fs/btrfs/inode.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/pagemap.h>
  25 #include <linux/highmem.h>
  26 #include <linux/time.h>
  27 #include <linux/init.h>
  28 #include <linux/string.h>
  29 #include <linux/backing-dev.h>
  30 #include <linux/mpage.h>
  31 #include <linux/swap.h>
  32 #include <linux/writeback.h>
  33 #include <linux/statfs.h>
  34 #include <linux/compat.h>
  35 #include <linux/bit_spinlock.h>
  36 #include <linux/xattr.h>
  37 #include <linux/posix_acl.h>
  38 #include <linux/falloc.h>
  39 #include <linux/slab.h>
  40 #include "compat.h"
  41 #include "ctree.h"
  42 #include "disk-io.h"
  43 #include "transaction.h"
  44 #include "btrfs_inode.h"
  45 #include "ioctl.h"
  46 #include "print-tree.h"
  47 #include "volumes.h"
  48 #include "ordered-data.h"
  49 #include "xattr.h"
  50 #include "tree-log.h"
  51 #include "compression.h"
  52 #include "locking.h"
  53 #include "free-space-cache.h"
  54
  55 struct btrfs_iget_args {
  56         u64 ino;
  57         struct btrfs_root *root;
  58 };
  59
  60 static const struct inode_operations btrfs_dir_inode_operations;
  61 static const struct inode_operations btrfs_symlink_inode_operations;
  62 static const struct inode_operations btrfs_dir_ro_inode_operations;
  63 static const struct inode_operations btrfs_special_inode_operations;
  64 static const struct inode_operations btrfs_file_inode_operations;
  65 static const struct address_space_operations btrfs_aops;
  66 static const struct address_space_operations btrfs_symlink_aops;
  67 static const struct file_operations btrfs_dir_file_operations;
  68 static struct extent_io_ops btrfs_extent_io_ops;
  69
  70 static struct kmem_cache *btrfs_inode_cachep;
  71 struct kmem_cache *btrfs_trans_handle_cachep;
  72 struct kmem_cache *btrfs_transaction_cachep;
  73 struct kmem_cache *btrfs_path_cachep;
  74 struct kmem_cache *btrfs_free_space_cachep;
  75
  76 #define S_SHIFT 12
  77 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  78         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  79         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  80         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  81         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  82         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  83         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  84         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  85 };
  86
  87 static int btrfs_setsize(struct inode *inode, loff_t newsize);
  88 static int btrfs_truncate(struct inode *inode);
  89 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
  90 static noinline int cow_file_range(struct inode *inode,
  91                                    struct page *locked_page,
  92                                    u64 start, u64 end, int *page_started,
  93                                    unsigned long *nr_written, int unlock);
  94
  95 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
  96                                      struct inode *inode,  struct inode *dir,
  97                                      const struct qstr *qstr)
  98 {
  99         int err;
 100
 101         err = btrfs_init_acl(trans, inode, dir);
 102         if (!err)
 103                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 104         return err;
 105 }
 106
 107 /*
 108  * this does all the hard work for inserting an inline extent into
 109  * the btree.  The caller should have done a btrfs_drop_extents so that
 110  * no overlapping inline items exist in the btree
 111  */
 112 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 113                                 struct btrfs_root *root, struct inode *inode,
 114                                 u64 start, size_t size, size_t compressed_size,
 115                                 struct page **compressed_pages)
 116 {
 117         struct btrfs_key key;
 118         struct btrfs_path *path;
 119         struct extent_buffer *leaf;
 120         struct page *page = NULL;
 121         char *kaddr;
 122         unsigned long ptr;
 123         struct btrfs_file_extent_item *ei;
 124         int err = 0;
 125         int ret;
 126         size_t cur_size = size;
 127         size_t datasize;
 128         unsigned long offset;
 129         int compress_type = BTRFS_COMPRESS_NONE;
 130
 131         if (compressed_size && compressed_pages) {
 132                 compress_type = root->fs_info->compress_type;
 133                 cur_size = compressed_size;
 134         }
 135
 136         path = btrfs_alloc_path();
 137         if (!path)
 138                 return -ENOMEM;
 139
 140         path->leave_spinning = 1;
 141         btrfs_set_trans_block_group(trans, inode);
 142
 143         key.objectid = inode->i_ino;
 144         key.offset = start;
 145         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 146         datasize = btrfs_file_extent_calc_inline_size(cur_size);
 147
 148         inode_add_bytes(inode, size);
 149         ret = btrfs_insert_empty_item(trans, root, path, &key,
 150                                       datasize);
 151         BUG_ON(ret);
 152         if (ret) {
 153                 err = ret;
 154                 goto fail;
 155         }
 156         leaf = path->nodes[0];
 157         ei = btrfs_item_ptr(leaf, path->slots[0],
 158                             struct btrfs_file_extent_item);
 159         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 160         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 161         btrfs_set_file_extent_encryption(leaf, ei, 0);
 162         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 163         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 164         ptr = btrfs_file_extent_inline_start(ei);
 165
 166         if (compress_type != BTRFS_COMPRESS_NONE) {
 167                 struct page *cpage;
 168                 int i = 0;
 169                 while (compressed_size > 0) {
 170                         cpage = compressed_pages[i];
 171                         cur_size = min_t(unsigned long, compressed_size,
 172                                        PAGE_CACHE_SIZE);
 173
 174                         kaddr = kmap_atomic(cpage, KM_USER0);
 175                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 176                         kunmap_atomic(kaddr, KM_USER0);
 177
 178                         i++;
 179                         ptr += cur_size;
 180                         compressed_size -= cur_size;
 181                 }
 182                 btrfs_set_file_extent_compression(leaf, ei,
 183                                                   compress_type);
 184         } else {
 185                 page = find_get_page(inode->i_mapping,
 186                                      start >> PAGE_CACHE_SHIFT);
 187                 btrfs_set_file_extent_compression(leaf, ei, 0);
 188                 kaddr = kmap_atomic(page, KM_USER0);
 189                 offset = start & (PAGE_CACHE_SIZE - 1);
 190                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 191                 kunmap_atomic(kaddr, KM_USER0);
 192                 page_cache_release(page);
 193         }
 194         btrfs_mark_buffer_dirty(leaf);
 195         btrfs_free_path(path);
 196
 197         /*
 198          * we're an inline extent, so nobody can
 199          * extend the file past i_size without locking
 200          * a page we already have locked.
 201          *
 202          * We must do any isize and inode updates
 203          * before we unlock the pages.  Otherwise we
 204          * could end up racing with unlink.
 205          */
 206         BTRFS_I(inode)->disk_i_size = inode->i_size;
 207         btrfs_update_inode(trans, root, inode);
 208
 209         return 0;
 210 fail:
 211         btrfs_free_path(path);
 212         return err;
 213 }
 214
 215
 216 /*
 217  * conditionally insert an inline extent into the file.  This
 218  * does the checks required to make sure the data is small enough
 219  * to fit as an inline extent.
 220  */
 221 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 222                                  struct btrfs_root *root,
 223                                  struct inode *inode, u64 start, u64 end,
 224                                  size_t compressed_size,
 225                                  struct page **compressed_pages)
 226 {
 227         u64 isize = i_size_read(inode);
 228         u64 actual_end = min(end + 1, isize);
 229         u64 inline_len = actual_end - start;
 230         u64 aligned_end = (end + root->sectorsize - 1) &
 231                         ~((u64)root->sectorsize - 1);
 232         u64 hint_byte;
 233         u64 data_len = inline_len;
 234         int ret;
 235
 236         if (compressed_size)
 237                 data_len = compressed_size;
 238
 239         if (start > 0 ||
 240             actual_end >= PAGE_CACHE_SIZE ||
 241             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 242             (!compressed_size &&
 243             (actual_end & (root->sectorsize - 1)) == 0) ||
 244             end + 1 < isize ||
 245             data_len > root->fs_info->max_inline) {
 246                 return 1;
 247         }
 248
 249         ret = btrfs_drop_extents(trans, inode, start, aligned_end,
 250                                  &hint_byte, 1);
 251         BUG_ON(ret);
 252
 253         if (isize > actual_end)
 254                 inline_len = min_t(u64, isize, actual_end);
 255         ret = insert_inline_extent(trans, root, inode, start,
 256                                    inline_len, compressed_size,
 257                                    compressed_pages);
 258         BUG_ON(ret);
 259         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 260         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 261         return 0;
 262 }
 263
 264 struct async_extent {
 265         u64 start;
 266         u64 ram_size;
 267         u64 compressed_size;
 268         struct page **pages;
 269         unsigned long nr_pages;
 270         int compress_type;
 271         struct list_head list;
 272 };
 273
 274 struct async_cow {
 275         struct inode *inode;
 276         struct btrfs_root *root;
 277         struct page *locked_page;
 278         u64 start;
 279         u64 end;
 280         struct list_head extents;
 281         struct btrfs_work work;
 282 };
 283
 284 static noinline int add_async_extent(struct async_cow *cow,
 285                                      u64 start, u64 ram_size,
 286                                      u64 compressed_size,
 287                                      struct page **pages,
 288                                      unsigned long nr_pages,
 289                                      int compress_type)
 290 {
 291         struct async_extent *async_extent;
 292
 293         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 294         BUG_ON(!async_extent);
 295         async_extent->start = start;
 296         async_extent->ram_size = ram_size;
 297         async_extent->compressed_size = compressed_size;
 298         async_extent->pages = pages;
 299         async_extent->nr_pages = nr_pages;
 300         async_extent->compress_type = compress_type;
 301         list_add_tail(&async_extent->list, &cow->extents);
 302         return 0;
 303 }
 304
 305 /*
 306  * we create compressed extents in two phases.  The first
 307  * phase compresses a range of pages that have already been
 308  * locked (both pages and state bits are locked).
 309  *
 310  * This is done inside an ordered work queue, and the compression
 311  * is spread across many cpus.  The actual IO submission is step
 312  * two, and the ordered work queue takes care of making sure that
 313  * happens in the same order things were put onto the queue by
 314  * writepages and friends.
 315  *
 316  * If this code finds it can't get good compression, it puts an
 317  * entry onto the work queue to write the uncompressed bytes.  This
 318  * makes sure that both compressed inodes and uncompressed inodes
 319  * are written in the same order that pdflush sent them down.
 320  */
 321 static noinline int compress_file_range(struct inode *inode,
 322                                         struct page *locked_page,
 323                                         u64 start, u64 end,
 324                                         struct async_cow *async_cow,
 325                                         int *num_added)
 326 {
 327         struct btrfs_root *root = BTRFS_I(inode)->root;
 328         struct btrfs_trans_handle *trans;
 329         u64 num_bytes;
 330         u64 blocksize = root->sectorsize;
 331         u64 actual_end;
 332         u64 isize = i_size_read(inode);
 333         int ret = 0;
 334         struct page **pages = NULL;
 335         unsigned long nr_pages;
 336         unsigned long nr_pages_ret = 0;
 337         unsigned long total_compressed = 0;
 338         unsigned long total_in = 0;
 339         unsigned long max_compressed = 128 * 1024;
 340         unsigned long max_uncompressed = 128 * 1024;
 341         int i;
 342         int will_compress;
 343         int compress_type = root->fs_info->compress_type;
 344
 345         actual_end = min_t(u64, isize, end + 1);
 346 again:
 347         will_compress = 0;
 348         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 349         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 350
 351         /*
 352          * we don't want to send crud past the end of i_size through
 353          * compression, that's just a waste of CPU time.  So, if the
 354          * end of the file is before the start of our current
 355          * requested range of bytes, we bail out to the uncompressed
 356          * cleanup code that can deal with all of this.
 357          *
 358          * It isn't really the fastest way to fix things, but this is a
 359          * very uncommon corner.
 360          */
 361         if (actual_end <= start)
 362                 goto cleanup_and_bail_uncompressed;
 363
 364         total_compressed = actual_end - start;
 365
 366         /* we want to make sure that amount of ram required to uncompress
 367          * an extent is reasonable, so we limit the total size in ram
 368          * of a compressed extent to 128k.  This is a crucial number
 369          * because it also controls how easily we can spread reads across
 370          * cpus for decompression.
 371          *
 372          * We also want to make sure the amount of IO required to do
 373          * a random read is reasonably small, so we limit the size of
 374          * a compressed extent to 128k.
 375          */
 376         total_compressed = min(total_compressed, max_uncompressed);
 377         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 378         num_bytes = max(blocksize,  num_bytes);
 379         total_in = 0;
 380         ret = 0;
 381
 382         /*
 383          * we do compression for mount -o compress and when the
 384          * inode has not been flagged as nocompress.  This flag can
 385          * change at any time if we discover bad compression ratios.
 386          */
 387         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 388             (btrfs_test_opt(root, COMPRESS) ||
 389              (BTRFS_I(inode)->force_compress) ||
 390              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 391                 WARN_ON(pages);
 392                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 393                 BUG_ON(!pages);
 394
 395                 if (BTRFS_I(inode)->force_compress)
 396                         compress_type = BTRFS_I(inode)->force_compress;
 397
 398                 ret = btrfs_compress_pages(compress_type,
 399                                            inode->i_mapping, start,
 400                                            total_compressed, pages,
 401                                            nr_pages, &nr_pages_ret,
 402                                            &total_in,
 403                                            &total_compressed,
 404                                            max_compressed);
 405
 406                 if (!ret) {
 407                         unsigned long offset = total_compressed &
 408                                 (PAGE_CACHE_SIZE - 1);
 409                         struct page *page = pages[nr_pages_ret - 1];
 410                         char *kaddr;
 411
 412                         /* zero the tail end of the last page, we might be
 413                          * sending it down to disk
 414                          */
 415                         if (offset) {
 416                                 kaddr = kmap_atomic(page, KM_USER0);
 417                                 memset(kaddr + offset, 0,
 418                                        PAGE_CACHE_SIZE - offset);
 419                                 kunmap_atomic(kaddr, KM_USER0);
 420                         }
 421                         will_compress = 1;
 422                 }
 423         }
 424         if (start == 0) {
 425                 trans = btrfs_join_transaction(root, 1);
 426                 BUG_ON(IS_ERR(trans));
 427                 btrfs_set_trans_block_group(trans, inode);
 428                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 429
 430                 /* lets try to make an inline extent */
 431                 if (ret || total_in < (actual_end - start)) {
 432                         /* we didn't compress the entire range, try
 433                          * to make an uncompressed inline extent.
 434                          */
 435                         ret = cow_file_range_inline(trans, root, inode,
 436                                                     start, end, 0, NULL);
 437                 } else {
 438                         /* try making a compressed inline extent */
 439                         ret = cow_file_range_inline(trans, root, inode,
 440                                                     start, end,
 441                                                     total_compressed, pages);
 442                 }
 443                 if (ret == 0) {
 444                         /*
 445                          * inline extent creation worked, we don't need
 446                          * to create any more async work items.  Unlock
 447                          * and free up our temp pages.
 448                          */
 449                         extent_clear_unlock_delalloc(inode,
 450                              &BTRFS_I(inode)->io_tree,
 451                              start, end, NULL,
 452                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
 453                              EXTENT_CLEAR_DELALLOC |
 454                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
 455
 456                         btrfs_end_transaction(trans, root);
 457                         goto free_pages_out;
 458                 }
 459                 btrfs_end_transaction(trans, root);
 460         }
 461
 462         if (will_compress) {
 463                 /*
 464                  * we aren't doing an inline extent round the compressed size
 465                  * up to a block size boundary so the allocator does sane
 466                  * things
 467                  */
 468                 total_compressed = (total_compressed + blocksize - 1) &
 469                         ~(blocksize - 1);
 470
 471                 /*
 472                  * one last check to make sure the compression is really a
 473                  * win, compare the page count read with the blocks on disk
 474                  */
 475                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
 476                         ~(PAGE_CACHE_SIZE - 1);
 477                 if (total_compressed >= total_in) {
 478                         will_compress = 0;
 479                 } else {
 480                         num_bytes = total_in;
 481                 }
 482         }
 483         if (!will_compress && pages) {
 484                 /*
 485                  * the compression code ran but failed to make things smaller,
 486                  * free any pages it allocated and our page pointer array
 487                  */
 488                 for (i = 0; i < nr_pages_ret; i++) {
 489                         WARN_ON(pages[i]->mapping);
 490                         page_cache_release(pages[i]);
 491                 }
 492                 kfree(pages);
 493                 pages = NULL;
 494                 total_compressed = 0;
 495                 nr_pages_ret = 0;
 496
 497                 /* flag the file so we don't compress in the future */
 498                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 499                     !(BTRFS_I(inode)->force_compress)) {
 500                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 501                 }
 502         }
 503         if (will_compress) {
 504                 *num_added += 1;
 505
 506                 /* the async work queues will take care of doing actual
 507                  * allocation on disk for these compressed pages,
 508                  * and will submit them to the elevator.
 509                  */
 510                 add_async_extent(async_cow, start, num_bytes,
 511                                  total_compressed, pages, nr_pages_ret,
 512                                  compress_type);
 513
 514                 if (start + num_bytes < end) {
 515                         start += num_bytes;
 516                         pages = NULL;
 517                         cond_resched();
 518                         goto again;
 519                 }
 520         } else {
 521 cleanup_and_bail_uncompressed:
 522                 /*
 523                  * No compression, but we still need to write the pages in
 524                  * the file we've been given so far.  redirty the locked
 525                  * page if it corresponds to our extent and set things up
 526                  * for the async work queue to run cow_file_range to do
 527                  * the normal delalloc dance
 528                  */
 529                 if (page_offset(locked_page) >= start &&
 530                     page_offset(locked_page) <= end) {
 531                         __set_page_dirty_nobuffers(locked_page);
 532                         /* unlocked later on in the async handlers */
 533                 }
 534                 add_async_extent(async_cow, start, end - start + 1,
 535                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 536                 *num_added += 1;
 537         }
 538
 539 out:
 540         return 0;
 541
 542 free_pages_out:
 543         for (i = 0; i < nr_pages_ret; i++) {
 544                 WARN_ON(pages[i]->mapping);
 545                 page_cache_release(pages[i]);
 546         }
 547         kfree(pages);
 548
 549         goto out;
 550 }
 551
 552 /*
 553  * phase two of compressed writeback.  This is the ordered portion
 554  * of the code, which only gets called in the order the work was
 555  * queued.  We walk all the async extents created by compress_file_range
 556  * and send them down to the disk.
 557  */
 558 static noinline int submit_compressed_extents(struct inode *inode,
 559                                               struct async_cow *async_cow)
 560 {
 561         struct async_extent *async_extent;
 562         u64 alloc_hint = 0;
 563         struct btrfs_trans_handle *trans;
 564         struct btrfs_key ins;
 565         struct extent_map *em;
 566         struct btrfs_root *root = BTRFS_I(inode)->root;
 567         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 568         struct extent_io_tree *io_tree;
 569         int ret = 0;
 570
 571         if (list_empty(&async_cow->extents))
 572                 return 0;
 573
 574
 575         while (!list_empty(&async_cow->extents)) {
 576                 async_extent = list_entry(async_cow->extents.next,
 577                                           struct async_extent, list);
 578                 list_del(&async_extent->list);
 579
 580                 io_tree = &BTRFS_I(inode)->io_tree;
 581
 582 retry:
 583                 /* did the compression code fall back to uncompressed IO? */
 584                 if (!async_extent->pages) {
 585                         int page_started = 0;
 586                         unsigned long nr_written = 0;
 587
 588                         lock_extent(io_tree, async_extent->start,
 589                                          async_extent->start +
 590                                          async_extent->ram_size - 1, GFP_NOFS);
 591
 592                         /* allocate blocks */
 593                         ret = cow_file_range(inode, async_cow->locked_page,
 594                                              async_extent->start,
 595                                              async_extent->start +
 596                                              async_extent->ram_size - 1,
 597                                              &page_started, &nr_written, 0);
 598
 599                         /*
 600                          * if page_started, cow_file_range inserted an
 601                          * inline extent and took care of all the unlocking
 602                          * and IO for us.  Otherwise, we need to submit
 603                          * all those pages down to the drive.
 604                          */
 605                         if (!page_started && !ret)
 606                                 extent_write_locked_range(io_tree,
 607                                                   inode, async_extent->start,
 608                                                   async_extent->start +
 609                                                   async_extent->ram_size - 1,
 610                                                   btrfs_get_extent,
 611                                                   WB_SYNC_ALL);
 612                         kfree(async_extent);
 613                         cond_resched();
 614                         continue;
 615                 }
 616
 617                 lock_extent(io_tree, async_extent->start,
 618                             async_extent->start + async_extent->ram_size - 1,
 619                             GFP_NOFS);
 620
 621                 trans = btrfs_join_transaction(root, 1);
 622                 BUG_ON(IS_ERR(trans));
 623                 ret = btrfs_reserve_extent(trans, root,
 624                                            async_extent->compressed_size,
 625                                            async_extent->compressed_size,
 626                                            0, alloc_hint,
 627                                            (u64)-1, &ins, 1);
 628                 btrfs_end_transaction(trans, root);
 629
 630                 if (ret) {
 631                         int i;
 632                         for (i = 0; i < async_extent->nr_pages; i++) {
 633                                 WARN_ON(async_extent->pages[i]->mapping);
 634                                 page_cache_release(async_extent->pages[i]);
 635                         }
 636                         kfree(async_extent->pages);
 637                         async_extent->nr_pages = 0;
 638                         async_extent->pages = NULL;
 639                         unlock_extent(io_tree, async_extent->start,
 640                                       async_extent->start +
 641                                       async_extent->ram_size - 1, GFP_NOFS);
 642                         goto retry;
 643                 }
 644
 645                 /*
 646                  * here we're doing allocation and writeback of the
 647                  * compressed pages
 648                  */
 649                 btrfs_drop_extent_cache(inode, async_extent->start,
 650                                         async_extent->start +
 651                                         async_extent->ram_size - 1, 0);
 652
 653                 em = alloc_extent_map(GFP_NOFS);
 654                 BUG_ON(!em);
 655                 em->start = async_extent->start;
 656                 em->len = async_extent->ram_size;
 657                 em->orig_start = em->start;
 658
 659                 em->block_start = ins.objectid;
 660                 em->block_len = ins.offset;
 661                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 662                 em->compress_type = async_extent->compress_type;
 663                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 664                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 665
 666                 while (1) {
 667                         write_lock(&em_tree->lock);
 668                         ret = add_extent_mapping(em_tree, em);
 669                         write_unlock(&em_tree->lock);
 670                         if (ret != -EEXIST) {
 671                                 free_extent_map(em);
 672                                 break;
 673                         }
 674                         btrfs_drop_extent_cache(inode, async_extent->start,
 675                                                 async_extent->start +
 676                                                 async_extent->ram_size - 1, 0);
 677                 }
 678
 679                 ret = btrfs_add_ordered_extent_compress(inode,
 680                                                 async_extent->start,
 681                                                 ins.objectid,
 682                                                 async_extent->ram_size,
 683                                                 ins.offset,
 684                                                 BTRFS_ORDERED_COMPRESSED,
 685                                                 async_extent->compress_type);
 686                 BUG_ON(ret);
 687
 688                 /*
 689                  * clear dirty, set writeback and unlock the pages.
 690                  */
 691                 extent_clear_unlock_delalloc(inode,
 692                                 &BTRFS_I(inode)->io_tree,
 693                                 async_extent->start,
 694                                 async_extent->start +
 695                                 async_extent->ram_size - 1,
 696                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 697                                 EXTENT_CLEAR_UNLOCK |
 698                                 EXTENT_CLEAR_DELALLOC |
 699                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
 700
 701                 ret = btrfs_submit_compressed_write(inode,
 702                                     async_extent->start,
 703                                     async_extent->ram_size,
 704                                     ins.objectid,
 705                                     ins.offset, async_extent->pages,
 706                                     async_extent->nr_pages);
 707
 708                 BUG_ON(ret);
 709                 alloc_hint = ins.objectid + ins.offset;
 710                 kfree(async_extent);
 711                 cond_resched();
 712         }
 713
 714         return 0;
 715 }
 716
 717 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 718                                       u64 num_bytes)
 719 {
 720         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 721         struct extent_map *em;
 722         u64 alloc_hint = 0;
 723
 724         read_lock(&em_tree->lock);
 725         em = search_extent_mapping(em_tree, start, num_bytes);
 726         if (em) {
 727                 /*
 728                  * if block start isn't an actual block number then find the
 729                  * first block in this inode and use that as a hint.  If that
 730                  * block is also bogus then just don't worry about it.
