fs/btrfs/inode.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/pagemap.h>
  25 #include <linux/highmem.h>
  26 #include <linux/time.h>
  27 #include <linux/init.h>
  28 #include <linux/string.h>
  29 #include <linux/backing-dev.h>
  30 #include <linux/mpage.h>
  31 #include <linux/swap.h>
  32 #include <linux/writeback.h>
  33 #include <linux/statfs.h>
  34 #include <linux/compat.h>
  35 #include <linux/bit_spinlock.h>
  36 #include <linux/xattr.h>
  37 #include <linux/posix_acl.h>
  38 #include <linux/falloc.h>
  39 #include <linux/slab.h>
  40 #include "compat.h"
  41 #include "ctree.h"
  42 #include "disk-io.h"
  43 #include "transaction.h"
  44 #include "btrfs_inode.h"
  45 #include "ioctl.h"
  46 #include "print-tree.h"
  47 #include "volumes.h"
  48 #include "ordered-data.h"
  49 #include "xattr.h"
  50 #include "tree-log.h"
  51 #include "compression.h"
  52 #include "locking.h"
  53 #include "free-space-cache.h"
  54
  55 struct btrfs_iget_args {
  56         u64 ino;
  57         struct btrfs_root *root;
  58 };
  59
  60 static const struct inode_operations btrfs_dir_inode_operations;
  61 static const struct inode_operations btrfs_symlink_inode_operations;
  62 static const struct inode_operations btrfs_dir_ro_inode_operations;
  63 static const struct inode_operations btrfs_special_inode_operations;
  64 static const struct inode_operations btrfs_file_inode_operations;
  65 static const struct address_space_operations btrfs_aops;
  66 static const struct address_space_operations btrfs_symlink_aops;
  67 static const struct file_operations btrfs_dir_file_operations;
  68 static struct extent_io_ops btrfs_extent_io_ops;
  69
  70 static struct kmem_cache *btrfs_inode_cachep;
  71 struct kmem_cache *btrfs_trans_handle_cachep;
  72 struct kmem_cache *btrfs_transaction_cachep;
  73 struct kmem_cache *btrfs_path_cachep;
  74 struct kmem_cache *btrfs_free_space_cachep;
  75
  76 #define S_SHIFT 12
  77 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  78         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  79         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  80         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  81         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  82         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  83         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  84         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  85 };
  86
  87 static int btrfs_setsize(struct inode *inode, loff_t newsize);
  88 static int btrfs_truncate(struct inode *inode);
  89 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
  90 static noinline int cow_file_range(struct inode *inode,
  91                                    struct page *locked_page,
  92                                    u64 start, u64 end, int *page_started,
  93                                    unsigned long *nr_written, int unlock);
  94
  95 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
  96                                      struct inode *inode,  struct inode *dir)
  97 {
  98         int err;
  99
 100         err = btrfs_init_acl(trans, inode, dir);
 101         if (!err)
 102                 err = btrfs_xattr_security_init(trans, inode, dir);
 103         return err;
 104 }
 105
 106 /*
 107  * this does all the hard work for inserting an inline extent into
 108  * the btree.  The caller should have done a btrfs_drop_extents so that
 109  * no overlapping inline items exist in the btree
 110  */
 111 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 112                                 struct btrfs_root *root, struct inode *inode,
 113                                 u64 start, size_t size, size_t compressed_size,
 114                                 struct page **compressed_pages)
 115 {
 116         struct btrfs_key key;
 117         struct btrfs_path *path;
 118         struct extent_buffer *leaf;
 119         struct page *page = NULL;
 120         char *kaddr;
 121         unsigned long ptr;
 122         struct btrfs_file_extent_item *ei;
 123         int err = 0;
 124         int ret;
 125         size_t cur_size = size;
 126         size_t datasize;
 127         unsigned long offset;
 128         int compress_type = BTRFS_COMPRESS_NONE;
 129
 130         if (compressed_size && compressed_pages) {
 131                 compress_type = root->fs_info->compress_type;
 132                 cur_size = compressed_size;
 133         }
 134
 135         path = btrfs_alloc_path();
 136         if (!path)
 137                 return -ENOMEM;
 138
 139         path->leave_spinning = 1;
 140         btrfs_set_trans_block_group(trans, inode);
 141
 142         key.objectid = inode->i_ino;
 143         key.offset = start;
 144         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 145         datasize = btrfs_file_extent_calc_inline_size(cur_size);
 146
 147         inode_add_bytes(inode, size);
 148         ret = btrfs_insert_empty_item(trans, root, path, &key,
 149                                       datasize);
 150         BUG_ON(ret);
 151         if (ret) {
 152                 err = ret;
 153                 goto fail;
 154         }
 155         leaf = path->nodes[0];
 156         ei = btrfs_item_ptr(leaf, path->slots[0],
 157                             struct btrfs_file_extent_item);
 158         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 159         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 160         btrfs_set_file_extent_encryption(leaf, ei, 0);
 161         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 162         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 163         ptr = btrfs_file_extent_inline_start(ei);
 164
 165         if (compress_type != BTRFS_COMPRESS_NONE) {
 166                 struct page *cpage;
 167                 int i = 0;
 168                 while (compressed_size > 0) {
 169                         cpage = compressed_pages[i];
 170                         cur_size = min_t(unsigned long, compressed_size,
 171                                        PAGE_CACHE_SIZE);
 172
 173                         kaddr = kmap_atomic(cpage, KM_USER0);
 174                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 175                         kunmap_atomic(kaddr, KM_USER0);
 176
 177                         i++;
 178                         ptr += cur_size;
 179                         compressed_size -= cur_size;
 180                 }
 181                 btrfs_set_file_extent_compression(leaf, ei,
 182                                                   compress_type);
 183         } else {
 184                 page = find_get_page(inode->i_mapping,
 185                                      start >> PAGE_CACHE_SHIFT);
 186                 btrfs_set_file_extent_compression(leaf, ei, 0);
 187                 kaddr = kmap_atomic(page, KM_USER0);
 188                 offset = start & (PAGE_CACHE_SIZE - 1);
 189                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 190                 kunmap_atomic(kaddr, KM_USER0);
 191                 page_cache_release(page);
 192         }
 193         btrfs_mark_buffer_dirty(leaf);
 194         btrfs_free_path(path);
 195
 196         /*
 197          * we're an inline extent, so nobody can
 198          * extend the file past i_size without locking
 199          * a page we already have locked.
 200          *
 201          * We must do any isize and inode updates
 202          * before we unlock the pages.  Otherwise we
 203          * could end up racing with unlink.
 204          */
 205         BTRFS_I(inode)->disk_i_size = inode->i_size;
 206         btrfs_update_inode(trans, root, inode);
 207
 208         return 0;
 209 fail:
 210         btrfs_free_path(path);
 211         return err;
 212 }
 213
 214
 215 /*
 216  * conditionally insert an inline extent into the file.  This
 217  * does the checks required to make sure the data is small enough
 218  * to fit as an inline extent.
 219  */
 220 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 221                                  struct btrfs_root *root,
 222                                  struct inode *inode, u64 start, u64 end,
 223                                  size_t compressed_size,
 224                                  struct page **compressed_pages)
 225 {
 226         u64 isize = i_size_read(inode);
 227         u64 actual_end = min(end + 1, isize);
 228         u64 inline_len = actual_end - start;
 229         u64 aligned_end = (end + root->sectorsize - 1) &
 230                         ~((u64)root->sectorsize - 1);
 231         u64 hint_byte;
 232         u64 data_len = inline_len;
 233         int ret;
 234
 235         if (compressed_size)
 236                 data_len = compressed_size;
 237
 238         if (start > 0 ||
 239             actual_end >= PAGE_CACHE_SIZE ||
 240             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 241             (!compressed_size &&
 242             (actual_end & (root->sectorsize - 1)) == 0) ||
 243             end + 1 < isize ||
 244             data_len > root->fs_info->max_inline) {
 245                 return 1;
 246         }
 247
 248         ret = btrfs_drop_extents(trans, inode, start, aligned_end,
 249                                  &hint_byte, 1);
 250         BUG_ON(ret);
 251
 252         if (isize > actual_end)
 253                 inline_len = min_t(u64, isize, actual_end);
 254         ret = insert_inline_extent(trans, root, inode, start,
 255                                    inline_len, compressed_size,
 256                                    compressed_pages);
 257         BUG_ON(ret);
 258         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 259         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 260         return 0;
 261 }
 262
 263 struct async_extent {
 264         u64 start;
 265         u64 ram_size;
 266         u64 compressed_size;
 267         struct page **pages;
 268         unsigned long nr_pages;
 269         int compress_type;
 270         struct list_head list;
 271 };
 272
 273 struct async_cow {
 274         struct inode *inode;
 275         struct btrfs_root *root;
 276         struct page *locked_page;
 277         u64 start;
 278         u64 end;
 279         struct list_head extents;
 280         struct btrfs_work work;
 281 };
 282
 283 static noinline int add_async_extent(struct async_cow *cow,
 284                                      u64 start, u64 ram_size,
 285                                      u64 compressed_size,
 286                                      struct page **pages,
 287                                      unsigned long nr_pages,
 288                                      int compress_type)
 289 {
 290         struct async_extent *async_extent;
 291
 292         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 293         async_extent->start = start;
 294         async_extent->ram_size = ram_size;
 295         async_extent->compressed_size = compressed_size;
 296         async_extent->pages = pages;
 297         async_extent->nr_pages = nr_pages;
 298         async_extent->compress_type = compress_type;
 299         list_add_tail(&async_extent->list, &cow->extents);
 300         return 0;
 301 }
 302
 303 /*
 304  * we create compressed extents in two phases.  The first
 305  * phase compresses a range of pages that have already been
 306  * locked (both pages and state bits are locked).
 307  *
 308  * This is done inside an ordered work queue, and the compression
 309  * is spread across many cpus.  The actual IO submission is step
 310  * two, and the ordered work queue takes care of making sure that
 311  * happens in the same order things were put onto the queue by
 312  * writepages and friends.
 313  *
 314  * If this code finds it can't get good compression, it puts an
 315  * entry onto the work queue to write the uncompressed bytes.  This
 316  * makes sure that both compressed inodes and uncompressed inodes
 317  * are written in the same order that pdflush sent them down.
 318  */
 319 static noinline int compress_file_range(struct inode *inode,
 320                                         struct page *locked_page,
 321                                         u64 start, u64 end,
 322                                         struct async_cow *async_cow,
 323                                         int *num_added)
 324 {
 325         struct btrfs_root *root = BTRFS_I(inode)->root;
 326         struct btrfs_trans_handle *trans;
 327         u64 num_bytes;
 328         u64 blocksize = root->sectorsize;
 329         u64 actual_end;
 330         u64 isize = i_size_read(inode);
 331         int ret = 0;
 332         struct page **pages = NULL;
 333         unsigned long nr_pages;
 334         unsigned long nr_pages_ret = 0;
 335         unsigned long total_compressed = 0;
 336         unsigned long total_in = 0;
 337         unsigned long max_compressed = 128 * 1024;
 338         unsigned long max_uncompressed = 128 * 1024;
 339         int i;
 340         int will_compress;
 341         int compress_type = root->fs_info->compress_type;
 342
 343         actual_end = min_t(u64, isize, end + 1);
 344 again:
 345         will_compress = 0;
 346         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 347         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 348
 349         /*
 350          * we don't want to send crud past the end of i_size through
 351          * compression, that's just a waste of CPU time.  So, if the
 352          * end of the file is before the start of our current
 353          * requested range of bytes, we bail out to the uncompressed
 354          * cleanup code that can deal with all of this.
 355          *
 356          * It isn't really the fastest way to fix things, but this is a
 357          * very uncommon corner.
 358          */
 359         if (actual_end <= start)
 360                 goto cleanup_and_bail_uncompressed;
 361
 362         total_compressed = actual_end - start;
 363
 364         /* we want to make sure that amount of ram required to uncompress
 365          * an extent is reasonable, so we limit the total size in ram
 366          * of a compressed extent to 128k.  This is a crucial number
 367          * because it also controls how easily we can spread reads across
 368          * cpus for decompression.
 369          *
 370          * We also want to make sure the amount of IO required to do
 371          * a random read is reasonably small, so we limit the size of
 372          * a compressed extent to 128k.
 373          */
 374         total_compressed = min(total_compressed, max_uncompressed);
 375         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 376         num_bytes = max(blocksize,  num_bytes);
 377         total_in = 0;
 378         ret = 0;
 379
 380         /*
 381          * we do compression for mount -o compress and when the
 382          * inode has not been flagged as nocompress.  This flag can
 383          * change at any time if we discover bad compression ratios.
 384          */
 385         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 386             (btrfs_test_opt(root, COMPRESS) ||
 387              (BTRFS_I(inode)->force_compress))) {
 388                 WARN_ON(pages);
 389                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 390
 391                 if (BTRFS_I(inode)->force_compress)
 392                         compress_type = BTRFS_I(inode)->force_compress;
 393
 394                 ret = btrfs_compress_pages(compress_type,
 395                                            inode->i_mapping, start,
 396                                            total_compressed, pages,
 397                                            nr_pages, &nr_pages_ret,
 398                                            &total_in,
 399                                            &total_compressed,
 400                                            max_compressed);
 401
 402                 if (!ret) {
 403                         unsigned long offset = total_compressed &
 404                                 (PAGE_CACHE_SIZE - 1);
 405                         struct page *page = pages[nr_pages_ret - 1];
 406                         char *kaddr;
 407
 408                         /* zero the tail end of the last page, we might be
 409                          * sending it down to disk
 410                          */
 411                         if (offset) {
 412                                 kaddr = kmap_atomic(page, KM_USER0);
 413                                 memset(kaddr + offset, 0,
 414                                        PAGE_CACHE_SIZE - offset);
 415                                 kunmap_atomic(kaddr, KM_USER0);
 416                         }
 417                         will_compress = 1;
 418                 }
 419         }
 420         if (start == 0) {
 421                 trans = btrfs_join_transaction(root, 1);
 422                 BUG_ON(IS_ERR(trans));
 423                 btrfs_set_trans_block_group(trans, inode);
 424                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 425
 426                 /* lets try to make an inline extent */
 427                 if (ret || total_in < (actual_end - start)) {
 428                         /* we didn't compress the entire range, try
 429                          * to make an uncompressed inline extent.
 430                          */
 431                         ret = cow_file_range_inline(trans, root, inode,
 432                                                     start, end, 0, NULL);
 433                 } else {
 434                         /* try making a compressed inline extent */
 435                         ret = cow_file_range_inline(trans, root, inode,
 436                                                     start, end,
 437                                                     total_compressed, pages);
 438                 }
 439                 if (ret == 0) {
 440                         /*
 441                          * inline extent creation worked, we don't need
 442                          * to create any more async work items.  Unlock
 443                          * and free up our temp pages.
 444                          */
 445                         extent_clear_unlock_delalloc(inode,
 446                              &BTRFS_I(inode)->io_tree,
 447                              start, end, NULL,
 448                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
 449                              EXTENT_CLEAR_DELALLOC |
 450                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
 451
 452                         btrfs_end_transaction(trans, root);
 453                         goto free_pages_out;
 454                 }
 455                 btrfs_end_transaction(trans, root);
 456         }
 457
 458         if (will_compress) {
 459                 /*
 460                  * we aren't doing an inline extent round the compressed size
 461                  * up to a block size boundary so the allocator does sane
 462                  * things
 463                  */
 464                 total_compressed = (total_compressed + blocksize - 1) &
 465                         ~(blocksize - 1);
 466
 467                 /*
 468                  * one last check to make sure the compression is really a
 469                  * win, compare the page count read with the blocks on disk
 470                  */
 471                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
 472                         ~(PAGE_CACHE_SIZE - 1);
 473                 if (total_compressed >= total_in) {
 474                         will_compress = 0;
 475                 } else {
 476                         num_bytes = total_in;
 477                 }
 478         }
 479         if (!will_compress && pages) {
 480                 /*
 481                  * the compression code ran but failed to make things smaller,
 482                  * free any pages it allocated and our page pointer array
 483                  */
 484                 for (i = 0; i < nr_pages_ret; i++) {
 485                         WARN_ON(pages[i]->mapping);
 486                         page_cache_release(pages[i]);
 487                 }
 488                 kfree(pages);
 489                 pages = NULL;
 490                 total_compressed = 0;
 491                 nr_pages_ret = 0;
 492
 493                 /* flag the file so we don't compress in the future */
 494                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 495                     !(BTRFS_I(inode)->force_compress)) {
 496                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 497                 }
 498         }
 499         if (will_compress) {
 500                 *num_added += 1;
 501
 502                 /* the async work queues will take care of doing actual
 503                  * allocation on disk for these compressed pages,
 504                  * and will submit them to the elevator.
 505                  */
 506                 add_async_extent(async_cow, start, num_bytes,
 507                                  total_compressed, pages, nr_pages_ret,
 508                                  compress_type);
 509
 510                 if (start + num_bytes < end) {
 511                         start += num_bytes;
 512                         pages = NULL;
 513                         cond_resched();
 514                         goto again;
 515                 }
 516         } else {
 517 cleanup_and_bail_uncompressed:
 518                 /*
 519                  * No compression, but we still need to write the pages in
 520                  * the file we've been given so far.  redirty the locked
 521                  * page if it corresponds to our extent and set things up
 522                  * for the async work queue to run cow_file_range to do
 523                  * the normal delalloc dance
 524                  */
 525                 if (page_offset(locked_page) >= start &&
 526                     page_offset(locked_page) <= end) {
 527                         __set_page_dirty_nobuffers(locked_page);
 528                         /* unlocked later on in the async handlers */
 529                 }
 530                 add_async_extent(async_cow, start, end - start + 1,
 531                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 532                 *num_added += 1;
 533         }
 534
 535 out:
 536         return 0;
 537
 538 free_pages_out:
 539         for (i = 0; i < nr_pages_ret; i++) {
 540                 WARN_ON(pages[i]->mapping);
 541                 page_cache_release(pages[i]);
 542         }
 543         kfree(pages);
 544
 545         goto out;
 546 }
 547
 548 /*
 549  * phase two of compressed writeback.  This is the ordered portion
 550  * of the code, which only gets called in the order the work was
 551  * queued.  We walk all the async extents created by compress_file_range
 552  * and send them down to the disk.
 553  */
 554 static noinline int submit_compressed_extents(struct inode *inode,
 555                                               struct async_cow *async_cow)
 556 {
 557         struct async_extent *async_extent;
 558         u64 alloc_hint = 0;
 559         struct btrfs_trans_handle *trans;
 560         struct btrfs_key ins;
 561         struct extent_map *em;
 562         struct btrfs_root *root = BTRFS_I(inode)->root;
 563         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 564         struct extent_io_tree *io_tree;
 565         int ret = 0;
 566
 567         if (list_empty(&async_cow->extents))
 568                 return 0;
 569
 570
 571         while (!list_empty(&async_cow->extents)) {
 572                 async_extent = list_entry(async_cow->extents.next,
 573                                           struct async_extent, list);
 574                 list_del(&async_extent->list);
 575
 576                 io_tree = &BTRFS_I(inode)->io_tree;
 577
 578 retry:
 579                 /* did the compression code fall back to uncompressed IO? */
 580                 if (!async_extent->pages) {
 581                         int page_started = 0;
 582                         unsigned long nr_written = 0;
 583
 584                         lock_extent(io_tree, async_extent->start,
 585                                          async_extent->start +
 586                                          async_extent->ram_size - 1, GFP_NOFS);
 587
 588                         /* allocate blocks */
 589                         ret = cow_file_range(inode, async_cow->locked_page,
 590                                              async_extent->start,
 591                                              async_extent->start +
 592                                              async_extent->ram_size - 1,
 593                                              &page_started, &nr_written, 0);
 594
 595                         /*
 596                          * if page_started, cow_file_range inserted an
 597                          * inline extent and took care of all the unlocking
 598                          * and IO for us.  Otherwise, we need to submit
 599                          * all those pages down to the drive.
 600                          */
 601                         if (!page_started && !ret)
 602                                 extent_write_locked_range(io_tree,
 603                                                   inode, async_extent->start,
 604                                                   async_extent->start +
 605                                                   async_extent->ram_size - 1,
 606                                                   btrfs_get_extent,
 607                                                   WB_SYNC_ALL);
 608                         kfree(async_extent);
 609                         cond_resched();
 610                         continue;
 611                 }
 612
 613                 lock_extent(io_tree, async_extent->start,
 614                             async_extent->start + async_extent->ram_size - 1,
 615                             GFP_NOFS);
 616
 617                 trans = btrfs_join_transaction(root, 1);
 618                 BUG_ON(IS_ERR(trans));
 619                 ret = btrfs_reserve_extent(trans, root,
 620                                            async_extent->compressed_size,
 621                                            async_extent->compressed_size,
 622                                            0, alloc_hint,
 623                                            (u64)-1, &ins, 1);
 624                 btrfs_end_transaction(trans, root);
 625
 626                 if (ret) {
 627                         int i;
 628                         for (i = 0; i < async_extent->nr_pages; i++) {
 629                                 WARN_ON(async_extent->pages[i]->mapping);
 630                                 page_cache_release(async_extent->pages[i]);
 631                         }
 632                         kfree(async_extent->pages);
 633                         async_extent->nr_pages = 0;
 634                         async_extent->pages = NULL;
 635                         unlock_extent(io_tree, async_extent->start,
 636                                       async_extent->start +
 637                                       async_extent->ram_size - 1, GFP_NOFS);
 638                         goto retry;
 639                 }
 640
 641                 /*
 642                  * here we're doing allocation and writeback of the
 643                  * compressed pages
 644                  */
 645                 btrfs_drop_extent_cache(inode, async_extent->start,
 646                                         async_extent->start +
 647                                         async_extent->ram_size - 1, 0);
 648
 649                 em = alloc_extent_map(GFP_NOFS);
 650                 BUG_ON(!em);
 651                 em->start = async_extent->start;
 652                 em->len = async_extent->ram_size;
 653                 em->orig_start = em->start;
 654
 655                 em->block_start = ins.objectid;
 656                 em->block_len = ins.offset;
 657                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 658                 em->compress_type = async_extent->compress_type;
 659                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 660                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 661
 662                 while (1) {
 663                         write_lock(&em_tree->lock);
 664                         ret = add_extent_mapping(em_tree, em);
 665                         write_unlock(&em_tree->lock);
 666                         if (ret != -EEXIST) {
 667                                 free_extent_map(em);
 668                                 break;
 669                         }
 670                         btrfs_drop_extent_cache(inode, async_extent->start,
 671                                                 async_extent->start +
 672                                                 async_extent->ram_size - 1, 0);
 673                 }
 674
 675                 ret = btrfs_add_ordered_extent_compress(inode,
 676                                                 async_extent->start,
 677                                                 ins.objectid,
 678                                                 async_extent->ram_size,
 679                                                 ins.offset,
 680                                                 BTRFS_ORDERED_COMPRESSED,
 681                                                 async_extent->compress_type);
 682                 BUG_ON(ret);
 683
 684                 /*
 685                  * clear dirty, set writeback and unlock the pages.
 686                  */
 687                 extent_clear_unlock_delalloc(inode,
 688                                 &BTRFS_I(inode)->io_tree,
 689                                 async_extent->start,
 690                                 async_extent->start +
 691                                 async_extent->ram_size - 1,
 692                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 693                                 EXTENT_CLEAR_UNLOCK |
 694                                 EXTENT_CLEAR_DELALLOC |
 695                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
 696
 697                 ret = btrfs_submit_compressed_write(inode,
 698                                     async_extent->start,
 699                                     async_extent->ram_size,
 700                                     ins.objectid,
 701                                     ins.offset, async_extent->pages,
 702                                     async_extent->nr_pages);
 703
 704                 BUG_ON(ret);
 705                 alloc_hint = ins.objectid + ins.offset;
 706                 kfree(async_extent);
 707                 cond_resched();
 708         }
 709
 710         return 0;
 711 }
 712
 713 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 714                                       u64 num_bytes)
 715 {
 716         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 717         struct extent_map *em;
 718         u64 alloc_hint = 0;
 719
 720         read_lock(&em_tree->lock);
 721         em = search_extent_mapping(em_tree, start, num_bytes);
 722         if (em) {
 723                 /*
 724                  * if block start isn't an actual block number then find the
 725                  * first block in this inode and use that as a hint.  If that
 726                  * block is also bogus then just don't worry about it.
 727                  */
 728                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 729                         free_extent_map(em);
 730                         em = search_extent_mapping(em_tree, 0, 0);
 731                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 732                                 alloc_hint = em->block_start;
 733                         if (em)
 734                                 free_extent_map(em);
 735                 } else {
 736                         alloc_hint = em->block_start;
 737                         free_extent_map(em);
 738                 }
 739         }
 740         read_unlock(&em_tree->lock);
 741
 742         return alloc_hint;
 743 }
 744
 745 /*
 746  * when extent_io.c finds a delayed allocation range in the file,
 747  * the call backs end up in this code.  The basic idea is to
 748  * allocate extents on disk for the range, and create ordered data structs
 749  * in ram to track those extents.
 750  *
 751  * locked_page is the page that writepage had locked already.  We use
 752  * it to make sure we don't do extra locks or unlocks.
 753  *
 754  * *page_started is set to one if we unlock locked_page and do everything
 755  * required to start IO on it.  It may be clean and already done with
 756  * IO when we return.
