes;
1970         u64 generation;
1971 };
1972
1973 struct old_sa_defrag_extent {
1974         struct list_head list;
1975         struct new_sa_defrag_extent *new;
1976
1977         u64 extent_offset;
1978         u64 bytenr;
1979         u64 offset;
1980         u64 len;
1981         int count;
1982 };
1983
1984 struct new_sa_defrag_extent {
1985         struct rb_root root;
1986         struct list_head head;
1987         struct btrfs_path *path;
1988         struct inode *inode;
1989         u64 file_pos;
1990         u64 len;
1991         u64 bytenr;
1992         u64 disk_len;
1993         u8 compress_type;
1994 };
1995
1996 static int backref_comp(struct sa_defrag_extent_backref *b1,
1997                         struct sa_defrag_extent_backref *b2)
1998 {
1999         if (b1->root_id < b2->root_id)
2000                 return -1;
2001         else if (b1->root_id > b2->root_id)
2002                 return 1;
2003
2004         if (b1->inum < b2->inum)
2005                 return -1;
2006         else if (b1->inum > b2->inum)
2007                 return 1;
2008
2009         if (b1->file_pos < b2->file_pos)
2010                 return -1;
2011         else if (b1->file_pos > b2->file_pos)
2012                 return 1;
2013
2014         /*
2015          * [------------------------------] ===> (a range of space)
2016          *     |<--->|   |<---->| =============> (fs/file tree A)
2017          * |<---------------------------->| ===> (fs/file tree B)
2018          *
2019          * A range of space can refer to two file extents in one tree while
2020          * refer to only one file extent in another tree.
2021          *
2022          * So we may process a disk offset more than one time(two extents in A)
2023          * and locate at the same extent(one extent in B), then insert two same
2024          * backrefs(both refer to the extent in B).
2025          */
2026         return 0;
2027 }
2028
2029 static void backref_insert(struct rb_root *root,
2030                            struct sa_defrag_extent_backref *backref)
2031 {
2032         struct rb_node **p = &root->rb_node;
2033         struct rb_node *parent = NULL;
2034         struct sa_defrag_extent_backref *entry;
2035         int ret;
2036
2037         while (*p) {
2038                 parent = *p;
2039                 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2040
2041                 ret = backref_comp(backref, entry);
2042                 if (ret < 0)
2043                         p = &(*p)->rb_left;
2044                 else
2045                         p = &(*p)->rb_right;
2046         }
2047
2048         rb_link_node(&backref->node, parent, p);
2049         rb_insert_color(&backref->node, root);
2050 }
2051
2052 /*
2053  * Note the backref might has changed, and in this case we just return 0.
2054  */
2055 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2056                                        void *ctx)
2057 {
2058         struct btrfs_file_extent_item *extent;
2059         struct btrfs_fs_info *fs_info;
2060         struct old_sa_defrag_extent *old = ctx;
2061         struct new_sa_defrag_extent *new = old->new;
2062         struct btrfs_path *path = new->path;
2063         struct btrfs_key key;
2064         struct btrfs_root *root;
2065         struct sa_defrag_extent_backref *backref;
2066         struct extent_buffer *leaf;
2067         struct inode *inode = new->inode;
2068         int slot;
2069         int ret;
2070         u64 extent_offset;
2071         u64 num_bytes;
2072
2073         if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2074             inum == btrfs_ino(inode))
2075                 return 0;
2076
2077         key.objectid = root_id;
2078         key.type = BTRFS_ROOT_ITEM_KEY;
2079         key.offset = (u64)-1;
2080
2081         fs_info = BTRFS_I(inode)->root->fs_info;
2082         root = btrfs_read_fs_root_no_name(fs_info, &key);
2083         if (IS_ERR(root)) {
2084                 if (PTR_ERR(root) == -ENOENT)
2085                         return 0;
2086                 WARN_ON(1);
2087                 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2088                          inum, offset, root_id);
2089                 return PTR_ERR(root);
2090         }
2091
2092         key.objectid = inum;
2093         key.type = BTRFS_EXTENT_DATA_KEY;
2094         if (offset > (u64)-1 << 32)
2095                 key.offset = 0;
2096         else
2097                 key.offset = offset;
2098
2099         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2100         if (ret < 0) {
2101                 WARN_ON(1);
2102                 return ret;
2103         }
2104
2105         while (1) {
2106                 cond_resched();
2107
2108                 leaf = path->nodes[0];
2109                 slot = path->slots[0];
2110
2111                 if (slot >= btrfs_header_nritems(leaf)) {
2112                         ret = btrfs_next_leaf(root, path);
2113                         if (ret < 0) {
2114                                 goto out;
2115                         } else if (ret > 0) {
2116                                 ret = 0;
2117                                 goto out;
2118                         }
2119                         continue;
2120                 }
2121
2122                 path->slots[0]++;
2123
2124                 btrfs_item_key_to_cpu(leaf, &key, slot);
2125
2126                 if (key.objectid > inum)
2127                         goto out;
2128
2129                 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2130                         continue;
2131
2132                 extent = btrfs_item_ptr(leaf, slot,
2133                                         struct btrfs_file_extent_item);
2134
2135                 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2136                         continue;
2137
2138                 extent_offset = btrfs_file_extent_offset(leaf, extent);
2139                 if (key.offset - extent_offset != offset)
2140                         continue;
2141
2142                 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2143                 if (extent_offset >= old->extent_offset + old->offset +
2144                     old->len || extent_offset + num_bytes <=
2145                     old->extent_offset + old->offset)
2146                         continue;
2147
2148                 break;
2149         }
2150
2151         backref = kmalloc(sizeof(*backref), GFP_NOFS);
2152         if (!backref) {
2153                 ret = -ENOENT;
2154                 goto out;
2155         }
2156
2157         backref->root_id = root_id;
2158         backref->inum = inum;
2159         backref->file_pos = offset + extent_offset;
2160         backref->num_bytes = num_bytes;
2161         backref->extent_offset = extent_offset;
2162         backref->generation = btrfs_file_extent_generation(leaf, extent);
2163         backref->old = old;
2164         backref_insert(&new->root, backref);
2165         old->count++;
2166 out:
2167         btrfs_release_path(path);
2168         WARN_ON(ret);
2169         return ret;
2170 }
2171
2172 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2173                                    struct new_sa_defrag_extent *new)
2174 {
2175         struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2176         struct old_sa_defrag_extent *old, *tmp;
2177         int ret;
2178
2179         new->path = path;
2180
2181         list_for_each_entry_safe(old, tmp, &new->head, list) {
2182                 ret = iterate_inodes_from_logical(old->bytenr, fs_info,
2183                                                   path, record_one_backref,
2184                                                   old);
2185                 BUG_ON(ret < 0 && ret != -ENOENT);
2186
2187                 /* no backref to be processed for this extent */
2188                 if (!old->count) {
2189                         list_del(&old->list);
2190                         kfree(old);
2191                 }
2192         }
2193
2194         if (list_empty(&new->head))
2195                 return false;
2196
2197         return true;
2198 }
2199
2200 static int relink_is_mergable(struct extent_buffer *leaf,
2201                               struct btrfs_file_extent_item *fi,
2202                               u64 disk_bytenr)
2203 {
2204         if (btrfs_file_extent_disk_bytenr(leaf, fi) != disk_bytenr)
2205                 return 0;
2206
2207         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2208                 return 0;
2209
2210         if (btrfs_file_extent_compression(leaf, fi) ||
2211             btrfs_file_extent_encryption(leaf, fi) ||
2212             btrfs_file_extent_other_encoding(leaf, fi))
2213                 return 0;
2214
2215         return 1;
2216 }
2217
2218 /*
2219  * Note the backref might has changed, and in this case we just return 0.
2220  */
2221 static noinline int relink_extent_backref(struct btrfs_path *path,
2222                                  struct sa_defrag_extent_backref *prev,
2223                                  struct sa_defrag_extent_backref *backref)
2224 {
2225         struct btrfs_file_extent_item *extent;
2226         struct btrfs_file_extent_item *item;
2227         struct btrfs_ordered_extent *ordered;
2228         struct btrfs_trans_handle *trans;
2229         struct btrfs_fs_info *fs_info;
2230         struct btrfs_root *root;
2231         struct btrfs_key key;
2232         struct extent_buffer *leaf;
2233         struct old_sa_defrag_extent *old = backref->old;
2234         struct new_sa_defrag_extent *new = old->new;
2235         struct inode *src_inode = new->inode;
2236         struct inode *inode;
2237         struct extent_state *cached = NULL;
2238         int ret = 0;
2239         u64 start;
2240         u64 len;
2241         u64 lock_start;
2242         u64 lock_end;
2243         bool merge = false;
2244         int index;
2245
2246         if (prev && prev->root_id == backref->root_id &&
2247             prev->inum == backref->inum &&
2248             prev->file_pos + prev->num_bytes == backref->file_pos)
2249                 merge = true;
2250
2251         /* step 1: get root */
2252         key.objectid = backref->root_id;
2253         key.type = BTRFS_ROOT_ITEM_KEY;
2254         key.offset = (u64)-1;
2255
2256         fs_info = BTRFS_I(src_inode)->root->fs_info;
2257         index = srcu_read_lock(&fs_info->subvol_srcu);
2258
2259         root = btrfs_read_fs_root_no_name(fs_info, &key);
2260         if (IS_ERR(root)) {
2261                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2262                 if (PTR_ERR(root) == -ENOENT)
2263                         return 0;
2264                 return PTR_ERR(root);
2265         }
2266         if (btrfs_root_refs(&root->root_item) == 0) {
2267                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2268                 /* parse ENOENT to 0 */
2269                 return 0;
2270         }
2271
2272         /* step 2: get inode */
2273         key.objectid = backref->inum;
2274         key.type = BTRFS_INODE_ITEM_KEY;
2275         key.offset = 0;
2276
2277         inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2278         if (IS_ERR(inode)) {
2279                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2280                 return 0;
2281         }
2282
2283         srcu_read_unlock(&fs_info->subvol_srcu, index);
2284
2285         /* step 3: relink backref */
2286         lock_start = backref->file_pos;
2287         lock_end = backref->file_pos + backref->num_bytes - 1;
2288         lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2289                          0, &cached);
2290
2291         ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2292         if (ordered) {
2293                 btrfs_put_ordered_extent(ordered);
2294                 goto out_unlock;
2295         }
2296
2297         trans = btrfs_join_transaction(root);
2298         if (IS_ERR(trans)) {
2299                 ret = PTR_ERR(trans);
2300                 goto out_unlock;
2301         }
2302
2303         key.objectid = backref->inum;
2304         key.type = BTRFS_EXTENT_DATA_KEY;
2305         key.offset = backref->file_pos;
2306
2307         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2308         if (ret < 0) {
2309                 goto out_free_path;
2310         } else if (ret > 0) {
2311                 ret = 0;
2312                 goto out_free_path;
2313         }
2314
2315         extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2316                                 struct btrfs_file_extent_item);
2317
2318         if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2319             backref->generation)
2320                 goto out_free_path;
2321
2322         btrfs_release_path(path);
2323
2324         start = backref->file_pos;
2325         if (backref->extent_offset < old->extent_offset + old->offset)
2326                 start += old->extent_offset + old->offset -
2327                          backref->extent_offset;
2328
2329         len = min(backref->extent_offset + backref->num_bytes,
2330                   old->extent_offset + old->offset + old->len);
2331         len -= max(backref->extent_offset, old->extent_offset + old->offset);
2332
2333         ret = btrfs_drop_extents(trans, root, inode, start,
2334                                  start + len, 1);
2335         if (ret)
2336                 goto out_free_path;
2337 again:
2338         key.objectid = btrfs_ino(inode);
2339         key.type = BTRFS_EXTENT_DATA_KEY;
2340         key.offset = start;
2341
2342         path->leave_spinning = 1;
2343         if (merge) {
2344                 struct btrfs_file_extent_item *fi;
2345                 u64 extent_len;
2346                 struct btrfs_key found_key;
2347
2348                 ret = btrfs_search_slot(trans, root, &key, path, 1, 1);
2349                 if (ret < 0)
2350                         goto out_free_path;
2351
2352                 path->slots[0]--;
2353                 leaf = path->nodes[0];
2354                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2355
2356                 fi = btrfs_item_ptr(leaf, path->slots[0],
2357                                     struct btrfs_file_extent_item);
2358                 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2359
2360                 if (relink_is_mergable(leaf, fi, new->bytenr) &&
2361                     extent_len + found_key.offset == start) {
2362                         btrfs_set_file_extent_num_bytes(leaf, fi,
2363                                                         extent_len + len);
2364                         btrfs_mark_buffer_dirty(leaf);
2365                         inode_add_bytes(inode, len);
2366
2367                         ret = 1;
2368                         goto out_free_path;
2369                 } else {
2370                         merge = false;
2371                         btrfs_release_path(path);
2372                         goto again;
2373                 }
2374         }
2375
2376         ret = btrfs_insert_empty_item(trans, root, path, &key,
2377                                         sizeof(*extent));
2378         if (ret) {
2379                 btrfs_abort_transaction(trans, root, ret);
2380                 goto out_free_path;
2381         }
2382
2383         leaf = path->nodes[0];
2384         item = btrfs_item_ptr(leaf, path->slots[0],
2385                                 struct btrfs_file_extent_item);
2386         btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2387         btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2388         btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2389         btrfs_set_file_extent_num_bytes(leaf, item, len);
2390         btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2391         btrfs_set_file_extent_generation(leaf, item, trans->transid);
2392         btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2393         btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2394         btrfs_set_file_extent_encryption(leaf, item, 0);
2395         btrfs_set_file_extent_other_encoding(leaf, item, 0);
2396
2397         btrfs_mark_buffer_dirty(leaf);
2398         inode_add_bytes(inode, len);
2399         btrfs_release_path(path);
2400
2401         ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2402                         new->disk_len, 0,
2403                         backref->root_id, backref->inum,
2404                         new->file_pos, 0);      /* start - extent_offset */
2405         if (ret) {
2406                 btrfs_abort_transaction(trans, root, ret);
2407                 goto out_free_path;
2408         }
2409
2410         ret = 1;
2411 out_free_path:
2412         btrfs_release_path(path);
2413         path->leave_spinning = 0;
2414         btrfs_end_transaction(trans, root);
2415 out_unlock:
2416         unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2417                              &cached, GFP_NOFS);
2418         iput(inode);
2419         return ret;
2420 }
2421
2422 static void relink_file_extents(struct new_sa_defrag_extent *new)
2423 {
2424         struct btrfs_path *path;
2425         struct old_sa_defrag_extent *old, *tmp;
2426         struct sa_defrag_extent_backref *backref;
2427         struct sa_defrag_extent_backref *prev = NULL;
2428         struct inode *inode;
2429         struct btrfs_root *root;
2430         struct rb_node *node;
2431         int ret;
2432
2433         inode = new->inode;
2434         root = BTRFS_I(inode)->root;
2435
2436         path = btrfs_alloc_path();
2437         if (!path)
2438                 return;
2439
2440         if (!record_extent_backrefs(path, new)) {
2441                 btrfs_free_path(path);
2442                 goto out;
2443         }
2444         btrfs_release_path(path);
2445
2446         while (1) {
2447                 node = rb_first(&new->root);
2448                 if (!node)
2449                         break;
2450                 rb_erase(node, &new->root);
2451
2452                 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2453
2454                 ret = relink_extent_backref(path, prev, backref);
2455                 WARN_ON(ret < 0);
2456
2457                 kfree(prev);
2458
2459                 if (ret == 1)
2460                         prev = backref;
2461                 else
2462                         prev = NULL;
2463                 cond_resched();
2464         }
2465         kfree(prev);
2466
2467         btrfs_free_path(path);
2468
2469         list_for_each_entry_safe(old, tmp, &new->head, list) {
2470                 list_del(&old->list);
2471                 kfree(old);
2472         }
2473 out:
2474         atomic_dec(&root->fs_info->defrag_running);
2475         wake_up(&root->fs_info->transaction_wait);
2476
2477         kfree(new);
2478 }
2479
2480 static struct new_sa_defrag_extent *
2481 record_old_file_extents(struct inode *inode,
2482                         struct btrfs_ordered_extent *ordered)
2483 {
2484         struct btrfs_root *root = BTRFS_I(inode)->root;
2485         struct btrfs_path *path;
2486         struct btrfs_key key;
2487         struct old_sa_defrag_extent *old, *tmp;
2488         struct new_sa_defrag_extent *new;
2489         int ret;
2490
2491         new = kmalloc(sizeof(*new), GFP_NOFS);
2492         if (!new)
2493                 return NULL;
2494
2495         new->inode = inode;
2496         new->file_pos = ordered->file_offset;
2497         new->len = ordered->len;
2498         new->bytenr = ordered->start;
2499         new->disk_len = ordered->disk_len;
2500         new->compress_type = ordered->compress_type;
2501         new->root = RB_ROOT;
2502         INIT_LIST_HEAD(&new->head);
2503
2504         path = btrfs_alloc_path();
2505         if (!path)
2506                 goto out_kfree;
2507
2508         key.objectid = btrfs_ino(inode);
2509         key.type = BTRFS_EXTENT_DATA_KEY;
2510         key.offset = new->file_pos;
2511
2512         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2513         if (ret < 0)
2514                 goto out_free_path;
2515         if (ret > 0 && path->slots[0] > 0)
2516                 path->slots[0]--;
2517
2518         /* find out all the old extents for the file range */
2519         while (1) {
2520                 struct btrfs_file_extent_item *extent;
2521                 struct extent_buffer *l;
2522                 int slot;
2523                 u64 num_bytes;
2524                 u64 offset;
2525                 u64 end;
2526                 u64 disk_bytenr;
2527                 u64 extent_offset;
2528
2529                 l = path->nodes[0];
2530                 slot = path->slots[0];
2531
2532                 if (slot >= btrfs_header_nritems(l)) {
2533                         ret = btrfs_next_leaf(root, path);
2534                         if (ret < 0)
2535                                 goto out_free_list;
2536                         else if (ret > 0)
2537                                 break;
2538                         continue;
2539                 }
2540
2541                 btrfs_item_key_to_cpu(l, &key, slot);
2542
2543                 if (key.objectid != btrfs_ino(inode))
2544                         break;
2545                 if (key.type != BTRFS_EXTENT_DATA_KEY)
2546                         break;
2547                 if (key.offset >= new->file_pos + new->len)
2548                         break;
2549
2550                 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2551
2552                 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2553                 if (key.offset + num_bytes < new->file_pos)
2554                         goto next;
2555
2556                 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2557                 if (!disk_bytenr)
2558                         goto next;
2559
2560                 extent_offset = btrfs_file_extent_offset(l, extent);
2561
2562                 old = kmalloc(sizeof(*old), GFP_NOFS);
2563                 if (!old)
2564                         goto out_free_list;
2565
2566                 offset = max(new->file_pos, key.offset);
2567                 end = min(new->file_pos + new->len, key.offset + num_bytes);
2568
2569                 old->bytenr = disk_bytenr;
2570                 old->extent_offset = extent_offset;
2571                 old->offset = offset - key.offset;
2572                 old->len = end - offset;
2573                 old->new = new;
2574                 old->count = 0;
2575                 list_add_tail(&old->list, &new->head);
2576 next:
2577                 path->slots[0]++;
2578                 cond_resched();
2579         }
2580
2581         btrfs_free_path(path);
2582         atomic_inc(&root->fs_info->defrag_running);
2583
2584         return new;
2585
2586 out_free_list:
2587         list_for_each_entry_safe(old, tmp, &new->head, list) {
2588                 list_del(&old->list);
2589                 kfree(old);
2590         }
2591 out_free_path:
2592         btrfs_free_path(path);
2593 out_kfree:
2594         kfree(new);
2595         return NULL;
2596 }
2597
2598 /*
2599  * helper function for btrfs_finish_ordered_io, this
2600  * just reads in some of the csum leaves to prime them into ram
2601  * before we start the transaction.  It limits the amount of btree
2602  * reads required while inside the transaction.
2603  */
2604 /* as ordered data IO finishes, this gets called so we can finish
2605  * an ordered extent if the range of bytes in the file it covers are
2606  * fully written.
2607  */
2608 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2609 {
2610         struct inode *inode = ordered_extent->inode;
2611         struct btrfs_root *root = BTRFS_I(inode)->root;
2612         struct btrfs_trans_handle *trans = NULL;
2613         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2614         struct extent_state *cached_state = NULL;
2615         struct new_sa_defrag_extent *new = NULL;
2616         int compress_type = 0;
2617         int ret;
2618         bool nolock;
2619
2620         nolock = btrfs_is_free_space_inode(inode);
2621
2622         if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2623                 ret = -EIO;
2624                 goto out;
2625         }
2626
2627         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2628                 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2629                 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2630                 if (nolock)
2631                         trans = btrfs_join_transaction_nolock(root);
2632                 else
2633                         trans = btrfs_join_transaction(root);
2634                 if (IS_ERR(trans)) {
2635                         ret = PTR_ERR(trans);
2636                         trans = NULL;
2637                         goto out;
2638                 }
2639                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2640                 ret = btrfs_update_inode_fallback(trans, root, inode);
2641                 if (ret) /* -ENOMEM or corruption */
2642                         btrfs_abort_transaction(trans, root, ret);
2643                 goto out;
2644         }
2645
2646         lock_extent_bits(io_tree, ordered_extent->file_offset,
2647                          ordered_extent->file_offset + ordered_extent->len - 1,
2648                          0, &cached_state);
2649
2650         ret = test_range_bit(io_tree, ordered_extent->file_offset,
2651                         ordered_extent->file_offset + ordered_extent->len - 1,
2652                         EXTENT_DEFRAG, 1, cached_state);
2653         if (ret) {
2654                 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2655                 if (last_snapshot >= BTRFS_I(inode)->generation)
2656                         /* the inode is shared */
2657                         new = record_old_file_extents(inode, ordered_extent);
2658
2659                 clear_extent_bit(io_tree, ordered_extent->file_offset,
2660                         ordered_extent->file_offset + ordered_extent->len - 1,
2661                         EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2662         }
2663
2664         if (nolock)
2665                 trans = btrfs_join_transaction_nolock(root);
2666         else
2667                 trans = btrfs_join_transaction(root);
2668         if (IS_ERR(trans)) {
2669                 ret = PTR_ERR(trans);
2670                 trans = NULL;
2671                 goto out_unlock;
2672         }
2673         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2674
2675         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2676                 compress_type = ordered_extent->compress_type;
2677         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2678                 BUG_ON(compress_type);
2679                 ret = btrfs_mark_extent_written(trans, inode,
2680                                                 ordered_extent->file_offset,
2681                                                 ordered_extent->file_offset +
2682                                                 ordered_extent->len);
2683         } else {
2684                 BUG_ON(root == root->fs_info->tree_root);
2685                 ret = insert_reserved_file_extent(trans, inode,
2686                                                 ordered_extent->file_offset,
2687                                                 ordered_extent->start,
2688                                                 ordered_extent->disk_len,
2689                                                 ordered_extent->len,
2690                                                 ordered_extent->len,
2691                                                 compress_type, 0, 0,
2692                                                 BTRFS_FILE_EXTENT_REG);
2693         }
2694         unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2695                            ordered_extent->file_offset, ordered_extent->len,
2696                            trans->transid);
2697         if (ret < 0) {
2698                 btrfs_abort_transaction(trans, root, ret);
2699                 goto out_unlock;
2700         }
2701
2702         add_pending_csums(trans, inode, ordered_extent->file_offset,
2703                           &ordered_extent->list);
2704
2705         btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2706         ret = btrfs_update_inode_fallback(trans, root, inode);
2707         if (ret) { /* -ENOMEM or corruption */
2708                 btrfs_abort_transaction(trans, root, ret);
2709                 goto out_unlock;
2710         }
2711         ret = 0;
2712 out_unlock:
2713         unlock_extent_cached(io_tree, ordered_extent->file_offset,
2714                              ordered_extent->file_offset +
2715                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
2716 out:
2717         if (root != root->fs_info->tree_root)
2718                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2719         if (trans)
2720                 btrfs_end_transaction(trans, root);
2721
2722         if (ret) {
2723                 clear_extent_uptodate(io_tree, ordered_extent->file_offset,
2724                                       ordered_extent->file_offset +
2725                                       ordered_extent->len - 1, NULL, GFP_NOFS);
2726
2727                 /*
2728                  * If the ordered extent had an IOERR or something else went
2729                  * wrong we need to return the space for this ordered extent
2730                  * back to the allocator.
2731                  */
2732                 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2733                     !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2734                         btrfs_free_reserved_extent(root, ordered_extent->start,
2735                                                    ordered_extent->disk_len);
2736         }
2737
2738
2739         /*
2740          * This needs to be done to make sure anybody waiting knows we are done
2741          * updating everything for this ordered extent.
2742          */
2743         btrfs_remove_ordered_extent(inode, ordered_extent);
2744
2745         /* for snapshot-aware defrag */
2746         if (new)
2747                 relink_file_extents(new);
2748
2749         /* once for us */
2750         btrfs_put_ordered_extent(ordered_extent);
2751         /* once for the tree */
2752         btrfs_put_ordered_extent(ordered_extent);
2753
2754         return ret;
2755 }
2756
2757 static void finish_ordered_fn(struct btrfs_work *work)
2758 {
2759         struct btrfs_ordered_extent *ordered_extent;
2760         ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2761         btrfs_finish_ordered_io(ordered_extent);
2762 }
2763
2764 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2765                                 struct extent_state *state, int uptodate)
2766 {
2767         struct inode *inode = page->mapping->host;
2768         struct btrfs_root *root = BTRFS_I(inode)->root;
2769         struct btrfs_ordered_extent *ordered_extent = NULL;
2770         struct btrfs_workers *workers;
2771
2772         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2773
2774         ClearPagePrivate2(page);
2775         if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2776                                             end - start + 1, uptodate))
2777                 return 0;
2778
2779         ordered_extent->work.func = finish_ordered_fn;
2780         ordered_extent->work.flags = 0;
2781
2782         if (btrfs_is_free_space_inode(inode))
2783                 workers = &root->fs_info->endio_freespace_worker;
2784         else
2785                 workers = &root->fs_info->endio_write_workers;
2786         btrfs_queue_worker(workers, &ordered_extent->work);
2787
2788         return 0;
2789 }
2790
2791 /*
2792  * when reads are done, we need to check csums to verify the data is correct
2793  * if there's a match, we allow the bio to finish.  If not, the code in
2794  * extent_io.c will try to find good copies for us.