 731                  */
 732                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 733                         free_extent_map(em);
 734                         em = search_extent_mapping(em_tree, 0, 0);
 735                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 736                                 alloc_hint = em->block_start;
 737                         if (em)
 738                                 free_extent_map(em);
 739                 } else {
 740                         alloc_hint = em->block_start;
 741                         free_extent_map(em);
 742                 }
 743         }
 744         read_unlock(&em_tree->lock);
 745
 746         return alloc_hint;
 747 }
 748
 749 /*
 750  * when extent_io.c finds a delayed allocation range in the file,
 751  * the call backs end up in this code.  The basic idea is to
 752  * allocate extents on disk for the range, and create ordered data structs
 753  * in ram to track those extents.
 754  *
 755  * locked_page is the page that writepage had locked already.  We use
 756  * it to make sure we don't do extra locks or unlocks.
 757  *
 758  * *page_started is set to one if we unlock locked_page and do everything
 759  * required to start IO on it.  It may be clean and already done with
 760  * IO when we return.
 761  */
 762 static noinline int cow_file_range(struct inode *inode,
 763                                    struct page *locked_page,
 764                                    u64 start, u64 end, int *page_started,
 765                                    unsigned long *nr_written,
 766                                    int unlock)
 767 {
 768         struct btrfs_root *root = BTRFS_I(inode)->root;
 769         struct btrfs_trans_handle *trans;
 770         u64 alloc_hint = 0;
 771         u64 num_bytes;
 772         unsigned long ram_size;
 773         u64 disk_num_bytes;
 774         u64 cur_alloc_size;
 775         u64 blocksize = root->sectorsize;
 776         struct btrfs_key ins;
 777         struct extent_map *em;
 778         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 779         int ret = 0;
 780
 781         BUG_ON(root == root->fs_info->tree_root);
 782         trans = btrfs_join_transaction(root, 1);
 783         BUG_ON(IS_ERR(trans));
 784         btrfs_set_trans_block_group(trans, inode);
 785         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 786
 787         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 788         num_bytes = max(blocksize,  num_bytes);
 789         disk_num_bytes = num_bytes;
 790         ret = 0;
 791
 792         if (start == 0) {
 793                 /* lets try to make an inline extent */
 794                 ret = cow_file_range_inline(trans, root, inode,
 795                                             start, end, 0, NULL);
 796                 if (ret == 0) {
 797                         extent_clear_unlock_delalloc(inode,
 798                                      &BTRFS_I(inode)->io_tree,
 799                                      start, end, NULL,
 800                                      EXTENT_CLEAR_UNLOCK_PAGE |
 801                                      EXTENT_CLEAR_UNLOCK |
 802                                      EXTENT_CLEAR_DELALLOC |
 803                                      EXTENT_CLEAR_DIRTY |
 804                                      EXTENT_SET_WRITEBACK |
 805                                      EXTENT_END_WRITEBACK);
 806
 807                         *nr_written = *nr_written +
 808                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 809                         *page_started = 1;
 810                         ret = 0;
 811                         goto out;
 812                 }
 813         }
 814
 815         BUG_ON(disk_num_bytes >
 816                btrfs_super_total_bytes(&root->fs_info->super_copy));
 817
 818         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 819         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 820
 821         while (disk_num_bytes > 0) {
 822                 unsigned long op;
 823
 824                 cur_alloc_size = disk_num_bytes;
 825                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 826                                            root->sectorsize, 0, alloc_hint,
 827                                            (u64)-1, &ins, 1);
 828                 BUG_ON(ret);
 829
 830                 em = alloc_extent_map(GFP_NOFS);
 831                 BUG_ON(!em);
 832                 em->start = start;
 833                 em->orig_start = em->start;
 834                 ram_size = ins.offset;
 835                 em->len = ins.offset;
 836
 837                 em->block_start = ins.objectid;
 838                 em->block_len = ins.offset;
 839                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 840                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 841
 842                 while (1) {
 843                         write_lock(&em_tree->lock);
 844                         ret = add_extent_mapping(em_tree, em);
 845                         write_unlock(&em_tree->lock);
 846                         if (ret != -EEXIST) {
 847                                 free_extent_map(em);
 848                                 break;
 849                         }
 850                         btrfs_drop_extent_cache(inode, start,
 851                                                 start + ram_size - 1, 0);
 852                 }
 853
 854                 cur_alloc_size = ins.offset;
 855                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 856                                                ram_size, cur_alloc_size, 0);
 857                 BUG_ON(ret);
 858
 859                 if (root->root_key.objectid ==
 860                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 861                         ret = btrfs_reloc_clone_csums(inode, start,
 862                                                       cur_alloc_size);
 863                         BUG_ON(ret);
 864                 }
 865
 866                 if (disk_num_bytes < cur_alloc_size)
 867                         break;
 868
 869                 /* we're not doing compressed IO, don't unlock the first
 870                  * page (which the caller expects to stay locked), don't
 871                  * clear any dirty bits and don't set any writeback bits
 872                  *
 873                  * Do set the Private2 bit so we know this page was properly
 874                  * setup for writepage
 875                  */
 876                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
 877                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
 878                         EXTENT_SET_PRIVATE2;
 879
 880                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 881                                              start, start + ram_size - 1,
 882                                              locked_page, op);
 883                 disk_num_bytes -= cur_alloc_size;
 884                 num_bytes -= cur_alloc_size;
 885                 alloc_hint = ins.objectid + ins.offset;
 886                 start += cur_alloc_size;
 887         }
 888 out:
 889         ret = 0;
 890         btrfs_end_transaction(trans, root);
 891
 892         return ret;
 893 }
 894
 895 /*
 896  * work queue call back to started compression on a file and pages
 897  */
 898 static noinline void async_cow_start(struct btrfs_work *work)
 899 {
 900         struct async_cow *async_cow;
 901         int num_added = 0;
 902         async_cow = container_of(work, struct async_cow, work);
 903
 904         compress_file_range(async_cow->inode, async_cow->locked_page,
 905                             async_cow->start, async_cow->end, async_cow,
 906                             &num_added);
 907         if (num_added == 0)
 908                 async_cow->inode = NULL;
 909 }
 910
 911 /*
 912  * work queue call back to submit previously compressed pages
 913  */
 914 static noinline void async_cow_submit(struct btrfs_work *work)
 915 {
 916         struct async_cow *async_cow;
 917         struct btrfs_root *root;
 918         unsigned long nr_pages;
 919
 920         async_cow = container_of(work, struct async_cow, work);
 921
 922         root = async_cow->root;
 923         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
 924                 PAGE_CACHE_SHIFT;
 925
 926         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
 927
 928         if (atomic_read(&root->fs_info->async_delalloc_pages) <
 929             5 * 1042 * 1024 &&
 930             waitqueue_active(&root->fs_info->async_submit_wait))
 931                 wake_up(&root->fs_info->async_submit_wait);
 932
 933         if (async_cow->inode)
 934                 submit_compressed_extents(async_cow->inode, async_cow);
 935 }
 936
 937 static noinline void async_cow_free(struct btrfs_work *work)
 938 {
 939         struct async_cow *async_cow;
 940         async_cow = container_of(work, struct async_cow, work);
 941         kfree(async_cow);
 942 }
 943
 944 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
 945                                 u64 start, u64 end, int *page_started,
 946                                 unsigned long *nr_written)
 947 {
 948         struct async_cow *async_cow;
 949         struct btrfs_root *root = BTRFS_I(inode)->root;
 950         unsigned long nr_pages;
 951         u64 cur_end;
 952         int limit = 10 * 1024 * 1042;
 953
 954         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
 955                          1, 0, NULL, GFP_NOFS);
 956         while (start < end) {
 957                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
 958                 async_cow->inode = inode;
 959                 async_cow->root = root;
 960                 async_cow->locked_page = locked_page;
 961                 async_cow->start = start;
 962
 963                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
 964                         cur_end = end;
 965                 else
 966                         cur_end = min(end, start + 512 * 1024 - 1);
 967
 968                 async_cow->end = cur_end;
 969                 INIT_LIST_HEAD(&async_cow->extents);
 970
 971                 async_cow->work.func = async_cow_start;
 972                 async_cow->work.ordered_func = async_cow_submit;
 973                 async_cow->work.ordered_free = async_cow_free;
 974                 async_cow->work.flags = 0;
 975
 976                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
 977                         PAGE_CACHE_SHIFT;
 978                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
 979
 980                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
 981                                    &async_cow->work);
 982
 983                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
 984                         wait_event(root->fs_info->async_submit_wait,
 985                            (atomic_read(&root->fs_info->async_delalloc_pages) <
 986                             limit));
 987                 }
 988
 989                 while (atomic_read(&root->fs_info->async_submit_draining) &&
 990                       atomic_read(&root->fs_info->async_delalloc_pages)) {
 991                         wait_event(root->fs_info->async_submit_wait,
 992                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
 993                            0));
 994                 }
 995
 996                 *nr_written += nr_pages;
 997                 start = cur_end + 1;
 998         }
 999         *page_started = 1;
1000         return 0;
1001 }
1002
1003 static noinline int csum_exist_in_range(struct btrfs_root *root,
1004                                         u64 bytenr, u64 num_bytes)
1005 {
1006         int ret;
1007         struct btrfs_ordered_sum *sums;
1008         LIST_HEAD(list);
1009
1010         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1011                                        bytenr + num_bytes - 1, &list);
1012         if (ret == 0 && list_empty(&list))
1013                 return 0;
1014
1015         while (!list_empty(&list)) {
1016                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1017                 list_del(&sums->list);
1018                 kfree(sums);
1019         }
1020         return 1;
1021 }
1022
1023 /*
1024  * when nowcow writeback call back.  This checks for snapshots or COW copies
1025  * of the extents that exist in the file, and COWs the file as required.
1026  *
1027  * If no cow copies or snapshots exist, we write directly to the existing
1028  * blocks on disk
1029  */
1030 static noinline int run_delalloc_nocow(struct inode *inode,
1031                                        struct page *locked_page,
1032                               u64 start, u64 end, int *page_started, int force,
1033                               unsigned long *nr_written)
1034 {
1035         struct btrfs_root *root = BTRFS_I(inode)->root;
1036         struct btrfs_trans_handle *trans;
1037         struct extent_buffer *leaf;
1038         struct btrfs_path *path;
1039         struct btrfs_file_extent_item *fi;
1040         struct btrfs_key found_key;
1041         u64 cow_start;
1042         u64 cur_offset;
1043         u64 extent_end;
1044         u64 extent_offset;
1045         u64 disk_bytenr;
1046         u64 num_bytes;
1047         int extent_type;
1048         int ret;
1049         int type;
1050         int nocow;
1051         int check_prev = 1;
1052         bool nolock = false;
1053
1054         path = btrfs_alloc_path();
1055         BUG_ON(!path);
1056         if (root == root->fs_info->tree_root) {
1057                 nolock = true;
1058                 trans = btrfs_join_transaction_nolock(root, 1);
1059         } else {
1060                 trans = btrfs_join_transaction(root, 1);
1061         }
1062         BUG_ON(IS_ERR(trans));
1063
1064         cow_start = (u64)-1;
1065         cur_offset = start;
1066         while (1) {
1067                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
1068                                                cur_offset, 0);
1069                 BUG_ON(ret < 0);
1070                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1071                         leaf = path->nodes[0];
1072                         btrfs_item_key_to_cpu(leaf, &found_key,
1073                                               path->slots[0] - 1);
1074                         if (found_key.objectid == inode->i_ino &&
1075                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1076                                 path->slots[0]--;
1077                 }
1078                 check_prev = 0;
1079 next_slot:
1080                 leaf = path->nodes[0];
1081                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1082                         ret = btrfs_next_leaf(root, path);
1083                         if (ret < 0)
1084                                 BUG_ON(1);
1085                         if (ret > 0)
1086                                 break;
1087                         leaf = path->nodes[0];
1088                 }
1089
1090                 nocow = 0;
1091                 disk_bytenr = 0;
1092                 num_bytes = 0;
1093                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1094
1095                 if (found_key.objectid > inode->i_ino ||
1096                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1097                     found_key.offset > end)
1098                         break;
1099
1100                 if (found_key.offset > cur_offset) {
1101                         extent_end = found_key.offset;
1102                         extent_type = 0;
1103                         goto out_check;
1104                 }
1105
1106                 fi = btrfs_item_ptr(leaf, path->slots[0],
1107                                     struct btrfs_file_extent_item);
1108                 extent_type = btrfs_file_extent_type(leaf, fi);
1109
1110                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1111                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1112                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1113                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1114                         extent_end = found_key.offset +
1115                                 btrfs_file_extent_num_bytes(leaf, fi);
1116                         if (extent_end <= start) {
1117                                 path->slots[0]++;
1118                                 goto next_slot;
1119                         }
1120                         if (disk_bytenr == 0)
1121                                 goto out_check;
1122                         if (btrfs_file_extent_compression(leaf, fi) ||
1123                             btrfs_file_extent_encryption(leaf, fi) ||
1124                             btrfs_file_extent_other_encoding(leaf, fi))
1125                                 goto out_check;
1126                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1127                                 goto out_check;
1128                         if (btrfs_extent_readonly(root, disk_bytenr))
1129                                 goto out_check;
1130                         if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1131                                                   found_key.offset -
1132                                                   extent_offset, disk_bytenr))
1133                                 goto out_check;
1134                         disk_bytenr += extent_offset;
1135                         disk_bytenr += cur_offset - found_key.offset;
1136                         num_bytes = min(end + 1, extent_end) - cur_offset;
1137                         /*
1138                          * force cow if csum exists in the range.
1139                          * this ensure that csum for a given extent are
1140                          * either valid or do not exist.
1141                          */
1142                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1143                                 goto out_check;
1144                         nocow = 1;
1145                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1146                         extent_end = found_key.offset +
1147                                 btrfs_file_extent_inline_len(leaf, fi);
1148                         extent_end = ALIGN(extent_end, root->sectorsize);
1149                 } else {
1150                         BUG_ON(1);
1151                 }
1152 out_check:
1153                 if (extent_end <= start) {
1154                         path->slots[0]++;
1155                         goto next_slot;
1156                 }
1157                 if (!nocow) {
1158                         if (cow_start == (u64)-1)
1159                                 cow_start = cur_offset;
1160                         cur_offset = extent_end;
1161                         if (cur_offset > end)
1162                                 break;
1163                         path->slots[0]++;
1164                         goto next_slot;
1165                 }
1166
1167                 btrfs_release_path(root, path);
1168                 if (cow_start != (u64)-1) {
1169                         ret = cow_file_range(inode, locked_page, cow_start,
1170                                         found_key.offset - 1, page_started,
1171                                         nr_written, 1);
1172                         BUG_ON(ret);
1173                         cow_start = (u64)-1;
1174                 }
1175
1176                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1177                         struct extent_map *em;
1178                         struct extent_map_tree *em_tree;
1179                         em_tree = &BTRFS_I(inode)->extent_tree;
1180                         em = alloc_extent_map(GFP_NOFS);
1181                         BUG_ON(!em);
1182                         em->start = cur_offset;
1183                         em->orig_start = em->start;
1184                         em->len = num_bytes;
1185                         em->block_len = num_bytes;
1186                         em->block_start = disk_bytenr;
1187                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1188                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1189                         while (1) {
1190                                 write_lock(&em_tree->lock);
1191                                 ret = add_extent_mapping(em_tree, em);
1192                                 write_unlock(&em_tree->lock);
1193                                 if (ret != -EEXIST) {
1194                                         free_extent_map(em);
1195                                         break;
1196                                 }
1197                                 btrfs_drop_extent_cache(inode, em->start,
1198                                                 em->start + em->len - 1, 0);
1199                         }
1200                         type = BTRFS_ORDERED_PREALLOC;
1201                 } else {
1202                         type = BTRFS_ORDERED_NOCOW;
1203                 }
1204
1205                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1206                                                num_bytes, num_bytes, type);
1207                 BUG_ON(ret);
1208
1209                 if (root->root_key.objectid ==
1210                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1211                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1212                                                       num_bytes);
1213                         BUG_ON(ret);
1214                 }
1215
1216                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1217                                 cur_offset, cur_offset + num_bytes - 1,
1218                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1219                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1220                                 EXTENT_SET_PRIVATE2);
1221                 cur_offset = extent_end;
1222                 if (cur_offset > end)
1223                         break;
1224         }
1225         btrfs_release_path(root, path);
1226
1227         if (cur_offset <= end && cow_start == (u64)-1)
1228                 cow_start = cur_offset;
1229         if (cow_start != (u64)-1) {
1230                 ret = cow_file_range(inode, locked_page, cow_start, end,
1231                                      page_started, nr_written, 1);
1232                 BUG_ON(ret);
1233         }
1234
1235         if (nolock) {
1236                 ret = btrfs_end_transaction_nolock(trans, root);
1237                 BUG_ON(ret);
1238         } else {
1239                 ret = btrfs_end_transaction(trans, root);
1240                 BUG_ON(ret);
1241         }
1242         btrfs_free_path(path);
1243         return 0;
1244 }
1245
1246 /*
1247  * extent_io.c call back to do delayed allocation processing
1248  */
1249 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1250                               u64 start, u64 end, int *page_started,
1251                               unsigned long *nr_written)
1252 {
1253         int ret;
1254         struct btrfs_root *root = BTRFS_I(inode)->root;
1255
1256         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1257                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1258                                          page_started, 1, nr_written);
1259         else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1260                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1261                                          page_started, 0, nr_written);
1262         else if (!btrfs_test_opt(root, COMPRESS) &&
1263                  !(BTRFS_I(inode)->force_compress) &&
1264                  !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1265                 ret = cow_file_range(inode, locked_page, start, end,
1266                                       page_started, nr_written, 1);
1267         else
1268                 ret = cow_file_range_async(inode, locked_page, start, end,
1269                                            page_started, nr_written);
1270         return ret;
1271 }
1272
1273 static int btrfs_split_extent_hook(struct inode *inode,
1274                                    struct extent_state *orig, u64 split)
1275 {
1276         /* not delalloc, ignore it */
1277         if (!(orig->state & EXTENT_DELALLOC))
1278                 return 0;
1279
1280         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1281         return 0;
1282 }
1283
1284 /*
1285  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1286  * extents so we can keep track of new extents that are just merged onto old
1287  * extents, such as when we are doing sequential writes, so we can properly
1288  * account for the metadata space we'll need.
1289  */
1290 static int btrfs_merge_extent_hook(struct inode *inode,
1291                                    struct extent_state *new,
1292                                    struct extent_state *other)
1293 {
1294         /* not delalloc, ignore it */
1295         if (!(other->state & EXTENT_DELALLOC))
1296                 return 0;
1297
1298         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1299         return 0;
1300 }
1301
1302 /*
1303  * extent_io.c set_bit_hook, used to track delayed allocation
1304  * bytes in this file, and to maintain the list of inodes that
1305  * have pending delalloc work to be done.
1306  */
1307 static int btrfs_set_bit_hook(struct inode *inode,
1308                               struct extent_state *state, int *bits)
1309 {
1310
1311         /*
1312          * set_bit and clear bit hooks normally require _irqsave/restore
1313          * but in this case, we are only testeing for the DELALLOC
1314          * bit, which is only set or cleared with irqs on
1315          */
1316         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1317                 struct btrfs_root *root = BTRFS_I(inode)->root;
1318                 u64 len = state->end + 1 - state->start;
1319                 int do_list = (root->root_key.objectid !=
1320                                BTRFS_ROOT_TREE_OBJECTID);
1321
1322                 if (*bits & EXTENT_FIRST_DELALLOC)
1323                         *bits &= ~EXTENT_FIRST_DELALLOC;
1324                 else
1325                         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1326
1327                 spin_lock(&root->fs_info->delalloc_lock);
1328                 BTRFS_I(inode)->delalloc_bytes += len;
1329                 root->fs_info->delalloc_bytes += len;
1330                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1331                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1332                                       &root->fs_info->delalloc_inodes);
1333                 }
1334                 spin_unlock(&root->fs_info->delalloc_lock);
1335         }
1336         return 0;
1337 }
1338
1339 /*
1340  * extent_io.c clear_bit_hook, see set_bit_hook for why
1341  */
1342 static int btrfs_clear_bit_hook(struct inode *inode,
1343                                 struct extent_state *state, int *bits)
1344 {
1345         /*
1346          * set_bit and clear bit hooks normally require _irqsave/restore
1347          * but in this case, we are only testeing for the DELALLOC
1348          * bit, which is only set or cleared with irqs on
1349          */
1350         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1351                 struct btrfs_root *root = BTRFS_I(inode)->root;
1352                 u64 len = state->end + 1 - state->start;
1353                 int do_list = (root->root_key.objectid !=
1354                                BTRFS_ROOT_TREE_OBJECTID);
1355
1356                 if (*bits & EXTENT_FIRST_DELALLOC)
1357                         *bits &= ~EXTENT_FIRST_DELALLOC;
1358                 else if (!(*bits & EXTENT_DO_ACCOUNTING))
1359                         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1360
1361                 if (*bits & EXTENT_DO_ACCOUNTING)
1362                         btrfs_delalloc_release_metadata(inode, len);
1363
1364                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1365                     && do_list)
1366                         btrfs_free_reserved_data_space(inode, len);
1367
1368                 spin_lock(&root->fs_info->delalloc_lock);
1369                 root->fs_info->delalloc_bytes -= len;
1370                 BTRFS_I(inode)->delalloc_bytes -= len;
1371
1372                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1373                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1374                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1375                 }
1376                 spin_unlock(&root->fs_info->delalloc_lock);
1377         }
1378         return 0;
1379 }
1380
1381 /*
1382  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1383  * we don't create bios that span stripes or chunks
1384  */
1385 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1386                          size_t size, struct bio *bio,
1387                          unsigned long bio_flags)
1388 {
1389         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1390         struct btrfs_mapping_tree *map_tree;
1391         u64 logical = (u64)bio->bi_sector << 9;
1392         u64 length = 0;
1393         u64 map_length;
1394         int ret;
1395
1396         if (bio_flags & EXTENT_BIO_COMPRESSED)
1397                 return 0;
1398
1399         length = bio->bi_size;
1400         map_tree = &root->fs_info->mapping_tree;
1401         map_length = length;
1402         ret = btrfs_map_block(map_tree, READ, logical,
1403                               &map_length, NULL, 0);
1404
1405         if (map_length < length + size)
1406                 return 1;
1407         return ret;
1408 }
1409
1410 /*
1411  * in order to insert checksums into the metadata in large chunks,
1412  * we wait until bio submission time.   All the pages in the bio are
1413  * checksummed and sums are attached onto the ordered extent record.
1414  *
1415  * At IO completion time the cums attached on the ordered extent record
1416  * are inserted into the btree
1417  */
1418 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1419                                     struct bio *bio, int mirror_num,
1420                                     unsigned long bio_flags,
1421                                     u64 bio_offset)
1422 {
1423         struct btrfs_root *root = BTRFS_I(inode)->root;
1424         int ret = 0;
1425
1426         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1427         BUG_ON(ret);
1428         return 0;
1429 }
1430
1431 /*
1432  * in order to insert checksums into the metadata in large chunks,
1433  * we wait until bio submission time.   All the pages in the bio are
1434  * checksummed and sums are attached onto the ordered extent record.
1435  *
1436  * At IO completion time the cums attached on the ordered extent record
1437  * are inserted into the btree
1438  */
1439 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1440                           int mirror_num, unsigned long bio_flags,
1441                           u64 bio_offset)
1442 {
1443         struct btrfs_root *root = BTRFS_I(inode)->root;
1444         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1445 }
1446
1447 /*
1448  * extent_io.c submission hook. This does the right thing for csum calculation
1449  * on write, or reading the csums from the tree before a read
1450  */
1451 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1452                           int mirror_num, unsigned long bio_flags,
1453                           u64 bio_offset)
1454 {
1455         struct btrfs_root *root = BTRFS_I(inode)->root;
1456         int ret = 0;
1457         int skip_sum;
1458
1459         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1460
1461         if (root == root->fs_info->tree_root)
1462                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1463         else
1464                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1465         BUG_ON(ret);
1466
1467         if (!(rw & REQ_WRITE)) {
1468                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1469                         return btrfs_submit_compressed_read(inode, bio,
1470                                                     mirror_num, bio_flags);
1471                 } else if (!skip_sum) {
1472                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1473                         if (ret)
1474                                 return ret;
1475                 }
1476                 goto mapit;
1477         } else if (!skip_sum) {
1478                 /* csum items have already been cloned */
1479                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1480                         goto mapit;
1481                 /* we're doing a write, do the async checksumming */
1482                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1483                                    inode, rw, bio, mirror_num,
1484                                    bio_flags, bio_offset,
1485                                    __btrfs_submit_bio_start,
1486                                    __btrfs_submit_bio_done);
1487         }
1488
1489 mapit:
1490         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1491 }
1492
1493 /*
1494  * given a list of ordered sums record them in the inode.  This happens
1495  * at IO completion time based on sums calculated at bio submission time.