 757  */
 758 static noinline int cow_file_range(struct inode *inode,
 759                                    struct page *locked_page,
 760                                    u64 start, u64 end, int *page_started,
 761                                    unsigned long *nr_written,
 762                                    int unlock)
 763 {
 764         struct btrfs_root *root = BTRFS_I(inode)->root;
 765         struct btrfs_trans_handle *trans;
 766         u64 alloc_hint = 0;
 767         u64 num_bytes;
 768         unsigned long ram_size;
 769         u64 disk_num_bytes;
 770         u64 cur_alloc_size;
 771         u64 blocksize = root->sectorsize;
 772         struct btrfs_key ins;
 773         struct extent_map *em;
 774         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 775         int ret = 0;
 776
 777         BUG_ON(root == root->fs_info->tree_root);
 778         trans = btrfs_join_transaction(root, 1);
 779         BUG_ON(IS_ERR(trans));
 780         btrfs_set_trans_block_group(trans, inode);
 781         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 782
 783         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 784         num_bytes = max(blocksize,  num_bytes);
 785         disk_num_bytes = num_bytes;
 786         ret = 0;
 787
 788         if (start == 0) {
 789                 /* lets try to make an inline extent */
 790                 ret = cow_file_range_inline(trans, root, inode,
 791                                             start, end, 0, NULL);
 792                 if (ret == 0) {
 793                         extent_clear_unlock_delalloc(inode,
 794                                      &BTRFS_I(inode)->io_tree,
 795                                      start, end, NULL,
 796                                      EXTENT_CLEAR_UNLOCK_PAGE |
 797                                      EXTENT_CLEAR_UNLOCK |
 798                                      EXTENT_CLEAR_DELALLOC |
 799                                      EXTENT_CLEAR_DIRTY |
 800                                      EXTENT_SET_WRITEBACK |
 801                                      EXTENT_END_WRITEBACK);
 802
 803                         *nr_written = *nr_written +
 804                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 805                         *page_started = 1;
 806                         ret = 0;
 807                         goto out;
 808                 }
 809         }
 810
 811         BUG_ON(disk_num_bytes >
 812                btrfs_super_total_bytes(&root->fs_info->super_copy));
 813
 814         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 815         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 816
 817         while (disk_num_bytes > 0) {
 818                 unsigned long op;
 819
 820                 cur_alloc_size = disk_num_bytes;
 821                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 822                                            root->sectorsize, 0, alloc_hint,
 823                                            (u64)-1, &ins, 1);
 824                 BUG_ON(ret);
 825
 826                 em = alloc_extent_map(GFP_NOFS);
 827                 BUG_ON(!em);
 828                 em->start = start;
 829                 em->orig_start = em->start;
 830                 ram_size = ins.offset;
 831                 em->len = ins.offset;
 832
 833                 em->block_start = ins.objectid;
 834                 em->block_len = ins.offset;
 835                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 836                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 837
 838                 while (1) {
 839                         write_lock(&em_tree->lock);
 840                         ret = add_extent_mapping(em_tree, em);
 841                         write_unlock(&em_tree->lock);
 842                         if (ret != -EEXIST) {
 843                                 free_extent_map(em);
 844                                 break;
 845                         }
 846                         btrfs_drop_extent_cache(inode, start,
 847                                                 start + ram_size - 1, 0);
 848                 }
 849
 850                 cur_alloc_size = ins.offset;
 851                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 852                                                ram_size, cur_alloc_size, 0);
 853                 BUG_ON(ret);
 854
 855                 if (root->root_key.objectid ==
 856                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 857                         ret = btrfs_reloc_clone_csums(inode, start,
 858                                                       cur_alloc_size);
 859                         BUG_ON(ret);
 860                 }
 861
 862                 if (disk_num_bytes < cur_alloc_size)
 863                         break;
 864
 865                 /* we're not doing compressed IO, don't unlock the first
 866                  * page (which the caller expects to stay locked), don't
 867                  * clear any dirty bits and don't set any writeback bits
 868                  *
 869                  * Do set the Private2 bit so we know this page was properly
 870                  * setup for writepage
 871                  */
 872                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
 873                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
 874                         EXTENT_SET_PRIVATE2;
 875
 876                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 877                                              start, start + ram_size - 1,
 878                                              locked_page, op);
 879                 disk_num_bytes -= cur_alloc_size;
 880                 num_bytes -= cur_alloc_size;
 881                 alloc_hint = ins.objectid + ins.offset;
 882                 start += cur_alloc_size;
 883         }
 884 out:
 885         ret = 0;
 886         btrfs_end_transaction(trans, root);
 887
 888         return ret;
 889 }
 890
 891 /*
 892  * work queue call back to started compression on a file and pages
 893  */
 894 static noinline void async_cow_start(struct btrfs_work *work)
 895 {
 896         struct async_cow *async_cow;
 897         int num_added = 0;
 898         async_cow = container_of(work, struct async_cow, work);
 899
 900         compress_file_range(async_cow->inode, async_cow->locked_page,
 901                             async_cow->start, async_cow->end, async_cow,
 902                             &num_added);
 903         if (num_added == 0)
 904                 async_cow->inode = NULL;
 905 }
 906
 907 /*
 908  * work queue call back to submit previously compressed pages
 909  */
 910 static noinline void async_cow_submit(struct btrfs_work *work)
 911 {
 912         struct async_cow *async_cow;
 913         struct btrfs_root *root;
 914         unsigned long nr_pages;
 915
 916         async_cow = container_of(work, struct async_cow, work);
 917
 918         root = async_cow->root;
 919         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
 920                 PAGE_CACHE_SHIFT;
 921
 922         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
 923
 924         if (atomic_read(&root->fs_info->async_delalloc_pages) <
 925             5 * 1042 * 1024 &&
 926             waitqueue_active(&root->fs_info->async_submit_wait))
 927                 wake_up(&root->fs_info->async_submit_wait);
 928
 929         if (async_cow->inode)
 930                 submit_compressed_extents(async_cow->inode, async_cow);
 931 }
 932
 933 static noinline void async_cow_free(struct btrfs_work *work)
 934 {
 935         struct async_cow *async_cow;
 936         async_cow = container_of(work, struct async_cow, work);
 937         kfree(async_cow);
 938 }
 939
 940 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
 941                                 u64 start, u64 end, int *page_started,
 942                                 unsigned long *nr_written)
 943 {
 944         struct async_cow *async_cow;
 945         struct btrfs_root *root = BTRFS_I(inode)->root;
 946         unsigned long nr_pages;
 947         u64 cur_end;
 948         int limit = 10 * 1024 * 1042;
 949
 950         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
 951                          1, 0, NULL, GFP_NOFS);
 952         while (start < end) {
 953                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
 954                 async_cow->inode = inode;
 955                 async_cow->root = root;
 956                 async_cow->locked_page = locked_page;
 957                 async_cow->start = start;
 958
 959                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
 960                         cur_end = end;
 961                 else
 962                         cur_end = min(end, start + 512 * 1024 - 1);
 963
 964                 async_cow->end = cur_end;
 965                 INIT_LIST_HEAD(&async_cow->extents);
 966
 967                 async_cow->work.func = async_cow_start;
 968                 async_cow->work.ordered_func = async_cow_submit;
 969                 async_cow->work.ordered_free = async_cow_free;
 970                 async_cow->work.flags = 0;
 971
 972                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
 973                         PAGE_CACHE_SHIFT;
 974                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
 975
 976                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
 977                                    &async_cow->work);
 978
 979                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
 980                         wait_event(root->fs_info->async_submit_wait,
 981                            (atomic_read(&root->fs_info->async_delalloc_pages) <
 982                             limit));
 983                 }
 984
 985                 while (atomic_read(&root->fs_info->async_submit_draining) &&
 986                       atomic_read(&root->fs_info->async_delalloc_pages)) {
 987                         wait_event(root->fs_info->async_submit_wait,
 988                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
 989                            0));
 990                 }
 991
 992                 *nr_written += nr_pages;
 993                 start = cur_end + 1;
 994         }
 995         *page_started = 1;
 996         return 0;
 997 }
 998
 999 static noinline int csum_exist_in_range(struct btrfs_root *root,
1000                                         u64 bytenr, u64 num_bytes)
1001 {
1002         int ret;
1003         struct btrfs_ordered_sum *sums;
1004         LIST_HEAD(list);
1005
1006         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1007                                        bytenr + num_bytes - 1, &list);
1008         if (ret == 0 && list_empty(&list))
1009                 return 0;
1010
1011         while (!list_empty(&list)) {
1012                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1013                 list_del(&sums->list);
1014                 kfree(sums);
1015         }
1016         return 1;
1017 }
1018
1019 /*
1020  * when nowcow writeback call back.  This checks for snapshots or COW copies
1021  * of the extents that exist in the file, and COWs the file as required.
1022  *
1023  * If no cow copies or snapshots exist, we write directly to the existing
1024  * blocks on disk
1025  */
1026 static noinline int run_delalloc_nocow(struct inode *inode,
1027                                        struct page *locked_page,
1028                               u64 start, u64 end, int *page_started, int force,
1029                               unsigned long *nr_written)
1030 {
1031         struct btrfs_root *root = BTRFS_I(inode)->root;
1032         struct btrfs_trans_handle *trans;
1033         struct extent_buffer *leaf;
1034         struct btrfs_path *path;
1035         struct btrfs_file_extent_item *fi;
1036         struct btrfs_key found_key;
1037         u64 cow_start;
1038         u64 cur_offset;
1039         u64 extent_end;
1040         u64 extent_offset;
1041         u64 disk_bytenr;
1042         u64 num_bytes;
1043         int extent_type;
1044         int ret;
1045         int type;
1046         int nocow;
1047         int check_prev = 1;
1048         bool nolock = false;
1049
1050         path = btrfs_alloc_path();
1051         BUG_ON(!path);
1052         if (root == root->fs_info->tree_root) {
1053                 nolock = true;
1054                 trans = btrfs_join_transaction_nolock(root, 1);
1055         } else {
1056                 trans = btrfs_join_transaction(root, 1);
1057         }
1058         BUG_ON(IS_ERR(trans));
1059
1060         cow_start = (u64)-1;
1061         cur_offset = start;
1062         while (1) {
1063                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
1064                                                cur_offset, 0);
1065                 BUG_ON(ret < 0);
1066                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1067                         leaf = path->nodes[0];
1068                         btrfs_item_key_to_cpu(leaf, &found_key,
1069                                               path->slots[0] - 1);
1070                         if (found_key.objectid == inode->i_ino &&
1071                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1072                                 path->slots[0]--;
1073                 }
1074                 check_prev = 0;
1075 next_slot:
1076                 leaf = path->nodes[0];
1077                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1078                         ret = btrfs_next_leaf(root, path);
1079                         if (ret < 0)
1080                                 BUG_ON(1);
1081                         if (ret > 0)
1082                                 break;
1083                         leaf = path->nodes[0];
1084                 }
1085
1086                 nocow = 0;
1087                 disk_bytenr = 0;
1088                 num_bytes = 0;
1089                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1090
1091                 if (found_key.objectid > inode->i_ino ||
1092                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1093                     found_key.offset > end)
1094                         break;
1095
1096                 if (found_key.offset > cur_offset) {
1097                         extent_end = found_key.offset;
1098                         extent_type = 0;
1099                         goto out_check;
1100                 }
1101
1102                 fi = btrfs_item_ptr(leaf, path->slots[0],
1103                                     struct btrfs_file_extent_item);
1104                 extent_type = btrfs_file_extent_type(leaf, fi);
1105
1106                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1107                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1108                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1109                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1110                         extent_end = found_key.offset +
1111                                 btrfs_file_extent_num_bytes(leaf, fi);
1112                         if (extent_end <= start) {
1113                                 path->slots[0]++;
1114                                 goto next_slot;
1115                         }
1116                         if (disk_bytenr == 0)
1117                                 goto out_check;
1118                         if (btrfs_file_extent_compression(leaf, fi) ||
1119                             btrfs_file_extent_encryption(leaf, fi) ||
1120                             btrfs_file_extent_other_encoding(leaf, fi))
1121                                 goto out_check;
1122                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1123                                 goto out_check;
1124                         if (btrfs_extent_readonly(root, disk_bytenr))
1125                                 goto out_check;
1126                         if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1127                                                   found_key.offset -
1128                                                   extent_offset, disk_bytenr))
1129                                 goto out_check;
1130                         disk_bytenr += extent_offset;
1131                         disk_bytenr += cur_offset - found_key.offset;
1132                         num_bytes = min(end + 1, extent_end) - cur_offset;
1133                         /*
1134                          * force cow if csum exists in the range.
1135                          * this ensure that csum for a given extent are
1136                          * either valid or do not exist.
1137                          */
1138                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1139                                 goto out_check;
1140                         nocow = 1;
1141                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1142                         extent_end = found_key.offset +
1143                                 btrfs_file_extent_inline_len(leaf, fi);
1144                         extent_end = ALIGN(extent_end, root->sectorsize);
1145                 } else {
1146                         BUG_ON(1);
1147                 }
1148 out_check:
1149                 if (extent_end <= start) {
1150                         path->slots[0]++;
1151                         goto next_slot;
1152                 }
1153                 if (!nocow) {
1154                         if (cow_start == (u64)-1)
1155                                 cow_start = cur_offset;
1156                         cur_offset = extent_end;
1157                         if (cur_offset > end)
1158                                 break;
1159                         path->slots[0]++;
1160                         goto next_slot;
1161                 }
1162
1163                 btrfs_release_path(root, path);
1164                 if (cow_start != (u64)-1) {
1165                         ret = cow_file_range(inode, locked_page, cow_start,
1166                                         found_key.offset - 1, page_started,
1167                                         nr_written, 1);
1168                         BUG_ON(ret);
1169                         cow_start = (u64)-1;
1170                 }
1171
1172                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1173                         struct extent_map *em;
1174                         struct extent_map_tree *em_tree;
1175                         em_tree = &BTRFS_I(inode)->extent_tree;
1176                         em = alloc_extent_map(GFP_NOFS);
1177                         BUG_ON(!em);
1178                         em->start = cur_offset;
1179                         em->orig_start = em->start;
1180                         em->len = num_bytes;
1181                         em->block_len = num_bytes;
1182                         em->block_start = disk_bytenr;
1183                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1184                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1185                         while (1) {
1186                                 write_lock(&em_tree->lock);
1187                                 ret = add_extent_mapping(em_tree, em);
1188                                 write_unlock(&em_tree->lock);
1189                                 if (ret != -EEXIST) {
1190                                         free_extent_map(em);
1191                                         break;
1192                                 }
1193                                 btrfs_drop_extent_cache(inode, em->start,
1194                                                 em->start + em->len - 1, 0);
1195                         }
1196                         type = BTRFS_ORDERED_PREALLOC;
1197                 } else {
1198                         type = BTRFS_ORDERED_NOCOW;
1199                 }
1200
1201                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1202                                                num_bytes, num_bytes, type);
1203                 BUG_ON(ret);
1204
1205                 if (root->root_key.objectid ==
1206                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1207                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1208                                                       num_bytes);
1209                         BUG_ON(ret);
1210                 }
1211
1212                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1213                                 cur_offset, cur_offset + num_bytes - 1,
1214                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1215                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1216                                 EXTENT_SET_PRIVATE2);
1217                 cur_offset = extent_end;
1218                 if (cur_offset > end)
1219                         break;
1220         }
1221         btrfs_release_path(root, path);
1222
1223         if (cur_offset <= end && cow_start == (u64)-1)
1224                 cow_start = cur_offset;
1225         if (cow_start != (u64)-1) {
1226                 ret = cow_file_range(inode, locked_page, cow_start, end,
1227                                      page_started, nr_written, 1);
1228                 BUG_ON(ret);
1229         }
1230
1231         if (nolock) {
1232                 ret = btrfs_end_transaction_nolock(trans, root);
1233                 BUG_ON(ret);
1234         } else {
1235                 ret = btrfs_end_transaction(trans, root);
1236                 BUG_ON(ret);
1237         }
1238         btrfs_free_path(path);
1239         return 0;
1240 }
1241
1242 /*
1243  * extent_io.c call back to do delayed allocation processing
1244  */
1245 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1246                               u64 start, u64 end, int *page_started,
1247                               unsigned long *nr_written)
1248 {
1249         int ret;
1250         struct btrfs_root *root = BTRFS_I(inode)->root;
1251
1252         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1253                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1254                                          page_started, 1, nr_written);
1255         else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1256                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1257                                          page_started, 0, nr_written);
1258         else if (!btrfs_test_opt(root, COMPRESS) &&
1259                  !(BTRFS_I(inode)->force_compress))
1260                 ret = cow_file_range(inode, locked_page, start, end,
1261                                       page_started, nr_written, 1);
1262         else
1263                 ret = cow_file_range_async(inode, locked_page, start, end,
1264                                            page_started, nr_written);
1265         return ret;
1266 }
1267
1268 static int btrfs_split_extent_hook(struct inode *inode,
1269                                    struct extent_state *orig, u64 split)
1270 {
1271         /* not delalloc, ignore it */
1272         if (!(orig->state & EXTENT_DELALLOC))
1273                 return 0;
1274
1275         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1276         return 0;
1277 }
1278
1279 /*
1280  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1281  * extents so we can keep track of new extents that are just merged onto old
1282  * extents, such as when we are doing sequential writes, so we can properly
1283  * account for the metadata space we'll need.
1284  */
1285 static int btrfs_merge_extent_hook(struct inode *inode,
1286                                    struct extent_state *new,
1287                                    struct extent_state *other)
1288 {
1289         /* not delalloc, ignore it */
1290         if (!(other->state & EXTENT_DELALLOC))
1291                 return 0;
1292
1293         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1294         return 0;
1295 }
1296
1297 /*
1298  * extent_io.c set_bit_hook, used to track delayed allocation
1299  * bytes in this file, and to maintain the list of inodes that
1300  * have pending delalloc work to be done.
1301  */
1302 static int btrfs_set_bit_hook(struct inode *inode,
1303                               struct extent_state *state, int *bits)
1304 {
1305
1306         /*
1307          * set_bit and clear bit hooks normally require _irqsave/restore
1308          * but in this case, we are only testeing for the DELALLOC
1309          * bit, which is only set or cleared with irqs on
1310          */
1311         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1312                 struct btrfs_root *root = BTRFS_I(inode)->root;
1313                 u64 len = state->end + 1 - state->start;
1314                 int do_list = (root->root_key.objectid !=
1315                                BTRFS_ROOT_TREE_OBJECTID);
1316
1317                 if (*bits & EXTENT_FIRST_DELALLOC)
1318                         *bits &= ~EXTENT_FIRST_DELALLOC;
1319                 else
1320                         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1321
1322                 spin_lock(&root->fs_info->delalloc_lock);
1323                 BTRFS_I(inode)->delalloc_bytes += len;
1324                 root->fs_info->delalloc_bytes += len;
1325                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1326                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1327                                       &root->fs_info->delalloc_inodes);
1328                 }
1329                 spin_unlock(&root->fs_info->delalloc_lock);
1330         }
1331         return 0;
1332 }
1333
1334 /*
1335  * extent_io.c clear_bit_hook, see set_bit_hook for why
1336  */
1337 static int btrfs_clear_bit_hook(struct inode *inode,
1338                                 struct extent_state *state, int *bits)
1339 {
1340         /*
1341          * set_bit and clear bit hooks normally require _irqsave/restore
1342          * but in this case, we are only testeing for the DELALLOC
1343          * bit, which is only set or cleared with irqs on
1344          */
1345         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1346                 struct btrfs_root *root = BTRFS_I(inode)->root;
1347                 u64 len = state->end + 1 - state->start;
1348                 int do_list = (root->root_key.objectid !=
1349                                BTRFS_ROOT_TREE_OBJECTID);
1350
1351                 if (*bits & EXTENT_FIRST_DELALLOC)
1352                         *bits &= ~EXTENT_FIRST_DELALLOC;
1353                 else if (!(*bits & EXTENT_DO_ACCOUNTING))
1354                         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1355
1356                 if (*bits & EXTENT_DO_ACCOUNTING)
1357                         btrfs_delalloc_release_metadata(inode, len);
1358
1359                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1360                     && do_list)
1361                         btrfs_free_reserved_data_space(inode, len);
1362
1363                 spin_lock(&root->fs_info->delalloc_lock);
1364                 root->fs_info->delalloc_bytes -= len;
1365                 BTRFS_I(inode)->delalloc_bytes -= len;
1366
1367                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1368                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1369                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1370                 }
1371                 spin_unlock(&root->fs_info->delalloc_lock);
1372         }
1373         return 0;
1374 }
1375
1376 /*
1377  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1378  * we don't create bios that span stripes or chunks
1379  */
1380 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1381                          size_t size, struct bio *bio,
1382                          unsigned long bio_flags)
1383 {
1384         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1385         struct btrfs_mapping_tree *map_tree;
1386         u64 logical = (u64)bio->bi_sector << 9;
1387         u64 length = 0;
1388         u64 map_length;
1389         int ret;
1390
1391         if (bio_flags & EXTENT_BIO_COMPRESSED)
1392                 return 0;
1393
1394         length = bio->bi_size;
1395         map_tree = &root->fs_info->mapping_tree;
1396         map_length = length;
1397         ret = btrfs_map_block(map_tree, READ, logical,
1398                               &map_length, NULL, 0);
1399
1400         if (map_length < length + size)
1401                 return 1;
1402         return ret;
1403 }
1404
1405 /*
1406  * in order to insert checksums into the metadata in large chunks,
1407  * we wait until bio submission time.   All the pages in the bio are
1408  * checksummed and sums are attached onto the ordered extent record.
1409  *
1410  * At IO completion time the cums attached on the ordered extent record
1411  * are inserted into the btree
1412  */
1413 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1414                                     struct bio *bio, int mirror_num,
1415                                     unsigned long bio_flags,
1416                                     u64 bio_offset)
1417 {
1418         struct btrfs_root *root = BTRFS_I(inode)->root;
1419         int ret = 0;
1420
1421         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1422         BUG_ON(ret);
1423         return 0;
1424 }
1425
1426 /*
1427  * in order to insert checksums into the metadata in large chunks,
1428  * we wait until bio submission time.   All the pages in the bio are
1429  * checksummed and sums are attached onto the ordered extent record.
1430  *
1431  * At IO completion time the cums attached on the ordered extent record
1432  * are inserted into the btree
1433  */
1434 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1435                           int mirror_num, unsigned long bio_flags,
1436                           u64 bio_offset)
1437 {
1438         struct btrfs_root *root = BTRFS_I(inode)->root;
1439         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1440 }
1441
1442 /*
1443  * extent_io.c submission hook. This does the right thing for csum calculation
1444  * on write, or reading the csums from the tree before a read
1445  */
1446 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1447                           int mirror_num, unsigned long bio_flags,
1448                           u64 bio_offset)
1449 {
1450         struct btrfs_root *root = BTRFS_I(inode)->root;
1451         int ret = 0;
1452         int skip_sum;
1453
1454         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1455
1456         if (root == root->fs_info->tree_root)
1457                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1458         else
1459                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1460         BUG_ON(ret);
1461
1462         if (!(rw & REQ_WRITE)) {
1463                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1464                         return btrfs_submit_compressed_read(inode, bio,
1465                                                     mirror_num, bio_flags);
1466                 } else if (!skip_sum)
1467                         btrfs_lookup_bio_sums(root, inode, bio, NULL);
1468                 goto mapit;
1469         } else if (!skip_sum) {
1470                 /* csum items have already been cloned */
1471                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1472                         goto mapit;
1473                 /* we're doing a write, do the async checksumming */
1474                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1475                                    inode, rw, bio, mirror_num,
1476                                    bio_flags, bio_offset,
1477                                    __btrfs_submit_bio_start,
1478                                    __btrfs_submit_bio_done);
1479         }
1480
1481 mapit:
1482         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1483 }
1484
1485 /*
1486  * given a list of ordered sums record them in the inode.  This happens
1487  * at IO completion time based on sums calculated at bio submission time.
1488  */
1489 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1490                              struct inode *inode, u64 file_offset,
1491                              struct list_head *list)
1492 {
1493         struct btrfs_ordered_sum *sum;
1494
1495         btrfs_set_trans_block_group(trans, inode);
1496
1497         list_for_each_entry(sum, list, list) {
1498                 btrfs_csum_file_blocks(trans,
1499                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1500         }
1501         return 0;
1502 }
1503
1504 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1505                               struct extent_state **cached_state)
1506 {
1507         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1508                 WARN_ON(1);
1509         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1510                                    cached_state, GFP_NOFS);
1511 }
1512
1513 /* see btrfs_writepage_start_hook for details on why this is required */
1514 struct btrfs_writepage_fixup {
1515         struct page *page;
1516         struct btrfs_work work;
1517 };
1518
1519 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1520 {
1521         struct btrfs_writepage_fixup *fixup;
1522         struct btrfs_ordered_extent *ordered;
1523         struct extent_state *cached_state = NULL;
1524         struct page *page;
1525         struct inode *inode;
1526         u64 page_start;
1527         u64 page_end;
1528
1529         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1530         page = fixup->page;
1531 again:
1532         lock_page(page);
1533         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1534                 ClearPageChecked(page);
1535                 goto out_page;
1536         }
1537
1538         inode = page->mapping->host;
1539         page_start = page_offset(page);
1540         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1541
1542         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1543                          &cached_state, GFP_NOFS);
1544
1545         /* already ordered? We're done */
1546         if (PagePrivate2(page))
1547                 goto out;
1548
1549         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1550         if (ordered) {
1551                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1552                                      page_end, &cached_state, GFP_NOFS);
1553                 unlock_page(page);
1554                 btrfs_start_ordered_extent(inode, ordered, 1);
1555                 goto again;
1556         }
1557
1558         BUG();
1559         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1560         ClearPageChecked(page);
1561 out:
1562         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1563                              &cached_state, GFP_NOFS);
1564 out_page:
1565         unlock_page(page);
1566         page_cache_release(page);
1567         kfree(fixup);
1568 }
1569
1570 /*
1571  * There are a few paths in the higher layers of the kernel that directly
1572  * set the page dirty bit without asking the filesystem if it is a
1573  * good idea.  This causes problems because we want to make sure COW
1574  * properly happens and the data=ordered rules are followed.
1575  *
1576  * In our case any range that doesn't have the ORDERED bit set
1577  * hasn't been properly setup for IO.  We kick off an async process
1578  * to fix it up.  The async helper will wait for ordered extents, set
1579  * the delalloc bit and make it safe to write the page.
1580  */
1581 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1582 {
1583         struct inode *inode = page->mapping->host;
1584         struct btrfs_writepage_fixup *fixup;
1585         struct btrfs_root *root = BTRFS_I(inode)->root;
1586
1587         /* this page is properly in the ordered list */
1588         if (TestClearPagePrivate2(page))
1589                 return 0;
1590
1591         if (PageChecked(page))
1592                 return -EAGAIN;
1593
1594         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1595         if (!fixup)
1596                 return -EAGAIN;
1597
1598         SetPageChecked(page);
1599         page_cache_get(page);
1600         fixup->work.func = btrfs_writepage_fixup_worker;
1601         fixup->page = page;
1602         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1603         return -EAGAIN;
1604 }
1605
1606 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1607                                        struct inode *inode, u64 file_pos,
1608                                        u64 disk_bytenr, u64 disk_num_bytes,
1609                                        u64 num_bytes, u64 ram_bytes,
1610                                        u8 compression, u8 encryption,
1611                                        u16 other_encoding, int extent_type)
1612 {
1613         struct btrfs_root *root = BTRFS_I(inode)->root;
1614         struct btrfs_file_extent_item *fi;
1615         struct btrfs_path *path;
1616         struct extent_buffer *leaf;
1617         struct btrfs_key ins;
1618         u64 hint;
1619         int ret;
1620
1621         path = btrfs_alloc_path();
1622         BUG_ON(!path);
1623
1624         path->leave_spinning = 1;
1625
1626         /*
1627          * we may be replacing one extent in the tree with another.
1628          * The new extent is pinned in the extent map, and we don't want
1629          * to drop it from the cache until it is completely in the btree.
1630          *
1631          * So, tell btrfs_drop_extents to leave this extent in the cache.
1632          * the caller is expected to unpin it and allow it to be merged
1633          * with the others.
1634          */
1635         ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1636                                  &hint, 0);
1637         BUG_ON(ret);
1638
1639         ins.objectid = inode->i_ino;
1640         ins.offset = file_pos;
1641         ins.type = BTRFS_EXTENT_DATA_KEY;
1642         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1643         BUG_ON(ret);
1644         leaf = path->nodes[0];
1645         fi = btrfs_item_ptr(leaf, path->slots[0],
1646                             struct btrfs_file_extent_item);
1647         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1648         btrfs_set_file_extent_type(leaf, fi, extent_type);
1649         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1650         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1651         btrfs_set_file_extent_offset(leaf, fi, 0);
1652         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1653         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1654         btrfs_set_file_extent_compression(leaf, fi, compression);
1655         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1656         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1657
1658         btrfs_unlock_up_safe(path, 1);
1659         btrfs_set_lock_blocking(leaf);
1660
1661         btrfs_mark_buffer_dirty(leaf);
1662
1663         inode_add_bytes(inode, num_bytes);
1664
1665         ins.objectid = disk_bytenr;
1666         ins.offset = disk_num_bytes;
1667         ins.type = BTRFS_EXTENT_ITEM_KEY;
1668         ret = btrfs_alloc_reserved_file_extent(trans, root,
1669                                         root->root_key.objectid,
1670                                         inode->i_ino, file_pos, &ins);
1671         BUG_ON(ret);
1672         btrfs_free_path(path);
1673
1674         return 0;
1675 }
1676
1677 /*
1678  * helper function for btrfs_finish_ordered_io, this
1679  * just reads in some of the csum leaves to prime them into ram
1680  * before we start the transaction.  It limits the amount of btree
1681  * reads required while inside the transaction.
1682  */
1683 /* as ordered data IO finishes, this gets called so we can finish
1684  * an ordered extent if the range of bytes in the file it covers are
1685  * fully written.
1686  */
1687 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1688 {
1689         struct btrfs_root *root = BTRFS_I(inode)->root;
1690         struct btrfs_trans_handle *trans = NULL;
1691         struct btrfs_ordered_extent *ordered_extent = NULL;
1692         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1693         struct extent_state *cached_state = NULL;
1694         int compress_type = 0;
1695         int ret;
1696         bool nolock = false;
1697
1698         ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1699                                              end - start + 1);
1700         if (!ret)
1701                 return 0;
1702         BUG_ON(!ordered_extent);
1703
1704         nolock = (root == root->fs_info->tree_root);
1705
1706         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1707                 BUG_ON(!list_empty(&ordered_extent->list));
1708                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1709                 if (!ret) {
1710                         if (nolock)
1711                                 trans = btrfs_join_transaction_nolock(root, 1);
1712                         else
1713                                 trans = btrfs_join_transaction(root, 1);
1714                         BUG_ON(IS_ERR(trans));
1715                         btrfs_set_trans_block_group(trans, inode);
1716                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1717                         ret = btrfs_update_inode(trans, root, inode);
1718                         BUG_ON(ret);
1719                 }
1720                 goto out;
1721         }
1722
1723         lock_extent_bits(io_tree, ordered_extent->file_offset,
1724                          ordered_extent->file_offset + ordered_extent->len - 1,
1725                          0, &cached_state, GFP_NOFS);
1726
1727         if (nolock)
1728                 trans = btrfs_join_transaction_nolock(root, 1);
1729         else
1730                 trans = btrfs_join_transaction(root, 1);
1731         BUG_ON(IS_ERR(trans));
1732         btrfs_set_trans_block_group(trans, inode);
1733         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1734
1735         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1736                 compress_type = ordered_extent->compress_type;
1737         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1738                 BUG_ON(compress_type);
1739                 ret = btrfs_mark_extent_written(trans, inode,
1740                                                 ordered_extent->file_offset,
1741                                                 ordered_extent->file_offset +
1742                                                 ordered_extent->len);
1743                 BUG_ON(ret);
1744         } else {
1745                 BUG_ON(root == root->fs_info->tree_root);
1746                 ret = insert_reserved_file_extent(trans, inode,
1747                                                 ordered_extent->file_offset,
1748                                                 ordered_extent->start,
1749                                                 ordered_extent->disk_len,
1750                                                 ordered_extent->len,
1751                                                 ordered_extent->len,
1752                                                 compress_type, 0, 0,
1753                                                 BTRFS_FILE_EXTENT_REG);
1754                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1755                                    ordered_extent->file_offset,
1756                                    ordered_extent->len);
1757                 BUG_ON(ret);
1758         }
1759         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1760                              ordered_extent->file_offset +
1761                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1762
1763         add_pending_csums(trans, inode, ordered_extent->file_offset,
1764                           &ordered_extent->list);
1765
1766         btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1767         ret = btrfs_update_inode(trans, root, inode);
1768         BUG_ON(ret);
1769 out:
1770         if (nolock) {
1771                 if (trans)
1772                         btrfs_end_transaction_nolock(trans, root);
1773         } else {
1774                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1775                 if (trans)
1776                         btrfs_end_transaction(trans, root);
1777         }
1778
1779         /* once for us */
1780         btrfs_put_ordered_extent(ordered_extent);
1781         /* once for the tree */
1782         btrfs_put_ordered_extent(ordered_extent);
1783
1784         return 0;
1785 }
1786
1787 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1788                                 struct extent_state *state, int uptodate)
1789 {
1790         ClearPagePrivate2(page);
1791         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1792 }
1793
1794 /*
1795  * When IO fails, either with EIO or csum verification fails, we
1796  * try other mirrors that might have a good copy of the data.  This
1797  * io_failure_record is used to record state as we go through all the
1798  * mirrors.  If another mirror has good data, the page is set up to date
1799  * and things continue.  If a good mirror can't be found, the original
1800  * bio end_io callback is called to indicate things have failed.
