fs/btrfs/inode.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/pagemap.h>
  25 #include <linux/highmem.h>
  26 #include <linux/time.h>
  27 #include <linux/init.h>
  28 #include <linux/string.h>
  29 #include <linux/backing-dev.h>
  30 #include <linux/mpage.h>
  31 #include <linux/swap.h>
  32 #include <linux/writeback.h>
  33 #include <linux/statfs.h>
  34 #include <linux/compat.h>
  35 #include <linux/bit_spinlock.h>
  36 #include <linux/xattr.h>
  37 #include <linux/posix_acl.h>
  38 #include <linux/falloc.h>
  39 #include <linux/slab.h>
  40 #include <linux/ratelimit.h>
  41 #include "compat.h"
  42 #include "ctree.h"
  43 #include "disk-io.h"
  44 #include "transaction.h"
  45 #include "btrfs_inode.h"
  46 #include "ioctl.h"
  47 #include "print-tree.h"
  48 #include "volumes.h"
  49 #include "ordered-data.h"
  50 #include "xattr.h"
  51 #include "tree-log.h"
  52 #include "compression.h"
  53 #include "locking.h"
  54 #include "free-space-cache.h"
  55 #include "inode-map.h"
  56
  57 struct btrfs_iget_args {
  58         u64 ino;
  59         struct btrfs_root *root;
  60 };
  61
  62 static const struct inode_operations btrfs_dir_inode_operations;
  63 static const struct inode_operations btrfs_symlink_inode_operations;
  64 static const struct inode_operations btrfs_dir_ro_inode_operations;
  65 static const struct inode_operations btrfs_special_inode_operations;
  66 static const struct inode_operations btrfs_file_inode_operations;
  67 static const struct address_space_operations btrfs_aops;
  68 static const struct address_space_operations btrfs_symlink_aops;
  69 static const struct file_operations btrfs_dir_file_operations;
  70 static struct extent_io_ops btrfs_extent_io_ops;
  71
  72 static struct kmem_cache *btrfs_inode_cachep;
  73 struct kmem_cache *btrfs_trans_handle_cachep;
  74 struct kmem_cache *btrfs_transaction_cachep;
  75 struct kmem_cache *btrfs_path_cachep;
  76 struct kmem_cache *btrfs_free_space_cachep;
  77
  78 #define S_SHIFT 12
  79 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  80         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  81         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  82         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  83         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  84         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  85         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  86         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  87 };
  88
  89 static int btrfs_setsize(struct inode *inode, loff_t newsize);
  90 static int btrfs_truncate(struct inode *inode);
  91 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
  92 static noinline int cow_file_range(struct inode *inode,
  93                                    struct page *locked_page,
  94                                    u64 start, u64 end, int *page_started,
  95                                    unsigned long *nr_written, int unlock);
  96
  97 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
  98                                      struct inode *inode,  struct inode *dir,
  99                                      const struct qstr *qstr)
 100 {
 101         int err;
 102
 103         err = btrfs_init_acl(trans, inode, dir);
 104         if (!err)
 105                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 106         return err;
 107 }
 108
 109 /*
 110  * this does all the hard work for inserting an inline extent into
 111  * the btree.  The caller should have done a btrfs_drop_extents so that
 112  * no overlapping inline items exist in the btree
 113  */
 114 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 115                                 struct btrfs_root *root, struct inode *inode,
 116                                 u64 start, size_t size, size_t compressed_size,
 117                                 int compress_type,
 118                                 struct page **compressed_pages)
 119 {
 120         struct btrfs_key key;
 121         struct btrfs_path *path;
 122         struct extent_buffer *leaf;
 123         struct page *page = NULL;
 124         char *kaddr;
 125         unsigned long ptr;
 126         struct btrfs_file_extent_item *ei;
 127         int err = 0;
 128         int ret;
 129         size_t cur_size = size;
 130         size_t datasize;
 131         unsigned long offset;
 132
 133         if (compressed_size && compressed_pages)
 134                 cur_size = compressed_size;
 135
 136         path = btrfs_alloc_path();
 137         if (!path)
 138                 return -ENOMEM;
 139
 140         path->leave_spinning = 1;
 141
 142         key.objectid = btrfs_ino(inode);
 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, int compress_type,
 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                                    compress_type, 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         BUG_ON(!async_extent);
 294         async_extent->start = start;
 295         async_extent->ram_size = ram_size;
 296         async_extent->compressed_size = compressed_size;
 297         async_extent->pages = pages;
 298         async_extent->nr_pages = nr_pages;
 299         async_extent->compress_type = compress_type;
 300         list_add_tail(&async_extent->list, &cow->extents);
 301         return 0;
 302 }
 303
 304 /*
 305  * we create compressed extents in two phases.  The first
 306  * phase compresses a range of pages that have already been
 307  * locked (both pages and state bits are locked).
 308  *
 309  * This is done inside an ordered work queue, and the compression
 310  * is spread across many cpus.  The actual IO submission is step
 311  * two, and the ordered work queue takes care of making sure that
 312  * happens in the same order things were put onto the queue by
 313  * writepages and friends.
 314  *
 315  * If this code finds it can't get good compression, it puts an
 316  * entry onto the work queue to write the uncompressed bytes.  This
 317  * makes sure that both compressed inodes and uncompressed inodes
 318  * are written in the same order that pdflush sent them down.
 319  */
 320 static noinline int compress_file_range(struct inode *inode,
 321                                         struct page *locked_page,
 322                                         u64 start, u64 end,
 323                                         struct async_cow *async_cow,
 324                                         int *num_added)
 325 {
 326         struct btrfs_root *root = BTRFS_I(inode)->root;
 327         struct btrfs_trans_handle *trans;
 328         u64 num_bytes;
 329         u64 blocksize = root->sectorsize;
 330         u64 actual_end;
 331         u64 isize = i_size_read(inode);
 332         int ret = 0;
 333         struct page **pages = NULL;
 334         unsigned long nr_pages;
 335         unsigned long nr_pages_ret = 0;
 336         unsigned long total_compressed = 0;
 337         unsigned long total_in = 0;
 338         unsigned long max_compressed = 128 * 1024;
 339         unsigned long max_uncompressed = 128 * 1024;
 340         int i;
 341         int will_compress;
 342         int compress_type = root->fs_info->compress_type;
 343
 344         /* if this is a small write inside eof, kick off a defragbot */
 345         if (end <= BTRFS_I(inode)->disk_i_size && (end - start + 1) < 16 * 1024)
 346                 btrfs_add_inode_defrag(NULL, inode);
 347
 348         actual_end = min_t(u64, isize, end + 1);
 349 again:
 350         will_compress = 0;
 351         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 352         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 353
 354         /*
 355          * we don't want to send crud past the end of i_size through
 356          * compression, that's just a waste of CPU time.  So, if the
 357          * end of the file is before the start of our current
 358          * requested range of bytes, we bail out to the uncompressed
 359          * cleanup code that can deal with all of this.
 360          *
 361          * It isn't really the fastest way to fix things, but this is a
 362          * very uncommon corner.
 363          */
 364         if (actual_end <= start)
 365                 goto cleanup_and_bail_uncompressed;
 366
 367         total_compressed = actual_end - start;
 368
 369         /* we want to make sure that amount of ram required to uncompress
 370          * an extent is reasonable, so we limit the total size in ram
 371          * of a compressed extent to 128k.  This is a crucial number
 372          * because it also controls how easily we can spread reads across
 373          * cpus for decompression.
 374          *
 375          * We also want to make sure the amount of IO required to do
 376          * a random read is reasonably small, so we limit the size of
 377          * a compressed extent to 128k.
 378          */
 379         total_compressed = min(total_compressed, max_uncompressed);
 380         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 381         num_bytes = max(blocksize,  num_bytes);
 382         total_in = 0;
 383         ret = 0;
 384
 385         /*
 386          * we do compression for mount -o compress and when the
 387          * inode has not been flagged as nocompress.  This flag can
 388          * change at any time if we discover bad compression ratios.
 389          */
 390         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 391             (btrfs_test_opt(root, COMPRESS) ||
 392              (BTRFS_I(inode)->force_compress) ||
 393              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 394                 WARN_ON(pages);
 395                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 396                 BUG_ON(!pages);
 397
 398                 if (BTRFS_I(inode)->force_compress)
 399                         compress_type = BTRFS_I(inode)->force_compress;
 400
 401                 ret = btrfs_compress_pages(compress_type,
 402                                            inode->i_mapping, start,
 403                                            total_compressed, pages,
 404                                            nr_pages, &nr_pages_ret,
 405                                            &total_in,
 406                                            &total_compressed,
 407                                            max_compressed);
 408
 409                 if (!ret) {
 410                         unsigned long offset = total_compressed &
 411                                 (PAGE_CACHE_SIZE - 1);
 412                         struct page *page = pages[nr_pages_ret - 1];
 413                         char *kaddr;
 414
 415                         /* zero the tail end of the last page, we might be
 416                          * sending it down to disk
 417                          */
 418                         if (offset) {
 419                                 kaddr = kmap_atomic(page, KM_USER0);
 420                                 memset(kaddr + offset, 0,
 421                                        PAGE_CACHE_SIZE - offset);
 422                                 kunmap_atomic(kaddr, KM_USER0);
 423                         }
 424                         will_compress = 1;
 425                 }
 426         }
 427         if (start == 0) {
 428                 trans = btrfs_join_transaction(root);
 429                 BUG_ON(IS_ERR(trans));
 430                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 431
 432                 /* lets try to make an inline extent */
 433                 if (ret || total_in < (actual_end - start)) {
 434                         /* we didn't compress the entire range, try
 435                          * to make an uncompressed inline extent.
 436                          */
 437                         ret = cow_file_range_inline(trans, root, inode,
 438                                                     start, end, 0, 0, NULL);
 439                 } else {
 440                         /* try making a compressed inline extent */
 441                         ret = cow_file_range_inline(trans, root, inode,
 442                                                     start, end,
 443                                                     total_compressed,
 444                                                     compress_type, pages);
 445                 }
 446                 if (ret == 0) {
 447                         /*
 448                          * inline extent creation worked, we don't need
 449                          * to create any more async work items.  Unlock
 450                          * and free up our temp pages.
 451                          */
 452                         extent_clear_unlock_delalloc(inode,
 453                              &BTRFS_I(inode)->io_tree,
 454                              start, end, NULL,
 455                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
 456                              EXTENT_CLEAR_DELALLOC |
 457                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
 458
 459                         btrfs_end_transaction(trans, root);
 460                         goto free_pages_out;
 461                 }
 462                 btrfs_end_transaction(trans, root);
 463         }
 464
 465         if (will_compress) {
 466                 /*
 467                  * we aren't doing an inline extent round the compressed size
 468                  * up to a block size boundary so the allocator does sane
 469                  * things
 470                  */
 471                 total_compressed = (total_compressed + blocksize - 1) &
 472                         ~(blocksize - 1);
 473
 474                 /*
 475                  * one last check to make sure the compression is really a
 476                  * win, compare the page count read with the blocks on disk
 477                  */
 478                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
 479                         ~(PAGE_CACHE_SIZE - 1);
 480                 if (total_compressed >= total_in) {
 481                         will_compress = 0;
 482                 } else {
 483                         num_bytes = total_in;
 484                 }
 485         }
 486         if (!will_compress && pages) {
 487                 /*
 488                  * the compression code ran but failed to make things smaller,
 489                  * free any pages it allocated and our page pointer array
 490                  */
 491                 for (i = 0; i < nr_pages_ret; i++) {
 492                         WARN_ON(pages[i]->mapping);
 493                         page_cache_release(pages[i]);
 494                 }
 495                 kfree(pages);
 496                 pages = NULL;
 497                 total_compressed = 0;
 498                 nr_pages_ret = 0;
 499
 500                 /* flag the file so we don't compress in the future */
 501                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 502                     !(BTRFS_I(inode)->force_compress)) {
 503                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 504                 }
 505         }
 506         if (will_compress) {
 507                 *num_added += 1;
 508
 509                 /* the async work queues will take care of doing actual
 510                  * allocation on disk for these compressed pages,
 511                  * and will submit them to the elevator.
 512                  */
 513                 add_async_extent(async_cow, start, num_bytes,
 514                                  total_compressed, pages, nr_pages_ret,
 515                                  compress_type);
 516
 517                 if (start + num_bytes < end) {
 518                         start += num_bytes;
 519                         pages = NULL;
 520                         cond_resched();
 521                         goto again;
 522                 }
 523         } else {
 524 cleanup_and_bail_uncompressed:
 525                 /*
 526                  * No compression, but we still need to write the pages in
 527                  * the file we've been given so far.  redirty the locked
 528                  * page if it corresponds to our extent and set things up
 529                  * for the async work queue to run cow_file_range to do
 530                  * the normal delalloc dance
 531                  */
 532                 if (page_offset(locked_page) >= start &&
 533                     page_offset(locked_page) <= end) {
 534                         __set_page_dirty_nobuffers(locked_page);
 535                         /* unlocked later on in the async handlers */
 536                 }
 537                 add_async_extent(async_cow, start, end - start + 1,
 538                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 539                 *num_added += 1;
 540         }
 541
 542 out:
 543         return 0;
 544
 545 free_pages_out:
 546         for (i = 0; i < nr_pages_ret; i++) {
 547                 WARN_ON(pages[i]->mapping);
 548                 page_cache_release(pages[i]);
 549         }
 550         kfree(pages);
 551
 552         goto out;
 553 }
 554
 555 /*
 556  * phase two of compressed writeback.  This is the ordered portion
 557  * of the code, which only gets called in the order the work was
 558  * queued.  We walk all the async extents created by compress_file_range
 559  * and send them down to the disk.
 560  */
 561 static noinline int submit_compressed_extents(struct inode *inode,
 562                                               struct async_cow *async_cow)
 563 {
 564         struct async_extent *async_extent;
 565         u64 alloc_hint = 0;
 566         struct btrfs_trans_handle *trans;
 567         struct btrfs_key ins;
 568         struct extent_map *em;
 569         struct btrfs_root *root = BTRFS_I(inode)->root;
 570         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 571         struct extent_io_tree *io_tree;
 572         int ret = 0;
 573
 574         if (list_empty(&async_cow->extents))
 575                 return 0;
 576
 577
 578         while (!list_empty(&async_cow->extents)) {
 579                 async_extent = list_entry(async_cow->extents.next,
 580                                           struct async_extent, list);
 581                 list_del(&async_extent->list);
 582
 583                 io_tree = &BTRFS_I(inode)->io_tree;
 584
 585 retry:
 586                 /* did the compression code fall back to uncompressed IO? */
 587                 if (!async_extent->pages) {
 588                         int page_started = 0;
 589                         unsigned long nr_written = 0;
 590
 591                         lock_extent(io_tree, async_extent->start,
 592                                          async_extent->start +
 593                                          async_extent->ram_size - 1, GFP_NOFS);
 594
 595                         /* allocate blocks */
 596                         ret = cow_file_range(inode, async_cow->locked_page,
 597                                              async_extent->start,
 598                                              async_extent->start +
 599                                              async_extent->ram_size - 1,
 600                                              &page_started, &nr_written, 0);
 601
 602                         /*
 603                          * if page_started, cow_file_range inserted an
 604                          * inline extent and took care of all the unlocking
 605                          * and IO for us.  Otherwise, we need to submit
 606                          * all those pages down to the drive.
 607                          */
 608                         if (!page_started && !ret)
 609                                 extent_write_locked_range(io_tree,
 610                                                   inode, async_extent->start,
 611                                                   async_extent->start +
 612                                                   async_extent->ram_size - 1,
 613                                                   btrfs_get_extent,
 614                                                   WB_SYNC_ALL);
 615                         kfree(async_extent);
 616                         cond_resched();
 617                         continue;
 618                 }
 619
 620                 lock_extent(io_tree, async_extent->start,
 621                             async_extent->start + async_extent->ram_size - 1,
 622                             GFP_NOFS);
 623
 624                 trans = btrfs_join_transaction(root);
 625                 BUG_ON(IS_ERR(trans));
 626                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 627                 ret = btrfs_reserve_extent(trans, root,
 628                                            async_extent->compressed_size,
 629                                            async_extent->compressed_size,
 630                                            0, alloc_hint,
 631                                            (u64)-1, &ins, 1);
 632                 btrfs_end_transaction(trans, root);
 633
 634                 if (ret) {
 635                         int i;
 636                         for (i = 0; i < async_extent->nr_pages; i++) {
 637                                 WARN_ON(async_extent->pages[i]->mapping);
 638                                 page_cache_release(async_extent->pages[i]);
 639                         }
 640                         kfree(async_extent->pages);
 641                         async_extent->nr_pages = 0;
 642                         async_extent->pages = NULL;
 643                         unlock_extent(io_tree, async_extent->start,
 644                                       async_extent->start +
 645                                       async_extent->ram_size - 1, GFP_NOFS);
 646                         goto retry;
 647                 }
 648
 649                 /*
 650                  * here we're doing allocation and writeback of the
 651                  * compressed pages
 652                  */
 653                 btrfs_drop_extent_cache(inode, async_extent->start,
 654                                         async_extent->start +
 655                                         async_extent->ram_size - 1, 0);
 656
 657                 em = alloc_extent_map();
 658                 BUG_ON(!em);
 659                 em->start = async_extent->start;
 660                 em->len = async_extent->ram_size;
 661                 em->orig_start = em->start;
 662
 663                 em->block_start = ins.objectid;
 664                 em->block_len = ins.offset;
 665                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 666                 em->compress_type = async_extent->compress_type;
 667                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 668                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 669
 670                 while (1) {
 671                         write_lock(&em_tree->lock);
 672                         ret = add_extent_mapping(em_tree, em);
 673                         write_unlock(&em_tree->lock);
 674                         if (ret != -EEXIST) {
 675                                 free_extent_map(em);
 676                                 break;
 677                         }
 678                         btrfs_drop_extent_cache(inode, async_extent->start,
 679                                                 async_extent->start +
 680                                                 async_extent->ram_size - 1, 0);
 681                 }
 682
 683                 ret = btrfs_add_ordered_extent_compress(inode,
 684                                                 async_extent->start,
 685                                                 ins.objectid,
 686                                                 async_extent->ram_size,
 687                                                 ins.offset,
 688                                                 BTRFS_ORDERED_COMPRESSED,
 689                                                 async_extent->compress_type);
 690                 BUG_ON(ret);
 691
 692                 /*
 693                  * clear dirty, set writeback and unlock the pages.
 694                  */
 695                 extent_clear_unlock_delalloc(inode,
 696                                 &BTRFS_I(inode)->io_tree,
 697                                 async_extent->start,
 698                                 async_extent->start +
 699                                 async_extent->ram_size - 1,
 700                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 701                                 EXTENT_CLEAR_UNLOCK |
 702                                 EXTENT_CLEAR_DELALLOC |
 703                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
 704
 705                 ret = btrfs_submit_compressed_write(inode,
 706                                     async_extent->start,
 707                                     async_extent->ram_size,
 708                                     ins.objectid,
 709                                     ins.offset, async_extent->pages,
 710                                     async_extent->nr_pages);
 711
 712                 BUG_ON(ret);
 713                 alloc_hint = ins.objectid + ins.offset;
 714                 kfree(async_extent);
 715                 cond_resched();
 716         }
 717
 718         return 0;
 719 }
 720
 721 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 722                                       u64 num_bytes)
 723 {
 724         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 725         struct extent_map *em;
 726         u64 alloc_hint = 0;
 727
 728         read_lock(&em_tree->lock);
 729         em = search_extent_mapping(em_tree, start, num_bytes);
 730         if (em) {
 731                 /*
 732                  * if block start isn't an actual block number then find the
 733                  * first block in this inode and use that as a hint.  If that
 734                  * block is also bogus then just don't worry about it.
 735                  */
 736                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 737                         free_extent_map(em);
 738                         em = search_extent_mapping(em_tree, 0, 0);
 739                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 740                                 alloc_hint = em->block_start;
 741                         if (em)
 742                                 free_extent_map(em);
 743                 } else {
 744                         alloc_hint = em->block_start;
 745                         free_extent_map(em);
 746                 }
 747         }
 748         read_unlock(&em_tree->lock);
 749
 750         return alloc_hint;
 751 }
 752
 753 static inline bool is_free_space_inode(struct btrfs_root *root,
 754                                        struct inode *inode)
 755 {
 756         if (root == root->fs_info->tree_root ||
 757             BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID)
 758                 return true;
 759         return false;
 760 }
 761
 762 /*
 763  * when extent_io.c finds a delayed allocation range in the file,
 764  * the call backs end up in this code.  The basic idea is to
 765  * allocate extents on disk for the range, and create ordered data structs
 766  * in ram to track those extents.
 767  *
 768  * locked_page is the page that writepage had locked already.  We use
 769  * it to make sure we don't do extra locks or unlocks.
 770  *
 771  * *page_started is set to one if we unlock locked_page and do everything
 772  * required to start IO on it.  It may be clean and already done with
 773  * IO when we return.
 774  */
 775 static noinline int cow_file_range(struct inode *inode,
 776                                    struct page *locked_page,
 777                                    u64 start, u64 end, int *page_started,
 778                                    unsigned long *nr_written,
 779                                    int unlock)
 780 {
 781         struct btrfs_root *root = BTRFS_I(inode)->root;
 782         struct btrfs_trans_handle *trans;
 783         u64 alloc_hint = 0;
 784         u64 num_bytes;
 785         unsigned long ram_size;
 786         u64 disk_num_bytes;
 787         u64 cur_alloc_size;
 788         u64 blocksize = root->sectorsize;
 789         struct btrfs_key ins;
 790         struct extent_map *em;
 791         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 792         int ret = 0;
 793
 794         BUG_ON(is_free_space_inode(root, inode));
 795         trans = btrfs_join_transaction(root);
 796         BUG_ON(IS_ERR(trans));
 797         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 798
 799         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 800         num_bytes = max(blocksize,  num_bytes);
 801         disk_num_bytes = num_bytes;
 802         ret = 0;
 803
 804         /* if this is a small write inside eof, kick off defrag */
 805         if (end <= BTRFS_I(inode)->disk_i_size && num_bytes < 64 * 1024)
 806                 btrfs_add_inode_defrag(trans, inode);
 807
 808         if (start == 0) {
 809                 /* lets try to make an inline extent */
 810                 ret = cow_file_range_inline(trans, root, inode,
 811                                             start, end, 0, 0, NULL);
 812                 if (ret == 0) {
 813                         extent_clear_unlock_delalloc(inode,
 814                                      &BTRFS_I(inode)->io_tree,
 815                                      start, end, NULL,
 816                                      EXTENT_CLEAR_UNLOCK_PAGE |
 817                                      EXTENT_CLEAR_UNLOCK |
 818                                      EXTENT_CLEAR_DELALLOC |
 819                                      EXTENT_CLEAR_DIRTY |
 820                                      EXTENT_SET_WRITEBACK |
 821                                      EXTENT_END_WRITEBACK);
 822
 823                         *nr_written = *nr_written +
 824                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 825                         *page_started = 1;
 826                         ret = 0;
 827                         goto out;
 828                 }
 829         }
 830
 831         BUG_ON(disk_num_bytes >
 832                btrfs_super_total_bytes(&root->fs_info->super_copy));
 833
 834         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 835         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 836
 837         while (disk_num_bytes > 0) {
 838                 unsigned long op;
 839
 840                 cur_alloc_size = disk_num_bytes;
 841                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 842                                            root->sectorsize, 0, alloc_hint,
 843                                            (u64)-1, &ins, 1);
 844                 BUG_ON(ret);
 845
 846                 em = alloc_extent_map();
 847                 BUG_ON(!em);
 848                 em->start = start;
 849                 em->orig_start = em->start;
 850                 ram_size = ins.offset;
 851                 em->len = ins.offset;
 852
 853                 em->block_start = ins.objectid;
 854                 em->block_len = ins.offset;
 855                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 856                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 857
 858                 while (1) {
 859                         write_lock(&em_tree->lock);
 860                         ret = add_extent_mapping(em_tree, em);
 861                         write_unlock(&em_tree->lock);
 862                         if (ret != -EEXIST) {
 863                                 free_extent_map(em);
 864                                 break;
 865                         }
 866                         btrfs_drop_extent_cache(inode, start,
 867                                                 start + ram_size - 1, 0);
 868                 }
 869
 870                 cur_alloc_size = ins.offset;
 871                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 872                                                ram_size, cur_alloc_size, 0);
 873                 BUG_ON(ret);
 874
 875                 if (root->root_key.objectid ==
 876                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 877                         ret = btrfs_reloc_clone_csums(inode, start,
 878                                                       cur_alloc_size);
 879                         BUG_ON(ret);
 880                 }
 881
 882                 if (disk_num_bytes < cur_alloc_size)
 883                         break;
 884
 885                 /* we're not doing compressed IO, don't unlock the first
 886                  * page (which the caller expects to stay locked), don't
 887                  * clear any dirty bits and don't set any writeback bits
 888                  *
 889                  * Do set the Private2 bit so we know this page was properly
 890                  * setup for writepage
 891                  */
 892                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
 893                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
 894                         EXTENT_SET_PRIVATE2;
 895
 896                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 897                                              start, start + ram_size - 1,
 898                                              locked_page, op);
 899                 disk_num_bytes -= cur_alloc_size;
 900                 num_bytes -= cur_alloc_size;
 901                 alloc_hint = ins.objectid + ins.offset;
 902                 start += cur_alloc_size;
 903         }
 904 out:
 905         ret = 0;
 906         btrfs_end_transaction(trans, root);
 907
 908         return ret;
 909 }
 910
 911 /*
 912  * work queue call back to started compression on a file and pages
 913  */
 914 static noinline void async_cow_start(struct btrfs_work *work)
 915 {
 916         struct async_cow *async_cow;
 917         int num_added = 0;
 918         async_cow = container_of(work, struct async_cow, work);
 919
 920         compress_file_range(async_cow->inode, async_cow->locked_page,
 921                             async_cow->start, async_cow->end, async_cow,
 922                             &num_added);
 923         if (num_added == 0)
 924                 async_cow->inode = NULL;
 925 }
 926
 927 /*
 928  * work queue call back to submit previously compressed pages
 929  */
 930 static noinline void async_cow_submit(struct btrfs_work *work)
 931 {
 932         struct async_cow *async_cow;
 933         struct btrfs_root *root;
 934         unsigned long nr_pages;
 935
 936         async_cow = container_of(work, struct async_cow, work);
 937
 938         root = async_cow->root;
 939         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
 940                 PAGE_CACHE_SHIFT;
 941
 942         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
 943
 944         if (atomic_read(&root->fs_info->async_delalloc_pages) <
 945             5 * 1042 * 1024 &&
 946             waitqueue_active(&root->fs_info->async_submit_wait))
 947                 wake_up(&root->fs_info->async_submit_wait);
 948
 949         if (async_cow->inode)
 950                 submit_compressed_extents(async_cow->inode, async_cow);
 951 }
 952
 953 static noinline void async_cow_free(struct btrfs_work *work)
 954 {
 955         struct async_cow *async_cow;
 956         async_cow = container_of(work, struct async_cow, work);
 957         kfree(async_cow);
 958 }
 959
 960 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
 961                                 u64 start, u64 end, int *page_started,
 962                                 unsigned long *nr_written)
 963 {
 964         struct async_cow *async_cow;
 965         struct btrfs_root *root = BTRFS_I(inode)->root;
 966         unsigned long nr_pages;
 967         u64 cur_end;
 968         int limit = 10 * 1024 * 1042;
 969
 970         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
 971                          1, 0, NULL, GFP_NOFS);
 972         while (start < end) {
 973                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
 974                 BUG_ON(!async_cow);
 975                 async_cow->inode = inode;
 976                 async_cow->root = root;
 977                 async_cow->locked_page = locked_page;
 978                 async_cow->start = start;
 979
 980                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
 981                         cur_end = end;
 982                 else
 983                         cur_end = min(end, start + 512 * 1024 - 1);
 984
 985                 async_cow->end = cur_end;
 986                 INIT_LIST_HEAD(&async_cow->extents);
 987
 988                 async_cow->work.func = async_cow_start;
 989                 async_cow->work.ordered_func = async_cow_submit;
 990                 async_cow->work.ordered_free = async_cow_free;
 991                 async_cow->work.flags = 0;
 992
 993                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
 994                         PAGE_CACHE_SHIFT;
 995                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
 996
 997                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
 998                                    &async_cow->work);
 999
1000                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1001                         wait_event(root->fs_info->async_submit_wait,
1002                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1003                             limit));
1004                 }
1005
1006                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1007                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1008                         wait_event(root->fs_info->async_submit_wait,
1009                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1010                            0));
1011                 }
1012
1013                 *nr_written += nr_pages;
1014                 start = cur_end + 1;
1015         }
1016         *page_started = 1;
1017         return 0;
1018 }
1019
1020 static noinline int csum_exist_in_range(struct btrfs_root *root,
1021                                         u64 bytenr, u64 num_bytes)
1022 {
1023         int ret;
1024         struct btrfs_ordered_sum *sums;
1025         LIST_HEAD(list);
1026
1027         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1028                                        bytenr + num_bytes - 1, &list, 0);
1029         if (ret == 0 && list_empty(&list))
1030                 return 0;
1031
1032         while (!list_empty(&list)) {
1033                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1034                 list_del(&sums->list);
1035                 kfree(sums);
1036         }
1037         return 1;
1038 }
1039
1040 /*
1041  * when nowcow writeback call back.  This checks for snapshots or COW copies
1042  * of the extents that exist in the file, and COWs the file as required.