2795  */
2796 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
2797                                struct extent_state *state, int mirror)
2798 {
2799         size_t offset = start - page_offset(page);
2800         struct inode *inode = page->mapping->host;
2801         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2802         char *kaddr;
2803         u64 private = ~(u32)0;
2804         int ret;
2805         struct btrfs_root *root = BTRFS_I(inode)->root;
2806         u32 csum = ~(u32)0;
2807         static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
2808                                       DEFAULT_RATELIMIT_BURST);
2809
2810         if (PageChecked(page)) {
2811                 ClearPageChecked(page);
2812                 goto good;
2813         }
2814
2815         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2816                 goto good;
2817
2818         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2819             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2820                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2821                                   GFP_NOFS);
2822                 return 0;
2823         }
2824
2825         if (state && state->start == start) {
2826                 private = state->private;
2827                 ret = 0;
2828         } else {
2829                 ret = get_state_private(io_tree, start, &private);
2830         }
2831         kaddr = kmap_atomic(page);
2832         if (ret)
2833                 goto zeroit;
2834
2835         csum = btrfs_csum_data(kaddr + offset, csum,  end - start + 1);
2836         btrfs_csum_final(csum, (char *)&csum);
2837         if (csum != private)
2838                 goto zeroit;
2839
2840         kunmap_atomic(kaddr);
2841 good:
2842         return 0;
2843
2844 zeroit:
2845         if (__ratelimit(&_rs))
2846                 btrfs_info(root->fs_info, "csum failed ino %llu off %llu csum %u private %llu",
2847                         (unsigned long long)btrfs_ino(page->mapping->host),
2848                         (unsigned long long)start, csum,
2849                         (unsigned long long)private);
2850         memset(kaddr + offset, 1, end - start + 1);
2851         flush_dcache_page(page);
2852         kunmap_atomic(kaddr);
2853         if (private == 0)
2854                 return 0;
2855         return -EIO;
2856 }
2857
2858 struct delayed_iput {
2859         struct list_head list;
2860         struct inode *inode;
2861 };
2862
2863 /* JDM: If this is fs-wide, why can't we add a pointer to
2864  * btrfs_inode instead and avoid the allocation? */
2865 void btrfs_add_delayed_iput(struct inode *inode)
2866 {
2867         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2868         struct delayed_iput *delayed;
2869
2870         if (atomic_add_unless(&inode->i_count, -1, 1))
2871                 return;
2872
2873         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2874         delayed->inode = inode;
2875
2876         spin_lock(&fs_info->delayed_iput_lock);
2877         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2878         spin_unlock(&fs_info->delayed_iput_lock);
2879 }
2880
2881 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2882 {
2883         LIST_HEAD(list);
2884         struct btrfs_fs_info *fs_info = root->fs_info;
2885         struct delayed_iput *delayed;
2886         int empty;
2887
2888         spin_lock(&fs_info->delayed_iput_lock);
2889         empty = list_empty(&fs_info->delayed_iputs);
2890         spin_unlock(&fs_info->delayed_iput_lock);
2891         if (empty)
2892                 return;
2893
2894         spin_lock(&fs_info->delayed_iput_lock);
2895         list_splice_init(&fs_info->delayed_iputs, &list);
2896         spin_unlock(&fs_info->delayed_iput_lock);
2897
2898         while (!list_empty(&list)) {
2899                 delayed = list_entry(list.next, struct delayed_iput, list);
2900                 list_del(&delayed->list);
2901                 iput(delayed->inode);
2902                 kfree(delayed);
2903         }
2904 }
2905
2906 /*
2907  * This is called in transaction commit time. If there are no orphan
2908  * files in the subvolume, it removes orphan item and frees block_rsv
2909  * structure.
2910  */
2911 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2912                               struct btrfs_root *root)
2913 {
2914         struct btrfs_block_rsv *block_rsv;
2915         int ret;
2916
2917         if (atomic_read(&root->orphan_inodes) ||
2918             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2919                 return;
2920
2921         spin_lock(&root->orphan_lock);
2922         if (atomic_read(&root->orphan_inodes)) {
2923                 spin_unlock(&root->orphan_lock);
2924                 return;
2925         }
2926
2927         if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2928                 spin_unlock(&root->orphan_lock);
2929                 return;
2930         }
2931
2932         block_rsv = root->orphan_block_rsv;
2933         root->orphan_block_rsv = NULL;
2934         spin_unlock(&root->orphan_lock);
2935
2936         if (root->orphan_item_inserted &&
2937             btrfs_root_refs(&root->root_item) > 0) {
2938                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2939                                             root->root_key.objectid);
2940                 BUG_ON(ret);
2941                 root->orphan_item_inserted = 0;
2942         }
2943
2944         if (block_rsv) {
2945                 WARN_ON(block_rsv->size > 0);
2946                 btrfs_free_block_rsv(root, block_rsv);
2947         }
2948 }
2949
2950 /*
2951  * This creates an orphan entry for the given inode in case something goes
2952  * wrong in the middle of an unlink/truncate.
2953  *
2954  * NOTE: caller of this function should reserve 5 units of metadata for
2955  *       this function.
2956  */
2957 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2958 {
2959         struct btrfs_root *root = BTRFS_I(inode)->root;
2960         struct btrfs_block_rsv *block_rsv = NULL;
2961         int reserve = 0;
2962         int insert = 0;
2963         int ret;
2964
2965         if (!root->orphan_block_rsv) {
2966                 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2967                 if (!block_rsv)
2968                         return -ENOMEM;
2969         }
2970
2971         spin_lock(&root->orphan_lock);
2972         if (!root->orphan_block_rsv) {
2973                 root->orphan_block_rsv = block_rsv;
2974         } else if (block_rsv) {
2975                 btrfs_free_block_rsv(root, block_rsv);
2976                 block_rsv = NULL;
2977         }
2978
2979         if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2980                               &BTRFS_I(inode)->runtime_flags)) {
2981 #if 0
2982                 /*
2983                  * For proper ENOSPC handling, we should do orphan
2984                  * cleanup when mounting. But this introduces backward
2985                  * compatibility issue.
2986                  */
2987                 if (!xchg(&root->orphan_item_inserted, 1))
2988                         insert = 2;
2989                 else
2990                         insert = 1;
2991 #endif
2992                 insert = 1;
2993                 atomic_inc(&root->orphan_inodes);
2994         }
2995
2996         if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2997                               &BTRFS_I(inode)->runtime_flags))
2998                 reserve = 1;
2999         spin_unlock(&root->orphan_lock);
3000
3001         /* grab metadata reservation from transaction handle */
3002         if (reserve) {
3003                 ret = btrfs_orphan_reserve_metadata(trans, inode);
3004                 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
3005         }
3006
3007         /* insert an orphan item to track this unlinked/truncated file */
3008         if (insert >= 1) {
3009                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3010                 if (ret && ret != -EEXIST) {
3011                         clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3012                                   &BTRFS_I(inode)->runtime_flags);
3013                         btrfs_abort_transaction(trans, root, ret);
3014                         return ret;
3015                 }
3016                 ret = 0;
3017         }
3018
3019         /* insert an orphan item to track subvolume contains orphan files */
3020         if (insert >= 2) {
3021                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3022                                                root->root_key.objectid);
3023                 if (ret && ret != -EEXIST) {
3024                         btrfs_abort_transaction(trans, root, ret);
3025                         return ret;
3026                 }
3027         }
3028         return 0;
3029 }
3030
3031 /*
3032  * We have done the truncate/delete so we can go ahead and remove the orphan
3033  * item for this particular inode.
3034  */
3035 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3036                             struct inode *inode)
3037 {
3038         struct btrfs_root *root = BTRFS_I(inode)->root;
3039         int delete_item = 0;
3040         int release_rsv = 0;
3041         int ret = 0;
3042
3043         spin_lock(&root->orphan_lock);
3044         if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3045                                &BTRFS_I(inode)->runtime_flags))
3046                 delete_item = 1;
3047
3048         if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3049                                &BTRFS_I(inode)->runtime_flags))
3050                 release_rsv = 1;
3051         spin_unlock(&root->orphan_lock);
3052
3053         if (trans && delete_item) {
3054                 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
3055                 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
3056         }
3057
3058         if (release_rsv) {
3059                 btrfs_orphan_release_metadata(inode);
3060                 atomic_dec(&root->orphan_inodes);
3061         }
3062
3063         return 0;
3064 }
3065
3066 /*
3067  * this cleans up any orphans that may be left on the list from the last use
3068  * of this root.
3069  */
3070 int btrfs_orphan_cleanup(struct btrfs_root *root)
3071 {
3072         struct btrfs_path *path;
3073         struct extent_buffer *leaf;
3074         struct btrfs_key key, found_key;
3075         struct btrfs_trans_handle *trans;
3076         struct inode *inode;
3077         u64 last_objectid = 0;
3078         int ret = 0, nr_unlink = 0, nr_truncate = 0;
3079
3080         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3081                 return 0;
3082
3083         path = btrfs_alloc_path();
3084         if (!path) {
3085                 ret = -ENOMEM;
3086                 goto out;
3087         }
3088         path->reada = -1;
3089
3090         key.objectid = BTRFS_ORPHAN_OBJECTID;
3091         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
3092         key.offset = (u64)-1;
3093
3094         while (1) {
3095                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3096                 if (ret < 0)
3097                         goto out;
3098
3099                 /*
3100                  * if ret == 0 means we found what we were searching for, which
3101                  * is weird, but possible, so only screw with path if we didn't
3102                  * find the key and see if we have stuff that matches
3103                  */
3104                 if (ret > 0) {
3105                         ret = 0;
3106                         if (path->slots[0] == 0)
3107                                 break;
3108                         path->slots[0]--;
3109                 }
3110
3111                 /* pull out the item */
3112                 leaf = path->nodes[0];
3113                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3114
3115                 /* make sure the item matches what we want */
3116                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3117                         break;
3118                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
3119                         break;
3120
3121                 /* release the path since we're done with it */
3122                 btrfs_release_path(path);
3123
3124                 /*
3125                  * this is where we are basically btrfs_lookup, without the
3126                  * crossing root thing.  we store the inode number in the
3127                  * offset of the orphan item.
3128                  */
3129
3130                 if (found_key.offset == last_objectid) {
3131                         btrfs_err(root->fs_info,
3132                                 "Error removing orphan entry, stopping orphan cleanup");
3133                         ret = -EINVAL;
3134                         goto out;
3135                 }
3136
3137                 last_objectid = found_key.offset;
3138
3139                 found_key.objectid = found_key.offset;
3140                 found_key.type = BTRFS_INODE_ITEM_KEY;
3141                 found_key.offset = 0;
3142                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3143                 ret = PTR_RET(inode);
3144                 if (ret && ret != -ESTALE)
3145                         goto out;
3146
3147                 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3148                         struct btrfs_root *dead_root;
3149                         struct btrfs_fs_info *fs_info = root->fs_info;
3150                         int is_dead_root = 0;
3151
3152                         /*
3153                          * this is an orphan in the tree root. Currently these
3154                          * could come from 2 sources:
3155                          *  a) a snapshot deletion in progress
3156                          *  b) a free space cache inode
3157                          * We need to distinguish those two, as the snapshot
3158                          * orphan must not get deleted.
3159                          * find_dead_roots already ran before us, so if this
3160                          * is a snapshot deletion, we should find the root
3161                          * in the dead_roots list
3162                          */
3163                         spin_lock(&fs_info->trans_lock);
3164                         list_for_each_entry(dead_root, &fs_info->dead_roots,
3165                                             root_list) {
3166                                 if (dead_root->root_key.objectid ==
3167                                     found_key.objectid) {
3168                                         is_dead_root = 1;
3169                                         break;
3170                                 }
3171                         }
3172                         spin_unlock(&fs_info->trans_lock);
3173                         if (is_dead_root) {
3174                                 /* prevent this orphan from being found again */
3175                                 key.offset = found_key.objectid - 1;
3176                                 continue;
3177                         }
3178                 }
3179                 /*
3180                  * Inode is already gone but the orphan item is still there,
3181                  * kill the orphan item.
3182                  */
3183                 if (ret == -ESTALE) {
3184                         trans = btrfs_start_transaction(root, 1);
3185                         if (IS_ERR(trans)) {
3186                                 ret = PTR_ERR(trans);
3187                                 goto out;
3188                         }
3189                         btrfs_debug(root->fs_info, "auto deleting %Lu",
3190                                 found_key.objectid);
3191                         ret = btrfs_del_orphan_item(trans, root,
3192                                                     found_key.objectid);
3193                         BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
3194                         btrfs_end_transaction(trans, root);
3195                         continue;
3196                 }
3197
3198                 /*
3199                  * add this inode to the orphan list so btrfs_orphan_del does
3200                  * the proper thing when we hit it
3201                  */
3202                 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3203                         &BTRFS_I(inode)->runtime_flags);
3204                 atomic_inc(&root->orphan_inodes);
3205
3206                 /* if we have links, this was a truncate, lets do that */
3207                 if (inode->i_nlink) {
3208                         if (!S_ISREG(inode->i_mode)) {
3209                                 WARN_ON(1);
3210                                 iput(inode);
3211                                 continue;
3212                         }
3213                         nr_truncate++;
3214
3215                         /* 1 for the orphan item deletion. */
3216                         trans = btrfs_start_transaction(root, 1);
3217                         if (IS_ERR(trans)) {
3218                                 ret = PTR_ERR(trans);
3219                                 goto out;
3220                         }
3221                         ret = btrfs_orphan_add(trans, inode);
3222                         btrfs_end_transaction(trans, root);
3223                         if (ret)
3224                                 goto out;
3225
3226                         ret = btrfs_truncate(inode);
3227                         if (ret)
3228                                 btrfs_orphan_del(NULL, inode);
3229                 } else {
3230                         nr_unlink++;
3231                 }
3232
3233                 /* this will do delete_inode and everything for us */
3234                 iput(inode);
3235                 if (ret)
3236                         goto out;
3237         }
3238         /* release the path since we're done with it */
3239         btrfs_release_path(path);
3240
3241         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3242
3243         if (root->orphan_block_rsv)
3244                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3245                                         (u64)-1);
3246
3247         if (root->orphan_block_rsv || root->orphan_item_inserted) {
3248                 trans = btrfs_join_transaction(root);
3249                 if (!IS_ERR(trans))
3250                         btrfs_end_transaction(trans, root);
3251         }
3252
3253         if (nr_unlink)
3254                 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3255         if (nr_truncate)
3256                 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3257
3258 out:
3259         if (ret)
3260                 btrfs_crit(root->fs_info,
3261                         "could not do orphan cleanup %d", ret);
3262         btrfs_free_path(path);
3263         return ret;
3264 }
3265
3266 /*
3267  * very simple check to peek ahead in the leaf looking for xattrs.  If we
3268  * don't find any xattrs, we know there can't be any acls.
3269  *
3270  * slot is the slot the inode is in, objectid is the objectid of the inode
3271  */
3272 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3273                                           int slot, u64 objectid)
3274 {
3275         u32 nritems = btrfs_header_nritems(leaf);
3276         struct btrfs_key found_key;
3277         int scanned = 0;
3278
3279         slot++;
3280         while (slot < nritems) {
3281                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3282
3283                 /* we found a different objectid, there must not be acls */
3284                 if (found_key.objectid != objectid)
3285                         return 0;
3286
3287                 /* we found an xattr, assume we've got an acl */
3288                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
3289                         return 1;
3290
3291                 /*
3292                  * we found a key greater than an xattr key, there can't
3293                  * be any acls later on
3294                  */
3295                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3296                         return 0;
3297
3298                 slot++;
3299                 scanned++;
3300
3301                 /*
3302                  * it goes inode, inode backrefs, xattrs, extents,
3303                  * so if there are a ton of hard links to an inode there can
3304                  * be a lot of backrefs.  Don't waste time searching too hard,
3305                  * this is just an optimization
3306                  */
3307                 if (scanned >= 8)
3308                         break;
3309         }
3310         /* we hit the end of the leaf before we found an xattr or
3311          * something larger than an xattr.  We have to assume the inode
3312          * has acls
3313          */
3314         return 1;
3315 }
3316
3317 /*
3318  * read an inode from the btree into the in-memory inode
3319  */
3320 static void btrfs_read_locked_inode(struct inode *inode)
3321 {
3322         struct btrfs_path *path;
3323         struct extent_buffer *leaf;
3324         struct btrfs_inode_item *inode_item;
3325         struct btrfs_timespec *tspec;
3326         struct btrfs_root *root = BTRFS_I(inode)->root;
3327         struct btrfs_key location;
3328         int maybe_acls;
3329         u32 rdev;
3330         int ret;
3331         bool filled = false;
3332
3333         ret = btrfs_fill_inode(inode, &rdev);
3334         if (!ret)
3335                 filled = true;
3336
3337         path = btrfs_alloc_path();
3338         if (!path)
3339                 goto make_bad;
3340
3341         path->leave_spinning = 1;
3342         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3343
3344         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3345         if (ret)
3346                 goto make_bad;
3347
3348         leaf = path->nodes[0];
3349
3350         if (filled)
3351                 goto cache_acl;
3352
3353         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3354                                     struct btrfs_inode_item);
3355         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3356         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3357         i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3358         i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3359         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3360
3361         tspec = btrfs_inode_atime(inode_item);
3362         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3363         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3364
3365         tspec = btrfs_inode_mtime(inode_item);
3366         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3367         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3368
3369         tspec = btrfs_inode_ctime(inode_item);
3370         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3371         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3372
3373         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3374         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3375         BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3376
3377         /*
3378          * If we were modified in the current generation and evicted from memory
3379          * and then re-read we need to do a full sync since we don't have any
3380          * idea about which extents were modified before we were evicted from
3381          * cache.
3382          */
3383         if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3384                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3385                         &BTRFS_I(inode)->runtime_flags);
3386
3387         inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3388         inode->i_generation = BTRFS_I(inode)->generation;
3389         inode->i_rdev = 0;
3390         rdev = btrfs_inode_rdev(leaf, inode_item);
3391
3392         BTRFS_I(inode)->index_cnt = (u64)-1;
3393         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3394 cache_acl:
3395         /*
3396          * try to precache a NULL acl entry for files that don't have
3397          * any xattrs or acls
3398          */
3399         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3400                                            btrfs_ino(inode));
3401         if (!maybe_acls)
3402                 cache_no_acl(inode);
3403
3404         btrfs_free_path(path);
3405
3406         switch (inode->i_mode & S_IFMT) {
3407         case S_IFREG:
3408                 inode->i_mapping->a_ops = &btrfs_aops;
3409                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3410                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3411                 inode->i_fop = &btrfs_file_operations;
3412                 inode->i_op = &btrfs_file_inode_operations;
3413                 break;
3414         case S_IFDIR:
3415                 inode->i_fop = &btrfs_dir_file_operations;
3416                 if (root == root->fs_info->tree_root)
3417                         inode->i_op = &btrfs_dir_ro_inode_operations;
3418                 else
3419                         inode->i_op = &btrfs_dir_inode_operations;
3420                 break;
3421         case S_IFLNK:
3422                 inode->i_op = &btrfs_symlink_inode_operations;
3423                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3424                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3425                 break;
3426         default:
3427                 inode->i_op = &btrfs_special_inode_operations;
3428                 init_special_inode(inode, inode->i_mode, rdev);
3429                 break;
3430         }
3431
3432         btrfs_update_iflags(inode);
3433         return;
3434
3435 make_bad:
3436         btrfs_free_path(path);
3437         make_bad_inode(inode);
3438 }
3439
3440 /*
3441  * given a leaf and an inode, copy the inode fields into the leaf
3442  */
3443 static void fill_inode_item(struct btrfs_trans_handle *trans,
3444                             struct extent_buffer *leaf,
3445                             struct btrfs_inode_item *item,
3446                             struct inode *inode)
3447 {
3448         struct btrfs_map_token token;
3449
3450         btrfs_init_map_token(&token);
3451
3452         btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3453         btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3454         btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3455                                    &token);
3456         btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3457         btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3458
3459         btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3460                                      inode->i_atime.tv_sec, &token);
3461         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3462                                       inode->i_atime.tv_nsec, &token);
3463
3464         btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3465                                      inode->i_mtime.tv_sec, &token);
3466         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3467                                       inode->i_mtime.tv_nsec, &token);
3468
3469         btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3470                                      inode->i_ctime.tv_sec, &token);
3471         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3472                                       inode->i_ctime.tv_nsec, &token);
3473
3474         btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3475                                      &token);
3476         btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3477                                          &token);
3478         btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3479         btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3480         btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3481         btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3482         btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3483 }
3484
3485 /*
3486  * copy everything in the in-memory inode into the btree.
3487  */
3488 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3489                                 struct btrfs_root *root, struct inode *inode)
3490 {
3491         struct btrfs_inode_item *inode_item;
3492         struct btrfs_path *path;
3493         struct extent_buffer *leaf;
3494         int ret;
3495
3496         path = btrfs_alloc_path();
3497         if (!path)
3498                 return -ENOMEM;
3499
3500         path->leave_spinning = 1;
3501         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3502                                  1);
3503         if (ret) {
3504                 if (ret > 0)
3505                         ret = -ENOENT;
3506                 goto failed;
3507         }
3508
3509         btrfs_unlock_up_safe(path, 1);
3510         leaf = path->nodes[0];
3511         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3512                                     struct btrfs_inode_item);
3513
3514         fill_inode_item(trans, leaf, inode_item, inode);
3515         btrfs_mark_buffer_dirty(leaf);
3516         btrfs_set_inode_last_trans(trans, inode);
3517         ret = 0;
3518 failed:
3519         btrfs_free_path(path);
3520         return ret;
3521 }
3522
3523 /*
3524  * copy everything in the in-memory inode into the btree.
3525  */
3526 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3527                                 struct btrfs_root *root, struct inode *inode)
3528 {
3529         int ret;
3530
3531         /*
3532          * If the inode is a free space inode, we can deadlock during commit
3533          * if we put it into the delayed code.
3534          *
3535          * The data relocation inode should also be directly updated
3536          * without delay
3537          */
3538         if (!btrfs_is_free_space_inode(inode)
3539             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
3540                 btrfs_update_root_times(trans, root);
3541
3542                 ret = btrfs_delayed_update_inode(trans, root, inode);
3543                 if (!ret)
3544                         btrfs_set_inode_last_trans(trans, inode);
3545                 return ret;
3546         }
3547
3548         return btrfs_update_inode_item(trans, root, inode);
3549 }
3550
3551 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3552                                          struct btrfs_root *root,
3553                                          struct inode *inode)
3554 {
3555         int ret;
3556
3557         ret = btrfs_update_inode(trans, root, inode);
3558         if (ret == -ENOSPC)
3559                 return btrfs_update_inode_item(trans, root, inode);
3560         return ret;
3561 }
3562
3563 /*
3564  * unlink helper that gets used here in inode.c and in the tree logging
3565  * recovery code.  It remove a link in a directory with a given name, and
3566  * also drops the back refs in the inode to the directory
3567  */
3568 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3569                                 struct btrfs_root *root,
3570                                 struct inode *dir, struct inode *inode,
3571                                 const char *name, int name_len)
3572 {
3573         struct btrfs_path *path;
3574         int ret = 0;
3575         struct extent_buffer *leaf;
3576         struct btrfs_dir_item *di;
3577         struct btrfs_key key;
3578         u64 index;
3579         u64 ino = btrfs_ino(inode);
3580         u64 dir_ino = btrfs_ino(dir);
3581
3582         path = btrfs_alloc_path();
3583         if (!path) {
3584                 ret = -ENOMEM;
3585                 goto out;
3586         }
3587
3588         path->leave_spinning = 1;
3589         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3590                                     name, name_len, -1);
3591         if (IS_ERR(di)) {
3592                 ret = PTR_ERR(di);
3593                 goto err;
3594         }
3595         if (!di) {
3596                 ret = -ENOENT;
3597                 goto err;
3598         }
3599         leaf = path->nodes[0];
3600         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3601         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3602         if (ret)
3603                 goto err;
3604         btrfs_release_path(path);
3605
3606         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3607                                   dir_ino, &index);
3608         if (ret) {
3609                 btrfs_info(root->fs_info,
3610                         "failed to delete reference to %.*s, inode %llu parent %llu",
3611                         name_len, name,
3612                         (unsigned long long)ino, (unsigned long long)dir_ino);
3613                 btrfs_abort_transaction(trans, root, ret);
3614                 goto err;
3615         }
3616
3617         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3618         if (ret) {
3619                 btrfs_abort_transaction(trans, root, ret);
3620                 goto err;
3621         }
3622
3623         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3624                                          inode, dir_ino);
3625         if (ret != 0 && ret != -ENOENT) {
3626                 btrfs_abort_transaction(trans, root, ret);
3627                 goto err;
3628         }
3629
3630         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3631                                            dir, index);
3632         if (ret == -ENOENT)
3633                 ret = 0;
3634         else if (ret)
3635                 btrfs_abort_transaction(trans, root, ret);
3636 err:
3637         btrfs_free_path(path);
3638         if (ret)
3639                 goto out;
3640
3641         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3642         inode_inc_iversion(inode);
3643         inode_inc_iversion(dir);
3644         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3645         ret = btrfs_update_inode(trans, root, dir);
3646 out:
3647         return ret;
3648 }
3649
3650 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3651                        struct btrfs_root *root,
3652                        struct inode *dir, struct inode *inode,
3653                        const char *name, int name_len)
3654 {
3655         int ret;
3656         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3657         if (!ret) {
3658                 btrfs_drop_nlink(inode);
3659                 ret = btrfs_update_inode(trans, root, inode);
3660         }
3661         return ret;
3662 }
3663
3664
3665 /* helper to check if there is any shared block in the path */
3666 static int check_path_shared(struct btrfs_root *root,
3667                              struct btrfs_path *path)
3668 {
3669         struct extent_buffer *eb;
3670         int level;
3671         u64 refs = 1;
3672
3673         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
3674                 int ret;
3675
3676                 if (!path->nodes[level])
3677                         break;
3678                 eb = path->nodes[level];
3679                 if (!btrfs_block_can_be_shared(root, eb))
3680                         continue;
3681                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, level, 1,
3682                                                &refs, NULL);
3683                 if (refs > 1)
3684                         return 1;
3685         }
3686         return 0;
3687 }
3688
3689 /*
3690  * helper to start transaction for unlink and rmdir.
3691  *
3692  * unlink and rmdir are special in btrfs, they do not always free space.
3693  * so in enospc case, we should make sure they will free space before
3694  * allowing them to use the global metadata reservation.