1801  */
1802 struct io_failure_record {
1803         struct page *page;
1804         u64 start;
1805         u64 len;
1806         u64 logical;
1807         unsigned long bio_flags;
1808         int last_mirror;
1809 };
1810
1811 static int btrfs_io_failed_hook(struct bio *failed_bio,
1812                          struct page *page, u64 start, u64 end,
1813                          struct extent_state *state)
1814 {
1815         struct io_failure_record *failrec = NULL;
1816         u64 private;
1817         struct extent_map *em;
1818         struct inode *inode = page->mapping->host;
1819         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1820         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1821         struct bio *bio;
1822         int num_copies;
1823         int ret;
1824         int rw;
1825         u64 logical;
1826
1827         ret = get_state_private(failure_tree, start, &private);
1828         if (ret) {
1829                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1830                 if (!failrec)
1831                         return -ENOMEM;
1832                 failrec->start = start;
1833                 failrec->len = end - start + 1;
1834                 failrec->last_mirror = 0;
1835                 failrec->bio_flags = 0;
1836
1837                 read_lock(&em_tree->lock);
1838                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1839                 if (em->start > start || em->start + em->len < start) {
1840                         free_extent_map(em);
1841                         em = NULL;
1842                 }
1843                 read_unlock(&em_tree->lock);
1844
1845                 if (!em || IS_ERR(em)) {
1846                         kfree(failrec);
1847                         return -EIO;
1848                 }
1849                 logical = start - em->start;
1850                 logical = em->block_start + logical;
1851                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1852                         logical = em->block_start;
1853                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1854                         extent_set_compress_type(&failrec->bio_flags,
1855                                                  em->compress_type);
1856                 }
1857                 failrec->logical = logical;
1858                 free_extent_map(em);
1859                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1860                                 EXTENT_DIRTY, GFP_NOFS);
1861                 set_state_private(failure_tree, start,
1862                                  (u64)(unsigned long)failrec);
1863         } else {
1864                 failrec = (struct io_failure_record *)(unsigned long)private;
1865         }
1866         num_copies = btrfs_num_copies(
1867                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1868                               failrec->logical, failrec->len);
1869         failrec->last_mirror++;
1870         if (!state) {
1871                 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1872                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1873                                                     failrec->start,
1874                                                     EXTENT_LOCKED);
1875                 if (state && state->start != failrec->start)
1876                         state = NULL;
1877                 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1878         }
1879         if (!state || failrec->last_mirror > num_copies) {
1880                 set_state_private(failure_tree, failrec->start, 0);
1881                 clear_extent_bits(failure_tree, failrec->start,
1882                                   failrec->start + failrec->len - 1,
1883                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1884                 kfree(failrec);
1885                 return -EIO;
1886         }
1887         bio = bio_alloc(GFP_NOFS, 1);
1888         bio->bi_private = state;
1889         bio->bi_end_io = failed_bio->bi_end_io;
1890         bio->bi_sector = failrec->logical >> 9;
1891         bio->bi_bdev = failed_bio->bi_bdev;
1892         bio->bi_size = 0;
1893
1894         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1895         if (failed_bio->bi_rw & REQ_WRITE)
1896                 rw = WRITE;
1897         else
1898                 rw = READ;
1899
1900         BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1901                                                       failrec->last_mirror,
1902                                                       failrec->bio_flags, 0);
1903         return 0;
1904 }
1905
1906 /*
1907  * each time an IO finishes, we do a fast check in the IO failure tree
1908  * to see if we need to process or clean up an io_failure_record
1909  */
1910 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1911 {
1912         u64 private;
1913         u64 private_failure;
1914         struct io_failure_record *failure;
1915         int ret;
1916
1917         private = 0;
1918         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1919                              (u64)-1, 1, EXTENT_DIRTY, 0)) {
1920                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1921                                         start, &private_failure);
1922                 if (ret == 0) {
1923                         failure = (struct io_failure_record *)(unsigned long)
1924                                    private_failure;
1925                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1926                                           failure->start, 0);
1927                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1928                                           failure->start,
1929                                           failure->start + failure->len - 1,
1930                                           EXTENT_DIRTY | EXTENT_LOCKED,
1931                                           GFP_NOFS);
1932                         kfree(failure);
1933                 }
1934         }
1935         return 0;
1936 }
1937
1938 /*
1939  * when reads are done, we need to check csums to verify the data is correct
1940  * if there's a match, we allow the bio to finish.  If not, we go through
1941  * the io_failure_record routines to find good copies
1942  */
1943 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1944                                struct extent_state *state)
1945 {
1946         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1947         struct inode *inode = page->mapping->host;
1948         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1949         char *kaddr;
1950         u64 private = ~(u32)0;
1951         int ret;
1952         struct btrfs_root *root = BTRFS_I(inode)->root;
1953         u32 csum = ~(u32)0;
1954
1955         if (PageChecked(page)) {
1956                 ClearPageChecked(page);
1957                 goto good;
1958         }
1959
1960         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1961                 return 0;
1962
1963         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1964             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1965                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1966                                   GFP_NOFS);
1967                 return 0;
1968         }
1969
1970         if (state && state->start == start) {
1971                 private = state->private;
1972                 ret = 0;
1973         } else {
1974                 ret = get_state_private(io_tree, start, &private);
1975         }
1976         kaddr = kmap_atomic(page, KM_USER0);
1977         if (ret)
1978                 goto zeroit;
1979
1980         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
1981         btrfs_csum_final(csum, (char *)&csum);
1982         if (csum != private)
1983                 goto zeroit;
1984
1985         kunmap_atomic(kaddr, KM_USER0);
1986 good:
1987         /* if the io failure tree for this inode is non-empty,
1988          * check to see if we've recovered from a failed IO
1989          */
1990         btrfs_clean_io_failures(inode, start);
1991         return 0;
1992
1993 zeroit:
1994         if (printk_ratelimit()) {
1995                 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
1996                        "private %llu\n", page->mapping->host->i_ino,
1997                        (unsigned long long)start, csum,
1998                        (unsigned long long)private);
1999         }
2000         memset(kaddr + offset, 1, end - start + 1);
2001         flush_dcache_page(page);
2002         kunmap_atomic(kaddr, KM_USER0);
2003         if (private == 0)
2004                 return 0;
2005         return -EIO;
2006 }
2007
2008 struct delayed_iput {
2009         struct list_head list;
2010         struct inode *inode;
2011 };
2012
2013 void btrfs_add_delayed_iput(struct inode *inode)
2014 {
2015         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2016         struct delayed_iput *delayed;
2017
2018         if (atomic_add_unless(&inode->i_count, -1, 1))
2019                 return;
2020
2021         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2022         delayed->inode = inode;
2023
2024         spin_lock(&fs_info->delayed_iput_lock);
2025         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2026         spin_unlock(&fs_info->delayed_iput_lock);
2027 }
2028
2029 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2030 {
2031         LIST_HEAD(list);
2032         struct btrfs_fs_info *fs_info = root->fs_info;
2033         struct delayed_iput *delayed;
2034         int empty;
2035
2036         spin_lock(&fs_info->delayed_iput_lock);
2037         empty = list_empty(&fs_info->delayed_iputs);
2038         spin_unlock(&fs_info->delayed_iput_lock);
2039         if (empty)
2040                 return;
2041
2042         down_read(&root->fs_info->cleanup_work_sem);
2043         spin_lock(&fs_info->delayed_iput_lock);
2044         list_splice_init(&fs_info->delayed_iputs, &list);
2045         spin_unlock(&fs_info->delayed_iput_lock);
2046
2047         while (!list_empty(&list)) {
2048                 delayed = list_entry(list.next, struct delayed_iput, list);
2049                 list_del(&delayed->list);
2050                 iput(delayed->inode);
2051                 kfree(delayed);
2052         }
2053         up_read(&root->fs_info->cleanup_work_sem);
2054 }
2055
2056 /*
2057  * calculate extra metadata reservation when snapshotting a subvolume
2058  * contains orphan files.
2059  */
2060 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans,
2061                                 struct btrfs_pending_snapshot *pending,
2062                                 u64 *bytes_to_reserve)
2063 {
2064         struct btrfs_root *root;
2065         struct btrfs_block_rsv *block_rsv;
2066         u64 num_bytes;
2067         int index;
2068
2069         root = pending->root;
2070         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2071                 return;
2072
2073         block_rsv = root->orphan_block_rsv;
2074
2075         /* orphan block reservation for the snapshot */
2076         num_bytes = block_rsv->size;
2077
2078         /*
2079          * after the snapshot is created, COWing tree blocks may use more
2080          * space than it frees. So we should make sure there is enough
2081          * reserved space.
2082          */
2083         index = trans->transid & 0x1;
2084         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2085                 num_bytes += block_rsv->size -
2086                              (block_rsv->reserved + block_rsv->freed[index]);
2087         }
2088
2089         *bytes_to_reserve += num_bytes;
2090 }
2091
2092 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans,
2093                                 struct btrfs_pending_snapshot *pending)
2094 {
2095         struct btrfs_root *root = pending->root;
2096         struct btrfs_root *snap = pending->snap;
2097         struct btrfs_block_rsv *block_rsv;
2098         u64 num_bytes;
2099         int index;
2100         int ret;
2101
2102         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2103                 return;
2104
2105         /* refill source subvolume's orphan block reservation */
2106         block_rsv = root->orphan_block_rsv;
2107         index = trans->transid & 0x1;
2108         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2109                 num_bytes = block_rsv->size -
2110                             (block_rsv->reserved + block_rsv->freed[index]);
2111                 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2112                                               root->orphan_block_rsv,
2113                                               num_bytes);
2114                 BUG_ON(ret);
2115         }
2116
2117         /* setup orphan block reservation for the snapshot */
2118         block_rsv = btrfs_alloc_block_rsv(snap);
2119         BUG_ON(!block_rsv);
2120
2121         btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2122         snap->orphan_block_rsv = block_rsv;
2123
2124         num_bytes = root->orphan_block_rsv->size;
2125         ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2126                                       block_rsv, num_bytes);
2127         BUG_ON(ret);
2128
2129 #if 0
2130         /* insert orphan item for the snapshot */
2131         WARN_ON(!root->orphan_item_inserted);
2132         ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2133                                        snap->root_key.objectid);
2134         BUG_ON(ret);
2135         snap->orphan_item_inserted = 1;
2136 #endif
2137 }
2138
2139 enum btrfs_orphan_cleanup_state {
2140         ORPHAN_CLEANUP_STARTED  = 1,
2141         ORPHAN_CLEANUP_DONE     = 2,
2142 };
2143
2144 /*
2145  * This is called in transaction commmit time. If there are no orphan
2146  * files in the subvolume, it removes orphan item and frees block_rsv
2147  * structure.
2148  */
2149 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2150                               struct btrfs_root *root)
2151 {
2152         int ret;
2153
2154         if (!list_empty(&root->orphan_list) ||
2155             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2156                 return;
2157
2158         if (root->orphan_item_inserted &&
2159             btrfs_root_refs(&root->root_item) > 0) {
2160                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2161                                             root->root_key.objectid);
2162                 BUG_ON(ret);
2163                 root->orphan_item_inserted = 0;
2164         }
2165
2166         if (root->orphan_block_rsv) {
2167                 WARN_ON(root->orphan_block_rsv->size > 0);
2168                 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2169                 root->orphan_block_rsv = NULL;
2170         }
2171 }
2172
2173 /*
2174  * This creates an orphan entry for the given inode in case something goes
2175  * wrong in the middle of an unlink/truncate.
2176  *
2177  * NOTE: caller of this function should reserve 5 units of metadata for
2178  *       this function.
2179  */
2180 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2181 {
2182         struct btrfs_root *root = BTRFS_I(inode)->root;
2183         struct btrfs_block_rsv *block_rsv = NULL;
2184         int reserve = 0;
2185         int insert = 0;
2186         int ret;
2187
2188         if (!root->orphan_block_rsv) {
2189                 block_rsv = btrfs_alloc_block_rsv(root);
2190                 BUG_ON(!block_rsv);
2191         }
2192
2193         spin_lock(&root->orphan_lock);
2194         if (!root->orphan_block_rsv) {
2195                 root->orphan_block_rsv = block_rsv;
2196         } else if (block_rsv) {
2197                 btrfs_free_block_rsv(root, block_rsv);
2198                 block_rsv = NULL;
2199         }
2200
2201         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2202                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2203 #if 0
2204                 /*
2205                  * For proper ENOSPC handling, we should do orphan
2206                  * cleanup when mounting. But this introduces backward
2207                  * compatibility issue.
2208                  */
2209                 if (!xchg(&root->orphan_item_inserted, 1))
2210                         insert = 2;
2211                 else
2212                         insert = 1;
2213 #endif
2214                 insert = 1;
2215         } else {
2216                 WARN_ON(!BTRFS_I(inode)->orphan_meta_reserved);
2217         }
2218
2219         if (!BTRFS_I(inode)->orphan_meta_reserved) {
2220                 BTRFS_I(inode)->orphan_meta_reserved = 1;
2221                 reserve = 1;
2222         }
2223         spin_unlock(&root->orphan_lock);
2224
2225         if (block_rsv)
2226                 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2227
2228         /* grab metadata reservation from transaction handle */
2229         if (reserve) {
2230                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2231                 BUG_ON(ret);
2232         }
2233
2234         /* insert an orphan item to track this unlinked/truncated file */
2235         if (insert >= 1) {
2236                 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
2237                 BUG_ON(ret);
2238         }
2239
2240         /* insert an orphan item to track subvolume contains orphan files */
2241         if (insert >= 2) {
2242                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2243                                                root->root_key.objectid);
2244                 BUG_ON(ret);
2245         }
2246         return 0;
2247 }
2248
2249 /*
2250  * We have done the truncate/delete so we can go ahead and remove the orphan
2251  * item for this particular inode.
2252  */
2253 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2254 {
2255         struct btrfs_root *root = BTRFS_I(inode)->root;
2256         int delete_item = 0;
2257         int release_rsv = 0;
2258         int ret = 0;
2259
2260         spin_lock(&root->orphan_lock);
2261         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2262                 list_del_init(&BTRFS_I(inode)->i_orphan);
2263                 delete_item = 1;
2264         }
2265
2266         if (BTRFS_I(inode)->orphan_meta_reserved) {
2267                 BTRFS_I(inode)->orphan_meta_reserved = 0;
2268                 release_rsv = 1;
2269         }
2270         spin_unlock(&root->orphan_lock);
2271
2272         if (trans && delete_item) {
2273                 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
2274                 BUG_ON(ret);
2275         }
2276
2277         if (release_rsv)
2278                 btrfs_orphan_release_metadata(inode);
2279
2280         return 0;
2281 }
2282
2283 /*
2284  * this cleans up any orphans that may be left on the list from the last use
2285  * of this root.
2286  */
2287 int btrfs_orphan_cleanup(struct btrfs_root *root)
2288 {
2289         struct btrfs_path *path;
2290         struct extent_buffer *leaf;
2291         struct btrfs_key key, found_key;
2292         struct btrfs_trans_handle *trans;
2293         struct inode *inode;
2294         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2295
2296         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2297                 return 0;
2298
2299         path = btrfs_alloc_path();
2300         if (!path) {
2301                 ret = -ENOMEM;
2302                 goto out;
2303         }
2304         path->reada = -1;
2305
2306         key.objectid = BTRFS_ORPHAN_OBJECTID;
2307         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2308         key.offset = (u64)-1;
2309
2310         while (1) {
2311                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2312                 if (ret < 0)
2313                         goto out;
2314
2315                 /*
2316                  * if ret == 0 means we found what we were searching for, which
2317                  * is weird, but possible, so only screw with path if we didnt
2318                  * find the key and see if we have stuff that matches
2319                  */
2320                 if (ret > 0) {
2321                         ret = 0;
2322                         if (path->slots[0] == 0)
2323                                 break;
2324                         path->slots[0]--;
2325                 }
2326
2327                 /* pull out the item */
2328                 leaf = path->nodes[0];
2329                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2330
2331                 /* make sure the item matches what we want */
2332                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2333                         break;
2334                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2335                         break;
2336
2337                 /* release the path since we're done with it */
2338                 btrfs_release_path(root, path);
2339
2340                 /*
2341                  * this is where we are basically btrfs_lookup, without the
2342                  * crossing root thing.  we store the inode number in the
2343                  * offset of the orphan item.
2344                  */
2345                 found_key.objectid = found_key.offset;
2346                 found_key.type = BTRFS_INODE_ITEM_KEY;
2347                 found_key.offset = 0;
2348                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2349                 if (IS_ERR(inode)) {
2350                         ret = PTR_ERR(inode);
2351                         goto out;
2352                 }
2353
2354                 /*
2355                  * add this inode to the orphan list so btrfs_orphan_del does
2356                  * the proper thing when we hit it
2357                  */
2358                 spin_lock(&root->orphan_lock);
2359                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2360                 spin_unlock(&root->orphan_lock);
2361
2362                 /*
2363                  * if this is a bad inode, means we actually succeeded in
2364                  * removing the inode, but not the orphan record, which means
2365                  * we need to manually delete the orphan since iput will just
2366                  * do a destroy_inode
2367                  */
2368                 if (is_bad_inode(inode)) {
2369                         trans = btrfs_start_transaction(root, 0);
2370                         if (IS_ERR(trans)) {
2371                                 ret = PTR_ERR(trans);
2372                                 goto out;
2373                         }
2374                         btrfs_orphan_del(trans, inode);
2375                         btrfs_end_transaction(trans, root);
2376                         iput(inode);
2377                         continue;
2378                 }
2379
2380                 /* if we have links, this was a truncate, lets do that */
2381                 if (inode->i_nlink) {
2382                         if (!S_ISREG(inode->i_mode)) {
2383                                 WARN_ON(1);
2384                                 iput(inode);
2385                                 continue;
2386                         }
2387                         nr_truncate++;
2388                         ret = btrfs_truncate(inode);
2389                 } else {
2390                         nr_unlink++;
2391                 }
2392
2393                 /* this will do delete_inode and everything for us */
2394                 iput(inode);
2395                 if (ret)
2396                         goto out;
2397         }
2398         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2399
2400         if (root->orphan_block_rsv)
2401                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2402                                         (u64)-1);
2403
2404         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2405                 trans = btrfs_join_transaction(root, 1);
2406                 if (!IS_ERR(trans))
2407                         btrfs_end_transaction(trans, root);
2408         }
2409
2410         if (nr_unlink)
2411                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2412         if (nr_truncate)
2413                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2414
2415 out:
2416         if (ret)
2417                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2418         btrfs_free_path(path);
2419         return ret;
2420 }
2421
2422 /*
2423  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2424  * don't find any xattrs, we know there can't be any acls.
2425  *
2426  * slot is the slot the inode is in, objectid is the objectid of the inode
2427  */
2428 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2429                                           int slot, u64 objectid)
2430 {
2431         u32 nritems = btrfs_header_nritems(leaf);
2432         struct btrfs_key found_key;
2433         int scanned = 0;
2434
2435         slot++;
2436         while (slot < nritems) {
2437                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2438
2439                 /* we found a different objectid, there must not be acls */
2440                 if (found_key.objectid != objectid)
2441                         return 0;
2442
2443                 /* we found an xattr, assume we've got an acl */
2444                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2445                         return 1;
2446
2447                 /*
2448                  * we found a key greater than an xattr key, there can't
2449                  * be any acls later on
2450                  */
2451                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2452                         return 0;
2453
2454                 slot++;
2455                 scanned++;
2456
2457                 /*
2458                  * it goes inode, inode backrefs, xattrs, extents,
2459                  * so if there are a ton of hard links to an inode there can
2460                  * be a lot of backrefs.  Don't waste time searching too hard,
2461                  * this is just an optimization
2462                  */
2463                 if (scanned >= 8)
2464                         break;
2465         }
2466         /* we hit the end of the leaf before we found an xattr or
2467          * something larger than an xattr.  We have to assume the inode
2468          * has acls
2469          */
2470         return 1;
2471 }
2472
2473 /*
2474  * read an inode from the btree into the in-memory inode
2475  */
2476 static void btrfs_read_locked_inode(struct inode *inode)
2477 {
2478         struct btrfs_path *path;
2479         struct extent_buffer *leaf;
2480         struct btrfs_inode_item *inode_item;
2481         struct btrfs_timespec *tspec;
2482         struct btrfs_root *root = BTRFS_I(inode)->root;
2483         struct btrfs_key location;
2484         int maybe_acls;
2485         u64 alloc_group_block;
2486         u32 rdev;
2487         int ret;
2488
2489         path = btrfs_alloc_path();
2490         BUG_ON(!path);
2491         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2492
2493         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2494         if (ret)
2495                 goto make_bad;
2496
2497         leaf = path->nodes[0];
2498         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2499                                     struct btrfs_inode_item);
2500
2501         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2502         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2503         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2504         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2505         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2506
2507         tspec = btrfs_inode_atime(inode_item);
2508         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2509         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2510
2511         tspec = btrfs_inode_mtime(inode_item);
2512         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2513         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2514
2515         tspec = btrfs_inode_ctime(inode_item);
2516         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2517         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2518
2519         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2520         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2521         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2522         inode->i_generation = BTRFS_I(inode)->generation;
2523         inode->i_rdev = 0;
2524         rdev = btrfs_inode_rdev(leaf, inode_item);
2525
2526         BTRFS_I(inode)->index_cnt = (u64)-1;
2527         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2528
2529         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2530
2531         /*
2532          * try to precache a NULL acl entry for files that don't have
2533          * any xattrs or acls
2534          */
2535         maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2536         if (!maybe_acls)
2537                 cache_no_acl(inode);
2538
2539         BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2540                                                 alloc_group_block, 0);
2541         btrfs_free_path(path);
2542         inode_item = NULL;
2543
2544         switch (inode->i_mode & S_IFMT) {
2545         case S_IFREG:
2546                 inode->i_mapping->a_ops = &btrfs_aops;
2547                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2548                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2549                 inode->i_fop = &btrfs_file_operations;
2550                 inode->i_op = &btrfs_file_inode_operations;
2551                 break;
2552         case S_IFDIR:
2553                 inode->i_fop = &btrfs_dir_file_operations;
2554                 if (root == root->fs_info->tree_root)
2555                         inode->i_op = &btrfs_dir_ro_inode_operations;
2556                 else
2557                         inode->i_op = &btrfs_dir_inode_operations;
2558                 break;
2559         case S_IFLNK:
2560                 inode->i_op = &btrfs_symlink_inode_operations;
2561                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2562                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2563                 break;
2564         default:
2565                 inode->i_op = &btrfs_special_inode_operations;
2566                 init_special_inode(inode, inode->i_mode, rdev);
2567                 break;
2568         }
2569
2570         btrfs_update_iflags(inode);
2571         return;
2572
2573 make_bad:
2574         btrfs_free_path(path);
2575         make_bad_inode(inode);
2576 }
2577
2578 /*
2579  * given a leaf and an inode, copy the inode fields into the leaf
2580  */
2581 static void fill_inode_item(struct btrfs_trans_handle *trans,
2582                             struct extent_buffer *leaf,
2583                             struct btrfs_inode_item *item,
2584                             struct inode *inode)
2585 {
2586         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2587         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2588         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2589         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2590         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2591
2592         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2593                                inode->i_atime.tv_sec);
2594         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2595                                 inode->i_atime.tv_nsec);
2596
2597         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2598                                inode->i_mtime.tv_sec);
2599         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2600                                 inode->i_mtime.tv_nsec);
2601
2602         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2603                                inode->i_ctime.tv_sec);
2604         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2605                                 inode->i_ctime.tv_nsec);
2606
2607         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2608         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2609         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2610         btrfs_set_inode_transid(leaf, item, trans->transid);
2611         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2612         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2613         btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2614 }
2615
2616 /*
2617  * copy everything in the in-memory inode into the btree.
2618  */
2619 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2620                                 struct btrfs_root *root, struct inode *inode)
2621 {
2622         struct btrfs_inode_item *inode_item;
2623         struct btrfs_path *path;
2624         struct extent_buffer *leaf;
2625         int ret;
2626
2627         path = btrfs_alloc_path();
2628         BUG_ON(!path);
2629         path->leave_spinning = 1;
2630         ret = btrfs_lookup_inode(trans, root, path,
2631                                  &BTRFS_I(inode)->location, 1);
2632         if (ret) {
2633                 if (ret > 0)
2634                         ret = -ENOENT;
2635                 goto failed;
2636         }
2637
2638         btrfs_unlock_up_safe(path, 1);
2639         leaf = path->nodes[0];
2640         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2641                                   struct btrfs_inode_item);
2642
2643         fill_inode_item(trans, leaf, inode_item, inode);
2644         btrfs_mark_buffer_dirty(leaf);
2645         btrfs_set_inode_last_trans(trans, inode);
2646         ret = 0;
2647 failed:
2648         btrfs_free_path(path);
2649         return ret;
2650 }
2651
2652
2653 /*
2654  * unlink helper that gets used here in inode.c and in the tree logging
2655  * recovery code.  It remove a link in a directory with a given name, and
2656  * also drops the back refs in the inode to the directory
2657  */
2658 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2659                        struct btrfs_root *root,
2660                        struct inode *dir, struct inode *inode,
2661                        const char *name, int name_len)
2662 {
2663         struct btrfs_path *path;
2664         int ret = 0;
2665         struct extent_buffer *leaf;
2666         struct btrfs_dir_item *di;
2667         struct btrfs_key key;
2668         u64 index;
2669
2670         path = btrfs_alloc_path();
2671         if (!path) {
2672                 ret = -ENOMEM;
2673                 goto out;
2674         }
2675
2676         path->leave_spinning = 1;
2677         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2678                                     name, name_len, -1);
2679         if (IS_ERR(di)) {
2680                 ret = PTR_ERR(di);
2681                 goto err;
2682         }
2683         if (!di) {
2684                 ret = -ENOENT;
2685                 goto err;
2686         }
2687         leaf = path->nodes[0];
2688         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2689         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2690         if (ret)
2691                 goto err;
2692         btrfs_release_path(root, path);
2693
2694         ret = btrfs_del_inode_ref(trans, root, name, name_len,
2695                                   inode->i_ino,
2696                                   dir->i_ino, &index);
2697         if (ret) {
2698                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2699                        "inode %lu parent %lu\n", name_len, name,
2700                        inode->i_ino, dir->i_ino);
2701                 goto err;
2702         }
2703
2704         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2705                                          index, name, name_len, -1);
2706         if (IS_ERR(di)) {
2707                 ret = PTR_ERR(di);
2708                 goto err;
2709         }
2710         if (!di) {
2711                 ret = -ENOENT;
2712                 goto err;
2713         }
2714         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2715         btrfs_release_path(root, path);
2716
2717         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2718                                          inode, dir->i_ino);
2719         BUG_ON(ret != 0 && ret != -ENOENT);
2720
2721         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2722                                            dir, index);
2723         if (ret == -ENOENT)
2724                 ret = 0;
2725 err:
2726         btrfs_free_path(path);
2727         if (ret)
2728                 goto out;
2729
2730         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2731         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2732         btrfs_update_inode(trans, root, dir);
2733         btrfs_drop_nlink(inode);
2734         ret = btrfs_update_inode(trans, root, inode);
2735 out:
2736         return ret;
2737 }
2738
2739 /* helper to check if there is any shared block in the path */
2740 static int check_path_shared(struct btrfs_root *root,
2741                              struct btrfs_path *path)
2742 {
2743         struct extent_buffer *eb;
2744         int level;
2745         u64 refs = 1;
2746
2747         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2748                 int ret;
2749
2750                 if (!path->nodes[level])
2751                         break;
2752                 eb = path->nodes[level];
2753                 if (!btrfs_block_can_be_shared(root, eb))
2754                         continue;
2755                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2756                                                &refs, NULL);
2757                 if (refs > 1)
2758                         return 1;
2759         }
2760         return 0;
2761 }
2762
2763 /*
2764  * helper to start transaction for unlink and rmdir.
2765  *
2766  * unlink and rmdir are special in btrfs, they do not always free space.
2767  * so in enospc case, we should make sure they will free space before
2768  * allowing them to use the global metadata reservation.
