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