3695  */
3696 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
3697                                                        struct dentry *dentry)
3698 {
3699         struct btrfs_trans_handle *trans;
3700         struct btrfs_root *root = BTRFS_I(dir)->root;
3701         struct btrfs_path *path;
3702         struct btrfs_dir_item *di;
3703         struct inode *inode = dentry->d_inode;
3704         u64 index;
3705         int check_link = 1;
3706         int err = -ENOSPC;
3707         int ret;
3708         u64 ino = btrfs_ino(inode);
3709         u64 dir_ino = btrfs_ino(dir);
3710
3711         /*
3712          * 1 for the possible orphan item
3713          * 1 for the dir item
3714          * 1 for the dir index
3715          * 1 for the inode ref
3716          * 1 for the inode
3717          */
3718         trans = btrfs_start_transaction(root, 5);
3719         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3720                 return trans;
3721
3722         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
3723                 return ERR_PTR(-ENOSPC);
3724
3725         /* check if there is someone else holds reference */
3726         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
3727                 return ERR_PTR(-ENOSPC);
3728
3729         if (atomic_read(&inode->i_count) > 2)
3730                 return ERR_PTR(-ENOSPC);
3731
3732         if (xchg(&root->fs_info->enospc_unlink, 1))
3733                 return ERR_PTR(-ENOSPC);
3734
3735         path = btrfs_alloc_path();
3736         if (!path) {
3737                 root->fs_info->enospc_unlink = 0;
3738                 return ERR_PTR(-ENOMEM);
3739         }
3740
3741         /* 1 for the orphan item */
3742         trans = btrfs_start_transaction(root, 1);
3743         if (IS_ERR(trans)) {
3744                 btrfs_free_path(path);
3745                 root->fs_info->enospc_unlink = 0;
3746                 return trans;
3747         }
3748
3749         path->skip_locking = 1;
3750         path->search_commit_root = 1;
3751
3752         ret = btrfs_lookup_inode(trans, root, path,
3753                                 &BTRFS_I(dir)->location, 0);
3754         if (ret < 0) {
3755                 err = ret;
3756                 goto out;
3757         }
3758         if (ret == 0) {
3759                 if (check_path_shared(root, path))
3760                         goto out;
3761         } else {
3762                 check_link = 0;
3763         }
3764         btrfs_release_path(path);
3765
3766         ret = btrfs_lookup_inode(trans, root, path,
3767                                 &BTRFS_I(inode)->location, 0);
3768         if (ret < 0) {
3769                 err = ret;
3770                 goto out;
3771         }
3772         if (ret == 0) {
3773                 if (check_path_shared(root, path))
3774                         goto out;
3775         } else {
3776                 check_link = 0;
3777         }
3778         btrfs_release_path(path);
3779
3780         if (ret == 0 && S_ISREG(inode->i_mode)) {
3781                 ret = btrfs_lookup_file_extent(trans, root, path,
3782                                                ino, (u64)-1, 0);
3783                 if (ret < 0) {
3784                         err = ret;
3785                         goto out;
3786                 }
3787                 BUG_ON(ret == 0); /* Corruption */
3788                 if (check_path_shared(root, path))
3789                         goto out;
3790                 btrfs_release_path(path);
3791         }
3792
3793         if (!check_link) {
3794                 err = 0;
3795                 goto out;
3796         }
3797
3798         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3799                                 dentry->d_name.name, dentry->d_name.len, 0);
3800         if (IS_ERR(di)) {
3801                 err = PTR_ERR(di);
3802                 goto out;
3803         }
3804         if (di) {
3805                 if (check_path_shared(root, path))
3806                         goto out;
3807         } else {
3808                 err = 0;
3809                 goto out;
3810         }
3811         btrfs_release_path(path);
3812
3813         ret = btrfs_get_inode_ref_index(trans, root, path, dentry->d_name.name,
3814                                         dentry->d_name.len, ino, dir_ino, 0,
3815                                         &index);
3816         if (ret) {
3817                 err = ret;
3818                 goto out;
3819         }
3820
3821         if (check_path_shared(root, path))
3822                 goto out;
3823
3824         btrfs_release_path(path);
3825
3826         /*
3827          * This is a commit root search, if we can lookup inode item and other
3828          * relative items in the commit root, it means the transaction of
3829          * dir/file creation has been committed, and the dir index item that we
3830          * delay to insert has also been inserted into the commit root. So
3831          * we needn't worry about the delayed insertion of the dir index item
3832          * here.
3833          */
3834         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
3835                                 dentry->d_name.name, dentry->d_name.len, 0);
3836         if (IS_ERR(di)) {
3837                 err = PTR_ERR(di);
3838                 goto out;
3839         }
3840         BUG_ON(ret == -ENOENT);
3841         if (check_path_shared(root, path))
3842                 goto out;
3843
3844         err = 0;
3845 out:
3846         btrfs_free_path(path);
3847         /* Migrate the orphan reservation over */
3848         if (!err)
3849                 err = btrfs_block_rsv_migrate(trans->block_rsv,
3850                                 &root->fs_info->global_block_rsv,
3851                                 trans->bytes_reserved);
3852
3853         if (err) {
3854                 btrfs_end_transaction(trans, root);
3855                 root->fs_info->enospc_unlink = 0;
3856                 return ERR_PTR(err);
3857         }
3858
3859         trans->block_rsv = &root->fs_info->global_block_rsv;
3860         return trans;
3861 }
3862
3863 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
3864                                struct btrfs_root *root)
3865 {
3866         if (trans->block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL) {
3867                 btrfs_block_rsv_release(root, trans->block_rsv,
3868                                         trans->bytes_reserved);
3869                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3870                 BUG_ON(!root->fs_info->enospc_unlink);
3871                 root->fs_info->enospc_unlink = 0;
3872         }
3873         btrfs_end_transaction(trans, root);
3874 }
3875
3876 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3877 {
3878         struct btrfs_root *root = BTRFS_I(dir)->root;
3879         struct btrfs_trans_handle *trans;
3880         struct inode *inode = dentry->d_inode;
3881         int ret;
3882
3883         trans = __unlink_start_trans(dir, dentry);
3884         if (IS_ERR(trans))
3885                 return PTR_ERR(trans);
3886
3887         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3888
3889         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3890                                  dentry->d_name.name, dentry->d_name.len);
3891         if (ret)
3892                 goto out;
3893
3894         if (inode->i_nlink == 0) {
3895                 ret = btrfs_orphan_add(trans, inode);
3896                 if (ret)
3897                         goto out;
3898         }
3899
3900 out:
3901         __unlink_end_trans(trans, root);
3902         btrfs_btree_balance_dirty(root);
3903         return ret;
3904 }
3905
3906 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3907                         struct btrfs_root *root,
3908                         struct inode *dir, u64 objectid,
3909                         const char *name, int name_len)
3910 {
3911         struct btrfs_path *path;
3912         struct extent_buffer *leaf;
3913         struct btrfs_dir_item *di;
3914         struct btrfs_key key;
3915         u64 index;
3916         int ret;
3917         u64 dir_ino = btrfs_ino(dir);
3918
3919         path = btrfs_alloc_path();
3920         if (!path)
3921                 return -ENOMEM;
3922
3923         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3924                                    name, name_len, -1);
3925         if (IS_ERR_OR_NULL(di)) {
3926                 if (!di)
3927                         ret = -ENOENT;
3928                 else
3929                         ret = PTR_ERR(di);
3930                 goto out;
3931         }
3932
3933         leaf = path->nodes[0];
3934         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3935         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3936         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3937         if (ret) {
3938                 btrfs_abort_transaction(trans, root, ret);
3939                 goto out;
3940         }
3941         btrfs_release_path(path);
3942
3943         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3944                                  objectid, root->root_key.objectid,
3945                                  dir_ino, &index, name, name_len);
3946         if (ret < 0) {
3947                 if (ret != -ENOENT) {
3948                         btrfs_abort_transaction(trans, root, ret);
3949                         goto out;
3950                 }
3951                 di = btrfs_search_dir_index_item(root, path, dir_ino,
3952                                                  name, name_len);
3953                 if (IS_ERR_OR_NULL(di)) {
3954                         if (!di)
3955                                 ret = -ENOENT;
3956                         else
3957                                 ret = PTR_ERR(di);
3958                         btrfs_abort_transaction(trans, root, ret);
3959                         goto out;
3960                 }
3961
3962                 leaf = path->nodes[0];
3963                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3964                 btrfs_release_path(path);
3965                 index = key.offset;
3966         }
3967         btrfs_release_path(path);
3968
3969         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3970         if (ret) {
3971                 btrfs_abort_transaction(trans, root, ret);
3972                 goto out;
3973         }
3974
3975         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3976         inode_inc_iversion(dir);
3977         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3978         ret = btrfs_update_inode_fallback(trans, root, dir);
3979         if (ret)
3980                 btrfs_abort_transaction(trans, root, ret);
3981 out:
3982         btrfs_free_path(path);
3983         return ret;
3984 }
3985
3986 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3987 {
3988         struct inode *inode = dentry->d_inode;
3989         int err = 0;
3990         struct btrfs_root *root = BTRFS_I(dir)->root;
3991         struct btrfs_trans_handle *trans;
3992
3993         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
3994                 return -ENOTEMPTY;
3995         if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3996                 return -EPERM;
3997
3998         trans = __unlink_start_trans(dir, dentry);
3999         if (IS_ERR(trans))
4000                 return PTR_ERR(trans);
4001
4002         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4003                 err = btrfs_unlink_subvol(trans, root, dir,
4004                                           BTRFS_I(inode)->location.objectid,
4005                                           dentry->d_name.name,
4006                                           dentry->d_name.len);
4007                 goto out;
4008         }
4009
4010         err = btrfs_orphan_add(trans, inode);
4011         if (err)
4012                 goto out;
4013
4014         /* now the directory is empty */
4015         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
4016                                  dentry->d_name.name, dentry->d_name.len);
4017         if (!err)
4018                 btrfs_i_size_write(inode, 0);
4019 out:
4020         __unlink_end_trans(trans, root);
4021         btrfs_btree_balance_dirty(root);
4022
4023         return err;
4024 }
4025
4026 /*
4027  * this can truncate away extent items, csum items and directory items.
4028  * It starts at a high offset and removes keys until it can't find
4029  * any higher than new_size
4030  *
4031  * csum items that cross the new i_size are truncated to the new size
4032  * as well.
4033  *
4034  * min_type is the minimum key type to truncate down to.  If set to 0, this
4035  * will kill all the items on this inode, including the INODE_ITEM_KEY.
4036  */
4037 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4038                                struct btrfs_root *root,
4039                                struct inode *inode,
4040                                u64 new_size, u32 min_type)
4041 {
4042         struct btrfs_path *path;
4043         struct extent_buffer *leaf;
4044         struct btrfs_file_extent_item *fi;
4045         struct btrfs_key key;
4046         struct btrfs_key found_key;
4047         u64 extent_start = 0;
4048         u64 extent_num_bytes = 0;
4049         u64 extent_offset = 0;
4050         u64 item_end = 0;
4051         u32 found_type = (u8)-1;
4052         int found_extent;
4053         int del_item;
4054         int pending_del_nr = 0;
4055         int pending_del_slot = 0;
4056         int extent_type = -1;
4057         int ret;
4058         int err = 0;
4059         u64 ino = btrfs_ino(inode);
4060
4061         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4062
4063         path = btrfs_alloc_path();
4064         if (!path)
4065                 return -ENOMEM;
4066         path->reada = -1;
4067
4068         /*
4069          * We want to drop from the next block forward in case this new size is
4070          * not block aligned since we will be keeping the last block of the
4071          * extent just the way it is.
4072          */
4073         if (root->ref_cows || root == root->fs_info->tree_root)
4074                 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4075                                         root->sectorsize), (u64)-1, 0);
4076
4077         /*
4078          * This function is also used to drop the items in the log tree before
4079          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4080          * it is used to drop the loged items. So we shouldn't kill the delayed
4081          * items.
4082          */
4083         if (min_type == 0 && root == BTRFS_I(inode)->root)
4084                 btrfs_kill_delayed_inode_items(inode);
4085
4086         key.objectid = ino;
4087         key.offset = (u64)-1;
4088         key.type = (u8)-1;
4089
4090 search_again:
4091         path->leave_spinning = 1;
4092         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4093         if (ret < 0) {
4094                 err = ret;
4095                 goto out;
4096         }
4097
4098         if (ret > 0) {
4099                 /* there are no items in the tree for us to truncate, we're
4100                  * done
4101                  */
4102                 if (path->slots[0] == 0)
4103                         goto out;
4104                 path->slots[0]--;
4105         }
4106
4107         while (1) {
4108                 fi = NULL;
4109                 leaf = path->nodes[0];
4110                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4111                 found_type = btrfs_key_type(&found_key);
4112
4113                 if (found_key.objectid != ino)
4114                         break;
4115
4116                 if (found_type < min_type)
4117                         break;
4118
4119                 item_end = found_key.offset;
4120                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4121                         fi = btrfs_item_ptr(leaf, path->slots[0],
4122                                             struct btrfs_file_extent_item);
4123                         extent_type = btrfs_file_extent_type(leaf, fi);
4124                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4125                                 item_end +=
4126                                     btrfs_file_extent_num_bytes(leaf, fi);
4127                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4128                                 item_end += btrfs_file_extent_inline_len(leaf,
4129                                                                          fi);
4130                         }
4131                         item_end--;
4132                 }
4133                 if (found_type > min_type) {
4134                         del_item = 1;
4135                 } else {
4136                         if (item_end < new_size)
4137                                 break;
4138                         if (found_key.offset >= new_size)
4139                                 del_item = 1;
4140                         else
4141                                 del_item = 0;
4142                 }
4143                 found_extent = 0;
4144                 /* FIXME, shrink the extent if the ref count is only 1 */
4145                 if (found_type != BTRFS_EXTENT_DATA_KEY)
4146                         goto delete;
4147
4148                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4149                         u64 num_dec;
4150                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4151                         if (!del_item) {
4152                                 u64 orig_num_bytes =
4153                                         btrfs_file_extent_num_bytes(leaf, fi);
4154                                 extent_num_bytes = ALIGN(new_size -
4155                                                 found_key.offset,
4156                                                 root->sectorsize);
4157                                 btrfs_set_file_extent_num_bytes(leaf, fi,
4158                                                          extent_num_bytes);
4159                                 num_dec = (orig_num_bytes -
4160                                            extent_num_bytes);
4161                                 if (root->ref_cows && extent_start != 0)
4162                                         inode_sub_bytes(inode, num_dec);
4163                                 btrfs_mark_buffer_dirty(leaf);
4164                         } else {
4165                                 extent_num_bytes =
4166                                         btrfs_file_extent_disk_num_bytes(leaf,
4167                                                                          fi);
4168                                 extent_offset = found_key.offset -
4169                                         btrfs_file_extent_offset(leaf, fi);
4170
4171                                 /* FIXME blocksize != 4096 */
4172                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4173                                 if (extent_start != 0) {
4174                                         found_extent = 1;
4175                                         if (root->ref_cows)
4176                                                 inode_sub_bytes(inode, num_dec);
4177                                 }
4178                         }
4179                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4180                         /*
4181                          * we can't truncate inline items that have had
4182                          * special encodings
4183                          */
4184                         if (!del_item &&
4185                             btrfs_file_extent_compression(leaf, fi) == 0 &&
4186                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
4187                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4188                                 u32 size = new_size - found_key.offset;
4189
4190                                 if (root->ref_cows) {
4191                                         inode_sub_bytes(inode, item_end + 1 -
4192                                                         new_size);
4193                                 }
4194                                 size =
4195                                     btrfs_file_extent_calc_inline_size(size);
4196                                 btrfs_truncate_item(root, path, size, 1);
4197                         } else if (root->ref_cows) {
4198                                 inode_sub_bytes(inode, item_end + 1 -
4199                                                 found_key.offset);
4200                         }
4201                 }
4202 delete:
4203                 if (del_item) {
4204                         if (!pending_del_nr) {
4205                                 /* no pending yet, add ourselves */
4206                                 pending_del_slot = path->slots[0];
4207                                 pending_del_nr = 1;
4208                         } else if (pending_del_nr &&
4209                                    path->slots[0] + 1 == pending_del_slot) {
4210                                 /* hop on the pending chunk */
4211                                 pending_del_nr++;
4212                                 pending_del_slot = path->slots[0];
4213                         } else {
4214                                 BUG();
4215                         }
4216                 } else {
4217                         break;
4218                 }
4219                 if (found_extent && (root->ref_cows ||
4220                                      root == root->fs_info->tree_root)) {
4221                         btrfs_set_path_blocking(path);
4222                         ret = btrfs_free_extent(trans, root, extent_start,
4223                                                 extent_num_bytes, 0,
4224                                                 btrfs_header_owner(leaf),
4225                                                 ino, extent_offset, 0);
4226                         BUG_ON(ret);
4227                 }
4228
4229                 if (found_type == BTRFS_INODE_ITEM_KEY)
4230                         break;
4231
4232                 if (path->slots[0] == 0 ||
4233                     path->slots[0] != pending_del_slot) {
4234                         if (pending_del_nr) {
4235                                 ret = btrfs_del_items(trans, root, path,
4236                                                 pending_del_slot,
4237                                                 pending_del_nr);
4238                                 if (ret) {
4239                                         btrfs_abort_transaction(trans,
4240                                                                 root, ret);
4241                                         goto error;
4242                                 }
4243                                 pending_del_nr = 0;
4244                         }
4245                         btrfs_release_path(path);
4246                         goto search_again;
4247                 } else {
4248                         path->slots[0]--;
4249                 }
4250         }
4251 out:
4252         if (pending_del_nr) {
4253                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4254                                       pending_del_nr);
4255                 if (ret)
4256                         btrfs_abort_transaction(trans, root, ret);
4257         }
4258 error:
4259         btrfs_free_path(path);
4260         return err;
4261 }
4262
4263 /*
4264  * btrfs_truncate_page - read, zero a chunk and write a page
4265  * @inode - inode that we're zeroing
4266  * @from - the offset to start zeroing
4267  * @len - the length to zero, 0 to zero the entire range respective to the
4268  *      offset
4269  * @front - zero up to the offset instead of from the offset on
4270  *
4271  * This will find the page for the "from" offset and cow the page and zero the
4272  * part we want to zero.  This is used with truncate and hole punching.
4273  */
4274 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4275                         int front)
4276 {
4277         struct address_space *mapping = inode->i_mapping;
4278         struct btrfs_root *root = BTRFS_I(inode)->root;
4279         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4280         struct btrfs_ordered_extent *ordered;
4281         struct extent_state *cached_state = NULL;
4282         char *kaddr;
4283         u32 blocksize = root->sectorsize;
4284         pgoff_t index = from >> PAGE_CACHE_SHIFT;
4285         unsigned offset = from & (PAGE_CACHE_SIZE-1);
4286         struct page *page;
4287         gfp_t mask = btrfs_alloc_write_mask(mapping);
4288         int ret = 0;
4289         u64 page_start;
4290         u64 page_end;
4291
4292         if ((offset & (blocksize - 1)) == 0 &&
4293             (!len || ((len & (blocksize - 1)) == 0)))
4294                 goto out;
4295         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
4296         if (ret)
4297                 goto out;
4298
4299 again:
4300         page = find_or_create_page(mapping, index, mask);
4301         if (!page) {
4302                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4303                 ret = -ENOMEM;
4304                 goto out;
4305         }
4306
4307         page_start = page_offset(page);
4308         page_end = page_start + PAGE_CACHE_SIZE - 1;
4309
4310         if (!PageUptodate(page)) {
4311                 ret = btrfs_readpage(NULL, page);
4312                 lock_page(page);
4313                 if (page->mapping != mapping) {
4314                         unlock_page(page);
4315                         page_cache_release(page);
4316                         goto again;
4317                 }
4318                 if (!PageUptodate(page)) {
4319                         ret = -EIO;
4320                         goto out_unlock;
4321                 }
4322         }
4323         wait_on_page_writeback(page);
4324
4325         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
4326         set_page_extent_mapped(page);
4327
4328         ordered = btrfs_lookup_ordered_extent(inode, page_start);
4329         if (ordered) {
4330                 unlock_extent_cached(io_tree, page_start, page_end,
4331                                      &cached_state, GFP_NOFS);
4332                 unlock_page(page);
4333                 page_cache_release(page);
4334                 btrfs_start_ordered_extent(inode, ordered, 1);
4335                 btrfs_put_ordered_extent(ordered);
4336                 goto again;
4337         }
4338
4339         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
4340                           EXTENT_DIRTY | EXTENT_DELALLOC |
4341                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4342                           0, 0, &cached_state, GFP_NOFS);
4343
4344         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4345                                         &cached_state);
4346         if (ret) {
4347                 unlock_extent_cached(io_tree, page_start, page_end,
4348                                      &cached_state, GFP_NOFS);
4349                 goto out_unlock;
4350         }
4351
4352         if (offset != PAGE_CACHE_SIZE) {
4353                 if (!len)
4354                         len = PAGE_CACHE_SIZE - offset;
4355                 kaddr = kmap(page);
4356                 if (front)
4357                         memset(kaddr, 0, offset);
4358                 else
4359                         memset(kaddr + offset, 0, len);
4360                 flush_dcache_page(page);
4361                 kunmap(page);
4362         }
4363         ClearPageChecked(page);
4364         set_page_dirty(page);
4365         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4366                              GFP_NOFS);
4367
4368 out_unlock:
4369         if (ret)
4370                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4371         unlock_page(page);
4372         page_cache_release(page);
4373 out:
4374         return ret;
4375 }
4376
4377 /*
4378  * This function puts in dummy file extents for the area we're creating a hole
4379  * for.  So if we are truncating this file to a larger size we need to insert
4380  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4381  * the range between oldsize and size
4382  */
4383 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4384 {
4385         struct btrfs_trans_handle *trans;
4386         struct btrfs_root *root = BTRFS_I(inode)->root;
4387         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4388         struct extent_map *em = NULL;
4389         struct extent_state *cached_state = NULL;
4390         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4391         u64 hole_start = ALIGN(oldsize, root->sectorsize);
4392         u64 block_end = ALIGN(size, root->sectorsize);
4393         u64 last_byte;
4394         u64 cur_offset;
4395         u64 hole_size;
4396         int err = 0;
4397
4398         if (size <= hole_start)
4399                 return 0;
4400
4401         while (1) {
4402                 struct btrfs_ordered_extent *ordered;
4403                 btrfs_wait_ordered_range(inode, hole_start,
4404                                          block_end - hole_start);
4405                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
4406                                  &cached_state);
4407                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
4408                 if (!ordered)
4409                         break;
4410                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4411                                      &cached_state, GFP_NOFS);
4412                 btrfs_put_ordered_extent(ordered);
4413         }
4414
4415         cur_offset = hole_start;
4416         while (1) {
4417                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4418                                 block_end - cur_offset, 0);
4419                 if (IS_ERR(em)) {
4420                         err = PTR_ERR(em);
4421                         em = NULL;
4422                         break;
4423                 }
4424                 last_byte = min(extent_map_end(em), block_end);
4425                 last_byte = ALIGN(last_byte , root->sectorsize);
4426                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4427                         struct extent_map *hole_em;
4428                         hole_size = last_byte - cur_offset;
4429
4430                         trans = btrfs_start_transaction(root, 3);
4431                         if (IS_ERR(trans)) {
4432                                 err = PTR_ERR(trans);
4433                                 break;
4434                         }
4435
4436                         err = btrfs_drop_extents(trans, root, inode,
4437                                                  cur_offset,
4438                                                  cur_offset + hole_size, 1);
4439                         if (err) {
4440                                 btrfs_abort_transaction(trans, root, err);
4441                                 btrfs_end_transaction(trans, root);
4442                                 break;
4443                         }
4444
4445                         err = btrfs_insert_file_extent(trans, root,
4446                                         btrfs_ino(inode), cur_offset, 0,
4447                                         0, hole_size, 0, hole_size,
4448                                         0, 0, 0);
4449                         if (err) {
4450                                 btrfs_abort_transaction(trans, root, err);
4451                                 btrfs_end_transaction(trans, root);
4452                                 break;
4453                         }
4454
4455                         btrfs_drop_extent_cache(inode, cur_offset,
4456                                                 cur_offset + hole_size - 1, 0);
4457                         hole_em = alloc_extent_map();
4458                         if (!hole_em) {
4459                                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4460                                         &BTRFS_I(inode)->runtime_flags);
4461                                 goto next;
4462                         }
4463                         hole_em->start = cur_offset;
4464                         hole_em->len = hole_size;
4465                         hole_em->orig_start = cur_offset;
4466
4467                         hole_em->block_start = EXTENT_MAP_HOLE;
4468                         hole_em->block_len = 0;
4469                         hole_em->orig_block_len = 0;
4470                         hole_em->ram_bytes = hole_size;
4471                         hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4472                         hole_em->compress_type = BTRFS_COMPRESS_NONE;
4473                         hole_em->generation = trans->transid;
4474
4475                         while (1) {
4476                                 write_lock(&em_tree->lock);
4477                                 err = add_extent_mapping(em_tree, hole_em, 1);
4478                                 write_unlock(&em_tree->lock);
4479                                 if (err != -EEXIST)
4480                                         break;
4481                                 btrfs_drop_extent_cache(inode, cur_offset,
4482                                                         cur_offset +
4483                                                         hole_size - 1, 0);
4484                         }
4485                         free_extent_map(hole_em);
4486 next:
4487                         btrfs_update_inode(trans, root, inode);
4488                         btrfs_end_transaction(trans, root);
4489                 }
4490                 free_extent_map(em);
4491                 em = NULL;
4492                 cur_offset = last_byte;
4493                 if (cur_offset >= block_end)
4494                         break;
4495         }
4496
4497         free_extent_map(em);
4498         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4499                              GFP_NOFS);
4500         return err;
4501 }
4502
4503 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4504 {
4505         struct btrfs_root *root = BTRFS_I(inode)->root;
4506         struct btrfs_trans_handle *trans;
4507         loff_t oldsize = i_size_read(inode);
4508         loff_t newsize = attr->ia_size;
4509         int mask = attr->ia_valid;
4510         int ret;
4511
4512         if (newsize == oldsize)
4513                 return 0;
4514
4515         /*
4516          * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4517          * special case where we need to update the times despite not having
4518          * these flags set.  For all other operations the VFS set these flags
4519          * explicitly if it wants a timestamp update.
4520          */
4521         if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME))))
4522                 inode->i_ctime = inode->i_mtime = current_fs_time(inode->i_sb);
4523
4524         if (newsize > oldsize) {
4525                 truncate_pagecache(inode, oldsize, newsize);
4526                 ret = btrfs_cont_expand(inode, oldsize, newsize);
4527                 if (ret)
4528                         return ret;
4529
4530                 trans = btrfs_start_transaction(root, 1);
4531                 if (IS_ERR(trans))
4532                         return PTR_ERR(trans);
4533
4534                 i_size_write(inode, newsize);
4535                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4536                 ret = btrfs_update_inode(trans, root, inode);
4537                 btrfs_end_transaction(trans, root);
4538         } else {
4539
4540                 /*
4541                  * We're truncating a file that used to have good data down to
4542                  * zero. Make sure it gets into the ordered flush list so that
4543                  * any new writes get down to disk quickly.
4544                  */
4545                 if (newsize == 0)
4546                         set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4547                                 &BTRFS_I(inode)->runtime_flags);
4548
4549                 /*
4550                  * 1 for the orphan item we're going to add
4551                  * 1 for the orphan item deletion.
4552                  */
4553                 trans = btrfs_start_transaction(root, 2);
4554                 if (IS_ERR(trans))
4555                         return PTR_ERR(trans);
4556
4557                 /*
4558                  * We need to do this in case we fail at _any_ point during the
4559                  * actual truncate.  Once we do the truncate_setsize we could
4560                  * invalidate pages which forces any outstanding ordered io to
4561                  * be instantly completed which will give us extents that need
4562                  * to be truncated.  If we fail to get an orphan inode down we
4563                  * could have left over extents that were never meant to live,
4564                  * so we need to garuntee from this point on that everything
4565                  * will be consistent.