2769  */
2770 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2771                                                        struct dentry *dentry)
2772 {
2773         struct btrfs_trans_handle *trans;
2774         struct btrfs_root *root = BTRFS_I(dir)->root;
2775         struct btrfs_path *path;
2776         struct btrfs_inode_ref *ref;
2777         struct btrfs_dir_item *di;
2778         struct inode *inode = dentry->d_inode;
2779         u64 index;
2780         int check_link = 1;
2781         int err = -ENOSPC;
2782         int ret;
2783
2784         trans = btrfs_start_transaction(root, 10);
2785         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2786                 return trans;
2787
2788         if (inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2789                 return ERR_PTR(-ENOSPC);
2790
2791         /* check if there is someone else holds reference */
2792         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2793                 return ERR_PTR(-ENOSPC);
2794
2795         if (atomic_read(&inode->i_count) > 2)
2796                 return ERR_PTR(-ENOSPC);
2797
2798         if (xchg(&root->fs_info->enospc_unlink, 1))
2799                 return ERR_PTR(-ENOSPC);
2800
2801         path = btrfs_alloc_path();
2802         if (!path) {
2803                 root->fs_info->enospc_unlink = 0;
2804                 return ERR_PTR(-ENOMEM);
2805         }
2806
2807         trans = btrfs_start_transaction(root, 0);
2808         if (IS_ERR(trans)) {
2809                 btrfs_free_path(path);
2810                 root->fs_info->enospc_unlink = 0;
2811                 return trans;
2812         }
2813
2814         path->skip_locking = 1;
2815         path->search_commit_root = 1;
2816
2817         ret = btrfs_lookup_inode(trans, root, path,
2818                                 &BTRFS_I(dir)->location, 0);
2819         if (ret < 0) {
2820                 err = ret;
2821                 goto out;
2822         }
2823         if (ret == 0) {
2824                 if (check_path_shared(root, path))
2825                         goto out;
2826         } else {
2827                 check_link = 0;
2828         }
2829         btrfs_release_path(root, path);
2830
2831         ret = btrfs_lookup_inode(trans, root, path,
2832                                 &BTRFS_I(inode)->location, 0);
2833         if (ret < 0) {
2834                 err = ret;
2835                 goto out;
2836         }
2837         if (ret == 0) {
2838                 if (check_path_shared(root, path))
2839                         goto out;
2840         } else {
2841                 check_link = 0;
2842         }
2843         btrfs_release_path(root, path);
2844
2845         if (ret == 0 && S_ISREG(inode->i_mode)) {
2846                 ret = btrfs_lookup_file_extent(trans, root, path,
2847                                                inode->i_ino, (u64)-1, 0);
2848                 if (ret < 0) {
2849                         err = ret;
2850                         goto out;
2851                 }
2852                 BUG_ON(ret == 0);
2853                 if (check_path_shared(root, path))
2854                         goto out;
2855                 btrfs_release_path(root, path);
2856         }
2857
2858         if (!check_link) {
2859                 err = 0;
2860                 goto out;
2861         }
2862
2863         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2864                                 dentry->d_name.name, dentry->d_name.len, 0);
2865         if (IS_ERR(di)) {
2866                 err = PTR_ERR(di);
2867                 goto out;
2868         }
2869         if (di) {
2870                 if (check_path_shared(root, path))
2871                         goto out;
2872         } else {
2873                 err = 0;
2874                 goto out;
2875         }
2876         btrfs_release_path(root, path);
2877
2878         ref = btrfs_lookup_inode_ref(trans, root, path,
2879                                 dentry->d_name.name, dentry->d_name.len,
2880                                 inode->i_ino, dir->i_ino, 0);
2881         if (IS_ERR(ref)) {
2882                 err = PTR_ERR(ref);
2883                 goto out;
2884         }
2885         BUG_ON(!ref);
2886         if (check_path_shared(root, path))
2887                 goto out;
2888         index = btrfs_inode_ref_index(path->nodes[0], ref);
2889         btrfs_release_path(root, path);
2890
2891         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, index,
2892                                 dentry->d_name.name, dentry->d_name.len, 0);
2893         if (IS_ERR(di)) {
2894                 err = PTR_ERR(di);
2895                 goto out;
2896         }
2897         BUG_ON(ret == -ENOENT);
2898         if (check_path_shared(root, path))
2899                 goto out;
2900
2901         err = 0;
2902 out:
2903         btrfs_free_path(path);
2904         if (err) {
2905                 btrfs_end_transaction(trans, root);
2906                 root->fs_info->enospc_unlink = 0;
2907                 return ERR_PTR(err);
2908         }
2909
2910         trans->block_rsv = &root->fs_info->global_block_rsv;
2911         return trans;
2912 }
2913
2914 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2915                                struct btrfs_root *root)
2916 {
2917         if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2918                 BUG_ON(!root->fs_info->enospc_unlink);
2919                 root->fs_info->enospc_unlink = 0;
2920         }
2921         btrfs_end_transaction_throttle(trans, root);
2922 }
2923
2924 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2925 {
2926         struct btrfs_root *root = BTRFS_I(dir)->root;
2927         struct btrfs_trans_handle *trans;
2928         struct inode *inode = dentry->d_inode;
2929         int ret;
2930         unsigned long nr = 0;
2931
2932         trans = __unlink_start_trans(dir, dentry);
2933         if (IS_ERR(trans))
2934                 return PTR_ERR(trans);
2935
2936         btrfs_set_trans_block_group(trans, dir);
2937
2938         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2939
2940         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2941                                  dentry->d_name.name, dentry->d_name.len);
2942         BUG_ON(ret);
2943
2944         if (inode->i_nlink == 0) {
2945                 ret = btrfs_orphan_add(trans, inode);
2946                 BUG_ON(ret);
2947         }
2948
2949         nr = trans->blocks_used;
2950         __unlink_end_trans(trans, root);
2951         btrfs_btree_balance_dirty(root, nr);
2952         return ret;
2953 }
2954
2955 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2956                         struct btrfs_root *root,
2957                         struct inode *dir, u64 objectid,
2958                         const char *name, int name_len)
2959 {
2960         struct btrfs_path *path;
2961         struct extent_buffer *leaf;
2962         struct btrfs_dir_item *di;
2963         struct btrfs_key key;
2964         u64 index;
2965         int ret;
2966
2967         path = btrfs_alloc_path();
2968         if (!path)
2969                 return -ENOMEM;
2970
2971         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2972                                    name, name_len, -1);
2973         BUG_ON(!di || IS_ERR(di));
2974
2975         leaf = path->nodes[0];
2976         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2977         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2978         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2979         BUG_ON(ret);
2980         btrfs_release_path(root, path);
2981
2982         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2983                                  objectid, root->root_key.objectid,
2984                                  dir->i_ino, &index, name, name_len);
2985         if (ret < 0) {
2986                 BUG_ON(ret != -ENOENT);
2987                 di = btrfs_search_dir_index_item(root, path, dir->i_ino,
2988                                                  name, name_len);
2989                 BUG_ON(!di || IS_ERR(di));
2990
2991                 leaf = path->nodes[0];
2992                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2993                 btrfs_release_path(root, path);
2994                 index = key.offset;
2995         }
2996
2997         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2998                                          index, name, name_len, -1);
2999         BUG_ON(!di || IS_ERR(di));
3000
3001         leaf = path->nodes[0];
3002         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3003         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3004         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3005         BUG_ON(ret);
3006         btrfs_release_path(root, path);
3007
3008         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3009         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3010         ret = btrfs_update_inode(trans, root, dir);
3011         BUG_ON(ret);
3012
3013         btrfs_free_path(path);
3014         return 0;
3015 }
3016
3017 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3018 {
3019         struct inode *inode = dentry->d_inode;
3020         int err = 0;
3021         struct btrfs_root *root = BTRFS_I(dir)->root;
3022         struct btrfs_trans_handle *trans;
3023         unsigned long nr = 0;
3024
3025         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3026             inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
3027                 return -ENOTEMPTY;
3028
3029         trans = __unlink_start_trans(dir, dentry);
3030         if (IS_ERR(trans))
3031                 return PTR_ERR(trans);
3032
3033         btrfs_set_trans_block_group(trans, dir);
3034
3035         if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3036                 err = btrfs_unlink_subvol(trans, root, dir,
3037                                           BTRFS_I(inode)->location.objectid,
3038                                           dentry->d_name.name,
3039                                           dentry->d_name.len);
3040                 goto out;
3041         }
3042
3043         err = btrfs_orphan_add(trans, inode);
3044         if (err)
3045                 goto out;
3046
3047         /* now the directory is empty */
3048         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3049                                  dentry->d_name.name, dentry->d_name.len);
3050         if (!err)
3051                 btrfs_i_size_write(inode, 0);
3052 out:
3053         nr = trans->blocks_used;
3054         __unlink_end_trans(trans, root);
3055         btrfs_btree_balance_dirty(root, nr);
3056
3057         return err;
3058 }
3059
3060 #if 0
3061 /*
3062  * when truncating bytes in a file, it is possible to avoid reading
3063  * the leaves that contain only checksum items.  This can be the
3064  * majority of the IO required to delete a large file, but it must
3065  * be done carefully.
3066  *
3067  * The keys in the level just above the leaves are checked to make sure
3068  * the lowest key in a given leaf is a csum key, and starts at an offset
3069  * after the new  size.
3070  *
3071  * Then the key for the next leaf is checked to make sure it also has
3072  * a checksum item for the same file.  If it does, we know our target leaf
3073  * contains only checksum items, and it can be safely freed without reading
3074  * it.
3075  *
3076  * This is just an optimization targeted at large files.  It may do
3077  * nothing.  It will return 0 unless things went badly.
3078  */
3079 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
3080                                      struct btrfs_root *root,
3081                                      struct btrfs_path *path,
3082                                      struct inode *inode, u64 new_size)
3083 {
3084         struct btrfs_key key;
3085         int ret;
3086         int nritems;
3087         struct btrfs_key found_key;
3088         struct btrfs_key other_key;
3089         struct btrfs_leaf_ref *ref;
3090         u64 leaf_gen;
3091         u64 leaf_start;
3092
3093         path->lowest_level = 1;
3094         key.objectid = inode->i_ino;
3095         key.type = BTRFS_CSUM_ITEM_KEY;
3096         key.offset = new_size;
3097 again:
3098         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3099         if (ret < 0)
3100                 goto out;
3101
3102         if (path->nodes[1] == NULL) {
3103                 ret = 0;
3104                 goto out;
3105         }
3106         ret = 0;
3107         btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
3108         nritems = btrfs_header_nritems(path->nodes[1]);
3109
3110         if (!nritems)
3111                 goto out;
3112
3113         if (path->slots[1] >= nritems)
3114                 goto next_node;
3115
3116         /* did we find a key greater than anything we want to delete? */
3117         if (found_key.objectid > inode->i_ino ||
3118            (found_key.objectid == inode->i_ino && found_key.type > key.type))
3119                 goto out;
3120
3121         /* we check the next key in the node to make sure the leave contains
3122          * only checksum items.  This comparison doesn't work if our
3123          * leaf is the last one in the node
3124          */
3125         if (path->slots[1] + 1 >= nritems) {
3126 next_node:
3127                 /* search forward from the last key in the node, this
3128                  * will bring us into the next node in the tree
3129                  */
3130                 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
3131
3132                 /* unlikely, but we inc below, so check to be safe */
3133                 if (found_key.offset == (u64)-1)
3134                         goto out;
3135
3136                 /* search_forward needs a path with locks held, do the
3137                  * search again for the original key.  It is possible
3138                  * this will race with a balance and return a path that
3139                  * we could modify, but this drop is just an optimization
3140                  * and is allowed to miss some leaves.
3141                  */
3142                 btrfs_release_path(root, path);
3143                 found_key.offset++;
3144
3145                 /* setup a max key for search_forward */
3146                 other_key.offset = (u64)-1;
3147                 other_key.type = key.type;
3148                 other_key.objectid = key.objectid;
3149
3150                 path->keep_locks = 1;
3151                 ret = btrfs_search_forward(root, &found_key, &other_key,
3152                                            path, 0, 0);
3153                 path->keep_locks = 0;
3154                 if (ret || found_key.objectid != key.objectid ||
3155                     found_key.type != key.type) {
3156                         ret = 0;
3157                         goto out;
3158                 }
3159
3160                 key.offset = found_key.offset;
3161                 btrfs_release_path(root, path);
3162                 cond_resched();
3163                 goto again;
3164         }
3165
3166         /* we know there's one more slot after us in the tree,
3167          * read that key so we can verify it is also a checksum item
3168          */
3169         btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
3170
3171         if (found_key.objectid < inode->i_ino)
3172                 goto next_key;
3173
3174         if (found_key.type != key.type || found_key.offset < new_size)
3175                 goto next_key;
3176
3177         /*
3178          * if the key for the next leaf isn't a csum key from this objectid,
3179          * we can't be sure there aren't good items inside this leaf.
3180          * Bail out
3181          */
3182         if (other_key.objectid != inode->i_ino || other_key.type != key.type)
3183                 goto out;
3184
3185         leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
3186         leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
3187         /*
3188          * it is safe to delete this leaf, it contains only
3189          * csum items from this inode at an offset >= new_size
3190          */
3191         ret = btrfs_del_leaf(trans, root, path, leaf_start);
3192         BUG_ON(ret);
3193
3194         if (root->ref_cows && leaf_gen < trans->transid) {
3195                 ref = btrfs_alloc_leaf_ref(root, 0);
3196                 if (ref) {
3197                         ref->root_gen = root->root_key.offset;
3198                         ref->bytenr = leaf_start;
3199                         ref->owner = 0;
3200                         ref->generation = leaf_gen;
3201                         ref->nritems = 0;
3202
3203                         btrfs_sort_leaf_ref(ref);
3204
3205                         ret = btrfs_add_leaf_ref(root, ref, 0);
3206                         WARN_ON(ret);
3207                         btrfs_free_leaf_ref(root, ref);
3208                 } else {
3209                         WARN_ON(1);
3210                 }
3211         }
3212 next_key:
3213         btrfs_release_path(root, path);
3214
3215         if (other_key.objectid == inode->i_ino &&
3216             other_key.type == key.type && other_key.offset > key.offset) {
3217                 key.offset = other_key.offset;
3218                 cond_resched();
3219                 goto again;
3220         }
3221         ret = 0;
3222 out:
3223         /* fixup any changes we've made to the path */
3224         path->lowest_level = 0;
3225         path->keep_locks = 0;
3226         btrfs_release_path(root, path);
3227         return ret;
3228 }
3229
3230 #endif
3231
3232 /*
3233  * this can truncate away extent items, csum items and directory items.
3234  * It starts at a high offset and removes keys until it can't find
3235  * any higher than new_size
3236  *
3237  * csum items that cross the new i_size are truncated to the new size
3238  * as well.
3239  *
3240  * min_type is the minimum key type to truncate down to.  If set to 0, this
3241  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3242  */
3243 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3244                                struct btrfs_root *root,
3245                                struct inode *inode,
3246                                u64 new_size, u32 min_type)
3247 {
3248         struct btrfs_path *path;
3249         struct extent_buffer *leaf;
3250         struct btrfs_file_extent_item *fi;
3251         struct btrfs_key key;
3252         struct btrfs_key found_key;
3253         u64 extent_start = 0;
3254         u64 extent_num_bytes = 0;
3255         u64 extent_offset = 0;
3256         u64 item_end = 0;
3257         u64 mask = root->sectorsize - 1;
3258         u32 found_type = (u8)-1;
3259         int found_extent;
3260         int del_item;
3261         int pending_del_nr = 0;
3262         int pending_del_slot = 0;
3263         int extent_type = -1;
3264         int encoding;
3265         int ret;
3266         int err = 0;
3267
3268         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3269
3270         if (root->ref_cows || root == root->fs_info->tree_root)
3271                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3272
3273         path = btrfs_alloc_path();
3274         BUG_ON(!path);
3275         path->reada = -1;
3276
3277         key.objectid = inode->i_ino;
3278         key.offset = (u64)-1;
3279         key.type = (u8)-1;
3280
3281 search_again:
3282         path->leave_spinning = 1;
3283         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3284         if (ret < 0) {
3285                 err = ret;
3286                 goto out;
3287         }
3288
3289         if (ret > 0) {
3290                 /* there are no items in the tree for us to truncate, we're
3291                  * done
3292                  */
3293                 if (path->slots[0] == 0)
3294                         goto out;
3295                 path->slots[0]--;
3296         }
3297
3298         while (1) {
3299                 fi = NULL;
3300                 leaf = path->nodes[0];
3301                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3302                 found_type = btrfs_key_type(&found_key);
3303                 encoding = 0;
3304
3305                 if (found_key.objectid != inode->i_ino)
3306                         break;
3307
3308                 if (found_type < min_type)
3309                         break;
3310
3311                 item_end = found_key.offset;
3312                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3313                         fi = btrfs_item_ptr(leaf, path->slots[0],
3314                                             struct btrfs_file_extent_item);
3315                         extent_type = btrfs_file_extent_type(leaf, fi);
3316                         encoding = btrfs_file_extent_compression(leaf, fi);
3317                         encoding |= btrfs_file_extent_encryption(leaf, fi);
3318                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3319
3320                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3321                                 item_end +=
3322                                     btrfs_file_extent_num_bytes(leaf, fi);
3323                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3324                                 item_end += btrfs_file_extent_inline_len(leaf,
3325                                                                          fi);
3326                         }
3327                         item_end--;
3328                 }
3329                 if (found_type > min_type) {
3330                         del_item = 1;
3331                 } else {
3332                         if (item_end < new_size)
3333                                 break;
3334                         if (found_key.offset >= new_size)
3335                                 del_item = 1;
3336                         else
3337                                 del_item = 0;
3338                 }
3339                 found_extent = 0;
3340                 /* FIXME, shrink the extent if the ref count is only 1 */
3341                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3342                         goto delete;
3343
3344                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3345                         u64 num_dec;
3346                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3347                         if (!del_item && !encoding) {
3348                                 u64 orig_num_bytes =
3349                                         btrfs_file_extent_num_bytes(leaf, fi);
3350                                 extent_num_bytes = new_size -
3351                                         found_key.offset + root->sectorsize - 1;
3352                                 extent_num_bytes = extent_num_bytes &
3353                                         ~((u64)root->sectorsize - 1);
3354                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3355                                                          extent_num_bytes);
3356                                 num_dec = (orig_num_bytes -
3357                                            extent_num_bytes);
3358                                 if (root->ref_cows && extent_start != 0)
3359                                         inode_sub_bytes(inode, num_dec);
3360                                 btrfs_mark_buffer_dirty(leaf);
3361                         } else {
3362                                 extent_num_bytes =
3363                                         btrfs_file_extent_disk_num_bytes(leaf,
3364                                                                          fi);
3365                                 extent_offset = found_key.offset -
3366                                         btrfs_file_extent_offset(leaf, fi);
3367
3368                                 /* FIXME blocksize != 4096 */
3369                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3370                                 if (extent_start != 0) {
3371                                         found_extent = 1;
3372                                         if (root->ref_cows)
3373                                                 inode_sub_bytes(inode, num_dec);
3374                                 }
3375                         }
3376                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3377                         /*
3378                          * we can't truncate inline items that have had
3379                          * special encodings
3380                          */
3381                         if (!del_item &&
3382                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3383                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3384                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3385                                 u32 size = new_size - found_key.offset;
3386
3387                                 if (root->ref_cows) {
3388                                         inode_sub_bytes(inode, item_end + 1 -
3389                                                         new_size);
3390                                 }
3391                                 size =
3392                                     btrfs_file_extent_calc_inline_size(size);
3393                                 ret = btrfs_truncate_item(trans, root, path,
3394                                                           size, 1);
3395                                 BUG_ON(ret);
3396                         } else if (root->ref_cows) {
3397                                 inode_sub_bytes(inode, item_end + 1 -
3398                                                 found_key.offset);
3399                         }
3400                 }
3401 delete:
3402                 if (del_item) {
3403                         if (!pending_del_nr) {
3404                                 /* no pending yet, add ourselves */
3405                                 pending_del_slot = path->slots[0];
3406                                 pending_del_nr = 1;
3407                         } else if (pending_del_nr &&
3408                                    path->slots[0] + 1 == pending_del_slot) {
3409                                 /* hop on the pending chunk */
3410                                 pending_del_nr++;
3411                                 pending_del_slot = path->slots[0];
3412                         } else {
3413                                 BUG();
3414                         }
3415                 } else {
3416                         break;
3417                 }
3418                 if (found_extent && (root->ref_cows ||
3419                                      root == root->fs_info->tree_root)) {
3420                         btrfs_set_path_blocking(path);
3421                         ret = btrfs_free_extent(trans, root, extent_start,
3422                                                 extent_num_bytes, 0,
3423                                                 btrfs_header_owner(leaf),
3424                                                 inode->i_ino, extent_offset);
3425                         BUG_ON(ret);
3426                 }
3427
3428                 if (found_type == BTRFS_INODE_ITEM_KEY)
3429                         break;
3430
3431                 if (path->slots[0] == 0 ||
3432                     path->slots[0] != pending_del_slot) {
3433                         if (root->ref_cows) {
3434                                 err = -EAGAIN;
3435                                 goto out;
3436                         }
3437                         if (pending_del_nr) {
3438                                 ret = btrfs_del_items(trans, root, path,
3439                                                 pending_del_slot,
3440                                                 pending_del_nr);
3441                                 BUG_ON(ret);
3442                                 pending_del_nr = 0;
3443                         }
3444                         btrfs_release_path(root, path);
3445                         goto search_again;
3446                 } else {
3447                         path->slots[0]--;
3448                 }
3449         }
3450 out:
3451         if (pending_del_nr) {
3452                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3453                                       pending_del_nr);
3454                 BUG_ON(ret);
3455         }
3456         btrfs_free_path(path);
3457         return err;
3458 }
3459
3460 /*
3461  * taken from block_truncate_page, but does cow as it zeros out
3462  * any bytes left in the last page in the file.
3463  */
3464 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3465 {
3466         struct inode *inode = mapping->host;
3467         struct btrfs_root *root = BTRFS_I(inode)->root;
3468         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3469         struct btrfs_ordered_extent *ordered;
3470         struct extent_state *cached_state = NULL;
3471         char *kaddr;
3472         u32 blocksize = root->sectorsize;
3473         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3474         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3475         struct page *page;
3476         int ret = 0;
3477         u64 page_start;
3478         u64 page_end;
3479
3480         if ((offset & (blocksize - 1)) == 0)
3481                 goto out;
3482         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3483         if (ret)
3484                 goto out;
3485
3486         ret = -ENOMEM;
3487 again:
3488         page = grab_cache_page(mapping, index);
3489         if (!page) {
3490                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3491                 goto out;
3492         }
3493
3494         page_start = page_offset(page);
3495         page_end = page_start + PAGE_CACHE_SIZE - 1;
3496
3497         if (!PageUptodate(page)) {
3498                 ret = btrfs_readpage(NULL, page);
3499                 lock_page(page);
3500                 if (page->mapping != mapping) {
3501                         unlock_page(page);
3502                         page_cache_release(page);
3503                         goto again;
3504                 }
3505                 if (!PageUptodate(page)) {
3506                         ret = -EIO;
3507                         goto out_unlock;
3508                 }
3509         }
3510         wait_on_page_writeback(page);
3511
3512         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3513                          GFP_NOFS);
3514         set_page_extent_mapped(page);
3515
3516         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3517         if (ordered) {
3518                 unlock_extent_cached(io_tree, page_start, page_end,
3519                                      &cached_state, GFP_NOFS);
3520                 unlock_page(page);
3521                 page_cache_release(page);
3522                 btrfs_start_ordered_extent(inode, ordered, 1);
3523                 btrfs_put_ordered_extent(ordered);
3524                 goto again;
3525         }
3526
3527         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3528                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3529                           0, 0, &cached_state, GFP_NOFS);
3530
3531         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3532                                         &cached_state);
3533         if (ret) {
3534                 unlock_extent_cached(io_tree, page_start, page_end,
3535                                      &cached_state, GFP_NOFS);
3536                 goto out_unlock;
3537         }
3538
3539         ret = 0;
3540         if (offset != PAGE_CACHE_SIZE) {
3541                 kaddr = kmap(page);
3542                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3543                 flush_dcache_page(page);
3544                 kunmap(page);
3545         }
3546         ClearPageChecked(page);
3547         set_page_dirty(page);
3548         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3549                              GFP_NOFS);
3550
3551 out_unlock:
3552         if (ret)
3553                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3554         unlock_page(page);
3555         page_cache_release(page);
3556 out:
3557         return ret;
3558 }
3559
3560 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3561 {
3562         struct btrfs_trans_handle *trans;
3563         struct btrfs_root *root = BTRFS_I(inode)->root;
3564         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3565         struct extent_map *em = NULL;
3566         struct extent_state *cached_state = NULL;
3567         u64 mask = root->sectorsize - 1;
3568         u64 hole_start = (oldsize + mask) & ~mask;
3569         u64 block_end = (size + mask) & ~mask;
3570         u64 last_byte;
3571         u64 cur_offset;
3572         u64 hole_size;
3573         int err = 0;
3574
3575         if (size <= hole_start)
3576                 return 0;
3577
3578         while (1) {
3579                 struct btrfs_ordered_extent *ordered;
3580                 btrfs_wait_ordered_range(inode, hole_start,
3581                                          block_end - hole_start);
3582                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3583                                  &cached_state, GFP_NOFS);
3584                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3585                 if (!ordered)
3586                         break;
3587                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3588                                      &cached_state, GFP_NOFS);
3589                 btrfs_put_ordered_extent(ordered);
3590         }
3591
3592         cur_offset = hole_start;
3593         while (1) {
3594                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3595                                 block_end - cur_offset, 0);
3596                 BUG_ON(IS_ERR(em) || !em);
3597                 last_byte = min(extent_map_end(em), block_end);
3598                 last_byte = (last_byte + mask) & ~mask;
3599                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3600                         u64 hint_byte = 0;
3601                         hole_size = last_byte - cur_offset;
3602
3603                         trans = btrfs_start_transaction(root, 2);
3604                         if (IS_ERR(trans)) {
3605                                 err = PTR_ERR(trans);
3606                                 break;
3607                         }
3608                         btrfs_set_trans_block_group(trans, inode);
3609
3610                         err = btrfs_drop_extents(trans, inode, cur_offset,
3611                                                  cur_offset + hole_size,
3612                                                  &hint_byte, 1);
3613                         if (err)
3614                                 break;
3615
3616                         err = btrfs_insert_file_extent(trans, root,
3617                                         inode->i_ino, cur_offset, 0,
3618                                         0, hole_size, 0, hole_size,
3619                                         0, 0, 0);
3620                         if (err)
3621                                 break;
3622
3623                         btrfs_drop_extent_cache(inode, hole_start,
3624                                         last_byte - 1, 0);
3625
3626                         btrfs_end_transaction(trans, root);
3627                 }
3628                 free_extent_map(em);
3629                 em = NULL;
3630                 cur_offset = last_byte;
3631                 if (cur_offset >= block_end)
3632                         break;
3633         }
3634
3635         free_extent_map(em);
3636         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3637                              GFP_NOFS);
3638         return err;
3639 }
3640
3641 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3642 {
3643         struct btrfs_root *root = BTRFS_I(inode)->root;
3644         struct btrfs_trans_handle *trans;
3645         loff_t oldsize = i_size_read(inode);
3646         unsigned long nr;
3647         int ret;
3648
3649         if (newsize == oldsize)
3650                 return 0;
3651
3652         if (newsize > oldsize) {
3653                 i_size_write(inode, newsize);
3654                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3655                 truncate_pagecache(inode, oldsize, newsize);
3656                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3657                 if (ret) {
3658                         btrfs_setsize(inode, oldsize);
3659                         return ret;
3660                 }
3661
3662                 trans = btrfs_start_transaction(root, 1);
3663                 if (IS_ERR(trans))
3664                         return PTR_ERR(trans);
3665
3666                 ret = btrfs_update_inode(trans, root, inode);
3667                 if (ret) {
3668                         btrfs_end_transaction(trans, root);
3669                         return ret;
3670                 }
3671                 nr = trans->blocks_used;
3672                 btrfs_end_transaction(trans, root);
3673                 btrfs_btree_balance_dirty(root, nr);
3674         } else {
3675
3676                 /*
3677                  * We're truncating a file that used to have good data down to
3678                  * zero. Make sure it gets into the ordered flush list so that
3679                  * any new writes get down to disk quickly.
3680                  */
3681                 if (newsize == 0)
3682                         BTRFS_I(inode)->ordered_data_close = 1;
3683
3684                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3685                 truncate_setsize(inode, newsize);
3686                 ret = btrfs_truncate(inode);
3687         }
3688
3689         return ret;
3690 }
3691
3692 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3693 {
3694         struct inode *inode = dentry->d_inode;
3695         struct btrfs_root *root = BTRFS_I(inode)->root;
3696         int err;
3697
3698         if (btrfs_root_readonly(root))
3699                 return -EROFS;
3700
3701         err = inode_change_ok(inode, attr);
3702         if (err)
3703                 return err;
3704
3705         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3706                 err = btrfs_setsize(inode, attr->ia_size);
3707                 if (err)
3708                         return err;
3709         }
3710
3711         if (attr->ia_valid) {
3712                 setattr_copy(inode, attr);
3713                 mark_inode_dirty(inode);
3714
3715                 if (attr->ia_valid & ATTR_MODE)
3716                         err = btrfs_acl_chmod(inode);
3717         }
3718
3719         return err;
3720 }
3721
3722 void btrfs_evict_inode(struct inode *inode)
3723 {
3724         struct btrfs_trans_handle *trans;
3725         struct btrfs_root *root = BTRFS_I(inode)->root;
3726         unsigned long nr;
3727         int ret;
3728
3729         truncate_inode_pages(&inode->i_data, 0);
3730         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3731                                root == root->fs_info->tree_root))
3732                 goto no_delete;
3733
3734         if (is_bad_inode(inode)) {
3735                 btrfs_orphan_del(NULL, inode);
3736                 goto no_delete;
3737         }
3738         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3739         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3740
3741         if (root->fs_info->log_root_recovering) {
3742                 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3743                 goto no_delete;
3744         }
3745
3746         if (inode->i_nlink > 0) {
3747                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3748                 goto no_delete;
3749         }
3750
3751         btrfs_i_size_write(inode, 0);
3752
3753         while (1) {
3754                 trans = btrfs_start_transaction(root, 0);
3755                 BUG_ON(IS_ERR(trans));
3756                 btrfs_set_trans_block_group(trans, inode);
3757                 trans->block_rsv = root->orphan_block_rsv;
3758
3759                 ret = btrfs_block_rsv_check(trans, root,
3760                                             root->orphan_block_rsv, 0, 5);
3761                 if (ret) {
3762                         BUG_ON(ret != -EAGAIN);
3763                         ret = btrfs_commit_transaction(trans, root);
3764                         BUG_ON(ret);
3765                         continue;
3766                 }
3767
3768                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3769                 if (ret != -EAGAIN)
3770                         break;
3771
3772                 nr = trans->blocks_used;
3773                 btrfs_end_transaction(trans, root);
3774                 trans = NULL;
3775                 btrfs_btree_balance_dirty(root, nr);
3776
3777         }
3778
3779         if (ret == 0) {
3780                 ret = btrfs_orphan_del(trans, inode);
3781                 BUG_ON(ret);
3782         }
3783
3784         nr = trans->blocks_used;
3785         btrfs_end_transaction(trans, root);
3786         btrfs_btree_balance_dirty(root, nr);
3787 no_delete:
3788         end_writeback(inode);
3789         return;
3790 }
3791
3792 /*
3793  * this returns the key found in the dir entry in the location pointer.