1043  *
1044  * If no cow copies or snapshots exist, we write directly to the existing
1045  * blocks on disk
1046  */
1047 static noinline int run_delalloc_nocow(struct inode *inode,
1048                                        struct page *locked_page,
1049                               u64 start, u64 end, int *page_started, int force,
1050                               unsigned long *nr_written)
1051 {
1052         struct btrfs_root *root = BTRFS_I(inode)->root;
1053         struct btrfs_trans_handle *trans;
1054         struct extent_buffer *leaf;
1055         struct btrfs_path *path;
1056         struct btrfs_file_extent_item *fi;
1057         struct btrfs_key found_key;
1058         u64 cow_start;
1059         u64 cur_offset;
1060         u64 extent_end;
1061         u64 extent_offset;
1062         u64 disk_bytenr;
1063         u64 num_bytes;
1064         int extent_type;
1065         int ret;
1066         int type;
1067         int nocow;
1068         int check_prev = 1;
1069         bool nolock;
1070         u64 ino = btrfs_ino(inode);
1071
1072         path = btrfs_alloc_path();
1073         BUG_ON(!path);
1074
1075         nolock = is_free_space_inode(root, inode);
1076
1077         if (nolock)
1078                 trans = btrfs_join_transaction_nolock(root);
1079         else
1080                 trans = btrfs_join_transaction(root);
1081
1082         BUG_ON(IS_ERR(trans));
1083         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1084
1085         cow_start = (u64)-1;
1086         cur_offset = start;
1087         while (1) {
1088                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1089                                                cur_offset, 0);
1090                 BUG_ON(ret < 0);
1091                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1092                         leaf = path->nodes[0];
1093                         btrfs_item_key_to_cpu(leaf, &found_key,
1094                                               path->slots[0] - 1);
1095                         if (found_key.objectid == ino &&
1096                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1097                                 path->slots[0]--;
1098                 }
1099                 check_prev = 0;
1100 next_slot:
1101                 leaf = path->nodes[0];
1102                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1103                         ret = btrfs_next_leaf(root, path);
1104                         if (ret < 0)
1105                                 BUG_ON(1);
1106                         if (ret > 0)
1107                                 break;
1108                         leaf = path->nodes[0];
1109                 }
1110
1111                 nocow = 0;
1112                 disk_bytenr = 0;
1113                 num_bytes = 0;
1114                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1115
1116                 if (found_key.objectid > ino ||
1117                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1118                     found_key.offset > end)
1119                         break;
1120
1121                 if (found_key.offset > cur_offset) {
1122                         extent_end = found_key.offset;
1123                         extent_type = 0;
1124                         goto out_check;
1125                 }
1126
1127                 fi = btrfs_item_ptr(leaf, path->slots[0],
1128                                     struct btrfs_file_extent_item);
1129                 extent_type = btrfs_file_extent_type(leaf, fi);
1130
1131                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1132                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1133                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1134                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1135                         extent_end = found_key.offset +
1136                                 btrfs_file_extent_num_bytes(leaf, fi);
1137                         if (extent_end <= start) {
1138                                 path->slots[0]++;
1139                                 goto next_slot;
1140                         }
1141                         if (disk_bytenr == 0)
1142                                 goto out_check;
1143                         if (btrfs_file_extent_compression(leaf, fi) ||
1144                             btrfs_file_extent_encryption(leaf, fi) ||
1145                             btrfs_file_extent_other_encoding(leaf, fi))
1146                                 goto out_check;
1147                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1148                                 goto out_check;
1149                         if (btrfs_extent_readonly(root, disk_bytenr))
1150                                 goto out_check;
1151                         if (btrfs_cross_ref_exist(trans, root, ino,
1152                                                   found_key.offset -
1153                                                   extent_offset, disk_bytenr))
1154                                 goto out_check;
1155                         disk_bytenr += extent_offset;
1156                         disk_bytenr += cur_offset - found_key.offset;
1157                         num_bytes = min(end + 1, extent_end) - cur_offset;
1158                         /*
1159                          * force cow if csum exists in the range.
1160                          * this ensure that csum for a given extent are
1161                          * either valid or do not exist.
1162                          */
1163                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1164                                 goto out_check;
1165                         nocow = 1;
1166                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1167                         extent_end = found_key.offset +
1168                                 btrfs_file_extent_inline_len(leaf, fi);
1169                         extent_end = ALIGN(extent_end, root->sectorsize);
1170                 } else {
1171                         BUG_ON(1);
1172                 }
1173 out_check:
1174                 if (extent_end <= start) {
1175                         path->slots[0]++;
1176                         goto next_slot;
1177                 }
1178                 if (!nocow) {
1179                         if (cow_start == (u64)-1)
1180                                 cow_start = cur_offset;
1181                         cur_offset = extent_end;
1182                         if (cur_offset > end)
1183                                 break;
1184                         path->slots[0]++;
1185                         goto next_slot;
1186                 }
1187
1188                 btrfs_release_path(path);
1189                 if (cow_start != (u64)-1) {
1190                         ret = cow_file_range(inode, locked_page, cow_start,
1191                                         found_key.offset - 1, page_started,
1192                                         nr_written, 1);
1193                         BUG_ON(ret);
1194                         cow_start = (u64)-1;
1195                 }
1196
1197                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1198                         struct extent_map *em;
1199                         struct extent_map_tree *em_tree;
1200                         em_tree = &BTRFS_I(inode)->extent_tree;
1201                         em = alloc_extent_map();
1202                         BUG_ON(!em);
1203                         em->start = cur_offset;
1204                         em->orig_start = em->start;
1205                         em->len = num_bytes;
1206                         em->block_len = num_bytes;
1207                         em->block_start = disk_bytenr;
1208                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1209                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1210                         while (1) {
1211                                 write_lock(&em_tree->lock);
1212                                 ret = add_extent_mapping(em_tree, em);
1213                                 write_unlock(&em_tree->lock);
1214                                 if (ret != -EEXIST) {
1215                                         free_extent_map(em);
1216                                         break;
1217                                 }
1218                                 btrfs_drop_extent_cache(inode, em->start,
1219                                                 em->start + em->len - 1, 0);
1220                         }
1221                         type = BTRFS_ORDERED_PREALLOC;
1222                 } else {
1223                         type = BTRFS_ORDERED_NOCOW;
1224                 }
1225
1226                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1227                                                num_bytes, num_bytes, type);
1228                 BUG_ON(ret);
1229
1230                 if (root->root_key.objectid ==
1231                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1232                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1233                                                       num_bytes);
1234                         BUG_ON(ret);
1235                 }
1236
1237                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1238                                 cur_offset, cur_offset + num_bytes - 1,
1239                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1240                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1241                                 EXTENT_SET_PRIVATE2);
1242                 cur_offset = extent_end;
1243                 if (cur_offset > end)
1244                         break;
1245         }
1246         btrfs_release_path(path);
1247
1248         if (cur_offset <= end && cow_start == (u64)-1)
1249                 cow_start = cur_offset;
1250         if (cow_start != (u64)-1) {
1251                 ret = cow_file_range(inode, locked_page, cow_start, end,
1252                                      page_started, nr_written, 1);
1253                 BUG_ON(ret);
1254         }
1255
1256         if (nolock) {
1257                 ret = btrfs_end_transaction_nolock(trans, root);
1258                 BUG_ON(ret);
1259         } else {
1260                 ret = btrfs_end_transaction(trans, root);
1261                 BUG_ON(ret);
1262         }
1263         btrfs_free_path(path);
1264         return 0;
1265 }
1266
1267 /*
1268  * extent_io.c call back to do delayed allocation processing
1269  */
1270 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1271                               u64 start, u64 end, int *page_started,
1272                               unsigned long *nr_written)
1273 {
1274         int ret;
1275         struct btrfs_root *root = BTRFS_I(inode)->root;
1276
1277         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1278                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1279                                          page_started, 1, nr_written);
1280         else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1281                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1282                                          page_started, 0, nr_written);
1283         else if (!btrfs_test_opt(root, COMPRESS) &&
1284                  !(BTRFS_I(inode)->force_compress) &&
1285                  !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1286                 ret = cow_file_range(inode, locked_page, start, end,
1287                                       page_started, nr_written, 1);
1288         else
1289                 ret = cow_file_range_async(inode, locked_page, start, end,
1290                                            page_started, nr_written);
1291         return ret;
1292 }
1293
1294 static int btrfs_split_extent_hook(struct inode *inode,
1295                                    struct extent_state *orig, u64 split)
1296 {
1297         /* not delalloc, ignore it */
1298         if (!(orig->state & EXTENT_DELALLOC))
1299                 return 0;
1300
1301         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1302         return 0;
1303 }
1304
1305 /*
1306  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1307  * extents so we can keep track of new extents that are just merged onto old
1308  * extents, such as when we are doing sequential writes, so we can properly
1309  * account for the metadata space we'll need.
1310  */
1311 static int btrfs_merge_extent_hook(struct inode *inode,
1312                                    struct extent_state *new,
1313                                    struct extent_state *other)
1314 {
1315         /* not delalloc, ignore it */
1316         if (!(other->state & EXTENT_DELALLOC))
1317                 return 0;
1318
1319         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1320         return 0;
1321 }
1322
1323 /*
1324  * extent_io.c set_bit_hook, used to track delayed allocation
1325  * bytes in this file, and to maintain the list of inodes that
1326  * have pending delalloc work to be done.
1327  */
1328 static int btrfs_set_bit_hook(struct inode *inode,
1329                               struct extent_state *state, int *bits)
1330 {
1331
1332         /*
1333          * set_bit and clear bit hooks normally require _irqsave/restore
1334          * but in this case, we are only testing for the DELALLOC
1335          * bit, which is only set or cleared with irqs on
1336          */
1337         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1338                 struct btrfs_root *root = BTRFS_I(inode)->root;
1339                 u64 len = state->end + 1 - state->start;
1340                 bool do_list = !is_free_space_inode(root, inode);
1341
1342                 if (*bits & EXTENT_FIRST_DELALLOC)
1343                         *bits &= ~EXTENT_FIRST_DELALLOC;
1344                 else
1345                         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1346
1347                 spin_lock(&root->fs_info->delalloc_lock);
1348                 BTRFS_I(inode)->delalloc_bytes += len;
1349                 root->fs_info->delalloc_bytes += len;
1350                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1351                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1352                                       &root->fs_info->delalloc_inodes);
1353                 }
1354                 spin_unlock(&root->fs_info->delalloc_lock);
1355         }
1356         return 0;
1357 }
1358
1359 /*
1360  * extent_io.c clear_bit_hook, see set_bit_hook for why
1361  */
1362 static int btrfs_clear_bit_hook(struct inode *inode,
1363                                 struct extent_state *state, int *bits)
1364 {
1365         /*
1366          * set_bit and clear bit hooks normally require _irqsave/restore
1367          * but in this case, we are only testing for the DELALLOC
1368          * bit, which is only set or cleared with irqs on
1369          */
1370         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1371                 struct btrfs_root *root = BTRFS_I(inode)->root;
1372                 u64 len = state->end + 1 - state->start;
1373                 bool do_list = !is_free_space_inode(root, inode);
1374
1375                 if (*bits & EXTENT_FIRST_DELALLOC)
1376                         *bits &= ~EXTENT_FIRST_DELALLOC;
1377                 else if (!(*bits & EXTENT_DO_ACCOUNTING))
1378                         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1379
1380                 if (*bits & EXTENT_DO_ACCOUNTING)
1381                         btrfs_delalloc_release_metadata(inode, len);
1382
1383                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1384                     && do_list)
1385                         btrfs_free_reserved_data_space(inode, len);
1386
1387                 spin_lock(&root->fs_info->delalloc_lock);
1388                 root->fs_info->delalloc_bytes -= len;
1389                 BTRFS_I(inode)->delalloc_bytes -= len;
1390
1391                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1392                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1393                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1394                 }
1395                 spin_unlock(&root->fs_info->delalloc_lock);
1396         }
1397         return 0;
1398 }
1399
1400 /*
1401  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1402  * we don't create bios that span stripes or chunks
1403  */
1404 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1405                          size_t size, struct bio *bio,
1406                          unsigned long bio_flags)
1407 {
1408         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1409         struct btrfs_mapping_tree *map_tree;
1410         u64 logical = (u64)bio->bi_sector << 9;
1411         u64 length = 0;
1412         u64 map_length;
1413         int ret;
1414
1415         if (bio_flags & EXTENT_BIO_COMPRESSED)
1416                 return 0;
1417
1418         length = bio->bi_size;
1419         map_tree = &root->fs_info->mapping_tree;
1420         map_length = length;
1421         ret = btrfs_map_block(map_tree, READ, logical,
1422                               &map_length, NULL, 0);
1423
1424         if (map_length < length + size)
1425                 return 1;
1426         return ret;
1427 }
1428
1429 /*
1430  * in order to insert checksums into the metadata in large chunks,
1431  * we wait until bio submission time.   All the pages in the bio are
1432  * checksummed and sums are attached onto the ordered extent record.
1433  *
1434  * At IO completion time the cums attached on the ordered extent record
1435  * are inserted into the btree
1436  */
1437 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1438                                     struct bio *bio, int mirror_num,
1439                                     unsigned long bio_flags,
1440                                     u64 bio_offset)
1441 {
1442         struct btrfs_root *root = BTRFS_I(inode)->root;
1443         int ret = 0;
1444
1445         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1446         BUG_ON(ret);
1447         return 0;
1448 }
1449
1450 /*
1451  * in order to insert checksums into the metadata in large chunks,
1452  * we wait until bio submission time.   All the pages in the bio are
1453  * checksummed and sums are attached onto the ordered extent record.
1454  *
1455  * At IO completion time the cums attached on the ordered extent record
1456  * are inserted into the btree
1457  */
1458 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1459                           int mirror_num, unsigned long bio_flags,
1460                           u64 bio_offset)
1461 {
1462         struct btrfs_root *root = BTRFS_I(inode)->root;
1463         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1464 }
1465
1466 /*
1467  * extent_io.c submission hook. This does the right thing for csum calculation
1468  * on write, or reading the csums from the tree before a read
1469  */
1470 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1471                           int mirror_num, unsigned long bio_flags,
1472                           u64 bio_offset)
1473 {
1474         struct btrfs_root *root = BTRFS_I(inode)->root;
1475         int ret = 0;
1476         int skip_sum;
1477
1478         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1479
1480         if (is_free_space_inode(root, inode))
1481                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1482         else
1483                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1484         BUG_ON(ret);
1485
1486         if (!(rw & REQ_WRITE)) {
1487                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1488                         return btrfs_submit_compressed_read(inode, bio,
1489                                                     mirror_num, bio_flags);
1490                 } else if (!skip_sum) {
1491                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1492                         if (ret)
1493                                 return ret;
1494                 }
1495                 goto mapit;
1496         } else if (!skip_sum) {
1497                 /* csum items have already been cloned */
1498                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1499                         goto mapit;
1500                 /* we're doing a write, do the async checksumming */
1501                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1502                                    inode, rw, bio, mirror_num,
1503                                    bio_flags, bio_offset,
1504                                    __btrfs_submit_bio_start,
1505                                    __btrfs_submit_bio_done);
1506         }
1507
1508 mapit:
1509         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1510 }
1511
1512 /*
1513  * given a list of ordered sums record them in the inode.  This happens
1514  * at IO completion time based on sums calculated at bio submission time.
1515  */
1516 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1517                              struct inode *inode, u64 file_offset,
1518                              struct list_head *list)
1519 {
1520         struct btrfs_ordered_sum *sum;
1521
1522         list_for_each_entry(sum, list, list) {
1523                 btrfs_csum_file_blocks(trans,
1524                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1525         }
1526         return 0;
1527 }
1528
1529 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1530                               struct extent_state **cached_state)
1531 {
1532         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1533                 WARN_ON(1);
1534         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1535                                    cached_state, GFP_NOFS);
1536 }
1537
1538 /* see btrfs_writepage_start_hook for details on why this is required */
1539 struct btrfs_writepage_fixup {
1540         struct page *page;
1541         struct btrfs_work work;
1542 };
1543
1544 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1545 {
1546         struct btrfs_writepage_fixup *fixup;
1547         struct btrfs_ordered_extent *ordered;
1548         struct extent_state *cached_state = NULL;
1549         struct page *page;
1550         struct inode *inode;
1551         u64 page_start;
1552         u64 page_end;
1553
1554         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1555         page = fixup->page;
1556 again:
1557         lock_page(page);
1558         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1559                 ClearPageChecked(page);
1560                 goto out_page;
1561         }
1562
1563         inode = page->mapping->host;
1564         page_start = page_offset(page);
1565         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1566
1567         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1568                          &cached_state, GFP_NOFS);
1569
1570         /* already ordered? We're done */
1571         if (PagePrivate2(page))
1572                 goto out;
1573
1574         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1575         if (ordered) {
1576                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1577                                      page_end, &cached_state, GFP_NOFS);
1578                 unlock_page(page);
1579                 btrfs_start_ordered_extent(inode, ordered, 1);
1580                 goto again;
1581         }
1582
1583         BUG();
1584         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1585         ClearPageChecked(page);
1586 out:
1587         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1588                              &cached_state, GFP_NOFS);
1589 out_page:
1590         unlock_page(page);
1591         page_cache_release(page);
1592         kfree(fixup);
1593 }
1594
1595 /*
1596  * There are a few paths in the higher layers of the kernel that directly
1597  * set the page dirty bit without asking the filesystem if it is a
1598  * good idea.  This causes problems because we want to make sure COW
1599  * properly happens and the data=ordered rules are followed.
1600  *
1601  * In our case any range that doesn't have the ORDERED bit set
1602  * hasn't been properly setup for IO.  We kick off an async process
1603  * to fix it up.  The async helper will wait for ordered extents, set
1604  * the delalloc bit and make it safe to write the page.
1605  */
1606 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1607 {
1608         struct inode *inode = page->mapping->host;
1609         struct btrfs_writepage_fixup *fixup;
1610         struct btrfs_root *root = BTRFS_I(inode)->root;
1611
1612         /* this page is properly in the ordered list */
1613         if (TestClearPagePrivate2(page))
1614                 return 0;
1615
1616         if (PageChecked(page))
1617                 return -EAGAIN;
1618
1619         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1620         if (!fixup)
1621                 return -EAGAIN;
1622
1623         SetPageChecked(page);
1624         page_cache_get(page);
1625         fixup->work.func = btrfs_writepage_fixup_worker;
1626         fixup->page = page;
1627         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1628         return -EAGAIN;
1629 }
1630
1631 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1632                                        struct inode *inode, u64 file_pos,
1633                                        u64 disk_bytenr, u64 disk_num_bytes,
1634                                        u64 num_bytes, u64 ram_bytes,
1635                                        u8 compression, u8 encryption,
1636                                        u16 other_encoding, int extent_type)
1637 {
1638         struct btrfs_root *root = BTRFS_I(inode)->root;
1639         struct btrfs_file_extent_item *fi;
1640         struct btrfs_path *path;
1641         struct extent_buffer *leaf;
1642         struct btrfs_key ins;
1643         u64 hint;
1644         int ret;
1645
1646         path = btrfs_alloc_path();
1647         BUG_ON(!path);
1648
1649         path->leave_spinning = 1;
1650
1651         /*
1652          * we may be replacing one extent in the tree with another.
1653          * The new extent is pinned in the extent map, and we don't want
1654          * to drop it from the cache until it is completely in the btree.
1655          *
1656          * So, tell btrfs_drop_extents to leave this extent in the cache.
1657          * the caller is expected to unpin it and allow it to be merged
1658          * with the others.
1659          */
1660         ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1661                                  &hint, 0);
1662         BUG_ON(ret);
1663
1664         ins.objectid = btrfs_ino(inode);
1665         ins.offset = file_pos;
1666         ins.type = BTRFS_EXTENT_DATA_KEY;
1667         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1668         BUG_ON(ret);
1669         leaf = path->nodes[0];
1670         fi = btrfs_item_ptr(leaf, path->slots[0],
1671                             struct btrfs_file_extent_item);
1672         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1673         btrfs_set_file_extent_type(leaf, fi, extent_type);
1674         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1675         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1676         btrfs_set_file_extent_offset(leaf, fi, 0);
1677         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1678         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1679         btrfs_set_file_extent_compression(leaf, fi, compression);
1680         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1681         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1682
1683         btrfs_unlock_up_safe(path, 1);
1684         btrfs_set_lock_blocking(leaf);
1685
1686         btrfs_mark_buffer_dirty(leaf);
1687
1688         inode_add_bytes(inode, num_bytes);
1689
1690         ins.objectid = disk_bytenr;
1691         ins.offset = disk_num_bytes;
1692         ins.type = BTRFS_EXTENT_ITEM_KEY;
1693         ret = btrfs_alloc_reserved_file_extent(trans, root,
1694                                         root->root_key.objectid,
1695                                         btrfs_ino(inode), file_pos, &ins);
1696         BUG_ON(ret);
1697         btrfs_free_path(path);
1698
1699         return 0;
1700 }
1701
1702 /*
1703  * helper function for btrfs_finish_ordered_io, this
1704  * just reads in some of the csum leaves to prime them into ram
1705  * before we start the transaction.  It limits the amount of btree
1706  * reads required while inside the transaction.
1707  */
1708 /* as ordered data IO finishes, this gets called so we can finish
1709  * an ordered extent if the range of bytes in the file it covers are
1710  * fully written.
1711  */
1712 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1713 {
1714         struct btrfs_root *root = BTRFS_I(inode)->root;
1715         struct btrfs_trans_handle *trans = NULL;
1716         struct btrfs_ordered_extent *ordered_extent = NULL;
1717         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1718         struct extent_state *cached_state = NULL;
1719         int compress_type = 0;
1720         int ret;
1721         bool nolock;
1722
1723         ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1724                                              end - start + 1);
1725         if (!ret)
1726                 return 0;
1727         BUG_ON(!ordered_extent);
1728
1729         nolock = is_free_space_inode(root, inode);
1730
1731         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1732                 BUG_ON(!list_empty(&ordered_extent->list));
1733                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1734                 if (!ret) {
1735                         if (nolock)
1736                                 trans = btrfs_join_transaction_nolock(root);
1737                         else
1738                                 trans = btrfs_join_transaction(root);
1739                         BUG_ON(IS_ERR(trans));
1740                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1741                         ret = btrfs_update_inode(trans, root, inode);
1742                         BUG_ON(ret);
1743                 }
1744                 goto out;
1745         }
1746
1747         lock_extent_bits(io_tree, ordered_extent->file_offset,
1748                          ordered_extent->file_offset + ordered_extent->len - 1,
1749                          0, &cached_state, GFP_NOFS);
1750
1751         if (nolock)
1752                 trans = btrfs_join_transaction_nolock(root);
1753         else
1754                 trans = btrfs_join_transaction(root);
1755         BUG_ON(IS_ERR(trans));
1756         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1757
1758         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1759                 compress_type = ordered_extent->compress_type;
1760         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1761                 BUG_ON(compress_type);
1762                 ret = btrfs_mark_extent_written(trans, inode,
1763                                                 ordered_extent->file_offset,
1764                                                 ordered_extent->file_offset +
1765                                                 ordered_extent->len);
1766                 BUG_ON(ret);
1767         } else {
1768                 BUG_ON(root == root->fs_info->tree_root);
1769                 ret = insert_reserved_file_extent(trans, inode,
1770                                                 ordered_extent->file_offset,
1771                                                 ordered_extent->start,
1772                                                 ordered_extent->disk_len,
1773                                                 ordered_extent->len,
1774                                                 ordered_extent->len,
1775                                                 compress_type, 0, 0,
1776                                                 BTRFS_FILE_EXTENT_REG);
1777                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1778                                    ordered_extent->file_offset,
1779                                    ordered_extent->len);
1780                 BUG_ON(ret);
1781         }
1782         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1783                              ordered_extent->file_offset +
1784                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1785
1786         add_pending_csums(trans, inode, ordered_extent->file_offset,
1787                           &ordered_extent->list);
1788
1789         ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1790         if (!ret) {
1791                 ret = btrfs_update_inode(trans, root, inode);
1792                 BUG_ON(ret);
1793         }
1794         ret = 0;
1795 out:
1796         if (nolock) {
1797                 if (trans)
1798                         btrfs_end_transaction_nolock(trans, root);
1799         } else {
1800                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1801                 if (trans)
1802                         btrfs_end_transaction(trans, root);
1803         }
1804
1805         /* once for us */
1806         btrfs_put_ordered_extent(ordered_extent);
1807         /* once for the tree */
1808         btrfs_put_ordered_extent(ordered_extent);
1809
1810         return 0;
1811 }
1812
1813 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1814                                 struct extent_state *state, int uptodate)
1815 {
1816         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1817
1818         ClearPagePrivate2(page);
1819         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1820 }
1821
1822 /*
1823  * When IO fails, either with EIO or csum verification fails, we
1824  * try other mirrors that might have a good copy of the data.  This
1825  * io_failure_record is used to record state as we go through all the
1826  * mirrors.  If another mirror has good data, the page is set up to date
1827  * and things continue.  If a good mirror can't be found, the original
1828  * bio end_io callback is called to indicate things have failed.