4566                  */
4567                 ret = btrfs_orphan_add(trans, inode);
4568                 btrfs_end_transaction(trans, root);
4569                 if (ret)
4570                         return ret;
4571
4572                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4573                 truncate_setsize(inode, newsize);
4574
4575                 /* Disable nonlocked read DIO to avoid the end less truncate */
4576                 btrfs_inode_block_unlocked_dio(inode);
4577                 inode_dio_wait(inode);
4578                 btrfs_inode_resume_unlocked_dio(inode);
4579
4580                 ret = btrfs_truncate(inode);
4581                 if (ret && inode->i_nlink)
4582                         btrfs_orphan_del(NULL, inode);
4583         }
4584
4585         return ret;
4586 }
4587
4588 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4589 {
4590         struct inode *inode = dentry->d_inode;
4591         struct btrfs_root *root = BTRFS_I(inode)->root;
4592         int err;
4593
4594         if (btrfs_root_readonly(root))
4595                 return -EROFS;
4596
4597         err = inode_change_ok(inode, attr);
4598         if (err)
4599                 return err;
4600
4601         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
4602                 err = btrfs_setsize(inode, attr);
4603                 if (err)
4604                         return err;
4605         }
4606
4607         if (attr->ia_valid) {
4608                 setattr_copy(inode, attr);
4609                 inode_inc_iversion(inode);
4610                 err = btrfs_dirty_inode(inode);
4611
4612                 if (!err && attr->ia_valid & ATTR_MODE)
4613                         err = btrfs_acl_chmod(inode);
4614         }
4615
4616         return err;
4617 }
4618
4619 void btrfs_evict_inode(struct inode *inode)
4620 {
4621         struct btrfs_trans_handle *trans;
4622         struct btrfs_root *root = BTRFS_I(inode)->root;
4623         struct btrfs_block_rsv *rsv, *global_rsv;
4624         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
4625         int ret;
4626
4627         trace_btrfs_inode_evict(inode);
4628
4629         truncate_inode_pages(&inode->i_data, 0);
4630         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
4631                                btrfs_is_free_space_inode(inode)))
4632                 goto no_delete;
4633
4634         if (is_bad_inode(inode)) {
4635                 btrfs_orphan_del(NULL, inode);
4636                 goto no_delete;
4637         }
4638         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4639         btrfs_wait_ordered_range(inode, 0, (u64)-1);
4640
4641         if (root->fs_info->log_root_recovering) {
4642                 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
4643                                  &BTRFS_I(inode)->runtime_flags));
4644                 goto no_delete;
4645         }
4646
4647         if (inode->i_nlink > 0) {
4648                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
4649                 goto no_delete;
4650         }
4651
4652         ret = btrfs_commit_inode_delayed_inode(inode);
4653         if (ret) {
4654                 btrfs_orphan_del(NULL, inode);
4655                 goto no_delete;
4656         }
4657
4658         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4659         if (!rsv) {
4660                 btrfs_orphan_del(NULL, inode);
4661                 goto no_delete;
4662         }
4663         rsv->size = min_size;
4664         rsv->failfast = 1;
4665         global_rsv = &root->fs_info->global_block_rsv;
4666
4667         btrfs_i_size_write(inode, 0);
4668
4669         /*
4670          * This is a bit simpler than btrfs_truncate since we've already
4671          * reserved our space for our orphan item in the unlink, so we just
4672          * need to reserve some slack space in case we add bytes and update
4673          * inode item when doing the truncate.
4674          */
4675         while (1) {
4676                 ret = btrfs_block_rsv_refill(root, rsv, min_size,
4677                                              BTRFS_RESERVE_FLUSH_LIMIT);
4678
4679                 /*
4680                  * Try and steal from the global reserve since we will
4681                  * likely not use this space anyway, we want to try as
4682                  * hard as possible to get this to work.
4683                  */
4684                 if (ret)
4685                         ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
4686
4687                 if (ret) {
4688                         btrfs_warn(root->fs_info,
4689                                 "Could not get space for a delete, will truncate on mount %d",
4690                                 ret);
4691                         btrfs_orphan_del(NULL, inode);
4692                         btrfs_free_block_rsv(root, rsv);
4693                         goto no_delete;
4694                 }
4695
4696                 trans = btrfs_join_transaction(root);
4697                 if (IS_ERR(trans)) {
4698                         btrfs_orphan_del(NULL, inode);
4699                         btrfs_free_block_rsv(root, rsv);
4700                         goto no_delete;
4701                 }
4702
4703                 trans->block_rsv = rsv;
4704
4705                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
4706                 if (ret != -ENOSPC)
4707                         break;
4708
4709                 trans->block_rsv = &root->fs_info->trans_block_rsv;
4710                 btrfs_end_transaction(trans, root);
4711                 trans = NULL;
4712                 btrfs_btree_balance_dirty(root);
4713         }
4714
4715         btrfs_free_block_rsv(root, rsv);
4716
4717         if (ret == 0) {
4718                 trans->block_rsv = root->orphan_block_rsv;
4719                 ret = btrfs_orphan_del(trans, inode);
4720                 BUG_ON(ret);
4721         }
4722
4723         trans->block_rsv = &root->fs_info->trans_block_rsv;
4724         if (!(root == root->fs_info->tree_root ||
4725               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
4726                 btrfs_return_ino(root, btrfs_ino(inode));
4727
4728         btrfs_end_transaction(trans, root);
4729         btrfs_btree_balance_dirty(root);
4730 no_delete:
4731         btrfs_remove_delayed_node(inode);
4732         clear_inode(inode);
4733         return;
4734 }
4735
4736 /*
4737  * this returns the key found in the dir entry in the location pointer.
4738  * If no dir entries were found, location->objectid is 0.
4739  */
4740 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4741                                struct btrfs_key *location)
4742 {
4743         const char *name = dentry->d_name.name;
4744         int namelen = dentry->d_name.len;
4745         struct btrfs_dir_item *di;
4746         struct btrfs_path *path;
4747         struct btrfs_root *root = BTRFS_I(dir)->root;
4748         int ret = 0;
4749
4750         path = btrfs_alloc_path();
4751         if (!path)
4752                 return -ENOMEM;
4753
4754         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
4755                                     namelen, 0);
4756         if (IS_ERR(di))
4757                 ret = PTR_ERR(di);
4758
4759         if (IS_ERR_OR_NULL(di))
4760                 goto out_err;
4761
4762         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
4763 out:
4764         btrfs_free_path(path);
4765         return ret;
4766 out_err:
4767         location->objectid = 0;
4768         goto out;
4769 }
4770
4771 /*
4772  * when we hit a tree root in a directory, the btrfs part of the inode
4773  * needs to be changed to reflect the root directory of the tree root.  This
4774  * is kind of like crossing a mount point.
4775  */
4776 static int fixup_tree_root_location(struct btrfs_root *root,
4777                                     struct inode *dir,
4778                                     struct dentry *dentry,
4779                                     struct btrfs_key *location,
4780                                     struct btrfs_root **sub_root)
4781 {
4782         struct btrfs_path *path;
4783         struct btrfs_root *new_root;
4784         struct btrfs_root_ref *ref;
4785         struct extent_buffer *leaf;
4786         int ret;
4787         int err = 0;
4788
4789         path = btrfs_alloc_path();
4790         if (!path) {
4791                 err = -ENOMEM;
4792                 goto out;
4793         }
4794
4795         err = -ENOENT;
4796         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
4797                                   BTRFS_I(dir)->root->root_key.objectid,
4798                                   location->objectid);
4799         if (ret) {
4800                 if (ret < 0)
4801                         err = ret;
4802                 goto out;
4803         }
4804
4805         leaf = path->nodes[0];
4806         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
4807         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4808             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4809                 goto out;
4810
4811         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4812                                    (unsigned long)(ref + 1),
4813                                    dentry->d_name.len);
4814         if (ret)
4815                 goto out;
4816
4817         btrfs_release_path(path);
4818
4819         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4820         if (IS_ERR(new_root)) {
4821                 err = PTR_ERR(new_root);
4822                 goto out;
4823         }
4824
4825         if (btrfs_root_refs(&new_root->root_item) == 0) {
4826                 err = -ENOENT;
4827                 goto out;
4828         }
4829
4830         *sub_root = new_root;
4831         location->objectid = btrfs_root_dirid(&new_root->root_item);
4832         location->type = BTRFS_INODE_ITEM_KEY;
4833         location->offset = 0;
4834         err = 0;
4835 out:
4836         btrfs_free_path(path);
4837         return err;
4838 }
4839
4840 static void inode_tree_add(struct inode *inode)
4841 {
4842         struct btrfs_root *root = BTRFS_I(inode)->root;
4843         struct btrfs_inode *entry;
4844         struct rb_node **p;
4845         struct rb_node *parent;
4846         u64 ino = btrfs_ino(inode);
4847
4848         if (inode_unhashed(inode))
4849                 return;
4850 again:
4851         parent = NULL;
4852         spin_lock(&root->inode_lock);
4853         p = &root->inode_tree.rb_node;
4854         while (*p) {
4855                 parent = *p;
4856                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4857
4858                 if (ino < btrfs_ino(&entry->vfs_inode))
4859                         p = &parent->rb_left;
4860                 else if (ino > btrfs_ino(&entry->vfs_inode))
4861                         p = &parent->rb_right;
4862                 else {
4863                         WARN_ON(!(entry->vfs_inode.i_state &
4864                                   (I_WILL_FREE | I_FREEING)));
4865                         rb_erase(parent, &root->inode_tree);
4866                         RB_CLEAR_NODE(parent);
4867                         spin_unlock(&root->inode_lock);
4868                         goto again;
4869                 }
4870         }
4871         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
4872         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4873         spin_unlock(&root->inode_lock);
4874 }
4875
4876 static void inode_tree_del(struct inode *inode)
4877 {
4878         struct btrfs_root *root = BTRFS_I(inode)->root;
4879         int empty = 0;
4880
4881         spin_lock(&root->inode_lock);
4882         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
4883                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4884                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4885                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4886         }
4887         spin_unlock(&root->inode_lock);
4888
4889         /*
4890          * Free space cache has inodes in the tree root, but the tree root has a
4891          * root_refs of 0, so this could end up dropping the tree root as a
4892          * snapshot, so we need the extra !root->fs_info->tree_root check to
4893          * make sure we don't drop it.
4894          */
4895         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
4896             root != root->fs_info->tree_root) {
4897                 synchronize_srcu(&root->fs_info->subvol_srcu);
4898                 spin_lock(&root->inode_lock);
4899                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4900                 spin_unlock(&root->inode_lock);
4901                 if (empty)
4902                         btrfs_add_dead_root(root);
4903         }
4904 }
4905
4906 void btrfs_invalidate_inodes(struct btrfs_root *root)
4907 {
4908         struct rb_node *node;
4909         struct rb_node *prev;
4910         struct btrfs_inode *entry;
4911         struct inode *inode;
4912         u64 objectid = 0;
4913
4914         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4915
4916         spin_lock(&root->inode_lock);
4917 again:
4918         node = root->inode_tree.rb_node;
4919         prev = NULL;
4920         while (node) {
4921                 prev = node;
4922                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4923
4924                 if (objectid < btrfs_ino(&entry->vfs_inode))
4925                         node = node->rb_left;
4926                 else if (objectid > btrfs_ino(&entry->vfs_inode))
4927                         node = node->rb_right;
4928                 else
4929                         break;
4930         }
4931         if (!node) {
4932                 while (prev) {
4933                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
4934                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4935                                 node = prev;
4936                                 break;
4937                         }
4938                         prev = rb_next(prev);
4939                 }
4940         }
4941         while (node) {
4942                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4943                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4944                 inode = igrab(&entry->vfs_inode);
4945                 if (inode) {
4946                         spin_unlock(&root->inode_lock);
4947                         if (atomic_read(&inode->i_count) > 1)
4948                                 d_prune_aliases(inode);
4949                         /*
4950                          * btrfs_drop_inode will have it removed from
4951                          * the inode cache when its usage count
4952                          * hits zero.
4953                          */
4954                         iput(inode);
4955                         cond_resched();
4956                         spin_lock(&root->inode_lock);
4957                         goto again;
4958                 }
4959
4960                 if (cond_resched_lock(&root->inode_lock))
4961                         goto again;
4962
4963                 node = rb_next(node);
4964         }
4965         spin_unlock(&root->inode_lock);
4966 }
4967
4968 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4969 {
4970         struct btrfs_iget_args *args = p;
4971         inode->i_ino = args->ino;
4972         BTRFS_I(inode)->root = args->root;
4973         return 0;
4974 }
4975
4976 static int btrfs_find_actor(struct inode *inode, void *opaque)
4977 {
4978         struct btrfs_iget_args *args = opaque;
4979         return args->ino == btrfs_ino(inode) &&
4980                 args->root == BTRFS_I(inode)->root;
4981 }
4982
4983 static struct inode *btrfs_iget_locked(struct super_block *s,
4984                                        u64 objectid,
4985                                        struct btrfs_root *root)
4986 {
4987         struct inode *inode;
4988         struct btrfs_iget_args args;
4989         args.ino = objectid;
4990         args.root = root;
4991
4992         inode = iget5_locked(s, objectid, btrfs_find_actor,
4993                              btrfs_init_locked_inode,
4994                              (void *)&args);
4995         return inode;
4996 }
4997
4998 /* Get an inode object given its location and corresponding root.
4999  * Returns in *is_new if the inode was read from disk
5000  */
5001 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5002                          struct btrfs_root *root, int *new)
5003 {
5004         struct inode *inode;
5005
5006         inode = btrfs_iget_locked(s, location->objectid, root);
5007         if (!inode)
5008                 return ERR_PTR(-ENOMEM);
5009
5010         if (inode->i_state & I_NEW) {
5011                 BTRFS_I(inode)->root = root;
5012                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
5013                 btrfs_read_locked_inode(inode);
5014                 if (!is_bad_inode(inode)) {
5015                         inode_tree_add(inode);
5016                         unlock_new_inode(inode);
5017                         if (new)
5018                                 *new = 1;
5019                 } else {
5020                         unlock_new_inode(inode);
5021                         iput(inode);
5022                         inode = ERR_PTR(-ESTALE);
5023                 }
5024         }
5025
5026         return inode;
5027 }
5028
5029 static struct inode *new_simple_dir(struct super_block *s,
5030                                     struct btrfs_key *key,
5031                                     struct btrfs_root *root)
5032 {
5033         struct inode *inode = new_inode(s);
5034
5035         if (!inode)
5036                 return ERR_PTR(-ENOMEM);
5037
5038         BTRFS_I(inode)->root = root;
5039         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5040         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5041
5042         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5043         inode->i_op = &btrfs_dir_ro_inode_operations;
5044         inode->i_fop = &simple_dir_operations;
5045         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5046         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5047
5048         return inode;
5049 }
5050
5051 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5052 {
5053         struct inode *inode;
5054         struct btrfs_root *root = BTRFS_I(dir)->root;
5055         struct btrfs_root *sub_root = root;
5056         struct btrfs_key location;
5057         int index;
5058         int ret = 0;
5059
5060         if (dentry->d_name.len > BTRFS_NAME_LEN)
5061                 return ERR_PTR(-ENAMETOOLONG);
5062
5063         ret = btrfs_inode_by_name(dir, dentry, &location);
5064         if (ret < 0)
5065                 return ERR_PTR(ret);
5066
5067         if (location.objectid == 0)
5068                 return NULL;
5069
5070         if (location.type == BTRFS_INODE_ITEM_KEY) {
5071                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5072                 return inode;
5073         }
5074
5075         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5076
5077         index = srcu_read_lock(&root->fs_info->subvol_srcu);
5078         ret = fixup_tree_root_location(root, dir, dentry,
5079                                        &location, &sub_root);
5080         if (ret < 0) {
5081                 if (ret != -ENOENT)
5082                         inode = ERR_PTR(ret);
5083                 else
5084                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
5085         } else {
5086                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5087         }
5088         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5089
5090         if (!IS_ERR(inode) && root != sub_root) {
5091                 down_read(&root->fs_info->cleanup_work_sem);
5092                 if (!(inode->i_sb->s_flags & MS_RDONLY))
5093                         ret = btrfs_orphan_cleanup(sub_root);
5094                 up_read(&root->fs_info->cleanup_work_sem);
5095                 if (ret)
5096                         inode = ERR_PTR(ret);
5097         }
5098
5099         return inode;
5100 }
5101
5102 static int btrfs_dentry_delete(const struct dentry *dentry)
5103 {
5104         struct btrfs_root *root;
5105         struct inode *inode = dentry->d_inode;
5106
5107         if (!inode && !IS_ROOT(dentry))
5108                 inode = dentry->d_parent->d_inode;
5109
5110         if (inode) {
5111                 root = BTRFS_I(inode)->root;
5112                 if (btrfs_root_refs(&root->root_item) == 0)
5113                         return 1;
5114
5115                 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5116                         return 1;
5117         }
5118         return 0;
5119 }
5120
5121 static void btrfs_dentry_release(struct dentry *dentry)
5122 {
5123         if (dentry->d_fsdata)
5124                 kfree(dentry->d_fsdata);
5125 }
5126
5127 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5128                                    unsigned int flags)
5129 {
5130         struct dentry *ret;
5131
5132         ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
5133         return ret;
5134 }
5135
5136 unsigned char btrfs_filetype_table[] = {
5137         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5138 };
5139
5140 static int btrfs_real_readdir(struct file *filp, void *dirent,
5141                               filldir_t filldir)
5142 {
5143         struct inode *inode = file_inode(filp);
5144         struct btrfs_root *root = BTRFS_I(inode)->root;
5145         struct btrfs_item *item;
5146         struct btrfs_dir_item *di;
5147         struct btrfs_key key;
5148         struct btrfs_key found_key;
5149         struct btrfs_path *path;
5150         struct list_head ins_list;
5151         struct list_head del_list;
5152         int ret;
5153         struct extent_buffer *leaf;
5154         int slot;
5155         unsigned char d_type;
5156         int over = 0;
5157         u32 di_cur;
5158         u32 di_total;
5159         u32 di_len;
5160         int key_type = BTRFS_DIR_INDEX_KEY;
5161         char tmp_name[32];
5162         char *name_ptr;
5163         int name_len;
5164         int is_curr = 0;        /* filp->f_pos points to the current index? */
5165
5166         /* FIXME, use a real flag for deciding about the key type */
5167         if (root->fs_info->tree_root == root)
5168                 key_type = BTRFS_DIR_ITEM_KEY;
5169
5170         /* special case for "." */
5171         if (filp->f_pos == 0) {
5172                 over = filldir(dirent, ".", 1,
5173                                filp->f_pos, btrfs_ino(inode), DT_DIR);
5174                 if (over)
5175                         return 0;
5176                 filp->f_pos = 1;
5177         }
5178         /* special case for .., just use the back ref */
5179         if (filp->f_pos == 1) {
5180                 u64 pino = parent_ino(filp->f_path.dentry);
5181                 over = filldir(dirent, "..", 2,
5182                                filp->f_pos, pino, DT_DIR);
5183                 if (over)
5184                         return 0;
5185                 filp->f_pos = 2;
5186         }
5187         path = btrfs_alloc_path();
5188         if (!path)
5189                 return -ENOMEM;
5190
5191         path->reada = 1;
5192
5193         if (key_type == BTRFS_DIR_INDEX_KEY) {
5194                 INIT_LIST_HEAD(&ins_list);
5195                 INIT_LIST_HEAD(&del_list);
5196                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5197         }
5198
5199         btrfs_set_key_type(&key, key_type);
5200         key.offset = filp->f_pos;
5201         key.objectid = btrfs_ino(inode);
5202
5203         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5204         if (ret < 0)
5205                 goto err;
5206
5207         while (1) {
5208                 leaf = path->nodes[0];
5209                 slot = path->slots[0];
5210                 if (slot >= btrfs_header_nritems(leaf)) {
5211                         ret = btrfs_next_leaf(root, path);
5212                         if (ret < 0)
5213                                 goto err;
5214                         else if (ret > 0)
5215                                 break;
5216                         continue;
5217                 }
5218
5219                 item = btrfs_item_nr(leaf, slot);
5220                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5221
5222                 if (found_key.objectid != key.objectid)
5223                         break;
5224                 if (btrfs_key_type(&found_key) != key_type)
5225                         break;
5226                 if (found_key.offset < filp->f_pos)
5227                         goto next;
5228                 if (key_type == BTRFS_DIR_INDEX_KEY &&
5229                     btrfs_should_delete_dir_index(&del_list,
5230                                                   found_key.offset))
5231                         goto next;
5232
5233                 filp->f_pos = found_key.offset;
5234                 is_curr = 1;
5235
5236                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5237                 di_cur = 0;
5238                 di_total = btrfs_item_size(leaf, item);
5239
5240                 while (di_cur < di_total) {
5241                         struct btrfs_key location;
5242
5243                         if (verify_dir_item(root, leaf, di))
5244                                 break;
5245
5246                         name_len = btrfs_dir_name_len(leaf, di);
5247                         if (name_len <= sizeof(tmp_name)) {
5248                                 name_ptr = tmp_name;
5249                         } else {
5250                                 name_ptr = kmalloc(name_len, GFP_NOFS);
5251                                 if (!name_ptr) {
5252                                         ret = -ENOMEM;
5253                                         goto err;
5254                                 }
5255                         }
5256                         read_extent_buffer(leaf, name_ptr,
5257                                            (unsigned long)(di + 1), name_len);
5258
5259                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5260                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
5261
5262
5263                         /* is this a reference to our own snapshot? If so
5264                          * skip it.
5265                          *
5266                          * In contrast to old kernels, we insert the snapshot's
5267                          * dir item and dir index after it has been created, so
5268                          * we won't find a reference to our own snapshot. We
5269                          * still keep the following code for backward
5270                          * compatibility.
5271                          */
5272                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
5273                             location.objectid == root->root_key.objectid) {
5274                                 over = 0;
5275                                 goto skip;
5276                         }
5277                         over = filldir(dirent, name_ptr, name_len,
5278                                        found_key.offset, location.objectid,
5279                                        d_type);
5280
5281 skip:
5282                         if (name_ptr != tmp_name)
5283                                 kfree(name_ptr);
5284
5285                         if (over)
5286                                 goto nopos;
5287                         di_len = btrfs_dir_name_len(leaf, di) +
5288                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
5289                         di_cur += di_len;
5290                         di = (struct btrfs_dir_item *)((char *)di + di_len);
5291                 }
5292 next:
5293                 path->slots[0]++;
5294         }
5295
5296         if (key_type == BTRFS_DIR_INDEX_KEY) {
5297                 if (is_curr)
5298                         filp->f_pos++;
5299                 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
5300                                                       &ins_list);
5301                 if (ret)
5302                         goto nopos;
5303         }
5304
5305         /* Reached end of directory/root. Bump pos past the last item. */
5306         if (key_type == BTRFS_DIR_INDEX_KEY)
5307                 /*
5308                  * 32-bit glibc will use getdents64, but then strtol -
5309                  * so the last number we can serve is this.
5310                  */
5311                 filp->f_pos = 0x7fffffff;
5312         else
5313                 filp->f_pos++;
5314 nopos:
5315         ret = 0;
5316 err:
5317         if (key_type == BTRFS_DIR_INDEX_KEY)
5318                 btrfs_put_delayed_items(&ins_list, &del_list);
5319         btrfs_free_path(path);
5320         return ret;
5321 }
5322
5323 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5324 {
5325         struct btrfs_root *root = BTRFS_I(inode)->root;
5326         struct btrfs_trans_handle *trans;
5327         int ret = 0;
5328         bool nolock = false;
5329
5330         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5331                 return 0;
5332
5333         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5334                 nolock = true;
5335
5336         if (wbc->sync_mode == WB_SYNC_ALL) {
5337                 if (nolock)
5338                         trans = btrfs_join_transaction_nolock(root);
5339                 else
5340                         trans = btrfs_join_transaction(root);
5341                 if (IS_ERR(trans))
5342                         return PTR_ERR(trans);
5343                 ret = btrfs_commit_transaction(trans, root);
5344         }
5345         return ret;
5346 }
5347
5348 /*
5349  * This is somewhat expensive, updating the tree every time the
5350  * inode changes.  But, it is most likely to find the inode in cache.
5351  * FIXME, needs more benchmarking...there are no reasons other than performance
5352  * to keep or drop this code.
5353  */
5354 static int btrfs_dirty_inode(struct inode *inode)
5355 {
5356         struct btrfs_root *root = BTRFS_I(inode)->root;
5357         struct btrfs_trans_handle *trans;
5358         int ret;
5359
5360         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5361                 return 0;
5362
5363         trans = btrfs_join_transaction(root);
5364         if (IS_ERR(trans))
5365                 return PTR_ERR(trans);
5366
5367         ret = btrfs_update_inode(trans, root, inode);
5368         if (ret && ret == -ENOSPC) {
5369                 /* whoops, lets try again with the full transaction */
5370                 btrfs_end_transaction(trans, root);
5371                 trans = btrfs_start_transaction(root, 1);
5372                 if (IS_ERR(trans))
5373                         return PTR_ERR(trans);
5374
5375                 ret = btrfs_update_inode(trans, root, inode);
5376         }
5377         btrfs_end_transaction(trans, root);
5378         if (BTRFS_I(inode)->delayed_node)
5379                 btrfs_balance_delayed_items(root);
5380
5381         return ret;
5382 }
5383
5384 /*
5385  * This is a copy of file_update_time.  We need this so we can return error on
5386  * ENOSPC for updating the inode in the case of file write and mmap writes.
5387  */
5388 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5389                              int flags)
5390 {
5391         struct btrfs_root *root = BTRFS_I(inode)->root;
5392
5393         if (btrfs_root_readonly(root))
5394                 return -EROFS;
5395
5396         if (flags & S_VERSION)
5397                 inode_inc_iversion(inode);
5398         if (flags & S_CTIME)
5399                 inode->i_ctime = *now;
5400         if (flags & S_MTIME)
5401                 inode->i_mtime = *now;
5402         if (flags & S_ATIME)
5403                 inode->i_atime = *now;
5404         return btrfs_dirty_inode(inode);
5405 }
5406
5407 /*
5408  * find the highest existing sequence number in a directory
5409  * and then set the in-memory index_cnt variable to reflect
5410  * free sequence numbers
5411  */
5412 static int btrfs_set_inode_index_count(struct inode *inode)
5413 {
5414         struct btrfs_root *root = BTRFS_I(inode)->root;
5415         struct btrfs_key key, found_key;
5416         struct btrfs_path *path;
5417         struct extent_buffer *leaf;
5418         int ret;
5419
5420         key.objectid = btrfs_ino(inode);
5421         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
5422         key.offset = (u64)-1;
5423
5424         path = btrfs_alloc_path();
5425         if (!path)
5426                 return -ENOMEM;
5427
5428         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5429         if (ret < 0)
5430                 goto out;
5431         /* FIXME: we should be able to handle this */
5432         if (ret == 0)
5433                 goto out;
5434         ret = 0;
5435
5436         /*
5437          * MAGIC NUMBER EXPLANATION:
5438          * since we search a directory based on f_pos we have to start at 2
5439          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5440          * else has to start at 2
5441          */
5442         if (path->slots[0] == 0) {
5443                 BTRFS_I(inode)->index_cnt = 2;
5444                 goto out;
5445         }
5446
5447         path->slots[0]--;
5448
5449         leaf = path->nodes[0];
5450         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5451
5452         if (found_key.objectid != btrfs_ino(inode) ||
5453             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
5454                 BTRFS_I(inode)->index_cnt = 2;
5455                 goto out;
5456         }
5457
5458         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5459 out:
5460         btrfs_free_path(path);
5461         return ret;
5462 }
5463
5464 /*
5465  * helper to find a free sequence number in a given directory.  This current
5466  * code is very simple, later versions will do smarter things in the btree
5467  */
5468 int btrfs_set_inode_index(struct inode *dir, u64 *index)
5469 {
5470         int ret = 0;
5471
5472         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
5473                 ret = btrfs_inode_delayed_dir_index_count(dir);
5474                 if (ret) {
5475                         ret = btrfs_set_inode_index_count(dir);
5476                         if (ret)
5477                                 return ret;
5478                 }
5479         }
5480
5481         *index = BTRFS_I(dir)->index_cnt;
5482         BTRFS_I(dir)->index_cnt++;
5483
5484         return ret;
5485 }
5486
5487 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5488                                      struct btrfs_root *root,
5489                                      struct inode *dir,
5490                                      const char *name, int name_len,
5491                                      u64 ref_objectid, u64 objectid,
5492                                      umode_t mode, u64 *index)
5493 {
5494         struct inode *inode;
5495         struct btrfs_inode_item *inode_item;
5496         struct btrfs_key *location;
5497         struct btrfs_path *path;
5498         struct btrfs_inode_ref *ref;
5499         struct btrfs_key key[2];
5500         u32 sizes[2];
5501         unsigned long ptr;
5502         int ret;
5503         int owner;
5504
5505         path = btrfs_alloc_path();
5506         if (!path)
5507                 return ERR_PTR(-ENOMEM);
5508
5509         inode = new_inode(root->fs_info->sb);
5510         if (!inode) {
5511                 btrfs_free_path(path);
5512                 return ERR_PTR(-ENOMEM);
5513         }
5514
5515         /*
5516          * we have to initialize this early, so we can reclaim the inode
5517          * number if we fail afterwards in this function.