1496  */
1497 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1498                              struct inode *inode, u64 file_offset,
1499                              struct list_head *list)
1500 {
1501         struct btrfs_ordered_sum *sum;
1502
1503         btrfs_set_trans_block_group(trans, inode);
1504
1505         list_for_each_entry(sum, list, list) {
1506                 btrfs_csum_file_blocks(trans,
1507                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1508         }
1509         return 0;
1510 }
1511
1512 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1513                               struct extent_state **cached_state)
1514 {
1515         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1516                 WARN_ON(1);
1517         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1518                                    cached_state, GFP_NOFS);
1519 }
1520
1521 /* see btrfs_writepage_start_hook for details on why this is required */
1522 struct btrfs_writepage_fixup {
1523         struct page *page;
1524         struct btrfs_work work;
1525 };
1526
1527 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1528 {
1529         struct btrfs_writepage_fixup *fixup;
1530         struct btrfs_ordered_extent *ordered;
1531         struct extent_state *cached_state = NULL;
1532         struct page *page;
1533         struct inode *inode;
1534         u64 page_start;
1535         u64 page_end;
1536
1537         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1538         page = fixup->page;
1539 again:
1540         lock_page(page);
1541         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1542                 ClearPageChecked(page);
1543                 goto out_page;
1544         }
1545
1546         inode = page->mapping->host;
1547         page_start = page_offset(page);
1548         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1549
1550         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1551                          &cached_state, GFP_NOFS);
1552
1553         /* already ordered? We're done */
1554         if (PagePrivate2(page))
1555                 goto out;
1556
1557         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1558         if (ordered) {
1559                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1560                                      page_end, &cached_state, GFP_NOFS);
1561                 unlock_page(page);
1562                 btrfs_start_ordered_extent(inode, ordered, 1);
1563                 goto again;
1564         }
1565
1566         BUG();
1567         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1568         ClearPageChecked(page);
1569 out:
1570         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1571                              &cached_state, GFP_NOFS);
1572 out_page:
1573         unlock_page(page);
1574         page_cache_release(page);
1575         kfree(fixup);
1576 }
1577
1578 /*
1579  * There are a few paths in the higher layers of the kernel that directly
1580  * set the page dirty bit without asking the filesystem if it is a
1581  * good idea.  This causes problems because we want to make sure COW
1582  * properly happens and the data=ordered rules are followed.
1583  *
1584  * In our case any range that doesn't have the ORDERED bit set
1585  * hasn't been properly setup for IO.  We kick off an async process
1586  * to fix it up.  The async helper will wait for ordered extents, set
1587  * the delalloc bit and make it safe to write the page.
1588  */
1589 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1590 {
1591         struct inode *inode = page->mapping->host;
1592         struct btrfs_writepage_fixup *fixup;
1593         struct btrfs_root *root = BTRFS_I(inode)->root;
1594
1595         /* this page is properly in the ordered list */
1596         if (TestClearPagePrivate2(page))
1597                 return 0;
1598
1599         if (PageChecked(page))
1600                 return -EAGAIN;
1601
1602         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1603         if (!fixup)
1604                 return -EAGAIN;
1605
1606         SetPageChecked(page);
1607         page_cache_get(page);
1608         fixup->work.func = btrfs_writepage_fixup_worker;
1609         fixup->page = page;
1610         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1611         return -EAGAIN;
1612 }
1613
1614 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1615                                        struct inode *inode, u64 file_pos,
1616                                        u64 disk_bytenr, u64 disk_num_bytes,
1617                                        u64 num_bytes, u64 ram_bytes,
1618                                        u8 compression, u8 encryption,
1619                                        u16 other_encoding, int extent_type)
1620 {
1621         struct btrfs_root *root = BTRFS_I(inode)->root;
1622         struct btrfs_file_extent_item *fi;
1623         struct btrfs_path *path;
1624         struct extent_buffer *leaf;
1625         struct btrfs_key ins;
1626         u64 hint;
1627         int ret;
1628
1629         path = btrfs_alloc_path();
1630         BUG_ON(!path);
1631
1632         path->leave_spinning = 1;
1633
1634         /*
1635          * we may be replacing one extent in the tree with another.
1636          * The new extent is pinned in the extent map, and we don't want
1637          * to drop it from the cache until it is completely in the btree.
1638          *
1639          * So, tell btrfs_drop_extents to leave this extent in the cache.
1640          * the caller is expected to unpin it and allow it to be merged
1641          * with the others.
1642          */
1643         ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1644                                  &hint, 0);
1645         BUG_ON(ret);
1646
1647         ins.objectid = inode->i_ino;
1648         ins.offset = file_pos;
1649         ins.type = BTRFS_EXTENT_DATA_KEY;
1650         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1651         BUG_ON(ret);
1652         leaf = path->nodes[0];
1653         fi = btrfs_item_ptr(leaf, path->slots[0],
1654                             struct btrfs_file_extent_item);
1655         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1656         btrfs_set_file_extent_type(leaf, fi, extent_type);
1657         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1658         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1659         btrfs_set_file_extent_offset(leaf, fi, 0);
1660         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1661         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1662         btrfs_set_file_extent_compression(leaf, fi, compression);
1663         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1664         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1665
1666         btrfs_unlock_up_safe(path, 1);
1667         btrfs_set_lock_blocking(leaf);
1668
1669         btrfs_mark_buffer_dirty(leaf);
1670
1671         inode_add_bytes(inode, num_bytes);
1672
1673         ins.objectid = disk_bytenr;
1674         ins.offset = disk_num_bytes;
1675         ins.type = BTRFS_EXTENT_ITEM_KEY;
1676         ret = btrfs_alloc_reserved_file_extent(trans, root,
1677                                         root->root_key.objectid,
1678                                         inode->i_ino, file_pos, &ins);
1679         BUG_ON(ret);
1680         btrfs_free_path(path);
1681
1682         return 0;
1683 }
1684
1685 /*
1686  * helper function for btrfs_finish_ordered_io, this
1687  * just reads in some of the csum leaves to prime them into ram
1688  * before we start the transaction.  It limits the amount of btree
1689  * reads required while inside the transaction.
1690  */
1691 /* as ordered data IO finishes, this gets called so we can finish
1692  * an ordered extent if the range of bytes in the file it covers are
1693  * fully written.
1694  */
1695 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1696 {
1697         struct btrfs_root *root = BTRFS_I(inode)->root;
1698         struct btrfs_trans_handle *trans = NULL;
1699         struct btrfs_ordered_extent *ordered_extent = NULL;
1700         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1701         struct extent_state *cached_state = NULL;
1702         int compress_type = 0;
1703         int ret;
1704         bool nolock = false;
1705
1706         ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1707                                              end - start + 1);
1708         if (!ret)
1709                 return 0;
1710         BUG_ON(!ordered_extent);
1711
1712         nolock = (root == root->fs_info->tree_root);
1713
1714         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1715                 BUG_ON(!list_empty(&ordered_extent->list));
1716                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1717                 if (!ret) {
1718                         if (nolock)
1719                                 trans = btrfs_join_transaction_nolock(root, 1);
1720                         else
1721                                 trans = btrfs_join_transaction(root, 1);
1722                         BUG_ON(IS_ERR(trans));
1723                         btrfs_set_trans_block_group(trans, inode);
1724                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1725                         ret = btrfs_update_inode(trans, root, inode);
1726                         BUG_ON(ret);
1727                 }
1728                 goto out;
1729         }
1730
1731         lock_extent_bits(io_tree, ordered_extent->file_offset,
1732                          ordered_extent->file_offset + ordered_extent->len - 1,
1733                          0, &cached_state, GFP_NOFS);
1734
1735         if (nolock)
1736                 trans = btrfs_join_transaction_nolock(root, 1);
1737         else
1738                 trans = btrfs_join_transaction(root, 1);
1739         BUG_ON(IS_ERR(trans));
1740         btrfs_set_trans_block_group(trans, inode);
1741         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1742
1743         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1744                 compress_type = ordered_extent->compress_type;
1745         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1746                 BUG_ON(compress_type);
1747                 ret = btrfs_mark_extent_written(trans, inode,
1748                                                 ordered_extent->file_offset,
1749                                                 ordered_extent->file_offset +
1750                                                 ordered_extent->len);
1751                 BUG_ON(ret);
1752         } else {
1753                 BUG_ON(root == root->fs_info->tree_root);
1754                 ret = insert_reserved_file_extent(trans, inode,
1755                                                 ordered_extent->file_offset,
1756                                                 ordered_extent->start,
1757                                                 ordered_extent->disk_len,
1758                                                 ordered_extent->len,
1759                                                 ordered_extent->len,
1760                                                 compress_type, 0, 0,
1761                                                 BTRFS_FILE_EXTENT_REG);
1762                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1763                                    ordered_extent->file_offset,
1764                                    ordered_extent->len);
1765                 BUG_ON(ret);
1766         }
1767         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1768                              ordered_extent->file_offset +
1769                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1770
1771         add_pending_csums(trans, inode, ordered_extent->file_offset,
1772                           &ordered_extent->list);
1773
1774         btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1775         ret = btrfs_update_inode(trans, root, inode);
1776         BUG_ON(ret);
1777 out:
1778         if (nolock) {
1779                 if (trans)
1780                         btrfs_end_transaction_nolock(trans, root);
1781         } else {
1782                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1783                 if (trans)
1784                         btrfs_end_transaction(trans, root);
1785         }
1786
1787         /* once for us */
1788         btrfs_put_ordered_extent(ordered_extent);
1789         /* once for the tree */
1790         btrfs_put_ordered_extent(ordered_extent);
1791
1792         return 0;
1793 }
1794
1795 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1796                                 struct extent_state *state, int uptodate)
1797 {
1798         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1799
1800         ClearPagePrivate2(page);
1801         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1802 }
1803
1804 /*
1805  * When IO fails, either with EIO or csum verification fails, we
1806  * try other mirrors that might have a good copy of the data.  This
1807  * io_failure_record is used to record state as we go through all the
1808  * mirrors.  If another mirror has good data, the page is set up to date
1809  * and things continue.  If a good mirror can't be found, the original
1810  * bio end_io callback is called to indicate things have failed.
1811  */
1812 struct io_failure_record {
1813         struct page *page;
1814         u64 start;
1815         u64 len;
1816         u64 logical;
1817         unsigned long bio_flags;
1818         int last_mirror;
1819 };
1820
1821 static int btrfs_io_failed_hook(struct bio *failed_bio,
1822                          struct page *page, u64 start, u64 end,
1823                          struct extent_state *state)
1824 {
1825         struct io_failure_record *failrec = NULL;
1826         u64 private;
1827         struct extent_map *em;
1828         struct inode *inode = page->mapping->host;
1829         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1830         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1831         struct bio *bio;
1832         int num_copies;
1833         int ret;
1834         int rw;
1835         u64 logical;
1836
1837         ret = get_state_private(failure_tree, start, &private);
1838         if (ret) {
1839                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1840                 if (!failrec)
1841                         return -ENOMEM;
1842                 failrec->start = start;
1843                 failrec->len = end - start + 1;
1844                 failrec->last_mirror = 0;
1845                 failrec->bio_flags = 0;
1846
1847                 read_lock(&em_tree->lock);
1848                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1849                 if (em->start > start || em->start + em->len < start) {
1850                         free_extent_map(em);
1851                         em = NULL;
1852                 }
1853                 read_unlock(&em_tree->lock);
1854
1855                 if (!em || IS_ERR(em)) {
1856                         kfree(failrec);
1857                         return -EIO;
1858                 }
1859                 logical = start - em->start;
1860                 logical = em->block_start + logical;
1861                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1862                         logical = em->block_start;
1863                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1864                         extent_set_compress_type(&failrec->bio_flags,
1865                                                  em->compress_type);
1866                 }
1867                 failrec->logical = logical;
1868                 free_extent_map(em);
1869                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1870                                 EXTENT_DIRTY, GFP_NOFS);
1871                 set_state_private(failure_tree, start,
1872                                  (u64)(unsigned long)failrec);
1873         } else {
1874                 failrec = (struct io_failure_record *)(unsigned long)private;
1875         }
1876         num_copies = btrfs_num_copies(
1877                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1878                               failrec->logical, failrec->len);
1879         failrec->last_mirror++;
1880         if (!state) {
1881                 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1882                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1883                                                     failrec->start,
1884                                                     EXTENT_LOCKED);
1885                 if (state && state->start != failrec->start)
1886                         state = NULL;
1887                 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1888         }
1889         if (!state || failrec->last_mirror > num_copies) {
1890                 set_state_private(failure_tree, failrec->start, 0);
1891                 clear_extent_bits(failure_tree, failrec->start,
1892                                   failrec->start + failrec->len - 1,
1893                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1894                 kfree(failrec);
1895                 return -EIO;
1896         }
1897         bio = bio_alloc(GFP_NOFS, 1);
1898         bio->bi_private = state;
1899         bio->bi_end_io = failed_bio->bi_end_io;
1900         bio->bi_sector = failrec->logical >> 9;
1901         bio->bi_bdev = failed_bio->bi_bdev;
1902         bio->bi_size = 0;
1903
1904         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1905         if (failed_bio->bi_rw & REQ_WRITE)
1906                 rw = WRITE;
1907         else
1908                 rw = READ;
1909
1910         ret = BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1911                                                       failrec->last_mirror,
1912                                                       failrec->bio_flags, 0);
1913         return ret;
1914 }
1915
1916 /*
1917  * each time an IO finishes, we do a fast check in the IO failure tree
1918  * to see if we need to process or clean up an io_failure_record
1919  */
1920 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1921 {
1922         u64 private;
1923         u64 private_failure;
1924         struct io_failure_record *failure;
1925         int ret;
1926
1927         private = 0;
1928         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1929                              (u64)-1, 1, EXTENT_DIRTY, 0)) {
1930                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1931                                         start, &private_failure);
1932                 if (ret == 0) {
1933                         failure = (struct io_failure_record *)(unsigned long)
1934                                    private_failure;
1935                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1936                                           failure->start, 0);
1937                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1938                                           failure->start,
1939                                           failure->start + failure->len - 1,
1940                                           EXTENT_DIRTY | EXTENT_LOCKED,
1941                                           GFP_NOFS);
1942                         kfree(failure);
1943                 }
1944         }
1945         return 0;
1946 }
1947
1948 /*
1949  * when reads are done, we need to check csums to verify the data is correct
1950  * if there's a match, we allow the bio to finish.  If not, we go through
1951  * the io_failure_record routines to find good copies
1952  */
1953 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1954                                struct extent_state *state)
1955 {
1956         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1957         struct inode *inode = page->mapping->host;
1958         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1959         char *kaddr;
1960         u64 private = ~(u32)0;
1961         int ret;
1962         struct btrfs_root *root = BTRFS_I(inode)->root;
1963         u32 csum = ~(u32)0;
1964
1965         if (PageChecked(page)) {
1966                 ClearPageChecked(page);
1967                 goto good;
1968         }
1969
1970         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1971                 return 0;
1972
1973         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1974             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1975                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1976                                   GFP_NOFS);
1977                 return 0;
1978         }
1979
1980         if (state && state->start == start) {
1981                 private = state->private;
1982                 ret = 0;
1983         } else {
1984                 ret = get_state_private(io_tree, start, &private);
1985         }
1986         kaddr = kmap_atomic(page, KM_USER0);
1987         if (ret)
1988                 goto zeroit;
1989
1990         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
1991         btrfs_csum_final(csum, (char *)&csum);
1992         if (csum != private)
1993                 goto zeroit;
1994
1995         kunmap_atomic(kaddr, KM_USER0);
1996 good:
1997         /* if the io failure tree for this inode is non-empty,
1998          * check to see if we've recovered from a failed IO
1999          */
2000         btrfs_clean_io_failures(inode, start);
2001         return 0;
2002
2003 zeroit:
2004         if (printk_ratelimit()) {
2005                 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
2006                        "private %llu\n", page->mapping->host->i_ino,
2007                        (unsigned long long)start, csum,
2008                        (unsigned long long)private);
2009         }
2010         memset(kaddr + offset, 1, end - start + 1);
2011         flush_dcache_page(page);
2012         kunmap_atomic(kaddr, KM_USER0);
2013         if (private == 0)
2014                 return 0;
2015         return -EIO;
2016 }
2017
2018 struct delayed_iput {
2019         struct list_head list;
2020         struct inode *inode;
2021 };
2022
2023 void btrfs_add_delayed_iput(struct inode *inode)
2024 {
2025         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2026         struct delayed_iput *delayed;
2027
2028         if (atomic_add_unless(&inode->i_count, -1, 1))
2029                 return;
2030
2031         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2032         delayed->inode = inode;
2033
2034         spin_lock(&fs_info->delayed_iput_lock);
2035         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2036         spin_unlock(&fs_info->delayed_iput_lock);
2037 }
2038
2039 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2040 {
2041         LIST_HEAD(list);
2042         struct btrfs_fs_info *fs_info = root->fs_info;
2043         struct delayed_iput *delayed;
2044         int empty;
2045
2046         spin_lock(&fs_info->delayed_iput_lock);
2047         empty = list_empty(&fs_info->delayed_iputs);
2048         spin_unlock(&fs_info->delayed_iput_lock);
2049         if (empty)
2050                 return;
2051
2052         down_read(&root->fs_info->cleanup_work_sem);
2053         spin_lock(&fs_info->delayed_iput_lock);
2054         list_splice_init(&fs_info->delayed_iputs, &list);
2055         spin_unlock(&fs_info->delayed_iput_lock);
2056
2057         while (!list_empty(&list)) {
2058                 delayed = list_entry(list.next, struct delayed_iput, list);
2059                 list_del(&delayed->list);
2060                 iput(delayed->inode);
2061                 kfree(delayed);
2062         }
2063         up_read(&root->fs_info->cleanup_work_sem);
2064 }
2065
2066 /*
2067  * calculate extra metadata reservation when snapshotting a subvolume
2068  * contains orphan files.
2069  */
2070 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans,
2071                                 struct btrfs_pending_snapshot *pending,
2072                                 u64 *bytes_to_reserve)
2073 {
2074         struct btrfs_root *root;
2075         struct btrfs_block_rsv *block_rsv;
2076         u64 num_bytes;
2077         int index;
2078
2079         root = pending->root;
2080         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2081                 return;
2082
2083         block_rsv = root->orphan_block_rsv;
2084
2085         /* orphan block reservation for the snapshot */
2086         num_bytes = block_rsv->size;
2087
2088         /*
2089          * after the snapshot is created, COWing tree blocks may use more
2090          * space than it frees. So we should make sure there is enough
2091          * reserved space.
2092          */
2093         index = trans->transid & 0x1;
2094         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2095                 num_bytes += block_rsv->size -
2096                              (block_rsv->reserved + block_rsv->freed[index]);
2097         }
2098
2099         *bytes_to_reserve += num_bytes;
2100 }
2101
2102 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans,
2103                                 struct btrfs_pending_snapshot *pending)
2104 {
2105         struct btrfs_root *root = pending->root;
2106         struct btrfs_root *snap = pending->snap;
2107         struct btrfs_block_rsv *block_rsv;
2108         u64 num_bytes;
2109         int index;
2110         int ret;
2111
2112         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2113                 return;
2114
2115         /* refill source subvolume's orphan block reservation */
2116         block_rsv = root->orphan_block_rsv;
2117         index = trans->transid & 0x1;
2118         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2119                 num_bytes = block_rsv->size -
2120                             (block_rsv->reserved + block_rsv->freed[index]);
2121                 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2122                                               root->orphan_block_rsv,
2123                                               num_bytes);
2124                 BUG_ON(ret);
2125         }
2126
2127         /* setup orphan block reservation for the snapshot */
2128         block_rsv = btrfs_alloc_block_rsv(snap);
2129         BUG_ON(!block_rsv);
2130
2131         btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2132         snap->orphan_block_rsv = block_rsv;
2133
2134         num_bytes = root->orphan_block_rsv->size;
2135         ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2136                                       block_rsv, num_bytes);
2137         BUG_ON(ret);
2138
2139 #if 0
2140         /* insert orphan item for the snapshot */
2141         WARN_ON(!root->orphan_item_inserted);
2142         ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2143                                        snap->root_key.objectid);
2144         BUG_ON(ret);
2145         snap->orphan_item_inserted = 1;
2146 #endif
2147 }
2148
2149 enum btrfs_orphan_cleanup_state {
2150         ORPHAN_CLEANUP_STARTED  = 1,
2151         ORPHAN_CLEANUP_DONE     = 2,
2152 };
2153
2154 /*
2155  * This is called in transaction commmit time. If there are no orphan
2156  * files in the subvolume, it removes orphan item and frees block_rsv
2157  * structure.
2158  */
2159 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2160                               struct btrfs_root *root)
2161 {
2162         int ret;
2163
2164         if (!list_empty(&root->orphan_list) ||
2165             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2166                 return;
2167
2168         if (root->orphan_item_inserted &&
2169             btrfs_root_refs(&root->root_item) > 0) {
2170                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2171                                             root->root_key.objectid);
2172                 BUG_ON(ret);
2173                 root->orphan_item_inserted = 0;
2174         }
2175
2176         if (root->orphan_block_rsv) {
2177                 WARN_ON(root->orphan_block_rsv->size > 0);
2178                 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2179                 root->orphan_block_rsv = NULL;
2180         }
2181 }
2182
2183 /*
2184  * This creates an orphan entry for the given inode in case something goes
2185  * wrong in the middle of an unlink/truncate.
2186  *
2187  * NOTE: caller of this function should reserve 5 units of metadata for
2188  *       this function.
2189  */
2190 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2191 {
2192         struct btrfs_root *root = BTRFS_I(inode)->root;
2193         struct btrfs_block_rsv *block_rsv = NULL;
2194         int reserve = 0;
2195         int insert = 0;
2196         int ret;
2197
2198         if (!root->orphan_block_rsv) {
2199                 block_rsv = btrfs_alloc_block_rsv(root);
2200                 BUG_ON(!block_rsv);
2201         }
2202
2203         spin_lock(&root->orphan_lock);
2204         if (!root->orphan_block_rsv) {
2205                 root->orphan_block_rsv = block_rsv;
2206         } else if (block_rsv) {
2207                 btrfs_free_block_rsv(root, block_rsv);
2208                 block_rsv = NULL;
2209         }
2210
2211         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2212                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2213 #if 0
2214                 /*
2215                  * For proper ENOSPC handling, we should do orphan
2216                  * cleanup when mounting. But this introduces backward
2217                  * compatibility issue.
2218                  */
2219                 if (!xchg(&root->orphan_item_inserted, 1))
2220                         insert = 2;
2221                 else
2222                         insert = 1;
2223 #endif
2224                 insert = 1;
2225         } else {
2226                 WARN_ON(!BTRFS_I(inode)->orphan_meta_reserved);
2227         }
2228
2229         if (!BTRFS_I(inode)->orphan_meta_reserved) {
2230                 BTRFS_I(inode)->orphan_meta_reserved = 1;
2231                 reserve = 1;
2232         }
2233         spin_unlock(&root->orphan_lock);
2234
2235         if (block_rsv)
2236                 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2237
2238         /* grab metadata reservation from transaction handle */
2239         if (reserve) {
2240                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2241                 BUG_ON(ret);
2242         }
2243
2244         /* insert an orphan item to track this unlinked/truncated file */
2245         if (insert >= 1) {
2246                 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
2247                 BUG_ON(ret);
2248         }
2249
2250         /* insert an orphan item to track subvolume contains orphan files */
2251         if (insert >= 2) {
2252                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2253                                                root->root_key.objectid);
2254                 BUG_ON(ret);
2255         }
2256         return 0;
2257 }
2258
2259 /*
2260  * We have done the truncate/delete so we can go ahead and remove the orphan
2261  * item for this particular inode.
2262  */
2263 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2264 {
2265         struct btrfs_root *root = BTRFS_I(inode)->root;
2266         int delete_item = 0;
2267         int release_rsv = 0;
2268         int ret = 0;
2269
2270         spin_lock(&root->orphan_lock);
2271         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2272                 list_del_init(&BTRFS_I(inode)->i_orphan);
2273                 delete_item = 1;
2274         }
2275
2276         if (BTRFS_I(inode)->orphan_meta_reserved) {
2277                 BTRFS_I(inode)->orphan_meta_reserved = 0;
2278                 release_rsv = 1;
2279         }
2280         spin_unlock(&root->orphan_lock);
2281
2282         if (trans && delete_item) {
2283                 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
2284                 BUG_ON(ret);
2285         }
2286
2287         if (release_rsv)
2288                 btrfs_orphan_release_metadata(inode);
2289
2290         return 0;
2291 }
2292
2293 /*
2294  * this cleans up any orphans that may be left on the list from the last use
2295  * of this root.