1829  */
1830 struct io_failure_record {
1831         struct page *page;
1832         u64 start;
1833         u64 len;
1834         u64 logical;
1835         unsigned long bio_flags;
1836         int last_mirror;
1837 };
1838
1839 static int btrfs_io_failed_hook(struct bio *failed_bio,
1840                          struct page *page, u64 start, u64 end,
1841                          struct extent_state *state)
1842 {
1843         struct io_failure_record *failrec = NULL;
1844         u64 private;
1845         struct extent_map *em;
1846         struct inode *inode = page->mapping->host;
1847         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1848         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1849         struct bio *bio;
1850         int num_copies;
1851         int ret;
1852         int rw;
1853         u64 logical;
1854
1855         ret = get_state_private(failure_tree, start, &private);
1856         if (ret) {
1857                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1858                 if (!failrec)
1859                         return -ENOMEM;
1860                 failrec->start = start;
1861                 failrec->len = end - start + 1;
1862                 failrec->last_mirror = 0;
1863                 failrec->bio_flags = 0;
1864
1865                 read_lock(&em_tree->lock);
1866                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1867                 if (em->start > start || em->start + em->len < start) {
1868                         free_extent_map(em);
1869                         em = NULL;
1870                 }
1871                 read_unlock(&em_tree->lock);
1872
1873                 if (IS_ERR_OR_NULL(em)) {
1874                         kfree(failrec);
1875                         return -EIO;
1876                 }
1877                 logical = start - em->start;
1878                 logical = em->block_start + logical;
1879                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1880                         logical = em->block_start;
1881                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1882                         extent_set_compress_type(&failrec->bio_flags,
1883                                                  em->compress_type);
1884                 }
1885                 failrec->logical = logical;
1886                 free_extent_map(em);
1887                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1888                                 EXTENT_DIRTY, GFP_NOFS);
1889                 set_state_private(failure_tree, start,
1890                                  (u64)(unsigned long)failrec);
1891         } else {
1892                 failrec = (struct io_failure_record *)(unsigned long)private;
1893         }
1894         num_copies = btrfs_num_copies(
1895                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1896                               failrec->logical, failrec->len);
1897         failrec->last_mirror++;
1898         if (!state) {
1899                 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1900                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1901                                                     failrec->start,
1902                                                     EXTENT_LOCKED);
1903                 if (state && state->start != failrec->start)
1904                         state = NULL;
1905                 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1906         }
1907         if (!state || failrec->last_mirror > num_copies) {
1908                 set_state_private(failure_tree, failrec->start, 0);
1909                 clear_extent_bits(failure_tree, failrec->start,
1910                                   failrec->start + failrec->len - 1,
1911                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1912                 kfree(failrec);
1913                 return -EIO;
1914         }
1915         bio = bio_alloc(GFP_NOFS, 1);
1916         bio->bi_private = state;
1917         bio->bi_end_io = failed_bio->bi_end_io;
1918         bio->bi_sector = failrec->logical >> 9;
1919         bio->bi_bdev = failed_bio->bi_bdev;
1920         bio->bi_size = 0;
1921
1922         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1923         if (failed_bio->bi_rw & REQ_WRITE)
1924                 rw = WRITE;
1925         else
1926                 rw = READ;
1927
1928         ret = BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1929                                                       failrec->last_mirror,
1930                                                       failrec->bio_flags, 0);
1931         return ret;
1932 }
1933
1934 /*
1935  * each time an IO finishes, we do a fast check in the IO failure tree
1936  * to see if we need to process or clean up an io_failure_record
1937  */
1938 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1939 {
1940         u64 private;
1941         u64 private_failure;
1942         struct io_failure_record *failure;
1943         int ret;
1944
1945         private = 0;
1946         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1947                              (u64)-1, 1, EXTENT_DIRTY, 0)) {
1948                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1949                                         start, &private_failure);
1950                 if (ret == 0) {
1951                         failure = (struct io_failure_record *)(unsigned long)
1952                                    private_failure;
1953                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1954                                           failure->start, 0);
1955                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1956                                           failure->start,
1957                                           failure->start + failure->len - 1,
1958                                           EXTENT_DIRTY | EXTENT_LOCKED,
1959                                           GFP_NOFS);
1960                         kfree(failure);
1961                 }
1962         }
1963         return 0;
1964 }
1965
1966 /*
1967  * when reads are done, we need to check csums to verify the data is correct
1968  * if there's a match, we allow the bio to finish.  If not, we go through
1969  * the io_failure_record routines to find good copies
1970  */
1971 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1972                                struct extent_state *state)
1973 {
1974         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1975         struct inode *inode = page->mapping->host;
1976         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1977         char *kaddr;
1978         u64 private = ~(u32)0;
1979         int ret;
1980         struct btrfs_root *root = BTRFS_I(inode)->root;
1981         u32 csum = ~(u32)0;
1982
1983         if (PageChecked(page)) {
1984                 ClearPageChecked(page);
1985                 goto good;
1986         }
1987
1988         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1989                 goto good;
1990
1991         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1992             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1993                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1994                                   GFP_NOFS);
1995                 return 0;
1996         }
1997
1998         if (state && state->start == start) {
1999                 private = state->private;
2000                 ret = 0;
2001         } else {
2002                 ret = get_state_private(io_tree, start, &private);
2003         }
2004         kaddr = kmap_atomic(page, KM_USER0);
2005         if (ret)
2006                 goto zeroit;
2007
2008         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
2009         btrfs_csum_final(csum, (char *)&csum);
2010         if (csum != private)
2011                 goto zeroit;
2012
2013         kunmap_atomic(kaddr, KM_USER0);
2014 good:
2015         /* if the io failure tree for this inode is non-empty,
2016          * check to see if we've recovered from a failed IO
2017          */
2018         btrfs_clean_io_failures(inode, start);
2019         return 0;
2020
2021 zeroit:
2022         printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
2023                        "private %llu\n",
2024                        (unsigned long long)btrfs_ino(page->mapping->host),
2025                        (unsigned long long)start, csum,
2026                        (unsigned long long)private);
2027         memset(kaddr + offset, 1, end - start + 1);
2028         flush_dcache_page(page);
2029         kunmap_atomic(kaddr, KM_USER0);
2030         if (private == 0)
2031                 return 0;
2032         return -EIO;
2033 }
2034
2035 struct delayed_iput {
2036         struct list_head list;
2037         struct inode *inode;
2038 };
2039
2040 void btrfs_add_delayed_iput(struct inode *inode)
2041 {
2042         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2043         struct delayed_iput *delayed;
2044
2045         if (atomic_add_unless(&inode->i_count, -1, 1))
2046                 return;
2047
2048         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2049         delayed->inode = inode;
2050
2051         spin_lock(&fs_info->delayed_iput_lock);
2052         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2053         spin_unlock(&fs_info->delayed_iput_lock);
2054 }
2055
2056 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2057 {
2058         LIST_HEAD(list);
2059         struct btrfs_fs_info *fs_info = root->fs_info;
2060         struct delayed_iput *delayed;
2061         int empty;
2062
2063         spin_lock(&fs_info->delayed_iput_lock);
2064         empty = list_empty(&fs_info->delayed_iputs);
2065         spin_unlock(&fs_info->delayed_iput_lock);
2066         if (empty)
2067                 return;
2068
2069         down_read(&root->fs_info->cleanup_work_sem);
2070         spin_lock(&fs_info->delayed_iput_lock);
2071         list_splice_init(&fs_info->delayed_iputs, &list);
2072         spin_unlock(&fs_info->delayed_iput_lock);
2073
2074         while (!list_empty(&list)) {
2075                 delayed = list_entry(list.next, struct delayed_iput, list);
2076                 list_del(&delayed->list);
2077                 iput(delayed->inode);
2078                 kfree(delayed);
2079         }
2080         up_read(&root->fs_info->cleanup_work_sem);
2081 }
2082
2083 /*
2084  * calculate extra metadata reservation when snapshotting a subvolume
2085  * contains orphan files.
2086  */
2087 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans,
2088                                 struct btrfs_pending_snapshot *pending,
2089                                 u64 *bytes_to_reserve)
2090 {
2091         struct btrfs_root *root;
2092         struct btrfs_block_rsv *block_rsv;
2093         u64 num_bytes;
2094         int index;
2095
2096         root = pending->root;
2097         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2098                 return;
2099
2100         block_rsv = root->orphan_block_rsv;
2101
2102         /* orphan block reservation for the snapshot */
2103         num_bytes = block_rsv->size;
2104
2105         /*
2106          * after the snapshot is created, COWing tree blocks may use more
2107          * space than it frees. So we should make sure there is enough
2108          * reserved space.
2109          */
2110         index = trans->transid & 0x1;
2111         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2112                 num_bytes += block_rsv->size -
2113                              (block_rsv->reserved + block_rsv->freed[index]);
2114         }
2115
2116         *bytes_to_reserve += num_bytes;
2117 }
2118
2119 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans,
2120                                 struct btrfs_pending_snapshot *pending)
2121 {
2122         struct btrfs_root *root = pending->root;
2123         struct btrfs_root *snap = pending->snap;
2124         struct btrfs_block_rsv *block_rsv;
2125         u64 num_bytes;
2126         int index;
2127         int ret;
2128
2129         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2130                 return;
2131
2132         /* refill source subvolume's orphan block reservation */
2133         block_rsv = root->orphan_block_rsv;
2134         index = trans->transid & 0x1;
2135         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2136                 num_bytes = block_rsv->size -
2137                             (block_rsv->reserved + block_rsv->freed[index]);
2138                 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2139                                               root->orphan_block_rsv,
2140                                               num_bytes);
2141                 BUG_ON(ret);
2142         }
2143
2144         /* setup orphan block reservation for the snapshot */
2145         block_rsv = btrfs_alloc_block_rsv(snap);
2146         BUG_ON(!block_rsv);
2147
2148         btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2149         snap->orphan_block_rsv = block_rsv;
2150
2151         num_bytes = root->orphan_block_rsv->size;
2152         ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2153                                       block_rsv, num_bytes);
2154         BUG_ON(ret);
2155
2156 #if 0
2157         /* insert orphan item for the snapshot */
2158         WARN_ON(!root->orphan_item_inserted);
2159         ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2160                                        snap->root_key.objectid);
2161         BUG_ON(ret);
2162         snap->orphan_item_inserted = 1;
2163 #endif
2164 }
2165
2166 enum btrfs_orphan_cleanup_state {
2167         ORPHAN_CLEANUP_STARTED  = 1,
2168         ORPHAN_CLEANUP_DONE     = 2,
2169 };
2170
2171 /*
2172  * This is called in transaction commmit time. If there are no orphan
2173  * files in the subvolume, it removes orphan item and frees block_rsv
2174  * structure.
2175  */
2176 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2177                               struct btrfs_root *root)
2178 {
2179         int ret;
2180
2181         if (!list_empty(&root->orphan_list) ||
2182             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2183                 return;
2184
2185         if (root->orphan_item_inserted &&
2186             btrfs_root_refs(&root->root_item) > 0) {
2187                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2188                                             root->root_key.objectid);
2189                 BUG_ON(ret);
2190                 root->orphan_item_inserted = 0;
2191         }
2192
2193         if (root->orphan_block_rsv) {
2194                 WARN_ON(root->orphan_block_rsv->size > 0);
2195                 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2196                 root->orphan_block_rsv = NULL;
2197         }
2198 }
2199
2200 /*
2201  * This creates an orphan entry for the given inode in case something goes
2202  * wrong in the middle of an unlink/truncate.
2203  *
2204  * NOTE: caller of this function should reserve 5 units of metadata for
2205  *       this function.
2206  */
2207 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2208 {
2209         struct btrfs_root *root = BTRFS_I(inode)->root;
2210         struct btrfs_block_rsv *block_rsv = NULL;
2211         int reserve = 0;
2212         int insert = 0;
2213         int ret;
2214
2215         if (!root->orphan_block_rsv) {
2216                 block_rsv = btrfs_alloc_block_rsv(root);
2217                 BUG_ON(!block_rsv);
2218         }
2219
2220         spin_lock(&root->orphan_lock);
2221         if (!root->orphan_block_rsv) {
2222                 root->orphan_block_rsv = block_rsv;
2223         } else if (block_rsv) {
2224                 btrfs_free_block_rsv(root, block_rsv);
2225                 block_rsv = NULL;
2226         }
2227
2228         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2229                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2230 #if 0
2231                 /*
2232                  * For proper ENOSPC handling, we should do orphan
2233                  * cleanup when mounting. But this introduces backward
2234                  * compatibility issue.
2235                  */
2236                 if (!xchg(&root->orphan_item_inserted, 1))
2237                         insert = 2;
2238                 else
2239                         insert = 1;
2240 #endif
2241                 insert = 1;
2242         }
2243
2244         if (!BTRFS_I(inode)->orphan_meta_reserved) {
2245                 BTRFS_I(inode)->orphan_meta_reserved = 1;
2246                 reserve = 1;
2247         }
2248         spin_unlock(&root->orphan_lock);
2249
2250         if (block_rsv)
2251                 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2252
2253         /* grab metadata reservation from transaction handle */
2254         if (reserve) {
2255                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2256                 BUG_ON(ret);
2257         }
2258
2259         /* insert an orphan item to track this unlinked/truncated file */
2260         if (insert >= 1) {
2261                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2262                 BUG_ON(ret);
2263         }
2264
2265         /* insert an orphan item to track subvolume contains orphan files */
2266         if (insert >= 2) {
2267                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2268                                                root->root_key.objectid);
2269                 BUG_ON(ret);
2270         }
2271         return 0;
2272 }
2273
2274 /*
2275  * We have done the truncate/delete so we can go ahead and remove the orphan
2276  * item for this particular inode.
2277  */
2278 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2279 {
2280         struct btrfs_root *root = BTRFS_I(inode)->root;
2281         int delete_item = 0;
2282         int release_rsv = 0;
2283         int ret = 0;
2284
2285         spin_lock(&root->orphan_lock);
2286         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2287                 list_del_init(&BTRFS_I(inode)->i_orphan);
2288                 delete_item = 1;
2289         }
2290
2291         if (BTRFS_I(inode)->orphan_meta_reserved) {
2292                 BTRFS_I(inode)->orphan_meta_reserved = 0;
2293                 release_rsv = 1;
2294         }
2295         spin_unlock(&root->orphan_lock);
2296
2297         if (trans && delete_item) {
2298                 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2299                 BUG_ON(ret);
2300         }
2301
2302         if (release_rsv)
2303                 btrfs_orphan_release_metadata(inode);
2304
2305         return 0;
2306 }
2307
2308 /*
2309  * this cleans up any orphans that may be left on the list from the last use
2310  * of this root.
2311  */
2312 int btrfs_orphan_cleanup(struct btrfs_root *root)
2313 {
2314         struct btrfs_path *path;
2315         struct extent_buffer *leaf;
2316         struct btrfs_key key, found_key;
2317         struct btrfs_trans_handle *trans;
2318         struct inode *inode;
2319         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2320
2321         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2322                 return 0;
2323
2324         path = btrfs_alloc_path();
2325         if (!path) {
2326                 ret = -ENOMEM;
2327                 goto out;
2328         }
2329         path->reada = -1;
2330
2331         key.objectid = BTRFS_ORPHAN_OBJECTID;
2332         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2333         key.offset = (u64)-1;
2334
2335         while (1) {
2336                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2337                 if (ret < 0)
2338                         goto out;
2339
2340                 /*
2341                  * if ret == 0 means we found what we were searching for, which
2342                  * is weird, but possible, so only screw with path if we didn't
2343                  * find the key and see if we have stuff that matches
2344                  */
2345                 if (ret > 0) {
2346                         ret = 0;
2347                         if (path->slots[0] == 0)
2348                                 break;
2349                         path->slots[0]--;
2350                 }
2351
2352                 /* pull out the item */
2353                 leaf = path->nodes[0];
2354                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2355
2356                 /* make sure the item matches what we want */
2357                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2358                         break;
2359                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2360                         break;
2361
2362                 /* release the path since we're done with it */
2363                 btrfs_release_path(path);
2364
2365                 /*
2366                  * this is where we are basically btrfs_lookup, without the
2367                  * crossing root thing.  we store the inode number in the
2368                  * offset of the orphan item.
2369                  */
2370                 found_key.objectid = found_key.offset;
2371                 found_key.type = BTRFS_INODE_ITEM_KEY;
2372                 found_key.offset = 0;
2373                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2374                 if (IS_ERR(inode)) {
2375                         ret = PTR_ERR(inode);
2376                         goto out;
2377                 }
2378
2379                 /*
2380                  * add this inode to the orphan list so btrfs_orphan_del does
2381                  * the proper thing when we hit it
2382                  */
2383                 spin_lock(&root->orphan_lock);
2384                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2385                 spin_unlock(&root->orphan_lock);
2386
2387                 /*
2388                  * if this is a bad inode, means we actually succeeded in
2389                  * removing the inode, but not the orphan record, which means
2390                  * we need to manually delete the orphan since iput will just
2391                  * do a destroy_inode
2392                  */
2393                 if (is_bad_inode(inode)) {
2394                         trans = btrfs_start_transaction(root, 0);
2395                         if (IS_ERR(trans)) {
2396                                 ret = PTR_ERR(trans);
2397                                 goto out;
2398                         }
2399                         btrfs_orphan_del(trans, inode);
2400                         btrfs_end_transaction(trans, root);
2401                         iput(inode);
2402                         continue;
2403                 }
2404
2405                 /* if we have links, this was a truncate, lets do that */
2406                 if (inode->i_nlink) {
2407                         if (!S_ISREG(inode->i_mode)) {
2408                                 WARN_ON(1);
2409                                 iput(inode);
2410                                 continue;
2411                         }
2412                         nr_truncate++;
2413                         ret = btrfs_truncate(inode);
2414                 } else {
2415                         nr_unlink++;
2416                 }
2417
2418                 /* this will do delete_inode and everything for us */
2419                 iput(inode);
2420                 if (ret)
2421                         goto out;
2422         }
2423         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2424
2425         if (root->orphan_block_rsv)
2426                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2427                                         (u64)-1);
2428
2429         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2430                 trans = btrfs_join_transaction(root);
2431                 if (!IS_ERR(trans))
2432                         btrfs_end_transaction(trans, root);
2433         }
2434
2435         if (nr_unlink)
2436                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2437         if (nr_truncate)
2438                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2439
2440 out:
2441         if (ret)
2442                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2443         btrfs_free_path(path);
2444         return ret;
2445 }
2446
2447 /*
2448  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2449  * don't find any xattrs, we know there can't be any acls.
2450  *
2451  * slot is the slot the inode is in, objectid is the objectid of the inode
2452  */
2453 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2454                                           int slot, u64 objectid)
2455 {
2456         u32 nritems = btrfs_header_nritems(leaf);
2457         struct btrfs_key found_key;
2458         int scanned = 0;
2459
2460         slot++;
2461         while (slot < nritems) {
2462                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2463
2464                 /* we found a different objectid, there must not be acls */
2465                 if (found_key.objectid != objectid)
2466                         return 0;
2467
2468                 /* we found an xattr, assume we've got an acl */
2469                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2470                         return 1;
2471
2472                 /*
2473                  * we found a key greater than an xattr key, there can't
2474                  * be any acls later on
2475                  */
2476                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2477                         return 0;
2478
2479                 slot++;
2480                 scanned++;
2481
2482                 /*
2483                  * it goes inode, inode backrefs, xattrs, extents,
2484                  * so if there are a ton of hard links to an inode there can
2485                  * be a lot of backrefs.  Don't waste time searching too hard,
2486                  * this is just an optimization
2487                  */
2488                 if (scanned >= 8)
2489                         break;
2490         }
2491         /* we hit the end of the leaf before we found an xattr or
2492          * something larger than an xattr.  We have to assume the inode
2493          * has acls
2494          */
2495         return 1;
2496 }
2497
2498 /*
2499  * read an inode from the btree into the in-memory inode
2500  */
2501 static void btrfs_read_locked_inode(struct inode *inode)
2502 {
2503         struct btrfs_path *path;
2504         struct extent_buffer *leaf;
2505         struct btrfs_inode_item *inode_item;
2506         struct btrfs_timespec *tspec;
2507         struct btrfs_root *root = BTRFS_I(inode)->root;
2508         struct btrfs_key location;
2509         int maybe_acls;
2510         u32 rdev;
2511         int ret;
2512         bool filled = false;
2513
2514         ret = btrfs_fill_inode(inode, &rdev);
2515         if (!ret)
2516                 filled = true;
2517
2518         path = btrfs_alloc_path();
2519         BUG_ON(!path);
2520         path->leave_spinning = 1;
2521         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2522
2523         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2524         if (ret)
2525                 goto make_bad;
2526
2527         leaf = path->nodes[0];
2528
2529         if (filled)
2530                 goto cache_acl;
2531
2532         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2533                                     struct btrfs_inode_item);
2534         if (!leaf->map_token)
2535                 map_private_extent_buffer(leaf, (unsigned long)inode_item,
2536                                           sizeof(struct btrfs_inode_item),
2537                                           &leaf->map_token, &leaf->kaddr,
2538                                           &leaf->map_start, &leaf->map_len,
2539                                           KM_USER1);
2540
2541         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2542         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2543         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2544         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2545         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2546
2547         tspec = btrfs_inode_atime(inode_item);
2548         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2549         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2550
2551         tspec = btrfs_inode_mtime(inode_item);
2552         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2553         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2554
2555         tspec = btrfs_inode_ctime(inode_item);
2556         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2557         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2558
2559         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2560         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2561         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2562         inode->i_generation = BTRFS_I(inode)->generation;
2563         inode->i_rdev = 0;
2564         rdev = btrfs_inode_rdev(leaf, inode_item);
2565
2566         BTRFS_I(inode)->index_cnt = (u64)-1;
2567         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2568 cache_acl:
2569         /*
2570          * try to precache a NULL acl entry for files that don't have
2571          * any xattrs or acls
2572          */
2573         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2574                                            btrfs_ino(inode));
2575         if (!maybe_acls)
2576                 cache_no_acl(inode);
2577
2578         if (leaf->map_token) {
2579                 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2580                 leaf->map_token = NULL;
2581         }
2582
2583         btrfs_free_path(path);
2584
2585         switch (inode->i_mode & S_IFMT) {
2586         case S_IFREG:
2587                 inode->i_mapping->a_ops = &btrfs_aops;
2588                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2589                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2590                 inode->i_fop = &btrfs_file_operations;
2591                 inode->i_op = &btrfs_file_inode_operations;
2592                 break;
2593         case S_IFDIR:
2594                 inode->i_fop = &btrfs_dir_file_operations;
2595                 if (root == root->fs_info->tree_root)
2596                         inode->i_op = &btrfs_dir_ro_inode_operations;
2597                 else
2598                         inode->i_op = &btrfs_dir_inode_operations;
2599                 break;
2600         case S_IFLNK:
2601                 inode->i_op = &btrfs_symlink_inode_operations;
2602                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2603                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2604                 break;
2605         default:
2606                 inode->i_op = &btrfs_special_inode_operations;
2607                 init_special_inode(inode, inode->i_mode, rdev);
2608                 break;
2609         }
2610
2611         btrfs_update_iflags(inode);
2612         return;
2613
2614 make_bad:
2615         btrfs_free_path(path);
2616         make_bad_inode(inode);
2617 }
2618
2619 /*
2620  * given a leaf and an inode, copy the inode fields into the leaf
2621  */
2622 static void fill_inode_item(struct btrfs_trans_handle *trans,
2623                             struct extent_buffer *leaf,
2624                             struct btrfs_inode_item *item,
2625                             struct inode *inode)
2626 {
2627         if (!leaf->map_token)
2628                 map_private_extent_buffer(leaf, (unsigned long)item,
2629                                           sizeof(struct btrfs_inode_item),
2630                                           &leaf->map_token, &leaf->kaddr,
2631                                           &leaf->map_start, &leaf->map_len,
2632                                           KM_USER1);
2633
2634         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2635         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2636         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2637         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2638         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2639
2640         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2641                                inode->i_atime.tv_sec);
2642         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2643                                 inode->i_atime.tv_nsec);
2644
2645         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2646                                inode->i_mtime.tv_sec);
2647         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2648                                 inode->i_mtime.tv_nsec);
2649
2650         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2651                                inode->i_ctime.tv_sec);
2652         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2653                                 inode->i_ctime.tv_nsec);
2654
2655         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2656         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2657         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2658         btrfs_set_inode_transid(leaf, item, trans->transid);
2659         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2660         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2661         btrfs_set_inode_block_group(leaf, item, 0);
2662
2663         if (leaf->map_token) {
2664                 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2665                 leaf->map_token = NULL;
2666         }
2667 }
2668
2669 /*
2670  * copy everything in the in-memory inode into the btree.
2671  */
2672 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2673                                 struct btrfs_root *root, struct inode *inode)
2674 {
2675         struct btrfs_inode_item *inode_item;
2676         struct btrfs_path *path;
2677         struct extent_buffer *leaf;
2678         int ret;
2679
2680         /*
2681          * If root is tree root, it means this inode is used to
2682          * store free space information. And these inodes are updated
2683          * when committing the transaction, so they needn't delaye to
2684          * be updated, or deadlock will occured.
2685          */
2686         if (!is_free_space_inode(root, inode)) {
2687                 ret = btrfs_delayed_update_inode(trans, root, inode);
2688                 if (!ret)
2689                         btrfs_set_inode_last_trans(trans, inode);
2690                 return ret;
2691         }
2692
2693         path = btrfs_alloc_path();
2694         if (!path)
2695                 return -ENOMEM;
2696
2697         path->leave_spinning = 1;
2698         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2699                                  1);
2700         if (ret) {
2701                 if (ret > 0)
2702                         ret = -ENOENT;
2703                 goto failed;
2704         }
2705
2706         btrfs_unlock_up_safe(path, 1);
2707         leaf = path->nodes[0];
2708         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2709                                     struct btrfs_inode_item);
2710
2711         fill_inode_item(trans, leaf, inode_item, inode);
2712         btrfs_mark_buffer_dirty(leaf);
2713         btrfs_set_inode_last_trans(trans, inode);
2714         ret = 0;
2715 failed:
2716         btrfs_free_path(path);
2717         return ret;
2718 }
2719
2720 /*
2721  * unlink helper that gets used here in inode.c and in the tree logging
2722  * recovery code.  It remove a link in a directory with a given name, and
2723  * also drops the back refs in the inode to the directory
2724  */
2725 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2726                                 struct btrfs_root *root,
2727                                 struct inode *dir, struct inode *inode,
2728                                 const char *name, int name_len)
2729 {
2730         struct btrfs_path *path;
2731         int ret = 0;
2732         struct extent_buffer *leaf;
2733         struct btrfs_dir_item *di;
2734         struct btrfs_key key;
2735         u64 index;
2736         u64 ino = btrfs_ino(inode);
2737         u64 dir_ino = btrfs_ino(dir);
2738
2739         path = btrfs_alloc_path();
2740         if (!path) {
2741                 ret = -ENOMEM;
2742                 goto out;
2743         }
2744
2745         path->leave_spinning = 1;
2746         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2747                                     name, name_len, -1);
2748         if (IS_ERR(di)) {
2749                 ret = PTR_ERR(di);
2750                 goto err;
2751         }
2752         if (!di) {
2753                 ret = -ENOENT;
2754                 goto err;
2755         }
2756         leaf = path->nodes[0];
2757         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2758         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2759         if (ret)
2760                 goto err;
2761         btrfs_release_path(path);
2762
2763         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2764                                   dir_ino, &index);
2765         if (ret) {
2766                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2767                        "inode %llu parent %llu\n", name_len, name,
2768                        (unsigned long long)ino, (unsigned long long)dir_ino);
2769                 goto err;
2770         }
2771
2772         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2773         if (ret)
2774                 goto err;
2775
2776         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2777                                          inode, dir_ino);
2778         BUG_ON(ret != 0 && ret != -ENOENT);
2779
2780         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2781                                            dir, index);
2782         if (ret == -ENOENT)
2783                 ret = 0;
2784 err:
2785         btrfs_free_path(path);
2786         if (ret)
2787                 goto out;
2788
2789         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2790         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2791         btrfs_update_inode(trans, root, dir);
2792 out:
2793         return ret;
2794 }
2795
2796 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2797                        struct btrfs_root *root,
2798                        struct inode *dir, struct inode *inode,
2799                        const char *name, int name_len)
2800 {
2801         int ret;
2802         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2803         if (!ret) {
2804                 btrfs_drop_nlink(inode);
2805                 ret = btrfs_update_inode(trans, root, inode);
2806         }
2807         return ret;
2808 }
2809
2810
2811 /* helper to check if there is any shared block in the path */
2812 static int check_path_shared(struct btrfs_root *root,
2813                              struct btrfs_path *path)
2814 {
2815         struct extent_buffer *eb;
2816         int level;
2817         u64 refs = 1;
2818
2819         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2820                 int ret;
2821
2822                 if (!path->nodes[level])
2823                         break;
2824                 eb = path->nodes[level];
2825                 if (!btrfs_block_can_be_shared(root, eb))
2826                         continue;
2827                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2828                                                &refs, NULL);
2829                 if (refs > 1)
2830                         return 1;
2831         }
2832         return 0;
2833 }
2834
2835 /*
2836  * helper to start transaction for unlink and rmdir.
2837  *
2838  * unlink and rmdir are special in btrfs, they do not always free space.
2839  * so in enospc case, we should make sure they will free space before
2840  * allowing them to use the global metadata reservation.