5518          */
5519         inode->i_ino = objectid;
5520
5521         if (dir) {
5522                 trace_btrfs_inode_request(dir);
5523
5524                 ret = btrfs_set_inode_index(dir, index);
5525                 if (ret) {
5526                         btrfs_free_path(path);
5527                         iput(inode);
5528                         return ERR_PTR(ret);
5529                 }
5530         }
5531         /*
5532          * index_cnt is ignored for everything but a dir,
5533          * btrfs_get_inode_index_count has an explanation for the magic
5534          * number
5535          */
5536         BTRFS_I(inode)->index_cnt = 2;
5537         BTRFS_I(inode)->root = root;
5538         BTRFS_I(inode)->generation = trans->transid;
5539         inode->i_generation = BTRFS_I(inode)->generation;
5540
5541         /*
5542          * We could have gotten an inode number from somebody who was fsynced
5543          * and then removed in this same transaction, so let's just set full
5544          * sync since it will be a full sync anyway and this will blow away the
5545          * old info in the log.
5546          */
5547         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5548
5549         if (S_ISDIR(mode))
5550                 owner = 0;
5551         else
5552                 owner = 1;
5553
5554         key[0].objectid = objectid;
5555         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
5556         key[0].offset = 0;
5557
5558         /*
5559          * Start new inodes with an inode_ref. This is slightly more
5560          * efficient for small numbers of hard links since they will
5561          * be packed into one item. Extended refs will kick in if we
5562          * add more hard links than can fit in the ref item.
5563          */
5564         key[1].objectid = objectid;
5565         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
5566         key[1].offset = ref_objectid;
5567
5568         sizes[0] = sizeof(struct btrfs_inode_item);
5569         sizes[1] = name_len + sizeof(*ref);
5570
5571         path->leave_spinning = 1;
5572         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
5573         if (ret != 0)
5574                 goto fail;
5575
5576         inode_init_owner(inode, dir, mode);
5577         inode_set_bytes(inode, 0);
5578         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5579         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5580                                   struct btrfs_inode_item);
5581         memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
5582                              sizeof(*inode_item));
5583         fill_inode_item(trans, path->nodes[0], inode_item, inode);
5584
5585         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
5586                              struct btrfs_inode_ref);
5587         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
5588         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
5589         ptr = (unsigned long)(ref + 1);
5590         write_extent_buffer(path->nodes[0], name, ptr, name_len);
5591
5592         btrfs_mark_buffer_dirty(path->nodes[0]);
5593         btrfs_free_path(path);
5594
5595         location = &BTRFS_I(inode)->location;
5596         location->objectid = objectid;
5597         location->offset = 0;
5598         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
5599
5600         btrfs_inherit_iflags(inode, dir);
5601
5602         if (S_ISREG(mode)) {
5603                 if (btrfs_test_opt(root, NODATASUM))
5604                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
5605                 if (btrfs_test_opt(root, NODATACOW))
5606                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
5607                                 BTRFS_INODE_NODATASUM;
5608         }
5609
5610         insert_inode_hash(inode);
5611         inode_tree_add(inode);
5612
5613         trace_btrfs_inode_new(inode);
5614         btrfs_set_inode_last_trans(trans, inode);
5615
5616         btrfs_update_root_times(trans, root);
5617
5618         return inode;
5619 fail:
5620         if (dir)
5621                 BTRFS_I(dir)->index_cnt--;
5622         btrfs_free_path(path);
5623         iput(inode);
5624         return ERR_PTR(ret);
5625 }
5626
5627 static inline u8 btrfs_inode_type(struct inode *inode)
5628 {
5629         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
5630 }
5631
5632 /*
5633  * utility function to add 'inode' into 'parent_inode' with
5634  * a give name and a given sequence number.
5635  * if 'add_backref' is true, also insert a backref from the
5636  * inode to the parent directory.
5637  */
5638 int btrfs_add_link(struct btrfs_trans_handle *trans,
5639                    struct inode *parent_inode, struct inode *inode,
5640                    const char *name, int name_len, int add_backref, u64 index)
5641 {
5642         int ret = 0;
5643         struct btrfs_key key;
5644         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
5645         u64 ino = btrfs_ino(inode);
5646         u64 parent_ino = btrfs_ino(parent_inode);
5647
5648         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5649                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
5650         } else {
5651                 key.objectid = ino;
5652                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
5653                 key.offset = 0;
5654         }
5655
5656         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5657                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
5658                                          key.objectid, root->root_key.objectid,
5659                                          parent_ino, index, name, name_len);
5660         } else if (add_backref) {
5661                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
5662                                              parent_ino, index);
5663         }
5664
5665         /* Nothing to clean up yet */
5666         if (ret)
5667                 return ret;
5668
5669         ret = btrfs_insert_dir_item(trans, root, name, name_len,
5670                                     parent_inode, &key,
5671                                     btrfs_inode_type(inode), index);
5672         if (ret == -EEXIST || ret == -EOVERFLOW)
5673                 goto fail_dir_item;
5674         else if (ret) {
5675                 btrfs_abort_transaction(trans, root, ret);
5676                 return ret;
5677         }
5678
5679         btrfs_i_size_write(parent_inode, parent_inode->i_size +
5680                            name_len * 2);
5681         inode_inc_iversion(parent_inode);
5682         parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
5683         ret = btrfs_update_inode(trans, root, parent_inode);
5684         if (ret)
5685                 btrfs_abort_transaction(trans, root, ret);
5686         return ret;
5687
5688 fail_dir_item:
5689         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5690                 u64 local_index;
5691                 int err;
5692                 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
5693                                  key.objectid, root->root_key.objectid,
5694                                  parent_ino, &local_index, name, name_len);
5695
5696         } else if (add_backref) {
5697                 u64 local_index;
5698                 int err;
5699
5700                 err = btrfs_del_inode_ref(trans, root, name, name_len,
5701                                           ino, parent_ino, &local_index);
5702         }
5703         return ret;
5704 }
5705
5706 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
5707                             struct inode *dir, struct dentry *dentry,
5708                             struct inode *inode, int backref, u64 index)
5709 {
5710         int err = btrfs_add_link(trans, dir, inode,
5711                                  dentry->d_name.name, dentry->d_name.len,
5712                                  backref, index);
5713         if (err > 0)
5714                 err = -EEXIST;
5715         return err;
5716 }
5717
5718 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
5719                         umode_t mode, dev_t rdev)
5720 {
5721         struct btrfs_trans_handle *trans;
5722         struct btrfs_root *root = BTRFS_I(dir)->root;
5723         struct inode *inode = NULL;
5724         int err;
5725         int drop_inode = 0;
5726         u64 objectid;
5727         u64 index = 0;
5728
5729         if (!new_valid_dev(rdev))
5730                 return -EINVAL;
5731
5732         /*
5733          * 2 for inode item and ref
5734          * 2 for dir items
5735          * 1 for xattr if selinux is on
5736          */
5737         trans = btrfs_start_transaction(root, 5);
5738         if (IS_ERR(trans))
5739                 return PTR_ERR(trans);
5740
5741         err = btrfs_find_free_ino(root, &objectid);
5742         if (err)
5743                 goto out_unlock;
5744
5745         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5746                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5747                                 mode, &index);
5748         if (IS_ERR(inode)) {
5749                 err = PTR_ERR(inode);
5750                 goto out_unlock;
5751         }
5752
5753         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5754         if (err) {
5755                 drop_inode = 1;
5756                 goto out_unlock;
5757         }
5758
5759         /*
5760         * If the active LSM wants to access the inode during
5761         * d_instantiate it needs these. Smack checks to see
5762         * if the filesystem supports xattrs by looking at the
5763         * ops vector.
5764         */
5765
5766         inode->i_op = &btrfs_special_inode_operations;
5767         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5768         if (err)
5769                 drop_inode = 1;
5770         else {
5771                 init_special_inode(inode, inode->i_mode, rdev);
5772                 btrfs_update_inode(trans, root, inode);
5773                 d_instantiate(dentry, inode);
5774         }
5775 out_unlock:
5776         btrfs_end_transaction(trans, root);
5777         btrfs_btree_balance_dirty(root);
5778         if (drop_inode) {
5779                 inode_dec_link_count(inode);
5780                 iput(inode);
5781         }
5782         return err;
5783 }
5784
5785 static int btrfs_create(struct inode *dir, struct dentry *dentry,
5786                         umode_t mode, bool excl)
5787 {
5788         struct btrfs_trans_handle *trans;
5789         struct btrfs_root *root = BTRFS_I(dir)->root;
5790         struct inode *inode = NULL;
5791         int drop_inode_on_err = 0;
5792         int err;
5793         u64 objectid;
5794         u64 index = 0;
5795
5796         /*
5797          * 2 for inode item and ref
5798          * 2 for dir items
5799          * 1 for xattr if selinux is on
5800          */
5801         trans = btrfs_start_transaction(root, 5);
5802         if (IS_ERR(trans))
5803                 return PTR_ERR(trans);
5804
5805         err = btrfs_find_free_ino(root, &objectid);
5806         if (err)
5807                 goto out_unlock;
5808
5809         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5810                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5811                                 mode, &index);
5812         if (IS_ERR(inode)) {
5813                 err = PTR_ERR(inode);
5814                 goto out_unlock;
5815         }
5816         drop_inode_on_err = 1;
5817
5818         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5819         if (err)
5820                 goto out_unlock;
5821
5822         err = btrfs_update_inode(trans, root, inode);
5823         if (err)
5824                 goto out_unlock;
5825
5826         /*
5827         * If the active LSM wants to access the inode during
5828         * d_instantiate it needs these. Smack checks to see
5829         * if the filesystem supports xattrs by looking at the
5830         * ops vector.
5831         */
5832         inode->i_fop = &btrfs_file_operations;
5833         inode->i_op = &btrfs_file_inode_operations;
5834
5835         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5836         if (err)
5837                 goto out_unlock;
5838
5839         inode->i_mapping->a_ops = &btrfs_aops;
5840         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5841         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5842         d_instantiate(dentry, inode);
5843
5844 out_unlock:
5845         btrfs_end_transaction(trans, root);
5846         if (err && drop_inode_on_err) {
5847                 inode_dec_link_count(inode);
5848                 iput(inode);
5849         }
5850         btrfs_btree_balance_dirty(root);
5851         return err;
5852 }
5853
5854 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5855                       struct dentry *dentry)
5856 {
5857         struct btrfs_trans_handle *trans;
5858         struct btrfs_root *root = BTRFS_I(dir)->root;
5859         struct inode *inode = old_dentry->d_inode;
5860         u64 index;
5861         int err;
5862         int drop_inode = 0;
5863
5864         /* do not allow sys_link's with other subvols of the same device */
5865         if (root->objectid != BTRFS_I(inode)->root->objectid)
5866                 return -EXDEV;
5867
5868         if (inode->i_nlink >= BTRFS_LINK_MAX)
5869                 return -EMLINK;
5870
5871         err = btrfs_set_inode_index(dir, &index);
5872         if (err)
5873                 goto fail;
5874
5875         /*
5876          * 2 items for inode and inode ref
5877          * 2 items for dir items
5878          * 1 item for parent inode
5879          */
5880         trans = btrfs_start_transaction(root, 5);
5881         if (IS_ERR(trans)) {
5882                 err = PTR_ERR(trans);
5883                 goto fail;
5884         }
5885
5886         btrfs_inc_nlink(inode);
5887         inode_inc_iversion(inode);
5888         inode->i_ctime = CURRENT_TIME;
5889         ihold(inode);
5890         set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
5891
5892         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5893
5894         if (err) {
5895                 drop_inode = 1;
5896         } else {
5897                 struct dentry *parent = dentry->d_parent;
5898                 err = btrfs_update_inode(trans, root, inode);
5899                 if (err)
5900                         goto fail;
5901                 d_instantiate(dentry, inode);
5902                 btrfs_log_new_name(trans, inode, NULL, parent);
5903         }
5904
5905         btrfs_end_transaction(trans, root);
5906 fail:
5907         if (drop_inode) {
5908                 inode_dec_link_count(inode);
5909                 iput(inode);
5910         }
5911         btrfs_btree_balance_dirty(root);
5912         return err;
5913 }
5914
5915 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5916 {
5917         struct inode *inode = NULL;
5918         struct btrfs_trans_handle *trans;
5919         struct btrfs_root *root = BTRFS_I(dir)->root;
5920         int err = 0;
5921         int drop_on_err = 0;
5922         u64 objectid = 0;
5923         u64 index = 0;
5924
5925         /*
5926          * 2 items for inode and ref
5927          * 2 items for dir items
5928          * 1 for xattr if selinux is on
5929          */
5930         trans = btrfs_start_transaction(root, 5);
5931         if (IS_ERR(trans))
5932                 return PTR_ERR(trans);
5933
5934         err = btrfs_find_free_ino(root, &objectid);
5935         if (err)
5936                 goto out_fail;
5937
5938         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5939                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5940                                 S_IFDIR | mode, &index);
5941         if (IS_ERR(inode)) {
5942                 err = PTR_ERR(inode);
5943                 goto out_fail;
5944         }
5945
5946         drop_on_err = 1;
5947
5948         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5949         if (err)
5950                 goto out_fail;
5951
5952         inode->i_op = &btrfs_dir_inode_operations;
5953         inode->i_fop = &btrfs_dir_file_operations;
5954
5955         btrfs_i_size_write(inode, 0);
5956         err = btrfs_update_inode(trans, root, inode);
5957         if (err)
5958                 goto out_fail;
5959
5960         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5961                              dentry->d_name.len, 0, index);
5962         if (err)
5963                 goto out_fail;
5964
5965         d_instantiate(dentry, inode);
5966         drop_on_err = 0;
5967
5968 out_fail:
5969         btrfs_end_transaction(trans, root);
5970         if (drop_on_err)
5971                 iput(inode);
5972         btrfs_btree_balance_dirty(root);
5973         return err;
5974 }
5975
5976 /* helper for btfs_get_extent.  Given an existing extent in the tree,
5977  * and an extent that you want to insert, deal with overlap and insert
5978  * the new extent into the tree.
5979  */
5980 static int merge_extent_mapping(struct extent_map_tree *em_tree,
5981                                 struct extent_map *existing,
5982                                 struct extent_map *em,
5983                                 u64 map_start, u64 map_len)
5984 {
5985         u64 start_diff;
5986
5987         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5988         start_diff = map_start - em->start;
5989         em->start = map_start;
5990         em->len = map_len;
5991         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5992             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
5993                 em->block_start += start_diff;
5994                 em->block_len -= start_diff;
5995         }
5996         return add_extent_mapping(em_tree, em, 0);
5997 }
5998
5999 static noinline int uncompress_inline(struct btrfs_path *path,
6000                                       struct inode *inode, struct page *page,
6001                                       size_t pg_offset, u64 extent_offset,
6002                                       struct btrfs_file_extent_item *item)
6003 {
6004         int ret;
6005         struct extent_buffer *leaf = path->nodes[0];
6006         char *tmp;
6007         size_t max_size;
6008         unsigned long inline_size;
6009         unsigned long ptr;
6010         int compress_type;
6011
6012         WARN_ON(pg_offset != 0);
6013         compress_type = btrfs_file_extent_compression(leaf, item);
6014         max_size = btrfs_file_extent_ram_bytes(leaf, item);
6015         inline_size = btrfs_file_extent_inline_item_len(leaf,
6016                                         btrfs_item_nr(leaf, path->slots[0]));
6017         tmp = kmalloc(inline_size, GFP_NOFS);
6018         if (!tmp)
6019                 return -ENOMEM;
6020         ptr = btrfs_file_extent_inline_start(item);
6021
6022         read_extent_buffer(leaf, tmp, ptr, inline_size);
6023
6024         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
6025         ret = btrfs_decompress(compress_type, tmp, page,
6026                                extent_offset, inline_size, max_size);
6027         if (ret) {
6028                 char *kaddr = kmap_atomic(page);
6029                 unsigned long copy_size = min_t(u64,
6030                                   PAGE_CACHE_SIZE - pg_offset,
6031                                   max_size - extent_offset);
6032                 memset(kaddr + pg_offset, 0, copy_size);
6033                 kunmap_atomic(kaddr);
6034         }
6035         kfree(tmp);
6036         return 0;
6037 }
6038
6039 /*
6040  * a bit scary, this does extent mapping from logical file offset to the disk.
6041  * the ugly parts come from merging extents from the disk with the in-ram
6042  * representation.  This gets more complex because of the data=ordered code,
6043  * where the in-ram extents might be locked pending data=ordered completion.
6044  *
6045  * This also copies inline extents directly into the page.
6046  */
6047
6048 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6049                                     size_t pg_offset, u64 start, u64 len,
6050                                     int create)
6051 {
6052         int ret;
6053         int err = 0;
6054         u64 bytenr;
6055         u64 extent_start = 0;
6056         u64 extent_end = 0;
6057         u64 objectid = btrfs_ino(inode);
6058         u32 found_type;
6059         struct btrfs_path *path = NULL;
6060         struct btrfs_root *root = BTRFS_I(inode)->root;
6061         struct btrfs_file_extent_item *item;
6062         struct extent_buffer *leaf;
6063         struct btrfs_key found_key;
6064         struct extent_map *em = NULL;
6065         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6066         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6067         struct btrfs_trans_handle *trans = NULL;
6068         int compress_type;
6069
6070 again:
6071         read_lock(&em_tree->lock);
6072         em = lookup_extent_mapping(em_tree, start, len);
6073         if (em)
6074                 em->bdev = root->fs_info->fs_devices->latest_bdev;
6075         read_unlock(&em_tree->lock);
6076
6077         if (em) {
6078                 if (em->start > start || em->start + em->len <= start)
6079                         free_extent_map(em);
6080                 else if (em->block_start == EXTENT_MAP_INLINE && page)
6081                         free_extent_map(em);
6082                 else
6083                         goto out;
6084         }
6085         em = alloc_extent_map();
6086         if (!em) {
6087                 err = -ENOMEM;
6088                 goto out;
6089         }
6090         em->bdev = root->fs_info->fs_devices->latest_bdev;
6091         em->start = EXTENT_MAP_HOLE;
6092         em->orig_start = EXTENT_MAP_HOLE;
6093         em->len = (u64)-1;
6094         em->block_len = (u64)-1;
6095
6096         if (!path) {
6097                 path = btrfs_alloc_path();
6098                 if (!path) {
6099                         err = -ENOMEM;
6100                         goto out;
6101                 }
6102                 /*
6103                  * Chances are we'll be called again, so go ahead and do
6104                  * readahead
6105                  */
6106                 path->reada = 1;
6107         }
6108
6109         ret = btrfs_lookup_file_extent(trans, root, path,
6110                                        objectid, start, trans != NULL);
6111         if (ret < 0) {
6112                 err = ret;
6113                 goto out;
6114         }
6115
6116         if (ret != 0) {
6117                 if (path->slots[0] == 0)
6118                         goto not_found;
6119                 path->slots[0]--;
6120         }
6121
6122         leaf = path->nodes[0];
6123         item = btrfs_item_ptr(leaf, path->slots[0],
6124                               struct btrfs_file_extent_item);
6125         /* are we inside the extent that was found? */
6126         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6127         found_type = btrfs_key_type(&found_key);
6128         if (found_key.objectid != objectid ||
6129             found_type != BTRFS_EXTENT_DATA_KEY) {
6130                 goto not_found;
6131         }
6132
6133         found_type = btrfs_file_extent_type(leaf, item);
6134         extent_start = found_key.offset;
6135         compress_type = btrfs_file_extent_compression(leaf, item);
6136         if (found_type == BTRFS_FILE_EXTENT_REG ||
6137             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6138                 extent_end = extent_start +
6139                        btrfs_file_extent_num_bytes(leaf, item);
6140         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6141                 size_t size;
6142                 size = btrfs_file_extent_inline_len(leaf, item);
6143                 extent_end = ALIGN(extent_start + size, root->sectorsize);
6144         }
6145
6146         if (start >= extent_end) {
6147                 path->slots[0]++;
6148                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6149                         ret = btrfs_next_leaf(root, path);
6150                         if (ret < 0) {
6151                                 err = ret;
6152                                 goto out;
6153                         }
6154                         if (ret > 0)
6155                                 goto not_found;
6156                         leaf = path->nodes[0];
6157                 }
6158                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6159                 if (found_key.objectid != objectid ||
6160                     found_key.type != BTRFS_EXTENT_DATA_KEY)
6161                         goto not_found;
6162                 if (start + len <= found_key.offset)
6163                         goto not_found;
6164                 em->start = start;
6165                 em->orig_start = start;
6166                 em->len = found_key.offset - start;
6167                 goto not_found_em;
6168         }
6169
6170         em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
6171         if (found_type == BTRFS_FILE_EXTENT_REG ||
6172             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6173                 em->start = extent_start;
6174                 em->len = extent_end - extent_start;
6175                 em->orig_start = extent_start -
6176                                  btrfs_file_extent_offset(leaf, item);
6177                 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf,
6178                                                                       item);
6179                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
6180                 if (bytenr == 0) {
6181                         em->block_start = EXTENT_MAP_HOLE;
6182                         goto insert;
6183                 }
6184                 if (compress_type != BTRFS_COMPRESS_NONE) {
6185                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6186                         em->compress_type = compress_type;
6187                         em->block_start = bytenr;
6188                         em->block_len = em->orig_block_len;
6189                 } else {
6190                         bytenr += btrfs_file_extent_offset(leaf, item);
6191                         em->block_start = bytenr;
6192                         em->block_len = em->len;
6193                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
6194                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6195                 }
6196                 goto insert;
6197         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6198                 unsigned long ptr;
6199                 char *map;
6200                 size_t size;
6201                 size_t extent_offset;
6202                 size_t copy_size;
6203
6204                 em->block_start = EXTENT_MAP_INLINE;
6205                 if (!page || create) {
6206                         em->start = extent_start;
6207                         em->len = extent_end - extent_start;
6208                         goto out;
6209                 }
6210
6211                 size = btrfs_file_extent_inline_len(leaf, item);
6212                 extent_offset = page_offset(page) + pg_offset - extent_start;
6213                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6214                                 size - extent_offset);
6215                 em->start = extent_start + extent_offset;
6216                 em->len = ALIGN(copy_size, root->sectorsize);
6217                 em->orig_block_len = em->len;
6218                 em->orig_start = em->start;
6219                 if (compress_type) {
6220                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6221                         em->compress_type = compress_type;
6222                 }
6223                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6224                 if (create == 0 && !PageUptodate(page)) {
6225                         if (btrfs_file_extent_compression(leaf, item) !=
6226                             BTRFS_COMPRESS_NONE) {
6227                                 ret = uncompress_inline(path, inode, page,
6228                                                         pg_offset,
6229                                                         extent_offset, item);
6230                                 BUG_ON(ret); /* -ENOMEM */
6231                         } else {
6232                                 map = kmap(page);
6233                                 read_extent_buffer(leaf, map + pg_offset, ptr,
6234                                                    copy_size);
6235                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6236                                         memset(map + pg_offset + copy_size, 0,
6237                                                PAGE_CACHE_SIZE - pg_offset -
6238                                                copy_size);
6239                                 }
6240                                 kunmap(page);
6241                         }
6242                         flush_dcache_page(page);
6243                 } else if (create && PageUptodate(page)) {
6244                         BUG();
6245                         if (!trans) {
6246                                 kunmap(page);
6247                                 free_extent_map(em);
6248                                 em = NULL;
6249
6250                                 btrfs_release_path(path);
6251                                 trans = btrfs_join_transaction(root);
6252
6253                                 if (IS_ERR(trans))
6254                                         return ERR_CAST(trans);
6255                                 goto again;
6256                         }
6257                         map = kmap(page);
6258                         write_extent_buffer(leaf, map + pg_offset, ptr,
6259                                             copy_size);
6260                         kunmap(page);
6261                         btrfs_mark_buffer_dirty(leaf);
6262                 }
6263                 set_extent_uptodate(io_tree, em->start,
6264                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
6265                 goto insert;
6266         } else {
6267                 WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
6268         }
6269 not_found:
6270         em->start = start;
6271         em->orig_start = start;
6272         em->len = len;
6273 not_found_em:
6274         em->block_start = EXTENT_MAP_HOLE;
6275         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6276 insert:
6277         btrfs_release_path(path);
6278         if (em->start > start || extent_map_end(em) <= start) {
6279                 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6280                         (unsigned long long)em->start,
6281                         (unsigned long long)em->len,
6282                         (unsigned long long)start,
6283                         (unsigned long long)len);
6284                 err = -EIO;
6285                 goto out;
6286         }
6287
6288         err = 0;
6289         write_lock(&em_tree->lock);
6290         ret = add_extent_mapping(em_tree, em, 0);
6291         /* it is possible that someone inserted the extent into the tree
6292          * while we had the lock dropped.  It is also possible that
6293          * an overlapping map exists in the tree
6294          */
6295         if (ret == -EEXIST) {
6296                 struct extent_map *existing;
6297
6298                 ret = 0;
6299
6300                 existing = lookup_extent_mapping(em_tree, start, len);
6301                 if (existing && (existing->start > start ||
6302                     existing->start + existing->len <= start)) {
6303                         free_extent_map(existing);
6304                         existing = NULL;
6305                 }
6306                 if (!existing) {
6307                         existing = lookup_extent_mapping(em_tree, em->start,
6308                                                          em->len);
6309                         if (existing) {
6310                                 err = merge_extent_mapping(em_tree, existing,
6311                                                            em, start,
6312                                                            root->sectorsize);
6313                                 free_extent_map(existing);
6314                                 if (err) {
6315                                         free_extent_map(em);
6316                                         em = NULL;
6317                                 }
6318                         } else {
6319                                 err = -EIO;
6320                                 free_extent_map(em);
6321                                 em = NULL;
6322                         }
6323                 } else {
6324                         free_extent_map(em);
6325                         em = existing;
6326                         err = 0;
6327                 }
6328         }
6329         write_unlock(&em_tree->lock);
6330 out:
6331
6332         if (em)
6333                 trace_btrfs_get_extent(root, em);
6334
6335         if (path)
6336                 btrfs_free_path(path);
6337         if (trans) {
6338                 ret = btrfs_end_transaction(trans, root);
6339                 if (!err)
6340                         err = ret;
6341         }
6342         if (err) {
6343                 free_extent_map(em);
6344                 return ERR_PTR(err);
6345         }
6346         BUG_ON(!em); /* Error is always set */
6347         return em;
6348 }
6349
6350 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6351                                            size_t pg_offset, u64 start, u64 len,
6352                                            int create)
6353 {
6354         struct extent_map *em;
6355         struct extent_map *hole_em = NULL;
6356         u64 range_start = start;
6357         u64 end;
6358         u64 found;
6359         u64 found_end;
6360         int err = 0;
6361
6362         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6363         if (IS_ERR(em))
6364                 return em;
6365         if (em) {
6366                 /*
6367                  * if our em maps to
6368                  * -  a hole or
6369                  * -  a pre-alloc extent,
6370                  * there might actually be delalloc bytes behind it.