2296  */
2297 int btrfs_orphan_cleanup(struct btrfs_root *root)
2298 {
2299         struct btrfs_path *path;
2300         struct extent_buffer *leaf;
2301         struct btrfs_key key, found_key;
2302         struct btrfs_trans_handle *trans;
2303         struct inode *inode;
2304         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2305
2306         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2307                 return 0;
2308
2309         path = btrfs_alloc_path();
2310         if (!path) {
2311                 ret = -ENOMEM;
2312                 goto out;
2313         }
2314         path->reada = -1;
2315
2316         key.objectid = BTRFS_ORPHAN_OBJECTID;
2317         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2318         key.offset = (u64)-1;
2319
2320         while (1) {
2321                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2322                 if (ret < 0)
2323                         goto out;
2324
2325                 /*
2326                  * if ret == 0 means we found what we were searching for, which
2327                  * is weird, but possible, so only screw with path if we didnt
2328                  * find the key and see if we have stuff that matches
2329                  */
2330                 if (ret > 0) {
2331                         ret = 0;
2332                         if (path->slots[0] == 0)
2333                                 break;
2334                         path->slots[0]--;
2335                 }
2336
2337                 /* pull out the item */
2338                 leaf = path->nodes[0];
2339                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2340
2341                 /* make sure the item matches what we want */
2342                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2343                         break;
2344                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2345                         break;
2346
2347                 /* release the path since we're done with it */
2348                 btrfs_release_path(root, path);
2349
2350                 /*
2351                  * this is where we are basically btrfs_lookup, without the
2352                  * crossing root thing.  we store the inode number in the
2353                  * offset of the orphan item.
2354                  */
2355                 found_key.objectid = found_key.offset;
2356                 found_key.type = BTRFS_INODE_ITEM_KEY;
2357                 found_key.offset = 0;
2358                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2359                 if (IS_ERR(inode)) {
2360                         ret = PTR_ERR(inode);
2361                         goto out;
2362                 }
2363
2364                 /*
2365                  * add this inode to the orphan list so btrfs_orphan_del does
2366                  * the proper thing when we hit it
2367                  */
2368                 spin_lock(&root->orphan_lock);
2369                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2370                 spin_unlock(&root->orphan_lock);
2371
2372                 /*
2373                  * if this is a bad inode, means we actually succeeded in
2374                  * removing the inode, but not the orphan record, which means
2375                  * we need to manually delete the orphan since iput will just
2376                  * do a destroy_inode
2377                  */
2378                 if (is_bad_inode(inode)) {
2379                         trans = btrfs_start_transaction(root, 0);
2380                         if (IS_ERR(trans)) {
2381                                 ret = PTR_ERR(trans);
2382                                 goto out;
2383                         }
2384                         btrfs_orphan_del(trans, inode);
2385                         btrfs_end_transaction(trans, root);
2386                         iput(inode);
2387                         continue;
2388                 }
2389
2390                 /* if we have links, this was a truncate, lets do that */
2391                 if (inode->i_nlink) {
2392                         if (!S_ISREG(inode->i_mode)) {
2393                                 WARN_ON(1);
2394                                 iput(inode);
2395                                 continue;
2396                         }
2397                         nr_truncate++;
2398                         ret = btrfs_truncate(inode);
2399                 } else {
2400                         nr_unlink++;
2401                 }
2402
2403                 /* this will do delete_inode and everything for us */
2404                 iput(inode);
2405                 if (ret)
2406                         goto out;
2407         }
2408         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2409
2410         if (root->orphan_block_rsv)
2411                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2412                                         (u64)-1);
2413
2414         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2415                 trans = btrfs_join_transaction(root, 1);
2416                 if (!IS_ERR(trans))
2417                         btrfs_end_transaction(trans, root);
2418         }
2419
2420         if (nr_unlink)
2421                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2422         if (nr_truncate)
2423                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2424
2425 out:
2426         if (ret)
2427                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2428         btrfs_free_path(path);
2429         return ret;
2430 }
2431
2432 /*
2433  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2434  * don't find any xattrs, we know there can't be any acls.
2435  *
2436  * slot is the slot the inode is in, objectid is the objectid of the inode
2437  */
2438 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2439                                           int slot, u64 objectid)
2440 {
2441         u32 nritems = btrfs_header_nritems(leaf);
2442         struct btrfs_key found_key;
2443         int scanned = 0;
2444
2445         slot++;
2446         while (slot < nritems) {
2447                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2448
2449                 /* we found a different objectid, there must not be acls */
2450                 if (found_key.objectid != objectid)
2451                         return 0;
2452
2453                 /* we found an xattr, assume we've got an acl */
2454                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2455                         return 1;
2456
2457                 /*
2458                  * we found a key greater than an xattr key, there can't
2459                  * be any acls later on
2460                  */
2461                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2462                         return 0;
2463
2464                 slot++;
2465                 scanned++;
2466
2467                 /*
2468                  * it goes inode, inode backrefs, xattrs, extents,
2469                  * so if there are a ton of hard links to an inode there can
2470                  * be a lot of backrefs.  Don't waste time searching too hard,
2471                  * this is just an optimization
2472                  */
2473                 if (scanned >= 8)
2474                         break;
2475         }
2476         /* we hit the end of the leaf before we found an xattr or
2477          * something larger than an xattr.  We have to assume the inode
2478          * has acls
2479          */
2480         return 1;
2481 }
2482
2483 /*
2484  * read an inode from the btree into the in-memory inode
2485  */
2486 static void btrfs_read_locked_inode(struct inode *inode)
2487 {
2488         struct btrfs_path *path;
2489         struct extent_buffer *leaf;
2490         struct btrfs_inode_item *inode_item;
2491         struct btrfs_timespec *tspec;
2492         struct btrfs_root *root = BTRFS_I(inode)->root;
2493         struct btrfs_key location;
2494         int maybe_acls;
2495         u64 alloc_group_block;
2496         u32 rdev;
2497         int ret;
2498
2499         path = btrfs_alloc_path();
2500         BUG_ON(!path);
2501         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2502
2503         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2504         if (ret)
2505                 goto make_bad;
2506
2507         leaf = path->nodes[0];
2508         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2509                                     struct btrfs_inode_item);
2510
2511         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2512         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2513         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2514         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2515         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2516
2517         tspec = btrfs_inode_atime(inode_item);
2518         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2519         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2520
2521         tspec = btrfs_inode_mtime(inode_item);
2522         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2523         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2524
2525         tspec = btrfs_inode_ctime(inode_item);
2526         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2527         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2528
2529         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2530         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2531         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2532         inode->i_generation = BTRFS_I(inode)->generation;
2533         inode->i_rdev = 0;
2534         rdev = btrfs_inode_rdev(leaf, inode_item);
2535
2536         BTRFS_I(inode)->index_cnt = (u64)-1;
2537         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2538
2539         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2540         if (location.objectid == BTRFS_FREE_SPACE_OBJECTID)
2541                 inode->i_mapping->flags &= ~__GFP_FS;
2542
2543         /*
2544          * try to precache a NULL acl entry for files that don't have
2545          * any xattrs or acls
2546          */
2547         maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2548         if (!maybe_acls)
2549                 cache_no_acl(inode);
2550
2551         BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2552                                                 alloc_group_block, 0);
2553         btrfs_free_path(path);
2554         inode_item = NULL;
2555
2556         switch (inode->i_mode & S_IFMT) {
2557         case S_IFREG:
2558                 inode->i_mapping->a_ops = &btrfs_aops;
2559                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2560                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2561                 inode->i_fop = &btrfs_file_operations;
2562                 inode->i_op = &btrfs_file_inode_operations;
2563                 break;
2564         case S_IFDIR:
2565                 inode->i_fop = &btrfs_dir_file_operations;
2566                 if (root == root->fs_info->tree_root)
2567                         inode->i_op = &btrfs_dir_ro_inode_operations;
2568                 else
2569                         inode->i_op = &btrfs_dir_inode_operations;
2570                 break;
2571         case S_IFLNK:
2572                 inode->i_op = &btrfs_symlink_inode_operations;
2573                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2574                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2575                 break;
2576         default:
2577                 inode->i_op = &btrfs_special_inode_operations;
2578                 init_special_inode(inode, inode->i_mode, rdev);
2579                 break;
2580         }
2581
2582         btrfs_update_iflags(inode);
2583         return;
2584
2585 make_bad:
2586         btrfs_free_path(path);
2587         make_bad_inode(inode);
2588 }
2589
2590 /*
2591  * given a leaf and an inode, copy the inode fields into the leaf
2592  */
2593 static void fill_inode_item(struct btrfs_trans_handle *trans,
2594                             struct extent_buffer *leaf,
2595                             struct btrfs_inode_item *item,
2596                             struct inode *inode)
2597 {
2598         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2599         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2600         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2601         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2602         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2603
2604         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2605                                inode->i_atime.tv_sec);
2606         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2607                                 inode->i_atime.tv_nsec);
2608
2609         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2610                                inode->i_mtime.tv_sec);
2611         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2612                                 inode->i_mtime.tv_nsec);
2613
2614         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2615                                inode->i_ctime.tv_sec);
2616         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2617                                 inode->i_ctime.tv_nsec);
2618
2619         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2620         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2621         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2622         btrfs_set_inode_transid(leaf, item, trans->transid);
2623         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2624         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2625         btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2626 }
2627
2628 /*
2629  * copy everything in the in-memory inode into the btree.
2630  */
2631 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2632                                 struct btrfs_root *root, struct inode *inode)
2633 {
2634         struct btrfs_inode_item *inode_item;
2635         struct btrfs_path *path;
2636         struct extent_buffer *leaf;
2637         int ret;
2638
2639         path = btrfs_alloc_path();
2640         BUG_ON(!path);
2641         path->leave_spinning = 1;
2642         ret = btrfs_lookup_inode(trans, root, path,
2643                                  &BTRFS_I(inode)->location, 1);
2644         if (ret) {
2645                 if (ret > 0)
2646                         ret = -ENOENT;
2647                 goto failed;
2648         }
2649
2650         btrfs_unlock_up_safe(path, 1);
2651         leaf = path->nodes[0];
2652         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2653                                   struct btrfs_inode_item);
2654
2655         fill_inode_item(trans, leaf, inode_item, inode);
2656         btrfs_mark_buffer_dirty(leaf);
2657         btrfs_set_inode_last_trans(trans, inode);
2658         ret = 0;
2659 failed:
2660         btrfs_free_path(path);
2661         return ret;
2662 }
2663
2664
2665 /*
2666  * unlink helper that gets used here in inode.c and in the tree logging
2667  * recovery code.  It remove a link in a directory with a given name, and
2668  * also drops the back refs in the inode to the directory
2669  */
2670 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2671                                 struct btrfs_root *root,
2672                                 struct inode *dir, struct inode *inode,
2673                                 const char *name, int name_len)
2674 {
2675         struct btrfs_path *path;
2676         int ret = 0;
2677         struct extent_buffer *leaf;
2678         struct btrfs_dir_item *di;
2679         struct btrfs_key key;
2680         u64 index;
2681
2682         path = btrfs_alloc_path();
2683         if (!path) {
2684                 ret = -ENOMEM;
2685                 goto out;
2686         }
2687
2688         path->leave_spinning = 1;
2689         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2690                                     name, name_len, -1);
2691         if (IS_ERR(di)) {
2692                 ret = PTR_ERR(di);
2693                 goto err;
2694         }
2695         if (!di) {
2696                 ret = -ENOENT;
2697                 goto err;
2698         }
2699         leaf = path->nodes[0];
2700         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2701         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2702         if (ret)
2703                 goto err;
2704         btrfs_release_path(root, path);
2705
2706         ret = btrfs_del_inode_ref(trans, root, name, name_len,
2707                                   inode->i_ino,
2708                                   dir->i_ino, &index);
2709         if (ret) {
2710                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2711                        "inode %lu parent %lu\n", name_len, name,
2712                        inode->i_ino, dir->i_ino);
2713                 goto err;
2714         }
2715
2716         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2717                                          index, name, name_len, -1);
2718         if (IS_ERR(di)) {
2719                 ret = PTR_ERR(di);
2720                 goto err;
2721         }
2722         if (!di) {
2723                 ret = -ENOENT;
2724                 goto err;
2725         }
2726         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2727         btrfs_release_path(root, path);
2728
2729         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2730                                          inode, dir->i_ino);
2731         BUG_ON(ret != 0 && ret != -ENOENT);
2732
2733         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2734                                            dir, index);
2735         if (ret == -ENOENT)
2736                 ret = 0;
2737 err:
2738         btrfs_free_path(path);
2739         if (ret)
2740                 goto out;
2741
2742         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2743         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2744         btrfs_update_inode(trans, root, dir);
2745 out:
2746         return ret;
2747 }
2748
2749 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2750                        struct btrfs_root *root,
2751                        struct inode *dir, struct inode *inode,
2752                        const char *name, int name_len)
2753 {
2754         int ret;
2755         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2756         if (!ret) {
2757                 btrfs_drop_nlink(inode);
2758                 ret = btrfs_update_inode(trans, root, inode);
2759         }
2760         return ret;
2761 }
2762
2763
2764 /* helper to check if there is any shared block in the path */
2765 static int check_path_shared(struct btrfs_root *root,
2766                              struct btrfs_path *path)
2767 {
2768         struct extent_buffer *eb;
2769         int level;
2770         u64 refs = 1;
2771
2772         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2773                 int ret;
2774
2775                 if (!path->nodes[level])
2776                         break;
2777                 eb = path->nodes[level];
2778                 if (!btrfs_block_can_be_shared(root, eb))
2779                         continue;
2780                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2781                                                &refs, NULL);
2782                 if (refs > 1)
2783                         return 1;
2784         }
2785         return 0;
2786 }
2787
2788 /*
2789  * helper to start transaction for unlink and rmdir.
2790  *
2791  * unlink and rmdir are special in btrfs, they do not always free space.
2792  * so in enospc case, we should make sure they will free space before
2793  * allowing them to use the global metadata reservation.
2794  */
2795 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2796                                                        struct dentry *dentry)
2797 {
2798         struct btrfs_trans_handle *trans;
2799         struct btrfs_root *root = BTRFS_I(dir)->root;
2800         struct btrfs_path *path;
2801         struct btrfs_inode_ref *ref;
2802         struct btrfs_dir_item *di;
2803         struct inode *inode = dentry->d_inode;
2804         u64 index;
2805         int check_link = 1;
2806         int err = -ENOSPC;
2807         int ret;
2808
2809         trans = btrfs_start_transaction(root, 10);
2810         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2811                 return trans;
2812
2813         if (inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2814                 return ERR_PTR(-ENOSPC);
2815
2816         /* check if there is someone else holds reference */
2817         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2818                 return ERR_PTR(-ENOSPC);
2819
2820         if (atomic_read(&inode->i_count) > 2)
2821                 return ERR_PTR(-ENOSPC);
2822
2823         if (xchg(&root->fs_info->enospc_unlink, 1))
2824                 return ERR_PTR(-ENOSPC);
2825
2826         path = btrfs_alloc_path();
2827         if (!path) {
2828                 root->fs_info->enospc_unlink = 0;
2829                 return ERR_PTR(-ENOMEM);
2830         }
2831
2832         trans = btrfs_start_transaction(root, 0);
2833         if (IS_ERR(trans)) {
2834                 btrfs_free_path(path);
2835                 root->fs_info->enospc_unlink = 0;
2836                 return trans;
2837         }
2838
2839         path->skip_locking = 1;
2840         path->search_commit_root = 1;
2841
2842         ret = btrfs_lookup_inode(trans, root, path,
2843                                 &BTRFS_I(dir)->location, 0);
2844         if (ret < 0) {
2845                 err = ret;
2846                 goto out;
2847         }
2848         if (ret == 0) {
2849                 if (check_path_shared(root, path))
2850                         goto out;
2851         } else {
2852                 check_link = 0;
2853         }
2854         btrfs_release_path(root, path);
2855
2856         ret = btrfs_lookup_inode(trans, root, path,
2857                                 &BTRFS_I(inode)->location, 0);
2858         if (ret < 0) {
2859                 err = ret;
2860                 goto out;
2861         }
2862         if (ret == 0) {
2863                 if (check_path_shared(root, path))
2864                         goto out;
2865         } else {
2866                 check_link = 0;
2867         }
2868         btrfs_release_path(root, path);
2869
2870         if (ret == 0 && S_ISREG(inode->i_mode)) {
2871                 ret = btrfs_lookup_file_extent(trans, root, path,
2872                                                inode->i_ino, (u64)-1, 0);
2873                 if (ret < 0) {
2874                         err = ret;
2875                         goto out;
2876                 }
2877                 BUG_ON(ret == 0);
2878                 if (check_path_shared(root, path))
2879                         goto out;
2880                 btrfs_release_path(root, path);
2881         }
2882
2883         if (!check_link) {
2884                 err = 0;
2885                 goto out;
2886         }
2887
2888         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2889                                 dentry->d_name.name, dentry->d_name.len, 0);
2890         if (IS_ERR(di)) {
2891                 err = PTR_ERR(di);
2892                 goto out;
2893         }
2894         if (di) {
2895                 if (check_path_shared(root, path))
2896                         goto out;
2897         } else {
2898                 err = 0;
2899                 goto out;
2900         }
2901         btrfs_release_path(root, path);
2902
2903         ref = btrfs_lookup_inode_ref(trans, root, path,
2904                                 dentry->d_name.name, dentry->d_name.len,
2905                                 inode->i_ino, dir->i_ino, 0);
2906         if (IS_ERR(ref)) {
2907                 err = PTR_ERR(ref);
2908                 goto out;
2909         }
2910         BUG_ON(!ref);
2911         if (check_path_shared(root, path))
2912                 goto out;
2913         index = btrfs_inode_ref_index(path->nodes[0], ref);
2914         btrfs_release_path(root, path);
2915
2916         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, index,
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         BUG_ON(ret == -ENOENT);
2923         if (check_path_shared(root, path))
2924                 goto out;
2925
2926         err = 0;
2927 out:
2928         btrfs_free_path(path);
2929         if (err) {
2930                 btrfs_end_transaction(trans, root);
2931                 root->fs_info->enospc_unlink = 0;
2932                 return ERR_PTR(err);
2933         }
2934
2935         trans->block_rsv = &root->fs_info->global_block_rsv;
2936         return trans;
2937 }
2938
2939 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2940                                struct btrfs_root *root)
2941 {
2942         if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2943                 BUG_ON(!root->fs_info->enospc_unlink);
2944                 root->fs_info->enospc_unlink = 0;
2945         }
2946         btrfs_end_transaction_throttle(trans, root);
2947 }
2948
2949 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2950 {
2951         struct btrfs_root *root = BTRFS_I(dir)->root;
2952         struct btrfs_trans_handle *trans;
2953         struct inode *inode = dentry->d_inode;
2954         int ret;
2955         unsigned long nr = 0;
2956
2957         trans = __unlink_start_trans(dir, dentry);
2958         if (IS_ERR(trans))
2959                 return PTR_ERR(trans);
2960
2961         btrfs_set_trans_block_group(trans, dir);
2962
2963         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2964
2965         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2966                                  dentry->d_name.name, dentry->d_name.len);
2967         BUG_ON(ret);
2968
2969         if (inode->i_nlink == 0) {
2970                 ret = btrfs_orphan_add(trans, inode);
2971                 BUG_ON(ret);
2972         }
2973
2974         nr = trans->blocks_used;
2975         __unlink_end_trans(trans, root);
2976         btrfs_btree_balance_dirty(root, nr);
2977         return ret;
2978 }
2979
2980 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2981                         struct btrfs_root *root,
2982                         struct inode *dir, u64 objectid,
2983                         const char *name, int name_len)
2984 {
2985         struct btrfs_path *path;
2986         struct extent_buffer *leaf;
2987         struct btrfs_dir_item *di;
2988         struct btrfs_key key;
2989         u64 index;
2990         int ret;
2991
2992         path = btrfs_alloc_path();
2993         if (!path)
2994                 return -ENOMEM;
2995
2996         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2997                                    name, name_len, -1);
2998         BUG_ON(!di || IS_ERR(di));
2999
3000         leaf = path->nodes[0];
3001         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3002         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3003         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3004         BUG_ON(ret);
3005         btrfs_release_path(root, path);
3006
3007         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3008                                  objectid, root->root_key.objectid,
3009                                  dir->i_ino, &index, name, name_len);
3010         if (ret < 0) {
3011                 BUG_ON(ret != -ENOENT);
3012                 di = btrfs_search_dir_index_item(root, path, dir->i_ino,
3013                                                  name, name_len);
3014                 BUG_ON(!di || IS_ERR(di));
3015
3016                 leaf = path->nodes[0];
3017                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3018                 btrfs_release_path(root, path);
3019                 index = key.offset;
3020         }
3021
3022         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
3023                                          index, name, name_len, -1);
3024         BUG_ON(!di || IS_ERR(di));
3025
3026         leaf = path->nodes[0];
3027         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3028         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3029         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3030         BUG_ON(ret);
3031         btrfs_release_path(root, path);
3032
3033         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3034         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3035         ret = btrfs_update_inode(trans, root, dir);
3036         BUG_ON(ret);
3037
3038         btrfs_free_path(path);
3039         return 0;
3040 }
3041
3042 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3043 {
3044         struct inode *inode = dentry->d_inode;
3045         int err = 0;
3046         struct btrfs_root *root = BTRFS_I(dir)->root;
3047         struct btrfs_trans_handle *trans;
3048         unsigned long nr = 0;
3049
3050         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3051             inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
3052                 return -ENOTEMPTY;
3053
3054         trans = __unlink_start_trans(dir, dentry);
3055         if (IS_ERR(trans))
3056                 return PTR_ERR(trans);
3057
3058         btrfs_set_trans_block_group(trans, dir);
3059
3060         if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3061                 err = btrfs_unlink_subvol(trans, root, dir,
3062                                           BTRFS_I(inode)->location.objectid,
3063                                           dentry->d_name.name,
3064                                           dentry->d_name.len);
3065                 goto out;
3066         }
3067
3068         err = btrfs_orphan_add(trans, inode);
3069         if (err)
3070                 goto out;
3071
3072         /* now the directory is empty */
3073         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3074                                  dentry->d_name.name, dentry->d_name.len);
3075         if (!err)
3076                 btrfs_i_size_write(inode, 0);
3077 out:
3078         nr = trans->blocks_used;
3079         __unlink_end_trans(trans, root);
3080         btrfs_btree_balance_dirty(root, nr);
3081
3082         return err;
3083 }
3084
3085 #if 0
3086 /*
3087  * when truncating bytes in a file, it is possible to avoid reading
3088  * the leaves that contain only checksum items.  This can be the
3089  * majority of the IO required to delete a large file, but it must
3090  * be done carefully.
3091  *
3092  * The keys in the level just above the leaves are checked to make sure
3093  * the lowest key in a given leaf is a csum key, and starts at an offset
3094  * after the new  size.
3095  *
3096  * Then the key for the next leaf is checked to make sure it also has
3097  * a checksum item for the same file.  If it does, we know our target leaf
3098  * contains only checksum items, and it can be safely freed without reading
3099  * it.
3100  *
3101  * This is just an optimization targeted at large files.  It may do
3102  * nothing.  It will return 0 unless things went badly.
3103  */
3104 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
3105                                      struct btrfs_root *root,
3106                                      struct btrfs_path *path,
3107                                      struct inode *inode, u64 new_size)
3108 {
3109         struct btrfs_key key;
3110         int ret;
3111         int nritems;
3112         struct btrfs_key found_key;
3113         struct btrfs_key other_key;
3114         struct btrfs_leaf_ref *ref;
3115         u64 leaf_gen;
3116         u64 leaf_start;
3117
3118         path->lowest_level = 1;
3119         key.objectid = inode->i_ino;
3120         key.type = BTRFS_CSUM_ITEM_KEY;
3121         key.offset = new_size;
3122 again:
3123         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3124         if (ret < 0)
3125                 goto out;
3126
3127         if (path->nodes[1] == NULL) {
3128                 ret = 0;
3129                 goto out;
3130         }
3131         ret = 0;
3132         btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
3133         nritems = btrfs_header_nritems(path->nodes[1]);
3134
3135         if (!nritems)
3136                 goto out;
3137
3138         if (path->slots[1] >= nritems)
3139                 goto next_node;
3140
3141         /* did we find a key greater than anything we want to delete? */
3142         if (found_key.objectid > inode->i_ino ||
3143            (found_key.objectid == inode->i_ino && found_key.type > key.type))
3144                 goto out;
3145
3146         /* we check the next key in the node to make sure the leave contains
3147          * only checksum items.  This comparison doesn't work if our
3148          * leaf is the last one in the node
3149          */
3150         if (path->slots[1] + 1 >= nritems) {
3151 next_node:
3152                 /* search forward from the last key in the node, this
3153                  * will bring us into the next node in the tree
3154                  */
3155                 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
3156
3157                 /* unlikely, but we inc below, so check to be safe */
3158                 if (found_key.offset == (u64)-1)
3159                         goto out;
3160
3161                 /* search_forward needs a path with locks held, do the
3162                  * search again for the original key.  It is possible
3163                  * this will race with a balance and return a path that
3164                  * we could modify, but this drop is just an optimization
3165                  * and is allowed to miss some leaves.