2841  */
2842 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2843                                                        struct dentry *dentry)
2844 {
2845         struct btrfs_trans_handle *trans;
2846         struct btrfs_root *root = BTRFS_I(dir)->root;
2847         struct btrfs_path *path;
2848         struct btrfs_inode_ref *ref;
2849         struct btrfs_dir_item *di;
2850         struct inode *inode = dentry->d_inode;
2851         u64 index;
2852         int check_link = 1;
2853         int err = -ENOSPC;
2854         int ret;
2855         u64 ino = btrfs_ino(inode);
2856         u64 dir_ino = btrfs_ino(dir);
2857
2858         trans = btrfs_start_transaction(root, 10);
2859         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2860                 return trans;
2861
2862         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2863                 return ERR_PTR(-ENOSPC);
2864
2865         /* check if there is someone else holds reference */
2866         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2867                 return ERR_PTR(-ENOSPC);
2868
2869         if (atomic_read(&inode->i_count) > 2)
2870                 return ERR_PTR(-ENOSPC);
2871
2872         if (xchg(&root->fs_info->enospc_unlink, 1))
2873                 return ERR_PTR(-ENOSPC);
2874
2875         path = btrfs_alloc_path();
2876         if (!path) {
2877                 root->fs_info->enospc_unlink = 0;
2878                 return ERR_PTR(-ENOMEM);
2879         }
2880
2881         trans = btrfs_start_transaction(root, 0);
2882         if (IS_ERR(trans)) {
2883                 btrfs_free_path(path);
2884                 root->fs_info->enospc_unlink = 0;
2885                 return trans;
2886         }
2887
2888         path->skip_locking = 1;
2889         path->search_commit_root = 1;
2890
2891         ret = btrfs_lookup_inode(trans, root, path,
2892                                 &BTRFS_I(dir)->location, 0);
2893         if (ret < 0) {
2894                 err = ret;
2895                 goto out;
2896         }
2897         if (ret == 0) {
2898                 if (check_path_shared(root, path))
2899                         goto out;
2900         } else {
2901                 check_link = 0;
2902         }
2903         btrfs_release_path(path);
2904
2905         ret = btrfs_lookup_inode(trans, root, path,
2906                                 &BTRFS_I(inode)->location, 0);
2907         if (ret < 0) {
2908                 err = ret;
2909                 goto out;
2910         }
2911         if (ret == 0) {
2912                 if (check_path_shared(root, path))
2913                         goto out;
2914         } else {
2915                 check_link = 0;
2916         }
2917         btrfs_release_path(path);
2918
2919         if (ret == 0 && S_ISREG(inode->i_mode)) {
2920                 ret = btrfs_lookup_file_extent(trans, root, path,
2921                                                ino, (u64)-1, 0);
2922                 if (ret < 0) {
2923                         err = ret;
2924                         goto out;
2925                 }
2926                 BUG_ON(ret == 0);
2927                 if (check_path_shared(root, path))
2928                         goto out;
2929                 btrfs_release_path(path);
2930         }
2931
2932         if (!check_link) {
2933                 err = 0;
2934                 goto out;
2935         }
2936
2937         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2938                                 dentry->d_name.name, dentry->d_name.len, 0);
2939         if (IS_ERR(di)) {
2940                 err = PTR_ERR(di);
2941                 goto out;
2942         }
2943         if (di) {
2944                 if (check_path_shared(root, path))
2945                         goto out;
2946         } else {
2947                 err = 0;
2948                 goto out;
2949         }
2950         btrfs_release_path(path);
2951
2952         ref = btrfs_lookup_inode_ref(trans, root, path,
2953                                 dentry->d_name.name, dentry->d_name.len,
2954                                 ino, dir_ino, 0);
2955         if (IS_ERR(ref)) {
2956                 err = PTR_ERR(ref);
2957                 goto out;
2958         }
2959         BUG_ON(!ref);
2960         if (check_path_shared(root, path))
2961                 goto out;
2962         index = btrfs_inode_ref_index(path->nodes[0], ref);
2963         btrfs_release_path(path);
2964
2965         /*
2966          * This is a commit root search, if we can lookup inode item and other
2967          * relative items in the commit root, it means the transaction of
2968          * dir/file creation has been committed, and the dir index item that we
2969          * delay to insert has also been inserted into the commit root. So
2970          * we needn't worry about the delayed insertion of the dir index item
2971          * here.
2972          */
2973         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
2974                                 dentry->d_name.name, dentry->d_name.len, 0);
2975         if (IS_ERR(di)) {
2976                 err = PTR_ERR(di);
2977                 goto out;
2978         }
2979         BUG_ON(ret == -ENOENT);
2980         if (check_path_shared(root, path))
2981                 goto out;
2982
2983         err = 0;
2984 out:
2985         btrfs_free_path(path);
2986         if (err) {
2987                 btrfs_end_transaction(trans, root);
2988                 root->fs_info->enospc_unlink = 0;
2989                 return ERR_PTR(err);
2990         }
2991
2992         trans->block_rsv = &root->fs_info->global_block_rsv;
2993         return trans;
2994 }
2995
2996 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2997                                struct btrfs_root *root)
2998 {
2999         if (trans->block_rsv == &root->fs_info->global_block_rsv) {
3000                 BUG_ON(!root->fs_info->enospc_unlink);
3001                 root->fs_info->enospc_unlink = 0;
3002         }
3003         btrfs_end_transaction_throttle(trans, root);
3004 }
3005
3006 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3007 {
3008         struct btrfs_root *root = BTRFS_I(dir)->root;
3009         struct btrfs_trans_handle *trans;
3010         struct inode *inode = dentry->d_inode;
3011         int ret;
3012         unsigned long nr = 0;
3013
3014         trans = __unlink_start_trans(dir, dentry);
3015         if (IS_ERR(trans))
3016                 return PTR_ERR(trans);
3017
3018         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3019
3020         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3021                                  dentry->d_name.name, dentry->d_name.len);
3022         BUG_ON(ret);
3023
3024         if (inode->i_nlink == 0) {
3025                 ret = btrfs_orphan_add(trans, inode);
3026                 BUG_ON(ret);
3027         }
3028
3029         nr = trans->blocks_used;
3030         __unlink_end_trans(trans, root);
3031         btrfs_btree_balance_dirty(root, nr);
3032         return ret;
3033 }
3034
3035 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3036                         struct btrfs_root *root,
3037                         struct inode *dir, u64 objectid,
3038                         const char *name, int name_len)
3039 {
3040         struct btrfs_path *path;
3041         struct extent_buffer *leaf;
3042         struct btrfs_dir_item *di;
3043         struct btrfs_key key;
3044         u64 index;
3045         int ret;
3046         u64 dir_ino = btrfs_ino(dir);
3047
3048         path = btrfs_alloc_path();
3049         if (!path)
3050                 return -ENOMEM;
3051
3052         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3053                                    name, name_len, -1);
3054         BUG_ON(IS_ERR_OR_NULL(di));
3055
3056         leaf = path->nodes[0];
3057         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3058         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3059         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3060         BUG_ON(ret);
3061         btrfs_release_path(path);
3062
3063         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3064                                  objectid, root->root_key.objectid,
3065                                  dir_ino, &index, name, name_len);
3066         if (ret < 0) {
3067                 BUG_ON(ret != -ENOENT);
3068                 di = btrfs_search_dir_index_item(root, path, dir_ino,
3069                                                  name, name_len);
3070                 BUG_ON(IS_ERR_OR_NULL(di));
3071
3072                 leaf = path->nodes[0];
3073                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3074                 btrfs_release_path(path);
3075                 index = key.offset;
3076         }
3077         btrfs_release_path(path);
3078
3079         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3080         BUG_ON(ret);
3081
3082         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3083         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3084         ret = btrfs_update_inode(trans, root, dir);
3085         BUG_ON(ret);
3086
3087         btrfs_free_path(path);
3088         return 0;
3089 }
3090
3091 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3092 {
3093         struct inode *inode = dentry->d_inode;
3094         int err = 0;
3095         struct btrfs_root *root = BTRFS_I(dir)->root;
3096         struct btrfs_trans_handle *trans;
3097         unsigned long nr = 0;
3098
3099         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3100             btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3101                 return -ENOTEMPTY;
3102
3103         trans = __unlink_start_trans(dir, dentry);
3104         if (IS_ERR(trans))
3105                 return PTR_ERR(trans);
3106
3107         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3108                 err = btrfs_unlink_subvol(trans, root, dir,
3109                                           BTRFS_I(inode)->location.objectid,
3110                                           dentry->d_name.name,
3111                                           dentry->d_name.len);
3112                 goto out;
3113         }
3114
3115         err = btrfs_orphan_add(trans, inode);
3116         if (err)
3117                 goto out;
3118
3119         /* now the directory is empty */
3120         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3121                                  dentry->d_name.name, dentry->d_name.len);
3122         if (!err)
3123                 btrfs_i_size_write(inode, 0);
3124 out:
3125         nr = trans->blocks_used;
3126         __unlink_end_trans(trans, root);
3127         btrfs_btree_balance_dirty(root, nr);
3128
3129         return err;
3130 }
3131
3132 /*
3133  * this can truncate away extent items, csum items and directory items.
3134  * It starts at a high offset and removes keys until it can't find
3135  * any higher than new_size
3136  *
3137  * csum items that cross the new i_size are truncated to the new size
3138  * as well.
3139  *
3140  * min_type is the minimum key type to truncate down to.  If set to 0, this
3141  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3142  */
3143 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3144                                struct btrfs_root *root,
3145                                struct inode *inode,
3146                                u64 new_size, u32 min_type)
3147 {
3148         struct btrfs_path *path;
3149         struct extent_buffer *leaf;
3150         struct btrfs_file_extent_item *fi;
3151         struct btrfs_key key;
3152         struct btrfs_key found_key;
3153         u64 extent_start = 0;
3154         u64 extent_num_bytes = 0;
3155         u64 extent_offset = 0;
3156         u64 item_end = 0;
3157         u64 mask = root->sectorsize - 1;
3158         u32 found_type = (u8)-1;
3159         int found_extent;
3160         int del_item;
3161         int pending_del_nr = 0;
3162         int pending_del_slot = 0;
3163         int extent_type = -1;
3164         int encoding;
3165         int ret;
3166         int err = 0;
3167         u64 ino = btrfs_ino(inode);
3168
3169         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3170
3171         if (root->ref_cows || root == root->fs_info->tree_root)
3172                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3173
3174         /*
3175          * This function is also used to drop the items in the log tree before
3176          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3177          * it is used to drop the loged items. So we shouldn't kill the delayed
3178          * items.
3179          */
3180         if (min_type == 0 && root == BTRFS_I(inode)->root)
3181                 btrfs_kill_delayed_inode_items(inode);
3182
3183         path = btrfs_alloc_path();
3184         BUG_ON(!path);
3185         path->reada = -1;
3186
3187         key.objectid = ino;
3188         key.offset = (u64)-1;
3189         key.type = (u8)-1;
3190
3191 search_again:
3192         path->leave_spinning = 1;
3193         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3194         if (ret < 0) {
3195                 err = ret;
3196                 goto out;
3197         }
3198
3199         if (ret > 0) {
3200                 /* there are no items in the tree for us to truncate, we're
3201                  * done
3202                  */
3203                 if (path->slots[0] == 0)
3204                         goto out;
3205                 path->slots[0]--;
3206         }
3207
3208         while (1) {
3209                 fi = NULL;
3210                 leaf = path->nodes[0];
3211                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3212                 found_type = btrfs_key_type(&found_key);
3213                 encoding = 0;
3214
3215                 if (found_key.objectid != ino)
3216                         break;
3217
3218                 if (found_type < min_type)
3219                         break;
3220
3221                 item_end = found_key.offset;
3222                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3223                         fi = btrfs_item_ptr(leaf, path->slots[0],
3224                                             struct btrfs_file_extent_item);
3225                         extent_type = btrfs_file_extent_type(leaf, fi);
3226                         encoding = btrfs_file_extent_compression(leaf, fi);
3227                         encoding |= btrfs_file_extent_encryption(leaf, fi);
3228                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3229
3230                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3231                                 item_end +=
3232                                     btrfs_file_extent_num_bytes(leaf, fi);
3233                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3234                                 item_end += btrfs_file_extent_inline_len(leaf,
3235                                                                          fi);
3236                         }
3237                         item_end--;
3238                 }
3239                 if (found_type > min_type) {
3240                         del_item = 1;
3241                 } else {
3242                         if (item_end < new_size)
3243                                 break;
3244                         if (found_key.offset >= new_size)
3245                                 del_item = 1;
3246                         else
3247                                 del_item = 0;
3248                 }
3249                 found_extent = 0;
3250                 /* FIXME, shrink the extent if the ref count is only 1 */
3251                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3252                         goto delete;
3253
3254                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3255                         u64 num_dec;
3256                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3257                         if (!del_item && !encoding) {
3258                                 u64 orig_num_bytes =
3259                                         btrfs_file_extent_num_bytes(leaf, fi);
3260                                 extent_num_bytes = new_size -
3261                                         found_key.offset + root->sectorsize - 1;
3262                                 extent_num_bytes = extent_num_bytes &
3263                                         ~((u64)root->sectorsize - 1);
3264                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3265                                                          extent_num_bytes);
3266                                 num_dec = (orig_num_bytes -
3267                                            extent_num_bytes);
3268                                 if (root->ref_cows && extent_start != 0)
3269                                         inode_sub_bytes(inode, num_dec);
3270                                 btrfs_mark_buffer_dirty(leaf);
3271                         } else {
3272                                 extent_num_bytes =
3273                                         btrfs_file_extent_disk_num_bytes(leaf,
3274                                                                          fi);
3275                                 extent_offset = found_key.offset -
3276                                         btrfs_file_extent_offset(leaf, fi);
3277
3278                                 /* FIXME blocksize != 4096 */
3279                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3280                                 if (extent_start != 0) {
3281                                         found_extent = 1;
3282                                         if (root->ref_cows)
3283                                                 inode_sub_bytes(inode, num_dec);
3284                                 }
3285                         }
3286                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3287                         /*
3288                          * we can't truncate inline items that have had
3289                          * special encodings
3290                          */
3291                         if (!del_item &&
3292                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3293                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3294                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3295                                 u32 size = new_size - found_key.offset;
3296
3297                                 if (root->ref_cows) {
3298                                         inode_sub_bytes(inode, item_end + 1 -
3299                                                         new_size);
3300                                 }
3301                                 size =
3302                                     btrfs_file_extent_calc_inline_size(size);
3303                                 ret = btrfs_truncate_item(trans, root, path,
3304                                                           size, 1);
3305                         } else if (root->ref_cows) {
3306                                 inode_sub_bytes(inode, item_end + 1 -
3307                                                 found_key.offset);
3308                         }
3309                 }
3310 delete:
3311                 if (del_item) {
3312                         if (!pending_del_nr) {
3313                                 /* no pending yet, add ourselves */
3314                                 pending_del_slot = path->slots[0];
3315                                 pending_del_nr = 1;
3316                         } else if (pending_del_nr &&
3317                                    path->slots[0] + 1 == pending_del_slot) {
3318                                 /* hop on the pending chunk */
3319                                 pending_del_nr++;
3320                                 pending_del_slot = path->slots[0];
3321                         } else {
3322                                 BUG();
3323                         }
3324                 } else {
3325                         break;
3326                 }
3327                 if (found_extent && (root->ref_cows ||
3328                                      root == root->fs_info->tree_root)) {
3329                         btrfs_set_path_blocking(path);
3330                         ret = btrfs_free_extent(trans, root, extent_start,
3331                                                 extent_num_bytes, 0,
3332                                                 btrfs_header_owner(leaf),
3333                                                 ino, extent_offset);
3334                         BUG_ON(ret);
3335                 }
3336
3337                 if (found_type == BTRFS_INODE_ITEM_KEY)
3338                         break;
3339
3340                 if (path->slots[0] == 0 ||
3341                     path->slots[0] != pending_del_slot) {
3342                         if (root->ref_cows &&
3343                             BTRFS_I(inode)->location.objectid !=
3344                                                 BTRFS_FREE_INO_OBJECTID) {
3345                                 err = -EAGAIN;
3346                                 goto out;
3347                         }
3348                         if (pending_del_nr) {
3349                                 ret = btrfs_del_items(trans, root, path,
3350                                                 pending_del_slot,
3351                                                 pending_del_nr);
3352                                 BUG_ON(ret);
3353                                 pending_del_nr = 0;
3354                         }
3355                         btrfs_release_path(path);
3356                         goto search_again;
3357                 } else {
3358                         path->slots[0]--;
3359                 }
3360         }
3361 out:
3362         if (pending_del_nr) {
3363                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3364                                       pending_del_nr);
3365                 BUG_ON(ret);
3366         }
3367         btrfs_free_path(path);
3368         return err;
3369 }
3370
3371 /*
3372  * taken from block_truncate_page, but does cow as it zeros out
3373  * any bytes left in the last page in the file.
3374  */
3375 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3376 {
3377         struct inode *inode = mapping->host;
3378         struct btrfs_root *root = BTRFS_I(inode)->root;
3379         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3380         struct btrfs_ordered_extent *ordered;
3381         struct extent_state *cached_state = NULL;
3382         char *kaddr;
3383         u32 blocksize = root->sectorsize;
3384         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3385         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3386         struct page *page;
3387         int ret = 0;
3388         u64 page_start;
3389         u64 page_end;
3390
3391         if ((offset & (blocksize - 1)) == 0)
3392                 goto out;
3393         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3394         if (ret)
3395                 goto out;
3396
3397         ret = -ENOMEM;
3398 again:
3399         page = grab_cache_page(mapping, index);
3400         if (!page) {
3401                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3402                 goto out;
3403         }
3404
3405         page_start = page_offset(page);
3406         page_end = page_start + PAGE_CACHE_SIZE - 1;
3407
3408         if (!PageUptodate(page)) {
3409                 ret = btrfs_readpage(NULL, page);
3410                 lock_page(page);
3411                 if (page->mapping != mapping) {
3412                         unlock_page(page);
3413                         page_cache_release(page);
3414                         goto again;
3415                 }
3416                 if (!PageUptodate(page)) {
3417                         ret = -EIO;
3418                         goto out_unlock;
3419                 }
3420         }
3421         wait_on_page_writeback(page);
3422
3423         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3424                          GFP_NOFS);
3425         set_page_extent_mapped(page);
3426
3427         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3428         if (ordered) {
3429                 unlock_extent_cached(io_tree, page_start, page_end,
3430                                      &cached_state, GFP_NOFS);
3431                 unlock_page(page);
3432                 page_cache_release(page);
3433                 btrfs_start_ordered_extent(inode, ordered, 1);
3434                 btrfs_put_ordered_extent(ordered);
3435                 goto again;
3436         }
3437
3438         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3439                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3440                           0, 0, &cached_state, GFP_NOFS);
3441
3442         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3443                                         &cached_state);
3444         if (ret) {
3445                 unlock_extent_cached(io_tree, page_start, page_end,
3446                                      &cached_state, GFP_NOFS);
3447                 goto out_unlock;
3448         }
3449
3450         ret = 0;
3451         if (offset != PAGE_CACHE_SIZE) {
3452                 kaddr = kmap(page);
3453                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3454                 flush_dcache_page(page);
3455                 kunmap(page);
3456         }
3457         ClearPageChecked(page);
3458         set_page_dirty(page);
3459         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3460                              GFP_NOFS);
3461
3462 out_unlock:
3463         if (ret)
3464                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3465         unlock_page(page);
3466         page_cache_release(page);
3467 out:
3468         return ret;
3469 }
3470
3471 /*
3472  * This function puts in dummy file extents for the area we're creating a hole
3473  * for.  So if we are truncating this file to a larger size we need to insert
3474  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3475  * the range between oldsize and size
3476  */
3477 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3478 {
3479         struct btrfs_trans_handle *trans;
3480         struct btrfs_root *root = BTRFS_I(inode)->root;
3481         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3482         struct extent_map *em = NULL;
3483         struct extent_state *cached_state = NULL;
3484         u64 mask = root->sectorsize - 1;
3485         u64 hole_start = (oldsize + mask) & ~mask;
3486         u64 block_end = (size + mask) & ~mask;
3487         u64 last_byte;
3488         u64 cur_offset;
3489         u64 hole_size;
3490         int err = 0;
3491
3492         if (size <= hole_start)
3493                 return 0;
3494
3495         while (1) {
3496                 struct btrfs_ordered_extent *ordered;
3497                 btrfs_wait_ordered_range(inode, hole_start,
3498                                          block_end - hole_start);
3499                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3500                                  &cached_state, GFP_NOFS);
3501                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3502                 if (!ordered)
3503                         break;
3504                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3505                                      &cached_state, GFP_NOFS);
3506                 btrfs_put_ordered_extent(ordered);
3507         }
3508
3509         cur_offset = hole_start;
3510         while (1) {
3511                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3512                                 block_end - cur_offset, 0);
3513                 BUG_ON(IS_ERR_OR_NULL(em));
3514                 last_byte = min(extent_map_end(em), block_end);
3515                 last_byte = (last_byte + mask) & ~mask;
3516                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3517                         u64 hint_byte = 0;
3518                         hole_size = last_byte - cur_offset;
3519
3520                         trans = btrfs_start_transaction(root, 2);
3521                         if (IS_ERR(trans)) {
3522                                 err = PTR_ERR(trans);
3523                                 break;
3524                         }
3525
3526                         err = btrfs_drop_extents(trans, inode, cur_offset,
3527                                                  cur_offset + hole_size,
3528                                                  &hint_byte, 1);
3529                         if (err)
3530                                 break;
3531
3532                         err = btrfs_insert_file_extent(trans, root,
3533                                         btrfs_ino(inode), cur_offset, 0,
3534                                         0, hole_size, 0, hole_size,
3535                                         0, 0, 0);
3536                         if (err)
3537                                 break;
3538
3539                         btrfs_drop_extent_cache(inode, hole_start,
3540                                         last_byte - 1, 0);
3541
3542                         btrfs_end_transaction(trans, root);
3543                 }
3544                 free_extent_map(em);
3545                 em = NULL;
3546                 cur_offset = last_byte;
3547                 if (cur_offset >= block_end)
3548                         break;
3549         }
3550
3551         free_extent_map(em);
3552         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3553                              GFP_NOFS);
3554         return err;
3555 }
3556
3557 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3558 {
3559         loff_t oldsize = i_size_read(inode);
3560         int ret;
3561
3562         if (newsize == oldsize)
3563                 return 0;
3564
3565         if (newsize > oldsize) {
3566                 i_size_write(inode, newsize);
3567                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3568                 truncate_pagecache(inode, oldsize, newsize);
3569                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3570                 if (ret) {
3571                         btrfs_setsize(inode, oldsize);
3572                         return ret;
3573                 }
3574
3575                 mark_inode_dirty(inode);
3576         } else {
3577
3578                 /*
3579                  * We're truncating a file that used to have good data down to
3580                  * zero. Make sure it gets into the ordered flush list so that
3581                  * any new writes get down to disk quickly.
3582                  */
3583                 if (newsize == 0)
3584                         BTRFS_I(inode)->ordered_data_close = 1;
3585
3586                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3587                 truncate_setsize(inode, newsize);
3588                 ret = btrfs_truncate(inode);
3589         }
3590
3591         return ret;
3592 }
3593
3594 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3595 {
3596         struct inode *inode = dentry->d_inode;
3597         struct btrfs_root *root = BTRFS_I(inode)->root;
3598         int err;
3599
3600         if (btrfs_root_readonly(root))
3601                 return -EROFS;
3602
3603         err = inode_change_ok(inode, attr);
3604         if (err)
3605                 return err;
3606
3607         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3608                 err = btrfs_setsize(inode, attr->ia_size);
3609                 if (err)
3610                         return err;
3611         }
3612
3613         if (attr->ia_valid) {
3614                 setattr_copy(inode, attr);
3615                 mark_inode_dirty(inode);
3616
3617                 if (attr->ia_valid & ATTR_MODE)
3618                         err = btrfs_acl_chmod(inode);
3619         }
3620
3621         return err;
3622 }
3623
3624 void btrfs_evict_inode(struct inode *inode)
3625 {
3626         struct btrfs_trans_handle *trans;
3627         struct btrfs_root *root = BTRFS_I(inode)->root;
3628         unsigned long nr;
3629         int ret;
3630
3631         trace_btrfs_inode_evict(inode);
3632
3633         truncate_inode_pages(&inode->i_data, 0);
3634         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3635                                is_free_space_inode(root, inode)))
3636                 goto no_delete;
3637
3638         if (is_bad_inode(inode)) {
3639                 btrfs_orphan_del(NULL, inode);
3640                 goto no_delete;
3641         }
3642         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3643         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3644
3645         if (root->fs_info->log_root_recovering) {
3646                 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3647                 goto no_delete;
3648         }
3649
3650         if (inode->i_nlink > 0) {
3651                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3652                 goto no_delete;
3653         }
3654
3655         btrfs_i_size_write(inode, 0);
3656
3657         while (1) {
3658                 trans = btrfs_join_transaction(root);
3659                 BUG_ON(IS_ERR(trans));
3660                 trans->block_rsv = root->orphan_block_rsv;
3661
3662                 ret = btrfs_block_rsv_check(trans, root,
3663                                             root->orphan_block_rsv, 0, 5);
3664                 if (ret) {
3665                         BUG_ON(ret != -EAGAIN);
3666                         ret = btrfs_commit_transaction(trans, root);
3667                         BUG_ON(ret);
3668                         continue;
3669                 }
3670
3671                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3672                 if (ret != -EAGAIN)
3673                         break;
3674
3675                 nr = trans->blocks_used;
3676                 btrfs_end_transaction(trans, root);
3677                 trans = NULL;
3678                 btrfs_btree_balance_dirty(root, nr);
3679
3680         }
3681
3682         if (ret == 0) {
3683                 ret = btrfs_orphan_del(trans, inode);
3684                 BUG_ON(ret);
3685         }
3686
3687         if (!(root == root->fs_info->tree_root ||
3688               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3689                 btrfs_return_ino(root, btrfs_ino(inode));
3690
3691         nr = trans->blocks_used;
3692         btrfs_end_transaction(trans, root);
3693         btrfs_btree_balance_dirty(root, nr);
3694 no_delete:
3695         end_writeback(inode);
3696         return;
3697 }
3698
3699 /*
3700  * this returns the key found in the dir entry in the location pointer.
3701  * If no dir entries were found, location->objectid is 0.
3702  */
3703 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3704                                struct btrfs_key *location)
3705 {
3706         const char *name = dentry->d_name.name;
3707         int namelen = dentry->d_name.len;
3708         struct btrfs_dir_item *di;
3709         struct btrfs_path *path;
3710         struct btrfs_root *root = BTRFS_I(dir)->root;
3711         int ret = 0;
3712
3713         path = btrfs_alloc_path();
3714         BUG_ON(!path);
3715
3716         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3717                                     namelen, 0);
3718         if (IS_ERR(di))
3719                 ret = PTR_ERR(di);
3720
3721         if (IS_ERR_OR_NULL(di))
3722                 goto out_err;
3723
3724         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3725 out:
3726         btrfs_free_path(path);
3727         return ret;
3728 out_err:
3729         location->objectid = 0;
3730         goto out;
3731 }
3732
3733 /*
3734  * when we hit a tree root in a directory, the btrfs part of the inode
3735  * needs to be changed to reflect the root directory of the tree root.  This
3736  * is kind of like crossing a mount point.
3737  */
3738 static int fixup_tree_root_location(struct btrfs_root *root,
3739                                     struct inode *dir,
3740                                     struct dentry *dentry,
3741                                     struct btrfs_key *location,
3742                                     struct btrfs_root **sub_root)
3743 {
3744         struct btrfs_path *path;
3745         struct btrfs_root *new_root;
3746         struct btrfs_root_ref *ref;
3747         struct extent_buffer *leaf;
3748         int ret;
3749         int err = 0;
3750
3751         path = btrfs_alloc_path();
3752         if (!path) {
3753                 err = -ENOMEM;
3754                 goto out;
3755         }
3756
3757         err = -ENOENT;
3758         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3759                                   BTRFS_I(dir)->root->root_key.objectid,
3760                                   location->objectid);
3761         if (ret) {
3762                 if (ret < 0)
3763                         err = ret;
3764                 goto out;
3765         }
3766
3767         leaf = path->nodes[0];
3768         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3769         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3770             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3771                 goto out;
3772
3773         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3774                                    (unsigned long)(ref + 1),
3775                                    dentry->d_name.len);
3776         if (ret)
3777                 goto out;
3778
3779         btrfs_release_path(path);
3780
3781         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3782         if (IS_ERR(new_root)) {
3783                 err = PTR_ERR(new_root);
3784                 goto out;
3785         }
3786
3787         if (btrfs_root_refs(&new_root->root_item) == 0) {
3788                 err = -ENOENT;
3789                 goto out;
3790         }
3791
3792         *sub_root = new_root;
3793         location->objectid = btrfs_root_dirid(&new_root->root_item);
3794         location->type = BTRFS_INODE_ITEM_KEY;
3795         location->offset = 0;
3796         err = 0;
3797 out:
3798         btrfs_free_path(path);
3799         return err;
3800 }
3801
3802 static void inode_tree_add(struct inode *inode)
3803 {
3804         struct btrfs_root *root = BTRFS_I(inode)->root;
3805         struct btrfs_inode *entry;
3806         struct rb_node **p;
3807         struct rb_node *parent;
3808         u64 ino = btrfs_ino(inode);
3809 again:
3810         p = &root->inode_tree.rb_node;
3811         parent = NULL;
3812
3813         if (inode_unhashed(inode))
3814                 return;
3815
3816         spin_lock(&root->inode_lock);
3817         while (*p) {
3818                 parent = *p;
3819                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3820
3821                 if (ino < btrfs_ino(&entry->vfs_inode))
3822                         p = &parent->rb_left;
3823                 else if (ino > btrfs_ino(&entry->vfs_inode))
3824                         p = &parent->rb_right;
3825                 else {
3826                         WARN_ON(!(entry->vfs_inode.i_state &
3827                                   (I_WILL_FREE | I_FREEING)));
3828                         rb_erase(parent, &root->inode_tree);
3829                         RB_CLEAR_NODE(parent);
3830                         spin_unlock(&root->inode_lock);
3831                         goto again;
3832                 }
3833         }
3834         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3835         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3836         spin_unlock(&root->inode_lock);
3837 }
3838
3839 static void inode_tree_del(struct inode *inode)
3840 {
3841         struct btrfs_root *root = BTRFS_I(inode)->root;
3842         int empty = 0;
3843
3844         spin_lock(&root->inode_lock);
3845         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3846                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3847                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3848                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3849         }
3850         spin_unlock(&root->inode_lock);
3851
3852         /*
3853          * Free space cache has inodes in the tree root, but the tree root has a
3854          * root_refs of 0, so this could end up dropping the tree root as a
3855          * snapshot, so we need the extra !root->fs_info->tree_root check to
3856          * make sure we don't drop it.