6371                  */
6372                 if (em->block_start != EXTENT_MAP_HOLE &&
6373                     !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6374                         return em;
6375                 else
6376                         hole_em = em;
6377         }
6378
6379         /* check to see if we've wrapped (len == -1 or similar) */
6380         end = start + len;
6381         if (end < start)
6382                 end = (u64)-1;
6383         else
6384                 end -= 1;
6385
6386         em = NULL;
6387
6388         /* ok, we didn't find anything, lets look for delalloc */
6389         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6390                                  end, len, EXTENT_DELALLOC, 1);
6391         found_end = range_start + found;
6392         if (found_end < range_start)
6393                 found_end = (u64)-1;
6394
6395         /*
6396          * we didn't find anything useful, return
6397          * the original results from get_extent()
6398          */
6399         if (range_start > end || found_end <= start) {
6400                 em = hole_em;
6401                 hole_em = NULL;
6402                 goto out;
6403         }
6404
6405         /* adjust the range_start to make sure it doesn't
6406          * go backwards from the start they passed in
6407          */
6408         range_start = max(start,range_start);
6409         found = found_end - range_start;
6410
6411         if (found > 0) {
6412                 u64 hole_start = start;
6413                 u64 hole_len = len;
6414
6415                 em = alloc_extent_map();
6416                 if (!em) {
6417                         err = -ENOMEM;
6418                         goto out;
6419                 }
6420                 /*
6421                  * when btrfs_get_extent can't find anything it
6422                  * returns one huge hole
6423                  *
6424                  * make sure what it found really fits our range, and
6425                  * adjust to make sure it is based on the start from
6426                  * the caller
6427                  */
6428                 if (hole_em) {
6429                         u64 calc_end = extent_map_end(hole_em);
6430
6431                         if (calc_end <= start || (hole_em->start > end)) {
6432                                 free_extent_map(hole_em);
6433                                 hole_em = NULL;
6434                         } else {
6435                                 hole_start = max(hole_em->start, start);
6436                                 hole_len = calc_end - hole_start;
6437                         }
6438                 }
6439                 em->bdev = NULL;
6440                 if (hole_em && range_start > hole_start) {
6441                         /* our hole starts before our delalloc, so we
6442                          * have to return just the parts of the hole
6443                          * that go until  the delalloc starts
6444                          */
6445                         em->len = min(hole_len,
6446                                       range_start - hole_start);
6447                         em->start = hole_start;
6448                         em->orig_start = hole_start;
6449                         /*
6450                          * don't adjust block start at all,
6451                          * it is fixed at EXTENT_MAP_HOLE
6452                          */
6453                         em->block_start = hole_em->block_start;
6454                         em->block_len = hole_len;
6455                         if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6456                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6457                 } else {
6458                         em->start = range_start;
6459                         em->len = found;
6460                         em->orig_start = range_start;
6461                         em->block_start = EXTENT_MAP_DELALLOC;
6462                         em->block_len = found;
6463                 }
6464         } else if (hole_em) {
6465                 return hole_em;
6466         }
6467 out:
6468
6469         free_extent_map(hole_em);
6470         if (err) {
6471                 free_extent_map(em);
6472                 return ERR_PTR(err);
6473         }
6474         return em;
6475 }
6476
6477 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6478                                                   u64 start, u64 len)
6479 {
6480         struct btrfs_root *root = BTRFS_I(inode)->root;
6481         struct btrfs_trans_handle *trans;
6482         struct extent_map *em;
6483         struct btrfs_key ins;
6484         u64 alloc_hint;
6485         int ret;
6486
6487         trans = btrfs_join_transaction(root);
6488         if (IS_ERR(trans))
6489                 return ERR_CAST(trans);
6490
6491         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
6492
6493         alloc_hint = get_extent_allocation_hint(inode, start, len);
6494         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
6495                                    alloc_hint, &ins, 1);
6496         if (ret) {
6497                 em = ERR_PTR(ret);
6498                 goto out;
6499         }
6500
6501         em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
6502                               ins.offset, ins.offset, ins.offset, 0);
6503         if (IS_ERR(em))
6504                 goto out;
6505
6506         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
6507                                            ins.offset, ins.offset, 0);
6508         if (ret) {
6509                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
6510                 em = ERR_PTR(ret);
6511         }
6512 out:
6513         btrfs_end_transaction(trans, root);
6514         return em;
6515 }
6516
6517 /*
6518  * returns 1 when the nocow is safe, < 1 on error, 0 if the
6519  * block must be cow'd
6520  */
6521 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
6522                                       struct inode *inode, u64 offset, u64 *len,
6523                                       u64 *orig_start, u64 *orig_block_len,
6524                                       u64 *ram_bytes)
6525 {
6526         struct btrfs_path *path;
6527         int ret;
6528         struct extent_buffer *leaf;
6529         struct btrfs_root *root = BTRFS_I(inode)->root;
6530         struct btrfs_file_extent_item *fi;
6531         struct btrfs_key key;
6532         u64 disk_bytenr;
6533         u64 backref_offset;
6534         u64 extent_end;
6535         u64 num_bytes;
6536         int slot;
6537         int found_type;
6538
6539         path = btrfs_alloc_path();
6540         if (!path)
6541                 return -ENOMEM;
6542
6543         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
6544                                        offset, 0);
6545         if (ret < 0)
6546                 goto out;
6547
6548         slot = path->slots[0];
6549         if (ret == 1) {
6550                 if (slot == 0) {
6551                         /* can't find the item, must cow */
6552                         ret = 0;
6553                         goto out;
6554                 }
6555                 slot--;
6556         }
6557         ret = 0;
6558         leaf = path->nodes[0];
6559         btrfs_item_key_to_cpu(leaf, &key, slot);
6560         if (key.objectid != btrfs_ino(inode) ||
6561             key.type != BTRFS_EXTENT_DATA_KEY) {
6562                 /* not our file or wrong item type, must cow */
6563                 goto out;
6564         }
6565
6566         if (key.offset > offset) {
6567                 /* Wrong offset, must cow */
6568                 goto out;
6569         }
6570
6571         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6572         found_type = btrfs_file_extent_type(leaf, fi);
6573         if (found_type != BTRFS_FILE_EXTENT_REG &&
6574             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6575                 /* not a regular extent, must cow */
6576                 goto out;
6577         }
6578         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6579         backref_offset = btrfs_file_extent_offset(leaf, fi);
6580
6581         *orig_start = key.offset - backref_offset;
6582         *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
6583         *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6584
6585         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
6586         if (extent_end < offset + *len) {
6587                 /* extent doesn't include our full range, must cow */
6588                 goto out;
6589         }
6590
6591         if (btrfs_extent_readonly(root, disk_bytenr))
6592                 goto out;
6593
6594         /*
6595          * look for other files referencing this extent, if we
6596          * find any we must cow
6597          */
6598         if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
6599                                   key.offset - backref_offset, disk_bytenr))
6600                 goto out;
6601
6602         /*
6603          * adjust disk_bytenr and num_bytes to cover just the bytes
6604          * in this extent we are about to write.  If there
6605          * are any csums in that range we have to cow in order
6606          * to keep the csums correct
6607          */
6608         disk_bytenr += backref_offset;
6609         disk_bytenr += offset - key.offset;
6610         num_bytes = min(offset + *len, extent_end) - offset;
6611         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
6612                                 goto out;
6613         /*
6614          * all of the above have passed, it is safe to overwrite this extent
6615          * without cow
6616          */
6617         *len = num_bytes;
6618         ret = 1;
6619 out:
6620         btrfs_free_path(path);
6621         return ret;
6622 }
6623
6624 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
6625                               struct extent_state **cached_state, int writing)
6626 {
6627         struct btrfs_ordered_extent *ordered;
6628         int ret = 0;
6629
6630         while (1) {
6631                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6632                                  0, cached_state);
6633                 /*
6634                  * We're concerned with the entire range that we're going to be
6635                  * doing DIO to, so we need to make sure theres no ordered
6636                  * extents in this range.
6637                  */
6638                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6639                                                      lockend - lockstart + 1);
6640
6641                 /*
6642                  * We need to make sure there are no buffered pages in this
6643                  * range either, we could have raced between the invalidate in
6644                  * generic_file_direct_write and locking the extent.  The
6645                  * invalidate needs to happen so that reads after a write do not
6646                  * get stale data.
6647                  */
6648                 if (!ordered && (!writing ||
6649                     !test_range_bit(&BTRFS_I(inode)->io_tree,
6650                                     lockstart, lockend, EXTENT_UPTODATE, 0,
6651                                     *cached_state)))
6652                         break;
6653
6654                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6655                                      cached_state, GFP_NOFS);
6656
6657                 if (ordered) {
6658                         btrfs_start_ordered_extent(inode, ordered, 1);
6659                         btrfs_put_ordered_extent(ordered);
6660                 } else {
6661                         /* Screw you mmap */
6662                         ret = filemap_write_and_wait_range(inode->i_mapping,
6663                                                            lockstart,
6664                                                            lockend);
6665                         if (ret)
6666                                 break;
6667
6668                         /*
6669                          * If we found a page that couldn't be invalidated just
6670                          * fall back to buffered.
6671                          */
6672                         ret = invalidate_inode_pages2_range(inode->i_mapping,
6673                                         lockstart >> PAGE_CACHE_SHIFT,
6674                                         lockend >> PAGE_CACHE_SHIFT);
6675                         if (ret)
6676                                 break;
6677                 }
6678
6679                 cond_resched();
6680         }
6681
6682         return ret;
6683 }
6684
6685 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
6686                                            u64 len, u64 orig_start,
6687                                            u64 block_start, u64 block_len,
6688                                            u64 orig_block_len, u64 ram_bytes,
6689                                            int type)
6690 {
6691         struct extent_map_tree *em_tree;
6692         struct extent_map *em;
6693         struct btrfs_root *root = BTRFS_I(inode)->root;
6694         int ret;
6695
6696         em_tree = &BTRFS_I(inode)->extent_tree;
6697         em = alloc_extent_map();
6698         if (!em)
6699                 return ERR_PTR(-ENOMEM);
6700
6701         em->start = start;
6702         em->orig_start = orig_start;
6703         em->mod_start = start;
6704         em->mod_len = len;
6705         em->len = len;
6706         em->block_len = block_len;
6707         em->block_start = block_start;
6708         em->bdev = root->fs_info->fs_devices->latest_bdev;
6709         em->orig_block_len = orig_block_len;
6710         em->ram_bytes = ram_bytes;
6711         em->generation = -1;
6712         set_bit(EXTENT_FLAG_PINNED, &em->flags);
6713         if (type == BTRFS_ORDERED_PREALLOC)
6714                 set_bit(EXTENT_FLAG_FILLING, &em->flags);
6715
6716         do {
6717                 btrfs_drop_extent_cache(inode, em->start,
6718                                 em->start + em->len - 1, 0);
6719                 write_lock(&em_tree->lock);
6720                 ret = add_extent_mapping(em_tree, em, 1);
6721                 write_unlock(&em_tree->lock);
6722         } while (ret == -EEXIST);
6723
6724         if (ret) {
6725                 free_extent_map(em);
6726                 return ERR_PTR(ret);
6727         }
6728
6729         return em;
6730 }
6731
6732
6733 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
6734                                    struct buffer_head *bh_result, int create)
6735 {
6736         struct extent_map *em;
6737         struct btrfs_root *root = BTRFS_I(inode)->root;
6738         struct extent_state *cached_state = NULL;
6739         u64 start = iblock << inode->i_blkbits;
6740         u64 lockstart, lockend;
6741         u64 len = bh_result->b_size;
6742         struct btrfs_trans_handle *trans;
6743         int unlock_bits = EXTENT_LOCKED;
6744         int ret = 0;
6745
6746         if (create)
6747                 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
6748         else
6749                 len = min_t(u64, len, root->sectorsize);
6750
6751         lockstart = start;
6752         lockend = start + len - 1;
6753
6754         /*
6755          * If this errors out it's because we couldn't invalidate pagecache for
6756          * this range and we need to fallback to buffered.
6757          */
6758         if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
6759                 return -ENOTBLK;
6760
6761         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
6762         if (IS_ERR(em)) {
6763                 ret = PTR_ERR(em);
6764                 goto unlock_err;
6765         }
6766
6767         /*
6768          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6769          * io.  INLINE is special, and we could probably kludge it in here, but
6770          * it's still buffered so for safety lets just fall back to the generic
6771          * buffered path.
6772          *
6773          * For COMPRESSED we _have_ to read the entire extent in so we can
6774          * decompress it, so there will be buffering required no matter what we
6775          * do, so go ahead and fallback to buffered.
6776          *
6777          * We return -ENOTBLK because thats what makes DIO go ahead and go back
6778          * to buffered IO.  Don't blame me, this is the price we pay for using
6779          * the generic code.
6780          */
6781         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6782             em->block_start == EXTENT_MAP_INLINE) {
6783                 free_extent_map(em);
6784                 ret = -ENOTBLK;
6785                 goto unlock_err;
6786         }
6787
6788         /* Just a good old fashioned hole, return */
6789         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6790                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6791                 free_extent_map(em);
6792                 goto unlock_err;
6793         }
6794
6795         /*
6796          * We don't allocate a new extent in the following cases
6797          *
6798          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
6799          * existing extent.
6800          * 2) The extent is marked as PREALLOC.  We're good to go here and can
6801          * just use the extent.
6802          *
6803          */
6804         if (!create) {
6805                 len = min(len, em->len - (start - em->start));
6806                 lockstart = start + len;
6807                 goto unlock;
6808         }
6809
6810         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6811             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6812              em->block_start != EXTENT_MAP_HOLE)) {
6813                 int type;
6814                 int ret;
6815                 u64 block_start, orig_start, orig_block_len, ram_bytes;
6816
6817                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6818                         type = BTRFS_ORDERED_PREALLOC;
6819                 else
6820                         type = BTRFS_ORDERED_NOCOW;
6821                 len = min(len, em->len - (start - em->start));
6822                 block_start = em->block_start + (start - em->start);
6823
6824                 /*
6825                  * we're not going to log anything, but we do need
6826                  * to make sure the current transaction stays open
6827                  * while we look for nocow cross refs
6828                  */
6829                 trans = btrfs_join_transaction(root);
6830                 if (IS_ERR(trans))
6831                         goto must_cow;
6832
6833                 if (can_nocow_odirect(trans, inode, start, &len, &orig_start,
6834                                       &orig_block_len, &ram_bytes) == 1) {
6835                         if (type == BTRFS_ORDERED_PREALLOC) {
6836                                 free_extent_map(em);
6837                                 em = create_pinned_em(inode, start, len,
6838                                                        orig_start,
6839                                                        block_start, len,
6840                                                        orig_block_len,
6841                                                        ram_bytes, type);
6842                                 if (IS_ERR(em)) {
6843                                         btrfs_end_transaction(trans, root);
6844                                         goto unlock_err;
6845                                 }
6846                         }
6847
6848                         ret = btrfs_add_ordered_extent_dio(inode, start,
6849                                            block_start, len, len, type);
6850                         btrfs_end_transaction(trans, root);
6851                         if (ret) {
6852                                 free_extent_map(em);
6853                                 goto unlock_err;
6854                         }
6855                         goto unlock;
6856                 }
6857                 btrfs_end_transaction(trans, root);
6858         }
6859 must_cow:
6860         /*
6861          * this will cow the extent, reset the len in case we changed
6862          * it above
6863          */
6864         len = bh_result->b_size;
6865         free_extent_map(em);
6866         em = btrfs_new_extent_direct(inode, start, len);
6867         if (IS_ERR(em)) {
6868                 ret = PTR_ERR(em);
6869                 goto unlock_err;
6870         }
6871         len = min(len, em->len - (start - em->start));
6872 unlock:
6873         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6874                 inode->i_blkbits;
6875         bh_result->b_size = len;
6876         bh_result->b_bdev = em->bdev;
6877         set_buffer_mapped(bh_result);
6878         if (create) {
6879                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6880                         set_buffer_new(bh_result);
6881
6882                 /*
6883                  * Need to update the i_size under the extent lock so buffered
6884                  * readers will get the updated i_size when we unlock.
6885                  */
6886                 if (start + len > i_size_read(inode))
6887                         i_size_write(inode, start + len);
6888
6889                 spin_lock(&BTRFS_I(inode)->lock);
6890                 BTRFS_I(inode)->outstanding_extents++;
6891                 spin_unlock(&BTRFS_I(inode)->lock);
6892
6893                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6894                                      lockstart + len - 1, EXTENT_DELALLOC, NULL,
6895                                      &cached_state, GFP_NOFS);
6896                 BUG_ON(ret);
6897         }
6898
6899         /*
6900          * In the case of write we need to clear and unlock the entire range,
6901          * in the case of read we need to unlock only the end area that we
6902          * aren't using if there is any left over space.
6903          */
6904         if (lockstart < lockend) {
6905                 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6906                                  lockend, unlock_bits, 1, 0,
6907                                  &cached_state, GFP_NOFS);
6908         } else {
6909                 free_extent_state(cached_state);
6910         }
6911
6912         free_extent_map(em);
6913
6914         return 0;
6915
6916 unlock_err:
6917         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6918                          unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6919         return ret;
6920 }
6921
6922 struct btrfs_dio_private {
6923         struct inode *inode;
6924         u64 logical_offset;
6925         u64 disk_bytenr;
6926         u64 bytes;
6927         void *private;
6928
6929         /* number of bios pending for this dio */
6930         atomic_t pending_bios;
6931
6932         /* IO errors */
6933         int errors;
6934
6935         /* orig_bio is our btrfs_io_bio */
6936         struct bio *orig_bio;
6937
6938         /* dio_bio came from fs/direct-io.c */
6939         struct bio *dio_bio;
6940 };
6941
6942 static void btrfs_endio_direct_read(struct bio *bio, int err)
6943 {
6944         struct btrfs_dio_private *dip = bio->bi_private;
6945         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6946         struct bio_vec *bvec = bio->bi_io_vec;
6947         struct inode *inode = dip->inode;
6948         struct btrfs_root *root = BTRFS_I(inode)->root;
6949         struct bio *dio_bio;
6950         u64 start;
6951
6952         start = dip->logical_offset;
6953         do {
6954                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6955                         struct page *page = bvec->bv_page;
6956                         char *kaddr;
6957                         u32 csum = ~(u32)0;
6958                         u64 private = ~(u32)0;
6959                         unsigned long flags;
6960
6961                         if (get_state_private(&BTRFS_I(inode)->io_tree,
6962                                               start, &private))
6963                                 goto failed;
6964                         local_irq_save(flags);
6965                         kaddr = kmap_atomic(page);
6966                         csum = btrfs_csum_data(kaddr + bvec->bv_offset,
6967                                                csum, bvec->bv_len);
6968                         btrfs_csum_final(csum, (char *)&csum);
6969                         kunmap_atomic(kaddr);
6970                         local_irq_restore(flags);
6971
6972                         flush_dcache_page(bvec->bv_page);
6973                         if (csum != private) {
6974 failed:
6975                                 btrfs_err(root->fs_info, "csum failed ino %llu off %llu csum %u private %u",
6976                                         (unsigned long long)btrfs_ino(inode),
6977                                         (unsigned long long)start,
6978                                         csum, (unsigned)private);
6979                                 err = -EIO;
6980                         }
6981                 }
6982
6983                 start += bvec->bv_len;
6984                 bvec++;
6985         } while (bvec <= bvec_end);
6986
6987         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
6988                       dip->logical_offset + dip->bytes - 1);
6989         dio_bio = dip->dio_bio;
6990
6991         kfree(dip);
6992
6993         /* If we had a csum failure make sure to clear the uptodate flag */
6994         if (err)
6995                 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
6996         dio_end_io(dio_bio, err);
6997         bio_put(bio);
6998 }
6999
7000 static void btrfs_endio_direct_write(struct bio *bio, int err)
7001 {
7002         struct btrfs_dio_private *dip = bio->bi_private;
7003         struct inode *inode = dip->inode;
7004         struct btrfs_root *root = BTRFS_I(inode)->root;
7005         struct btrfs_ordered_extent *ordered = NULL;
7006         u64 ordered_offset = dip->logical_offset;
7007         u64 ordered_bytes = dip->bytes;
7008         struct bio *dio_bio;
7009         int ret;
7010
7011         if (err)
7012                 goto out_done;
7013 again:
7014         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
7015                                                    &ordered_offset,
7016                                                    ordered_bytes, !err);
7017         if (!ret)
7018                 goto out_test;
7019
7020         ordered->work.func = finish_ordered_fn;
7021         ordered->work.flags = 0;
7022         btrfs_queue_worker(&root->fs_info->endio_write_workers,
7023                            &ordered->work);
7024 out_test:
7025         /*
7026          * our bio might span multiple ordered extents.  If we haven't
7027          * completed the accounting for the whole dio, go back and try again
7028          */
7029         if (ordered_offset < dip->logical_offset + dip->bytes) {
7030                 ordered_bytes = dip->logical_offset + dip->bytes -
7031                         ordered_offset;
7032                 ordered = NULL;
7033                 goto again;
7034         }
7035 out_done:
7036         dio_bio = dip->dio_bio;
7037
7038         kfree(dip);
7039
7040         /* If we had an error make sure to clear the uptodate flag */
7041         if (err)
7042                 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7043         dio_end_io(dio_bio, err);
7044         bio_put(bio);
7045 }
7046
7047 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
7048                                     struct bio *bio, int mirror_num,
7049                                     unsigned long bio_flags, u64 offset)
7050 {
7051         int ret;
7052         struct btrfs_root *root = BTRFS_I(inode)->root;
7053         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
7054         BUG_ON(ret); /* -ENOMEM */
7055         return 0;
7056 }
7057
7058 static void btrfs_end_dio_bio(struct bio *bio, int err)
7059 {
7060         struct btrfs_dio_private *dip = bio->bi_private;
7061
7062         if (err) {
7063                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
7064                       "sector %#Lx len %u err no %d\n",
7065                       (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
7066                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
7067                 dip->errors = 1;
7068
7069                 /*
7070                  * before atomic variable goto zero, we must make sure
7071                  * dip->errors is perceived to be set.
7072                  */
7073                 smp_mb__before_atomic_dec();
7074         }
7075
7076         /* if there are more bios still pending for this dio, just exit */
7077         if (!atomic_dec_and_test(&dip->pending_bios))
7078                 goto out;
7079
7080         if (dip->errors) {
7081                 bio_io_error(dip->orig_bio);
7082         } else {
7083                 set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags);
7084                 bio_endio(dip->orig_bio, 0);
7085         }
7086 out:
7087         bio_put(bio);
7088 }
7089
7090 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
7091                                        u64 first_sector, gfp_t gfp_flags)
7092 {
7093         int nr_vecs = bio_get_nr_vecs(bdev);
7094         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
7095 }
7096
7097 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
7098                                          int rw, u64 file_offset, int skip_sum,
7099                                          int async_submit)
7100 {
7101         int write = rw & REQ_WRITE;
7102         struct btrfs_root *root = BTRFS_I(inode)->root;
7103         int ret;
7104
7105         if (async_submit)
7106                 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7107
7108         bio_get(bio);
7109
7110         if (!write) {
7111                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
7112                 if (ret)
7113                         goto err;
7114         }
7115
7116         if (skip_sum)
7117                 goto map;
7118
7119         if (write && async_submit) {
7120                 ret = btrfs_wq_submit_bio(root->fs_info,
7121                                    inode, rw, bio, 0, 0,
7122                                    file_offset,
7123                                    __btrfs_submit_bio_start_direct_io,
7124                                    __btrfs_submit_bio_done);
7125                 goto err;
7126         } else if (write) {
7127                 /*
7128                  * If we aren't doing async submit, calculate the csum of the
7129                  * bio now.
7130                  */
7131                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
7132                 if (ret)
7133                         goto err;
7134         } else if (!skip_sum) {
7135                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio, file_offset);
7136                 if (ret)
7137                         goto err;
7138         }
7139
7140 map:
7141         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
7142 err:
7143         bio_put(bio);
7144         return ret;
7145 }
7146
7147 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
7148                                     int skip_sum)
7149 {
7150         struct inode *inode = dip->inode;
7151         struct btrfs_root *root = BTRFS_I(inode)->root;
7152         struct bio *bio;
7153         struct bio *orig_bio = dip->orig_bio;
7154         struct bio_vec *bvec = orig_bio->bi_io_vec;
7155         u64 start_sector = orig_bio->bi_sector;
7156         u64 file_offset = dip->logical_offset;
7157         u64 submit_len = 0;
7158         u64 map_length;
7159         int nr_pages = 0;
7160         int ret = 0;
7161         int async_submit = 0;
7162
7163         map_length = orig_bio->bi_size;
7164         ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
7165                               &map_length, NULL, 0);
7166         if (ret) {
7167                 bio_put(orig_bio);
7168                 return -EIO;
7169         }
7170         if (map_length >= orig_bio->bi_size) {
7171                 bio = orig_bio;
7172                 goto submit;
7173         }
7174
7175         /* async crcs make it difficult to collect full stripe writes. */
7176         if (btrfs_get_alloc_profile(root, 1) &
7177             (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6))
7178                 async_submit = 0;
7179         else
7180                 async_submit = 1;
7181
7182         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
7183         if (!bio)
7184                 return -ENOMEM;
7185         bio->bi_private = dip;
7186         bio->bi_end_io = btrfs_end_dio_bio;
7187         atomic_inc(&dip->pending_bios);
7188
7189         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
7190                 if (unlikely(map_length < submit_len + bvec->bv_len ||
7191                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
7192                                  bvec->bv_offset) < bvec->bv_len)) {
7193                         /*
7194                          * inc the count before we submit the bio so
7195                          * we know the end IO handler won't happen before
7196                          * we inc the count. Otherwise, the dip might get freed
7197                          * before we're done setting it up
7198                          */
7199                         atomic_inc(&dip->pending_bios);
7200                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
7201                                                      file_offset, skip_sum,
7202                                                      async_submit);
7203                         if (ret) {
7204                                 bio_put(bio);
7205                                 atomic_dec(&dip->pending_bios);
7206                                 goto out_err;
7207                         }
7208
7209                         start_sector += submit_len >> 9;
7210                         file_offset += submit_len;
7211
7212                         submit_len = 0;
7213                         nr_pages = 0;
7214
7215                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
7216                                                   start_sector, GFP_NOFS);
7217                         if (!bio)
7218                                 goto out_err;
7219                         bio->bi_private = dip;
7220                         bio->bi_end_io = btrfs_end_dio_bio;
7221
7222                         map_length = orig_bio->bi_size;
7223                         ret = btrfs_map_block(root->fs_info, rw,
7224                                               start_sector << 9,
7225                                               &map_length, NULL, 0);
7226                         if (ret) {
7227                                 bio_put(bio);
7228                                 goto out_err;
7229                         }
7230                 } else {
7231                         submit_len += bvec->bv_len;
7232                         nr_pages ++;
7233                         bvec++;
7234                 }
7235         }
7236
7237 submit:
7238         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
7239                                      async_submit);
7240         if (!ret)
7241                 return 0;
7242
7243         bio_put(bio);
7244 out_err:
7245         dip->errors = 1;
7246         /*
7247          * before atomic variable goto zero, we must
7248          * make sure dip->errors is perceived to be set.