3166                  */
3167                 btrfs_release_path(root, path);
3168                 found_key.offset++;
3169
3170                 /* setup a max key for search_forward */
3171                 other_key.offset = (u64)-1;
3172                 other_key.type = key.type;
3173                 other_key.objectid = key.objectid;
3174
3175                 path->keep_locks = 1;
3176                 ret = btrfs_search_forward(root, &found_key, &other_key,
3177                                            path, 0, 0);
3178                 path->keep_locks = 0;
3179                 if (ret || found_key.objectid != key.objectid ||
3180                     found_key.type != key.type) {
3181                         ret = 0;
3182                         goto out;
3183                 }
3184
3185                 key.offset = found_key.offset;
3186                 btrfs_release_path(root, path);
3187                 cond_resched();
3188                 goto again;
3189         }
3190
3191         /* we know there's one more slot after us in the tree,
3192          * read that key so we can verify it is also a checksum item
3193          */
3194         btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
3195
3196         if (found_key.objectid < inode->i_ino)
3197                 goto next_key;
3198
3199         if (found_key.type != key.type || found_key.offset < new_size)
3200                 goto next_key;
3201
3202         /*
3203          * if the key for the next leaf isn't a csum key from this objectid,
3204          * we can't be sure there aren't good items inside this leaf.
3205          * Bail out
3206          */
3207         if (other_key.objectid != inode->i_ino || other_key.type != key.type)
3208                 goto out;
3209
3210         leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
3211         leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
3212         /*
3213          * it is safe to delete this leaf, it contains only
3214          * csum items from this inode at an offset >= new_size
3215          */
3216         ret = btrfs_del_leaf(trans, root, path, leaf_start);
3217         BUG_ON(ret);
3218
3219         if (root->ref_cows && leaf_gen < trans->transid) {
3220                 ref = btrfs_alloc_leaf_ref(root, 0);
3221                 if (ref) {
3222                         ref->root_gen = root->root_key.offset;
3223                         ref->bytenr = leaf_start;
3224                         ref->owner = 0;
3225                         ref->generation = leaf_gen;
3226                         ref->nritems = 0;
3227
3228                         btrfs_sort_leaf_ref(ref);
3229
3230                         ret = btrfs_add_leaf_ref(root, ref, 0);
3231                         WARN_ON(ret);
3232                         btrfs_free_leaf_ref(root, ref);
3233                 } else {
3234                         WARN_ON(1);
3235                 }
3236         }
3237 next_key:
3238         btrfs_release_path(root, path);
3239
3240         if (other_key.objectid == inode->i_ino &&
3241             other_key.type == key.type && other_key.offset > key.offset) {
3242                 key.offset = other_key.offset;
3243                 cond_resched();
3244                 goto again;
3245         }
3246         ret = 0;
3247 out:
3248         /* fixup any changes we've made to the path */
3249         path->lowest_level = 0;
3250         path->keep_locks = 0;
3251         btrfs_release_path(root, path);
3252         return ret;
3253 }
3254
3255 #endif
3256
3257 /*
3258  * this can truncate away extent items, csum items and directory items.
3259  * It starts at a high offset and removes keys until it can't find
3260  * any higher than new_size
3261  *
3262  * csum items that cross the new i_size are truncated to the new size
3263  * as well.
3264  *
3265  * min_type is the minimum key type to truncate down to.  If set to 0, this
3266  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3267  */
3268 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3269                                struct btrfs_root *root,
3270                                struct inode *inode,
3271                                u64 new_size, u32 min_type)
3272 {
3273         struct btrfs_path *path;
3274         struct extent_buffer *leaf;
3275         struct btrfs_file_extent_item *fi;
3276         struct btrfs_key key;
3277         struct btrfs_key found_key;
3278         u64 extent_start = 0;
3279         u64 extent_num_bytes = 0;
3280         u64 extent_offset = 0;
3281         u64 item_end = 0;
3282         u64 mask = root->sectorsize - 1;
3283         u32 found_type = (u8)-1;
3284         int found_extent;
3285         int del_item;
3286         int pending_del_nr = 0;
3287         int pending_del_slot = 0;
3288         int extent_type = -1;
3289         int encoding;
3290         int ret;
3291         int err = 0;
3292
3293         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3294
3295         if (root->ref_cows || root == root->fs_info->tree_root)
3296                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3297
3298         path = btrfs_alloc_path();
3299         BUG_ON(!path);
3300         path->reada = -1;
3301
3302         key.objectid = inode->i_ino;
3303         key.offset = (u64)-1;
3304         key.type = (u8)-1;
3305
3306 search_again:
3307         path->leave_spinning = 1;
3308         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3309         if (ret < 0) {
3310                 err = ret;
3311                 goto out;
3312         }
3313
3314         if (ret > 0) {
3315                 /* there are no items in the tree for us to truncate, we're
3316                  * done
3317                  */
3318                 if (path->slots[0] == 0)
3319                         goto out;
3320                 path->slots[0]--;
3321         }
3322
3323         while (1) {
3324                 fi = NULL;
3325                 leaf = path->nodes[0];
3326                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3327                 found_type = btrfs_key_type(&found_key);
3328                 encoding = 0;
3329
3330                 if (found_key.objectid != inode->i_ino)
3331                         break;
3332
3333                 if (found_type < min_type)
3334                         break;
3335
3336                 item_end = found_key.offset;
3337                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3338                         fi = btrfs_item_ptr(leaf, path->slots[0],
3339                                             struct btrfs_file_extent_item);
3340                         extent_type = btrfs_file_extent_type(leaf, fi);
3341                         encoding = btrfs_file_extent_compression(leaf, fi);
3342                         encoding |= btrfs_file_extent_encryption(leaf, fi);
3343                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3344
3345                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3346                                 item_end +=
3347                                     btrfs_file_extent_num_bytes(leaf, fi);
3348                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3349                                 item_end += btrfs_file_extent_inline_len(leaf,
3350                                                                          fi);
3351                         }
3352                         item_end--;
3353                 }
3354                 if (found_type > min_type) {
3355                         del_item = 1;
3356                 } else {
3357                         if (item_end < new_size)
3358                                 break;
3359                         if (found_key.offset >= new_size)
3360                                 del_item = 1;
3361                         else
3362                                 del_item = 0;
3363                 }
3364                 found_extent = 0;
3365                 /* FIXME, shrink the extent if the ref count is only 1 */
3366                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3367                         goto delete;
3368
3369                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3370                         u64 num_dec;
3371                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3372                         if (!del_item && !encoding) {
3373                                 u64 orig_num_bytes =
3374                                         btrfs_file_extent_num_bytes(leaf, fi);
3375                                 extent_num_bytes = new_size -
3376                                         found_key.offset + root->sectorsize - 1;
3377                                 extent_num_bytes = extent_num_bytes &
3378                                         ~((u64)root->sectorsize - 1);
3379                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3380                                                          extent_num_bytes);
3381                                 num_dec = (orig_num_bytes -
3382                                            extent_num_bytes);
3383                                 if (root->ref_cows && extent_start != 0)
3384                                         inode_sub_bytes(inode, num_dec);
3385                                 btrfs_mark_buffer_dirty(leaf);
3386                         } else {
3387                                 extent_num_bytes =
3388                                         btrfs_file_extent_disk_num_bytes(leaf,
3389                                                                          fi);
3390                                 extent_offset = found_key.offset -
3391                                         btrfs_file_extent_offset(leaf, fi);
3392
3393                                 /* FIXME blocksize != 4096 */
3394                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3395                                 if (extent_start != 0) {
3396                                         found_extent = 1;
3397                                         if (root->ref_cows)
3398                                                 inode_sub_bytes(inode, num_dec);
3399                                 }
3400                         }
3401                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3402                         /*
3403                          * we can't truncate inline items that have had
3404                          * special encodings
3405                          */
3406                         if (!del_item &&
3407                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3408                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3409                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3410                                 u32 size = new_size - found_key.offset;
3411
3412                                 if (root->ref_cows) {
3413                                         inode_sub_bytes(inode, item_end + 1 -
3414                                                         new_size);
3415                                 }
3416                                 size =
3417                                     btrfs_file_extent_calc_inline_size(size);
3418                                 ret = btrfs_truncate_item(trans, root, path,
3419                                                           size, 1);
3420                                 BUG_ON(ret);
3421                         } else if (root->ref_cows) {
3422                                 inode_sub_bytes(inode, item_end + 1 -
3423                                                 found_key.offset);
3424                         }
3425                 }
3426 delete:
3427                 if (del_item) {
3428                         if (!pending_del_nr) {
3429                                 /* no pending yet, add ourselves */
3430                                 pending_del_slot = path->slots[0];
3431                                 pending_del_nr = 1;
3432                         } else if (pending_del_nr &&
3433                                    path->slots[0] + 1 == pending_del_slot) {
3434                                 /* hop on the pending chunk */
3435                                 pending_del_nr++;
3436                                 pending_del_slot = path->slots[0];
3437                         } else {
3438                                 BUG();
3439                         }
3440                 } else {
3441                         break;
3442                 }
3443                 if (found_extent && (root->ref_cows ||
3444                                      root == root->fs_info->tree_root)) {
3445                         btrfs_set_path_blocking(path);
3446                         ret = btrfs_free_extent(trans, root, extent_start,
3447                                                 extent_num_bytes, 0,
3448                                                 btrfs_header_owner(leaf),
3449                                                 inode->i_ino, extent_offset);
3450                         BUG_ON(ret);
3451                 }
3452
3453                 if (found_type == BTRFS_INODE_ITEM_KEY)
3454                         break;
3455
3456                 if (path->slots[0] == 0 ||
3457                     path->slots[0] != pending_del_slot) {
3458                         if (root->ref_cows) {
3459                                 err = -EAGAIN;
3460                                 goto out;
3461                         }
3462                         if (pending_del_nr) {
3463                                 ret = btrfs_del_items(trans, root, path,
3464                                                 pending_del_slot,
3465                                                 pending_del_nr);
3466                                 BUG_ON(ret);
3467                                 pending_del_nr = 0;
3468                         }
3469                         btrfs_release_path(root, path);
3470                         goto search_again;
3471                 } else {
3472                         path->slots[0]--;
3473                 }
3474         }
3475 out:
3476         if (pending_del_nr) {
3477                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3478                                       pending_del_nr);
3479                 BUG_ON(ret);
3480         }
3481         btrfs_free_path(path);
3482         return err;
3483 }
3484
3485 /*
3486  * taken from block_truncate_page, but does cow as it zeros out
3487  * any bytes left in the last page in the file.
3488  */
3489 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3490 {
3491         struct inode *inode = mapping->host;
3492         struct btrfs_root *root = BTRFS_I(inode)->root;
3493         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3494         struct btrfs_ordered_extent *ordered;
3495         struct extent_state *cached_state = NULL;
3496         char *kaddr;
3497         u32 blocksize = root->sectorsize;
3498         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3499         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3500         struct page *page;
3501         int ret = 0;
3502         u64 page_start;
3503         u64 page_end;
3504
3505         if ((offset & (blocksize - 1)) == 0)
3506                 goto out;
3507         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3508         if (ret)
3509                 goto out;
3510
3511         ret = -ENOMEM;
3512 again:
3513         page = grab_cache_page(mapping, index);
3514         if (!page) {
3515                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3516                 goto out;
3517         }
3518
3519         page_start = page_offset(page);
3520         page_end = page_start + PAGE_CACHE_SIZE - 1;
3521
3522         if (!PageUptodate(page)) {
3523                 ret = btrfs_readpage(NULL, page);
3524                 lock_page(page);
3525                 if (page->mapping != mapping) {
3526                         unlock_page(page);
3527                         page_cache_release(page);
3528                         goto again;
3529                 }
3530                 if (!PageUptodate(page)) {
3531                         ret = -EIO;
3532                         goto out_unlock;
3533                 }
3534         }
3535         wait_on_page_writeback(page);
3536
3537         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3538                          GFP_NOFS);
3539         set_page_extent_mapped(page);
3540
3541         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3542         if (ordered) {
3543                 unlock_extent_cached(io_tree, page_start, page_end,
3544                                      &cached_state, GFP_NOFS);
3545                 unlock_page(page);
3546                 page_cache_release(page);
3547                 btrfs_start_ordered_extent(inode, ordered, 1);
3548                 btrfs_put_ordered_extent(ordered);
3549                 goto again;
3550         }
3551
3552         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3553                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3554                           0, 0, &cached_state, GFP_NOFS);
3555
3556         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3557                                         &cached_state);
3558         if (ret) {
3559                 unlock_extent_cached(io_tree, page_start, page_end,
3560                                      &cached_state, GFP_NOFS);
3561                 goto out_unlock;
3562         }
3563
3564         ret = 0;
3565         if (offset != PAGE_CACHE_SIZE) {
3566                 kaddr = kmap(page);
3567                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3568                 flush_dcache_page(page);
3569                 kunmap(page);
3570         }
3571         ClearPageChecked(page);
3572         set_page_dirty(page);
3573         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3574                              GFP_NOFS);
3575
3576 out_unlock:
3577         if (ret)
3578                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3579         unlock_page(page);
3580         page_cache_release(page);
3581 out:
3582         return ret;
3583 }
3584
3585 /*
3586  * This function puts in dummy file extents for the area we're creating a hole
3587  * for.  So if we are truncating this file to a larger size we need to insert
3588  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3589  * the range between oldsize and size
3590  */
3591 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3592 {
3593         struct btrfs_trans_handle *trans;
3594         struct btrfs_root *root = BTRFS_I(inode)->root;
3595         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3596         struct extent_map *em = NULL;
3597         struct extent_state *cached_state = NULL;
3598         u64 mask = root->sectorsize - 1;
3599         u64 hole_start = (oldsize + mask) & ~mask;
3600         u64 block_end = (size + mask) & ~mask;
3601         u64 last_byte;
3602         u64 cur_offset;
3603         u64 hole_size;
3604         int err = 0;
3605
3606         if (size <= hole_start)
3607                 return 0;
3608
3609         while (1) {
3610                 struct btrfs_ordered_extent *ordered;
3611                 btrfs_wait_ordered_range(inode, hole_start,
3612                                          block_end - hole_start);
3613                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3614                                  &cached_state, GFP_NOFS);
3615                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3616                 if (!ordered)
3617                         break;
3618                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3619                                      &cached_state, GFP_NOFS);
3620                 btrfs_put_ordered_extent(ordered);
3621         }
3622
3623         cur_offset = hole_start;
3624         while (1) {
3625                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3626                                 block_end - cur_offset, 0);
3627                 BUG_ON(IS_ERR(em) || !em);
3628                 last_byte = min(extent_map_end(em), block_end);
3629                 last_byte = (last_byte + mask) & ~mask;
3630                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3631                         u64 hint_byte = 0;
3632                         hole_size = last_byte - cur_offset;
3633
3634                         trans = btrfs_start_transaction(root, 2);
3635                         if (IS_ERR(trans)) {
3636                                 err = PTR_ERR(trans);
3637                                 break;
3638                         }
3639                         btrfs_set_trans_block_group(trans, inode);
3640
3641                         err = btrfs_drop_extents(trans, inode, cur_offset,
3642                                                  cur_offset + hole_size,
3643                                                  &hint_byte, 1);
3644                         if (err)
3645                                 break;
3646
3647                         err = btrfs_insert_file_extent(trans, root,
3648                                         inode->i_ino, cur_offset, 0,
3649                                         0, hole_size, 0, hole_size,
3650                                         0, 0, 0);
3651                         if (err)
3652                                 break;
3653
3654                         btrfs_drop_extent_cache(inode, hole_start,
3655                                         last_byte - 1, 0);
3656
3657                         btrfs_end_transaction(trans, root);
3658                 }
3659                 free_extent_map(em);
3660                 em = NULL;
3661                 cur_offset = last_byte;
3662                 if (cur_offset >= block_end)
3663                         break;
3664         }
3665
3666         free_extent_map(em);
3667         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3668                              GFP_NOFS);
3669         return err;
3670 }
3671
3672 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3673 {
3674         loff_t oldsize = i_size_read(inode);
3675         int ret;
3676
3677         if (newsize == oldsize)
3678                 return 0;
3679
3680         if (newsize > oldsize) {
3681                 i_size_write(inode, newsize);
3682                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3683                 truncate_pagecache(inode, oldsize, newsize);
3684                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3685                 if (ret) {
3686                         btrfs_setsize(inode, oldsize);
3687                         return ret;
3688                 }
3689
3690                 mark_inode_dirty(inode);
3691         } else {
3692
3693                 /*
3694                  * We're truncating a file that used to have good data down to
3695                  * zero. Make sure it gets into the ordered flush list so that
3696                  * any new writes get down to disk quickly.
3697                  */
3698                 if (newsize == 0)
3699                         BTRFS_I(inode)->ordered_data_close = 1;
3700
3701                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3702                 truncate_setsize(inode, newsize);
3703                 ret = btrfs_truncate(inode);
3704         }
3705
3706         return ret;
3707 }
3708
3709 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3710 {
3711         struct inode *inode = dentry->d_inode;
3712         struct btrfs_root *root = BTRFS_I(inode)->root;
3713         int err;
3714
3715         if (btrfs_root_readonly(root))
3716                 return -EROFS;
3717
3718         err = inode_change_ok(inode, attr);
3719         if (err)
3720                 return err;
3721
3722         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3723                 err = btrfs_setsize(inode, attr->ia_size);
3724                 if (err)
3725                         return err;
3726         }
3727
3728         if (attr->ia_valid) {
3729                 setattr_copy(inode, attr);
3730                 mark_inode_dirty(inode);
3731
3732                 if (attr->ia_valid & ATTR_MODE)
3733                         err = btrfs_acl_chmod(inode);
3734         }
3735
3736         return err;
3737 }
3738
3739 void btrfs_evict_inode(struct inode *inode)
3740 {
3741         struct btrfs_trans_handle *trans;
3742         struct btrfs_root *root = BTRFS_I(inode)->root;
3743         unsigned long nr;
3744         int ret;
3745
3746         trace_btrfs_inode_evict(inode);
3747
3748         truncate_inode_pages(&inode->i_data, 0);
3749         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3750                                root == root->fs_info->tree_root))
3751                 goto no_delete;
3752
3753         if (is_bad_inode(inode)) {
3754                 btrfs_orphan_del(NULL, inode);
3755                 goto no_delete;
3756         }
3757         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3758         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3759
3760         if (root->fs_info->log_root_recovering) {
3761                 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3762                 goto no_delete;
3763         }
3764
3765         if (inode->i_nlink > 0) {
3766                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3767                 goto no_delete;
3768         }
3769
3770         btrfs_i_size_write(inode, 0);
3771
3772         while (1) {
3773                 trans = btrfs_start_transaction(root, 0);
3774                 BUG_ON(IS_ERR(trans));
3775                 btrfs_set_trans_block_group(trans, inode);
3776                 trans->block_rsv = root->orphan_block_rsv;
3777
3778                 ret = btrfs_block_rsv_check(trans, root,
3779                                             root->orphan_block_rsv, 0, 5);
3780                 if (ret) {
3781                         BUG_ON(ret != -EAGAIN);
3782                         ret = btrfs_commit_transaction(trans, root);
3783                         BUG_ON(ret);
3784                         continue;
3785                 }
3786
3787                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3788                 if (ret != -EAGAIN)
3789                         break;
3790
3791                 nr = trans->blocks_used;
3792                 btrfs_end_transaction(trans, root);
3793                 trans = NULL;
3794                 btrfs_btree_balance_dirty(root, nr);
3795
3796         }
3797
3798         if (ret == 0) {
3799                 ret = btrfs_orphan_del(trans, inode);
3800                 BUG_ON(ret);
3801         }
3802
3803         nr = trans->blocks_used;
3804         btrfs_end_transaction(trans, root);
3805         btrfs_btree_balance_dirty(root, nr);
3806 no_delete:
3807         end_writeback(inode);
3808         return;
3809 }
3810
3811 /*
3812  * this returns the key found in the dir entry in the location pointer.
3813  * If no dir entries were found, location->objectid is 0.
3814  */
3815 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3816                                struct btrfs_key *location)
3817 {
3818         const char *name = dentry->d_name.name;
3819         int namelen = dentry->d_name.len;
3820         struct btrfs_dir_item *di;
3821         struct btrfs_path *path;
3822         struct btrfs_root *root = BTRFS_I(dir)->root;
3823         int ret = 0;
3824
3825         path = btrfs_alloc_path();
3826         BUG_ON(!path);
3827
3828         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3829                                     namelen, 0);
3830         if (IS_ERR(di))
3831                 ret = PTR_ERR(di);
3832
3833         if (!di || IS_ERR(di))
3834                 goto out_err;
3835
3836         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3837 out:
3838         btrfs_free_path(path);
3839         return ret;
3840 out_err:
3841         location->objectid = 0;
3842         goto out;
3843 }
3844
3845 /*
3846  * when we hit a tree root in a directory, the btrfs part of the inode
3847  * needs to be changed to reflect the root directory of the tree root.  This
3848  * is kind of like crossing a mount point.
3849  */
3850 static int fixup_tree_root_location(struct btrfs_root *root,
3851                                     struct inode *dir,
3852                                     struct dentry *dentry,
3853                                     struct btrfs_key *location,
3854                                     struct btrfs_root **sub_root)
3855 {
3856         struct btrfs_path *path;
3857         struct btrfs_root *new_root;
3858         struct btrfs_root_ref *ref;
3859         struct extent_buffer *leaf;
3860         int ret;
3861         int err = 0;
3862
3863         path = btrfs_alloc_path();
3864         if (!path) {
3865                 err = -ENOMEM;
3866                 goto out;
3867         }
3868
3869         err = -ENOENT;
3870         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3871                                   BTRFS_I(dir)->root->root_key.objectid,
3872                                   location->objectid);
3873         if (ret) {
3874                 if (ret < 0)
3875                         err = ret;
3876                 goto out;
3877         }
3878
3879         leaf = path->nodes[0];
3880         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3881         if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino ||
3882             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3883                 goto out;
3884
3885         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3886                                    (unsigned long)(ref + 1),
3887                                    dentry->d_name.len);
3888         if (ret)
3889                 goto out;
3890
3891         btrfs_release_path(root->fs_info->tree_root, path);
3892
3893         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3894         if (IS_ERR(new_root)) {
3895                 err = PTR_ERR(new_root);
3896                 goto out;
3897         }
3898
3899         if (btrfs_root_refs(&new_root->root_item) == 0) {
3900                 err = -ENOENT;
3901                 goto out;
3902         }
3903
3904         *sub_root = new_root;
3905         location->objectid = btrfs_root_dirid(&new_root->root_item);
3906         location->type = BTRFS_INODE_ITEM_KEY;
3907         location->offset = 0;
3908         err = 0;
3909 out:
3910         btrfs_free_path(path);
3911         return err;
3912 }
3913
3914 static void inode_tree_add(struct inode *inode)
3915 {
3916         struct btrfs_root *root = BTRFS_I(inode)->root;
3917         struct btrfs_inode *entry;
3918         struct rb_node **p;
3919         struct rb_node *parent;
3920 again:
3921         p = &root->inode_tree.rb_node;
3922         parent = NULL;
3923
3924         if (inode_unhashed(inode))
3925                 return;
3926
3927         spin_lock(&root->inode_lock);
3928         while (*p) {
3929                 parent = *p;
3930                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3931
3932                 if (inode->i_ino < entry->vfs_inode.i_ino)
3933                         p = &parent->rb_left;
3934                 else if (inode->i_ino > entry->vfs_inode.i_ino)
3935                         p = &parent->rb_right;
3936                 else {
3937                         WARN_ON(!(entry->vfs_inode.i_state &
3938                                   (I_WILL_FREE | I_FREEING)));
3939                         rb_erase(parent, &root->inode_tree);
3940                         RB_CLEAR_NODE(parent);
3941                         spin_unlock(&root->inode_lock);
3942                         goto again;
3943                 }
3944         }
3945         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3946         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3947         spin_unlock(&root->inode_lock);
3948 }
3949
3950 static void inode_tree_del(struct inode *inode)
3951 {
3952         struct btrfs_root *root = BTRFS_I(inode)->root;
3953         int empty = 0;
3954
3955         spin_lock(&root->inode_lock);
3956         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3957                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3958                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3959                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3960         }
3961         spin_unlock(&root->inode_lock);
3962
3963         /*
3964          * Free space cache has inodes in the tree root, but the tree root has a
3965          * root_refs of 0, so this could end up dropping the tree root as a
3966          * snapshot, so we need the extra !root->fs_info->tree_root check to
3967          * make sure we don't drop it.