3794  * If no dir entries were found, location->objectid is 0.
3795  */
3796 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3797                                struct btrfs_key *location)
3798 {
3799         const char *name = dentry->d_name.name;
3800         int namelen = dentry->d_name.len;
3801         struct btrfs_dir_item *di;
3802         struct btrfs_path *path;
3803         struct btrfs_root *root = BTRFS_I(dir)->root;
3804         int ret = 0;
3805
3806         path = btrfs_alloc_path();
3807         BUG_ON(!path);
3808
3809         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3810                                     namelen, 0);
3811         if (IS_ERR(di))
3812                 ret = PTR_ERR(di);
3813
3814         if (!di || IS_ERR(di))
3815                 goto out_err;
3816
3817         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3818 out:
3819         btrfs_free_path(path);
3820         return ret;
3821 out_err:
3822         location->objectid = 0;
3823         goto out;
3824 }
3825
3826 /*
3827  * when we hit a tree root in a directory, the btrfs part of the inode
3828  * needs to be changed to reflect the root directory of the tree root.  This
3829  * is kind of like crossing a mount point.
3830  */
3831 static int fixup_tree_root_location(struct btrfs_root *root,
3832                                     struct inode *dir,
3833                                     struct dentry *dentry,
3834                                     struct btrfs_key *location,
3835                                     struct btrfs_root **sub_root)
3836 {
3837         struct btrfs_path *path;
3838         struct btrfs_root *new_root;
3839         struct btrfs_root_ref *ref;
3840         struct extent_buffer *leaf;
3841         int ret;
3842         int err = 0;
3843
3844         path = btrfs_alloc_path();
3845         if (!path) {
3846                 err = -ENOMEM;
3847                 goto out;
3848         }
3849
3850         err = -ENOENT;
3851         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3852                                   BTRFS_I(dir)->root->root_key.objectid,
3853                                   location->objectid);
3854         if (ret) {
3855                 if (ret < 0)
3856                         err = ret;
3857                 goto out;
3858         }
3859
3860         leaf = path->nodes[0];
3861         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3862         if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino ||
3863             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3864                 goto out;
3865
3866         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3867                                    (unsigned long)(ref + 1),
3868                                    dentry->d_name.len);
3869         if (ret)
3870                 goto out;
3871
3872         btrfs_release_path(root->fs_info->tree_root, path);
3873
3874         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3875         if (IS_ERR(new_root)) {
3876                 err = PTR_ERR(new_root);
3877                 goto out;
3878         }
3879
3880         if (btrfs_root_refs(&new_root->root_item) == 0) {
3881                 err = -ENOENT;
3882                 goto out;
3883         }
3884
3885         *sub_root = new_root;
3886         location->objectid = btrfs_root_dirid(&new_root->root_item);
3887         location->type = BTRFS_INODE_ITEM_KEY;
3888         location->offset = 0;
3889         err = 0;
3890 out:
3891         btrfs_free_path(path);
3892         return err;
3893 }
3894
3895 static void inode_tree_add(struct inode *inode)
3896 {
3897         struct btrfs_root *root = BTRFS_I(inode)->root;
3898         struct btrfs_inode *entry;
3899         struct rb_node **p;
3900         struct rb_node *parent;
3901 again:
3902         p = &root->inode_tree.rb_node;
3903         parent = NULL;
3904
3905         if (inode_unhashed(inode))
3906                 return;
3907
3908         spin_lock(&root->inode_lock);
3909         while (*p) {
3910                 parent = *p;
3911                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3912
3913                 if (inode->i_ino < entry->vfs_inode.i_ino)
3914                         p = &parent->rb_left;
3915                 else if (inode->i_ino > entry->vfs_inode.i_ino)
3916                         p = &parent->rb_right;
3917                 else {
3918                         WARN_ON(!(entry->vfs_inode.i_state &
3919                                   (I_WILL_FREE | I_FREEING)));
3920                         rb_erase(parent, &root->inode_tree);
3921                         RB_CLEAR_NODE(parent);
3922                         spin_unlock(&root->inode_lock);
3923                         goto again;
3924                 }
3925         }
3926         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3927         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3928         spin_unlock(&root->inode_lock);
3929 }
3930
3931 static void inode_tree_del(struct inode *inode)
3932 {
3933         struct btrfs_root *root = BTRFS_I(inode)->root;
3934         int empty = 0;
3935
3936         spin_lock(&root->inode_lock);
3937         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3938                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3939                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3940                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3941         }
3942         spin_unlock(&root->inode_lock);
3943
3944         /*
3945          * Free space cache has inodes in the tree root, but the tree root has a
3946          * root_refs of 0, so this could end up dropping the tree root as a
3947          * snapshot, so we need the extra !root->fs_info->tree_root check to
3948          * make sure we don't drop it.
3949          */
3950         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3951             root != root->fs_info->tree_root) {
3952                 synchronize_srcu(&root->fs_info->subvol_srcu);
3953                 spin_lock(&root->inode_lock);
3954                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3955                 spin_unlock(&root->inode_lock);
3956                 if (empty)
3957                         btrfs_add_dead_root(root);
3958         }
3959 }
3960
3961 int btrfs_invalidate_inodes(struct btrfs_root *root)
3962 {
3963         struct rb_node *node;
3964         struct rb_node *prev;
3965         struct btrfs_inode *entry;
3966         struct inode *inode;
3967         u64 objectid = 0;
3968
3969         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3970
3971         spin_lock(&root->inode_lock);
3972 again:
3973         node = root->inode_tree.rb_node;
3974         prev = NULL;
3975         while (node) {
3976                 prev = node;
3977                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3978
3979                 if (objectid < entry->vfs_inode.i_ino)
3980                         node = node->rb_left;
3981                 else if (objectid > entry->vfs_inode.i_ino)
3982                         node = node->rb_right;
3983                 else
3984                         break;
3985         }
3986         if (!node) {
3987                 while (prev) {
3988                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
3989                         if (objectid <= entry->vfs_inode.i_ino) {
3990                                 node = prev;
3991                                 break;
3992                         }
3993                         prev = rb_next(prev);
3994                 }
3995         }
3996         while (node) {
3997                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3998                 objectid = entry->vfs_inode.i_ino + 1;
3999                 inode = igrab(&entry->vfs_inode);
4000                 if (inode) {
4001                         spin_unlock(&root->inode_lock);
4002                         if (atomic_read(&inode->i_count) > 1)
4003                                 d_prune_aliases(inode);
4004                         /*
4005                          * btrfs_drop_inode will have it removed from
4006                          * the inode cache when its usage count
4007                          * hits zero.
4008                          */
4009                         iput(inode);
4010                         cond_resched();
4011                         spin_lock(&root->inode_lock);
4012                         goto again;
4013                 }
4014
4015                 if (cond_resched_lock(&root->inode_lock))
4016                         goto again;
4017
4018                 node = rb_next(node);
4019         }
4020         spin_unlock(&root->inode_lock);
4021         return 0;
4022 }
4023
4024 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4025 {
4026         struct btrfs_iget_args *args = p;
4027         inode->i_ino = args->ino;
4028         BTRFS_I(inode)->root = args->root;
4029         btrfs_set_inode_space_info(args->root, inode);
4030         return 0;
4031 }
4032
4033 static int btrfs_find_actor(struct inode *inode, void *opaque)
4034 {
4035         struct btrfs_iget_args *args = opaque;
4036         return args->ino == inode->i_ino &&
4037                 args->root == BTRFS_I(inode)->root;
4038 }
4039
4040 static struct inode *btrfs_iget_locked(struct super_block *s,
4041                                        u64 objectid,
4042                                        struct btrfs_root *root)
4043 {
4044         struct inode *inode;
4045         struct btrfs_iget_args args;
4046         args.ino = objectid;
4047         args.root = root;
4048
4049         inode = iget5_locked(s, objectid, btrfs_find_actor,
4050                              btrfs_init_locked_inode,
4051                              (void *)&args);
4052         return inode;
4053 }
4054
4055 /* Get an inode object given its location and corresponding root.
4056  * Returns in *is_new if the inode was read from disk
4057  */
4058 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4059                          struct btrfs_root *root, int *new)
4060 {
4061         struct inode *inode;
4062
4063         inode = btrfs_iget_locked(s, location->objectid, root);
4064         if (!inode)
4065                 return ERR_PTR(-ENOMEM);
4066
4067         if (inode->i_state & I_NEW) {
4068                 BTRFS_I(inode)->root = root;
4069                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4070                 btrfs_read_locked_inode(inode);
4071
4072                 inode_tree_add(inode);
4073                 unlock_new_inode(inode);
4074                 if (new)
4075                         *new = 1;
4076         }
4077
4078         return inode;
4079 }
4080
4081 static struct inode *new_simple_dir(struct super_block *s,
4082                                     struct btrfs_key *key,
4083                                     struct btrfs_root *root)
4084 {
4085         struct inode *inode = new_inode(s);
4086
4087         if (!inode)
4088                 return ERR_PTR(-ENOMEM);
4089
4090         BTRFS_I(inode)->root = root;
4091         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4092         BTRFS_I(inode)->dummy_inode = 1;
4093
4094         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4095         inode->i_op = &simple_dir_inode_operations;
4096         inode->i_fop = &simple_dir_operations;
4097         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4098         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4099
4100         return inode;
4101 }
4102
4103 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4104 {
4105         struct inode *inode;
4106         struct btrfs_root *root = BTRFS_I(dir)->root;
4107         struct btrfs_root *sub_root = root;
4108         struct btrfs_key location;
4109         int index;
4110         int ret;
4111
4112         if (dentry->d_name.len > BTRFS_NAME_LEN)
4113                 return ERR_PTR(-ENAMETOOLONG);
4114
4115         ret = btrfs_inode_by_name(dir, dentry, &location);
4116
4117         if (ret < 0)
4118                 return ERR_PTR(ret);
4119
4120         if (location.objectid == 0)
4121                 return NULL;
4122
4123         if (location.type == BTRFS_INODE_ITEM_KEY) {
4124                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4125                 return inode;
4126         }
4127
4128         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4129
4130         index = srcu_read_lock(&root->fs_info->subvol_srcu);
4131         ret = fixup_tree_root_location(root, dir, dentry,
4132                                        &location, &sub_root);
4133         if (ret < 0) {
4134                 if (ret != -ENOENT)
4135                         inode = ERR_PTR(ret);
4136                 else
4137                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
4138         } else {
4139                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4140         }
4141         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4142
4143         if (!IS_ERR(inode) && root != sub_root) {
4144                 down_read(&root->fs_info->cleanup_work_sem);
4145                 if (!(inode->i_sb->s_flags & MS_RDONLY))
4146                         ret = btrfs_orphan_cleanup(sub_root);
4147                 up_read(&root->fs_info->cleanup_work_sem);
4148                 if (ret)
4149                         inode = ERR_PTR(ret);
4150         }
4151
4152         return inode;
4153 }
4154
4155 static int btrfs_dentry_delete(const struct dentry *dentry)
4156 {
4157         struct btrfs_root *root;
4158
4159         if (!dentry->d_inode && !IS_ROOT(dentry))
4160                 dentry = dentry->d_parent;
4161
4162         if (dentry->d_inode) {
4163                 root = BTRFS_I(dentry->d_inode)->root;
4164                 if (btrfs_root_refs(&root->root_item) == 0)
4165                         return 1;
4166         }
4167         return 0;
4168 }
4169
4170 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4171                                    struct nameidata *nd)
4172 {
4173         struct inode *inode;
4174
4175         inode = btrfs_lookup_dentry(dir, dentry);
4176         if (IS_ERR(inode))
4177                 return ERR_CAST(inode);
4178
4179         return d_splice_alias(inode, dentry);
4180 }
4181
4182 static unsigned char btrfs_filetype_table[] = {
4183         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4184 };
4185
4186 static int btrfs_real_readdir(struct file *filp, void *dirent,
4187                               filldir_t filldir)
4188 {
4189         struct inode *inode = filp->f_dentry->d_inode;
4190         struct btrfs_root *root = BTRFS_I(inode)->root;
4191         struct btrfs_item *item;
4192         struct btrfs_dir_item *di;
4193         struct btrfs_key key;
4194         struct btrfs_key found_key;
4195         struct btrfs_path *path;
4196         int ret;
4197         u32 nritems;
4198         struct extent_buffer *leaf;
4199         int slot;
4200         int advance;
4201         unsigned char d_type;
4202         int over = 0;
4203         u32 di_cur;
4204         u32 di_total;
4205         u32 di_len;
4206         int key_type = BTRFS_DIR_INDEX_KEY;
4207         char tmp_name[32];
4208         char *name_ptr;
4209         int name_len;
4210
4211         /* FIXME, use a real flag for deciding about the key type */
4212         if (root->fs_info->tree_root == root)
4213                 key_type = BTRFS_DIR_ITEM_KEY;
4214
4215         /* special case for "." */
4216         if (filp->f_pos == 0) {
4217                 over = filldir(dirent, ".", 1,
4218                                1, inode->i_ino,
4219                                DT_DIR);
4220                 if (over)
4221                         return 0;
4222                 filp->f_pos = 1;
4223         }
4224         /* special case for .., just use the back ref */
4225         if (filp->f_pos == 1) {
4226                 u64 pino = parent_ino(filp->f_path.dentry);
4227                 over = filldir(dirent, "..", 2,
4228                                2, pino, DT_DIR);
4229                 if (over)
4230                         return 0;
4231                 filp->f_pos = 2;
4232         }
4233         path = btrfs_alloc_path();
4234         path->reada = 2;
4235
4236         btrfs_set_key_type(&key, key_type);
4237         key.offset = filp->f_pos;
4238         key.objectid = inode->i_ino;
4239
4240         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4241         if (ret < 0)
4242                 goto err;
4243         advance = 0;
4244
4245         while (1) {
4246                 leaf = path->nodes[0];
4247                 nritems = btrfs_header_nritems(leaf);
4248                 slot = path->slots[0];
4249                 if (advance || slot >= nritems) {
4250                         if (slot >= nritems - 1) {
4251                                 ret = btrfs_next_leaf(root, path);
4252                                 if (ret)
4253                                         break;
4254                                 leaf = path->nodes[0];
4255                                 nritems = btrfs_header_nritems(leaf);
4256                                 slot = path->slots[0];
4257                         } else {
4258                                 slot++;
4259                                 path->slots[0]++;
4260                         }
4261                 }
4262
4263                 advance = 1;
4264                 item = btrfs_item_nr(leaf, slot);
4265                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4266
4267                 if (found_key.objectid != key.objectid)
4268                         break;
4269                 if (btrfs_key_type(&found_key) != key_type)
4270                         break;
4271                 if (found_key.offset < filp->f_pos)
4272                         continue;
4273
4274                 filp->f_pos = found_key.offset;
4275
4276                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4277                 di_cur = 0;
4278                 di_total = btrfs_item_size(leaf, item);
4279
4280                 while (di_cur < di_total) {
4281                         struct btrfs_key location;
4282
4283                         name_len = btrfs_dir_name_len(leaf, di);
4284                         if (name_len <= sizeof(tmp_name)) {
4285                                 name_ptr = tmp_name;
4286                         } else {
4287                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4288                                 if (!name_ptr) {
4289                                         ret = -ENOMEM;
4290                                         goto err;
4291                                 }
4292                         }
4293                         read_extent_buffer(leaf, name_ptr,
4294                                            (unsigned long)(di + 1), name_len);
4295
4296                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4297                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4298
4299                         /* is this a reference to our own snapshot? If so
4300                          * skip it
4301                          */
4302                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4303                             location.objectid == root->root_key.objectid) {
4304                                 over = 0;
4305                                 goto skip;
4306                         }
4307                         over = filldir(dirent, name_ptr, name_len,
4308                                        found_key.offset, location.objectid,
4309                                        d_type);
4310
4311 skip:
4312                         if (name_ptr != tmp_name)
4313                                 kfree(name_ptr);
4314
4315                         if (over)
4316                                 goto nopos;
4317                         di_len = btrfs_dir_name_len(leaf, di) +
4318                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4319                         di_cur += di_len;
4320                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4321                 }
4322         }
4323
4324         /* Reached end of directory/root. Bump pos past the last item. */
4325         if (key_type == BTRFS_DIR_INDEX_KEY)
4326                 /*
4327                  * 32-bit glibc will use getdents64, but then strtol -
4328                  * so the last number we can serve is this.
4329                  */
4330                 filp->f_pos = 0x7fffffff;
4331         else
4332                 filp->f_pos++;
4333 nopos:
4334         ret = 0;
4335 err:
4336         btrfs_free_path(path);
4337         return ret;
4338 }
4339
4340 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4341 {
4342         struct btrfs_root *root = BTRFS_I(inode)->root;
4343         struct btrfs_trans_handle *trans;
4344         int ret = 0;
4345         bool nolock = false;
4346
4347         if (BTRFS_I(inode)->dummy_inode)
4348                 return 0;
4349
4350         smp_mb();
4351         nolock = (root->fs_info->closing && root == root->fs_info->tree_root);
4352
4353         if (wbc->sync_mode == WB_SYNC_ALL) {
4354                 if (nolock)
4355                         trans = btrfs_join_transaction_nolock(root, 1);
4356                 else
4357                         trans = btrfs_join_transaction(root, 1);
4358                 if (IS_ERR(trans))
4359                         return PTR_ERR(trans);
4360                 btrfs_set_trans_block_group(trans, inode);
4361                 if (nolock)
4362                         ret = btrfs_end_transaction_nolock(trans, root);
4363                 else
4364                         ret = btrfs_commit_transaction(trans, root);
4365         }
4366         return ret;
4367 }
4368
4369 /*
4370  * This is somewhat expensive, updating the tree every time the
4371  * inode changes.  But, it is most likely to find the inode in cache.
4372  * FIXME, needs more benchmarking...there are no reasons other than performance
4373  * to keep or drop this code.
4374  */
4375 void btrfs_dirty_inode(struct inode *inode)
4376 {
4377         struct btrfs_root *root = BTRFS_I(inode)->root;
4378         struct btrfs_trans_handle *trans;
4379         int ret;
4380
4381         if (BTRFS_I(inode)->dummy_inode)
4382                 return;
4383
4384         trans = btrfs_join_transaction(root, 1);
4385         BUG_ON(IS_ERR(trans));
4386         btrfs_set_trans_block_group(trans, inode);
4387
4388         ret = btrfs_update_inode(trans, root, inode);
4389         if (ret && ret == -ENOSPC) {
4390                 /* whoops, lets try again with the full transaction */
4391                 btrfs_end_transaction(trans, root);
4392                 trans = btrfs_start_transaction(root, 1);
4393                 if (IS_ERR(trans)) {
4394                         if (printk_ratelimit()) {
4395                                 printk(KERN_ERR "btrfs: fail to "
4396                                        "dirty  inode %lu error %ld\n",
4397                                        inode->i_ino, PTR_ERR(trans));
4398                         }
4399                         return;
4400                 }
4401                 btrfs_set_trans_block_group(trans, inode);
4402
4403                 ret = btrfs_update_inode(trans, root, inode);
4404                 if (ret) {
4405                         if (printk_ratelimit()) {
4406                                 printk(KERN_ERR "btrfs: fail to "
4407                                        "dirty  inode %lu error %d\n",
4408                                        inode->i_ino, ret);
4409                         }
4410                 }
4411         }
4412         btrfs_end_transaction(trans, root);
4413 }
4414
4415 /*
4416  * find the highest existing sequence number in a directory
4417  * and then set the in-memory index_cnt variable to reflect
4418  * free sequence numbers
4419  */
4420 static int btrfs_set_inode_index_count(struct inode *inode)
4421 {
4422         struct btrfs_root *root = BTRFS_I(inode)->root;
4423         struct btrfs_key key, found_key;
4424         struct btrfs_path *path;
4425         struct extent_buffer *leaf;
4426         int ret;
4427
4428         key.objectid = inode->i_ino;
4429         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4430         key.offset = (u64)-1;
4431
4432         path = btrfs_alloc_path();
4433         if (!path)
4434                 return -ENOMEM;
4435
4436         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4437         if (ret < 0)
4438                 goto out;
4439         /* FIXME: we should be able to handle this */
4440         if (ret == 0)
4441                 goto out;
4442         ret = 0;
4443
4444         /*
4445          * MAGIC NUMBER EXPLANATION:
4446          * since we search a directory based on f_pos we have to start at 2
4447          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4448          * else has to start at 2
4449          */
4450         if (path->slots[0] == 0) {
4451                 BTRFS_I(inode)->index_cnt = 2;
4452                 goto out;
4453         }
4454
4455         path->slots[0]--;
4456
4457         leaf = path->nodes[0];
4458         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4459
4460         if (found_key.objectid != inode->i_ino ||
4461             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4462                 BTRFS_I(inode)->index_cnt = 2;
4463                 goto out;
4464         }
4465
4466         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4467 out:
4468         btrfs_free_path(path);
4469         return ret;
4470 }
4471
4472 /*
4473  * helper to find a free sequence number in a given directory.  This current
4474  * code is very simple, later versions will do smarter things in the btree
4475  */
4476 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4477 {
4478         int ret = 0;
4479
4480         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4481                 ret = btrfs_set_inode_index_count(dir);
4482                 if (ret)
4483                         return ret;
4484         }
4485
4486         *index = BTRFS_I(dir)->index_cnt;
4487         BTRFS_I(dir)->index_cnt++;
4488
4489         return ret;
4490 }
4491
4492 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4493                                      struct btrfs_root *root,
4494                                      struct inode *dir,
4495                                      const char *name, int name_len,
4496                                      u64 ref_objectid, u64 objectid,
4497                                      u64 alloc_hint, int mode, u64 *index)
4498 {
4499         struct inode *inode;
4500         struct btrfs_inode_item *inode_item;
4501         struct btrfs_key *location;
4502         struct btrfs_path *path;
4503         struct btrfs_inode_ref *ref;
4504         struct btrfs_key key[2];
4505         u32 sizes[2];
4506         unsigned long ptr;
4507         int ret;
4508         int owner;
4509
4510         path = btrfs_alloc_path();
4511         BUG_ON(!path);
4512
4513         inode = new_inode(root->fs_info->sb);
4514         if (!inode)
4515                 return ERR_PTR(-ENOMEM);
4516
4517         if (dir) {
4518                 ret = btrfs_set_inode_index(dir, index);
4519                 if (ret) {
4520                         iput(inode);
4521                         return ERR_PTR(ret);
4522                 }
4523         }
4524         /*
4525          * index_cnt is ignored for everything but a dir,
4526          * btrfs_get_inode_index_count has an explanation for the magic
4527          * number
4528          */
4529         BTRFS_I(inode)->index_cnt = 2;
4530         BTRFS_I(inode)->root = root;
4531         BTRFS_I(inode)->generation = trans->transid;
4532         inode->i_generation = BTRFS_I(inode)->generation;
4533         btrfs_set_inode_space_info(root, inode);
4534
4535         if (mode & S_IFDIR)
4536                 owner = 0;
4537         else
4538                 owner = 1;
4539         BTRFS_I(inode)->block_group =
4540                         btrfs_find_block_group(root, 0, alloc_hint, owner);
4541
4542         key[0].objectid = objectid;
4543         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4544         key[0].offset = 0;
4545
4546         key[1].objectid = objectid;
4547         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4548         key[1].offset = ref_objectid;
4549
4550         sizes[0] = sizeof(struct btrfs_inode_item);
4551         sizes[1] = name_len + sizeof(*ref);
4552
4553         path->leave_spinning = 1;
4554         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4555         if (ret != 0)
4556                 goto fail;
4557
4558         inode_init_owner(inode, dir, mode);
4559         inode->i_ino = objectid;
4560         inode_set_bytes(inode, 0);
4561         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4562         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4563                                   struct btrfs_inode_item);
4564         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4565
4566         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4567                              struct btrfs_inode_ref);
4568         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4569         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4570         ptr = (unsigned long)(ref + 1);
4571         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4572
4573         btrfs_mark_buffer_dirty(path->nodes[0]);
4574         btrfs_free_path(path);
4575
4576         location = &BTRFS_I(inode)->location;
4577         location->objectid = objectid;
4578         location->offset = 0;
4579         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4580
4581         btrfs_inherit_iflags(inode, dir);
4582
4583         if ((mode & S_IFREG)) {
4584                 if (btrfs_test_opt(root, NODATASUM))
4585                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4586                 if (btrfs_test_opt(root, NODATACOW))
4587                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4588         }
4589
4590         insert_inode_hash(inode);
4591         inode_tree_add(inode);
4592         return inode;
4593 fail:
4594         if (dir)
4595                 BTRFS_I(dir)->index_cnt--;
4596         btrfs_free_path(path);
4597         iput(inode);
4598         return ERR_PTR(ret);
4599 }
4600
4601 static inline u8 btrfs_inode_type(struct inode *inode)
4602 {
4603         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4604 }
4605
4606 /*
4607  * utility function to add 'inode' into 'parent_inode' with
4608  * a give name and a given sequence number.
4609  * if 'add_backref' is true, also insert a backref from the
4610  * inode to the parent directory.