3857          */
3858         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3859             root != root->fs_info->tree_root) {
3860                 synchronize_srcu(&root->fs_info->subvol_srcu);
3861                 spin_lock(&root->inode_lock);
3862                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3863                 spin_unlock(&root->inode_lock);
3864                 if (empty)
3865                         btrfs_add_dead_root(root);
3866         }
3867 }
3868
3869 int btrfs_invalidate_inodes(struct btrfs_root *root)
3870 {
3871         struct rb_node *node;
3872         struct rb_node *prev;
3873         struct btrfs_inode *entry;
3874         struct inode *inode;
3875         u64 objectid = 0;
3876
3877         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3878
3879         spin_lock(&root->inode_lock);
3880 again:
3881         node = root->inode_tree.rb_node;
3882         prev = NULL;
3883         while (node) {
3884                 prev = node;
3885                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3886
3887                 if (objectid < btrfs_ino(&entry->vfs_inode))
3888                         node = node->rb_left;
3889                 else if (objectid > btrfs_ino(&entry->vfs_inode))
3890                         node = node->rb_right;
3891                 else
3892                         break;
3893         }
3894         if (!node) {
3895                 while (prev) {
3896                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
3897                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
3898                                 node = prev;
3899                                 break;
3900                         }
3901                         prev = rb_next(prev);
3902                 }
3903         }
3904         while (node) {
3905                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3906                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
3907                 inode = igrab(&entry->vfs_inode);
3908                 if (inode) {
3909                         spin_unlock(&root->inode_lock);
3910                         if (atomic_read(&inode->i_count) > 1)
3911                                 d_prune_aliases(inode);
3912                         /*
3913                          * btrfs_drop_inode will have it removed from
3914                          * the inode cache when its usage count
3915                          * hits zero.
3916                          */
3917                         iput(inode);
3918                         cond_resched();
3919                         spin_lock(&root->inode_lock);
3920                         goto again;
3921                 }
3922
3923                 if (cond_resched_lock(&root->inode_lock))
3924                         goto again;
3925
3926                 node = rb_next(node);
3927         }
3928         spin_unlock(&root->inode_lock);
3929         return 0;
3930 }
3931
3932 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3933 {
3934         struct btrfs_iget_args *args = p;
3935         inode->i_ino = args->ino;
3936         BTRFS_I(inode)->root = args->root;
3937         btrfs_set_inode_space_info(args->root, inode);
3938         return 0;
3939 }
3940
3941 static int btrfs_find_actor(struct inode *inode, void *opaque)
3942 {
3943         struct btrfs_iget_args *args = opaque;
3944         return args->ino == btrfs_ino(inode) &&
3945                 args->root == BTRFS_I(inode)->root;
3946 }
3947
3948 static struct inode *btrfs_iget_locked(struct super_block *s,
3949                                        u64 objectid,
3950                                        struct btrfs_root *root)
3951 {
3952         struct inode *inode;
3953         struct btrfs_iget_args args;
3954         args.ino = objectid;
3955         args.root = root;
3956
3957         inode = iget5_locked(s, objectid, btrfs_find_actor,
3958                              btrfs_init_locked_inode,
3959                              (void *)&args);
3960         return inode;
3961 }
3962
3963 /* Get an inode object given its location and corresponding root.
3964  * Returns in *is_new if the inode was read from disk
3965  */
3966 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3967                          struct btrfs_root *root, int *new)
3968 {
3969         struct inode *inode;
3970
3971         inode = btrfs_iget_locked(s, location->objectid, root);
3972         if (!inode)
3973                 return ERR_PTR(-ENOMEM);
3974
3975         if (inode->i_state & I_NEW) {
3976                 BTRFS_I(inode)->root = root;
3977                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3978                 btrfs_read_locked_inode(inode);
3979                 inode_tree_add(inode);
3980                 unlock_new_inode(inode);
3981                 if (new)
3982                         *new = 1;
3983         }
3984
3985         return inode;
3986 }
3987
3988 static struct inode *new_simple_dir(struct super_block *s,
3989                                     struct btrfs_key *key,
3990                                     struct btrfs_root *root)
3991 {
3992         struct inode *inode = new_inode(s);
3993
3994         if (!inode)
3995                 return ERR_PTR(-ENOMEM);
3996
3997         BTRFS_I(inode)->root = root;
3998         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
3999         BTRFS_I(inode)->dummy_inode = 1;
4000
4001         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4002         inode->i_op = &simple_dir_inode_operations;
4003         inode->i_fop = &simple_dir_operations;
4004         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4005         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4006
4007         return inode;
4008 }
4009
4010 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4011 {
4012         struct inode *inode;
4013         struct btrfs_root *root = BTRFS_I(dir)->root;
4014         struct btrfs_root *sub_root = root;
4015         struct btrfs_key location;
4016         int index;
4017         int ret;
4018
4019         if (dentry->d_name.len > BTRFS_NAME_LEN)
4020                 return ERR_PTR(-ENAMETOOLONG);
4021
4022         ret = btrfs_inode_by_name(dir, dentry, &location);
4023
4024         if (ret < 0)
4025                 return ERR_PTR(ret);
4026
4027         if (location.objectid == 0)
4028                 return NULL;
4029
4030         if (location.type == BTRFS_INODE_ITEM_KEY) {
4031                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4032                 return inode;
4033         }
4034
4035         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4036
4037         index = srcu_read_lock(&root->fs_info->subvol_srcu);
4038         ret = fixup_tree_root_location(root, dir, dentry,
4039                                        &location, &sub_root);
4040         if (ret < 0) {
4041                 if (ret != -ENOENT)
4042                         inode = ERR_PTR(ret);
4043                 else
4044                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
4045         } else {
4046                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4047         }
4048         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4049
4050         if (!IS_ERR(inode) && root != sub_root) {
4051                 down_read(&root->fs_info->cleanup_work_sem);
4052                 if (!(inode->i_sb->s_flags & MS_RDONLY))
4053                         ret = btrfs_orphan_cleanup(sub_root);
4054                 up_read(&root->fs_info->cleanup_work_sem);
4055                 if (ret)
4056                         inode = ERR_PTR(ret);
4057         }
4058
4059         return inode;
4060 }
4061
4062 static int btrfs_dentry_delete(const struct dentry *dentry)
4063 {
4064         struct btrfs_root *root;
4065
4066         if (!dentry->d_inode && !IS_ROOT(dentry))
4067                 dentry = dentry->d_parent;
4068
4069         if (dentry->d_inode) {
4070                 root = BTRFS_I(dentry->d_inode)->root;
4071                 if (btrfs_root_refs(&root->root_item) == 0)
4072                         return 1;
4073         }
4074         return 0;
4075 }
4076
4077 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4078                                    struct nameidata *nd)
4079 {
4080         struct inode *inode;
4081
4082         inode = btrfs_lookup_dentry(dir, dentry);
4083         if (IS_ERR(inode))
4084                 return ERR_CAST(inode);
4085
4086         return d_splice_alias(inode, dentry);
4087 }
4088
4089 unsigned char btrfs_filetype_table[] = {
4090         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4091 };
4092
4093 static int btrfs_real_readdir(struct file *filp, void *dirent,
4094                               filldir_t filldir)
4095 {
4096         struct inode *inode = filp->f_dentry->d_inode;
4097         struct btrfs_root *root = BTRFS_I(inode)->root;
4098         struct btrfs_item *item;
4099         struct btrfs_dir_item *di;
4100         struct btrfs_key key;
4101         struct btrfs_key found_key;
4102         struct btrfs_path *path;
4103         struct list_head ins_list;
4104         struct list_head del_list;
4105         int ret;
4106         struct extent_buffer *leaf;
4107         int slot;
4108         unsigned char d_type;
4109         int over = 0;
4110         u32 di_cur;
4111         u32 di_total;
4112         u32 di_len;
4113         int key_type = BTRFS_DIR_INDEX_KEY;
4114         char tmp_name[32];
4115         char *name_ptr;
4116         int name_len;
4117         int is_curr = 0;        /* filp->f_pos points to the current index? */
4118
4119         /* FIXME, use a real flag for deciding about the key type */
4120         if (root->fs_info->tree_root == root)
4121                 key_type = BTRFS_DIR_ITEM_KEY;
4122
4123         /* special case for "." */
4124         if (filp->f_pos == 0) {
4125                 over = filldir(dirent, ".", 1, 1, btrfs_ino(inode), DT_DIR);
4126                 if (over)
4127                         return 0;
4128                 filp->f_pos = 1;
4129         }
4130         /* special case for .., just use the back ref */
4131         if (filp->f_pos == 1) {
4132                 u64 pino = parent_ino(filp->f_path.dentry);
4133                 over = filldir(dirent, "..", 2,
4134                                2, pino, DT_DIR);
4135                 if (over)
4136                         return 0;
4137                 filp->f_pos = 2;
4138         }
4139         path = btrfs_alloc_path();
4140         if (!path)
4141                 return -ENOMEM;
4142
4143         path->reada = 1;
4144
4145         if (key_type == BTRFS_DIR_INDEX_KEY) {
4146                 INIT_LIST_HEAD(&ins_list);
4147                 INIT_LIST_HEAD(&del_list);
4148                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4149         }
4150
4151         btrfs_set_key_type(&key, key_type);
4152         key.offset = filp->f_pos;
4153         key.objectid = btrfs_ino(inode);
4154
4155         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4156         if (ret < 0)
4157                 goto err;
4158
4159         while (1) {
4160                 leaf = path->nodes[0];
4161                 slot = path->slots[0];
4162                 if (slot >= btrfs_header_nritems(leaf)) {
4163                         ret = btrfs_next_leaf(root, path);
4164                         if (ret < 0)
4165                                 goto err;
4166                         else if (ret > 0)
4167                                 break;
4168                         continue;
4169                 }
4170
4171                 item = btrfs_item_nr(leaf, slot);
4172                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4173
4174                 if (found_key.objectid != key.objectid)
4175                         break;
4176                 if (btrfs_key_type(&found_key) != key_type)
4177                         break;
4178                 if (found_key.offset < filp->f_pos)
4179                         goto next;
4180                 if (key_type == BTRFS_DIR_INDEX_KEY &&
4181                     btrfs_should_delete_dir_index(&del_list,
4182                                                   found_key.offset))
4183                         goto next;
4184
4185                 filp->f_pos = found_key.offset;
4186                 is_curr = 1;
4187
4188                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4189                 di_cur = 0;
4190                 di_total = btrfs_item_size(leaf, item);
4191
4192                 while (di_cur < di_total) {
4193                         struct btrfs_key location;
4194
4195                         if (verify_dir_item(root, leaf, di))
4196                                 break;
4197
4198                         name_len = btrfs_dir_name_len(leaf, di);
4199                         if (name_len <= sizeof(tmp_name)) {
4200                                 name_ptr = tmp_name;
4201                         } else {
4202                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4203                                 if (!name_ptr) {
4204                                         ret = -ENOMEM;
4205                                         goto err;
4206                                 }
4207                         }
4208                         read_extent_buffer(leaf, name_ptr,
4209                                            (unsigned long)(di + 1), name_len);
4210
4211                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4212                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4213
4214                         /* is this a reference to our own snapshot? If so
4215                          * skip it
4216                          */
4217                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4218                             location.objectid == root->root_key.objectid) {
4219                                 over = 0;
4220                                 goto skip;
4221                         }
4222                         over = filldir(dirent, name_ptr, name_len,
4223                                        found_key.offset, location.objectid,
4224                                        d_type);
4225
4226 skip:
4227                         if (name_ptr != tmp_name)
4228                                 kfree(name_ptr);
4229
4230                         if (over)
4231                                 goto nopos;
4232                         di_len = btrfs_dir_name_len(leaf, di) +
4233                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4234                         di_cur += di_len;
4235                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4236                 }
4237 next:
4238                 path->slots[0]++;
4239         }
4240
4241         if (key_type == BTRFS_DIR_INDEX_KEY) {
4242                 if (is_curr)
4243                         filp->f_pos++;
4244                 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4245                                                       &ins_list);
4246                 if (ret)
4247                         goto nopos;
4248         }
4249
4250         /* Reached end of directory/root. Bump pos past the last item. */
4251         if (key_type == BTRFS_DIR_INDEX_KEY)
4252                 /*
4253                  * 32-bit glibc will use getdents64, but then strtol -
4254                  * so the last number we can serve is this.
4255                  */
4256                 filp->f_pos = 0x7fffffff;
4257         else
4258                 filp->f_pos++;
4259 nopos:
4260         ret = 0;
4261 err:
4262         if (key_type == BTRFS_DIR_INDEX_KEY)
4263                 btrfs_put_delayed_items(&ins_list, &del_list);
4264         btrfs_free_path(path);
4265         return ret;
4266 }
4267
4268 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4269 {
4270         struct btrfs_root *root = BTRFS_I(inode)->root;
4271         struct btrfs_trans_handle *trans;
4272         int ret = 0;
4273         bool nolock = false;
4274
4275         if (BTRFS_I(inode)->dummy_inode)
4276                 return 0;
4277
4278         if (btrfs_fs_closing(root->fs_info) && is_free_space_inode(root, inode))
4279                 nolock = true;
4280
4281         if (wbc->sync_mode == WB_SYNC_ALL) {
4282                 if (nolock)
4283                         trans = btrfs_join_transaction_nolock(root);
4284                 else
4285                         trans = btrfs_join_transaction(root);
4286                 if (IS_ERR(trans))
4287                         return PTR_ERR(trans);
4288                 if (nolock)
4289                         ret = btrfs_end_transaction_nolock(trans, root);
4290                 else
4291                         ret = btrfs_commit_transaction(trans, root);
4292         }
4293         return ret;
4294 }
4295
4296 /*
4297  * This is somewhat expensive, updating the tree every time the
4298  * inode changes.  But, it is most likely to find the inode in cache.
4299  * FIXME, needs more benchmarking...there are no reasons other than performance
4300  * to keep or drop this code.
4301  */
4302 void btrfs_dirty_inode(struct inode *inode, int flags)
4303 {
4304         struct btrfs_root *root = BTRFS_I(inode)->root;
4305         struct btrfs_trans_handle *trans;
4306         int ret;
4307
4308         if (BTRFS_I(inode)->dummy_inode)
4309                 return;
4310
4311         trans = btrfs_join_transaction(root);
4312         BUG_ON(IS_ERR(trans));
4313
4314         ret = btrfs_update_inode(trans, root, inode);
4315         if (ret && ret == -ENOSPC) {
4316                 /* whoops, lets try again with the full transaction */
4317                 btrfs_end_transaction(trans, root);
4318                 trans = btrfs_start_transaction(root, 1);
4319                 if (IS_ERR(trans)) {
4320                         printk_ratelimited(KERN_ERR "btrfs: fail to "
4321                                        "dirty  inode %llu error %ld\n",
4322                                        (unsigned long long)btrfs_ino(inode),
4323                                        PTR_ERR(trans));
4324                         return;
4325                 }
4326
4327                 ret = btrfs_update_inode(trans, root, inode);
4328                 if (ret) {
4329                         printk_ratelimited(KERN_ERR "btrfs: fail to "
4330                                        "dirty  inode %llu error %d\n",
4331                                        (unsigned long long)btrfs_ino(inode),
4332                                        ret);
4333                 }
4334         }
4335         btrfs_end_transaction(trans, root);
4336         if (BTRFS_I(inode)->delayed_node)
4337                 btrfs_balance_delayed_items(root);
4338 }
4339
4340 /*
4341  * find the highest existing sequence number in a directory
4342  * and then set the in-memory index_cnt variable to reflect
4343  * free sequence numbers
4344  */
4345 static int btrfs_set_inode_index_count(struct inode *inode)
4346 {
4347         struct btrfs_root *root = BTRFS_I(inode)->root;
4348         struct btrfs_key key, found_key;
4349         struct btrfs_path *path;
4350         struct extent_buffer *leaf;
4351         int ret;
4352
4353         key.objectid = btrfs_ino(inode);
4354         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4355         key.offset = (u64)-1;
4356
4357         path = btrfs_alloc_path();
4358         if (!path)
4359                 return -ENOMEM;
4360
4361         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4362         if (ret < 0)
4363                 goto out;
4364         /* FIXME: we should be able to handle this */
4365         if (ret == 0)
4366                 goto out;
4367         ret = 0;
4368
4369         /*
4370          * MAGIC NUMBER EXPLANATION:
4371          * since we search a directory based on f_pos we have to start at 2
4372          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4373          * else has to start at 2
4374          */
4375         if (path->slots[0] == 0) {
4376                 BTRFS_I(inode)->index_cnt = 2;
4377                 goto out;
4378         }
4379
4380         path->slots[0]--;
4381
4382         leaf = path->nodes[0];
4383         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4384
4385         if (found_key.objectid != btrfs_ino(inode) ||
4386             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4387                 BTRFS_I(inode)->index_cnt = 2;
4388                 goto out;
4389         }
4390
4391         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4392 out:
4393         btrfs_free_path(path);
4394         return ret;
4395 }
4396
4397 /*
4398  * helper to find a free sequence number in a given directory.  This current
4399  * code is very simple, later versions will do smarter things in the btree
4400  */
4401 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4402 {
4403         int ret = 0;
4404
4405         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4406                 ret = btrfs_inode_delayed_dir_index_count(dir);
4407                 if (ret) {
4408                         ret = btrfs_set_inode_index_count(dir);
4409                         if (ret)
4410                                 return ret;
4411                 }
4412         }
4413
4414         *index = BTRFS_I(dir)->index_cnt;
4415         BTRFS_I(dir)->index_cnt++;
4416
4417         return ret;
4418 }
4419
4420 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4421                                      struct btrfs_root *root,
4422                                      struct inode *dir,
4423                                      const char *name, int name_len,
4424                                      u64 ref_objectid, u64 objectid, int mode,
4425                                      u64 *index)
4426 {
4427         struct inode *inode;
4428         struct btrfs_inode_item *inode_item;
4429         struct btrfs_key *location;
4430         struct btrfs_path *path;
4431         struct btrfs_inode_ref *ref;
4432         struct btrfs_key key[2];
4433         u32 sizes[2];
4434         unsigned long ptr;
4435         int ret;
4436         int owner;
4437
4438         path = btrfs_alloc_path();
4439         BUG_ON(!path);
4440
4441         inode = new_inode(root->fs_info->sb);
4442         if (!inode) {
4443                 btrfs_free_path(path);
4444                 return ERR_PTR(-ENOMEM);
4445         }
4446
4447         /*
4448          * we have to initialize this early, so we can reclaim the inode
4449          * number if we fail afterwards in this function.
4450          */
4451         inode->i_ino = objectid;
4452
4453         if (dir) {
4454                 trace_btrfs_inode_request(dir);
4455
4456                 ret = btrfs_set_inode_index(dir, index);
4457                 if (ret) {
4458                         btrfs_free_path(path);
4459                         iput(inode);
4460                         return ERR_PTR(ret);
4461                 }
4462         }
4463         /*
4464          * index_cnt is ignored for everything but a dir,
4465          * btrfs_get_inode_index_count has an explanation for the magic
4466          * number
4467          */
4468         BTRFS_I(inode)->index_cnt = 2;
4469         BTRFS_I(inode)->root = root;
4470         BTRFS_I(inode)->generation = trans->transid;
4471         inode->i_generation = BTRFS_I(inode)->generation;
4472         btrfs_set_inode_space_info(root, inode);
4473
4474         if (mode & S_IFDIR)
4475                 owner = 0;
4476         else
4477                 owner = 1;
4478
4479         key[0].objectid = objectid;
4480         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4481         key[0].offset = 0;
4482
4483         key[1].objectid = objectid;
4484         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4485         key[1].offset = ref_objectid;
4486
4487         sizes[0] = sizeof(struct btrfs_inode_item);
4488         sizes[1] = name_len + sizeof(*ref);
4489
4490         path->leave_spinning = 1;
4491         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4492         if (ret != 0)
4493                 goto fail;
4494
4495         inode_init_owner(inode, dir, mode);
4496         inode_set_bytes(inode, 0);
4497         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4498         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4499                                   struct btrfs_inode_item);
4500         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4501
4502         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4503                              struct btrfs_inode_ref);
4504         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4505         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4506         ptr = (unsigned long)(ref + 1);
4507         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4508
4509         btrfs_mark_buffer_dirty(path->nodes[0]);
4510         btrfs_free_path(path);
4511
4512         location = &BTRFS_I(inode)->location;
4513         location->objectid = objectid;
4514         location->offset = 0;
4515         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4516
4517         btrfs_inherit_iflags(inode, dir);
4518
4519         if ((mode & S_IFREG)) {
4520                 if (btrfs_test_opt(root, NODATASUM))
4521                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4522                 if (btrfs_test_opt(root, NODATACOW) ||
4523                     (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4524                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4525         }
4526
4527         insert_inode_hash(inode);
4528         inode_tree_add(inode);
4529
4530         trace_btrfs_inode_new(inode);
4531         btrfs_set_inode_last_trans(trans, inode);
4532
4533         return inode;
4534 fail:
4535         if (dir)
4536                 BTRFS_I(dir)->index_cnt--;
4537         btrfs_free_path(path);
4538         iput(inode);
4539         return ERR_PTR(ret);
4540 }
4541
4542 static inline u8 btrfs_inode_type(struct inode *inode)
4543 {
4544         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4545 }
4546
4547 /*
4548  * utility function to add 'inode' into 'parent_inode' with
4549  * a give name and a given sequence number.
4550  * if 'add_backref' is true, also insert a backref from the
4551  * inode to the parent directory.
4552  */
4553 int btrfs_add_link(struct btrfs_trans_handle *trans,
4554                    struct inode *parent_inode, struct inode *inode,
4555                    const char *name, int name_len, int add_backref, u64 index)
4556 {
4557         int ret = 0;
4558         struct btrfs_key key;
4559         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4560         u64 ino = btrfs_ino(inode);
4561         u64 parent_ino = btrfs_ino(parent_inode);
4562
4563         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4564                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4565         } else {
4566                 key.objectid = ino;
4567                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4568                 key.offset = 0;
4569         }
4570
4571         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4572                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4573                                          key.objectid, root->root_key.objectid,
4574                                          parent_ino, index, name, name_len);
4575         } else if (add_backref) {
4576                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4577                                              parent_ino, index);
4578         }
4579
4580         if (ret == 0) {
4581                 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4582                                             parent_inode, &key,
4583                                             btrfs_inode_type(inode), index);
4584                 BUG_ON(ret);
4585
4586                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4587                                    name_len * 2);
4588                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4589                 ret = btrfs_update_inode(trans, root, parent_inode);
4590         }
4591         return ret;
4592 }
4593
4594 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4595                             struct inode *dir, struct dentry *dentry,
4596                             struct inode *inode, int backref, u64 index)
4597 {
4598         int err = btrfs_add_link(trans, dir, inode,
4599                                  dentry->d_name.name, dentry->d_name.len,
4600                                  backref, index);
4601         if (!err) {
4602                 d_instantiate(dentry, inode);
4603                 return 0;
4604         }
4605         if (err > 0)
4606                 err = -EEXIST;
4607         return err;
4608 }
4609
4610 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4611                         int mode, dev_t rdev)
4612 {
4613         struct btrfs_trans_handle *trans;
4614         struct btrfs_root *root = BTRFS_I(dir)->root;
4615         struct inode *inode = NULL;
4616         int err;
4617         int drop_inode = 0;
4618         u64 objectid;
4619         unsigned long nr = 0;
4620         u64 index = 0;
4621
4622         if (!new_valid_dev(rdev))
4623                 return -EINVAL;
4624
4625         /*
4626          * 2 for inode item and ref
4627          * 2 for dir items
4628          * 1 for xattr if selinux is on
4629          */
4630         trans = btrfs_start_transaction(root, 5);
4631         if (IS_ERR(trans))
4632                 return PTR_ERR(trans);
4633
4634         err = btrfs_find_free_ino(root, &objectid);
4635         if (err)
4636                 goto out_unlock;
4637
4638         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4639                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4640                                 mode, &index);
4641         if (IS_ERR(inode)) {
4642                 err = PTR_ERR(inode);
4643                 goto out_unlock;
4644         }
4645
4646         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4647         if (err) {
4648                 drop_inode = 1;
4649                 goto out_unlock;
4650         }
4651
4652         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4653         if (err)
4654                 drop_inode = 1;
4655         else {
4656                 inode->i_op = &btrfs_special_inode_operations;
4657                 init_special_inode(inode, inode->i_mode, rdev);
4658                 btrfs_update_inode(trans, root, inode);
4659         }
4660 out_unlock:
4661         nr = trans->blocks_used;
4662         btrfs_end_transaction_throttle(trans, root);
4663         btrfs_btree_balance_dirty(root, nr);
4664         if (drop_inode) {
4665                 inode_dec_link_count(inode);
4666                 iput(inode);
4667         }
4668         return err;
4669 }
4670
4671 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4672                         int mode, struct nameidata *nd)
4673 {
4674         struct btrfs_trans_handle *trans;
4675         struct btrfs_root *root = BTRFS_I(dir)->root;
4676         struct inode *inode = NULL;
4677         int drop_inode = 0;
4678         int err;
4679         unsigned long nr = 0;
4680         u64 objectid;
4681         u64 index = 0;
4682
4683         /*
4684          * 2 for inode item and ref
4685          * 2 for dir items
4686          * 1 for xattr if selinux is on
4687          */
4688         trans = btrfs_start_transaction(root, 5);
4689         if (IS_ERR(trans))
4690                 return PTR_ERR(trans);
4691
4692         err = btrfs_find_free_ino(root, &objectid);
4693         if (err)
4694                 goto out_unlock;
4695
4696         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4697                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4698                                 mode, &index);
4699         if (IS_ERR(inode)) {
4700                 err = PTR_ERR(inode);
4701                 goto out_unlock;
4702         }
4703
4704         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4705         if (err) {
4706                 drop_inode = 1;
4707                 goto out_unlock;
4708         }
4709
4710         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4711         if (err)
4712                 drop_inode = 1;
4713         else {
4714                 inode->i_mapping->a_ops = &btrfs_aops;
4715                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4716                 inode->i_fop = &btrfs_file_operations;
4717                 inode->i_op = &btrfs_file_inode_operations;
4718                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4719         }
4720 out_unlock:
4721         nr = trans->blocks_used;
4722         btrfs_end_transaction_throttle(trans, root);
4723         if (drop_inode) {
4724                 inode_dec_link_count(inode);
4725                 iput(inode);
4726         }
4727         btrfs_btree_balance_dirty(root, nr);
4728         return err;
4729 }
4730
4731 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4732                       struct dentry *dentry)
4733 {
4734         struct btrfs_trans_handle *trans;
4735         struct btrfs_root *root = BTRFS_I(dir)->root;
4736         struct inode *inode = old_dentry->d_inode;
4737         u64 index;
4738         unsigned long nr = 0;
4739         int err;
4740         int drop_inode = 0;
4741
4742         /* do not allow sys_link's with other subvols of the same device */
4743         if (root->objectid != BTRFS_I(inode)->root->objectid)
4744                 return -EXDEV;
4745
4746         if (inode->i_nlink == ~0U)
4747                 return -EMLINK;
4748
4749         err = btrfs_set_inode_index(dir, &index);
4750         if (err)
4751                 goto fail;
4752
4753         /*
4754          * 2 items for inode and inode ref
4755          * 2 items for dir items
4756          * 1 item for parent inode
4757          */
4758         trans = btrfs_start_transaction(root, 5);
4759         if (IS_ERR(trans)) {
4760                 err = PTR_ERR(trans);
4761                 goto fail;
4762         }
4763
4764         btrfs_inc_nlink(inode);
4765         inode->i_ctime = CURRENT_TIME;
4766         ihold(inode);
4767
4768         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4769
4770         if (err) {
4771                 drop_inode = 1;
4772         } else {
4773                 struct dentry *parent = dget_parent(dentry);
4774                 err = btrfs_update_inode(trans, root, inode);
4775                 BUG_ON(err);
4776                 btrfs_log_new_name(trans, inode, NULL, parent);
4777                 dput(parent);
4778         }
4779
4780         nr = trans->blocks_used;
4781         btrfs_end_transaction_throttle(trans, root);
4782 fail:
4783         if (drop_inode) {
4784                 inode_dec_link_count(inode);
4785                 iput(inode);
4786         }
4787         btrfs_btree_balance_dirty(root, nr);
4788         return err;
4789 }
4790
4791 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4792 {
4793         struct inode *inode = NULL;
4794         struct btrfs_trans_handle *trans;
4795         struct btrfs_root *root = BTRFS_I(dir)->root;
4796         int err = 0;
4797         int drop_on_err = 0;
4798         u64 objectid = 0;
4799         u64 index = 0;
4800         unsigned long nr = 1;
4801
4802         /*
4803          * 2 items for inode and ref
4804          * 2 items for dir items
4805          * 1 for xattr if selinux is on
4806          */
4807         trans = btrfs_start_transaction(root, 5);
4808         if (IS_ERR(trans))
4809                 return PTR_ERR(trans);
4810
4811         err = btrfs_find_free_ino(root, &objectid);
4812         if (err)
4813                 goto out_fail;
4814
4815         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4816                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4817                                 S_IFDIR | mode, &index);
4818         if (IS_ERR(inode)) {
4819                 err = PTR_ERR(inode);
4820                 goto out_fail;
4821         }
4822
4823         drop_on_err = 1;
4824
4825         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4826         if (err)
4827                 goto out_fail;
4828
4829         inode->i_op = &btrfs_dir_inode_operations;
4830         inode->i_fop = &btrfs_dir_file_operations;
4831
4832         btrfs_i_size_write(inode, 0);
4833         err = btrfs_update_inode(trans, root, inode);
4834         if (err)
4835                 goto out_fail;
4836
4837         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4838                              dentry->d_name.len, 0, index);
4839         if (err)
4840                 goto out_fail;
4841
4842         d_instantiate(dentry, inode);
4843         drop_on_err = 0;
4844
4845 out_fail:
4846         nr = trans->blocks_used;
4847         btrfs_end_transaction_throttle(trans, root);
4848         if (drop_on_err)
4849                 iput(inode);
4850         btrfs_btree_balance_dirty(root, nr);
4851         return err;
4852 }
4853
4854 /* helper for btfs_get_extent.  Given an existing extent in the tree,
4855  * and an extent that you want to insert, deal with overlap and insert
4856  * the new extent into the tree.