3968          */
3969         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3970             root != root->fs_info->tree_root) {
3971                 synchronize_srcu(&root->fs_info->subvol_srcu);
3972                 spin_lock(&root->inode_lock);
3973                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3974                 spin_unlock(&root->inode_lock);
3975                 if (empty)
3976                         btrfs_add_dead_root(root);
3977         }
3978 }
3979
3980 int btrfs_invalidate_inodes(struct btrfs_root *root)
3981 {
3982         struct rb_node *node;
3983         struct rb_node *prev;
3984         struct btrfs_inode *entry;
3985         struct inode *inode;
3986         u64 objectid = 0;
3987
3988         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3989
3990         spin_lock(&root->inode_lock);
3991 again:
3992         node = root->inode_tree.rb_node;
3993         prev = NULL;
3994         while (node) {
3995                 prev = node;
3996                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3997
3998                 if (objectid < entry->vfs_inode.i_ino)
3999                         node = node->rb_left;
4000                 else if (objectid > entry->vfs_inode.i_ino)
4001                         node = node->rb_right;
4002                 else
4003                         break;
4004         }
4005         if (!node) {
4006                 while (prev) {
4007                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
4008                         if (objectid <= entry->vfs_inode.i_ino) {
4009                                 node = prev;
4010                                 break;
4011                         }
4012                         prev = rb_next(prev);
4013                 }
4014         }
4015         while (node) {
4016                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4017                 objectid = entry->vfs_inode.i_ino + 1;
4018                 inode = igrab(&entry->vfs_inode);
4019                 if (inode) {
4020                         spin_unlock(&root->inode_lock);
4021                         if (atomic_read(&inode->i_count) > 1)
4022                                 d_prune_aliases(inode);
4023                         /*
4024                          * btrfs_drop_inode will have it removed from
4025                          * the inode cache when its usage count
4026                          * hits zero.
4027                          */
4028                         iput(inode);
4029                         cond_resched();
4030                         spin_lock(&root->inode_lock);
4031                         goto again;
4032                 }
4033
4034                 if (cond_resched_lock(&root->inode_lock))
4035                         goto again;
4036
4037                 node = rb_next(node);
4038         }
4039         spin_unlock(&root->inode_lock);
4040         return 0;
4041 }
4042
4043 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4044 {
4045         struct btrfs_iget_args *args = p;
4046         inode->i_ino = args->ino;
4047         BTRFS_I(inode)->root = args->root;
4048         btrfs_set_inode_space_info(args->root, inode);
4049         return 0;
4050 }
4051
4052 static int btrfs_find_actor(struct inode *inode, void *opaque)
4053 {
4054         struct btrfs_iget_args *args = opaque;
4055         return args->ino == inode->i_ino &&
4056                 args->root == BTRFS_I(inode)->root;
4057 }
4058
4059 static struct inode *btrfs_iget_locked(struct super_block *s,
4060                                        u64 objectid,
4061                                        struct btrfs_root *root)
4062 {
4063         struct inode *inode;
4064         struct btrfs_iget_args args;
4065         args.ino = objectid;
4066         args.root = root;
4067
4068         inode = iget5_locked(s, objectid, btrfs_find_actor,
4069                              btrfs_init_locked_inode,
4070                              (void *)&args);
4071         return inode;
4072 }
4073
4074 /* Get an inode object given its location and corresponding root.
4075  * Returns in *is_new if the inode was read from disk
4076  */
4077 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4078                          struct btrfs_root *root, int *new)
4079 {
4080         struct inode *inode;
4081
4082         inode = btrfs_iget_locked(s, location->objectid, root);
4083         if (!inode)
4084                 return ERR_PTR(-ENOMEM);
4085
4086         if (inode->i_state & I_NEW) {
4087                 BTRFS_I(inode)->root = root;
4088                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4089                 btrfs_read_locked_inode(inode);
4090                 inode_tree_add(inode);
4091                 unlock_new_inode(inode);
4092                 if (new)
4093                         *new = 1;
4094         }
4095
4096         return inode;
4097 }
4098
4099 static struct inode *new_simple_dir(struct super_block *s,
4100                                     struct btrfs_key *key,
4101                                     struct btrfs_root *root)
4102 {
4103         struct inode *inode = new_inode(s);
4104
4105         if (!inode)
4106                 return ERR_PTR(-ENOMEM);
4107
4108         BTRFS_I(inode)->root = root;
4109         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4110         BTRFS_I(inode)->dummy_inode = 1;
4111
4112         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4113         inode->i_op = &simple_dir_inode_operations;
4114         inode->i_fop = &simple_dir_operations;
4115         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4116         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4117
4118         return inode;
4119 }
4120
4121 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4122 {
4123         struct inode *inode;
4124         struct btrfs_root *root = BTRFS_I(dir)->root;
4125         struct btrfs_root *sub_root = root;
4126         struct btrfs_key location;
4127         int index;
4128         int ret;
4129
4130         if (dentry->d_name.len > BTRFS_NAME_LEN)
4131                 return ERR_PTR(-ENAMETOOLONG);
4132
4133         ret = btrfs_inode_by_name(dir, dentry, &location);
4134
4135         if (ret < 0)
4136                 return ERR_PTR(ret);
4137
4138         if (location.objectid == 0)
4139                 return NULL;
4140
4141         if (location.type == BTRFS_INODE_ITEM_KEY) {
4142                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4143                 return inode;
4144         }
4145
4146         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4147
4148         index = srcu_read_lock(&root->fs_info->subvol_srcu);
4149         ret = fixup_tree_root_location(root, dir, dentry,
4150                                        &location, &sub_root);
4151         if (ret < 0) {
4152                 if (ret != -ENOENT)
4153                         inode = ERR_PTR(ret);
4154                 else
4155                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
4156         } else {
4157                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4158         }
4159         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4160
4161         if (!IS_ERR(inode) && root != sub_root) {
4162                 down_read(&root->fs_info->cleanup_work_sem);
4163                 if (!(inode->i_sb->s_flags & MS_RDONLY))
4164                         ret = btrfs_orphan_cleanup(sub_root);
4165                 up_read(&root->fs_info->cleanup_work_sem);
4166                 if (ret)
4167                         inode = ERR_PTR(ret);
4168         }
4169
4170         return inode;
4171 }
4172
4173 static int btrfs_dentry_delete(const struct dentry *dentry)
4174 {
4175         struct btrfs_root *root;
4176
4177         if (!dentry->d_inode && !IS_ROOT(dentry))
4178                 dentry = dentry->d_parent;
4179
4180         if (dentry->d_inode) {
4181                 root = BTRFS_I(dentry->d_inode)->root;
4182                 if (btrfs_root_refs(&root->root_item) == 0)
4183                         return 1;
4184         }
4185         return 0;
4186 }
4187
4188 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4189                                    struct nameidata *nd)
4190 {
4191         struct inode *inode;
4192
4193         inode = btrfs_lookup_dentry(dir, dentry);
4194         if (IS_ERR(inode))
4195                 return ERR_CAST(inode);
4196
4197         return d_splice_alias(inode, dentry);
4198 }
4199
4200 static unsigned char btrfs_filetype_table[] = {
4201         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4202 };
4203
4204 static int btrfs_real_readdir(struct file *filp, void *dirent,
4205                               filldir_t filldir)
4206 {
4207         struct inode *inode = filp->f_dentry->d_inode;
4208         struct btrfs_root *root = BTRFS_I(inode)->root;
4209         struct btrfs_item *item;
4210         struct btrfs_dir_item *di;
4211         struct btrfs_key key;
4212         struct btrfs_key found_key;
4213         struct btrfs_path *path;
4214         int ret;
4215         u32 nritems;
4216         struct extent_buffer *leaf;
4217         int slot;
4218         int advance;
4219         unsigned char d_type;
4220         int over = 0;
4221         u32 di_cur;
4222         u32 di_total;
4223         u32 di_len;
4224         int key_type = BTRFS_DIR_INDEX_KEY;
4225         char tmp_name[32];
4226         char *name_ptr;
4227         int name_len;
4228
4229         /* FIXME, use a real flag for deciding about the key type */
4230         if (root->fs_info->tree_root == root)
4231                 key_type = BTRFS_DIR_ITEM_KEY;
4232
4233         /* special case for "." */
4234         if (filp->f_pos == 0) {
4235                 over = filldir(dirent, ".", 1,
4236                                1, inode->i_ino,
4237                                DT_DIR);
4238                 if (over)
4239                         return 0;
4240                 filp->f_pos = 1;
4241         }
4242         /* special case for .., just use the back ref */
4243         if (filp->f_pos == 1) {
4244                 u64 pino = parent_ino(filp->f_path.dentry);
4245                 over = filldir(dirent, "..", 2,
4246                                2, pino, DT_DIR);
4247                 if (over)
4248                         return 0;
4249                 filp->f_pos = 2;
4250         }
4251         path = btrfs_alloc_path();
4252         path->reada = 2;
4253
4254         btrfs_set_key_type(&key, key_type);
4255         key.offset = filp->f_pos;
4256         key.objectid = inode->i_ino;
4257
4258         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4259         if (ret < 0)
4260                 goto err;
4261         advance = 0;
4262
4263         while (1) {
4264                 leaf = path->nodes[0];
4265                 nritems = btrfs_header_nritems(leaf);
4266                 slot = path->slots[0];
4267                 if (advance || slot >= nritems) {
4268                         if (slot >= nritems - 1) {
4269                                 ret = btrfs_next_leaf(root, path);
4270                                 if (ret)
4271                                         break;
4272                                 leaf = path->nodes[0];
4273                                 nritems = btrfs_header_nritems(leaf);
4274                                 slot = path->slots[0];
4275                         } else {
4276                                 slot++;
4277                                 path->slots[0]++;
4278                         }
4279                 }
4280
4281                 advance = 1;
4282                 item = btrfs_item_nr(leaf, slot);
4283                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4284
4285                 if (found_key.objectid != key.objectid)
4286                         break;
4287                 if (btrfs_key_type(&found_key) != key_type)
4288                         break;
4289                 if (found_key.offset < filp->f_pos)
4290                         continue;
4291
4292                 filp->f_pos = found_key.offset;
4293
4294                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4295                 di_cur = 0;
4296                 di_total = btrfs_item_size(leaf, item);
4297
4298                 while (di_cur < di_total) {
4299                         struct btrfs_key location;
4300
4301                         if (verify_dir_item(root, leaf, di))
4302                                 break;
4303
4304                         name_len = btrfs_dir_name_len(leaf, di);
4305                         if (name_len <= sizeof(tmp_name)) {
4306                                 name_ptr = tmp_name;
4307                         } else {
4308                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4309                                 if (!name_ptr) {
4310                                         ret = -ENOMEM;
4311                                         goto err;
4312                                 }
4313                         }
4314                         read_extent_buffer(leaf, name_ptr,
4315                                            (unsigned long)(di + 1), name_len);
4316
4317                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4318                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4319
4320                         /* is this a reference to our own snapshot? If so
4321                          * skip it
4322                          */
4323                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4324                             location.objectid == root->root_key.objectid) {
4325                                 over = 0;
4326                                 goto skip;
4327                         }
4328                         over = filldir(dirent, name_ptr, name_len,
4329                                        found_key.offset, location.objectid,
4330                                        d_type);
4331
4332 skip:
4333                         if (name_ptr != tmp_name)
4334                                 kfree(name_ptr);
4335
4336                         if (over)
4337                                 goto nopos;
4338                         di_len = btrfs_dir_name_len(leaf, di) +
4339                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4340                         di_cur += di_len;
4341                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4342                 }
4343         }
4344
4345         /* Reached end of directory/root. Bump pos past the last item. */
4346         if (key_type == BTRFS_DIR_INDEX_KEY)
4347                 /*
4348                  * 32-bit glibc will use getdents64, but then strtol -
4349                  * so the last number we can serve is this.
4350                  */
4351                 filp->f_pos = 0x7fffffff;
4352         else
4353                 filp->f_pos++;
4354 nopos:
4355         ret = 0;
4356 err:
4357         btrfs_free_path(path);
4358         return ret;
4359 }
4360
4361 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4362 {
4363         struct btrfs_root *root = BTRFS_I(inode)->root;
4364         struct btrfs_trans_handle *trans;
4365         int ret = 0;
4366         bool nolock = false;
4367
4368         if (BTRFS_I(inode)->dummy_inode)
4369                 return 0;
4370
4371         smp_mb();
4372         nolock = (root->fs_info->closing && root == root->fs_info->tree_root);
4373
4374         if (wbc->sync_mode == WB_SYNC_ALL) {
4375                 if (nolock)
4376                         trans = btrfs_join_transaction_nolock(root, 1);
4377                 else
4378                         trans = btrfs_join_transaction(root, 1);
4379                 if (IS_ERR(trans))
4380                         return PTR_ERR(trans);
4381                 btrfs_set_trans_block_group(trans, inode);
4382                 if (nolock)
4383                         ret = btrfs_end_transaction_nolock(trans, root);
4384                 else
4385                         ret = btrfs_commit_transaction(trans, root);
4386         }
4387         return ret;
4388 }
4389
4390 /*
4391  * This is somewhat expensive, updating the tree every time the
4392  * inode changes.  But, it is most likely to find the inode in cache.
4393  * FIXME, needs more benchmarking...there are no reasons other than performance
4394  * to keep or drop this code.
4395  */
4396 void btrfs_dirty_inode(struct inode *inode)
4397 {
4398         struct btrfs_root *root = BTRFS_I(inode)->root;
4399         struct btrfs_trans_handle *trans;
4400         int ret;
4401
4402         if (BTRFS_I(inode)->dummy_inode)
4403                 return;
4404
4405         trans = btrfs_join_transaction(root, 1);
4406         BUG_ON(IS_ERR(trans));
4407         btrfs_set_trans_block_group(trans, inode);
4408
4409         ret = btrfs_update_inode(trans, root, inode);
4410         if (ret && ret == -ENOSPC) {
4411                 /* whoops, lets try again with the full transaction */
4412                 btrfs_end_transaction(trans, root);
4413                 trans = btrfs_start_transaction(root, 1);
4414                 if (IS_ERR(trans)) {
4415                         if (printk_ratelimit()) {
4416                                 printk(KERN_ERR "btrfs: fail to "
4417                                        "dirty  inode %lu error %ld\n",
4418                                        inode->i_ino, PTR_ERR(trans));
4419                         }
4420                         return;
4421                 }
4422                 btrfs_set_trans_block_group(trans, inode);
4423
4424                 ret = btrfs_update_inode(trans, root, inode);
4425                 if (ret) {
4426                         if (printk_ratelimit()) {
4427                                 printk(KERN_ERR "btrfs: fail to "
4428                                        "dirty  inode %lu error %d\n",
4429                                        inode->i_ino, ret);
4430                         }
4431                 }
4432         }
4433         btrfs_end_transaction(trans, root);
4434 }
4435
4436 /*
4437  * find the highest existing sequence number in a directory
4438  * and then set the in-memory index_cnt variable to reflect
4439  * free sequence numbers
4440  */
4441 static int btrfs_set_inode_index_count(struct inode *inode)
4442 {
4443         struct btrfs_root *root = BTRFS_I(inode)->root;
4444         struct btrfs_key key, found_key;
4445         struct btrfs_path *path;
4446         struct extent_buffer *leaf;
4447         int ret;
4448
4449         key.objectid = inode->i_ino;
4450         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4451         key.offset = (u64)-1;
4452
4453         path = btrfs_alloc_path();
4454         if (!path)
4455                 return -ENOMEM;
4456
4457         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4458         if (ret < 0)
4459                 goto out;
4460         /* FIXME: we should be able to handle this */
4461         if (ret == 0)
4462                 goto out;
4463         ret = 0;
4464
4465         /*
4466          * MAGIC NUMBER EXPLANATION:
4467          * since we search a directory based on f_pos we have to start at 2
4468          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4469          * else has to start at 2
4470          */
4471         if (path->slots[0] == 0) {
4472                 BTRFS_I(inode)->index_cnt = 2;
4473                 goto out;
4474         }
4475
4476         path->slots[0]--;
4477
4478         leaf = path->nodes[0];
4479         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4480
4481         if (found_key.objectid != inode->i_ino ||
4482             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4483                 BTRFS_I(inode)->index_cnt = 2;
4484                 goto out;
4485         }
4486
4487         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4488 out:
4489         btrfs_free_path(path);
4490         return ret;
4491 }
4492
4493 /*
4494  * helper to find a free sequence number in a given directory.  This current
4495  * code is very simple, later versions will do smarter things in the btree
4496  */
4497 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4498 {
4499         int ret = 0;
4500
4501         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4502                 ret = btrfs_set_inode_index_count(dir);
4503                 if (ret)
4504                         return ret;
4505         }
4506
4507         *index = BTRFS_I(dir)->index_cnt;
4508         BTRFS_I(dir)->index_cnt++;
4509
4510         return ret;
4511 }
4512
4513 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4514                                      struct btrfs_root *root,
4515                                      struct inode *dir,
4516                                      const char *name, int name_len,
4517                                      u64 ref_objectid, u64 objectid,
4518                                      u64 alloc_hint, int mode, u64 *index)
4519 {
4520         struct inode *inode;
4521         struct btrfs_inode_item *inode_item;
4522         struct btrfs_key *location;
4523         struct btrfs_path *path;
4524         struct btrfs_inode_ref *ref;
4525         struct btrfs_key key[2];
4526         u32 sizes[2];
4527         unsigned long ptr;
4528         int ret;
4529         int owner;
4530
4531         path = btrfs_alloc_path();
4532         BUG_ON(!path);
4533
4534         inode = new_inode(root->fs_info->sb);
4535         if (!inode)
4536                 return ERR_PTR(-ENOMEM);
4537
4538         if (dir) {
4539                 trace_btrfs_inode_request(dir);
4540
4541                 ret = btrfs_set_inode_index(dir, index);
4542                 if (ret) {
4543                         iput(inode);
4544                         return ERR_PTR(ret);
4545                 }
4546         }
4547         /*
4548          * index_cnt is ignored for everything but a dir,
4549          * btrfs_get_inode_index_count has an explanation for the magic
4550          * number
4551          */
4552         BTRFS_I(inode)->index_cnt = 2;
4553         BTRFS_I(inode)->root = root;
4554         BTRFS_I(inode)->generation = trans->transid;
4555         inode->i_generation = BTRFS_I(inode)->generation;
4556         btrfs_set_inode_space_info(root, inode);
4557
4558         if (mode & S_IFDIR)
4559                 owner = 0;
4560         else
4561                 owner = 1;
4562         BTRFS_I(inode)->block_group =
4563                         btrfs_find_block_group(root, 0, alloc_hint, owner);
4564
4565         key[0].objectid = objectid;
4566         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4567         key[0].offset = 0;
4568
4569         key[1].objectid = objectid;
4570         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4571         key[1].offset = ref_objectid;
4572
4573         sizes[0] = sizeof(struct btrfs_inode_item);
4574         sizes[1] = name_len + sizeof(*ref);
4575
4576         path->leave_spinning = 1;
4577         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4578         if (ret != 0)
4579                 goto fail;
4580
4581         inode_init_owner(inode, dir, mode);
4582         inode->i_ino = objectid;
4583         inode_set_bytes(inode, 0);
4584         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4585         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4586                                   struct btrfs_inode_item);
4587         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4588
4589         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4590                              struct btrfs_inode_ref);
4591         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4592         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4593         ptr = (unsigned long)(ref + 1);
4594         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4595
4596         btrfs_mark_buffer_dirty(path->nodes[0]);
4597         btrfs_free_path(path);
4598
4599         location = &BTRFS_I(inode)->location;
4600         location->objectid = objectid;
4601         location->offset = 0;
4602         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4603
4604         btrfs_inherit_iflags(inode, dir);
4605
4606         if ((mode & S_IFREG)) {
4607                 if (btrfs_test_opt(root, NODATASUM))
4608                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4609                 if (btrfs_test_opt(root, NODATACOW) ||
4610                     (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4611                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4612         }
4613
4614         insert_inode_hash(inode);
4615         inode_tree_add(inode);
4616
4617         trace_btrfs_inode_new(inode);
4618
4619         return inode;
4620 fail:
4621         if (dir)
4622                 BTRFS_I(dir)->index_cnt--;
4623         btrfs_free_path(path);
4624         iput(inode);
4625         return ERR_PTR(ret);
4626 }
4627
4628 static inline u8 btrfs_inode_type(struct inode *inode)
4629 {
4630         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4631 }
4632
4633 /*
4634  * utility function to add 'inode' into 'parent_inode' with
4635  * a give name and a given sequence number.
4636  * if 'add_backref' is true, also insert a backref from the
4637  * inode to the parent directory.