7249          */
7250         smp_mb__before_atomic_dec();
7251         if (atomic_dec_and_test(&dip->pending_bios))
7252                 bio_io_error(dip->orig_bio);
7253
7254         /* bio_end_io() will handle error, so we needn't return it */
7255         return 0;
7256 }
7257
7258 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
7259                                 struct inode *inode, loff_t file_offset)
7260 {
7261         struct btrfs_root *root = BTRFS_I(inode)->root;
7262         struct btrfs_dio_private *dip;
7263         struct bio_vec *bvec = dio_bio->bi_io_vec;
7264         struct bio *io_bio;
7265         int skip_sum;
7266         int write = rw & REQ_WRITE;
7267         int ret = 0;
7268
7269         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7270
7271         io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
7272
7273         if (!io_bio) {
7274                 ret = -ENOMEM;
7275                 goto free_ordered;
7276         }
7277
7278         dip = kmalloc(sizeof(*dip), GFP_NOFS);
7279         if (!dip) {
7280                 ret = -ENOMEM;
7281                 goto free_io_bio;
7282         }
7283
7284         dip->private = dio_bio->bi_private;
7285         io_bio->bi_private = dio_bio->bi_private;
7286         dip->inode = inode;
7287         dip->logical_offset = file_offset;
7288
7289         dip->bytes = 0;
7290         do {
7291                 dip->bytes += bvec->bv_len;
7292                 bvec++;
7293         } while (bvec <= (dio_bio->bi_io_vec + dio_bio->bi_vcnt - 1));
7294
7295         dip->disk_bytenr = (u64)dio_bio->bi_sector << 9;
7296         io_bio->bi_private = dip;
7297         dip->errors = 0;
7298         dip->orig_bio = io_bio;
7299         dip->dio_bio = dio_bio;
7300         atomic_set(&dip->pending_bios, 0);
7301
7302         if (write)
7303                 io_bio->bi_end_io = btrfs_endio_direct_write;
7304         else
7305                 io_bio->bi_end_io = btrfs_endio_direct_read;
7306
7307         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
7308         if (!ret)
7309                 return;
7310
7311 free_io_bio:
7312         bio_put(io_bio);
7313
7314 free_ordered:
7315         /*
7316          * If this is a write, we need to clean up the reserved space and kill
7317          * the ordered extent.
7318          */
7319         if (write) {
7320                 struct btrfs_ordered_extent *ordered;
7321                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
7322                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
7323                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
7324                         btrfs_free_reserved_extent(root, ordered->start,
7325                                                    ordered->disk_len);
7326                 btrfs_put_ordered_extent(ordered);
7327                 btrfs_put_ordered_extent(ordered);
7328         }
7329         bio_endio(dio_bio, ret);
7330 }
7331
7332 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
7333                         const struct iovec *iov, loff_t offset,
7334                         unsigned long nr_segs)
7335 {
7336         int seg;
7337         int i;
7338         size_t size;
7339         unsigned long addr;
7340         unsigned blocksize_mask = root->sectorsize - 1;
7341         ssize_t retval = -EINVAL;
7342         loff_t end = offset;
7343
7344         if (offset & blocksize_mask)
7345                 goto out;
7346
7347         /* Check the memory alignment.  Blocks cannot straddle pages */
7348         for (seg = 0; seg < nr_segs; seg++) {
7349                 addr = (unsigned long)iov[seg].iov_base;
7350                 size = iov[seg].iov_len;
7351                 end += size;
7352                 if ((addr & blocksize_mask) || (size & blocksize_mask))
7353                         goto out;
7354
7355                 /* If this is a write we don't need to check anymore */
7356                 if (rw & WRITE)
7357                         continue;
7358
7359                 /*
7360                  * Check to make sure we don't have duplicate iov_base's in this
7361                  * iovec, if so return EINVAL, otherwise we'll get csum errors
7362                  * when reading back.
7363                  */
7364                 for (i = seg + 1; i < nr_segs; i++) {
7365                         if (iov[seg].iov_base == iov[i].iov_base)
7366                                 goto out;
7367                 }
7368         }
7369         retval = 0;
7370 out:
7371         return retval;
7372 }
7373
7374 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
7375                         const struct iovec *iov, loff_t offset,
7376                         unsigned long nr_segs)
7377 {
7378         struct file *file = iocb->ki_filp;
7379         struct inode *inode = file->f_mapping->host;
7380         size_t count = 0;
7381         int flags = 0;
7382         bool wakeup = true;
7383         bool relock = false;
7384         ssize_t ret;
7385
7386         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
7387                             offset, nr_segs))
7388                 return 0;
7389
7390         atomic_inc(&inode->i_dio_count);
7391         smp_mb__after_atomic_inc();
7392
7393         if (rw & WRITE) {
7394                 count = iov_length(iov, nr_segs);
7395                 /*
7396                  * If the write DIO is beyond the EOF, we need update
7397                  * the isize, but it is protected by i_mutex. So we can
7398                  * not unlock the i_mutex at this case.
7399                  */
7400                 if (offset + count <= inode->i_size) {
7401                         mutex_unlock(&inode->i_mutex);
7402                         relock = true;
7403                 }
7404                 ret = btrfs_delalloc_reserve_space(inode, count);
7405                 if (ret)
7406                         goto out;
7407         } else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
7408                                      &BTRFS_I(inode)->runtime_flags))) {
7409                 inode_dio_done(inode);
7410                 flags = DIO_LOCKING | DIO_SKIP_HOLES;
7411                 wakeup = false;
7412         }
7413
7414         ret = __blockdev_direct_IO(rw, iocb, inode,
7415                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
7416                         iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
7417                         btrfs_submit_direct, flags);
7418         if (rw & WRITE) {
7419                 if (ret < 0 && ret != -EIOCBQUEUED)
7420                         btrfs_delalloc_release_space(inode, count);
7421                 else if (ret >= 0 && (size_t)ret < count)
7422                         btrfs_delalloc_release_space(inode,
7423                                                      count - (size_t)ret);
7424                 else
7425                         btrfs_delalloc_release_metadata(inode, 0);
7426         }
7427 out:
7428         if (wakeup)
7429                 inode_dio_done(inode);
7430         if (relock)
7431                 mutex_lock(&inode->i_mutex);
7432
7433         return ret;
7434 }
7435
7436 #define BTRFS_FIEMAP_FLAGS      (FIEMAP_FLAG_SYNC)
7437
7438 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7439                 __u64 start, __u64 len)
7440 {
7441         int     ret;
7442
7443         ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
7444         if (ret)
7445                 return ret;
7446
7447         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
7448 }
7449
7450 int btrfs_readpage(struct file *file, struct page *page)
7451 {
7452         struct extent_io_tree *tree;
7453         tree = &BTRFS_I(page->mapping->host)->io_tree;
7454         return extent_read_full_page(tree, page, btrfs_get_extent, 0);
7455 }
7456
7457 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
7458 {
7459         struct extent_io_tree *tree;
7460
7461
7462         if (current->flags & PF_MEMALLOC) {
7463                 redirty_page_for_writepage(wbc, page);
7464                 unlock_page(page);
7465                 return 0;
7466         }
7467         tree = &BTRFS_I(page->mapping->host)->io_tree;
7468         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
7469 }
7470
7471 static int btrfs_writepages(struct address_space *mapping,
7472                             struct writeback_control *wbc)
7473 {
7474         struct extent_io_tree *tree;
7475
7476         tree = &BTRFS_I(mapping->host)->io_tree;
7477         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
7478 }
7479
7480 static int
7481 btrfs_readpages(struct file *file, struct address_space *mapping,
7482                 struct list_head *pages, unsigned nr_pages)
7483 {
7484         struct extent_io_tree *tree;
7485         tree = &BTRFS_I(mapping->host)->io_tree;
7486         return extent_readpages(tree, mapping, pages, nr_pages,
7487                                 btrfs_get_extent);
7488 }
7489 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7490 {
7491         struct extent_io_tree *tree;
7492         struct extent_map_tree *map;
7493         int ret;
7494
7495         tree = &BTRFS_I(page->mapping->host)->io_tree;
7496         map = &BTRFS_I(page->mapping->host)->extent_tree;
7497         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
7498         if (ret == 1) {
7499                 ClearPagePrivate(page);
7500                 set_page_private(page, 0);
7501                 page_cache_release(page);
7502         }
7503         return ret;
7504 }
7505
7506 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7507 {
7508         if (PageWriteback(page) || PageDirty(page))
7509                 return 0;
7510         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
7511 }
7512
7513 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
7514 {
7515         struct inode *inode = page->mapping->host;
7516         struct extent_io_tree *tree;
7517         struct btrfs_ordered_extent *ordered;
7518         struct extent_state *cached_state = NULL;
7519         u64 page_start = page_offset(page);
7520         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
7521
7522         /*
7523          * we have the page locked, so new writeback can't start,
7524          * and the dirty bit won't be cleared while we are here.
7525          *
7526          * Wait for IO on this page so that we can safely clear
7527          * the PagePrivate2 bit and do ordered accounting
7528          */
7529         wait_on_page_writeback(page);
7530
7531         tree = &BTRFS_I(inode)->io_tree;
7532         if (offset) {
7533                 btrfs_releasepage(page, GFP_NOFS);
7534                 return;
7535         }
7536         lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7537         ordered = btrfs_lookup_ordered_extent(inode, page_offset(page));
7538         if (ordered) {
7539                 /*
7540                  * IO on this page will never be started, so we need
7541                  * to account for any ordered extents now
7542                  */
7543                 clear_extent_bit(tree, page_start, page_end,
7544                                  EXTENT_DIRTY | EXTENT_DELALLOC |
7545                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
7546                                  EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS);
7547                 /*
7548                  * whoever cleared the private bit is responsible
7549                  * for the finish_ordered_io
7550                  */
7551                 if (TestClearPagePrivate2(page) &&
7552                     btrfs_dec_test_ordered_pending(inode, &ordered, page_start,
7553                                                    PAGE_CACHE_SIZE, 1)) {
7554                         btrfs_finish_ordered_io(ordered);
7555                 }
7556                 btrfs_put_ordered_extent(ordered);
7557                 cached_state = NULL;
7558                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7559         }
7560         clear_extent_bit(tree, page_start, page_end,
7561                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
7562                  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
7563                  &cached_state, GFP_NOFS);
7564         __btrfs_releasepage(page, GFP_NOFS);
7565
7566         ClearPageChecked(page);
7567         if (PagePrivate(page)) {
7568                 ClearPagePrivate(page);
7569                 set_page_private(page, 0);
7570                 page_cache_release(page);
7571         }
7572 }
7573
7574 /*
7575  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
7576  * called from a page fault handler when a page is first dirtied. Hence we must
7577  * be careful to check for EOF conditions here. We set the page up correctly
7578  * for a written page which means we get ENOSPC checking when writing into
7579  * holes and correct delalloc and unwritten extent mapping on filesystems that
7580  * support these features.
7581  *
7582  * We are not allowed to take the i_mutex here so we have to play games to
7583  * protect against truncate races as the page could now be beyond EOF.  Because
7584  * vmtruncate() writes the inode size before removing pages, once we have the
7585  * page lock we can determine safely if the page is beyond EOF. If it is not
7586  * beyond EOF, then the page is guaranteed safe against truncation until we
7587  * unlock the page.
7588  */
7589 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
7590 {
7591         struct page *page = vmf->page;
7592         struct inode *inode = file_inode(vma->vm_file);
7593         struct btrfs_root *root = BTRFS_I(inode)->root;
7594         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7595         struct btrfs_ordered_extent *ordered;
7596         struct extent_state *cached_state = NULL;
7597         char *kaddr;
7598         unsigned long zero_start;
7599         loff_t size;
7600         int ret;
7601         int reserved = 0;
7602         u64 page_start;
7603         u64 page_end;
7604
7605         sb_start_pagefault(inode->i_sb);
7606         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
7607         if (!ret) {
7608                 ret = file_update_time(vma->vm_file);
7609                 reserved = 1;
7610         }
7611         if (ret) {
7612                 if (ret == -ENOMEM)
7613                         ret = VM_FAULT_OOM;
7614                 else /* -ENOSPC, -EIO, etc */
7615                         ret = VM_FAULT_SIGBUS;
7616                 if (reserved)
7617                         goto out;
7618                 goto out_noreserve;
7619         }
7620
7621         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
7622 again:
7623         lock_page(page);
7624         size = i_size_read(inode);
7625         page_start = page_offset(page);
7626         page_end = page_start + PAGE_CACHE_SIZE - 1;
7627
7628         if ((page->mapping != inode->i_mapping) ||
7629             (page_start >= size)) {
7630                 /* page got truncated out from underneath us */
7631                 goto out_unlock;
7632         }
7633         wait_on_page_writeback(page);
7634
7635         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
7636         set_page_extent_mapped(page);
7637
7638         /*
7639          * we can't set the delalloc bits if there are pending ordered
7640          * extents.  Drop our locks and wait for them to finish
7641          */
7642         ordered = btrfs_lookup_ordered_extent(inode, page_start);
7643         if (ordered) {
7644                 unlock_extent_cached(io_tree, page_start, page_end,
7645                                      &cached_state, GFP_NOFS);
7646                 unlock_page(page);
7647                 btrfs_start_ordered_extent(inode, ordered, 1);
7648                 btrfs_put_ordered_extent(ordered);
7649                 goto again;
7650         }
7651
7652         /*
7653          * XXX - page_mkwrite gets called every time the page is dirtied, even
7654          * if it was already dirty, so for space accounting reasons we need to
7655          * clear any delalloc bits for the range we are fixing to save.  There
7656          * is probably a better way to do this, but for now keep consistent with
7657          * prepare_pages in the normal write path.
7658          */
7659         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
7660                           EXTENT_DIRTY | EXTENT_DELALLOC |
7661                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
7662                           0, 0, &cached_state, GFP_NOFS);
7663
7664         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
7665                                         &cached_state);
7666         if (ret) {
7667                 unlock_extent_cached(io_tree, page_start, page_end,
7668                                      &cached_state, GFP_NOFS);
7669                 ret = VM_FAULT_SIGBUS;
7670                 goto out_unlock;
7671         }
7672         ret = 0;
7673
7674         /* page is wholly or partially inside EOF */
7675         if (page_start + PAGE_CACHE_SIZE > size)
7676                 zero_start = size & ~PAGE_CACHE_MASK;
7677         else
7678                 zero_start = PAGE_CACHE_SIZE;
7679
7680         if (zero_start != PAGE_CACHE_SIZE) {
7681                 kaddr = kmap(page);
7682                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
7683                 flush_dcache_page(page);
7684                 kunmap(page);
7685         }
7686         ClearPageChecked(page);
7687         set_page_dirty(page);
7688         SetPageUptodate(page);
7689
7690         BTRFS_I(inode)->last_trans = root->fs_info->generation;
7691         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
7692         BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
7693
7694         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
7695
7696 out_unlock:
7697         if (!ret) {
7698                 sb_end_pagefault(inode->i_sb);
7699                 return VM_FAULT_LOCKED;
7700         }
7701         unlock_page(page);
7702 out:
7703         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
7704 out_noreserve:
7705         sb_end_pagefault(inode->i_sb);
7706         return ret;
7707 }
7708
7709 static int btrfs_truncate(struct inode *inode)
7710 {
7711         struct btrfs_root *root = BTRFS_I(inode)->root;
7712         struct btrfs_block_rsv *rsv;
7713         int ret;
7714         int err = 0;
7715         struct btrfs_trans_handle *trans;
7716         u64 mask = root->sectorsize - 1;
7717         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
7718
7719         ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
7720         if (ret)
7721                 return ret;
7722
7723         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
7724         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
7725
7726         /*
7727          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
7728          * 3 things going on here
7729          *
7730          * 1) We need to reserve space for our orphan item and the space to
7731          * delete our orphan item.  Lord knows we don't want to have a dangling
7732          * orphan item because we didn't reserve space to remove it.
7733          *
7734          * 2) We need to reserve space to update our inode.
7735          *
7736          * 3) We need to have something to cache all the space that is going to
7737          * be free'd up by the truncate operation, but also have some slack
7738          * space reserved in case it uses space during the truncate (thank you
7739          * very much snapshotting).
7740          *
7741          * And we need these to all be seperate.  The fact is we can use alot of
7742          * space doing the truncate, and we have no earthly idea how much space
7743          * we will use, so we need the truncate reservation to be seperate so it
7744          * doesn't end up using space reserved for updating the inode or
7745          * removing the orphan item.  We also need to be able to stop the
7746          * transaction and start a new one, which means we need to be able to
7747          * update the inode several times, and we have no idea of knowing how
7748          * many times that will be, so we can't just reserve 1 item for the
7749          * entirety of the opration, so that has to be done seperately as well.
7750          * Then there is the orphan item, which does indeed need to be held on
7751          * to for the whole operation, and we need nobody to touch this reserved
7752          * space except the orphan code.
7753          *
7754          * So that leaves us with
7755          *
7756          * 1) root->orphan_block_rsv - for the orphan deletion.
7757          * 2) rsv - for the truncate reservation, which we will steal from the
7758          * transaction reservation.
7759          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
7760          * updating the inode.
7761          */
7762         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
7763         if (!rsv)
7764                 return -ENOMEM;
7765         rsv->size = min_size;
7766         rsv->failfast = 1;
7767
7768         /*
7769          * 1 for the truncate slack space
7770          * 1 for updating the inode.
7771          */
7772         trans = btrfs_start_transaction(root, 2);
7773         if (IS_ERR(trans)) {
7774                 err = PTR_ERR(trans);
7775                 goto out;
7776         }
7777
7778         /* Migrate the slack space for the truncate to our reserve */
7779         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
7780                                       min_size);
7781         BUG_ON(ret);
7782
7783         /*
7784          * setattr is responsible for setting the ordered_data_close flag,
7785          * but that is only tested during the last file release.  That
7786          * could happen well after the next commit, leaving a great big
7787          * window where new writes may get lost if someone chooses to write
7788          * to this file after truncating to zero
7789          *
7790          * The inode doesn't have any dirty data here, and so if we commit
7791          * this is a noop.  If someone immediately starts writing to the inode
7792          * it is very likely we'll catch some of their writes in this
7793          * transaction, and the commit will find this file on the ordered
7794          * data list with good things to send down.
7795          *
7796          * This is a best effort solution, there is still a window where
7797          * using truncate to replace the contents of the file will
7798          * end up with a zero length file after a crash.
7799          */
7800         if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
7801                                            &BTRFS_I(inode)->runtime_flags))
7802                 btrfs_add_ordered_operation(trans, root, inode);
7803
7804         /*
7805          * So if we truncate and then write and fsync we normally would just
7806          * write the extents that changed, which is a problem if we need to
7807          * first truncate that entire inode.  So set this flag so we write out
7808          * all of the extents in the inode to the sync log so we're completely
7809          * safe.
7810          */
7811         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
7812         trans->block_rsv = rsv;
7813
7814         while (1) {
7815                 ret = btrfs_truncate_inode_items(trans, root, inode,
7816                                                  inode->i_size,
7817                                                  BTRFS_EXTENT_DATA_KEY);
7818                 if (ret != -ENOSPC) {
7819                         err = ret;
7820                         break;
7821                 }
7822
7823                 trans->block_rsv = &root->fs_info->trans_block_rsv;
7824                 ret = btrfs_update_inode(trans, root, inode);
7825                 if (ret) {
7826                         err = ret;
7827                         break;
7828                 }
7829
7830                 btrfs_end_transaction(trans, root);
7831                 btrfs_btree_balance_dirty(root);
7832
7833                 trans = btrfs_start_transaction(root, 2);
7834                 if (IS_ERR(trans)) {
7835                         ret = err = PTR_ERR(trans);
7836                         trans = NULL;
7837                         break;
7838                 }
7839
7840                 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
7841                                               rsv, min_size);
7842                 BUG_ON(ret);    /* shouldn't happen */
7843                 trans->block_rsv = rsv;
7844         }
7845
7846         if (ret == 0 && inode->i_nlink > 0) {
7847                 trans->block_rsv = root->orphan_block_rsv;
7848                 ret = btrfs_orphan_del(trans, inode);
7849                 if (ret)
7850                         err = ret;
7851         }
7852
7853         if (trans) {
7854                 trans->block_rsv = &root->fs_info->trans_block_rsv;
7855                 ret = btrfs_update_inode(trans, root, inode);
7856                 if (ret && !err)
7857                         err = ret;
7858
7859                 ret = btrfs_end_transaction(trans, root);
7860                 btrfs_btree_balance_dirty(root);
7861         }
7862
7863 out:
7864         btrfs_free_block_rsv(root, rsv);
7865
7866         if (ret && !err)
7867                 err = ret;
7868
7869         return err;
7870 }
7871
7872 /*
7873  * create a new subvolume directory/inode (helper for the ioctl).
7874  */
7875 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
7876                              struct btrfs_root *new_root, u64 new_dirid)
7877 {
7878         struct inode *inode;
7879         int err;
7880         u64 index = 0;
7881
7882         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7883                                 new_dirid, new_dirid,
7884                                 S_IFDIR | (~current_umask() & S_IRWXUGO),
7885                                 &index);
7886         if (IS_ERR(inode))
7887                 return PTR_ERR(inode);
7888         inode->i_op = &btrfs_dir_inode_operations;
7889         inode->i_fop = &btrfs_dir_file_operations;
7890
7891         set_nlink(inode, 1);
7892         btrfs_i_size_write(inode, 0);
7893
7894         err = btrfs_update_inode(trans, new_root, inode);
7895
7896         iput(inode);
7897         return err;
7898 }
7899
7900 struct inode *btrfs_alloc_inode(struct super_block *sb)
7901 {
7902         struct btrfs_inode *ei;
7903         struct inode *inode;
7904
7905         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
7906         if (!ei)
7907                 return NULL;
7908
7909         ei->root = NULL;
7910         ei->generation = 0;
7911         ei->last_trans = 0;
7912         ei->last_sub_trans = 0;
7913         ei->logged_trans = 0;
7914         ei->delalloc_bytes = 0;
7915         ei->disk_i_size = 0;
7916         ei->flags = 0;
7917         ei->csum_bytes = 0;
7918         ei->index_cnt = (u64)-1;
7919         ei->last_unlink_trans = 0;
7920         ei->last_log_commit = 0;
7921
7922         spin_lock_init(&ei->lock);
7923         ei->outstanding_extents = 0;
7924         ei->reserved_extents = 0;
7925
7926         ei->runtime_flags = 0;
7927         ei->force_compress = BTRFS_COMPRESS_NONE;
7928
7929         ei->delayed_node = NULL;
7930
7931         inode = &ei->vfs_inode;
7932         extent_map_tree_init(&ei->extent_tree);
7933         extent_io_tree_init(&ei->io_tree, &inode->i_data);
7934         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
7935         ei->io_tree.track_uptodate = 1;
7936         ei->io_failure_tree.track_uptodate = 1;
7937         atomic_set(&ei->sync_writers, 0);
7938         mutex_init(&ei->log_mutex);
7939         mutex_init(&ei->delalloc_mutex);
7940         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
7941         INIT_LIST_HEAD(&ei->delalloc_inodes);
7942         INIT_LIST_HEAD(&ei->ordered_operations);
7943         RB_CLEAR_NODE(&ei->rb_node);
7944
7945         return inode;
7946 }
7947
7948 static void btrfs_i_callback(struct rcu_head *head)
7949 {
7950         struct inode *inode = container_of(head, struct inode, i_rcu);
7951         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7952 }
7953
7954 void btrfs_destroy_inode(struct inode *inode)
7955 {
7956         struct btrfs_ordered_extent *ordered;
7957         struct btrfs_root *root = BTRFS_I(inode)->root;
7958
7959         WARN_ON(!hlist_empty(&inode->i_dentry));
7960         WARN_ON(inode->i_data.nrpages);
7961         WARN_ON(BTRFS_I(inode)->outstanding_extents);
7962         WARN_ON(BTRFS_I(inode)->reserved_extents);
7963         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
7964         WARN_ON(BTRFS_I(inode)->csum_bytes);
7965
7966         /*
7967          * This can happen where we create an inode, but somebody else also
7968          * created the same inode and we need to destroy the one we already
7969          * created.
7970          */
7971         if (!root)
7972                 goto free;
7973
7974         /*
7975          * Make sure we're properly removed from the ordered operation
7976          * lists.
7977          */
7978         smp_mb();
7979         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
7980                 spin_lock(&root->fs_info->ordered_extent_lock);
7981                 list_del_init(&BTRFS_I(inode)->ordered_operations);
7982                 spin_unlock(&root->fs_info->ordered_extent_lock);
7983         }
7984
7985         if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
7986                      &BTRFS_I(inode)->runtime_flags)) {
7987                 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
7988                         (unsigned long long)btrfs_ino(inode));
7989                 atomic_dec(&root->orphan_inodes);
7990         }
7991
7992         while (1) {
7993                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7994                 if (!ordered)
7995                         break;
7996                 else {
7997                         btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
7998                                 (unsigned long long)ordered->file_offset,
7999                                 (unsigned long long)ordered->len);
8000                         btrfs_remove_ordered_extent(inode, ordered);
8001                         btrfs_put_ordered_extent(ordered);
8002                         btrfs_put_ordered_extent(ordered);
8003                 }
8004         }
8005         inode_tree_del(inode);
8006         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8007 free:
8008         call_rcu(&inode->i_rcu, btrfs_i_callback);
8009 }
8010
8011 int btrfs_drop_inode(struct inode *inode)
8012 {
8013         struct btrfs_root *root = BTRFS_I(inode)->root;
8014
8015         if (root == NULL)
8016                 return 1;
8017
8018         /* the snap/subvol tree is on deleting */
8019         if (btrfs_root_refs(&root->root_item) == 0 &&
8020             root != root->fs_info->tree_root)
8021                 return 1;
8022         else
8023                 return generic_drop_inode(inode);
8024 }
8025
8026 static void init_once(void *foo)
8027 {
8028         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8029
8030         inode_init_once(&ei->vfs_inode);
8031 }
8032
8033 void btrfs_destroy_cachep(void)
8034 {
8035         /*
8036          * Make sure all delayed rcu free inodes are flushed before we
8037          * destroy cache.