4857  */
4858 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4859                                 struct extent_map *existing,
4860                                 struct extent_map *em,
4861                                 u64 map_start, u64 map_len)
4862 {
4863         u64 start_diff;
4864
4865         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4866         start_diff = map_start - em->start;
4867         em->start = map_start;
4868         em->len = map_len;
4869         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4870             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4871                 em->block_start += start_diff;
4872                 em->block_len -= start_diff;
4873         }
4874         return add_extent_mapping(em_tree, em);
4875 }
4876
4877 static noinline int uncompress_inline(struct btrfs_path *path,
4878                                       struct inode *inode, struct page *page,
4879                                       size_t pg_offset, u64 extent_offset,
4880                                       struct btrfs_file_extent_item *item)
4881 {
4882         int ret;
4883         struct extent_buffer *leaf = path->nodes[0];
4884         char *tmp;
4885         size_t max_size;
4886         unsigned long inline_size;
4887         unsigned long ptr;
4888         int compress_type;
4889
4890         WARN_ON(pg_offset != 0);
4891         compress_type = btrfs_file_extent_compression(leaf, item);
4892         max_size = btrfs_file_extent_ram_bytes(leaf, item);
4893         inline_size = btrfs_file_extent_inline_item_len(leaf,
4894                                         btrfs_item_nr(leaf, path->slots[0]));
4895         tmp = kmalloc(inline_size, GFP_NOFS);
4896         if (!tmp)
4897                 return -ENOMEM;
4898         ptr = btrfs_file_extent_inline_start(item);
4899
4900         read_extent_buffer(leaf, tmp, ptr, inline_size);
4901
4902         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4903         ret = btrfs_decompress(compress_type, tmp, page,
4904                                extent_offset, inline_size, max_size);
4905         if (ret) {
4906                 char *kaddr = kmap_atomic(page, KM_USER0);
4907                 unsigned long copy_size = min_t(u64,
4908                                   PAGE_CACHE_SIZE - pg_offset,
4909                                   max_size - extent_offset);
4910                 memset(kaddr + pg_offset, 0, copy_size);
4911                 kunmap_atomic(kaddr, KM_USER0);
4912         }
4913         kfree(tmp);
4914         return 0;
4915 }
4916
4917 /*
4918  * a bit scary, this does extent mapping from logical file offset to the disk.
4919  * the ugly parts come from merging extents from the disk with the in-ram
4920  * representation.  This gets more complex because of the data=ordered code,
4921  * where the in-ram extents might be locked pending data=ordered completion.
4922  *
4923  * This also copies inline extents directly into the page.
4924  */
4925
4926 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4927                                     size_t pg_offset, u64 start, u64 len,
4928                                     int create)
4929 {
4930         int ret;
4931         int err = 0;
4932         u64 bytenr;
4933         u64 extent_start = 0;
4934         u64 extent_end = 0;
4935         u64 objectid = btrfs_ino(inode);
4936         u32 found_type;
4937         struct btrfs_path *path = NULL;
4938         struct btrfs_root *root = BTRFS_I(inode)->root;
4939         struct btrfs_file_extent_item *item;
4940         struct extent_buffer *leaf;
4941         struct btrfs_key found_key;
4942         struct extent_map *em = NULL;
4943         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4944         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4945         struct btrfs_trans_handle *trans = NULL;
4946         int compress_type;
4947
4948 again:
4949         read_lock(&em_tree->lock);
4950         em = lookup_extent_mapping(em_tree, start, len);
4951         if (em)
4952                 em->bdev = root->fs_info->fs_devices->latest_bdev;
4953         read_unlock(&em_tree->lock);
4954
4955         if (em) {
4956                 if (em->start > start || em->start + em->len <= start)
4957                         free_extent_map(em);
4958                 else if (em->block_start == EXTENT_MAP_INLINE && page)
4959                         free_extent_map(em);
4960                 else
4961                         goto out;
4962         }
4963         em = alloc_extent_map();
4964         if (!em) {
4965                 err = -ENOMEM;
4966                 goto out;
4967         }
4968         em->bdev = root->fs_info->fs_devices->latest_bdev;
4969         em->start = EXTENT_MAP_HOLE;
4970         em->orig_start = EXTENT_MAP_HOLE;
4971         em->len = (u64)-1;
4972         em->block_len = (u64)-1;
4973
4974         if (!path) {
4975                 path = btrfs_alloc_path();
4976                 if (!path) {
4977                         err = -ENOMEM;
4978                         goto out;
4979                 }
4980                 /*
4981                  * Chances are we'll be called again, so go ahead and do
4982                  * readahead
4983                  */
4984                 path->reada = 1;
4985         }
4986
4987         ret = btrfs_lookup_file_extent(trans, root, path,
4988                                        objectid, start, trans != NULL);
4989         if (ret < 0) {
4990                 err = ret;
4991                 goto out;
4992         }
4993
4994         if (ret != 0) {
4995                 if (path->slots[0] == 0)
4996                         goto not_found;
4997                 path->slots[0]--;
4998         }
4999
5000         leaf = path->nodes[0];
5001         item = btrfs_item_ptr(leaf, path->slots[0],
5002                               struct btrfs_file_extent_item);
5003         /* are we inside the extent that was found? */
5004         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5005         found_type = btrfs_key_type(&found_key);
5006         if (found_key.objectid != objectid ||
5007             found_type != BTRFS_EXTENT_DATA_KEY) {
5008                 goto not_found;
5009         }
5010
5011         found_type = btrfs_file_extent_type(leaf, item);
5012         extent_start = found_key.offset;
5013         compress_type = btrfs_file_extent_compression(leaf, item);
5014         if (found_type == BTRFS_FILE_EXTENT_REG ||
5015             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5016                 extent_end = extent_start +
5017                        btrfs_file_extent_num_bytes(leaf, item);
5018         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5019                 size_t size;
5020                 size = btrfs_file_extent_inline_len(leaf, item);
5021                 extent_end = (extent_start + size + root->sectorsize - 1) &
5022                         ~((u64)root->sectorsize - 1);
5023         }
5024
5025         if (start >= extent_end) {
5026                 path->slots[0]++;
5027                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5028                         ret = btrfs_next_leaf(root, path);
5029                         if (ret < 0) {
5030                                 err = ret;
5031                                 goto out;
5032                         }
5033                         if (ret > 0)
5034                                 goto not_found;
5035                         leaf = path->nodes[0];
5036                 }
5037                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5038                 if (found_key.objectid != objectid ||
5039                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5040                         goto not_found;
5041                 if (start + len <= found_key.offset)
5042                         goto not_found;
5043                 em->start = start;
5044                 em->len = found_key.offset - start;
5045                 goto not_found_em;
5046         }
5047
5048         if (found_type == BTRFS_FILE_EXTENT_REG ||
5049             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5050                 em->start = extent_start;
5051                 em->len = extent_end - extent_start;
5052                 em->orig_start = extent_start -
5053                                  btrfs_file_extent_offset(leaf, item);
5054                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5055                 if (bytenr == 0) {
5056                         em->block_start = EXTENT_MAP_HOLE;
5057                         goto insert;
5058                 }
5059                 if (compress_type != BTRFS_COMPRESS_NONE) {
5060                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5061                         em->compress_type = compress_type;
5062                         em->block_start = bytenr;
5063                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5064                                                                          item);
5065                 } else {
5066                         bytenr += btrfs_file_extent_offset(leaf, item);
5067                         em->block_start = bytenr;
5068                         em->block_len = em->len;
5069                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5070                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5071                 }
5072                 goto insert;
5073         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5074                 unsigned long ptr;
5075                 char *map;
5076                 size_t size;
5077                 size_t extent_offset;
5078                 size_t copy_size;
5079
5080                 em->block_start = EXTENT_MAP_INLINE;
5081                 if (!page || create) {
5082                         em->start = extent_start;
5083                         em->len = extent_end - extent_start;
5084                         goto out;
5085                 }
5086
5087                 size = btrfs_file_extent_inline_len(leaf, item);
5088                 extent_offset = page_offset(page) + pg_offset - extent_start;
5089                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5090                                 size - extent_offset);
5091                 em->start = extent_start + extent_offset;
5092                 em->len = (copy_size + root->sectorsize - 1) &
5093                         ~((u64)root->sectorsize - 1);
5094                 em->orig_start = EXTENT_MAP_INLINE;
5095                 if (compress_type) {
5096                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5097                         em->compress_type = compress_type;
5098                 }
5099                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5100                 if (create == 0 && !PageUptodate(page)) {
5101                         if (btrfs_file_extent_compression(leaf, item) !=
5102                             BTRFS_COMPRESS_NONE) {
5103                                 ret = uncompress_inline(path, inode, page,
5104                                                         pg_offset,
5105                                                         extent_offset, item);
5106                                 BUG_ON(ret);
5107                         } else {
5108                                 map = kmap(page);
5109                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5110                                                    copy_size);
5111                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5112                                         memset(map + pg_offset + copy_size, 0,
5113                                                PAGE_CACHE_SIZE - pg_offset -
5114                                                copy_size);
5115                                 }
5116                                 kunmap(page);
5117                         }
5118                         flush_dcache_page(page);
5119                 } else if (create && PageUptodate(page)) {
5120                         WARN_ON(1);
5121                         if (!trans) {
5122                                 kunmap(page);
5123                                 free_extent_map(em);
5124                                 em = NULL;
5125
5126                                 btrfs_release_path(path);
5127                                 trans = btrfs_join_transaction(root);
5128
5129                                 if (IS_ERR(trans))
5130                                         return ERR_CAST(trans);
5131                                 goto again;
5132                         }
5133                         map = kmap(page);
5134                         write_extent_buffer(leaf, map + pg_offset, ptr,
5135                                             copy_size);
5136                         kunmap(page);
5137                         btrfs_mark_buffer_dirty(leaf);
5138                 }
5139                 set_extent_uptodate(io_tree, em->start,
5140                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
5141                 goto insert;
5142         } else {
5143                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5144                 WARN_ON(1);
5145         }
5146 not_found:
5147         em->start = start;
5148         em->len = len;
5149 not_found_em:
5150         em->block_start = EXTENT_MAP_HOLE;
5151         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5152 insert:
5153         btrfs_release_path(path);
5154         if (em->start > start || extent_map_end(em) <= start) {
5155                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5156                        "[%llu %llu]\n", (unsigned long long)em->start,
5157                        (unsigned long long)em->len,
5158                        (unsigned long long)start,
5159                        (unsigned long long)len);
5160                 err = -EIO;
5161                 goto out;
5162         }
5163
5164         err = 0;
5165         write_lock(&em_tree->lock);
5166         ret = add_extent_mapping(em_tree, em);
5167         /* it is possible that someone inserted the extent into the tree
5168          * while we had the lock dropped.  It is also possible that
5169          * an overlapping map exists in the tree
5170          */
5171         if (ret == -EEXIST) {
5172                 struct extent_map *existing;
5173
5174                 ret = 0;
5175
5176                 existing = lookup_extent_mapping(em_tree, start, len);
5177                 if (existing && (existing->start > start ||
5178                     existing->start + existing->len <= start)) {
5179                         free_extent_map(existing);
5180                         existing = NULL;
5181                 }
5182                 if (!existing) {
5183                         existing = lookup_extent_mapping(em_tree, em->start,
5184                                                          em->len);
5185                         if (existing) {
5186                                 err = merge_extent_mapping(em_tree, existing,
5187                                                            em, start,
5188                                                            root->sectorsize);
5189                                 free_extent_map(existing);
5190                                 if (err) {
5191                                         free_extent_map(em);
5192                                         em = NULL;
5193                                 }
5194                         } else {
5195                                 err = -EIO;
5196                                 free_extent_map(em);
5197                                 em = NULL;
5198                         }
5199                 } else {
5200                         free_extent_map(em);
5201                         em = existing;
5202                         err = 0;
5203                 }
5204         }
5205         write_unlock(&em_tree->lock);
5206 out:
5207
5208         trace_btrfs_get_extent(root, em);
5209
5210         if (path)
5211                 btrfs_free_path(path);
5212         if (trans) {
5213                 ret = btrfs_end_transaction(trans, root);
5214                 if (!err)
5215                         err = ret;
5216         }
5217         if (err) {
5218                 free_extent_map(em);
5219                 return ERR_PTR(err);
5220         }
5221         return em;
5222 }
5223
5224 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5225                                            size_t pg_offset, u64 start, u64 len,
5226                                            int create)
5227 {
5228         struct extent_map *em;
5229         struct extent_map *hole_em = NULL;
5230         u64 range_start = start;
5231         u64 end;
5232         u64 found;
5233         u64 found_end;
5234         int err = 0;
5235
5236         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5237         if (IS_ERR(em))
5238                 return em;
5239         if (em) {
5240                 /*
5241                  * if our em maps to a hole, there might
5242                  * actually be delalloc bytes behind it
5243                  */
5244                 if (em->block_start != EXTENT_MAP_HOLE)
5245                         return em;
5246                 else
5247                         hole_em = em;
5248         }
5249
5250         /* check to see if we've wrapped (len == -1 or similar) */
5251         end = start + len;
5252         if (end < start)
5253                 end = (u64)-1;
5254         else
5255                 end -= 1;
5256
5257         em = NULL;
5258
5259         /* ok, we didn't find anything, lets look for delalloc */
5260         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5261                                  end, len, EXTENT_DELALLOC, 1);
5262         found_end = range_start + found;
5263         if (found_end < range_start)
5264                 found_end = (u64)-1;
5265
5266         /*
5267          * we didn't find anything useful, return
5268          * the original results from get_extent()
5269          */
5270         if (range_start > end || found_end <= start) {
5271                 em = hole_em;
5272                 hole_em = NULL;
5273                 goto out;
5274         }
5275
5276         /* adjust the range_start to make sure it doesn't
5277          * go backwards from the start they passed in
5278          */
5279         range_start = max(start,range_start);
5280         found = found_end - range_start;
5281
5282         if (found > 0) {
5283                 u64 hole_start = start;
5284                 u64 hole_len = len;
5285
5286                 em = alloc_extent_map();
5287                 if (!em) {
5288                         err = -ENOMEM;
5289                         goto out;
5290                 }
5291                 /*
5292                  * when btrfs_get_extent can't find anything it
5293                  * returns one huge hole
5294                  *
5295                  * make sure what it found really fits our range, and
5296                  * adjust to make sure it is based on the start from
5297                  * the caller
5298                  */
5299                 if (hole_em) {
5300                         u64 calc_end = extent_map_end(hole_em);
5301
5302                         if (calc_end <= start || (hole_em->start > end)) {
5303                                 free_extent_map(hole_em);
5304                                 hole_em = NULL;
5305                         } else {
5306                                 hole_start = max(hole_em->start, start);
5307                                 hole_len = calc_end - hole_start;
5308                         }
5309                 }
5310                 em->bdev = NULL;
5311                 if (hole_em && range_start > hole_start) {
5312                         /* our hole starts before our delalloc, so we
5313                          * have to return just the parts of the hole
5314                          * that go until  the delalloc starts
5315                          */
5316                         em->len = min(hole_len,
5317                                       range_start - hole_start);
5318                         em->start = hole_start;
5319                         em->orig_start = hole_start;
5320                         /*
5321                          * don't adjust block start at all,
5322                          * it is fixed at EXTENT_MAP_HOLE
5323                          */
5324                         em->block_start = hole_em->block_start;
5325                         em->block_len = hole_len;
5326                 } else {
5327                         em->start = range_start;
5328                         em->len = found;
5329                         em->orig_start = range_start;
5330                         em->block_start = EXTENT_MAP_DELALLOC;
5331                         em->block_len = found;
5332                 }
5333         } else if (hole_em) {
5334                 return hole_em;
5335         }
5336 out:
5337
5338         free_extent_map(hole_em);
5339         if (err) {
5340                 free_extent_map(em);
5341                 return ERR_PTR(err);
5342         }
5343         return em;
5344 }
5345
5346 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5347                                                   struct extent_map *em,
5348                                                   u64 start, u64 len)
5349 {
5350         struct btrfs_root *root = BTRFS_I(inode)->root;
5351         struct btrfs_trans_handle *trans;
5352         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5353         struct btrfs_key ins;
5354         u64 alloc_hint;
5355         int ret;
5356         bool insert = false;
5357
5358         /*
5359          * Ok if the extent map we looked up is a hole and is for the exact
5360          * range we want, there is no reason to allocate a new one, however if
5361          * it is not right then we need to free this one and drop the cache for
5362          * our range.
5363          */
5364         if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5365             em->len != len) {
5366                 free_extent_map(em);
5367                 em = NULL;
5368                 insert = true;
5369                 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5370         }
5371
5372         trans = btrfs_join_transaction(root);
5373         if (IS_ERR(trans))
5374                 return ERR_CAST(trans);
5375
5376         if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5377                 btrfs_add_inode_defrag(trans, inode);
5378
5379         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5380
5381         alloc_hint = get_extent_allocation_hint(inode, start, len);
5382         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5383                                    alloc_hint, (u64)-1, &ins, 1);
5384         if (ret) {
5385                 em = ERR_PTR(ret);
5386                 goto out;
5387         }
5388
5389         if (!em) {
5390                 em = alloc_extent_map();
5391                 if (!em) {
5392                         em = ERR_PTR(-ENOMEM);
5393                         goto out;
5394                 }
5395         }
5396
5397         em->start = start;
5398         em->orig_start = em->start;
5399         em->len = ins.offset;
5400
5401         em->block_start = ins.objectid;
5402         em->block_len = ins.offset;
5403         em->bdev = root->fs_info->fs_devices->latest_bdev;
5404
5405         /*
5406          * We need to do this because if we're using the original em we searched
5407          * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5408          */
5409         em->flags = 0;
5410         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5411
5412         while (insert) {
5413                 write_lock(&em_tree->lock);
5414                 ret = add_extent_mapping(em_tree, em);
5415                 write_unlock(&em_tree->lock);
5416                 if (ret != -EEXIST)
5417                         break;
5418                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5419         }
5420
5421         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5422                                            ins.offset, ins.offset, 0);
5423         if (ret) {
5424                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5425                 em = ERR_PTR(ret);
5426         }
5427 out:
5428         btrfs_end_transaction(trans, root);
5429         return em;
5430 }
5431
5432 /*
5433  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5434  * block must be cow'd
5435  */
5436 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5437                                       struct inode *inode, u64 offset, u64 len)
5438 {
5439         struct btrfs_path *path;
5440         int ret;
5441         struct extent_buffer *leaf;
5442         struct btrfs_root *root = BTRFS_I(inode)->root;
5443         struct btrfs_file_extent_item *fi;
5444         struct btrfs_key key;
5445         u64 disk_bytenr;
5446         u64 backref_offset;
5447         u64 extent_end;
5448         u64 num_bytes;
5449         int slot;
5450         int found_type;
5451
5452         path = btrfs_alloc_path();
5453         if (!path)
5454                 return -ENOMEM;
5455
5456         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5457                                        offset, 0);
5458         if (ret < 0)
5459                 goto out;
5460
5461         slot = path->slots[0];
5462         if (ret == 1) {
5463                 if (slot == 0) {
5464                         /* can't find the item, must cow */
5465                         ret = 0;
5466                         goto out;
5467                 }
5468                 slot--;
5469         }
5470         ret = 0;
5471         leaf = path->nodes[0];
5472         btrfs_item_key_to_cpu(leaf, &key, slot);
5473         if (key.objectid != btrfs_ino(inode) ||
5474             key.type != BTRFS_EXTENT_DATA_KEY) {
5475                 /* not our file or wrong item type, must cow */
5476                 goto out;
5477         }
5478
5479         if (key.offset > offset) {
5480                 /* Wrong offset, must cow */
5481                 goto out;
5482         }
5483
5484         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5485         found_type = btrfs_file_extent_type(leaf, fi);
5486         if (found_type != BTRFS_FILE_EXTENT_REG &&
5487             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5488                 /* not a regular extent, must cow */
5489                 goto out;
5490         }
5491         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5492         backref_offset = btrfs_file_extent_offset(leaf, fi);
5493
5494         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5495         if (extent_end < offset + len) {
5496                 /* extent doesn't include our full range, must cow */
5497                 goto out;
5498         }
5499
5500         if (btrfs_extent_readonly(root, disk_bytenr))
5501                 goto out;
5502
5503         /*
5504          * look for other files referencing this extent, if we
5505          * find any we must cow
5506          */
5507         if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5508                                   key.offset - backref_offset, disk_bytenr))
5509                 goto out;
5510
5511         /*
5512          * adjust disk_bytenr and num_bytes to cover just the bytes
5513          * in this extent we are about to write.  If there
5514          * are any csums in that range we have to cow in order
5515          * to keep the csums correct
5516          */
5517         disk_bytenr += backref_offset;
5518         disk_bytenr += offset - key.offset;
5519         num_bytes = min(offset + len, extent_end) - offset;
5520         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5521                                 goto out;
5522         /*
5523          * all of the above have passed, it is safe to overwrite this extent
5524          * without cow
5525          */
5526         ret = 1;
5527 out:
5528         btrfs_free_path(path);
5529         return ret;
5530 }
5531
5532 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5533                                    struct buffer_head *bh_result, int create)
5534 {
5535         struct extent_map *em;
5536         struct btrfs_root *root = BTRFS_I(inode)->root;
5537         u64 start = iblock << inode->i_blkbits;
5538         u64 len = bh_result->b_size;
5539         struct btrfs_trans_handle *trans;
5540
5541         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5542         if (IS_ERR(em))
5543                 return PTR_ERR(em);
5544
5545         /*
5546          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5547          * io.  INLINE is special, and we could probably kludge it in here, but
5548          * it's still buffered so for safety lets just fall back to the generic
5549          * buffered path.
5550          *
5551          * For COMPRESSED we _have_ to read the entire extent in so we can
5552          * decompress it, so there will be buffering required no matter what we
5553          * do, so go ahead and fallback to buffered.
5554          *
5555          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5556          * to buffered IO.  Don't blame me, this is the price we pay for using
5557          * the generic code.
5558          */
5559         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5560             em->block_start == EXTENT_MAP_INLINE) {
5561                 free_extent_map(em);
5562                 return -ENOTBLK;
5563         }
5564
5565         /* Just a good old fashioned hole, return */
5566         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5567                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5568                 free_extent_map(em);
5569                 /* DIO will do one hole at a time, so just unlock a sector */
5570                 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5571                               start + root->sectorsize - 1, GFP_NOFS);
5572                 return 0;
5573         }
5574
5575         /*
5576          * We don't allocate a new extent in the following cases
5577          *
5578          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
5579          * existing extent.
5580          * 2) The extent is marked as PREALLOC.  We're good to go here and can
5581          * just use the extent.