4638  */
4639 int btrfs_add_link(struct btrfs_trans_handle *trans,
4640                    struct inode *parent_inode, struct inode *inode,
4641                    const char *name, int name_len, int add_backref, u64 index)
4642 {
4643         int ret = 0;
4644         struct btrfs_key key;
4645         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4646
4647         if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4648                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4649         } else {
4650                 key.objectid = inode->i_ino;
4651                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4652                 key.offset = 0;
4653         }
4654
4655         if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4656                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4657                                          key.objectid, root->root_key.objectid,
4658                                          parent_inode->i_ino,
4659                                          index, name, name_len);
4660         } else if (add_backref) {
4661                 ret = btrfs_insert_inode_ref(trans, root,
4662                                              name, name_len, inode->i_ino,
4663                                              parent_inode->i_ino, index);
4664         }
4665
4666         if (ret == 0) {
4667                 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4668                                             parent_inode->i_ino, &key,
4669                                             btrfs_inode_type(inode), index);
4670                 BUG_ON(ret);
4671
4672                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4673                                    name_len * 2);
4674                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4675                 ret = btrfs_update_inode(trans, root, parent_inode);
4676         }
4677         return ret;
4678 }
4679
4680 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4681                             struct inode *dir, struct dentry *dentry,
4682                             struct inode *inode, int backref, u64 index)
4683 {
4684         int err = btrfs_add_link(trans, dir, inode,
4685                                  dentry->d_name.name, dentry->d_name.len,
4686                                  backref, index);
4687         if (!err) {
4688                 d_instantiate(dentry, inode);
4689                 return 0;
4690         }
4691         if (err > 0)
4692                 err = -EEXIST;
4693         return err;
4694 }
4695
4696 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4697                         int mode, dev_t rdev)
4698 {
4699         struct btrfs_trans_handle *trans;
4700         struct btrfs_root *root = BTRFS_I(dir)->root;
4701         struct inode *inode = NULL;
4702         int err;
4703         int drop_inode = 0;
4704         u64 objectid;
4705         unsigned long nr = 0;
4706         u64 index = 0;
4707
4708         if (!new_valid_dev(rdev))
4709                 return -EINVAL;
4710
4711         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4712         if (err)
4713                 return err;
4714
4715         /*
4716          * 2 for inode item and ref
4717          * 2 for dir items
4718          * 1 for xattr if selinux is on
4719          */
4720         trans = btrfs_start_transaction(root, 5);
4721         if (IS_ERR(trans))
4722                 return PTR_ERR(trans);
4723
4724         btrfs_set_trans_block_group(trans, dir);
4725
4726         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4727                                 dentry->d_name.len, dir->i_ino, objectid,
4728                                 BTRFS_I(dir)->block_group, mode, &index);
4729         err = PTR_ERR(inode);
4730         if (IS_ERR(inode))
4731                 goto out_unlock;
4732
4733         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4734         if (err) {
4735                 drop_inode = 1;
4736                 goto out_unlock;
4737         }
4738
4739         btrfs_set_trans_block_group(trans, inode);
4740         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4741         if (err)
4742                 drop_inode = 1;
4743         else {
4744                 inode->i_op = &btrfs_special_inode_operations;
4745                 init_special_inode(inode, inode->i_mode, rdev);
4746                 btrfs_update_inode(trans, root, inode);
4747         }
4748         btrfs_update_inode_block_group(trans, inode);
4749         btrfs_update_inode_block_group(trans, dir);
4750 out_unlock:
4751         nr = trans->blocks_used;
4752         btrfs_end_transaction_throttle(trans, root);
4753         btrfs_btree_balance_dirty(root, nr);
4754         if (drop_inode) {
4755                 inode_dec_link_count(inode);
4756                 iput(inode);
4757         }
4758         return err;
4759 }
4760
4761 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4762                         int mode, struct nameidata *nd)
4763 {
4764         struct btrfs_trans_handle *trans;
4765         struct btrfs_root *root = BTRFS_I(dir)->root;
4766         struct inode *inode = NULL;
4767         int drop_inode = 0;
4768         int err;
4769         unsigned long nr = 0;
4770         u64 objectid;
4771         u64 index = 0;
4772
4773         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4774         if (err)
4775                 return err;
4776         /*
4777          * 2 for inode item and ref
4778          * 2 for dir items
4779          * 1 for xattr if selinux is on
4780          */
4781         trans = btrfs_start_transaction(root, 5);
4782         if (IS_ERR(trans))
4783                 return PTR_ERR(trans);
4784
4785         btrfs_set_trans_block_group(trans, dir);
4786
4787         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4788                                 dentry->d_name.len, dir->i_ino, objectid,
4789                                 BTRFS_I(dir)->block_group, mode, &index);
4790         err = PTR_ERR(inode);
4791         if (IS_ERR(inode))
4792                 goto out_unlock;
4793
4794         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4795         if (err) {
4796                 drop_inode = 1;
4797                 goto out_unlock;
4798         }
4799
4800         btrfs_set_trans_block_group(trans, inode);
4801         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4802         if (err)
4803                 drop_inode = 1;
4804         else {
4805                 inode->i_mapping->a_ops = &btrfs_aops;
4806                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4807                 inode->i_fop = &btrfs_file_operations;
4808                 inode->i_op = &btrfs_file_inode_operations;
4809                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4810         }
4811         btrfs_update_inode_block_group(trans, inode);
4812         btrfs_update_inode_block_group(trans, dir);
4813 out_unlock:
4814         nr = trans->blocks_used;
4815         btrfs_end_transaction_throttle(trans, root);
4816         if (drop_inode) {
4817                 inode_dec_link_count(inode);
4818                 iput(inode);
4819         }
4820         btrfs_btree_balance_dirty(root, nr);
4821         return err;
4822 }
4823
4824 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4825                       struct dentry *dentry)
4826 {
4827         struct btrfs_trans_handle *trans;
4828         struct btrfs_root *root = BTRFS_I(dir)->root;
4829         struct inode *inode = old_dentry->d_inode;
4830         u64 index;
4831         unsigned long nr = 0;
4832         int err;
4833         int drop_inode = 0;
4834
4835         /* do not allow sys_link's with other subvols of the same device */
4836         if (root->objectid != BTRFS_I(inode)->root->objectid)
4837                 return -EXDEV;
4838
4839         if (inode->i_nlink == ~0U)
4840                 return -EMLINK;
4841
4842         btrfs_inc_nlink(inode);
4843         inode->i_ctime = CURRENT_TIME;
4844
4845         err = btrfs_set_inode_index(dir, &index);
4846         if (err)
4847                 goto fail;
4848
4849         /*
4850          * 2 items for inode and inode ref
4851          * 2 items for dir items
4852          * 1 item for parent inode
4853          */
4854         trans = btrfs_start_transaction(root, 5);
4855         if (IS_ERR(trans)) {
4856                 err = PTR_ERR(trans);
4857                 goto fail;
4858         }
4859
4860         btrfs_set_trans_block_group(trans, dir);
4861         ihold(inode);
4862
4863         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4864
4865         if (err) {
4866                 drop_inode = 1;
4867         } else {
4868                 struct dentry *parent = dget_parent(dentry);
4869                 btrfs_update_inode_block_group(trans, dir);
4870                 err = btrfs_update_inode(trans, root, inode);
4871                 BUG_ON(err);
4872                 btrfs_log_new_name(trans, inode, NULL, parent);
4873                 dput(parent);
4874         }
4875
4876         nr = trans->blocks_used;
4877         btrfs_end_transaction_throttle(trans, root);
4878 fail:
4879         if (drop_inode) {
4880                 inode_dec_link_count(inode);
4881                 iput(inode);
4882         }
4883         btrfs_btree_balance_dirty(root, nr);
4884         return err;
4885 }
4886
4887 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4888 {
4889         struct inode *inode = NULL;
4890         struct btrfs_trans_handle *trans;
4891         struct btrfs_root *root = BTRFS_I(dir)->root;
4892         int err = 0;
4893         int drop_on_err = 0;
4894         u64 objectid = 0;
4895         u64 index = 0;
4896         unsigned long nr = 1;
4897
4898         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4899         if (err)
4900                 return err;
4901
4902         /*
4903          * 2 items for inode and ref
4904          * 2 items for dir items
4905          * 1 for xattr if selinux is on
4906          */
4907         trans = btrfs_start_transaction(root, 5);
4908         if (IS_ERR(trans))
4909                 return PTR_ERR(trans);
4910         btrfs_set_trans_block_group(trans, dir);
4911
4912         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4913                                 dentry->d_name.len, dir->i_ino, objectid,
4914                                 BTRFS_I(dir)->block_group, S_IFDIR | mode,
4915                                 &index);
4916         if (IS_ERR(inode)) {
4917                 err = PTR_ERR(inode);
4918                 goto out_fail;
4919         }
4920
4921         drop_on_err = 1;
4922
4923         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4924         if (err)
4925                 goto out_fail;
4926
4927         inode->i_op = &btrfs_dir_inode_operations;
4928         inode->i_fop = &btrfs_dir_file_operations;
4929         btrfs_set_trans_block_group(trans, inode);
4930
4931         btrfs_i_size_write(inode, 0);
4932         err = btrfs_update_inode(trans, root, inode);
4933         if (err)
4934                 goto out_fail;
4935
4936         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4937                              dentry->d_name.len, 0, index);
4938         if (err)
4939                 goto out_fail;
4940
4941         d_instantiate(dentry, inode);
4942         drop_on_err = 0;
4943         btrfs_update_inode_block_group(trans, inode);
4944         btrfs_update_inode_block_group(trans, dir);
4945
4946 out_fail:
4947         nr = trans->blocks_used;
4948         btrfs_end_transaction_throttle(trans, root);
4949         if (drop_on_err)
4950                 iput(inode);
4951         btrfs_btree_balance_dirty(root, nr);
4952         return err;
4953 }
4954
4955 /* helper for btfs_get_extent.  Given an existing extent in the tree,
4956  * and an extent that you want to insert, deal with overlap and insert
4957  * the new extent into the tree.
4958  */
4959 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4960                                 struct extent_map *existing,
4961                                 struct extent_map *em,
4962                                 u64 map_start, u64 map_len)
4963 {
4964         u64 start_diff;
4965
4966         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4967         start_diff = map_start - em->start;
4968         em->start = map_start;
4969         em->len = map_len;
4970         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4971             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4972                 em->block_start += start_diff;
4973                 em->block_len -= start_diff;
4974         }
4975         return add_extent_mapping(em_tree, em);
4976 }
4977
4978 static noinline int uncompress_inline(struct btrfs_path *path,
4979                                       struct inode *inode, struct page *page,
4980                                       size_t pg_offset, u64 extent_offset,
4981                                       struct btrfs_file_extent_item *item)
4982 {
4983         int ret;
4984         struct extent_buffer *leaf = path->nodes[0];
4985         char *tmp;
4986         size_t max_size;
4987         unsigned long inline_size;
4988         unsigned long ptr;
4989         int compress_type;
4990
4991         WARN_ON(pg_offset != 0);
4992         compress_type = btrfs_file_extent_compression(leaf, item);
4993         max_size = btrfs_file_extent_ram_bytes(leaf, item);
4994         inline_size = btrfs_file_extent_inline_item_len(leaf,
4995                                         btrfs_item_nr(leaf, path->slots[0]));
4996         tmp = kmalloc(inline_size, GFP_NOFS);
4997         ptr = btrfs_file_extent_inline_start(item);
4998
4999         read_extent_buffer(leaf, tmp, ptr, inline_size);
5000
5001         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5002         ret = btrfs_decompress(compress_type, tmp, page,
5003                                extent_offset, inline_size, max_size);
5004         if (ret) {
5005                 char *kaddr = kmap_atomic(page, KM_USER0);
5006                 unsigned long copy_size = min_t(u64,
5007                                   PAGE_CACHE_SIZE - pg_offset,
5008                                   max_size - extent_offset);
5009                 memset(kaddr + pg_offset, 0, copy_size);
5010                 kunmap_atomic(kaddr, KM_USER0);
5011         }
5012         kfree(tmp);
5013         return 0;
5014 }
5015
5016 /*
5017  * a bit scary, this does extent mapping from logical file offset to the disk.
5018  * the ugly parts come from merging extents from the disk with the in-ram
5019  * representation.  This gets more complex because of the data=ordered code,
5020  * where the in-ram extents might be locked pending data=ordered completion.
5021  *
5022  * This also copies inline extents directly into the page.
5023  */
5024
5025 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5026                                     size_t pg_offset, u64 start, u64 len,
5027                                     int create)
5028 {
5029         int ret;
5030         int err = 0;
5031         u64 bytenr;
5032         u64 extent_start = 0;
5033         u64 extent_end = 0;
5034         u64 objectid = inode->i_ino;
5035         u32 found_type;
5036         struct btrfs_path *path = NULL;
5037         struct btrfs_root *root = BTRFS_I(inode)->root;
5038         struct btrfs_file_extent_item *item;
5039         struct extent_buffer *leaf;
5040         struct btrfs_key found_key;
5041         struct extent_map *em = NULL;
5042         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5043         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5044         struct btrfs_trans_handle *trans = NULL;
5045         int compress_type;
5046
5047 again:
5048         read_lock(&em_tree->lock);
5049         em = lookup_extent_mapping(em_tree, start, len);
5050         if (em)
5051                 em->bdev = root->fs_info->fs_devices->latest_bdev;
5052         read_unlock(&em_tree->lock);
5053
5054         if (em) {
5055                 if (em->start > start || em->start + em->len <= start)
5056                         free_extent_map(em);
5057                 else if (em->block_start == EXTENT_MAP_INLINE && page)
5058                         free_extent_map(em);
5059                 else
5060                         goto out;
5061         }
5062         em = alloc_extent_map(GFP_NOFS);
5063         if (!em) {
5064                 err = -ENOMEM;
5065                 goto out;
5066         }
5067         em->bdev = root->fs_info->fs_devices->latest_bdev;
5068         em->start = EXTENT_MAP_HOLE;
5069         em->orig_start = EXTENT_MAP_HOLE;
5070         em->len = (u64)-1;
5071         em->block_len = (u64)-1;
5072
5073         if (!path) {
5074                 path = btrfs_alloc_path();
5075                 BUG_ON(!path);
5076         }
5077
5078         ret = btrfs_lookup_file_extent(trans, root, path,
5079                                        objectid, start, trans != NULL);
5080         if (ret < 0) {
5081                 err = ret;
5082                 goto out;
5083         }
5084
5085         if (ret != 0) {
5086                 if (path->slots[0] == 0)
5087                         goto not_found;
5088                 path->slots[0]--;
5089         }
5090
5091         leaf = path->nodes[0];
5092         item = btrfs_item_ptr(leaf, path->slots[0],
5093                               struct btrfs_file_extent_item);
5094         /* are we inside the extent that was found? */
5095         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5096         found_type = btrfs_key_type(&found_key);
5097         if (found_key.objectid != objectid ||
5098             found_type != BTRFS_EXTENT_DATA_KEY) {
5099                 goto not_found;
5100         }
5101
5102         found_type = btrfs_file_extent_type(leaf, item);
5103         extent_start = found_key.offset;
5104         compress_type = btrfs_file_extent_compression(leaf, item);
5105         if (found_type == BTRFS_FILE_EXTENT_REG ||
5106             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5107                 extent_end = extent_start +
5108                        btrfs_file_extent_num_bytes(leaf, item);
5109         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5110                 size_t size;
5111                 size = btrfs_file_extent_inline_len(leaf, item);
5112                 extent_end = (extent_start + size + root->sectorsize - 1) &
5113                         ~((u64)root->sectorsize - 1);
5114         }
5115
5116         if (start >= extent_end) {
5117                 path->slots[0]++;
5118                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5119                         ret = btrfs_next_leaf(root, path);
5120                         if (ret < 0) {
5121                                 err = ret;
5122                                 goto out;
5123                         }
5124                         if (ret > 0)
5125                                 goto not_found;
5126                         leaf = path->nodes[0];
5127                 }
5128                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5129                 if (found_key.objectid != objectid ||
5130                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5131                         goto not_found;
5132                 if (start + len <= found_key.offset)
5133                         goto not_found;
5134                 em->start = start;
5135                 em->len = found_key.offset - start;
5136                 goto not_found_em;
5137         }
5138
5139         if (found_type == BTRFS_FILE_EXTENT_REG ||
5140             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5141                 em->start = extent_start;
5142                 em->len = extent_end - extent_start;
5143                 em->orig_start = extent_start -
5144                                  btrfs_file_extent_offset(leaf, item);
5145                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5146                 if (bytenr == 0) {
5147                         em->block_start = EXTENT_MAP_HOLE;
5148                         goto insert;
5149                 }
5150                 if (compress_type != BTRFS_COMPRESS_NONE) {
5151                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5152                         em->compress_type = compress_type;
5153                         em->block_start = bytenr;
5154                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5155                                                                          item);
5156                 } else {
5157                         bytenr += btrfs_file_extent_offset(leaf, item);
5158                         em->block_start = bytenr;
5159                         em->block_len = em->len;
5160                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5161                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5162                 }
5163                 goto insert;
5164         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5165                 unsigned long ptr;
5166                 char *map;
5167                 size_t size;
5168                 size_t extent_offset;
5169                 size_t copy_size;
5170
5171                 em->block_start = EXTENT_MAP_INLINE;
5172                 if (!page || create) {
5173                         em->start = extent_start;
5174                         em->len = extent_end - extent_start;
5175                         goto out;
5176                 }
5177
5178                 size = btrfs_file_extent_inline_len(leaf, item);
5179                 extent_offset = page_offset(page) + pg_offset - extent_start;
5180                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5181                                 size - extent_offset);
5182                 em->start = extent_start + extent_offset;
5183                 em->len = (copy_size + root->sectorsize - 1) &
5184                         ~((u64)root->sectorsize - 1);
5185                 em->orig_start = EXTENT_MAP_INLINE;
5186                 if (compress_type) {
5187                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5188                         em->compress_type = compress_type;
5189                 }
5190                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5191                 if (create == 0 && !PageUptodate(page)) {
5192                         if (btrfs_file_extent_compression(leaf, item) !=
5193                             BTRFS_COMPRESS_NONE) {
5194                                 ret = uncompress_inline(path, inode, page,
5195                                                         pg_offset,
5196                                                         extent_offset, item);
5197                                 BUG_ON(ret);
5198                         } else {
5199                                 map = kmap(page);
5200                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5201                                                    copy_size);
5202                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5203                                         memset(map + pg_offset + copy_size, 0,
5204                                                PAGE_CACHE_SIZE - pg_offset -
5205                                                copy_size);
5206                                 }
5207                                 kunmap(page);
5208                         }
5209                         flush_dcache_page(page);
5210                 } else if (create && PageUptodate(page)) {
5211                         WARN_ON(1);
5212                         if (!trans) {
5213                                 kunmap(page);
5214                                 free_extent_map(em);
5215                                 em = NULL;
5216                                 btrfs_release_path(root, path);
5217                                 trans = btrfs_join_transaction(root, 1);
5218                                 if (IS_ERR(trans))
5219                                         return ERR_CAST(trans);
5220                                 goto again;
5221                         }
5222                         map = kmap(page);
5223                         write_extent_buffer(leaf, map + pg_offset, ptr,
5224                                             copy_size);
5225                         kunmap(page);
5226                         btrfs_mark_buffer_dirty(leaf);
5227                 }
5228                 set_extent_uptodate(io_tree, em->start,
5229                                     extent_map_end(em) - 1, GFP_NOFS);
5230                 goto insert;
5231         } else {
5232                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5233                 WARN_ON(1);
5234         }
5235 not_found:
5236         em->start = start;
5237         em->len = len;
5238 not_found_em:
5239         em->block_start = EXTENT_MAP_HOLE;
5240         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5241 insert:
5242         btrfs_release_path(root, path);
5243         if (em->start > start || extent_map_end(em) <= start) {
5244                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5245                        "[%llu %llu]\n", (unsigned long long)em->start,
5246                        (unsigned long long)em->len,
5247                        (unsigned long long)start,
5248                        (unsigned long long)len);
5249                 err = -EIO;
5250                 goto out;
5251         }
5252
5253         err = 0;
5254         write_lock(&em_tree->lock);
5255         ret = add_extent_mapping(em_tree, em);
5256         /* it is possible that someone inserted the extent into the tree
5257          * while we had the lock dropped.  It is also possible that
5258          * an overlapping map exists in the tree
5259          */
5260         if (ret == -EEXIST) {
5261                 struct extent_map *existing;
5262
5263                 ret = 0;
5264
5265                 existing = lookup_extent_mapping(em_tree, start, len);
5266                 if (existing && (existing->start > start ||
5267                     existing->start + existing->len <= start)) {
5268                         free_extent_map(existing);
5269                         existing = NULL;
5270                 }
5271                 if (!existing) {
5272                         existing = lookup_extent_mapping(em_tree, em->start,
5273                                                          em->len);
5274                         if (existing) {
5275                                 err = merge_extent_mapping(em_tree, existing,
5276                                                            em, start,
5277                                                            root->sectorsize);
5278                                 free_extent_map(existing);
5279                                 if (err) {
5280                                         free_extent_map(em);
5281                                         em = NULL;
5282                                 }
5283                         } else {
5284                                 err = -EIO;
5285                                 free_extent_map(em);
5286                                 em = NULL;
5287                         }
5288                 } else {
5289                         free_extent_map(em);
5290                         em = existing;
5291                         err = 0;
5292                 }
5293         }
5294         write_unlock(&em_tree->lock);
5295 out:
5296
5297         trace_btrfs_get_extent(root, em);
5298
5299         if (path)
5300                 btrfs_free_path(path);
5301         if (trans) {
5302                 ret = btrfs_end_transaction(trans, root);
5303                 if (!err)
5304                         err = ret;
5305         }
5306         if (err) {
5307                 free_extent_map(em);
5308                 return ERR_PTR(err);
5309         }
5310         return em;
5311 }
5312
5313 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5314                                            size_t pg_offset, u64 start, u64 len,
5315                                            int create)
5316 {
5317         struct extent_map *em;
5318         struct extent_map *hole_em = NULL;
5319         u64 range_start = start;
5320         u64 end;
5321         u64 found;
5322         u64 found_end;
5323         int err = 0;
5324
5325         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5326         if (IS_ERR(em))
5327                 return em;
5328         if (em) {
5329                 /*
5330                  * if our em maps to a hole, there might
5331                  * actually be delalloc bytes behind it
5332                  */
5333                 if (em->block_start != EXTENT_MAP_HOLE)
5334                         return em;
5335                 else
5336                         hole_em = em;
5337         }
5338
5339         /* check to see if we've wrapped (len == -1 or similar) */
5340         end = start + len;
5341         if (end < start)
5342                 end = (u64)-1;
5343         else
5344                 end -= 1;
5345
5346         em = NULL;
5347
5348         /* ok, we didn't find anything, lets look for delalloc */
5349         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5350                                  end, len, EXTENT_DELALLOC, 1);
5351         found_end = range_start + found;
5352         if (found_end < range_start)
5353                 found_end = (u64)-1;
5354
5355         /*
5356          * we didn't find anything useful, return
5357          * the original results from get_extent()
5358          */
5359         if (range_start > end || found_end <= start) {
5360                 em = hole_em;
5361                 hole_em = NULL;
5362                 goto out;
5363         }
5364
5365         /* adjust the range_start to make sure it doesn't
5366          * go backwards from the start they passed in
5367          */
5368         range_start = max(start,range_start);
5369         found = found_end - range_start;
5370
5371         if (found > 0) {
5372                 u64 hole_start = start;
5373                 u64 hole_len = len;
5374
5375                 em = alloc_extent_map(GFP_NOFS);
5376                 if (!em) {
5377                         err = -ENOMEM;
5378                         goto out;
5379                 }
5380                 /*
5381                  * when btrfs_get_extent can't find anything it
5382                  * returns one huge hole
5383                  *
5384                  * make sure what it found really fits our range, and
5385                  * adjust to make sure it is based on the start from
5386                  * the caller
5387                  */
5388                 if (hole_em) {
5389                         u64 calc_end = extent_map_end(hole_em);
5390
5391                         if (calc_end <= start || (hole_em->start > end)) {
5392                                 free_extent_map(hole_em);
5393                                 hole_em = NULL;
5394                         } else {
5395                                 hole_start = max(hole_em->start, start);
5396                                 hole_len = calc_end - hole_start;
5397                         }
5398                 }
5399                 em->bdev = NULL;
5400                 if (hole_em && range_start > hole_start) {
5401                         /* our hole starts before our delalloc, so we
5402                          * have to return just the parts of the hole
5403                          * that go until  the delalloc starts
5404                          */
5405                         em->len = min(hole_len,
5406                                       range_start - hole_start);
5407                         em->start = hole_start;
5408                         em->orig_start = hole_start;
5409                         /*
5410                          * don't adjust block start at all,
5411                          * it is fixed at EXTENT_MAP_HOLE
5412                          */
5413                         em->block_start = hole_em->block_start;
5414                         em->block_len = hole_len;
5415                 } else {
5416                         em->start = range_start;
5417                         em->len = found;
5418                         em->orig_start = range_start;
5419                         em->block_start = EXTENT_MAP_DELALLOC;
5420                         em->block_len = found;
5421                 }
5422         } else if (hole_em) {
5423                 return hole_em;
5424         }
5425 out:
5426
5427         free_extent_map(hole_em);
5428         if (err) {
5429                 free_extent_map(em);
5430                 return ERR_PTR(err);
5431         }
5432         return em;
5433 }
5434
5435 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5436                                                   u64 start, u64 len)
5437 {
5438         struct btrfs_root *root = BTRFS_I(inode)->root;
5439         struct btrfs_trans_handle *trans;
5440         struct extent_map *em;
5441         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5442         struct btrfs_key ins;
5443         u64 alloc_hint;
5444         int ret;
5445
5446         btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5447
5448         trans = btrfs_join_transaction(root, 0);
5449         if (IS_ERR(trans))
5450                 return ERR_CAST(trans);
5451
5452         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5453
5454         alloc_hint = get_extent_allocation_hint(inode, start, len);
5455         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5456                                    alloc_hint, (u64)-1, &ins, 1);
5457         if (ret) {
5458                 em = ERR_PTR(ret);
5459                 goto out;
5460         }
5461
5462         em = alloc_extent_map(GFP_NOFS);
5463         if (!em) {
5464                 em = ERR_PTR(-ENOMEM);
5465                 goto out;
5466         }
5467
5468         em->start = start;
5469         em->orig_start = em->start;
5470         em->len = ins.offset;
5471
5472         em->block_start = ins.objectid;
5473         em->block_len = ins.offset;
5474         em->bdev = root->fs_info->fs_devices->latest_bdev;
5475         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5476
5477         while (1) {
5478                 write_lock(&em_tree->lock);
5479                 ret = add_extent_mapping(em_tree, em);
5480                 write_unlock(&em_tree->lock);
5481                 if (ret != -EEXIST)
5482                         break;
5483                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5484         }
5485
5486         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5487                                            ins.offset, ins.offset, 0);
5488         if (ret) {
5489                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5490                 em = ERR_PTR(ret);
5491         }
5492 out:
5493         btrfs_end_transaction(trans, root);
5494         return em;
5495 }
5496
5497 /*
5498  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5499  * block must be cow'd
5500  */
5501 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5502                                       struct inode *inode, u64 offset, u64 len)
5503 {
5504         struct btrfs_path *path;
5505         int ret;
5506         struct extent_buffer *leaf;
5507         struct btrfs_root *root = BTRFS_I(inode)->root;
5508         struct btrfs_file_extent_item *fi;
5509         struct btrfs_key key;
5510         u64 disk_bytenr;
5511         u64 backref_offset;
5512         u64 extent_end;
5513         u64 num_bytes;
5514         int slot;
5515         int found_type;
5516
5517         path = btrfs_alloc_path();
5518         if (!path)
5519                 return -ENOMEM;
5520
5521         ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
5522                                        offset, 0);
5523         if (ret < 0)
5524                 goto out;
5525
5526         slot = path->slots[0];
5527         if (ret == 1) {
5528                 if (slot == 0) {
5529                         /* can't find the item, must cow */
5530                         ret = 0;
5531                         goto out;
5532                 }
5533                 slot--;
5534         }
5535         ret = 0;
5536         leaf = path->nodes[0];
5537         btrfs_item_key_to_cpu(leaf, &key, slot);
5538         if (key.objectid != inode->i_ino ||
5539             key.type != BTRFS_EXTENT_DATA_KEY) {
5540                 /* not our file or wrong item type, must cow */
5541                 goto out;
5542         }
5543
5544         if (key.offset > offset) {
5545                 /* Wrong offset, must cow */
5546                 goto out;
5547         }
5548
5549         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5550         found_type = btrfs_file_extent_type(leaf, fi);
5551         if (found_type != BTRFS_FILE_EXTENT_REG &&
5552             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5553                 /* not a regular extent, must cow */
5554                 goto out;
5555         }
5556         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5557         backref_offset = btrfs_file_extent_offset(leaf, fi);
5558
5559         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5560         if (extent_end < offset + len) {
5561                 /* extent doesn't include our full range, must cow */
5562                 goto out;
5563         }
5564
5565         if (btrfs_extent_readonly(root, disk_bytenr))
5566                 goto out;
5567
5568         /*
5569          * look for other files referencing this extent, if we
5570          * find any we must cow
5571          */
5572         if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
5573                                   key.offset - backref_offset, disk_bytenr))
5574                 goto out;
5575
5576         /*
5577          * adjust disk_bytenr and num_bytes to cover just the bytes
5578          * in this extent we are about to write.  If there
5579          * are any csums in that range we have to cow in order
5580          * to keep the csums correct
5581          */
5582         disk_bytenr += backref_offset;
5583         disk_bytenr += offset - key.offset;
5584         num_bytes = min(offset + len, extent_end) - offset;
5585         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5586                                 goto out;
5587         /*
5588          * all of the above have passed, it is safe to overwrite this extent
5589          * without cow
5590          */
5591         ret = 1;
5592 out:
5593         btrfs_free_path(path);
5594         return ret;
5595 }
5596
5597 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5598                                    struct buffer_head *bh_result, int create)
5599 {
5600         struct extent_map *em;
5601         struct btrfs_root *root = BTRFS_I(inode)->root;
5602         u64 start = iblock << inode->i_blkbits;
5603         u64 len = bh_result->b_size;
5604         struct btrfs_trans_handle *trans;
5605
5606         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5607         if (IS_ERR(em))
5608                 return PTR_ERR(em);
5609
5610         /*
5611          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5612          * io.  INLINE is special, and we could probably kludge it in here, but
5613          * it's still buffered so for safety lets just fall back to the generic
5614          * buffered path.