8038          */
8039         rcu_barrier();
8040         if (btrfs_inode_cachep)
8041                 kmem_cache_destroy(btrfs_inode_cachep);
8042         if (btrfs_trans_handle_cachep)
8043                 kmem_cache_destroy(btrfs_trans_handle_cachep);
8044         if (btrfs_transaction_cachep)
8045                 kmem_cache_destroy(btrfs_transaction_cachep);
8046         if (btrfs_path_cachep)
8047                 kmem_cache_destroy(btrfs_path_cachep);
8048         if (btrfs_free_space_cachep)
8049                 kmem_cache_destroy(btrfs_free_space_cachep);
8050         if (btrfs_delalloc_work_cachep)
8051                 kmem_cache_destroy(btrfs_delalloc_work_cachep);
8052 }
8053
8054 int btrfs_init_cachep(void)
8055 {
8056         btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8057                         sizeof(struct btrfs_inode), 0,
8058                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
8059         if (!btrfs_inode_cachep)
8060                 goto fail;
8061
8062         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8063                         sizeof(struct btrfs_trans_handle), 0,
8064                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8065         if (!btrfs_trans_handle_cachep)
8066                 goto fail;
8067
8068         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
8069                         sizeof(struct btrfs_transaction), 0,
8070                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8071         if (!btrfs_transaction_cachep)
8072                 goto fail;
8073
8074         btrfs_path_cachep = kmem_cache_create("btrfs_path",
8075                         sizeof(struct btrfs_path), 0,
8076                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8077         if (!btrfs_path_cachep)
8078                 goto fail;
8079
8080         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
8081                         sizeof(struct btrfs_free_space), 0,
8082                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8083         if (!btrfs_free_space_cachep)
8084                 goto fail;
8085
8086         btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
8087                         sizeof(struct btrfs_delalloc_work), 0,
8088                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
8089                         NULL);
8090         if (!btrfs_delalloc_work_cachep)
8091                 goto fail;
8092
8093         return 0;
8094 fail:
8095         btrfs_destroy_cachep();
8096         return -ENOMEM;
8097 }
8098
8099 static int btrfs_getattr(struct vfsmount *mnt,
8100                          struct dentry *dentry, struct kstat *stat)
8101 {
8102         u64 delalloc_bytes;
8103         struct inode *inode = dentry->d_inode;
8104         u32 blocksize = inode->i_sb->s_blocksize;
8105
8106         generic_fillattr(inode, stat);
8107         stat->dev = BTRFS_I(inode)->root->anon_dev;
8108         stat->blksize = PAGE_CACHE_SIZE;
8109
8110         spin_lock(&BTRFS_I(inode)->lock);
8111         delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
8112         spin_unlock(&BTRFS_I(inode)->lock);
8113         stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
8114                         ALIGN(delalloc_bytes, blocksize)) >> 9;
8115         return 0;
8116 }
8117
8118 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
8119                            struct inode *new_dir, struct dentry *new_dentry)
8120 {
8121         struct btrfs_trans_handle *trans;
8122         struct btrfs_root *root = BTRFS_I(old_dir)->root;
8123         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8124         struct inode *new_inode = new_dentry->d_inode;
8125         struct inode *old_inode = old_dentry->d_inode;
8126         struct timespec ctime = CURRENT_TIME;
8127         u64 index = 0;
8128         u64 root_objectid;
8129         int ret;
8130         u64 old_ino = btrfs_ino(old_inode);
8131
8132         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
8133                 return -EPERM;
8134
8135         /* we only allow rename subvolume link between subvolumes */
8136         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8137                 return -EXDEV;
8138
8139         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8140             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
8141                 return -ENOTEMPTY;
8142
8143         if (S_ISDIR(old_inode->i_mode) && new_inode &&
8144             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8145                 return -ENOTEMPTY;
8146
8147
8148         /* check for collisions, even if the  name isn't there */
8149         ret = btrfs_check_dir_item_collision(root, new_dir->i_ino,
8150                              new_dentry->d_name.name,
8151                              new_dentry->d_name.len);
8152
8153         if (ret) {
8154                 if (ret == -EEXIST) {
8155                         /* we shouldn't get
8156                          * eexist without a new_inode */
8157                         if (!new_inode) {
8158                                 WARN_ON(1);
8159                                 return ret;
8160                         }
8161                 } else {
8162                         /* maybe -EOVERFLOW */
8163                         return ret;
8164                 }
8165         }
8166         ret = 0;
8167
8168         /*
8169          * we're using rename to replace one file with another.
8170          * and the replacement file is large.  Start IO on it now so
8171          * we don't add too much work to the end of the transaction
8172          */
8173         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
8174             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
8175                 filemap_flush(old_inode->i_mapping);
8176
8177         /* close the racy window with snapshot create/destroy ioctl */
8178         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8179                 down_read(&root->fs_info->subvol_sem);
8180         /*
8181          * We want to reserve the absolute worst case amount of items.  So if
8182          * both inodes are subvols and we need to unlink them then that would
8183          * require 4 item modifications, but if they are both normal inodes it
8184          * would require 5 item modifications, so we'll assume their normal
8185          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8186          * should cover the worst case number of items we'll modify.
8187          */
8188         trans = btrfs_start_transaction(root, 11);
8189         if (IS_ERR(trans)) {
8190                 ret = PTR_ERR(trans);
8191                 goto out_notrans;
8192         }
8193
8194         if (dest != root)
8195                 btrfs_record_root_in_trans(trans, dest);
8196
8197         ret = btrfs_set_inode_index(new_dir, &index);
8198         if (ret)
8199                 goto out_fail;
8200
8201         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8202                 /* force full log commit if subvolume involved. */
8203                 root->fs_info->last_trans_log_full_commit = trans->transid;
8204         } else {
8205                 ret = btrfs_insert_inode_ref(trans, dest,
8206                                              new_dentry->d_name.name,
8207                                              new_dentry->d_name.len,
8208                                              old_ino,
8209                                              btrfs_ino(new_dir), index);
8210                 if (ret)
8211                         goto out_fail;
8212                 /*
8213                  * this is an ugly little race, but the rename is required
8214                  * to make sure that if we crash, the inode is either at the
8215                  * old name or the new one.  pinning the log transaction lets
8216                  * us make sure we don't allow a log commit to come in after
8217                  * we unlink the name but before we add the new name back in.
8218                  */
8219                 btrfs_pin_log_trans(root);
8220         }
8221         /*
8222          * make sure the inode gets flushed if it is replacing
8223          * something.
8224          */
8225         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
8226                 btrfs_add_ordered_operation(trans, root, old_inode);
8227
8228         inode_inc_iversion(old_dir);
8229         inode_inc_iversion(new_dir);
8230         inode_inc_iversion(old_inode);
8231         old_dir->i_ctime = old_dir->i_mtime = ctime;
8232         new_dir->i_ctime = new_dir->i_mtime = ctime;
8233         old_inode->i_ctime = ctime;
8234
8235         if (old_dentry->d_parent != new_dentry->d_parent)
8236                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
8237
8238         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8239                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
8240                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
8241                                         old_dentry->d_name.name,
8242                                         old_dentry->d_name.len);
8243         } else {
8244                 ret = __btrfs_unlink_inode(trans, root, old_dir,
8245                                         old_dentry->d_inode,
8246                                         old_dentry->d_name.name,
8247                                         old_dentry->d_name.len);
8248                 if (!ret)
8249                         ret = btrfs_update_inode(trans, root, old_inode);
8250         }
8251         if (ret) {
8252                 btrfs_abort_transaction(trans, root, ret);
8253                 goto out_fail;
8254         }
8255
8256         if (new_inode) {
8257                 inode_inc_iversion(new_inode);
8258                 new_inode->i_ctime = CURRENT_TIME;
8259                 if (unlikely(btrfs_ino(new_inode) ==
8260                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
8261                         root_objectid = BTRFS_I(new_inode)->location.objectid;
8262                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
8263                                                 root_objectid,
8264                                                 new_dentry->d_name.name,
8265                                                 new_dentry->d_name.len);
8266                         BUG_ON(new_inode->i_nlink == 0);
8267                 } else {
8268                         ret = btrfs_unlink_inode(trans, dest, new_dir,
8269                                                  new_dentry->d_inode,
8270                                                  new_dentry->d_name.name,
8271                                                  new_dentry->d_name.len);
8272                 }
8273                 if (!ret && new_inode->i_nlink == 0) {
8274                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
8275                         BUG_ON(ret);
8276                 }
8277                 if (ret) {
8278                         btrfs_abort_transaction(trans, root, ret);
8279                         goto out_fail;
8280                 }
8281         }
8282
8283         ret = btrfs_add_link(trans, new_dir, old_inode,
8284                              new_dentry->d_name.name,
8285                              new_dentry->d_name.len, 0, index);
8286         if (ret) {
8287                 btrfs_abort_transaction(trans, root, ret);
8288                 goto out_fail;
8289         }
8290
8291         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
8292                 struct dentry *parent = new_dentry->d_parent;
8293                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
8294                 btrfs_end_log_trans(root);
8295         }
8296 out_fail:
8297         btrfs_end_transaction(trans, root);
8298 out_notrans:
8299         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8300                 up_read(&root->fs_info->subvol_sem);
8301
8302         return ret;
8303 }
8304
8305 static void btrfs_run_delalloc_work(struct btrfs_work *work)
8306 {
8307         struct btrfs_delalloc_work *delalloc_work;
8308
8309         delalloc_work = container_of(work, struct btrfs_delalloc_work,
8310                                      work);
8311         if (delalloc_work->wait)
8312                 btrfs_wait_ordered_range(delalloc_work->inode, 0, (u64)-1);
8313         else
8314                 filemap_flush(delalloc_work->inode->i_mapping);
8315
8316         if (delalloc_work->delay_iput)
8317                 btrfs_add_delayed_iput(delalloc_work->inode);
8318         else
8319                 iput(delalloc_work->inode);
8320         complete(&delalloc_work->completion);
8321 }
8322
8323 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
8324                                                     int wait, int delay_iput)
8325 {
8326         struct btrfs_delalloc_work *work;
8327
8328         work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
8329         if (!work)
8330                 return NULL;
8331
8332         init_completion(&work->completion);
8333         INIT_LIST_HEAD(&work->list);
8334         work->inode = inode;
8335         work->wait = wait;
8336         work->delay_iput = delay_iput;
8337         work->work.func = btrfs_run_delalloc_work;
8338
8339         return work;
8340 }
8341
8342 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
8343 {
8344         wait_for_completion(&work->completion);
8345         kmem_cache_free(btrfs_delalloc_work_cachep, work);
8346 }
8347
8348 /*
8349  * some fairly slow code that needs optimization. This walks the list
8350  * of all the inodes with pending delalloc and forces them to disk.
8351  */
8352 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8353 {
8354         struct btrfs_inode *binode;
8355         struct inode *inode;
8356         struct btrfs_delalloc_work *work, *next;
8357         struct list_head works;
8358         struct list_head splice;
8359         int ret = 0;
8360
8361         if (root->fs_info->sb->s_flags & MS_RDONLY)
8362                 return -EROFS;
8363
8364         INIT_LIST_HEAD(&works);
8365         INIT_LIST_HEAD(&splice);
8366
8367         spin_lock(&root->fs_info->delalloc_lock);
8368         list_splice_init(&root->fs_info->delalloc_inodes, &splice);
8369         while (!list_empty(&splice)) {
8370                 binode = list_entry(splice.next, struct btrfs_inode,
8371                                     delalloc_inodes);
8372
8373                 list_del_init(&binode->delalloc_inodes);
8374
8375                 inode = igrab(&binode->vfs_inode);
8376                 if (!inode) {
8377                         clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
8378                                   &binode->runtime_flags);
8379                         continue;
8380                 }
8381
8382                 list_add_tail(&binode->delalloc_inodes,
8383                               &root->fs_info->delalloc_inodes);
8384                 spin_unlock(&root->fs_info->delalloc_lock);
8385
8386                 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
8387                 if (unlikely(!work)) {
8388                         ret = -ENOMEM;
8389                         goto out;
8390                 }
8391                 list_add_tail(&work->list, &works);
8392                 btrfs_queue_worker(&root->fs_info->flush_workers,
8393                                    &work->work);
8394
8395                 cond_resched();
8396                 spin_lock(&root->fs_info->delalloc_lock);
8397         }
8398         spin_unlock(&root->fs_info->delalloc_lock);
8399
8400         list_for_each_entry_safe(work, next, &works, list) {
8401                 list_del_init(&work->list);
8402                 btrfs_wait_and_free_delalloc_work(work);
8403         }
8404
8405         /* the filemap_flush will queue IO into the worker threads, but
8406          * we have to make sure the IO is actually started and that
8407          * ordered extents get created before we return
8408          */
8409         atomic_inc(&root->fs_info->async_submit_draining);
8410         while (atomic_read(&root->fs_info->nr_async_submits) ||
8411               atomic_read(&root->fs_info->async_delalloc_pages)) {
8412                 wait_event(root->fs_info->async_submit_wait,
8413                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
8414                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8415         }
8416         atomic_dec(&root->fs_info->async_submit_draining);
8417         return 0;
8418 out:
8419         list_for_each_entry_safe(work, next, &works, list) {
8420                 list_del_init(&work->list);
8421                 btrfs_wait_and_free_delalloc_work(work);
8422         }
8423
8424         if (!list_empty_careful(&splice)) {
8425                 spin_lock(&root->fs_info->delalloc_lock);
8426                 list_splice_tail(&splice, &root->fs_info->delalloc_inodes);
8427                 spin_unlock(&root->fs_info->delalloc_lock);
8428         }
8429         return ret;
8430 }
8431
8432 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
8433                          const char *symname)
8434 {
8435         struct btrfs_trans_handle *trans;
8436         struct btrfs_root *root = BTRFS_I(dir)->root;
8437         struct btrfs_path *path;
8438         struct btrfs_key key;
8439         struct inode *inode = NULL;
8440         int err;
8441         int drop_inode = 0;
8442         u64 objectid;
8443         u64 index = 0 ;
8444         int name_len;
8445         int datasize;
8446         unsigned long ptr;
8447         struct btrfs_file_extent_item *ei;
8448         struct extent_buffer *leaf;
8449
8450         name_len = strlen(symname) + 1;
8451         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
8452                 return -ENAMETOOLONG;
8453
8454         /*
8455          * 2 items for inode item and ref
8456          * 2 items for dir items
8457          * 1 item for xattr if selinux is on
8458          */
8459         trans = btrfs_start_transaction(root, 5);
8460         if (IS_ERR(trans))
8461                 return PTR_ERR(trans);
8462
8463         err = btrfs_find_free_ino(root, &objectid);
8464         if (err)
8465                 goto out_unlock;
8466
8467         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
8468                                 dentry->d_name.len, btrfs_ino(dir), objectid,
8469                                 S_IFLNK|S_IRWXUGO, &index);
8470         if (IS_ERR(inode)) {
8471                 err = PTR_ERR(inode);
8472                 goto out_unlock;
8473         }
8474
8475         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
8476         if (err) {
8477                 drop_inode = 1;
8478                 goto out_unlock;
8479         }
8480
8481         /*
8482         * If the active LSM wants to access the inode during
8483         * d_instantiate it needs these. Smack checks to see
8484         * if the filesystem supports xattrs by looking at the
8485         * ops vector.
8486         */
8487         inode->i_fop = &btrfs_file_operations;
8488         inode->i_op = &btrfs_file_inode_operations;
8489
8490         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
8491         if (err)
8492                 drop_inode = 1;
8493         else {
8494                 inode->i_mapping->a_ops = &btrfs_aops;
8495                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8496                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
8497         }
8498         if (drop_inode)
8499                 goto out_unlock;
8500
8501         path = btrfs_alloc_path();
8502         if (!path) {
8503                 err = -ENOMEM;
8504                 drop_inode = 1;
8505                 goto out_unlock;
8506         }
8507         key.objectid = btrfs_ino(inode);
8508         key.offset = 0;
8509         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
8510         datasize = btrfs_file_extent_calc_inline_size(name_len);
8511         err = btrfs_insert_empty_item(trans, root, path, &key,
8512                                       datasize);
8513         if (err) {
8514                 drop_inode = 1;
8515                 btrfs_free_path(path);
8516                 goto out_unlock;
8517         }
8518         leaf = path->nodes[0];
8519         ei = btrfs_item_ptr(leaf, path->slots[0],
8520                             struct btrfs_file_extent_item);
8521         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
8522         btrfs_set_file_extent_type(leaf, ei,
8523                                    BTRFS_FILE_EXTENT_INLINE);
8524         btrfs_set_file_extent_encryption(leaf, ei, 0);
8525         btrfs_set_file_extent_compression(leaf, ei, 0);
8526         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
8527         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
8528
8529         ptr = btrfs_file_extent_inline_start(ei);
8530         write_extent_buffer(leaf, symname, ptr, name_len);
8531         btrfs_mark_buffer_dirty(leaf);
8532         btrfs_free_path(path);
8533
8534         inode->i_op = &btrfs_symlink_inode_operations;
8535         inode->i_mapping->a_ops = &btrfs_symlink_aops;
8536         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8537         inode_set_bytes(inode, name_len);
8538         btrfs_i_size_write(inode, name_len - 1);
8539         err = btrfs_update_inode(trans, root, inode);
8540         if (err)
8541                 drop_inode = 1;
8542
8543 out_unlock:
8544         if (!err)
8545                 d_instantiate(dentry, inode);
8546         btrfs_end_transaction(trans, root);
8547         if (drop_inode) {
8548                 inode_dec_link_count(inode);
8549                 iput(inode);
8550         }
8551         btrfs_btree_balance_dirty(root);
8552         return err;
8553 }
8554
8555 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
8556                                        u64 start, u64 num_bytes, u64 min_size,
8557                                        loff_t actual_len, u64 *alloc_hint,
8558                                        struct btrfs_trans_handle *trans)
8559 {
8560         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
8561         struct extent_map *em;
8562         struct btrfs_root *root = BTRFS_I(inode)->root;
8563         struct btrfs_key ins;
8564         u64 cur_offset = start;
8565         u64 i_size;
8566         u64 cur_bytes;
8567         int ret = 0;
8568         bool own_trans = true;
8569
8570         if (trans)
8571                 own_trans = false;
8572         while (num_bytes > 0) {
8573                 if (own_trans) {
8574                         trans = btrfs_start_transaction(root, 3);
8575                         if (IS_ERR(trans)) {
8576                                 ret = PTR_ERR(trans);
8577                                 break;
8578                         }
8579                 }
8580
8581                 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
8582                 cur_bytes = max(cur_bytes, min_size);
8583                 ret = btrfs_reserve_extent(trans, root, cur_bytes,
8584                                            min_size, 0, *alloc_hint, &ins, 1);
8585                 if (ret) {
8586                         if (own_trans)
8587                                 btrfs_end_transaction(trans, root);
8588                         break;
8589                 }
8590
8591                 ret = insert_reserved_file_extent(trans, inode,
8592                                                   cur_offset, ins.objectid,
8593                                                   ins.offset, ins.offset,
8594                                                   ins.offset, 0, 0, 0,
8595                                                   BTRFS_FILE_EXTENT_PREALLOC);
8596                 if (ret) {
8597                         btrfs_abort_transaction(trans, root, ret);
8598                         if (own_trans)
8599                                 btrfs_end_transaction(trans, root);
8600                         break;
8601                 }
8602                 btrfs_drop_extent_cache(inode, cur_offset,
8603                                         cur_offset + ins.offset -1, 0);
8604
8605                 em = alloc_extent_map();
8606                 if (!em) {
8607                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
8608                                 &BTRFS_I(inode)->runtime_flags);
8609                         goto next;
8610                 }
8611
8612                 em->start = cur_offset;
8613                 em->orig_start = cur_offset;
8614                 em->len = ins.offset;
8615                 em->block_start = ins.objectid;
8616                 em->block_len = ins.offset;
8617                 em->orig_block_len = ins.offset;
8618                 em->ram_bytes = ins.offset;
8619                 em->bdev = root->fs_info->fs_devices->latest_bdev;
8620                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
8621                 em->generation = trans->transid;
8622
8623                 while (1) {
8624                         write_lock(&em_tree->lock);
8625                         ret = add_extent_mapping(em_tree, em, 1);
8626                         write_unlock(&em_tree->lock);
8627                         if (ret != -EEXIST)
8628                                 break;
8629                         btrfs_drop_extent_cache(inode, cur_offset,
8630                                                 cur_offset + ins.offset - 1,
8631                                                 0);
8632                 }
8633                 free_extent_map(em);
8634 next:
8635                 num_bytes -= ins.offset;
8636                 cur_offset += ins.offset;
8637                 *alloc_hint = ins.objectid + ins.offset;
8638
8639                 inode_inc_iversion(inode);
8640                 inode->i_ctime = CURRENT_TIME;
8641                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
8642                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
8643                     (actual_len > inode->i_size) &&
8644                     (cur_offset > inode->i_size)) {
8645                         if (cur_offset > actual_len)
8646                                 i_size = actual_len;
8647                         else
8648                                 i_size = cur_offset;
8649                         i_size_write(inode, i_size);
8650                         btrfs_ordered_update_i_size(inode, i_size, NULL);
8651                 }
8652
8653                 ret = btrfs_update_inode(trans, root, inode);
8654
8655                 if (ret) {
8656                         btrfs_abort_transaction(trans, root, ret);
8657                         if (own_trans)
8658                                 btrfs_end_transaction(trans, root);
8659                         break;
8660                 }
8661
8662                 if (own_trans)
8663                         btrfs_end_transaction(trans, root);
8664         }
8665         return ret;
8666 }
8667
8668 int btrfs_prealloc_file_range(struct inode *inode, int mode,
8669                               u64 start, u64 num_bytes, u64 min_size,
8670                               loff_t actual_len, u64 *alloc_hint)
8671 {
8672         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8673                                            min_size, actual_len, alloc_hint,
8674                                            NULL);
8675 }
8676
8677 int btrfs_prealloc_file_range_trans(struct inode *inode,
8678                                     struct btrfs_trans_handle *trans, int mode,
8679                                     u64 start, u64 num_bytes, u64 min_size,
8680                                     loff_t actual_len, u64 *alloc_hint)
8681 {
8682         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8683                                            min_size, actual_len, alloc_hint, trans);
8684 }
8685
8686 static int btrfs_set_page_dirty(struct page *page)
8687 {
8688         return __set_page_dirty_nobuffers(page);
8689 }
8690
8691 static int btrfs_permission(struct inode *inode, int mask)
8692 {
8693         struct btrfs_root *root = BTRFS_I(inode)->root;
8694         umode_t mode = inode->i_mode;
8695
8696         if (mask & MAY_WRITE &&
8697             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
8698                 if (btrfs_root_readonly(root))
8699                         return -EROFS;
8700                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
8701                         return -EACCES;
8702         }
8703         return generic_permission(inode, mask);
8704 }
8705
8706 static const struct inode_operations btrfs_dir_inode_operations = {
8707         .getattr        = btrfs_getattr,
8708         .lookup         = btrfs_lookup,
8709         .create         = btrfs_create,
8710         .unlink         = btrfs_unlink,
8711         .link           = btrfs_link,
8712         .mkdir          = btrfs_mkdir,
8713         .rmdir          = btrfs_rmdir,
8714         .rename         = btrfs_rename,
8715         .symlink        = btrfs_symlink,
8716         .setattr        = btrfs_setattr,
8717         .mknod          = btrfs_mknod,
8718         .setxattr       = btrfs_setxattr,
8719         .getxattr       = btrfs_getxattr,
8720         .listxattr      = btrfs_listxattr,
8721         .removexattr    = btrfs_removexattr,
8722         .permission     = btrfs_permission,
8723         .get_acl        = btrfs_get_acl,
8724 };
8725 static const struct inode_operations btrfs_dir_ro_inode_operations = {
8726         .lookup         = btrfs_lookup,
8727         .permission     = btrfs_permission,
8728         .get_acl        = btrfs_get_acl,
8729 };
8730
8731 static const struct file_operations btrfs_dir_file_operations = {
8732         .llseek         = generic_file_llseek,
8733         .read           = generic_read_dir,
8734         .readdir        = btrfs_real_readdir,
8735         .unlocked_ioctl = btrfs_ioctl,
8736 #ifdef CONFIG_COMPAT
8737         .compat_ioctl   = btrfs_ioctl,
8738 #endif
8739         .release        = btrfs_release_file,
8740         .fsync          = btrfs_sync_file,
8741 };
8742
8743 static struct extent_io_ops btrfs_extent_io_ops = {
8744         .fill_delalloc = run_delalloc_range,
8745         .submit_bio_hook = btrfs_submit_bio_hook,
8746         .merge_bio_hook = btrfs_merge_bio_hook,
8747         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
8748         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
8749         .writepage_start_hook = btrfs_writepage_start_hook,
8750         .set_bit_hook = btrfs_set_bit_hook,
8751         .clear_bit_hook = btrfs_clear_bit_hook,
8752         .merge_extent_hook = btrfs_merge_extent_hook,
8753         .split_extent_hook = btrfs_split_extent_hook,
8754 };
8755
8756 /*
8757  * btrfs doesn't support the bmap operation because swapfiles
8758  * use bmap to make a mapping of extents in the file.  They assume
8759  * these extents won't change over the life of the file and they
8760  * use the bmap result to do IO directly to the drive.
8761  *
8762  * the btrfs bmap call would return logical addresses that aren't
8763  * suitable for IO and they also will change frequently as COW
8764  * operations happen.  So, swapfile + btrfs == corruption.
8765  *
8766  * For now we're avoiding this by dropping bmap.
8767  */
8768 static const struct address_space_operations btrfs_aops = {
8769         .readpage       = btrfs_readpage,
8770         .writepage      = btrfs_writepage,
8771         .writepages     = btrfs_writepages,
8772         .readpages      = btrfs_readpages,
8773         .direct_IO      = btrfs_direct_IO,
8774         .invalidatepage = btrfs_invalidatepage,
8775         .releasepage    = btrfs_releasepage,
8776         .set_page_dirty = btrfs_set_page_dirty,
8777         .error_remove_page = generic_error_remove_page,
8778 };
8779
8780 static const struct address_space_operations btrfs_symlink_aops = {
8781         .readpage       = btrfs_readpage,
8782         .writepage      = btrfs_writepage,
8783         .invalidatepage = btrfs_invalidatepage,
8784         .releasepage    = btrfs_releasepage,
8785 };
8786
8787 static const struct inode_operations btrfs_file_inode_operations = {
8788         .getattr        = btrfs_getattr,
8789         .setattr        = btrfs_setattr,
8790         .setxattr       = btrfs_setxattr,
8791         .getxattr       = btrfs_getxattr,
8792         .listxattr      = btrfs_listxattr,
8793         .removexattr    = btrfs_removexattr,
8794         .permission     = btrfs_permission,
8795         .fiemap         = btrfs_fiemap,
8796         .get_acl        = btrfs_get_acl,
8797         .update_time    = btrfs_update_time,
8798 };
8799 static const struct inode_operations btrfs_special_inode_operations = {
8800         .getattr        = btrfs_getattr,
8801         .setattr        = btrfs_setattr,
8802         .permission     = btrfs_permission,
8803         .setxattr       = btrfs_setxattr,
8804         .getxattr       = btrfs_getxattr,
8805         .listxattr      = btrfs_listxattr,
8806         .removexattr    = btrfs_removexattr,
8807         .get_acl        = btrfs_get_acl,
8808         .update_time    = btrfs_update_time,
8809 };
8810 static const struct inode_operations btrfs_symlink_inode_operations = {
8811         .readlink       = generic_readlink,
8812         .follow_link    = page_follow_link_light,
8813         .put_link       = page_put_link,
8814         .getattr        = btrfs_getattr,
8815         .setattr        = btrfs_setattr,
8816         .permission     = btrfs_permission,
8817         .setxattr       = btrfs_setxattr,
8818         .getxattr       = btrfs_getxattr,
8819         .listxattr      = btrfs_listxattr,
8820         .removexattr    = btrfs_removexattr,
8821         .get_acl        = btrfs_get_acl,
8822         .update_time    = btrfs_update_time,
8823 };
8824
8825 const struct dentry_operations btrfs_dentry_operations = {
8826         .d_delete       = btrfs_dentry_delete,
8827         .d_release      = btrfs_dentry_release,
8828 };