5582          *
5583          */
5584         if (!create) {
5585                 len = em->len - (start - em->start);
5586                 goto map;
5587         }
5588
5589         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5590             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5591              em->block_start != EXTENT_MAP_HOLE)) {
5592                 int type;
5593                 int ret;
5594                 u64 block_start;
5595
5596                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5597                         type = BTRFS_ORDERED_PREALLOC;
5598                 else
5599                         type = BTRFS_ORDERED_NOCOW;
5600                 len = min(len, em->len - (start - em->start));
5601                 block_start = em->block_start + (start - em->start);
5602
5603                 /*
5604                  * we're not going to log anything, but we do need
5605                  * to make sure the current transaction stays open
5606                  * while we look for nocow cross refs
5607                  */
5608                 trans = btrfs_join_transaction(root);
5609                 if (IS_ERR(trans))
5610                         goto must_cow;
5611
5612                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5613                         ret = btrfs_add_ordered_extent_dio(inode, start,
5614                                            block_start, len, len, type);
5615                         btrfs_end_transaction(trans, root);
5616                         if (ret) {
5617                                 free_extent_map(em);
5618                                 return ret;
5619                         }
5620                         goto unlock;
5621                 }
5622                 btrfs_end_transaction(trans, root);
5623         }
5624 must_cow:
5625         /*
5626          * this will cow the extent, reset the len in case we changed
5627          * it above
5628          */
5629         len = bh_result->b_size;
5630         em = btrfs_new_extent_direct(inode, em, start, len);
5631         if (IS_ERR(em))
5632                 return PTR_ERR(em);
5633         len = min(len, em->len - (start - em->start));
5634 unlock:
5635         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5636                           EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5637                           0, NULL, GFP_NOFS);
5638 map:
5639         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5640                 inode->i_blkbits;
5641         bh_result->b_size = len;
5642         bh_result->b_bdev = em->bdev;
5643         set_buffer_mapped(bh_result);
5644         if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5645                 set_buffer_new(bh_result);
5646
5647         free_extent_map(em);
5648
5649         return 0;
5650 }
5651
5652 struct btrfs_dio_private {
5653         struct inode *inode;
5654         u64 logical_offset;
5655         u64 disk_bytenr;
5656         u64 bytes;
5657         u32 *csums;
5658         void *private;
5659
5660         /* number of bios pending for this dio */
5661         atomic_t pending_bios;
5662
5663         /* IO errors */
5664         int errors;
5665
5666         struct bio *orig_bio;
5667 };
5668
5669 static void btrfs_endio_direct_read(struct bio *bio, int err)
5670 {
5671         struct btrfs_dio_private *dip = bio->bi_private;
5672         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5673         struct bio_vec *bvec = bio->bi_io_vec;
5674         struct inode *inode = dip->inode;
5675         struct btrfs_root *root = BTRFS_I(inode)->root;
5676         u64 start;
5677         u32 *private = dip->csums;
5678
5679         start = dip->logical_offset;
5680         do {
5681                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5682                         struct page *page = bvec->bv_page;
5683                         char *kaddr;
5684                         u32 csum = ~(u32)0;
5685                         unsigned long flags;
5686
5687                         local_irq_save(flags);
5688                         kaddr = kmap_atomic(page, KM_IRQ0);
5689                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5690                                                csum, bvec->bv_len);
5691                         btrfs_csum_final(csum, (char *)&csum);
5692                         kunmap_atomic(kaddr, KM_IRQ0);
5693                         local_irq_restore(flags);
5694
5695                         flush_dcache_page(bvec->bv_page);
5696                         if (csum != *private) {
5697                                 printk(KERN_ERR "btrfs csum failed ino %llu off"
5698                                       " %llu csum %u private %u\n",
5699                                       (unsigned long long)btrfs_ino(inode),
5700                                       (unsigned long long)start,
5701                                       csum, *private);
5702                                 err = -EIO;
5703                         }
5704                 }
5705
5706                 start += bvec->bv_len;
5707                 private++;
5708                 bvec++;
5709         } while (bvec <= bvec_end);
5710
5711         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5712                       dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5713         bio->bi_private = dip->private;
5714
5715         kfree(dip->csums);
5716         kfree(dip);
5717
5718         /* If we had a csum failure make sure to clear the uptodate flag */
5719         if (err)
5720                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5721         dio_end_io(bio, err);
5722 }
5723
5724 static void btrfs_endio_direct_write(struct bio *bio, int err)
5725 {
5726         struct btrfs_dio_private *dip = bio->bi_private;
5727         struct inode *inode = dip->inode;
5728         struct btrfs_root *root = BTRFS_I(inode)->root;
5729         struct btrfs_trans_handle *trans;
5730         struct btrfs_ordered_extent *ordered = NULL;
5731         struct extent_state *cached_state = NULL;
5732         u64 ordered_offset = dip->logical_offset;
5733         u64 ordered_bytes = dip->bytes;
5734         int ret;
5735
5736         if (err)
5737                 goto out_done;
5738 again:
5739         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5740                                                    &ordered_offset,
5741                                                    ordered_bytes);
5742         if (!ret)
5743                 goto out_test;
5744
5745         BUG_ON(!ordered);
5746
5747         trans = btrfs_join_transaction(root);
5748         if (IS_ERR(trans)) {
5749                 err = -ENOMEM;
5750                 goto out;
5751         }
5752         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5753
5754         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5755                 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5756                 if (!ret)
5757                         ret = btrfs_update_inode(trans, root, inode);
5758                 err = ret;
5759                 goto out;
5760         }
5761
5762         lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5763                          ordered->file_offset + ordered->len - 1, 0,
5764                          &cached_state, GFP_NOFS);
5765
5766         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5767                 ret = btrfs_mark_extent_written(trans, inode,
5768                                                 ordered->file_offset,
5769                                                 ordered->file_offset +
5770                                                 ordered->len);
5771                 if (ret) {
5772                         err = ret;
5773                         goto out_unlock;
5774                 }
5775         } else {
5776                 ret = insert_reserved_file_extent(trans, inode,
5777                                                   ordered->file_offset,
5778                                                   ordered->start,
5779                                                   ordered->disk_len,
5780                                                   ordered->len,
5781                                                   ordered->len,
5782                                                   0, 0, 0,
5783                                                   BTRFS_FILE_EXTENT_REG);
5784                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5785                                    ordered->file_offset, ordered->len);
5786                 if (ret) {
5787                         err = ret;
5788                         WARN_ON(1);
5789                         goto out_unlock;
5790                 }
5791         }
5792
5793         add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5794         ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5795         if (!ret)
5796                 btrfs_update_inode(trans, root, inode);
5797         ret = 0;
5798 out_unlock:
5799         unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5800                              ordered->file_offset + ordered->len - 1,
5801                              &cached_state, GFP_NOFS);
5802 out:
5803         btrfs_delalloc_release_metadata(inode, ordered->len);
5804         btrfs_end_transaction(trans, root);
5805         ordered_offset = ordered->file_offset + ordered->len;
5806         btrfs_put_ordered_extent(ordered);
5807         btrfs_put_ordered_extent(ordered);
5808
5809 out_test:
5810         /*
5811          * our bio might span multiple ordered extents.  If we haven't
5812          * completed the accounting for the whole dio, go back and try again
5813          */
5814         if (ordered_offset < dip->logical_offset + dip->bytes) {
5815                 ordered_bytes = dip->logical_offset + dip->bytes -
5816                         ordered_offset;
5817                 goto again;
5818         }
5819 out_done:
5820         bio->bi_private = dip->private;
5821
5822         kfree(dip->csums);
5823         kfree(dip);
5824
5825         /* If we had an error make sure to clear the uptodate flag */
5826         if (err)
5827                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5828         dio_end_io(bio, err);
5829 }
5830
5831 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5832                                     struct bio *bio, int mirror_num,
5833                                     unsigned long bio_flags, u64 offset)
5834 {
5835         int ret;
5836         struct btrfs_root *root = BTRFS_I(inode)->root;
5837         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5838         BUG_ON(ret);
5839         return 0;
5840 }
5841
5842 static void btrfs_end_dio_bio(struct bio *bio, int err)
5843 {
5844         struct btrfs_dio_private *dip = bio->bi_private;
5845
5846         if (err) {
5847                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
5848                       "sector %#Lx len %u err no %d\n",
5849                       (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
5850                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
5851                 dip->errors = 1;
5852
5853                 /*
5854                  * before atomic variable goto zero, we must make sure
5855                  * dip->errors is perceived to be set.
5856                  */
5857                 smp_mb__before_atomic_dec();
5858         }
5859
5860         /* if there are more bios still pending for this dio, just exit */
5861         if (!atomic_dec_and_test(&dip->pending_bios))
5862                 goto out;
5863
5864         if (dip->errors)
5865                 bio_io_error(dip->orig_bio);
5866         else {
5867                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5868                 bio_endio(dip->orig_bio, 0);
5869         }
5870 out:
5871         bio_put(bio);
5872 }
5873
5874 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5875                                        u64 first_sector, gfp_t gfp_flags)
5876 {
5877         int nr_vecs = bio_get_nr_vecs(bdev);
5878         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5879 }
5880
5881 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5882                                          int rw, u64 file_offset, int skip_sum,
5883                                          u32 *csums, int async_submit)
5884 {
5885         int write = rw & REQ_WRITE;
5886         struct btrfs_root *root = BTRFS_I(inode)->root;
5887         int ret;
5888
5889         bio_get(bio);
5890         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5891         if (ret)
5892                 goto err;
5893
5894         if (skip_sum)
5895                 goto map;
5896
5897         if (write && async_submit) {
5898                 ret = btrfs_wq_submit_bio(root->fs_info,
5899                                    inode, rw, bio, 0, 0,
5900                                    file_offset,
5901                                    __btrfs_submit_bio_start_direct_io,
5902                                    __btrfs_submit_bio_done);
5903                 goto err;
5904         } else if (write) {
5905                 /*
5906                  * If we aren't doing async submit, calculate the csum of the
5907                  * bio now.
5908                  */
5909                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
5910                 if (ret)
5911                         goto err;
5912         } else if (!skip_sum) {
5913                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
5914                                           file_offset, csums);
5915                 if (ret)
5916                         goto err;
5917         }
5918
5919 map:
5920         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
5921 err:
5922         bio_put(bio);
5923         return ret;
5924 }
5925
5926 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5927                                     int skip_sum)
5928 {
5929         struct inode *inode = dip->inode;
5930         struct btrfs_root *root = BTRFS_I(inode)->root;
5931         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5932         struct bio *bio;
5933         struct bio *orig_bio = dip->orig_bio;
5934         struct bio_vec *bvec = orig_bio->bi_io_vec;
5935         u64 start_sector = orig_bio->bi_sector;
5936         u64 file_offset = dip->logical_offset;
5937         u64 submit_len = 0;
5938         u64 map_length;
5939         int nr_pages = 0;
5940         u32 *csums = dip->csums;
5941         int ret = 0;
5942         int async_submit = 0;
5943         int write = rw & REQ_WRITE;
5944
5945         map_length = orig_bio->bi_size;
5946         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5947                               &map_length, NULL, 0);
5948         if (ret) {
5949                 bio_put(orig_bio);
5950                 return -EIO;
5951         }
5952
5953         if (map_length >= orig_bio->bi_size) {
5954                 bio = orig_bio;
5955                 goto submit;
5956         }
5957
5958         async_submit = 1;
5959         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5960         if (!bio)
5961                 return -ENOMEM;
5962         bio->bi_private = dip;
5963         bio->bi_end_io = btrfs_end_dio_bio;
5964         atomic_inc(&dip->pending_bios);
5965
5966         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
5967                 if (unlikely(map_length < submit_len + bvec->bv_len ||
5968                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5969                                  bvec->bv_offset) < bvec->bv_len)) {
5970                         /*
5971                          * inc the count before we submit the bio so
5972                          * we know the end IO handler won't happen before
5973                          * we inc the count. Otherwise, the dip might get freed
5974                          * before we're done setting it up
5975                          */
5976                         atomic_inc(&dip->pending_bios);
5977                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
5978                                                      file_offset, skip_sum,
5979                                                      csums, async_submit);
5980                         if (ret) {
5981                                 bio_put(bio);
5982                                 atomic_dec(&dip->pending_bios);
5983                                 goto out_err;
5984                         }
5985
5986                         /* Write's use the ordered csums */
5987                         if (!write && !skip_sum)
5988                                 csums = csums + nr_pages;
5989                         start_sector += submit_len >> 9;
5990                         file_offset += submit_len;
5991
5992                         submit_len = 0;
5993                         nr_pages = 0;
5994
5995                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
5996                                                   start_sector, GFP_NOFS);
5997                         if (!bio)
5998                                 goto out_err;
5999                         bio->bi_private = dip;
6000                         bio->bi_end_io = btrfs_end_dio_bio;
6001
6002                         map_length = orig_bio->bi_size;
6003                         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6004                                               &map_length, NULL, 0);
6005                         if (ret) {
6006                                 bio_put(bio);
6007                                 goto out_err;
6008                         }
6009                 } else {
6010                         submit_len += bvec->bv_len;
6011                         nr_pages ++;
6012                         bvec++;
6013                 }
6014         }
6015
6016 submit:
6017         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6018                                      csums, async_submit);
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         /* Write's use the ordered csum stuff, so we don't need dip->csums */
6057         if (!write && !skip_sum) {
6058                 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6059                 if (!dip->csums) {
6060                         kfree(dip);
6061                         ret = -ENOMEM;
6062                         goto free_ordered;
6063                 }
6064         }
6065
6066         dip->private = bio->bi_private;
6067         dip->inode = inode;
6068         dip->logical_offset = file_offset;
6069
6070         dip->bytes = 0;
6071         do {
6072                 dip->bytes += bvec->bv_len;
6073                 bvec++;
6074         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6075
6076         dip->disk_bytenr = (u64)bio->bi_sector << 9;
6077         bio->bi_private = dip;
6078         dip->errors = 0;
6079         dip->orig_bio = bio;
6080         atomic_set(&dip->pending_bios, 0);
6081
6082         if (write)
6083                 bio->bi_end_io = btrfs_endio_direct_write;
6084         else
6085                 bio->bi_end_io = btrfs_endio_direct_read;
6086
6087         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6088         if (!ret)
6089                 return;
6090 free_ordered:
6091         /*
6092          * If this is a write, we need to clean up the reserved space and kill
6093          * the ordered extent.
6094          */
6095         if (write) {
6096                 struct btrfs_ordered_extent *ordered;
6097                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6098                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6099                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6100                         btrfs_free_reserved_extent(root, ordered->start,
6101                                                    ordered->disk_len);
6102                 btrfs_put_ordered_extent(ordered);
6103                 btrfs_put_ordered_extent(ordered);
6104         }
6105         bio_endio(bio, ret);
6106 }
6107
6108 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6109                         const struct iovec *iov, loff_t offset,
6110                         unsigned long nr_segs)
6111 {
6112         int seg;
6113         int i;
6114         size_t size;
6115         unsigned long addr;
6116         unsigned blocksize_mask = root->sectorsize - 1;
6117         ssize_t retval = -EINVAL;
6118         loff_t end = offset;
6119
6120         if (offset & blocksize_mask)
6121                 goto out;
6122
6123         /* Check the memory alignment.  Blocks cannot straddle pages */
6124         for (seg = 0; seg < nr_segs; seg++) {
6125                 addr = (unsigned long)iov[seg].iov_base;
6126                 size = iov[seg].iov_len;
6127                 end += size;
6128                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6129                         goto out;
6130
6131                 /* If this is a write we don't need to check anymore */
6132                 if (rw & WRITE)
6133                         continue;
6134
6135                 /*
6136                  * Check to make sure we don't have duplicate iov_base's in this
6137                  * iovec, if so return EINVAL, otherwise we'll get csum errors
6138                  * when reading back.
6139                  */
6140                 for (i = seg + 1; i < nr_segs; i++) {
6141                         if (iov[seg].iov_base == iov[i].iov_base)
6142                                 goto out;
6143                 }
6144         }
6145         retval = 0;
6146 out:
6147         return retval;
6148 }
6149 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6150                         const struct iovec *iov, loff_t offset,
6151                         unsigned long nr_segs)
6152 {
6153         struct file *file = iocb->ki_filp;
6154         struct inode *inode = file->f_mapping->host;
6155         struct btrfs_ordered_extent *ordered;
6156         struct extent_state *cached_state = NULL;
6157         u64 lockstart, lockend;
6158         ssize_t ret;
6159         int writing = rw & WRITE;
6160         int write_bits = 0;
6161         size_t count = iov_length(iov, nr_segs);
6162
6163         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6164                             offset, nr_segs)) {
6165                 return 0;
6166         }
6167
6168         lockstart = offset;
6169         lockend = offset + count - 1;
6170
6171         if (writing) {
6172                 ret = btrfs_delalloc_reserve_space(inode, count);
6173                 if (ret)
6174                         goto out;
6175         }
6176
6177         while (1) {
6178                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6179                                  0, &cached_state, GFP_NOFS);
6180                 /*
6181                  * We're concerned with the entire range that we're going to be
6182                  * doing DIO to, so we need to make sure theres no ordered
6183                  * extents in this range.
6184                  */
6185                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6186                                                      lockend - lockstart + 1);
6187                 if (!ordered)
6188                         break;
6189                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6190                                      &cached_state, GFP_NOFS);
6191                 btrfs_start_ordered_extent(inode, ordered, 1);
6192                 btrfs_put_ordered_extent(ordered);
6193                 cond_resched();
6194         }
6195
6196         /*
6197          * we don't use btrfs_set_extent_delalloc because we don't want
6198          * the dirty or uptodate bits
6199          */
6200         if (writing) {
6201                 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6202                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6203                                      EXTENT_DELALLOC, 0, NULL, &cached_state,
6204                                      GFP_NOFS);
6205                 if (ret) {
6206                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6207                                          lockend, EXTENT_LOCKED | write_bits,
6208                                          1, 0, &cached_state, GFP_NOFS);
6209                         goto out;
6210                 }
6211         }
6212
6213         free_extent_state(cached_state);
6214         cached_state = NULL;
6215
6216         ret = __blockdev_direct_IO(rw, iocb, inode,
6217                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6218                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6219                    btrfs_submit_direct, 0);
6220
6221         if (ret < 0 && ret != -EIOCBQUEUED) {
6222                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6223                               offset + iov_length(iov, nr_segs) - 1,
6224                               EXTENT_LOCKED | write_bits, 1, 0,
6225                               &cached_state, GFP_NOFS);
6226         } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6227                 /*
6228                  * We're falling back to buffered, unlock the section we didn't
6229                  * do IO on.
6230                  */
6231                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6232                               offset + iov_length(iov, nr_segs) - 1,
6233                               EXTENT_LOCKED | write_bits, 1, 0,
6234                               &cached_state, GFP_NOFS);
6235         }
6236 out:
6237         free_extent_state(cached_state);
6238         return ret;
6239 }
6240
6241 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6242                 __u64 start, __u64 len)
6243 {
6244         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6245 }
6246
6247 int btrfs_readpage(struct file *file, struct page *page)
6248 {
6249         struct extent_io_tree *tree;
6250         tree = &BTRFS_I(page->mapping->host)->io_tree;
6251         return extent_read_full_page(tree, page, btrfs_get_extent);
6252 }
6253
6254 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6255 {
6256         struct extent_io_tree *tree;
6257
6258
6259         if (current->flags & PF_MEMALLOC) {
6260                 redirty_page_for_writepage(wbc, page);
6261                 unlock_page(page);
6262                 return 0;
6263         }
6264         tree = &BTRFS_I(page->mapping->host)->io_tree;
6265         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6266 }
6267
6268 int btrfs_writepages(struct address_space *mapping,
6269                      struct writeback_control *wbc)
6270 {
6271         struct extent_io_tree *tree;
6272
6273         tree = &BTRFS_I(mapping->host)->io_tree;
6274         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6275 }
6276
6277 static int
6278 btrfs_readpages(struct file *file, struct address_space *mapping,
6279                 struct list_head *pages, unsigned nr_pages)
6280 {
6281         struct extent_io_tree *tree;
6282         tree = &BTRFS_I(mapping->host)->io_tree;
6283         return extent_readpages(tree, mapping, pages, nr_pages,
6284                                 btrfs_get_extent);
6285 }
6286 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6287 {
6288         struct extent_io_tree *tree;
6289         struct extent_map_tree *map;
6290         int ret;
6291
6292         tree = &BTRFS_I(page->mapping->host)->io_tree;
6293         map = &BTRFS_I(page->mapping->host)->extent_tree;
6294         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6295         if (ret == 1) {
6296                 ClearPagePrivate(page);
6297                 set_page_private(page, 0);
6298                 page_cache_release(page);
6299         }
6300         return ret;
6301 }
6302
6303 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6304 {
6305         if (PageWriteback(page) || PageDirty(page))
6306                 return 0;
6307         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6308 }
6309
6310 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6311 {
6312         struct extent_io_tree *tree;
6313         struct btrfs_ordered_extent *ordered;
6314         struct extent_state *cached_state = NULL;
6315         u64 page_start = page_offset(page);
6316         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6317
6318
6319         /*
6320          * we have the page locked, so new writeback can't start,
6321          * and the dirty bit won't be cleared while we are here.
6322          *
6323          * Wait for IO on this page so that we can safely clear
6324          * the PagePrivate2 bit and do ordered accounting
6325          */
6326         wait_on_page_writeback(page);
6327
6328         tree = &BTRFS_I(page->mapping->host)->io_tree;
6329         if (offset) {
6330                 btrfs_releasepage(page, GFP_NOFS);
6331                 return;
6332         }
6333         lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6334                          GFP_NOFS);
6335         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6336                                            page_offset(page));
6337         if (ordered) {
6338                 /*
6339                  * IO on this page will never be started, so we need
6340                  * to account for any ordered extents now
6341                  */
6342                 clear_extent_bit(tree, page_start, page_end,
6343                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6344                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6345                                  &cached_state, GFP_NOFS);
6346                 /*
6347                  * whoever cleared the private bit is responsible
6348                  * for the finish_ordered_io
6349                  */
6350                 if (TestClearPagePrivate2(page)) {
6351                         btrfs_finish_ordered_io(page->mapping->host,
6352                                                 page_start, page_end);
6353                 }
6354                 btrfs_put_ordered_extent(ordered);
6355                 cached_state = NULL;
6356                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6357                                  GFP_NOFS);
6358         }
6359         clear_extent_bit(tree, page_start, page_end,
6360                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6361                  EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6362         __btrfs_releasepage(page, GFP_NOFS);
6363
6364         ClearPageChecked(page);
6365         if (PagePrivate(page)) {
6366                 ClearPagePrivate(page);
6367                 set_page_private(page, 0);
6368                 page_cache_release(page);
6369         }
6370 }
6371
6372 /*
6373  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6374  * called from a page fault handler when a page is first dirtied. Hence we must
6375  * be careful to check for EOF conditions here. We set the page up correctly
6376  * for a written page which means we get ENOSPC checking when writing into
6377  * holes and correct delalloc and unwritten extent mapping on filesystems that
6378  * support these features.
6379  *
6380  * We are not allowed to take the i_mutex here so we have to play games to
6381  * protect against truncate races as the page could now be beyond EOF.  Because
6382  * vmtruncate() writes the inode size before removing pages, once we have the
6383  * page lock we can determine safely if the page is beyond EOF. If it is not
6384  * beyond EOF, then the page is guaranteed safe against truncation until we
6385  * unlock the page.
6386  */
6387 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6388 {
6389         struct page *page = vmf->page;
6390         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6391         struct btrfs_root *root = BTRFS_I(inode)->root;
6392         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6393         struct btrfs_ordered_extent *ordered;
6394         struct extent_state *cached_state = NULL;
6395         char *kaddr;
6396         unsigned long zero_start;
6397         loff_t size;
6398         int ret;
6399         u64 page_start;
6400         u64 page_end;
6401
6402         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6403         if (ret) {
6404                 if (ret == -ENOMEM)
6405                         ret = VM_FAULT_OOM;
6406                 else /* -ENOSPC, -EIO, etc */
6407                         ret = VM_FAULT_SIGBUS;
6408                 goto out;
6409         }
6410
6411         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6412 again:
6413         lock_page(page);
6414         size = i_size_read(inode);
6415         page_start = page_offset(page);
6416         page_end = page_start + PAGE_CACHE_SIZE - 1;
6417
6418         if ((page->mapping != inode->i_mapping) ||
6419             (page_start >= size)) {
6420                 /* page got truncated out from underneath us */
6421                 goto out_unlock;
6422         }
6423         wait_on_page_writeback(page);
6424
6425         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6426                          GFP_NOFS);
6427         set_page_extent_mapped(page);
6428
6429         /*
6430          * we can't set the delalloc bits if there are pending ordered
6431          * extents.  Drop our locks and wait for them to finish
6432          */
6433         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6434         if (ordered) {
6435                 unlock_extent_cached(io_tree, page_start, page_end,
6436                                      &cached_state, GFP_NOFS);
6437                 unlock_page(page);
6438                 btrfs_start_ordered_extent(inode, ordered, 1);
6439                 btrfs_put_ordered_extent(ordered);
6440                 goto again;
6441         }
6442
6443         /*
6444          * XXX - page_mkwrite gets called every time the page is dirtied, even
6445          * if it was already dirty, so for space accounting reasons we need to
6446          * clear any delalloc bits for the range we are fixing to save.  There
6447          * is probably a better way to do this, but for now keep consistent with
6448          * prepare_pages in the normal write path.
6449          */
6450         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6451                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6452                           0, 0, &cached_state, GFP_NOFS);
6453
6454         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6455                                         &cached_state);
6456         if (ret) {
6457                 unlock_extent_cached(io_tree, page_start, page_end,
6458                                      &cached_state, GFP_NOFS);
6459                 ret = VM_FAULT_SIGBUS;
6460                 goto out_unlock;
6461         }
6462         ret = 0;
6463
6464         /* page is wholly or partially inside EOF */
6465         if (page_start + PAGE_CACHE_SIZE > size)
6466                 zero_start = size & ~PAGE_CACHE_MASK;
6467         else
6468                 zero_start = PAGE_CACHE_SIZE;
6469
6470         if (zero_start != PAGE_CACHE_SIZE) {
6471                 kaddr = kmap(page);
6472                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6473                 flush_dcache_page(page);
6474                 kunmap(page);
6475         }
6476         ClearPageChecked(page);
6477         set_page_dirty(page);
6478         SetPageUptodate(page);
6479
6480         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6481         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6482
6483         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6484
6485 out_unlock:
6486         if (!ret)
6487                 return VM_FAULT_LOCKED;
6488         unlock_page(page);
6489         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6490 out:
6491         return ret;
6492 }
6493
6494 static int btrfs_truncate(struct inode *inode)
6495 {
6496         struct btrfs_root *root = BTRFS_I(inode)->root;
6497         struct btrfs_block_rsv *rsv;
6498         int ret;
6499         int err = 0;
6500         struct btrfs_trans_handle *trans;
6501         unsigned long nr;
6502         u64 mask = root->sectorsize - 1;
6503
6504         ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6505         if (ret)
6506                 return ret;
6507
6508         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6509         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6510
6511         /*
6512          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
6513          * 3 things going on here
6514          *
6515          * 1) We need to reserve space for our orphan item and the space to
6516          * delete our orphan item.  Lord knows we don't want to have a dangling
6517          * orphan item because we didn't reserve space to remove it.
6518          *
6519          * 2) We need to reserve space to update our inode.
6520          *
6521          * 3) We need to have something to cache all the space that is going to
6522          * be free'd up by the truncate operation, but also have some slack
6523          * space reserved in case it uses space during the truncate (thank you
6524          * very much snapshotting).
6525          *
6526          * And we need these to all be seperate.  The fact is we can use alot of
6527          * space doing the truncate, and we have no earthly idea how much space
6528          * we will use, so we need the truncate reservation to be seperate so it
6529          * doesn't end up using space reserved for updating the inode or
6530          * removing the orphan item.  We also need to be able to stop the
6531          * transaction and start a new one, which means we need to be able to
6532          * update the inode several times, and we have no idea of knowing how
6533          * many times that will be, so we can't just reserve 1 item for the
6534          * entirety of the opration, so that has to be done seperately as well.
6535          * Then there is the orphan item, which does indeed need to be held on
6536          * to for the whole operation, and we need nobody to touch this reserved
6537          * space except the orphan code.
6538          *
6539          * So that leaves us with
6540          *
6541          * 1) root->orphan_block_rsv - for the orphan deletion.
6542          * 2) rsv - for the truncate reservation, which we will steal from the
6543          * transaction reservation.
6544          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6545          * updating the inode.
6546          */
6547         rsv = btrfs_alloc_block_rsv(root);
6548         if (!rsv)
6549                 return -ENOMEM;
6550         btrfs_add_durable_block_rsv(root->fs_info, rsv);
6551
6552         trans = btrfs_start_transaction(root, 4);
6553         if (IS_ERR(trans)) {
6554                 err = PTR_ERR(trans);
6555                 goto out;
6556         }
6557
6558         /*
6559          * Reserve space for the truncate process.  Truncate should be adding
6560          * space, but if there are snapshots it may end up using space.
6561          */
6562         ret = btrfs_truncate_reserve_metadata(trans, root, rsv);
6563         BUG_ON(ret);
6564
6565         ret = btrfs_orphan_add(trans, inode);
6566         if (ret) {
6567                 btrfs_end_transaction(trans, root);
6568                 goto out;
6569         }
6570
6571         nr = trans->blocks_used;
6572         btrfs_end_transaction(trans, root);
6573         btrfs_btree_balance_dirty(root, nr);
6574
6575         /*
6576          * Ok so we've already migrated our bytes over for the truncate, so here
6577          * just reserve the one slot we need for updating the inode.
6578          */
6579         trans = btrfs_start_transaction(root, 1);
6580         if (IS_ERR(trans)) {
6581                 err = PTR_ERR(trans);
6582                 goto out;
6583         }
6584         trans->block_rsv = rsv;
6585
6586         /*
6587          * setattr is responsible for setting the ordered_data_close flag,
6588          * but that is only tested during the last file release.  That
6589          * could happen well after the next commit, leaving a great big
6590          * window where new writes may get lost if someone chooses to write
6591          * to this file after truncating to zero
6592          *
6593          * The inode doesn't have any dirty data here, and so if we commit
6594          * this is a noop.  If someone immediately starts writing to the inode
6595          * it is very likely we'll catch some of their writes in this
6596          * transaction, and the commit will find this file on the ordered
6597          * data list with good things to send down.
6598          *
6599          * This is a best effort solution, there is still a window where
6600          * using truncate to replace the contents of the file will
6601          * end up with a zero length file after a crash.
6602          */
6603         if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6604                 btrfs_add_ordered_operation(trans, root, inode);
6605
6606         while (1) {
6607                 if (!trans) {
6608                         trans = btrfs_start_transaction(root, 3);
6609                         if (IS_ERR(trans)) {
6610                                 err = PTR_ERR(trans);
6611                                 goto out;
6612                         }
6613
6614                         ret = btrfs_truncate_reserve_metadata(trans, root,
6615                                                               rsv);
6616                         BUG_ON(ret);
6617
6618                         trans->block_rsv = rsv;
6619                 }
6620
6621                 ret = btrfs_truncate_inode_items(trans, root, inode,
6622                                                  inode->i_size,
6623                                                  BTRFS_EXTENT_DATA_KEY);
6624                 if (ret != -EAGAIN) {
6625                         err = ret;
6626                         break;
6627                 }
6628
6629                 trans->block_rsv = &root->fs_info->trans_block_rsv;
6630                 ret = btrfs_update_inode(trans, root, inode);
6631                 if (ret) {
6632                         err = ret;
6633                         break;
6634                 }
6635
6636                 nr = trans->blocks_used;
6637                 btrfs_end_transaction(trans, root);
6638                 trans = NULL;
6639                 btrfs_btree_balance_dirty(root, nr);
6640         }
6641
6642         if (ret == 0 && inode->i_nlink > 0) {
6643                 trans->block_rsv = root->orphan_block_rsv;
6644                 ret = btrfs_orphan_del(trans, inode);
6645                 if (ret)
6646                         err = ret;
6647         } else if (ret && inode->i_nlink > 0) {
6648                 /*
6649                  * Failed to do the truncate, remove us from the in memory
6650                  * orphan list.