4611  */
4612 int btrfs_add_link(struct btrfs_trans_handle *trans,
4613                    struct inode *parent_inode, struct inode *inode,
4614                    const char *name, int name_len, int add_backref, u64 index)
4615 {
4616         int ret = 0;
4617         struct btrfs_key key;
4618         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4619
4620         if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4621                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4622         } else {
4623                 key.objectid = inode->i_ino;
4624                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4625                 key.offset = 0;
4626         }
4627
4628         if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4629                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4630                                          key.objectid, root->root_key.objectid,
4631                                          parent_inode->i_ino,
4632                                          index, name, name_len);
4633         } else if (add_backref) {
4634                 ret = btrfs_insert_inode_ref(trans, root,
4635                                              name, name_len, inode->i_ino,
4636                                              parent_inode->i_ino, index);
4637         }
4638
4639         if (ret == 0) {
4640                 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4641                                             parent_inode->i_ino, &key,
4642                                             btrfs_inode_type(inode), index);
4643                 BUG_ON(ret);
4644
4645                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4646                                    name_len * 2);
4647                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4648                 ret = btrfs_update_inode(trans, root, parent_inode);
4649         }
4650         return ret;
4651 }
4652
4653 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4654                             struct inode *dir, struct dentry *dentry,
4655                             struct inode *inode, int backref, u64 index)
4656 {
4657         int err = btrfs_add_link(trans, dir, inode,
4658                                  dentry->d_name.name, dentry->d_name.len,
4659                                  backref, index);
4660         if (!err) {
4661                 d_instantiate(dentry, inode);
4662                 return 0;
4663         }
4664         if (err > 0)
4665                 err = -EEXIST;
4666         return err;
4667 }
4668
4669 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4670                         int mode, dev_t rdev)
4671 {
4672         struct btrfs_trans_handle *trans;
4673         struct btrfs_root *root = BTRFS_I(dir)->root;
4674         struct inode *inode = NULL;
4675         int err;
4676         int drop_inode = 0;
4677         u64 objectid;
4678         unsigned long nr = 0;
4679         u64 index = 0;
4680
4681         if (!new_valid_dev(rdev))
4682                 return -EINVAL;
4683
4684         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4685         if (err)
4686                 return err;
4687
4688         /*
4689          * 2 for inode item and ref
4690          * 2 for dir items
4691          * 1 for xattr if selinux is on
4692          */
4693         trans = btrfs_start_transaction(root, 5);
4694         if (IS_ERR(trans))
4695                 return PTR_ERR(trans);
4696
4697         btrfs_set_trans_block_group(trans, dir);
4698
4699         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4700                                 dentry->d_name.len, dir->i_ino, objectid,
4701                                 BTRFS_I(dir)->block_group, mode, &index);
4702         err = PTR_ERR(inode);
4703         if (IS_ERR(inode))
4704                 goto out_unlock;
4705
4706         err = btrfs_init_inode_security(trans, inode, dir);
4707         if (err) {
4708                 drop_inode = 1;
4709                 goto out_unlock;
4710         }
4711
4712         btrfs_set_trans_block_group(trans, inode);
4713         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4714         if (err)
4715                 drop_inode = 1;
4716         else {
4717                 inode->i_op = &btrfs_special_inode_operations;
4718                 init_special_inode(inode, inode->i_mode, rdev);
4719                 btrfs_update_inode(trans, root, inode);
4720         }
4721         btrfs_update_inode_block_group(trans, inode);
4722         btrfs_update_inode_block_group(trans, dir);
4723 out_unlock:
4724         nr = trans->blocks_used;
4725         btrfs_end_transaction_throttle(trans, root);
4726         btrfs_btree_balance_dirty(root, nr);
4727         if (drop_inode) {
4728                 inode_dec_link_count(inode);
4729                 iput(inode);
4730         }
4731         return err;
4732 }
4733
4734 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4735                         int mode, struct nameidata *nd)
4736 {
4737         struct btrfs_trans_handle *trans;
4738         struct btrfs_root *root = BTRFS_I(dir)->root;
4739         struct inode *inode = NULL;
4740         int drop_inode = 0;
4741         int err;
4742         unsigned long nr = 0;
4743         u64 objectid;
4744         u64 index = 0;
4745
4746         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4747         if (err)
4748                 return err;
4749         /*
4750          * 2 for inode item and ref
4751          * 2 for dir items
4752          * 1 for xattr if selinux is on
4753          */
4754         trans = btrfs_start_transaction(root, 5);
4755         if (IS_ERR(trans))
4756                 return PTR_ERR(trans);
4757
4758         btrfs_set_trans_block_group(trans, dir);
4759
4760         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4761                                 dentry->d_name.len, dir->i_ino, objectid,
4762                                 BTRFS_I(dir)->block_group, mode, &index);
4763         err = PTR_ERR(inode);
4764         if (IS_ERR(inode))
4765                 goto out_unlock;
4766
4767         err = btrfs_init_inode_security(trans, inode, dir);
4768         if (err) {
4769                 drop_inode = 1;
4770                 goto out_unlock;
4771         }
4772
4773         btrfs_set_trans_block_group(trans, inode);
4774         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4775         if (err)
4776                 drop_inode = 1;
4777         else {
4778                 inode->i_mapping->a_ops = &btrfs_aops;
4779                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4780                 inode->i_fop = &btrfs_file_operations;
4781                 inode->i_op = &btrfs_file_inode_operations;
4782                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4783         }
4784         btrfs_update_inode_block_group(trans, inode);
4785         btrfs_update_inode_block_group(trans, dir);
4786 out_unlock:
4787         nr = trans->blocks_used;
4788         btrfs_end_transaction_throttle(trans, root);
4789         if (drop_inode) {
4790                 inode_dec_link_count(inode);
4791                 iput(inode);
4792         }
4793         btrfs_btree_balance_dirty(root, nr);
4794         return err;
4795 }
4796
4797 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4798                       struct dentry *dentry)
4799 {
4800         struct btrfs_trans_handle *trans;
4801         struct btrfs_root *root = BTRFS_I(dir)->root;
4802         struct inode *inode = old_dentry->d_inode;
4803         u64 index;
4804         unsigned long nr = 0;
4805         int err;
4806         int drop_inode = 0;
4807
4808         if (inode->i_nlink == 0)
4809                 return -ENOENT;
4810
4811         /* do not allow sys_link's with other subvols of the same device */
4812         if (root->objectid != BTRFS_I(inode)->root->objectid)
4813                 return -EPERM;
4814
4815         btrfs_inc_nlink(inode);
4816         inode->i_ctime = CURRENT_TIME;
4817
4818         err = btrfs_set_inode_index(dir, &index);
4819         if (err)
4820                 goto fail;
4821
4822         /*
4823          * 2 items for inode and inode ref
4824          * 2 items for dir items
4825          * 1 item for parent inode
4826          */
4827         trans = btrfs_start_transaction(root, 5);
4828         if (IS_ERR(trans)) {
4829                 err = PTR_ERR(trans);
4830                 goto fail;
4831         }
4832
4833         btrfs_set_trans_block_group(trans, dir);
4834         ihold(inode);
4835
4836         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4837
4838         if (err) {
4839                 drop_inode = 1;
4840         } else {
4841                 struct dentry *parent = dget_parent(dentry);
4842                 btrfs_update_inode_block_group(trans, dir);
4843                 err = btrfs_update_inode(trans, root, inode);
4844                 BUG_ON(err);
4845                 btrfs_log_new_name(trans, inode, NULL, parent);
4846                 dput(parent);
4847         }
4848
4849         nr = trans->blocks_used;
4850         btrfs_end_transaction_throttle(trans, root);
4851 fail:
4852         if (drop_inode) {
4853                 inode_dec_link_count(inode);
4854                 iput(inode);
4855         }
4856         btrfs_btree_balance_dirty(root, nr);
4857         return err;
4858 }
4859
4860 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4861 {
4862         struct inode *inode = NULL;
4863         struct btrfs_trans_handle *trans;
4864         struct btrfs_root *root = BTRFS_I(dir)->root;
4865         int err = 0;
4866         int drop_on_err = 0;
4867         u64 objectid = 0;
4868         u64 index = 0;
4869         unsigned long nr = 1;
4870
4871         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4872         if (err)
4873                 return err;
4874
4875         /*
4876          * 2 items for inode and ref
4877          * 2 items for dir items
4878          * 1 for xattr if selinux is on
4879          */
4880         trans = btrfs_start_transaction(root, 5);
4881         if (IS_ERR(trans))
4882                 return PTR_ERR(trans);
4883         btrfs_set_trans_block_group(trans, dir);
4884
4885         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4886                                 dentry->d_name.len, dir->i_ino, objectid,
4887                                 BTRFS_I(dir)->block_group, S_IFDIR | mode,
4888                                 &index);
4889         if (IS_ERR(inode)) {
4890                 err = PTR_ERR(inode);
4891                 goto out_fail;
4892         }
4893
4894         drop_on_err = 1;
4895
4896         err = btrfs_init_inode_security(trans, inode, dir);
4897         if (err)
4898                 goto out_fail;
4899
4900         inode->i_op = &btrfs_dir_inode_operations;
4901         inode->i_fop = &btrfs_dir_file_operations;
4902         btrfs_set_trans_block_group(trans, inode);
4903
4904         btrfs_i_size_write(inode, 0);
4905         err = btrfs_update_inode(trans, root, inode);
4906         if (err)
4907                 goto out_fail;
4908
4909         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4910                              dentry->d_name.len, 0, index);
4911         if (err)
4912                 goto out_fail;
4913
4914         d_instantiate(dentry, inode);
4915         drop_on_err = 0;
4916         btrfs_update_inode_block_group(trans, inode);
4917         btrfs_update_inode_block_group(trans, dir);
4918
4919 out_fail:
4920         nr = trans->blocks_used;
4921         btrfs_end_transaction_throttle(trans, root);
4922         if (drop_on_err)
4923                 iput(inode);
4924         btrfs_btree_balance_dirty(root, nr);
4925         return err;
4926 }
4927
4928 /* helper for btfs_get_extent.  Given an existing extent in the tree,
4929  * and an extent that you want to insert, deal with overlap and insert
4930  * the new extent into the tree.
4931  */
4932 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4933                                 struct extent_map *existing,
4934                                 struct extent_map *em,
4935                                 u64 map_start, u64 map_len)
4936 {
4937         u64 start_diff;
4938
4939         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4940         start_diff = map_start - em->start;
4941         em->start = map_start;
4942         em->len = map_len;
4943         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4944             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4945                 em->block_start += start_diff;
4946                 em->block_len -= start_diff;
4947         }
4948         return add_extent_mapping(em_tree, em);
4949 }
4950
4951 static noinline int uncompress_inline(struct btrfs_path *path,
4952                                       struct inode *inode, struct page *page,
4953                                       size_t pg_offset, u64 extent_offset,
4954                                       struct btrfs_file_extent_item *item)
4955 {
4956         int ret;
4957         struct extent_buffer *leaf = path->nodes[0];
4958         char *tmp;
4959         size_t max_size;
4960         unsigned long inline_size;
4961         unsigned long ptr;
4962         int compress_type;
4963
4964         WARN_ON(pg_offset != 0);
4965         compress_type = btrfs_file_extent_compression(leaf, item);
4966         max_size = btrfs_file_extent_ram_bytes(leaf, item);
4967         inline_size = btrfs_file_extent_inline_item_len(leaf,
4968                                         btrfs_item_nr(leaf, path->slots[0]));
4969         tmp = kmalloc(inline_size, GFP_NOFS);
4970         ptr = btrfs_file_extent_inline_start(item);
4971
4972         read_extent_buffer(leaf, tmp, ptr, inline_size);
4973
4974         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4975         ret = btrfs_decompress(compress_type, tmp, page,
4976                                extent_offset, inline_size, max_size);
4977         if (ret) {
4978                 char *kaddr = kmap_atomic(page, KM_USER0);
4979                 unsigned long copy_size = min_t(u64,
4980                                   PAGE_CACHE_SIZE - pg_offset,
4981                                   max_size - extent_offset);
4982                 memset(kaddr + pg_offset, 0, copy_size);
4983                 kunmap_atomic(kaddr, KM_USER0);
4984         }
4985         kfree(tmp);
4986         return 0;
4987 }
4988
4989 /*
4990  * a bit scary, this does extent mapping from logical file offset to the disk.
4991  * the ugly parts come from merging extents from the disk with the in-ram
4992  * representation.  This gets more complex because of the data=ordered code,
4993  * where the in-ram extents might be locked pending data=ordered completion.
4994  *
4995  * This also copies inline extents directly into the page.
4996  */
4997
4998 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4999                                     size_t pg_offset, u64 start, u64 len,
5000                                     int create)
5001 {
5002         int ret;
5003         int err = 0;
5004         u64 bytenr;
5005         u64 extent_start = 0;
5006         u64 extent_end = 0;
5007         u64 objectid = inode->i_ino;
5008         u32 found_type;
5009         struct btrfs_path *path = NULL;
5010         struct btrfs_root *root = BTRFS_I(inode)->root;
5011         struct btrfs_file_extent_item *item;
5012         struct extent_buffer *leaf;
5013         struct btrfs_key found_key;
5014         struct extent_map *em = NULL;
5015         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5016         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5017         struct btrfs_trans_handle *trans = NULL;
5018         int compress_type;
5019
5020 again:
5021         read_lock(&em_tree->lock);
5022         em = lookup_extent_mapping(em_tree, start, len);
5023         if (em)
5024                 em->bdev = root->fs_info->fs_devices->latest_bdev;
5025         read_unlock(&em_tree->lock);
5026
5027         if (em) {
5028                 if (em->start > start || em->start + em->len <= start)
5029                         free_extent_map(em);
5030                 else if (em->block_start == EXTENT_MAP_INLINE && page)
5031                         free_extent_map(em);
5032                 else
5033                         goto out;
5034         }
5035         em = alloc_extent_map(GFP_NOFS);
5036         if (!em) {
5037                 err = -ENOMEM;
5038                 goto out;
5039         }
5040         em->bdev = root->fs_info->fs_devices->latest_bdev;
5041         em->start = EXTENT_MAP_HOLE;
5042         em->orig_start = EXTENT_MAP_HOLE;
5043         em->len = (u64)-1;
5044         em->block_len = (u64)-1;
5045
5046         if (!path) {
5047                 path = btrfs_alloc_path();
5048                 BUG_ON(!path);
5049         }
5050
5051         ret = btrfs_lookup_file_extent(trans, root, path,
5052                                        objectid, start, trans != NULL);
5053         if (ret < 0) {
5054                 err = ret;
5055                 goto out;
5056         }
5057
5058         if (ret != 0) {
5059                 if (path->slots[0] == 0)
5060                         goto not_found;
5061                 path->slots[0]--;
5062         }
5063
5064         leaf = path->nodes[0];
5065         item = btrfs_item_ptr(leaf, path->slots[0],
5066                               struct btrfs_file_extent_item);
5067         /* are we inside the extent that was found? */
5068         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5069         found_type = btrfs_key_type(&found_key);
5070         if (found_key.objectid != objectid ||
5071             found_type != BTRFS_EXTENT_DATA_KEY) {
5072                 goto not_found;
5073         }
5074
5075         found_type = btrfs_file_extent_type(leaf, item);
5076         extent_start = found_key.offset;
5077         compress_type = btrfs_file_extent_compression(leaf, item);
5078         if (found_type == BTRFS_FILE_EXTENT_REG ||
5079             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5080                 extent_end = extent_start +
5081                        btrfs_file_extent_num_bytes(leaf, item);
5082         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5083                 size_t size;
5084                 size = btrfs_file_extent_inline_len(leaf, item);
5085                 extent_end = (extent_start + size + root->sectorsize - 1) &
5086                         ~((u64)root->sectorsize - 1);
5087         }
5088
5089         if (start >= extent_end) {
5090                 path->slots[0]++;
5091                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5092                         ret = btrfs_next_leaf(root, path);
5093                         if (ret < 0) {
5094                                 err = ret;
5095                                 goto out;
5096                         }
5097                         if (ret > 0)
5098                                 goto not_found;
5099                         leaf = path->nodes[0];
5100                 }
5101                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5102                 if (found_key.objectid != objectid ||
5103                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5104                         goto not_found;
5105                 if (start + len <= found_key.offset)
5106                         goto not_found;
5107                 em->start = start;
5108                 em->len = found_key.offset - start;
5109                 goto not_found_em;
5110         }
5111
5112         if (found_type == BTRFS_FILE_EXTENT_REG ||
5113             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5114                 em->start = extent_start;
5115                 em->len = extent_end - extent_start;
5116                 em->orig_start = extent_start -
5117                                  btrfs_file_extent_offset(leaf, item);
5118                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5119                 if (bytenr == 0) {
5120                         em->block_start = EXTENT_MAP_HOLE;
5121                         goto insert;
5122                 }
5123                 if (compress_type != BTRFS_COMPRESS_NONE) {
5124                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5125                         em->compress_type = compress_type;
5126                         em->block_start = bytenr;
5127                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5128                                                                          item);
5129                 } else {
5130                         bytenr += btrfs_file_extent_offset(leaf, item);
5131                         em->block_start = bytenr;
5132                         em->block_len = em->len;
5133                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5134                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5135                 }
5136                 goto insert;
5137         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5138                 unsigned long ptr;
5139                 char *map;
5140                 size_t size;
5141                 size_t extent_offset;
5142                 size_t copy_size;
5143
5144                 em->block_start = EXTENT_MAP_INLINE;
5145                 if (!page || create) {
5146                         em->start = extent_start;
5147                         em->len = extent_end - extent_start;
5148                         goto out;
5149                 }
5150
5151                 size = btrfs_file_extent_inline_len(leaf, item);
5152                 extent_offset = page_offset(page) + pg_offset - extent_start;
5153                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5154                                 size - extent_offset);
5155                 em->start = extent_start + extent_offset;
5156                 em->len = (copy_size + root->sectorsize - 1) &
5157                         ~((u64)root->sectorsize - 1);
5158                 em->orig_start = EXTENT_MAP_INLINE;
5159                 if (compress_type) {
5160                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5161                         em->compress_type = compress_type;
5162                 }
5163                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5164                 if (create == 0 && !PageUptodate(page)) {
5165                         if (btrfs_file_extent_compression(leaf, item) !=
5166                             BTRFS_COMPRESS_NONE) {
5167                                 ret = uncompress_inline(path, inode, page,
5168                                                         pg_offset,
5169                                                         extent_offset, item);
5170                                 BUG_ON(ret);
5171                         } else {
5172                                 map = kmap(page);
5173                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5174                                                    copy_size);
5175                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5176                                         memset(map + pg_offset + copy_size, 0,
5177                                                PAGE_CACHE_SIZE - pg_offset -
5178                                                copy_size);
5179                                 }
5180                                 kunmap(page);
5181                         }
5182                         flush_dcache_page(page);
5183                 } else if (create && PageUptodate(page)) {
5184                         WARN_ON(1);
5185                         if (!trans) {
5186                                 kunmap(page);
5187                                 free_extent_map(em);
5188                                 em = NULL;
5189                                 btrfs_release_path(root, path);
5190                                 trans = btrfs_join_transaction(root, 1);
5191                                 if (IS_ERR(trans))
5192                                         return ERR_CAST(trans);
5193                                 goto again;
5194                         }
5195                         map = kmap(page);
5196                         write_extent_buffer(leaf, map + pg_offset, ptr,
5197                                             copy_size);
5198                         kunmap(page);
5199                         btrfs_mark_buffer_dirty(leaf);
5200                 }
5201                 set_extent_uptodate(io_tree, em->start,
5202                                     extent_map_end(em) - 1, GFP_NOFS);
5203                 goto insert;
5204         } else {
5205                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5206                 WARN_ON(1);
5207         }
5208 not_found:
5209         em->start = start;
5210         em->len = len;
5211 not_found_em:
5212         em->block_start = EXTENT_MAP_HOLE;
5213         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5214 insert:
5215         btrfs_release_path(root, path);
5216         if (em->start > start || extent_map_end(em) <= start) {
5217                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5218                        "[%llu %llu]\n", (unsigned long long)em->start,
5219                        (unsigned long long)em->len,
5220                        (unsigned long long)start,
5221                        (unsigned long long)len);
5222                 err = -EIO;
5223                 goto out;
5224         }
5225
5226         err = 0;
5227         write_lock(&em_tree->lock);
5228         ret = add_extent_mapping(em_tree, em);
5229         /* it is possible that someone inserted the extent into the tree
5230          * while we had the lock dropped.  It is also possible that
5231          * an overlapping map exists in the tree
5232          */
5233         if (ret == -EEXIST) {
5234                 struct extent_map *existing;
5235
5236                 ret = 0;
5237
5238                 existing = lookup_extent_mapping(em_tree, start, len);
5239                 if (existing && (existing->start > start ||
5240                     existing->start + existing->len <= start)) {
5241                         free_extent_map(existing);
5242                         existing = NULL;
5243                 }
5244                 if (!existing) {
5245                         existing = lookup_extent_mapping(em_tree, em->start,
5246                                                          em->len);
5247                         if (existing) {
5248                                 err = merge_extent_mapping(em_tree, existing,
5249                                                            em, start,
5250                                                            root->sectorsize);
5251                                 free_extent_map(existing);
5252                                 if (err) {
5253                                         free_extent_map(em);
5254                                         em = NULL;
5255                                 }
5256                         } else {
5257                                 err = -EIO;
5258                                 free_extent_map(em);
5259                                 em = NULL;
5260                         }
5261                 } else {
5262                         free_extent_map(em);
5263                         em = existing;
5264                         err = 0;
5265                 }
5266         }
5267         write_unlock(&em_tree->lock);
5268 out:
5269         if (path)
5270                 btrfs_free_path(path);
5271         if (trans) {
5272                 ret = btrfs_end_transaction(trans, root);
5273                 if (!err)
5274                         err = ret;
5275         }
5276         if (err) {
5277                 free_extent_map(em);
5278                 return ERR_PTR(err);
5279         }
5280         return em;
5281 }
5282
5283 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5284                                            size_t pg_offset, u64 start, u64 len,
5285                                            int create)
5286 {
5287         struct extent_map *em;
5288         struct extent_map *hole_em = NULL;
5289         u64 range_start = start;
5290         u64 end;
5291         u64 found;
5292         u64 found_end;
5293         int err = 0;
5294
5295         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5296         if (IS_ERR(em))
5297                 return em;
5298         if (em) {
5299                 /*
5300                  * if our em maps to a hole, there might
5301                  * actually be delalloc bytes behind it
5302                  */
5303                 if (em->block_start != EXTENT_MAP_HOLE)
5304                         return em;
5305                 else
5306                         hole_em = em;
5307         }
5308
5309         /* check to see if we've wrapped (len == -1 or similar) */
5310         end = start + len;
5311         if (end < start)
5312                 end = (u64)-1;
5313         else
5314                 end -= 1;
5315
5316         em = NULL;
5317
5318         /* ok, we didn't find anything, lets look for delalloc */
5319         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5320                                  end, len, EXTENT_DELALLOC, 1);
5321         found_end = range_start + found;
5322         if (found_end < range_start)
5323                 found_end = (u64)-1;
5324
5325         /*
5326          * we didn't find anything useful, return
5327          * the original results from get_extent()
5328          */
5329         if (range_start > end || found_end <= start) {
5330                 em = hole_em;
5331                 hole_em = NULL;
5332                 goto out;
5333         }
5334
5335         /* adjust the range_start to make sure it doesn't
5336          * go backwards from the start they passed in
5337          */
5338         range_start = max(start,range_start);
5339         found = found_end - range_start;
5340
5341         if (found > 0) {
5342                 u64 hole_start = start;
5343                 u64 hole_len = len;
5344
5345                 em = alloc_extent_map(GFP_NOFS);
5346                 if (!em) {
5347                         err = -ENOMEM;
5348                         goto out;
5349                 }
5350                 /*
5351                  * when btrfs_get_extent can't find anything it
5352                  * returns one huge hole
5353                  *
5354                  * make sure what it found really fits our range, and
5355                  * adjust to make sure it is based on the start from
5356                  * the caller
5357                  */
5358                 if (hole_em) {
5359                         u64 calc_end = extent_map_end(hole_em);
5360
5361                         if (calc_end <= start || (hole_em->start > end)) {
5362                                 free_extent_map(hole_em);
5363                                 hole_em = NULL;
5364                         } else {
5365                                 hole_start = max(hole_em->start, start);
5366                                 hole_len = calc_end - hole_start;
5367                         }
5368                 }
5369                 em->bdev = NULL;
5370                 if (hole_em && range_start > hole_start) {
5371                         /* our hole starts before our delalloc, so we
5372                          * have to return just the parts of the hole
5373                          * that go until  the delalloc starts
5374                          */
5375                         em->len = min(hole_len,
5376                                       range_start - hole_start);
5377                         em->start = hole_start;
5378                         em->orig_start = hole_start;
5379                         /*
5380                          * don't adjust block start at all,
5381                          * it is fixed at EXTENT_MAP_HOLE
5382                          */
5383                         em->block_start = hole_em->block_start;
5384                         em->block_len = hole_len;
5385                 } else {
5386                         em->start = range_start;
5387                         em->len = found;
5388                         em->orig_start = range_start;
5389                         em->block_start = EXTENT_MAP_DELALLOC;
5390                         em->block_len = found;
5391                 }
5392         } else if (hole_em) {
5393                 return hole_em;
5394         }
5395 out:
5396
5397         free_extent_map(hole_em);
5398         if (err) {
5399                 free_extent_map(em);
5400                 return ERR_PTR(err);
5401         }
5402         return em;
5403 }
5404
5405 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5406                                                   u64 start, u64 len)
5407 {
5408         struct btrfs_root *root = BTRFS_I(inode)->root;
5409         struct btrfs_trans_handle *trans;
5410         struct extent_map *em;
5411         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5412         struct btrfs_key ins;
5413         u64 alloc_hint;
5414         int ret;
5415
5416         btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5417
5418         trans = btrfs_join_transaction(root, 0);
5419         if (IS_ERR(trans))
5420                 return ERR_CAST(trans);
5421
5422         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5423
5424         alloc_hint = get_extent_allocation_hint(inode, start, len);
5425         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5426                                    alloc_hint, (u64)-1, &ins, 1);
5427         if (ret) {
5428                 em = ERR_PTR(ret);
5429                 goto out;
5430         }
5431
5432         em = alloc_extent_map(GFP_NOFS);
5433         if (!em) {
5434                 em = ERR_PTR(-ENOMEM);
5435                 goto out;
5436         }
5437
5438         em->start = start;
5439         em->orig_start = em->start;
5440         em->len = ins.offset;
5441
5442         em->block_start = ins.objectid;
5443         em->block_len = ins.offset;
5444         em->bdev = root->fs_info->fs_devices->latest_bdev;
5445         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5446
5447         while (1) {
5448                 write_lock(&em_tree->lock);
5449                 ret = add_extent_mapping(em_tree, em);
5450                 write_unlock(&em_tree->lock);
5451                 if (ret != -EEXIST)
5452                         break;
5453                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5454         }
5455
5456         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5457                                            ins.offset, ins.offset, 0);
5458         if (ret) {
5459                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5460                 em = ERR_PTR(ret);
5461         }
5462 out:
5463         btrfs_end_transaction(trans, root);
5464         return em;
5465 }
5466
5467 /*
5468  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5469  * block must be cow'd
5470  */
5471 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5472                                       struct inode *inode, u64 offset, u64 len)
5473 {
5474         struct btrfs_path *path;
5475         int ret;
5476         struct extent_buffer *leaf;
5477         struct btrfs_root *root = BTRFS_I(inode)->root;
5478         struct btrfs_file_extent_item *fi;
5479         struct btrfs_key key;
5480         u64 disk_bytenr;
5481         u64 backref_offset;
5482         u64 extent_end;
5483         u64 num_bytes;
5484         int slot;
5485         int found_type;
5486
5487         path = btrfs_alloc_path();
5488         if (!path)
5489                 return -ENOMEM;
5490
5491         ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
5492                                        offset, 0);
5493         if (ret < 0)
5494                 goto out;
5495
5496         slot = path->slots[0];
5497         if (ret == 1) {
5498                 if (slot == 0) {
5499                         /* can't find the item, must cow */
5500                         ret = 0;
5501                         goto out;
5502                 }
5503                 slot--;
5504         }
5505         ret = 0;
5506         leaf = path->nodes[0];
5507         btrfs_item_key_to_cpu(leaf, &key, slot);
5508         if (key.objectid != inode->i_ino ||
5509             key.type != BTRFS_EXTENT_DATA_KEY) {
5510                 /* not our file or wrong item type, must cow */
5511                 goto out;
5512         }
5513
5514         if (key.offset > offset) {
5515                 /* Wrong offset, must cow */
5516                 goto out;
5517         }
5518
5519         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5520         found_type = btrfs_file_extent_type(leaf, fi);
5521         if (found_type != BTRFS_FILE_EXTENT_REG &&
5522             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5523                 /* not a regular extent, must cow */
5524                 goto out;
5525         }
5526         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5527         backref_offset = btrfs_file_extent_offset(leaf, fi);
5528
5529         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5530         if (extent_end < offset + len) {
5531                 /* extent doesn't include our full range, must cow */
5532                 goto out;
5533         }
5534
5535         if (btrfs_extent_readonly(root, disk_bytenr))
5536                 goto out;
5537
5538         /*
5539          * look for other files referencing this extent, if we
5540          * find any we must cow
5541          */
5542         if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
5543                                   key.offset - backref_offset, disk_bytenr))
5544                 goto out;
5545
5546         /*
5547          * adjust disk_bytenr and num_bytes to cover just the bytes
5548          * in this extent we are about to write.  If there
5549          * are any csums in that range we have to cow in order
5550          * to keep the csums correct
5551          */
5552         disk_bytenr += backref_offset;
5553         disk_bytenr += offset - key.offset;
5554         num_bytes = min(offset + len, extent_end) - offset;
5555         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5556                                 goto out;
5557         /*
5558          * all of the above have passed, it is safe to overwrite this extent
5559          * without cow
5560          */
5561         ret = 1;
5562 out:
5563         btrfs_free_path(path);
5564         return ret;
5565 }
5566
5567 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5568                                    struct buffer_head *bh_result, int create)
5569 {
5570         struct extent_map *em;
5571         struct btrfs_root *root = BTRFS_I(inode)->root;
5572         u64 start = iblock << inode->i_blkbits;
5573         u64 len = bh_result->b_size;
5574         struct btrfs_trans_handle *trans;
5575
5576         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5577         if (IS_ERR(em))
5578                 return PTR_ERR(em);
5579
5580         /*
5581          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5582          * io.  INLINE is special, and we could probably kludge it in here, but
5583          * it's still buffered so for safety lets just fall back to the generic
5584          * buffered path.
5585          *
5586          * For COMPRESSED we _have_ to read the entire extent in so we can
5587          * decompress it, so there will be buffering required no matter what we
5588          * do, so go ahead and fallback to buffered.
5589          *
5590          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5591          * to buffered IO.  Don't blame me, this is the price we pay for using
5592          * the generic code.
5593          */
5594         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5595             em->block_start == EXTENT_MAP_INLINE) {
5596                 free_extent_map(em);
5597                 return -ENOTBLK;
5598         }
5599
5600         /* Just a good old fashioned hole, return */
5601         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5602                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5603                 free_extent_map(em);
5604                 /* DIO will do one hole at a time, so just unlock a sector */
5605                 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5606                               start + root->sectorsize - 1, GFP_NOFS);
5607                 return 0;
5608         }
5609
5610         /*
5611          * We don't allocate a new extent in the following cases
5612          *
5613          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
5614          * existing extent.
5615          * 2) The extent is marked as PREALLOC.  We're good to go here and can
5616          * just use the extent.
