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