6651                  */
6652                 ret = btrfs_orphan_del(NULL, inode);
6653         }
6654
6655         trans->block_rsv = &root->fs_info->trans_block_rsv;
6656         ret = btrfs_update_inode(trans, root, inode);
6657         if (ret && !err)
6658                 err = ret;
6659
6660         nr = trans->blocks_used;
6661         ret = btrfs_end_transaction_throttle(trans, root);
6662         btrfs_btree_balance_dirty(root, nr);
6663
6664 out:
6665         btrfs_free_block_rsv(root, rsv);
6666
6667         if (ret && !err)
6668                 err = ret;
6669
6670         return err;
6671 }
6672
6673 /*
6674  * create a new subvolume directory/inode (helper for the ioctl).
6675  */
6676 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6677                              struct btrfs_root *new_root, u64 new_dirid)
6678 {
6679         struct inode *inode;
6680         int err;
6681         u64 index = 0;
6682
6683         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6684                                 new_dirid, S_IFDIR | 0700, &index);
6685         if (IS_ERR(inode))
6686                 return PTR_ERR(inode);
6687         inode->i_op = &btrfs_dir_inode_operations;
6688         inode->i_fop = &btrfs_dir_file_operations;
6689
6690         inode->i_nlink = 1;
6691         btrfs_i_size_write(inode, 0);
6692
6693         err = btrfs_update_inode(trans, new_root, inode);
6694         BUG_ON(err);
6695
6696         iput(inode);
6697         return 0;
6698 }
6699
6700 /* helper function for file defrag and space balancing.  This
6701  * forces readahead on a given range of bytes in an inode
6702  */
6703 unsigned long btrfs_force_ra(struct address_space *mapping,
6704                               struct file_ra_state *ra, struct file *file,
6705                               pgoff_t offset, pgoff_t last_index)
6706 {
6707         pgoff_t req_size = last_index - offset + 1;
6708
6709         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
6710         return offset + req_size;
6711 }
6712
6713 struct inode *btrfs_alloc_inode(struct super_block *sb)
6714 {
6715         struct btrfs_inode *ei;
6716         struct inode *inode;
6717
6718         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6719         if (!ei)
6720                 return NULL;
6721
6722         ei->root = NULL;
6723         ei->space_info = NULL;
6724         ei->generation = 0;
6725         ei->sequence = 0;
6726         ei->last_trans = 0;
6727         ei->last_sub_trans = 0;
6728         ei->logged_trans = 0;
6729         ei->delalloc_bytes = 0;
6730         ei->reserved_bytes = 0;
6731         ei->disk_i_size = 0;
6732         ei->flags = 0;
6733         ei->index_cnt = (u64)-1;
6734         ei->last_unlink_trans = 0;
6735
6736         atomic_set(&ei->outstanding_extents, 0);
6737         atomic_set(&ei->reserved_extents, 0);
6738
6739         ei->ordered_data_close = 0;
6740         ei->orphan_meta_reserved = 0;
6741         ei->dummy_inode = 0;
6742         ei->in_defrag = 0;
6743         ei->force_compress = BTRFS_COMPRESS_NONE;
6744
6745         ei->delayed_node = NULL;
6746
6747         inode = &ei->vfs_inode;
6748         extent_map_tree_init(&ei->extent_tree);
6749         extent_io_tree_init(&ei->io_tree, &inode->i_data);
6750         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
6751         mutex_init(&ei->log_mutex);
6752         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6753         INIT_LIST_HEAD(&ei->i_orphan);
6754         INIT_LIST_HEAD(&ei->delalloc_inodes);
6755         INIT_LIST_HEAD(&ei->ordered_operations);
6756         RB_CLEAR_NODE(&ei->rb_node);
6757
6758         return inode;
6759 }
6760
6761 static void btrfs_i_callback(struct rcu_head *head)
6762 {
6763         struct inode *inode = container_of(head, struct inode, i_rcu);
6764         INIT_LIST_HEAD(&inode->i_dentry);
6765         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6766 }
6767
6768 void btrfs_destroy_inode(struct inode *inode)
6769 {
6770         struct btrfs_ordered_extent *ordered;
6771         struct btrfs_root *root = BTRFS_I(inode)->root;
6772
6773         WARN_ON(!list_empty(&inode->i_dentry));
6774         WARN_ON(inode->i_data.nrpages);
6775         WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents));
6776         WARN_ON(atomic_read(&BTRFS_I(inode)->reserved_extents));
6777
6778         /*
6779          * This can happen where we create an inode, but somebody else also
6780          * created the same inode and we need to destroy the one we already
6781          * created.
6782          */
6783         if (!root)
6784                 goto free;
6785
6786         /*
6787          * Make sure we're properly removed from the ordered operation
6788          * lists.
6789          */
6790         smp_mb();
6791         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6792                 spin_lock(&root->fs_info->ordered_extent_lock);
6793                 list_del_init(&BTRFS_I(inode)->ordered_operations);
6794                 spin_unlock(&root->fs_info->ordered_extent_lock);
6795         }
6796
6797         spin_lock(&root->orphan_lock);
6798         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6799                 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
6800                        (unsigned long long)btrfs_ino(inode));
6801                 list_del_init(&BTRFS_I(inode)->i_orphan);
6802         }
6803         spin_unlock(&root->orphan_lock);
6804
6805         while (1) {
6806                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6807                 if (!ordered)
6808                         break;
6809                 else {
6810                         printk(KERN_ERR "btrfs found ordered "
6811                                "extent %llu %llu on inode cleanup\n",
6812                                (unsigned long long)ordered->file_offset,
6813                                (unsigned long long)ordered->len);
6814                         btrfs_remove_ordered_extent(inode, ordered);
6815                         btrfs_put_ordered_extent(ordered);
6816                         btrfs_put_ordered_extent(ordered);
6817                 }
6818         }
6819         inode_tree_del(inode);
6820         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6821 free:
6822         btrfs_remove_delayed_node(inode);
6823         call_rcu(&inode->i_rcu, btrfs_i_callback);
6824 }
6825
6826 int btrfs_drop_inode(struct inode *inode)
6827 {
6828         struct btrfs_root *root = BTRFS_I(inode)->root;
6829
6830         if (btrfs_root_refs(&root->root_item) == 0 &&
6831             !is_free_space_inode(root, inode))
6832                 return 1;
6833         else
6834                 return generic_drop_inode(inode);
6835 }
6836
6837 static void init_once(void *foo)
6838 {
6839         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6840
6841         inode_init_once(&ei->vfs_inode);
6842 }
6843
6844 void btrfs_destroy_cachep(void)
6845 {
6846         if (btrfs_inode_cachep)
6847                 kmem_cache_destroy(btrfs_inode_cachep);
6848         if (btrfs_trans_handle_cachep)
6849                 kmem_cache_destroy(btrfs_trans_handle_cachep);
6850         if (btrfs_transaction_cachep)
6851                 kmem_cache_destroy(btrfs_transaction_cachep);
6852         if (btrfs_path_cachep)
6853                 kmem_cache_destroy(btrfs_path_cachep);
6854         if (btrfs_free_space_cachep)
6855                 kmem_cache_destroy(btrfs_free_space_cachep);
6856 }
6857
6858 int btrfs_init_cachep(void)
6859 {
6860         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6861                         sizeof(struct btrfs_inode), 0,
6862                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6863         if (!btrfs_inode_cachep)
6864                 goto fail;
6865
6866         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6867                         sizeof(struct btrfs_trans_handle), 0,
6868                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6869         if (!btrfs_trans_handle_cachep)
6870                 goto fail;
6871
6872         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6873                         sizeof(struct btrfs_transaction), 0,
6874                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6875         if (!btrfs_transaction_cachep)
6876                 goto fail;
6877
6878         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6879                         sizeof(struct btrfs_path), 0,
6880                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6881         if (!btrfs_path_cachep)
6882                 goto fail;
6883
6884         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6885                         sizeof(struct btrfs_free_space), 0,
6886                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6887         if (!btrfs_free_space_cachep)
6888                 goto fail;
6889
6890         return 0;
6891 fail:
6892         btrfs_destroy_cachep();
6893         return -ENOMEM;
6894 }
6895
6896 static int btrfs_getattr(struct vfsmount *mnt,
6897                          struct dentry *dentry, struct kstat *stat)
6898 {
6899         struct inode *inode = dentry->d_inode;
6900         generic_fillattr(inode, stat);
6901         stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
6902         stat->blksize = PAGE_CACHE_SIZE;
6903         stat->blocks = (inode_get_bytes(inode) +
6904                         BTRFS_I(inode)->delalloc_bytes) >> 9;
6905         return 0;
6906 }
6907
6908 /*
6909  * If a file is moved, it will inherit the cow and compression flags of the new
6910  * directory.
6911  */
6912 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6913 {
6914         struct btrfs_inode *b_dir = BTRFS_I(dir);
6915         struct btrfs_inode *b_inode = BTRFS_I(inode);
6916
6917         if (b_dir->flags & BTRFS_INODE_NODATACOW)
6918                 b_inode->flags |= BTRFS_INODE_NODATACOW;
6919         else
6920                 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6921
6922         if (b_dir->flags & BTRFS_INODE_COMPRESS)
6923                 b_inode->flags |= BTRFS_INODE_COMPRESS;
6924         else
6925                 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6926 }
6927
6928 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6929                            struct inode *new_dir, struct dentry *new_dentry)
6930 {
6931         struct btrfs_trans_handle *trans;
6932         struct btrfs_root *root = BTRFS_I(old_dir)->root;
6933         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6934         struct inode *new_inode = new_dentry->d_inode;
6935         struct inode *old_inode = old_dentry->d_inode;
6936         struct timespec ctime = CURRENT_TIME;
6937         u64 index = 0;
6938         u64 root_objectid;
6939         int ret;
6940         u64 old_ino = btrfs_ino(old_inode);
6941
6942         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6943                 return -EPERM;
6944
6945         /* we only allow rename subvolume link between subvolumes */
6946         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6947                 return -EXDEV;
6948
6949         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6950             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
6951                 return -ENOTEMPTY;
6952
6953         if (S_ISDIR(old_inode->i_mode) && new_inode &&
6954             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6955                 return -ENOTEMPTY;
6956         /*
6957          * we're using rename to replace one file with another.
6958          * and the replacement file is large.  Start IO on it now so
6959          * we don't add too much work to the end of the transaction
6960          */
6961         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
6962             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6963                 filemap_flush(old_inode->i_mapping);
6964
6965         /* close the racy window with snapshot create/destroy ioctl */
6966         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
6967                 down_read(&root->fs_info->subvol_sem);
6968         /*
6969          * We want to reserve the absolute worst case amount of items.  So if
6970          * both inodes are subvols and we need to unlink them then that would
6971          * require 4 item modifications, but if they are both normal inodes it
6972          * would require 5 item modifications, so we'll assume their normal
6973          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6974          * should cover the worst case number of items we'll modify.
6975          */
6976         trans = btrfs_start_transaction(root, 20);
6977         if (IS_ERR(trans)) {
6978                 ret = PTR_ERR(trans);
6979                 goto out_notrans;
6980         }
6981
6982         if (dest != root)
6983                 btrfs_record_root_in_trans(trans, dest);
6984
6985         ret = btrfs_set_inode_index(new_dir, &index);
6986         if (ret)
6987                 goto out_fail;
6988
6989         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6990                 /* force full log commit if subvolume involved. */
6991                 root->fs_info->last_trans_log_full_commit = trans->transid;
6992         } else {
6993                 ret = btrfs_insert_inode_ref(trans, dest,
6994                                              new_dentry->d_name.name,
6995                                              new_dentry->d_name.len,
6996                                              old_ino,
6997                                              btrfs_ino(new_dir), index);
6998                 if (ret)
6999                         goto out_fail;
7000                 /*
7001                  * this is an ugly little race, but the rename is required
7002                  * to make sure that if we crash, the inode is either at the
7003                  * old name or the new one.  pinning the log transaction lets
7004                  * us make sure we don't allow a log commit to come in after
7005                  * we unlink the name but before we add the new name back in.
7006                  */
7007                 btrfs_pin_log_trans(root);
7008         }
7009         /*
7010          * make sure the inode gets flushed if it is replacing
7011          * something.
7012          */
7013         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7014                 btrfs_add_ordered_operation(trans, root, old_inode);
7015
7016         old_dir->i_ctime = old_dir->i_mtime = ctime;
7017         new_dir->i_ctime = new_dir->i_mtime = ctime;
7018         old_inode->i_ctime = ctime;
7019
7020         if (old_dentry->d_parent != new_dentry->d_parent)
7021                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7022
7023         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7024                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7025                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7026                                         old_dentry->d_name.name,
7027                                         old_dentry->d_name.len);
7028         } else {
7029                 ret = __btrfs_unlink_inode(trans, root, old_dir,
7030                                         old_dentry->d_inode,
7031                                         old_dentry->d_name.name,
7032                                         old_dentry->d_name.len);
7033                 if (!ret)
7034                         ret = btrfs_update_inode(trans, root, old_inode);
7035         }
7036         BUG_ON(ret);
7037
7038         if (new_inode) {
7039                 new_inode->i_ctime = CURRENT_TIME;
7040                 if (unlikely(btrfs_ino(new_inode) ==
7041                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7042                         root_objectid = BTRFS_I(new_inode)->location.objectid;
7043                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
7044                                                 root_objectid,
7045                                                 new_dentry->d_name.name,
7046                                                 new_dentry->d_name.len);
7047                         BUG_ON(new_inode->i_nlink == 0);
7048                 } else {
7049                         ret = btrfs_unlink_inode(trans, dest, new_dir,
7050                                                  new_dentry->d_inode,
7051                                                  new_dentry->d_name.name,
7052                                                  new_dentry->d_name.len);
7053                 }
7054                 BUG_ON(ret);
7055                 if (new_inode->i_nlink == 0) {
7056                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7057                         BUG_ON(ret);
7058                 }
7059         }
7060
7061         fixup_inode_flags(new_dir, old_inode);
7062
7063         ret = btrfs_add_link(trans, new_dir, old_inode,
7064                              new_dentry->d_name.name,
7065                              new_dentry->d_name.len, 0, index);
7066         BUG_ON(ret);
7067
7068         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7069                 struct dentry *parent = dget_parent(new_dentry);
7070                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7071                 dput(parent);
7072                 btrfs_end_log_trans(root);
7073         }
7074 out_fail:
7075         btrfs_end_transaction_throttle(trans, root);
7076 out_notrans:
7077         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7078                 up_read(&root->fs_info->subvol_sem);
7079
7080         return ret;
7081 }
7082
7083 /*
7084  * some fairly slow code that needs optimization. This walks the list
7085  * of all the inodes with pending delalloc and forces them to disk.
7086  */
7087 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7088 {
7089         struct list_head *head = &root->fs_info->delalloc_inodes;
7090         struct btrfs_inode *binode;
7091         struct inode *inode;
7092
7093         if (root->fs_info->sb->s_flags & MS_RDONLY)
7094                 return -EROFS;
7095
7096         spin_lock(&root->fs_info->delalloc_lock);
7097         while (!list_empty(head)) {
7098                 binode = list_entry(head->next, struct btrfs_inode,
7099                                     delalloc_inodes);
7100                 inode = igrab(&binode->vfs_inode);
7101                 if (!inode)
7102                         list_del_init(&binode->delalloc_inodes);
7103                 spin_unlock(&root->fs_info->delalloc_lock);
7104                 if (inode) {
7105                         filemap_flush(inode->i_mapping);
7106                         if (delay_iput)
7107                                 btrfs_add_delayed_iput(inode);
7108                         else
7109                                 iput(inode);
7110                 }
7111                 cond_resched();
7112                 spin_lock(&root->fs_info->delalloc_lock);
7113         }
7114         spin_unlock(&root->fs_info->delalloc_lock);
7115
7116         /* the filemap_flush will queue IO into the worker threads, but
7117          * we have to make sure the IO is actually started and that
7118          * ordered extents get created before we return
7119          */
7120         atomic_inc(&root->fs_info->async_submit_draining);
7121         while (atomic_read(&root->fs_info->nr_async_submits) ||
7122               atomic_read(&root->fs_info->async_delalloc_pages)) {
7123                 wait_event(root->fs_info->async_submit_wait,
7124                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7125                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7126         }
7127         atomic_dec(&root->fs_info->async_submit_draining);
7128         return 0;
7129 }
7130
7131 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7132                          const char *symname)
7133 {
7134         struct btrfs_trans_handle *trans;
7135         struct btrfs_root *root = BTRFS_I(dir)->root;
7136         struct btrfs_path *path;
7137         struct btrfs_key key;
7138         struct inode *inode = NULL;
7139         int err;
7140         int drop_inode = 0;
7141         u64 objectid;
7142         u64 index = 0 ;
7143         int name_len;
7144         int datasize;
7145         unsigned long ptr;
7146         struct btrfs_file_extent_item *ei;
7147         struct extent_buffer *leaf;
7148         unsigned long nr = 0;
7149
7150         name_len = strlen(symname) + 1;
7151         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7152                 return -ENAMETOOLONG;
7153
7154         /*
7155          * 2 items for inode item and ref
7156          * 2 items for dir items
7157          * 1 item for xattr if selinux is on
7158          */
7159         trans = btrfs_start_transaction(root, 5);
7160         if (IS_ERR(trans))
7161                 return PTR_ERR(trans);
7162
7163         err = btrfs_find_free_ino(root, &objectid);
7164         if (err)
7165                 goto out_unlock;
7166
7167         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7168                                 dentry->d_name.len, btrfs_ino(dir), objectid,
7169                                 S_IFLNK|S_IRWXUGO, &index);
7170         if (IS_ERR(inode)) {
7171                 err = PTR_ERR(inode);
7172                 goto out_unlock;
7173         }
7174
7175         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7176         if (err) {
7177                 drop_inode = 1;
7178                 goto out_unlock;
7179         }
7180
7181         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7182         if (err)
7183                 drop_inode = 1;
7184         else {
7185                 inode->i_mapping->a_ops = &btrfs_aops;
7186                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7187                 inode->i_fop = &btrfs_file_operations;
7188                 inode->i_op = &btrfs_file_inode_operations;
7189                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7190         }
7191         if (drop_inode)
7192                 goto out_unlock;
7193
7194         path = btrfs_alloc_path();
7195         BUG_ON(!path);
7196         key.objectid = btrfs_ino(inode);
7197         key.offset = 0;
7198         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7199         datasize = btrfs_file_extent_calc_inline_size(name_len);
7200         err = btrfs_insert_empty_item(trans, root, path, &key,
7201                                       datasize);
7202         if (err) {
7203                 drop_inode = 1;
7204                 btrfs_free_path(path);
7205                 goto out_unlock;
7206         }
7207         leaf = path->nodes[0];
7208         ei = btrfs_item_ptr(leaf, path->slots[0],
7209                             struct btrfs_file_extent_item);
7210         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7211         btrfs_set_file_extent_type(leaf, ei,
7212                                    BTRFS_FILE_EXTENT_INLINE);
7213         btrfs_set_file_extent_encryption(leaf, ei, 0);
7214         btrfs_set_file_extent_compression(leaf, ei, 0);
7215         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7216         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7217
7218         ptr = btrfs_file_extent_inline_start(ei);
7219         write_extent_buffer(leaf, symname, ptr, name_len);
7220         btrfs_mark_buffer_dirty(leaf);
7221         btrfs_free_path(path);
7222
7223         inode->i_op = &btrfs_symlink_inode_operations;
7224         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7225         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7226         inode_set_bytes(inode, name_len);
7227         btrfs_i_size_write(inode, name_len - 1);
7228         err = btrfs_update_inode(trans, root, inode);
7229         if (err)
7230                 drop_inode = 1;
7231
7232 out_unlock:
7233         nr = trans->blocks_used;
7234         btrfs_end_transaction_throttle(trans, root);
7235         if (drop_inode) {
7236                 inode_dec_link_count(inode);
7237                 iput(inode);
7238         }
7239         btrfs_btree_balance_dirty(root, nr);
7240         return err;
7241 }
7242
7243 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7244                                        u64 start, u64 num_bytes, u64 min_size,
7245                                        loff_t actual_len, u64 *alloc_hint,
7246                                        struct btrfs_trans_handle *trans)
7247 {
7248         struct btrfs_root *root = BTRFS_I(inode)->root;
7249         struct btrfs_key ins;
7250         u64 cur_offset = start;
7251         u64 i_size;
7252         int ret = 0;
7253         bool own_trans = true;
7254
7255         if (trans)
7256                 own_trans = false;
7257         while (num_bytes > 0) {
7258                 if (own_trans) {
7259                         trans = btrfs_start_transaction(root, 3);
7260                         if (IS_ERR(trans)) {
7261                                 ret = PTR_ERR(trans);
7262                                 break;
7263                         }
7264                 }
7265
7266                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7267                                            0, *alloc_hint, (u64)-1, &ins, 1);
7268                 if (ret) {
7269                         if (own_trans)
7270                                 btrfs_end_transaction(trans, root);
7271                         break;
7272                 }
7273
7274                 ret = insert_reserved_file_extent(trans, inode,
7275                                                   cur_offset, ins.objectid,
7276                                                   ins.offset, ins.offset,
7277                                                   ins.offset, 0, 0, 0,
7278                                                   BTRFS_FILE_EXTENT_PREALLOC);
7279                 BUG_ON(ret);
7280                 btrfs_drop_extent_cache(inode, cur_offset,
7281                                         cur_offset + ins.offset -1, 0);
7282
7283                 num_bytes -= ins.offset;
7284                 cur_offset += ins.offset;
7285                 *alloc_hint = ins.objectid + ins.offset;
7286
7287                 inode->i_ctime = CURRENT_TIME;
7288                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7289                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7290                     (actual_len > inode->i_size) &&
7291                     (cur_offset > inode->i_size)) {
7292                         if (cur_offset > actual_len)
7293                                 i_size = actual_len;
7294                         else
7295                                 i_size = cur_offset;
7296                         i_size_write(inode, i_size);
7297                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7298                 }
7299
7300                 ret = btrfs_update_inode(trans, root, inode);
7301                 BUG_ON(ret);
7302
7303                 if (own_trans)
7304                         btrfs_end_transaction(trans, root);
7305         }
7306         return ret;
7307 }
7308
7309 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7310                               u64 start, u64 num_bytes, u64 min_size,
7311                               loff_t actual_len, u64 *alloc_hint)
7312 {
7313         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7314                                            min_size, actual_len, alloc_hint,
7315                                            NULL);
7316 }
7317
7318 int btrfs_prealloc_file_range_trans(struct inode *inode,
7319                                     struct btrfs_trans_handle *trans, int mode,
7320                                     u64 start, u64 num_bytes, u64 min_size,
7321                                     loff_t actual_len, u64 *alloc_hint)
7322 {
7323         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7324                                            min_size, actual_len, alloc_hint, trans);
7325 }
7326
7327 static int btrfs_set_page_dirty(struct page *page)
7328 {
7329         return __set_page_dirty_nobuffers(page);
7330 }
7331
7332 static int btrfs_permission(struct inode *inode, int mask, unsigned int flags)
7333 {
7334         struct btrfs_root *root = BTRFS_I(inode)->root;
7335
7336         if (btrfs_root_readonly(root) && (mask & MAY_WRITE))
7337                 return -EROFS;
7338         if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
7339                 return -EACCES;
7340         return generic_permission(inode, mask, flags, btrfs_check_acl);
7341 }
7342
7343 static const struct inode_operations btrfs_dir_inode_operations = {
7344         .getattr        = btrfs_getattr,
7345         .lookup         = btrfs_lookup,
7346         .create         = btrfs_create,
7347         .unlink         = btrfs_unlink,
7348         .link           = btrfs_link,
7349         .mkdir          = btrfs_mkdir,
7350         .rmdir          = btrfs_rmdir,
7351         .rename         = btrfs_rename,
7352         .symlink        = btrfs_symlink,
7353         .setattr        = btrfs_setattr,
7354         .mknod          = btrfs_mknod,
7355         .setxattr       = btrfs_setxattr,
7356         .getxattr       = btrfs_getxattr,
7357         .listxattr      = btrfs_listxattr,
7358         .removexattr    = btrfs_removexattr,
7359         .permission     = btrfs_permission,
7360 };
7361 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7362         .lookup         = btrfs_lookup,
7363         .permission     = btrfs_permission,
7364 };
7365
7366 static const struct file_operations btrfs_dir_file_operations = {
7367         .llseek         = generic_file_llseek,
7368         .read           = generic_read_dir,
7369         .readdir        = btrfs_real_readdir,
7370         .unlocked_ioctl = btrfs_ioctl,
7371 #ifdef CONFIG_COMPAT
7372         .compat_ioctl   = btrfs_ioctl,
7373 #endif
7374         .release        = btrfs_release_file,
7375         .fsync          = btrfs_sync_file,
7376 };
7377
7378 static struct extent_io_ops btrfs_extent_io_ops = {
7379         .fill_delalloc = run_delalloc_range,
7380         .submit_bio_hook = btrfs_submit_bio_hook,
7381         .merge_bio_hook = btrfs_merge_bio_hook,
7382         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7383         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7384         .writepage_start_hook = btrfs_writepage_start_hook,
7385         .readpage_io_failed_hook = btrfs_io_failed_hook,
7386         .set_bit_hook = btrfs_set_bit_hook,
7387         .clear_bit_hook = btrfs_clear_bit_hook,
7388         .merge_extent_hook = btrfs_merge_extent_hook,
7389         .split_extent_hook = btrfs_split_extent_hook,
7390 };
7391
7392 /*
7393  * btrfs doesn't support the bmap operation because swapfiles
7394  * use bmap to make a mapping of extents in the file.  They assume
7395  * these extents won't change over the life of the file and they
7396  * use the bmap result to do IO directly to the drive.
7397  *
7398  * the btrfs bmap call would return logical addresses that aren't
7399  * suitable for IO and they also will change frequently as COW
7400  * operations happen.  So, swapfile + btrfs == corruption.
7401  *
7402  * For now we're avoiding this by dropping bmap.
7403  */
7404 static const struct address_space_operations btrfs_aops = {
7405         .readpage       = btrfs_readpage,
7406         .writepage      = btrfs_writepage,
7407         .writepages     = btrfs_writepages,
7408         .readpages      = btrfs_readpages,
7409         .direct_IO      = btrfs_direct_IO,
7410         .invalidatepage = btrfs_invalidatepage,
7411         .releasepage    = btrfs_releasepage,
7412         .set_page_dirty = btrfs_set_page_dirty,
7413         .error_remove_page = generic_error_remove_page,
7414 };
7415
7416 static const struct address_space_operations btrfs_symlink_aops = {
7417         .readpage       = btrfs_readpage,
7418         .writepage      = btrfs_writepage,
7419         .invalidatepage = btrfs_invalidatepage,
7420         .releasepage    = btrfs_releasepage,
7421 };
7422
7423 static const struct inode_operations btrfs_file_inode_operations = {
7424         .getattr        = btrfs_getattr,
7425         .setattr        = btrfs_setattr,
7426         .setxattr       = btrfs_setxattr,
7427         .getxattr       = btrfs_getxattr,
7428         .listxattr      = btrfs_listxattr,
7429         .removexattr    = btrfs_removexattr,
7430         .permission     = btrfs_permission,
7431         .fiemap         = btrfs_fiemap,
7432 };
7433 static const struct inode_operations btrfs_special_inode_operations = {
7434         .getattr        = btrfs_getattr,
7435         .setattr        = btrfs_setattr,
7436         .permission     = btrfs_permission,
7437         .setxattr       = btrfs_setxattr,
7438         .getxattr       = btrfs_getxattr,
7439         .listxattr      = btrfs_listxattr,
7440         .removexattr    = btrfs_removexattr,
7441 };
7442 static const struct inode_operations btrfs_symlink_inode_operations = {
7443         .readlink       = generic_readlink,
7444         .follow_link    = page_follow_link_light,
7445         .put_link       = page_put_link,
7446         .getattr        = btrfs_getattr,
7447         .permission     = btrfs_permission,
7448         .setxattr       = btrfs_setxattr,
7449         .getxattr       = btrfs_getxattr,
7450         .listxattr      = btrfs_listxattr,
7451         .removexattr    = btrfs_removexattr,
7452 };
7453
7454 const struct dentry_operations btrfs_dentry_operations = {
7455         .d_delete       = btrfs_dentry_delete,
7456 };