5617          *
5618          */
5619         if (!create) {
5620                 len = em->len - (start - em->start);
5621                 goto map;
5622         }
5623
5624         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5625             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5626              em->block_start != EXTENT_MAP_HOLE)) {
5627                 int type;
5628                 int ret;
5629                 u64 block_start;
5630
5631                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5632                         type = BTRFS_ORDERED_PREALLOC;
5633                 else
5634                         type = BTRFS_ORDERED_NOCOW;
5635                 len = min(len, em->len - (start - em->start));
5636                 block_start = em->block_start + (start - em->start);
5637
5638                 /*
5639                  * we're not going to log anything, but we do need
5640                  * to make sure the current transaction stays open
5641                  * while we look for nocow cross refs
5642                  */
5643                 trans = btrfs_join_transaction(root, 0);
5644                 if (IS_ERR(trans))
5645                         goto must_cow;
5646
5647                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5648                         ret = btrfs_add_ordered_extent_dio(inode, start,
5649                                            block_start, len, len, type);
5650                         btrfs_end_transaction(trans, root);
5651                         if (ret) {
5652                                 free_extent_map(em);
5653                                 return ret;
5654                         }
5655                         goto unlock;
5656                 }
5657                 btrfs_end_transaction(trans, root);
5658         }
5659 must_cow:
5660         /*
5661          * this will cow the extent, reset the len in case we changed
5662          * it above
5663          */
5664         len = bh_result->b_size;
5665         free_extent_map(em);
5666         em = btrfs_new_extent_direct(inode, start, len);
5667         if (IS_ERR(em))
5668                 return PTR_ERR(em);
5669         len = min(len, em->len - (start - em->start));
5670 unlock:
5671         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5672                           EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5673                           0, NULL, GFP_NOFS);
5674 map:
5675         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5676                 inode->i_blkbits;
5677         bh_result->b_size = len;
5678         bh_result->b_bdev = em->bdev;
5679         set_buffer_mapped(bh_result);
5680         if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5681                 set_buffer_new(bh_result);
5682
5683         free_extent_map(em);
5684
5685         return 0;
5686 }
5687
5688 struct btrfs_dio_private {
5689         struct inode *inode;
5690         u64 logical_offset;
5691         u64 disk_bytenr;
5692         u64 bytes;
5693         u32 *csums;
5694         void *private;
5695
5696         /* number of bios pending for this dio */
5697         atomic_t pending_bios;
5698
5699         /* IO errors */
5700         int errors;
5701
5702         struct bio *orig_bio;
5703 };
5704
5705 static void btrfs_endio_direct_read(struct bio *bio, int err)
5706 {
5707         struct btrfs_dio_private *dip = bio->bi_private;
5708         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5709         struct bio_vec *bvec = bio->bi_io_vec;
5710         struct inode *inode = dip->inode;
5711         struct btrfs_root *root = BTRFS_I(inode)->root;
5712         u64 start;
5713         u32 *private = dip->csums;
5714
5715         start = dip->logical_offset;
5716         do {
5717                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5718                         struct page *page = bvec->bv_page;
5719                         char *kaddr;
5720                         u32 csum = ~(u32)0;
5721                         unsigned long flags;
5722
5723                         local_irq_save(flags);
5724                         kaddr = kmap_atomic(page, KM_IRQ0);
5725                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5726                                                csum, bvec->bv_len);
5727                         btrfs_csum_final(csum, (char *)&csum);
5728                         kunmap_atomic(kaddr, KM_IRQ0);
5729                         local_irq_restore(flags);
5730
5731                         flush_dcache_page(bvec->bv_page);
5732                         if (csum != *private) {
5733                                 printk(KERN_ERR "btrfs csum failed ino %lu off"
5734                                       " %llu csum %u private %u\n",
5735                                       inode->i_ino, (unsigned long long)start,
5736                                       csum, *private);
5737                                 err = -EIO;
5738                         }
5739                 }
5740
5741                 start += bvec->bv_len;
5742                 private++;
5743                 bvec++;
5744         } while (bvec <= bvec_end);
5745
5746         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5747                       dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5748         bio->bi_private = dip->private;
5749
5750         kfree(dip->csums);
5751         kfree(dip);
5752         dio_end_io(bio, err);
5753 }
5754
5755 static void btrfs_endio_direct_write(struct bio *bio, int err)
5756 {
5757         struct btrfs_dio_private *dip = bio->bi_private;
5758         struct inode *inode = dip->inode;
5759         struct btrfs_root *root = BTRFS_I(inode)->root;
5760         struct btrfs_trans_handle *trans;
5761         struct btrfs_ordered_extent *ordered = NULL;
5762         struct extent_state *cached_state = NULL;
5763         u64 ordered_offset = dip->logical_offset;
5764         u64 ordered_bytes = dip->bytes;
5765         int ret;
5766
5767         if (err)
5768                 goto out_done;
5769 again:
5770         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5771                                                    &ordered_offset,
5772                                                    ordered_bytes);
5773         if (!ret)
5774                 goto out_test;
5775
5776         BUG_ON(!ordered);
5777
5778         trans = btrfs_join_transaction(root, 1);
5779         if (IS_ERR(trans)) {
5780                 err = -ENOMEM;
5781                 goto out;
5782         }
5783         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5784
5785         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5786                 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5787                 if (!ret)
5788                         ret = btrfs_update_inode(trans, root, inode);
5789                 err = ret;
5790                 goto out;
5791         }
5792
5793         lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5794                          ordered->file_offset + ordered->len - 1, 0,
5795                          &cached_state, GFP_NOFS);
5796
5797         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5798                 ret = btrfs_mark_extent_written(trans, inode,
5799                                                 ordered->file_offset,
5800                                                 ordered->file_offset +
5801                                                 ordered->len);
5802                 if (ret) {
5803                         err = ret;
5804                         goto out_unlock;
5805                 }
5806         } else {
5807                 ret = insert_reserved_file_extent(trans, inode,
5808                                                   ordered->file_offset,
5809                                                   ordered->start,
5810                                                   ordered->disk_len,
5811                                                   ordered->len,
5812                                                   ordered->len,
5813                                                   0, 0, 0,
5814                                                   BTRFS_FILE_EXTENT_REG);
5815                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5816                                    ordered->file_offset, ordered->len);
5817                 if (ret) {
5818                         err = ret;
5819                         WARN_ON(1);
5820                         goto out_unlock;
5821                 }
5822         }
5823
5824         add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5825         btrfs_ordered_update_i_size(inode, 0, ordered);
5826         btrfs_update_inode(trans, root, inode);
5827 out_unlock:
5828         unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5829                              ordered->file_offset + ordered->len - 1,
5830                              &cached_state, GFP_NOFS);
5831 out:
5832         btrfs_delalloc_release_metadata(inode, ordered->len);
5833         btrfs_end_transaction(trans, root);
5834         ordered_offset = ordered->file_offset + ordered->len;
5835         btrfs_put_ordered_extent(ordered);
5836         btrfs_put_ordered_extent(ordered);
5837
5838 out_test:
5839         /*
5840          * our bio might span multiple ordered extents.  If we haven't
5841          * completed the accounting for the whole dio, go back and try again
5842          */
5843         if (ordered_offset < dip->logical_offset + dip->bytes) {
5844                 ordered_bytes = dip->logical_offset + dip->bytes -
5845                         ordered_offset;
5846                 goto again;
5847         }
5848 out_done:
5849         bio->bi_private = dip->private;
5850
5851         kfree(dip->csums);
5852         kfree(dip);
5853         dio_end_io(bio, err);
5854 }
5855
5856 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5857                                     struct bio *bio, int mirror_num,
5858                                     unsigned long bio_flags, u64 offset)
5859 {
5860         int ret;
5861         struct btrfs_root *root = BTRFS_I(inode)->root;
5862         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5863         BUG_ON(ret);
5864         return 0;
5865 }
5866
5867 static void btrfs_end_dio_bio(struct bio *bio, int err)
5868 {
5869         struct btrfs_dio_private *dip = bio->bi_private;
5870
5871         if (err) {
5872                 printk(KERN_ERR "btrfs direct IO failed ino %lu rw %lu "
5873                       "sector %#Lx len %u err no %d\n",
5874                       dip->inode->i_ino, bio->bi_rw,
5875                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
5876                 dip->errors = 1;
5877
5878                 /*
5879                  * before atomic variable goto zero, we must make sure
5880                  * dip->errors is perceived to be set.
5881                  */
5882                 smp_mb__before_atomic_dec();
5883         }
5884
5885         /* if there are more bios still pending for this dio, just exit */
5886         if (!atomic_dec_and_test(&dip->pending_bios))
5887                 goto out;
5888
5889         if (dip->errors)
5890                 bio_io_error(dip->orig_bio);
5891         else {
5892                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5893                 bio_endio(dip->orig_bio, 0);
5894         }
5895 out:
5896         bio_put(bio);
5897 }
5898
5899 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5900                                        u64 first_sector, gfp_t gfp_flags)
5901 {
5902         int nr_vecs = bio_get_nr_vecs(bdev);
5903         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5904 }
5905
5906 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5907                                          int rw, u64 file_offset, int skip_sum,
5908                                          u32 *csums)
5909 {
5910         int write = rw & REQ_WRITE;
5911         struct btrfs_root *root = BTRFS_I(inode)->root;
5912         int ret;
5913
5914         bio_get(bio);
5915         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5916         if (ret)
5917                 goto err;
5918
5919         if (write && !skip_sum) {
5920                 ret = btrfs_wq_submit_bio(root->fs_info,
5921                                    inode, rw, bio, 0, 0,
5922                                    file_offset,
5923                                    __btrfs_submit_bio_start_direct_io,
5924                                    __btrfs_submit_bio_done);
5925                 goto err;
5926         } else if (!skip_sum)
5927                 btrfs_lookup_bio_sums_dio(root, inode, bio,
5928                                           file_offset, csums);
5929
5930         ret = btrfs_map_bio(root, rw, bio, 0, 1);
5931 err:
5932         bio_put(bio);
5933         return ret;
5934 }
5935
5936 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5937                                     int skip_sum)
5938 {
5939         struct inode *inode = dip->inode;
5940         struct btrfs_root *root = BTRFS_I(inode)->root;
5941         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5942         struct bio *bio;
5943         struct bio *orig_bio = dip->orig_bio;
5944         struct bio_vec *bvec = orig_bio->bi_io_vec;
5945         u64 start_sector = orig_bio->bi_sector;
5946         u64 file_offset = dip->logical_offset;
5947         u64 submit_len = 0;
5948         u64 map_length;
5949         int nr_pages = 0;
5950         u32 *csums = dip->csums;
5951         int ret = 0;
5952
5953         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5954         if (!bio)
5955                 return -ENOMEM;
5956         bio->bi_private = dip;
5957         bio->bi_end_io = btrfs_end_dio_bio;
5958         atomic_inc(&dip->pending_bios);
5959
5960         map_length = orig_bio->bi_size;
5961         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5962                               &map_length, NULL, 0);
5963         if (ret) {
5964                 bio_put(bio);
5965                 return -EIO;
5966         }
5967
5968         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
5969                 if (unlikely(map_length < submit_len + bvec->bv_len ||
5970                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5971                                  bvec->bv_offset) < bvec->bv_len)) {
5972                         /*
5973                          * inc the count before we submit the bio so
5974                          * we know the end IO handler won't happen before
5975                          * we inc the count. Otherwise, the dip might get freed
5976                          * before we're done setting it up
5977                          */
5978                         atomic_inc(&dip->pending_bios);
5979                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
5980                                                      file_offset, skip_sum,
5981                                                      csums);
5982                         if (ret) {
5983                                 bio_put(bio);
5984                                 atomic_dec(&dip->pending_bios);
5985                                 goto out_err;
5986                         }
5987
5988                         if (!skip_sum)
5989                                 csums = csums + nr_pages;
5990                         start_sector += submit_len >> 9;
5991                         file_offset += submit_len;
5992
5993                         submit_len = 0;
5994                         nr_pages = 0;
5995
5996                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
5997                                                   start_sector, GFP_NOFS);
5998                         if (!bio)
5999                                 goto out_err;
6000                         bio->bi_private = dip;
6001                         bio->bi_end_io = btrfs_end_dio_bio;
6002
6003                         map_length = orig_bio->bi_size;
6004                         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6005                                               &map_length, NULL, 0);
6006                         if (ret) {
6007                                 bio_put(bio);
6008                                 goto out_err;
6009                         }
6010                 } else {
6011                         submit_len += bvec->bv_len;
6012                         nr_pages ++;
6013                         bvec++;
6014                 }
6015         }
6016
6017         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6018                                      csums);
6019         if (!ret)
6020                 return 0;
6021
6022         bio_put(bio);
6023 out_err:
6024         dip->errors = 1;
6025         /*
6026          * before atomic variable goto zero, we must
6027          * make sure dip->errors is perceived to be set.
6028          */
6029         smp_mb__before_atomic_dec();
6030         if (atomic_dec_and_test(&dip->pending_bios))
6031                 bio_io_error(dip->orig_bio);
6032
6033         /* bio_end_io() will handle error, so we needn't return it */
6034         return 0;
6035 }
6036
6037 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6038                                 loff_t file_offset)
6039 {
6040         struct btrfs_root *root = BTRFS_I(inode)->root;
6041         struct btrfs_dio_private *dip;
6042         struct bio_vec *bvec = bio->bi_io_vec;
6043         int skip_sum;
6044         int write = rw & REQ_WRITE;
6045         int ret = 0;
6046
6047         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6048
6049         dip = kmalloc(sizeof(*dip), GFP_NOFS);
6050         if (!dip) {
6051                 ret = -ENOMEM;
6052                 goto free_ordered;
6053         }
6054         dip->csums = NULL;
6055
6056         if (!skip_sum) {
6057                 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6058                 if (!dip->csums) {
6059                         kfree(dip);
6060                         ret = -ENOMEM;
6061                         goto free_ordered;
6062                 }
6063         }
6064
6065         dip->private = bio->bi_private;
6066         dip->inode = inode;
6067         dip->logical_offset = file_offset;
6068
6069         dip->bytes = 0;
6070         do {
6071                 dip->bytes += bvec->bv_len;
6072                 bvec++;
6073         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6074
6075         dip->disk_bytenr = (u64)bio->bi_sector << 9;
6076         bio->bi_private = dip;
6077         dip->errors = 0;
6078         dip->orig_bio = bio;
6079         atomic_set(&dip->pending_bios, 0);
6080
6081         if (write)
6082                 bio->bi_end_io = btrfs_endio_direct_write;
6083         else
6084                 bio->bi_end_io = btrfs_endio_direct_read;
6085
6086         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6087         if (!ret)
6088                 return;
6089 free_ordered:
6090         /*
6091          * If this is a write, we need to clean up the reserved space and kill
6092          * the ordered extent.
6093          */
6094         if (write) {
6095                 struct btrfs_ordered_extent *ordered;
6096                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6097                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6098                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6099                         btrfs_free_reserved_extent(root, ordered->start,
6100                                                    ordered->disk_len);
6101                 btrfs_put_ordered_extent(ordered);
6102                 btrfs_put_ordered_extent(ordered);
6103         }
6104         bio_endio(bio, ret);
6105 }
6106
6107 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6108                         const struct iovec *iov, loff_t offset,
6109                         unsigned long nr_segs)
6110 {
6111         int seg;
6112         size_t size;
6113         unsigned long addr;
6114         unsigned blocksize_mask = root->sectorsize - 1;
6115         ssize_t retval = -EINVAL;
6116         loff_t end = offset;
6117
6118         if (offset & blocksize_mask)
6119                 goto out;
6120
6121         /* Check the memory alignment.  Blocks cannot straddle pages */
6122         for (seg = 0; seg < nr_segs; seg++) {
6123                 addr = (unsigned long)iov[seg].iov_base;
6124                 size = iov[seg].iov_len;
6125                 end += size;
6126                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6127                         goto out;
6128         }
6129         retval = 0;
6130 out:
6131         return retval;
6132 }
6133 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6134                         const struct iovec *iov, loff_t offset,
6135                         unsigned long nr_segs)
6136 {
6137         struct file *file = iocb->ki_filp;
6138         struct inode *inode = file->f_mapping->host;
6139         struct btrfs_ordered_extent *ordered;
6140         struct extent_state *cached_state = NULL;
6141         u64 lockstart, lockend;
6142         ssize_t ret;
6143         int writing = rw & WRITE;
6144         int write_bits = 0;
6145         size_t count = iov_length(iov, nr_segs);
6146
6147         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6148                             offset, nr_segs)) {
6149                 return 0;
6150         }
6151
6152         lockstart = offset;
6153         lockend = offset + count - 1;
6154
6155         if (writing) {
6156                 ret = btrfs_delalloc_reserve_space(inode, count);
6157                 if (ret)
6158                         goto out;
6159         }
6160
6161         while (1) {
6162                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6163                                  0, &cached_state, GFP_NOFS);
6164                 /*
6165                  * We're concerned with the entire range that we're going to be
6166                  * doing DIO to, so we need to make sure theres no ordered
6167                  * extents in this range.
6168                  */
6169                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6170                                                      lockend - lockstart + 1);
6171                 if (!ordered)
6172                         break;
6173                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6174                                      &cached_state, GFP_NOFS);
6175                 btrfs_start_ordered_extent(inode, ordered, 1);
6176                 btrfs_put_ordered_extent(ordered);
6177                 cond_resched();
6178         }
6179
6180         /*
6181          * we don't use btrfs_set_extent_delalloc because we don't want
6182          * the dirty or uptodate bits
6183          */
6184         if (writing) {
6185                 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6186                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6187                                      EXTENT_DELALLOC, 0, NULL, &cached_state,
6188                                      GFP_NOFS);
6189                 if (ret) {
6190                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6191                                          lockend, EXTENT_LOCKED | write_bits,
6192                                          1, 0, &cached_state, GFP_NOFS);
6193                         goto out;
6194                 }
6195         }
6196
6197         free_extent_state(cached_state);
6198         cached_state = NULL;
6199
6200         ret = __blockdev_direct_IO(rw, iocb, inode,
6201                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6202                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6203                    btrfs_submit_direct, 0);
6204
6205         if (ret < 0 && ret != -EIOCBQUEUED) {
6206                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6207                               offset + iov_length(iov, nr_segs) - 1,
6208                               EXTENT_LOCKED | write_bits, 1, 0,
6209                               &cached_state, GFP_NOFS);
6210         } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6211                 /*
6212                  * We're falling back to buffered, unlock the section we didn't
6213                  * do IO on.
6214                  */
6215                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6216                               offset + iov_length(iov, nr_segs) - 1,
6217                               EXTENT_LOCKED | write_bits, 1, 0,
6218                               &cached_state, GFP_NOFS);
6219         }
6220 out:
6221         free_extent_state(cached_state);
6222         return ret;
6223 }
6224
6225 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6226                 __u64 start, __u64 len)
6227 {
6228         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6229 }
6230
6231 int btrfs_readpage(struct file *file, struct page *page)
6232 {
6233         struct extent_io_tree *tree;
6234         tree = &BTRFS_I(page->mapping->host)->io_tree;
6235         return extent_read_full_page(tree, page, btrfs_get_extent);
6236 }
6237
6238 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6239 {
6240         struct extent_io_tree *tree;
6241
6242
6243         if (current->flags & PF_MEMALLOC) {
6244                 redirty_page_for_writepage(wbc, page);
6245                 unlock_page(page);
6246                 return 0;
6247         }
6248         tree = &BTRFS_I(page->mapping->host)->io_tree;
6249         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6250 }
6251
6252 int btrfs_writepages(struct address_space *mapping,
6253                      struct writeback_control *wbc)
6254 {
6255         struct extent_io_tree *tree;
6256
6257         tree = &BTRFS_I(mapping->host)->io_tree;
6258         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6259 }
6260
6261 static int
6262 btrfs_readpages(struct file *file, struct address_space *mapping,
6263                 struct list_head *pages, unsigned nr_pages)
6264 {
6265         struct extent_io_tree *tree;
6266         tree = &BTRFS_I(mapping->host)->io_tree;
6267         return extent_readpages(tree, mapping, pages, nr_pages,
6268                                 btrfs_get_extent);
6269 }
6270 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6271 {
6272         struct extent_io_tree *tree;
6273         struct extent_map_tree *map;
6274         int ret;
6275
6276         tree = &BTRFS_I(page->mapping->host)->io_tree;
6277         map = &BTRFS_I(page->mapping->host)->extent_tree;
6278         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6279         if (ret == 1) {
6280                 ClearPagePrivate(page);
6281                 set_page_private(page, 0);
6282                 page_cache_release(page);
6283         }
6284         return ret;
6285 }
6286
6287 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6288 {
6289         if (PageWriteback(page) || PageDirty(page))
6290                 return 0;
6291         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6292 }
6293
6294 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6295 {
6296         struct extent_io_tree *tree;
6297         struct btrfs_ordered_extent *ordered;
6298         struct extent_state *cached_state = NULL;
6299         u64 page_start = page_offset(page);
6300         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6301
6302
6303         /*
6304          * we have the page locked, so new writeback can't start,
6305          * and the dirty bit won't be cleared while we are here.
6306          *
6307          * Wait for IO on this page so that we can safely clear
6308          * the PagePrivate2 bit and do ordered accounting
6309          */
6310         wait_on_page_writeback(page);
6311
6312         tree = &BTRFS_I(page->mapping->host)->io_tree;
6313         if (offset) {
6314                 btrfs_releasepage(page, GFP_NOFS);
6315                 return;
6316         }
6317         lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6318                          GFP_NOFS);
6319         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6320                                            page_offset(page));
6321         if (ordered) {
6322                 /*
6323                  * IO on this page will never be started, so we need
6324                  * to account for any ordered extents now
6325                  */
6326                 clear_extent_bit(tree, page_start, page_end,
6327                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6328                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6329                                  &cached_state, GFP_NOFS);
6330                 /*
6331                  * whoever cleared the private bit is responsible
6332                  * for the finish_ordered_io
6333                  */
6334                 if (TestClearPagePrivate2(page)) {
6335                         btrfs_finish_ordered_io(page->mapping->host,
6336                                                 page_start, page_end);
6337                 }
6338                 btrfs_put_ordered_extent(ordered);
6339                 cached_state = NULL;
6340                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6341                                  GFP_NOFS);
6342         }
6343         clear_extent_bit(tree, page_start, page_end,
6344                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6345                  EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6346         __btrfs_releasepage(page, GFP_NOFS);
6347
6348         ClearPageChecked(page);
6349         if (PagePrivate(page)) {
6350                 ClearPagePrivate(page);
6351                 set_page_private(page, 0);
6352                 page_cache_release(page);
6353         }
6354 }
6355
6356 /*
6357  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6358  * called from a page fault handler when a page is first dirtied. Hence we must
6359  * be careful to check for EOF conditions here. We set the page up correctly
6360  * for a written page which means we get ENOSPC checking when writing into
6361  * holes and correct delalloc and unwritten extent mapping on filesystems that
6362  * support these features.
6363  *
6364  * We are not allowed to take the i_mutex here so we have to play games to
6365  * protect against truncate races as the page could now be beyond EOF.  Because
6366  * vmtruncate() writes the inode size before removing pages, once we have the
6367  * page lock we can determine safely if the page is beyond EOF. If it is not
6368  * beyond EOF, then the page is guaranteed safe against truncation until we
6369  * unlock the page.
6370  */
6371 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6372 {
6373         struct page *page = vmf->page;
6374         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6375         struct btrfs_root *root = BTRFS_I(inode)->root;
6376         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6377         struct btrfs_ordered_extent *ordered;
6378         struct extent_state *cached_state = NULL;
6379         char *kaddr;
6380         unsigned long zero_start;
6381         loff_t size;
6382         int ret;
6383         u64 page_start;
6384         u64 page_end;
6385
6386         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6387         if (ret) {
6388                 if (ret == -ENOMEM)
6389                         ret = VM_FAULT_OOM;
6390                 else /* -ENOSPC, -EIO, etc */
6391                         ret = VM_FAULT_SIGBUS;
6392                 goto out;
6393         }
6394
6395         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6396 again:
6397         lock_page(page);
6398         size = i_size_read(inode);
6399         page_start = page_offset(page);
6400         page_end = page_start + PAGE_CACHE_SIZE - 1;
6401
6402         if ((page->mapping != inode->i_mapping) ||
6403             (page_start >= size)) {
6404                 /* page got truncated out from underneath us */
6405                 goto out_unlock;
6406         }
6407         wait_on_page_writeback(page);
6408
6409         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6410                          GFP_NOFS);
6411         set_page_extent_mapped(page);
6412
6413         /*
6414          * we can't set the delalloc bits if there are pending ordered
6415          * extents.  Drop our locks and wait for them to finish
6416          */
6417         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6418         if (ordered) {
6419                 unlock_extent_cached(io_tree, page_start, page_end,
6420                                      &cached_state, GFP_NOFS);
6421                 unlock_page(page);
6422                 btrfs_start_ordered_extent(inode, ordered, 1);
6423                 btrfs_put_ordered_extent(ordered);
6424                 goto again;
6425         }
6426
6427         /*
6428          * XXX - page_mkwrite gets called every time the page is dirtied, even
6429          * if it was already dirty, so for space accounting reasons we need to
6430          * clear any delalloc bits for the range we are fixing to save.  There
6431          * is probably a better way to do this, but for now keep consistent with
6432          * prepare_pages in the normal write path.
6433          */
6434         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6435                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6436                           0, 0, &cached_state, GFP_NOFS);
6437
6438         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6439                                         &cached_state);
6440         if (ret) {
6441                 unlock_extent_cached(io_tree, page_start, page_end,
6442                                      &cached_state, GFP_NOFS);
6443                 ret = VM_FAULT_SIGBUS;
6444                 goto out_unlock;
6445         }
6446         ret = 0;
6447
6448         /* page is wholly or partially inside EOF */
6449         if (page_start + PAGE_CACHE_SIZE > size)
6450                 zero_start = size & ~PAGE_CACHE_MASK;
6451         else
6452                 zero_start = PAGE_CACHE_SIZE;
6453
6454         if (zero_start != PAGE_CACHE_SIZE) {
6455                 kaddr = kmap(page);
6456                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6457                 flush_dcache_page(page);
6458                 kunmap(page);
6459         }
6460         ClearPageChecked(page);
6461         set_page_dirty(page);
6462         SetPageUptodate(page);
6463
6464         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6465         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6466
6467         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6468
6469 out_unlock:
6470         if (!ret)
6471                 return VM_FAULT_LOCKED;
6472         unlock_page(page);
6473         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6474 out:
6475         return ret;
6476 }
6477
6478 static int btrfs_truncate(struct inode *inode)
6479 {
6480         struct btrfs_root *root = BTRFS_I(inode)->root;
6481         int ret;
6482         int err = 0;
6483         struct btrfs_trans_handle *trans;
6484         unsigned long nr;
6485         u64 mask = root->sectorsize - 1;
6486
6487         ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6488         if (ret)
6489                 return ret;
6490
6491         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6492         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6493
6494         trans = btrfs_start_transaction(root, 5);
6495         if (IS_ERR(trans))
6496                 return PTR_ERR(trans);
6497
6498         btrfs_set_trans_block_group(trans, inode);
6499
6500         ret = btrfs_orphan_add(trans, inode);
6501         if (ret) {
6502                 btrfs_end_transaction(trans, root);
6503                 return ret;
6504         }
6505
6506         nr = trans->blocks_used;
6507         btrfs_end_transaction(trans, root);
6508         btrfs_btree_balance_dirty(root, nr);
6509
6510         /* Now start a transaction for the truncate */
6511         trans = btrfs_start_transaction(root, 0);
6512         if (IS_ERR(trans))
6513                 return PTR_ERR(trans);
6514         btrfs_set_trans_block_group(trans, inode);
6515         trans->block_rsv = root->orphan_block_rsv;
6516
6517         /*
6518          * setattr is responsible for setting the ordered_data_close flag,
6519          * but that is only tested during the last file release.  That
6520          * could happen well after the next commit, leaving a great big
6521          * window where new writes may get lost if someone chooses to write
6522          * to this file after truncating to zero
6523          *
6524          * The inode doesn't have any dirty data here, and so if we commit
6525          * this is a noop.  If someone immediately starts writing to the inode
6526          * it is very likely we'll catch some of their writes in this
6527          * transaction, and the commit will find this file on the ordered
6528          * data list with good things to send down.
6529          *
6530          * This is a best effort solution, there is still a window where
6531          * using truncate to replace the contents of the file will
6532          * end up with a zero length file after a crash.
6533          */
6534         if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6535                 btrfs_add_ordered_operation(trans, root, inode);
6536
6537         while (1) {
6538                 if (!trans) {
6539                         trans = btrfs_start_transaction(root, 0);
6540                         if (IS_ERR(trans))
6541                                 return PTR_ERR(trans);
6542                         btrfs_set_trans_block_group(trans, inode);
6543                         trans->block_rsv = root->orphan_block_rsv;
6544                 }
6545
6546                 ret = btrfs_block_rsv_check(trans, root,
6547                                             root->orphan_block_rsv, 0, 5);
6548                 if (ret == -EAGAIN) {
6549                         ret = btrfs_commit_transaction(trans, root);
6550                         if (ret)
6551                                 return ret;
6552                         trans = NULL;
6553                         continue;
6554                 } else if (ret) {
6555                         err = ret;
6556                         break;
6557                 }
6558
6559                 ret = btrfs_truncate_inode_items(trans, root, inode,
6560                                                  inode->i_size,
6561                                                  BTRFS_EXTENT_DATA_KEY);
6562                 if (ret != -EAGAIN) {
6563                         err = ret;
6564                         break;
6565                 }
6566
6567                 ret = btrfs_update_inode(trans, root, inode);
6568                 if (ret) {
6569                         err = ret;
6570                         break;
6571                 }
6572
6573                 nr = trans->blocks_used;
6574                 btrfs_end_transaction(trans, root);
6575                 trans = NULL;
6576                 btrfs_btree_balance_dirty(root, nr);
6577         }
6578
6579         if (ret == 0 && inode->i_nlink > 0) {
6580                 ret = btrfs_orphan_del(trans, inode);
6581                 if (ret)
6582                         err = ret;
6583         } else if (ret && inode->i_nlink > 0) {
6584                 /*
6585                  * Failed to do the truncate, remove us from the in memory
6586                  * orphan list.
6587                  */
6588                 ret = btrfs_orphan_del(NULL, inode);
6589         }
6590
6591         ret = btrfs_update_inode(trans, root, inode);
6592         if (ret && !err)
6593                 err = ret;
6594
6595         nr = trans->blocks_used;
6596         ret = btrfs_end_transaction_throttle(trans, root);
6597         if (ret && !err)
6598                 err = ret;
6599         btrfs_btree_balance_dirty(root, nr);
6600
6601         return err;
6602 }
6603
6604 /*
6605  * create a new subvolume directory/inode (helper for the ioctl).
6606  */
6607 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6608                              struct btrfs_root *new_root,
6609                              u64 new_dirid, u64 alloc_hint)
6610 {
6611         struct inode *inode;
6612         int err;
6613         u64 index = 0;
6614
6615         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6616                                 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
6617         if (IS_ERR(inode))
6618                 return PTR_ERR(inode);
6619         inode->i_op = &btrfs_dir_inode_operations;
6620         inode->i_fop = &btrfs_dir_file_operations;
6621
6622         inode->i_nlink = 1;
6623         btrfs_i_size_write(inode, 0);
6624
6625         err = btrfs_update_inode(trans, new_root, inode);
6626         BUG_ON(err);
6627
6628         iput(inode);
6629         return 0;
6630 }
6631
6632 /* helper function for file defrag and space balancing.  This
6633  * forces readahead on a given range of bytes in an inode
6634  */
6635 unsigned long btrfs_force_ra(struct address_space *mapping,
6636                               struct file_ra_state *ra, struct file *file,
6637                               pgoff_t offset, pgoff_t last_index)
6638 {
6639         pgoff_t req_size = last_index - offset + 1;
6640
6641         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
6642         return offset + req_size;
6643 }
6644
6645 struct inode *btrfs_alloc_inode(struct super_block *sb)
6646 {
6647         struct btrfs_inode *ei;
6648         struct inode *inode;
6649
6650         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6651         if (!ei)
6652                 return NULL;
6653
6654         ei->root = NULL;
6655         ei->space_info = NULL;
6656         ei->generation = 0;
6657         ei->sequence = 0;
6658         ei->last_trans = 0;
6659         ei->last_sub_trans = 0;
6660         ei->logged_trans = 0;
6661         ei->delalloc_bytes = 0;
6662         ei->reserved_bytes = 0;
6663         ei->disk_i_size = 0;
6664         ei->flags = 0;
6665         ei->index_cnt = (u64)-1;
6666         ei->last_unlink_trans = 0;
6667
6668         atomic_set(&ei->outstanding_extents, 0);
6669         atomic_set(&ei->reserved_extents, 0);
6670
6671         ei->ordered_data_close = 0;
6672         ei->orphan_meta_reserved = 0;
6673         ei->dummy_inode = 0;
6674         ei->force_compress = BTRFS_COMPRESS_NONE;
6675
6676         inode = &ei->vfs_inode;
6677         extent_map_tree_init(&ei->extent_tree, GFP_NOFS);
6678         extent_io_tree_init(&ei->io_tree, &inode->i_data, GFP_NOFS);
6679         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data, GFP_NOFS);
6680         mutex_init(&ei->log_mutex);
6681         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6682         INIT_LIST_HEAD(&ei->i_orphan);
6683         INIT_LIST_HEAD(&ei->delalloc_inodes);
6684         INIT_LIST_HEAD(&ei->ordered_operations);
6685         RB_CLEAR_NODE(&ei->rb_node);
6686
6687         return inode;
6688 }
6689
6690 static void btrfs_i_callback(struct rcu_head *head)
6691 {
6692         struct inode *inode = container_of(head, struct inode, i_rcu);
6693         INIT_LIST_HEAD(&inode->i_dentry);
6694         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6695 }
6696
6697 void btrfs_destroy_inode(struct inode *inode)
6698 {
6699         struct btrfs_ordered_extent *ordered;
6700         struct btrfs_root *root = BTRFS_I(inode)->root;
6701
6702         WARN_ON(!list_empty(&inode->i_dentry));
6703         WARN_ON(inode->i_data.nrpages);
6704         WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents));
6705         WARN_ON(atomic_read(&BTRFS_I(inode)->reserved_extents));
6706
6707         /*
6708          * This can happen where we create an inode, but somebody else also
6709          * created the same inode and we need to destroy the one we already
6710          * created.
6711          */
6712         if (!root)
6713                 goto free;
6714
6715         /*
6716          * Make sure we're properly removed from the ordered operation
6717          * lists.
6718          */
6719         smp_mb();
6720         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6721                 spin_lock(&root->fs_info->ordered_extent_lock);
6722                 list_del_init(&BTRFS_I(inode)->ordered_operations);
6723                 spin_unlock(&root->fs_info->ordered_extent_lock);
6724         }
6725
6726         if (root == root->fs_info->tree_root) {
6727                 struct btrfs_block_group_cache *block_group;
6728
6729                 block_group = btrfs_lookup_block_group(root->fs_info,
6730                                                 BTRFS_I(inode)->block_group);
6731                 if (block_group && block_group->inode == inode) {
6732                         spin_lock(&block_group->lock);
6733                         block_group->inode = NULL;
6734                         spin_unlock(&block_group->lock);
6735                         btrfs_put_block_group(block_group);
6736                 } else if (block_group) {
6737                         btrfs_put_block_group(block_group);
6738                 }
6739         }
6740
6741         spin_lock(&root->orphan_lock);
6742         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6743                 printk(KERN_INFO "BTRFS: inode %lu still on the orphan list\n",
6744                        inode->i_ino);
6745                 list_del_init(&BTRFS_I(inode)->i_orphan);
6746         }
6747         spin_unlock(&root->orphan_lock);
6748
6749         while (1) {
6750                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6751                 if (!ordered)
6752                         break;
6753                 else {
6754                         printk(KERN_ERR "btrfs found ordered "
6755                                "extent %llu %llu on inode cleanup\n",
6756                                (unsigned long long)ordered->file_offset,
6757                                (unsigned long long)ordered->len);
6758                         btrfs_remove_ordered_extent(inode, ordered);
6759                         btrfs_put_ordered_extent(ordered);
6760                         btrfs_put_ordered_extent(ordered);
6761                 }
6762         }
6763         inode_tree_del(inode);
6764         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6765 free:
6766         call_rcu(&inode->i_rcu, btrfs_i_callback);
6767 }
6768
6769 int btrfs_drop_inode(struct inode *inode)
6770 {
6771         struct btrfs_root *root = BTRFS_I(inode)->root;
6772
6773         if (btrfs_root_refs(&root->root_item) == 0 &&
6774             root != root->fs_info->tree_root)
6775                 return 1;
6776         else
6777                 return generic_drop_inode(inode);
6778 }
6779
6780 static void init_once(void *foo)
6781 {
6782         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6783
6784         inode_init_once(&ei->vfs_inode);
6785 }
6786
6787 void btrfs_destroy_cachep(void)
6788 {
6789         if (btrfs_inode_cachep)
6790                 kmem_cache_destroy(btrfs_inode_cachep);
6791         if (btrfs_trans_handle_cachep)
6792                 kmem_cache_destroy(btrfs_trans_handle_cachep);
6793         if (btrfs_transaction_cachep)
6794                 kmem_cache_destroy(btrfs_transaction_cachep);
6795         if (btrfs_path_cachep)
6796                 kmem_cache_destroy(btrfs_path_cachep);
6797         if (btrfs_free_space_cachep)
6798                 kmem_cache_destroy(btrfs_free_space_cachep);
6799 }
6800
6801 int btrfs_init_cachep(void)
6802 {
6803         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6804                         sizeof(struct btrfs_inode), 0,
6805                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6806         if (!btrfs_inode_cachep)
6807                 goto fail;
6808
6809         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6810                         sizeof(struct btrfs_trans_handle), 0,
6811                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6812         if (!btrfs_trans_handle_cachep)
6813                 goto fail;
6814
6815         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6816                         sizeof(struct btrfs_transaction), 0,
6817                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6818         if (!btrfs_transaction_cachep)
6819                 goto fail;
6820
6821         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6822                         sizeof(struct btrfs_path), 0,
6823                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6824         if (!btrfs_path_cachep)
6825                 goto fail;
6826
6827         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6828                         sizeof(struct btrfs_free_space), 0,
6829                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6830         if (!btrfs_free_space_cachep)
6831                 goto fail;
6832
6833         return 0;
6834 fail:
6835         btrfs_destroy_cachep();
6836         return -ENOMEM;
6837 }
6838
6839 static int btrfs_getattr(struct vfsmount *mnt,
6840                          struct dentry *dentry, struct kstat *stat)
6841 {
6842         struct inode *inode = dentry->d_inode;
6843         generic_fillattr(inode, stat);
6844         stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
6845         stat->blksize = PAGE_CACHE_SIZE;
6846         stat->blocks = (inode_get_bytes(inode) +
6847                         BTRFS_I(inode)->delalloc_bytes) >> 9;
6848         return 0;
6849 }
6850
6851 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6852                            struct inode *new_dir, struct dentry *new_dentry)
6853 {
6854         struct btrfs_trans_handle *trans;
6855         struct btrfs_root *root = BTRFS_I(old_dir)->root;
6856         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6857         struct inode *new_inode = new_dentry->d_inode;
6858         struct inode *old_inode = old_dentry->d_inode;
6859         struct timespec ctime = CURRENT_TIME;
6860         u64 index = 0;
6861         u64 root_objectid;
6862         int ret;
6863
6864         if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6865                 return -EPERM;
6866
6867         /* we only allow rename subvolume link between subvolumes */
6868         if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6869                 return -EXDEV;
6870
6871         if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6872             (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID))
6873                 return -ENOTEMPTY;
6874
6875         if (S_ISDIR(old_inode->i_mode) && new_inode &&
6876             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6877                 return -ENOTEMPTY;
6878         /*
6879          * we're using rename to replace one file with another.
6880          * and the replacement file is large.  Start IO on it now so
6881          * we don't add too much work to the end of the transaction
6882          */
6883         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
6884             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6885                 filemap_flush(old_inode->i_mapping);
6886
6887         /* close the racy window with snapshot create/destroy ioctl */
6888         if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
6889                 down_read(&root->fs_info->subvol_sem);
6890         /*
6891          * We want to reserve the absolute worst case amount of items.  So if
6892          * both inodes are subvols and we need to unlink them then that would
6893          * require 4 item modifications, but if they are both normal inodes it
6894          * would require 5 item modifications, so we'll assume their normal
6895          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6896          * should cover the worst case number of items we'll modify.
6897          */
6898         trans = btrfs_start_transaction(root, 20);
6899         if (IS_ERR(trans))
6900                 return PTR_ERR(trans);
6901
6902         btrfs_set_trans_block_group(trans, new_dir);
6903
6904         if (dest != root)
6905                 btrfs_record_root_in_trans(trans, dest);
6906
6907         ret = btrfs_set_inode_index(new_dir, &index);
6908         if (ret)
6909                 goto out_fail;
6910
6911         if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6912                 /* force full log commit if subvolume involved. */
6913                 root->fs_info->last_trans_log_full_commit = trans->transid;
6914         } else {
6915                 ret = btrfs_insert_inode_ref(trans, dest,
6916                                              new_dentry->d_name.name,
6917                                              new_dentry->d_name.len,
6918                                              old_inode->i_ino,
6919                                              new_dir->i_ino, index);
6920                 if (ret)
6921                         goto out_fail;
6922                 /*
6923                  * this is an ugly little race, but the rename is required
6924                  * to make sure that if we crash, the inode is either at the
6925                  * old name or the new one.  pinning the log transaction lets
6926                  * us make sure we don't allow a log commit to come in after
6927                  * we unlink the name but before we add the new name back in.
6928                  */
6929                 btrfs_pin_log_trans(root);
6930         }
6931         /*
6932          * make sure the inode gets flushed if it is replacing
6933          * something.
6934          */
6935         if (new_inode && new_inode->i_size &&
6936             old_inode && S_ISREG(old_inode->i_mode)) {
6937                 btrfs_add_ordered_operation(trans, root, old_inode);
6938         }
6939
6940         old_dir->i_ctime = old_dir->i_mtime = ctime;
6941         new_dir->i_ctime = new_dir->i_mtime = ctime;
6942         old_inode->i_ctime = ctime;
6943
6944         if (old_dentry->d_parent != new_dentry->d_parent)
6945                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
6946
6947         if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6948                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
6949                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
6950                                         old_dentry->d_name.name,
6951                                         old_dentry->d_name.len);
6952         } else {
6953                 btrfs_inc_nlink(old_dentry->d_inode);
6954                 ret = btrfs_unlink_inode(trans, root, old_dir,
6955                                          old_dentry->d_inode,
6956                                          old_dentry->d_name.name,
6957                                          old_dentry->d_name.len);
6958         }
6959         BUG_ON(ret);
6960
6961         if (new_inode) {
6962                 new_inode->i_ctime = CURRENT_TIME;
6963                 if (unlikely(new_inode->i_ino ==
6964                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
6965                         root_objectid = BTRFS_I(new_inode)->location.objectid;
6966                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
6967                                                 root_objectid,
6968                                                 new_dentry->d_name.name,
6969                                                 new_dentry->d_name.len);
6970                         BUG_ON(new_inode->i_nlink == 0);
6971                 } else {
6972                         ret = btrfs_unlink_inode(trans, dest, new_dir,
6973                                                  new_dentry->d_inode,
6974                                                  new_dentry->d_name.name,
6975                                                  new_dentry->d_name.len);
6976                 }
6977                 BUG_ON(ret);
6978                 if (new_inode->i_nlink == 0) {
6979                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
6980                         BUG_ON(ret);
6981                 }
6982         }
6983
6984         ret = btrfs_add_link(trans, new_dir, old_inode,
6985                              new_dentry->d_name.name,
6986                              new_dentry->d_name.len, 0, index);
6987         BUG_ON(ret);
6988
6989         if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
6990                 struct dentry *parent = dget_parent(new_dentry);
6991                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
6992                 dput(parent);
6993                 btrfs_end_log_trans(root);
6994         }
6995 out_fail:
6996         btrfs_end_transaction_throttle(trans, root);
6997
6998         if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
6999                 up_read(&root->fs_info->subvol_sem);
7000
7001         return ret;
7002 }
7003
7004 /*
7005  * some fairly slow code that needs optimization. This walks the list
7006  * of all the inodes with pending delalloc and forces them to disk.
7007  */
7008 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7009 {
7010         struct list_head *head = &root->fs_info->delalloc_inodes;
7011         struct btrfs_inode *binode;
7012         struct inode *inode;
7013
7014         if (root->fs_info->sb->s_flags & MS_RDONLY)
7015                 return -EROFS;
7016
7017         spin_lock(&root->fs_info->delalloc_lock);
7018         while (!list_empty(head)) {
7019                 binode = list_entry(head->next, struct btrfs_inode,
7020                                     delalloc_inodes);
7021                 inode = igrab(&binode->vfs_inode);
7022                 if (!inode)
7023                         list_del_init(&binode->delalloc_inodes);
7024                 spin_unlock(&root->fs_info->delalloc_lock);
7025                 if (inode) {
7026                         filemap_flush(inode->i_mapping);
7027                         if (delay_iput)
7028                                 btrfs_add_delayed_iput(inode);
7029                         else
7030                                 iput(inode);
7031                 }
7032                 cond_resched();
7033                 spin_lock(&root->fs_info->delalloc_lock);
7034         }
7035         spin_unlock(&root->fs_info->delalloc_lock);
7036
7037         /* the filemap_flush will queue IO into the worker threads, but
7038          * we have to make sure the IO is actually started and that
7039          * ordered extents get created before we return
7040          */
7041         atomic_inc(&root->fs_info->async_submit_draining);
7042         while (atomic_read(&root->fs_info->nr_async_submits) ||
7043               atomic_read(&root->fs_info->async_delalloc_pages)) {
7044                 wait_event(root->fs_info->async_submit_wait,
7045                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7046                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7047         }
7048         atomic_dec(&root->fs_info->async_submit_draining);
7049         return 0;
7050 }
7051
7052 int btrfs_start_one_delalloc_inode(struct btrfs_root *root, int delay_iput,
7053                                    int sync)
7054 {
7055         struct btrfs_inode *binode;
7056         struct inode *inode = NULL;
7057
7058         spin_lock(&root->fs_info->delalloc_lock);
7059         while (!list_empty(&root->fs_info->delalloc_inodes)) {
7060                 binode = list_entry(root->fs_info->delalloc_inodes.next,
7061                                     struct btrfs_inode, delalloc_inodes);
7062                 inode = igrab(&binode->vfs_inode);
7063                 if (inode) {
7064                         list_move_tail(&binode->delalloc_inodes,
7065                                        &root->fs_info->delalloc_inodes);
7066                         break;
7067                 }
7068
7069                 list_del_init(&binode->delalloc_inodes);
7070                 cond_resched_lock(&root->fs_info->delalloc_lock);
7071         }
7072         spin_unlock(&root->fs_info->delalloc_lock);
7073
7074         if (inode) {
7075                 if (sync) {
7076                         filemap_write_and_wait(inode->i_mapping);
7077                         /*
7078                          * We have to do this because compression doesn't
7079                          * actually set PG_writeback until it submits the pages
7080                          * for IO, which happens in an async thread, so we could
7081                          * race and not actually wait for any writeback pages
7082                          * because they've not been submitted yet.  Technically
7083                          * this could still be the case for the ordered stuff
7084                          * since the async thread may not have started to do its
7085                          * work yet.  If this becomes the case then we need to
7086                          * figure out a way to make sure that in writepage we
7087                          * wait for any async pages to be submitted before
7088                          * returning so that fdatawait does what its supposed to
7089                          * do.
7090                          */
7091                         btrfs_wait_ordered_range(inode, 0, (u64)-1);
7092                 } else {
7093                         filemap_flush(inode->i_mapping);
7094                 }
7095                 if (delay_iput)
7096                         btrfs_add_delayed_iput(inode);
7097                 else
7098                         iput(inode);
7099                 return 1;
7100         }
7101         return 0;
7102 }
7103
7104 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7105                          const char *symname)
7106 {
7107         struct btrfs_trans_handle *trans;
7108         struct btrfs_root *root = BTRFS_I(dir)->root;
7109         struct btrfs_path *path;
7110         struct btrfs_key key;
7111         struct inode *inode = NULL;
7112         int err;
7113         int drop_inode = 0;
7114         u64 objectid;
7115         u64 index = 0 ;
7116         int name_len;
7117         int datasize;
7118         unsigned long ptr;
7119         struct btrfs_file_extent_item *ei;
7120         struct extent_buffer *leaf;
7121         unsigned long nr = 0;
7122
7123         name_len = strlen(symname) + 1;
7124         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7125                 return -ENAMETOOLONG;
7126
7127         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
7128         if (err)
7129                 return err;
7130         /*
7131          * 2 items for inode item and ref
7132          * 2 items for dir items
7133          * 1 item for xattr if selinux is on
7134          */
7135         trans = btrfs_start_transaction(root, 5);
7136         if (IS_ERR(trans))
7137                 return PTR_ERR(trans);
7138
7139         btrfs_set_trans_block_group(trans, dir);
7140
7141         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7142                                 dentry->d_name.len, dir->i_ino, objectid,
7143                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
7144                                 &index);
7145         err = PTR_ERR(inode);
7146         if (IS_ERR(inode))
7147                 goto out_unlock;
7148
7149         err = btrfs_init_inode_security(trans, inode, dir);
7150         if (err) {
7151                 drop_inode = 1;
7152                 goto out_unlock;
7153         }
7154
7155         btrfs_set_trans_block_group(trans, inode);
7156         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7157         if (err)
7158                 drop_inode = 1;
7159         else {
7160                 inode->i_mapping->a_ops = &btrfs_aops;
7161                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7162                 inode->i_fop = &btrfs_file_operations;
7163                 inode->i_op = &btrfs_file_inode_operations;
7164                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7165         }
7166         btrfs_update_inode_block_group(trans, inode);
7167         btrfs_update_inode_block_group(trans, dir);
7168         if (drop_inode)
7169                 goto out_unlock;
7170
7171         path = btrfs_alloc_path();
7172         BUG_ON(!path);
7173         key.objectid = inode->i_ino;
7174         key.offset = 0;
7175         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7176         datasize = btrfs_file_extent_calc_inline_size(name_len);
7177         err = btrfs_insert_empty_item(trans, root, path, &key,
7178                                       datasize);
7179         if (err) {
7180                 drop_inode = 1;
7181                 goto out_unlock;
7182         }
7183         leaf = path->nodes[0];
7184         ei = btrfs_item_ptr(leaf, path->slots[0],
7185                             struct btrfs_file_extent_item);
7186         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7187         btrfs_set_file_extent_type(leaf, ei,
7188                                    BTRFS_FILE_EXTENT_INLINE);
7189         btrfs_set_file_extent_encryption(leaf, ei, 0);
7190         btrfs_set_file_extent_compression(leaf, ei, 0);
7191         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7192         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7193
7194         ptr = btrfs_file_extent_inline_start(ei);
7195         write_extent_buffer(leaf, symname, ptr, name_len);
7196         btrfs_mark_buffer_dirty(leaf);
7197         btrfs_free_path(path);
7198
7199         inode->i_op = &btrfs_symlink_inode_operations;
7200         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7201         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7202         inode_set_bytes(inode, name_len);
7203         btrfs_i_size_write(inode, name_len - 1);
7204         err = btrfs_update_inode(trans, root, inode);
7205         if (err)
7206                 drop_inode = 1;
7207
7208 out_unlock:
7209         nr = trans->blocks_used;
7210         btrfs_end_transaction_throttle(trans, root);
7211         if (drop_inode) {
7212                 inode_dec_link_count(inode);
7213                 iput(inode);
7214         }
7215         btrfs_btree_balance_dirty(root, nr);
7216         return err;
7217 }
7218
7219 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7220                                        u64 start, u64 num_bytes, u64 min_size,
7221                                        loff_t actual_len, u64 *alloc_hint,
7222                                        struct btrfs_trans_handle *trans)
7223 {
7224         struct btrfs_root *root = BTRFS_I(inode)->root;
7225         struct btrfs_key ins;
7226         u64 cur_offset = start;
7227         u64 i_size;
7228         int ret = 0;
7229         bool own_trans = true;
7230
7231         if (trans)
7232                 own_trans = false;
7233         while (num_bytes > 0) {
7234                 if (own_trans) {
7235                         trans = btrfs_start_transaction(root, 3);
7236                         if (IS_ERR(trans)) {
7237                                 ret = PTR_ERR(trans);
7238                                 break;
7239                         }
7240                 }
7241
7242                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7243                                            0, *alloc_hint, (u64)-1, &ins, 1);
7244                 if (ret) {
7245                         if (own_trans)
7246                                 btrfs_end_transaction(trans, root);
7247                         break;
7248                 }
7249
7250                 ret = insert_reserved_file_extent(trans, inode,
7251                                                   cur_offset, ins.objectid,
7252                                                   ins.offset, ins.offset,
7253                                                   ins.offset, 0, 0, 0,
7254                                                   BTRFS_FILE_EXTENT_PREALLOC);
7255                 BUG_ON(ret);
7256                 btrfs_drop_extent_cache(inode, cur_offset,
7257                                         cur_offset + ins.offset -1, 0);
7258
7259                 num_bytes -= ins.offset;
7260                 cur_offset += ins.offset;
7261                 *alloc_hint = ins.objectid + ins.offset;
7262
7263                 inode->i_ctime = CURRENT_TIME;
7264                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7265                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7266                     (actual_len > inode->i_size) &&
7267                     (cur_offset > inode->i_size)) {
7268                         if (cur_offset > actual_len)
7269                                 i_size = actual_len;
7270                         else
7271                                 i_size = cur_offset;
7272                         i_size_write(inode, i_size);
7273                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7274                 }
7275
7276                 ret = btrfs_update_inode(trans, root, inode);
7277                 BUG_ON(ret);
7278
7279                 if (own_trans)
7280                         btrfs_end_transaction(trans, root);
7281         }
7282         return ret;
7283 }
7284
7285 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7286                               u64 start, u64 num_bytes, u64 min_size,
7287                               loff_t actual_len, u64 *alloc_hint)
7288 {
7289         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7290                                            min_size, actual_len, alloc_hint,
7291                                            NULL);
7292 }
7293
7294 int btrfs_prealloc_file_range_trans(struct inode *inode,
7295                                     struct btrfs_trans_handle *trans, int mode,
7296                                     u64 start, u64 num_bytes, u64 min_size,
7297                                     loff_t actual_len, u64 *alloc_hint)
7298 {
7299         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7300                                            min_size, actual_len, alloc_hint, trans);
7301 }
7302
7303 static int btrfs_set_page_dirty(struct page *page)
7304 {
7305         return __set_page_dirty_nobuffers(page);
7306 }
7307
7308 static int btrfs_permission(struct inode *inode, int mask, unsigned int flags)
7309 {
7310         struct btrfs_root *root = BTRFS_I(inode)->root;
7311
7312         if (btrfs_root_readonly(root) && (mask & MAY_WRITE))
7313                 return -EROFS;
7314         if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
7315                 return -EACCES;
7316         return generic_permission(inode, mask, flags, btrfs_check_acl);
7317 }
7318
7319 static const struct inode_operations btrfs_dir_inode_operations = {
7320         .getattr        = btrfs_getattr,
7321         .lookup         = btrfs_lookup,
7322         .create         = btrfs_create,
7323         .unlink         = btrfs_unlink,
7324         .link           = btrfs_link,
7325         .mkdir          = btrfs_mkdir,
7326         .rmdir          = btrfs_rmdir,
7327         .rename         = btrfs_rename,
7328         .symlink        = btrfs_symlink,
7329         .setattr        = btrfs_setattr,
7330         .mknod          = btrfs_mknod,
7331         .setxattr       = btrfs_setxattr,
7332         .getxattr       = btrfs_getxattr,
7333         .listxattr      = btrfs_listxattr,
7334         .removexattr    = btrfs_removexattr,
7335         .permission     = btrfs_permission,
7336 };
7337 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7338         .lookup         = btrfs_lookup,
7339         .permission     = btrfs_permission,
7340 };
7341
7342 static const struct file_operations btrfs_dir_file_operations = {
7343         .llseek         = generic_file_llseek,
7344         .read           = generic_read_dir,
7345         .readdir        = btrfs_real_readdir,
7346         .unlocked_ioctl = btrfs_ioctl,
7347 #ifdef CONFIG_COMPAT
7348         .compat_ioctl   = btrfs_ioctl,
7349 #endif
7350         .release        = btrfs_release_file,
7351         .fsync          = btrfs_sync_file,
7352 };
7353
7354 static struct extent_io_ops btrfs_extent_io_ops = {
7355         .fill_delalloc = run_delalloc_range,
7356         .submit_bio_hook = btrfs_submit_bio_hook,
7357         .merge_bio_hook = btrfs_merge_bio_hook,
7358         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7359         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7360         .writepage_start_hook = btrfs_writepage_start_hook,
7361         .readpage_io_failed_hook = btrfs_io_failed_hook,
7362         .set_bit_hook = btrfs_set_bit_hook,
7363         .clear_bit_hook = btrfs_clear_bit_hook,
7364         .merge_extent_hook = btrfs_merge_extent_hook,
7365         .split_extent_hook = btrfs_split_extent_hook,
7366 };
7367
7368 /*
7369  * btrfs doesn't support the bmap operation because swapfiles
7370  * use bmap to make a mapping of extents in the file.  They assume
7371  * these extents won't change over the life of the file and they
7372  * use the bmap result to do IO directly to the drive.
7373  *
7374  * the btrfs bmap call would return logical addresses that aren't
7375  * suitable for IO and they also will change frequently as COW
7376  * operations happen.  So, swapfile + btrfs == corruption.
7377  *
7378  * For now we're avoiding this by dropping bmap.
7379  */
7380 static const struct address_space_operations btrfs_aops = {
7381         .readpage       = btrfs_readpage,
7382         .writepage      = btrfs_writepage,
7383         .writepages     = btrfs_writepages,
7384         .readpages      = btrfs_readpages,
7385         .sync_page      = block_sync_page,
7386         .direct_IO      = btrfs_direct_IO,
7387         .invalidatepage = btrfs_invalidatepage,
7388         .releasepage    = btrfs_releasepage,
7389         .set_page_dirty = btrfs_set_page_dirty,
7390         .error_remove_page = generic_error_remove_page,
7391 };
7392
7393 static const struct address_space_operations btrfs_symlink_aops = {
7394         .readpage       = btrfs_readpage,
7395         .writepage      = btrfs_writepage,
7396         .invalidatepage = btrfs_invalidatepage,
7397         .releasepage    = btrfs_releasepage,
7398 };
7399
7400 static const struct inode_operations btrfs_file_inode_operations = {
7401         .getattr        = btrfs_getattr,
7402         .setattr        = btrfs_setattr,
7403         .setxattr       = btrfs_setxattr,
7404         .getxattr       = btrfs_getxattr,
7405         .listxattr      = btrfs_listxattr,
7406         .removexattr    = btrfs_removexattr,
7407         .permission     = btrfs_permission,
7408         .fiemap         = btrfs_fiemap,
7409 };
7410 static const struct inode_operations btrfs_special_inode_operations = {
7411         .getattr        = btrfs_getattr,
7412         .setattr        = btrfs_setattr,
7413         .permission     = btrfs_permission,
7414         .setxattr       = btrfs_setxattr,
7415         .getxattr       = btrfs_getxattr,
7416         .listxattr      = btrfs_listxattr,
7417         .removexattr    = btrfs_removexattr,
7418 };
7419 static const struct inode_operations btrfs_symlink_inode_operations = {
7420         .readlink       = generic_readlink,
7421         .follow_link    = page_follow_link_light,
7422         .put_link       = page_put_link,
7423         .getattr        = btrfs_getattr,
7424         .permission     = btrfs_permission,
7425         .setxattr       = btrfs_setxattr,
7426         .getxattr       = btrfs_getxattr,
7427         .listxattr      = btrfs_listxattr,
7428         .removexattr    = btrfs_removexattr,
7429 };
7430
7431 const struct dentry_operations btrfs_dentry_operations = {
7432         .d_delete       = btrfs_dentry_delete,
7433 };