5615          *
5616          * For COMPRESSED we _have_ to read the entire extent in so we can
5617          * decompress it, so there will be buffering required no matter what we
5618          * do, so go ahead and fallback to buffered.
5619          *
5620          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5621          * to buffered IO.  Don't blame me, this is the price we pay for using
5622          * the generic code.
5623          */
5624         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5625             em->block_start == EXTENT_MAP_INLINE) {
5626                 free_extent_map(em);
5627                 return -ENOTBLK;
5628         }
5629
5630         /* Just a good old fashioned hole, return */
5631         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5632                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5633                 free_extent_map(em);
5634                 /* DIO will do one hole at a time, so just unlock a sector */
5635                 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5636                               start + root->sectorsize - 1, GFP_NOFS);
5637                 return 0;
5638         }
5639
5640         /*
5641          * We don't allocate a new extent in the following cases
5642          *
5643          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
5644          * existing extent.
5645          * 2) The extent is marked as PREALLOC.  We're good to go here and can
5646          * just use the extent.
5647          *
5648          */
5649         if (!create) {
5650                 len = em->len - (start - em->start);
5651                 goto map;
5652         }
5653
5654         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5655             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5656              em->block_start != EXTENT_MAP_HOLE)) {
5657                 int type;
5658                 int ret;
5659                 u64 block_start;
5660
5661                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5662                         type = BTRFS_ORDERED_PREALLOC;
5663                 else
5664                         type = BTRFS_ORDERED_NOCOW;
5665                 len = min(len, em->len - (start - em->start));
5666                 block_start = em->block_start + (start - em->start);
5667
5668                 /*
5669                  * we're not going to log anything, but we do need
5670                  * to make sure the current transaction stays open
5671                  * while we look for nocow cross refs
5672                  */
5673                 trans = btrfs_join_transaction(root, 0);
5674                 if (IS_ERR(trans))
5675                         goto must_cow;
5676
5677                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5678                         ret = btrfs_add_ordered_extent_dio(inode, start,
5679                                            block_start, len, len, type);
5680                         btrfs_end_transaction(trans, root);
5681                         if (ret) {
5682                                 free_extent_map(em);
5683                                 return ret;
5684                         }
5685                         goto unlock;
5686                 }
5687                 btrfs_end_transaction(trans, root);
5688         }
5689 must_cow:
5690         /*
5691          * this will cow the extent, reset the len in case we changed
5692          * it above
5693          */
5694         len = bh_result->b_size;
5695         free_extent_map(em);
5696         em = btrfs_new_extent_direct(inode, start, len);
5697         if (IS_ERR(em))
5698                 return PTR_ERR(em);
5699         len = min(len, em->len - (start - em->start));
5700 unlock:
5701         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5702                           EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5703                           0, NULL, GFP_NOFS);
5704 map:
5705         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5706                 inode->i_blkbits;
5707         bh_result->b_size = len;
5708         bh_result->b_bdev = em->bdev;
5709         set_buffer_mapped(bh_result);
5710         if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5711                 set_buffer_new(bh_result);
5712
5713         free_extent_map(em);
5714
5715         return 0;
5716 }
5717
5718 struct btrfs_dio_private {
5719         struct inode *inode;
5720         u64 logical_offset;
5721         u64 disk_bytenr;
5722         u64 bytes;
5723         u32 *csums;
5724         void *private;
5725
5726         /* number of bios pending for this dio */
5727         atomic_t pending_bios;
5728
5729         /* IO errors */
5730         int errors;
5731
5732         struct bio *orig_bio;
5733 };
5734
5735 static void btrfs_endio_direct_read(struct bio *bio, int err)
5736 {
5737         struct btrfs_dio_private *dip = bio->bi_private;
5738         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5739         struct bio_vec *bvec = bio->bi_io_vec;
5740         struct inode *inode = dip->inode;
5741         struct btrfs_root *root = BTRFS_I(inode)->root;
5742         u64 start;
5743         u32 *private = dip->csums;
5744
5745         start = dip->logical_offset;
5746         do {
5747                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5748                         struct page *page = bvec->bv_page;
5749                         char *kaddr;
5750                         u32 csum = ~(u32)0;
5751                         unsigned long flags;
5752
5753                         local_irq_save(flags);
5754                         kaddr = kmap_atomic(page, KM_IRQ0);
5755                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5756                                                csum, bvec->bv_len);
5757                         btrfs_csum_final(csum, (char *)&csum);
5758                         kunmap_atomic(kaddr, KM_IRQ0);
5759                         local_irq_restore(flags);
5760
5761                         flush_dcache_page(bvec->bv_page);
5762                         if (csum != *private) {
5763                                 printk(KERN_ERR "btrfs csum failed ino %lu off"
5764                                       " %llu csum %u private %u\n",
5765                                       inode->i_ino, (unsigned long long)start,
5766                                       csum, *private);
5767                                 err = -EIO;
5768                         }
5769                 }
5770
5771                 start += bvec->bv_len;
5772                 private++;
5773                 bvec++;
5774         } while (bvec <= bvec_end);
5775
5776         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5777                       dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5778         bio->bi_private = dip->private;
5779
5780         kfree(dip->csums);
5781         kfree(dip);
5782
5783         /* If we had a csum failure make sure to clear the uptodate flag */
5784         if (err)
5785                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5786         dio_end_io(bio, err);
5787 }
5788
5789 static void btrfs_endio_direct_write(struct bio *bio, int err)
5790 {
5791         struct btrfs_dio_private *dip = bio->bi_private;
5792         struct inode *inode = dip->inode;
5793         struct btrfs_root *root = BTRFS_I(inode)->root;
5794         struct btrfs_trans_handle *trans;
5795         struct btrfs_ordered_extent *ordered = NULL;
5796         struct extent_state *cached_state = NULL;
5797         u64 ordered_offset = dip->logical_offset;
5798         u64 ordered_bytes = dip->bytes;
5799         int ret;
5800
5801         if (err)
5802                 goto out_done;
5803 again:
5804         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5805                                                    &ordered_offset,
5806                                                    ordered_bytes);
5807         if (!ret)
5808                 goto out_test;
5809
5810         BUG_ON(!ordered);
5811
5812         trans = btrfs_join_transaction(root, 1);
5813         if (IS_ERR(trans)) {
5814                 err = -ENOMEM;
5815                 goto out;
5816         }
5817         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5818
5819         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5820                 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5821                 if (!ret)
5822                         ret = btrfs_update_inode(trans, root, inode);
5823                 err = ret;
5824                 goto out;
5825         }
5826
5827         lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5828                          ordered->file_offset + ordered->len - 1, 0,
5829                          &cached_state, GFP_NOFS);
5830
5831         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5832                 ret = btrfs_mark_extent_written(trans, inode,
5833                                                 ordered->file_offset,
5834                                                 ordered->file_offset +
5835                                                 ordered->len);
5836                 if (ret) {
5837                         err = ret;
5838                         goto out_unlock;
5839                 }
5840         } else {
5841                 ret = insert_reserved_file_extent(trans, inode,
5842                                                   ordered->file_offset,
5843                                                   ordered->start,
5844                                                   ordered->disk_len,
5845                                                   ordered->len,
5846                                                   ordered->len,
5847                                                   0, 0, 0,
5848                                                   BTRFS_FILE_EXTENT_REG);
5849                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5850                                    ordered->file_offset, ordered->len);
5851                 if (ret) {
5852                         err = ret;
5853                         WARN_ON(1);
5854                         goto out_unlock;
5855                 }
5856         }
5857
5858         add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5859         btrfs_ordered_update_i_size(inode, 0, ordered);
5860         btrfs_update_inode(trans, root, inode);
5861 out_unlock:
5862         unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5863                              ordered->file_offset + ordered->len - 1,
5864                              &cached_state, GFP_NOFS);
5865 out:
5866         btrfs_delalloc_release_metadata(inode, ordered->len);
5867         btrfs_end_transaction(trans, root);
5868         ordered_offset = ordered->file_offset + ordered->len;
5869         btrfs_put_ordered_extent(ordered);
5870         btrfs_put_ordered_extent(ordered);
5871
5872 out_test:
5873         /*
5874          * our bio might span multiple ordered extents.  If we haven't
5875          * completed the accounting for the whole dio, go back and try again
5876          */
5877         if (ordered_offset < dip->logical_offset + dip->bytes) {
5878                 ordered_bytes = dip->logical_offset + dip->bytes -
5879                         ordered_offset;
5880                 goto again;
5881         }
5882 out_done:
5883         bio->bi_private = dip->private;
5884
5885         kfree(dip->csums);
5886         kfree(dip);
5887
5888         /* If we had an error make sure to clear the uptodate flag */
5889         if (err)
5890                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5891         dio_end_io(bio, err);
5892 }
5893
5894 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5895                                     struct bio *bio, int mirror_num,
5896                                     unsigned long bio_flags, u64 offset)
5897 {
5898         int ret;
5899         struct btrfs_root *root = BTRFS_I(inode)->root;
5900         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5901         BUG_ON(ret);
5902         return 0;
5903 }
5904
5905 static void btrfs_end_dio_bio(struct bio *bio, int err)
5906 {
5907         struct btrfs_dio_private *dip = bio->bi_private;
5908
5909         if (err) {
5910                 printk(KERN_ERR "btrfs direct IO failed ino %lu rw %lu "
5911                       "sector %#Lx len %u err no %d\n",
5912                       dip->inode->i_ino, bio->bi_rw,
5913                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
5914                 dip->errors = 1;
5915
5916                 /*
5917                  * before atomic variable goto zero, we must make sure
5918                  * dip->errors is perceived to be set.
5919                  */
5920                 smp_mb__before_atomic_dec();
5921         }
5922
5923         /* if there are more bios still pending for this dio, just exit */
5924         if (!atomic_dec_and_test(&dip->pending_bios))
5925                 goto out;
5926
5927         if (dip->errors)
5928                 bio_io_error(dip->orig_bio);
5929         else {
5930                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5931                 bio_endio(dip->orig_bio, 0);
5932         }
5933 out:
5934         bio_put(bio);
5935 }
5936
5937 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5938                                        u64 first_sector, gfp_t gfp_flags)
5939 {
5940         int nr_vecs = bio_get_nr_vecs(bdev);
5941         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5942 }
5943
5944 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5945                                          int rw, u64 file_offset, int skip_sum,
5946                                          u32 *csums)
5947 {
5948         int write = rw & REQ_WRITE;
5949         struct btrfs_root *root = BTRFS_I(inode)->root;
5950         int ret;
5951
5952         bio_get(bio);
5953         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5954         if (ret)
5955                 goto err;
5956
5957         if (write && !skip_sum) {
5958                 ret = btrfs_wq_submit_bio(root->fs_info,
5959                                    inode, rw, bio, 0, 0,
5960                                    file_offset,
5961                                    __btrfs_submit_bio_start_direct_io,
5962                                    __btrfs_submit_bio_done);
5963                 goto err;
5964         } else if (!skip_sum) {
5965                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
5966                                           file_offset, csums);
5967                 if (ret)
5968                         goto err;
5969         }
5970
5971         ret = btrfs_map_bio(root, rw, bio, 0, 1);
5972 err:
5973         bio_put(bio);
5974         return ret;
5975 }
5976
5977 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5978                                     int skip_sum)
5979 {
5980         struct inode *inode = dip->inode;
5981         struct btrfs_root *root = BTRFS_I(inode)->root;
5982         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5983         struct bio *bio;
5984         struct bio *orig_bio = dip->orig_bio;
5985         struct bio_vec *bvec = orig_bio->bi_io_vec;
5986         u64 start_sector = orig_bio->bi_sector;
5987         u64 file_offset = dip->logical_offset;
5988         u64 submit_len = 0;
5989         u64 map_length;
5990         int nr_pages = 0;
5991         u32 *csums = dip->csums;
5992         int ret = 0;
5993         int write = rw & REQ_WRITE;
5994
5995         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5996         if (!bio)
5997                 return -ENOMEM;
5998         bio->bi_private = dip;
5999         bio->bi_end_io = btrfs_end_dio_bio;
6000         atomic_inc(&dip->pending_bios);
6001
6002         map_length = orig_bio->bi_size;
6003         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6004                               &map_length, NULL, 0);
6005         if (ret) {
6006                 bio_put(bio);
6007                 return -EIO;
6008         }
6009
6010         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6011                 if (unlikely(map_length < submit_len + bvec->bv_len ||
6012                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6013                                  bvec->bv_offset) < bvec->bv_len)) {
6014                         /*
6015                          * inc the count before we submit the bio so
6016                          * we know the end IO handler won't happen before
6017                          * we inc the count. Otherwise, the dip might get freed
6018                          * before we're done setting it up
6019                          */
6020                         atomic_inc(&dip->pending_bios);
6021                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
6022                                                      file_offset, skip_sum,
6023                                                      csums);
6024                         if (ret) {
6025                                 bio_put(bio);
6026                                 atomic_dec(&dip->pending_bios);
6027                                 goto out_err;
6028                         }
6029
6030                         /* Write's use the ordered csums */
6031                         if (!write && !skip_sum)
6032                                 csums = csums + nr_pages;
6033                         start_sector += submit_len >> 9;
6034                         file_offset += submit_len;
6035
6036                         submit_len = 0;
6037                         nr_pages = 0;
6038
6039                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6040                                                   start_sector, GFP_NOFS);
6041                         if (!bio)
6042                                 goto out_err;
6043                         bio->bi_private = dip;
6044                         bio->bi_end_io = btrfs_end_dio_bio;
6045
6046                         map_length = orig_bio->bi_size;
6047                         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6048                                               &map_length, NULL, 0);
6049                         if (ret) {
6050                                 bio_put(bio);
6051                                 goto out_err;
6052                         }
6053                 } else {
6054                         submit_len += bvec->bv_len;
6055                         nr_pages ++;
6056                         bvec++;
6057                 }
6058         }
6059
6060         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6061                                      csums);
6062         if (!ret)
6063                 return 0;
6064
6065         bio_put(bio);
6066 out_err:
6067         dip->errors = 1;
6068         /*
6069          * before atomic variable goto zero, we must
6070          * make sure dip->errors is perceived to be set.
6071          */
6072         smp_mb__before_atomic_dec();
6073         if (atomic_dec_and_test(&dip->pending_bios))
6074                 bio_io_error(dip->orig_bio);
6075
6076         /* bio_end_io() will handle error, so we needn't return it */
6077         return 0;
6078 }
6079
6080 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6081                                 loff_t file_offset)
6082 {
6083         struct btrfs_root *root = BTRFS_I(inode)->root;
6084         struct btrfs_dio_private *dip;
6085         struct bio_vec *bvec = bio->bi_io_vec;
6086         int skip_sum;
6087         int write = rw & REQ_WRITE;
6088         int ret = 0;
6089
6090         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6091
6092         dip = kmalloc(sizeof(*dip), GFP_NOFS);
6093         if (!dip) {
6094                 ret = -ENOMEM;
6095                 goto free_ordered;
6096         }
6097         dip->csums = NULL;
6098
6099         /* Write's use the ordered csum stuff, so we don't need dip->csums */
6100         if (!write && !skip_sum) {
6101                 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6102                 if (!dip->csums) {
6103                         kfree(dip);
6104                         ret = -ENOMEM;
6105                         goto free_ordered;
6106                 }
6107         }
6108
6109         dip->private = bio->bi_private;
6110         dip->inode = inode;
6111         dip->logical_offset = file_offset;
6112
6113         dip->bytes = 0;
6114         do {
6115                 dip->bytes += bvec->bv_len;
6116                 bvec++;
6117         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6118
6119         dip->disk_bytenr = (u64)bio->bi_sector << 9;
6120         bio->bi_private = dip;
6121         dip->errors = 0;
6122         dip->orig_bio = bio;
6123         atomic_set(&dip->pending_bios, 0);
6124
6125         if (write)
6126                 bio->bi_end_io = btrfs_endio_direct_write;
6127         else
6128                 bio->bi_end_io = btrfs_endio_direct_read;
6129
6130         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6131         if (!ret)
6132                 return;
6133 free_ordered:
6134         /*
6135          * If this is a write, we need to clean up the reserved space and kill
6136          * the ordered extent.
6137          */
6138         if (write) {
6139                 struct btrfs_ordered_extent *ordered;
6140                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6141                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6142                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6143                         btrfs_free_reserved_extent(root, ordered->start,
6144                                                    ordered->disk_len);
6145                 btrfs_put_ordered_extent(ordered);
6146                 btrfs_put_ordered_extent(ordered);
6147         }
6148         bio_endio(bio, ret);
6149 }
6150
6151 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6152                         const struct iovec *iov, loff_t offset,
6153                         unsigned long nr_segs)
6154 {
6155         int seg;
6156         size_t size;
6157         unsigned long addr;
6158         unsigned blocksize_mask = root->sectorsize - 1;
6159         ssize_t retval = -EINVAL;
6160         loff_t end = offset;
6161
6162         if (offset & blocksize_mask)
6163                 goto out;
6164
6165         /* Check the memory alignment.  Blocks cannot straddle pages */
6166         for (seg = 0; seg < nr_segs; seg++) {
6167                 addr = (unsigned long)iov[seg].iov_base;
6168                 size = iov[seg].iov_len;
6169                 end += size;
6170                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6171                         goto out;
6172         }
6173         retval = 0;
6174 out:
6175         return retval;
6176 }
6177 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6178                         const struct iovec *iov, loff_t offset,
6179                         unsigned long nr_segs)
6180 {
6181         struct file *file = iocb->ki_filp;
6182         struct inode *inode = file->f_mapping->host;
6183         struct btrfs_ordered_extent *ordered;
6184         struct extent_state *cached_state = NULL;
6185         u64 lockstart, lockend;
6186         ssize_t ret;
6187         int writing = rw & WRITE;
6188         int write_bits = 0;
6189         size_t count = iov_length(iov, nr_segs);
6190
6191         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6192                             offset, nr_segs)) {
6193                 return 0;
6194         }
6195
6196         lockstart = offset;
6197         lockend = offset + count - 1;
6198
6199         if (writing) {
6200                 ret = btrfs_delalloc_reserve_space(inode, count);
6201                 if (ret)
6202                         goto out;
6203         }
6204
6205         while (1) {
6206                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6207                                  0, &cached_state, GFP_NOFS);
6208                 /*
6209                  * We're concerned with the entire range that we're going to be
6210                  * doing DIO to, so we need to make sure theres no ordered
6211                  * extents in this range.
6212                  */
6213                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6214                                                      lockend - lockstart + 1);
6215                 if (!ordered)
6216                         break;
6217                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6218                                      &cached_state, GFP_NOFS);
6219                 btrfs_start_ordered_extent(inode, ordered, 1);
6220                 btrfs_put_ordered_extent(ordered);
6221                 cond_resched();
6222         }
6223
6224         /*
6225          * we don't use btrfs_set_extent_delalloc because we don't want
6226          * the dirty or uptodate bits
6227          */
6228         if (writing) {
6229                 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6230                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6231                                      EXTENT_DELALLOC, 0, NULL, &cached_state,
6232                                      GFP_NOFS);
6233                 if (ret) {
6234                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6235                                          lockend, EXTENT_LOCKED | write_bits,
6236                                          1, 0, &cached_state, GFP_NOFS);
6237                         goto out;
6238                 }
6239         }
6240
6241         free_extent_state(cached_state);
6242         cached_state = NULL;
6243
6244         ret = __blockdev_direct_IO(rw, iocb, inode,
6245                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6246                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6247                    btrfs_submit_direct, 0);
6248
6249         if (ret < 0 && ret != -EIOCBQUEUED) {
6250                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6251                               offset + iov_length(iov, nr_segs) - 1,
6252                               EXTENT_LOCKED | write_bits, 1, 0,
6253                               &cached_state, GFP_NOFS);
6254         } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6255                 /*
6256                  * We're falling back to buffered, unlock the section we didn't
6257                  * do IO on.
6258                  */
6259                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6260                               offset + iov_length(iov, nr_segs) - 1,
6261                               EXTENT_LOCKED | write_bits, 1, 0,
6262                               &cached_state, GFP_NOFS);
6263         }
6264 out:
6265         free_extent_state(cached_state);
6266         return ret;
6267 }
6268
6269 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6270                 __u64 start, __u64 len)
6271 {
6272         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6273 }
6274
6275 int btrfs_readpage(struct file *file, struct page *page)
6276 {
6277         struct extent_io_tree *tree;
6278         tree = &BTRFS_I(page->mapping->host)->io_tree;
6279         return extent_read_full_page(tree, page, btrfs_get_extent);
6280 }
6281
6282 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6283 {
6284         struct extent_io_tree *tree;
6285
6286
6287         if (current->flags & PF_MEMALLOC) {
6288                 redirty_page_for_writepage(wbc, page);
6289                 unlock_page(page);
6290                 return 0;
6291         }
6292         tree = &BTRFS_I(page->mapping->host)->io_tree;
6293         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6294 }
6295
6296 int btrfs_writepages(struct address_space *mapping,
6297                      struct writeback_control *wbc)
6298 {
6299         struct extent_io_tree *tree;
6300
6301         tree = &BTRFS_I(mapping->host)->io_tree;
6302         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6303 }
6304
6305 static int
6306 btrfs_readpages(struct file *file, struct address_space *mapping,
6307                 struct list_head *pages, unsigned nr_pages)
6308 {
6309         struct extent_io_tree *tree;
6310         tree = &BTRFS_I(mapping->host)->io_tree;
6311         return extent_readpages(tree, mapping, pages, nr_pages,
6312                                 btrfs_get_extent);
6313 }
6314 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6315 {
6316         struct extent_io_tree *tree;
6317         struct extent_map_tree *map;
6318         int ret;
6319
6320         tree = &BTRFS_I(page->mapping->host)->io_tree;
6321         map = &BTRFS_I(page->mapping->host)->extent_tree;
6322         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6323         if (ret == 1) {
6324                 ClearPagePrivate(page);
6325                 set_page_private(page, 0);
6326                 page_cache_release(page);
6327         }
6328         return ret;
6329 }
6330
6331 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6332 {
6333         if (PageWriteback(page) || PageDirty(page))
6334                 return 0;
6335         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6336 }
6337
6338 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6339 {
6340         struct extent_io_tree *tree;
6341         struct btrfs_ordered_extent *ordered;
6342         struct extent_state *cached_state = NULL;
6343         u64 page_start = page_offset(page);
6344         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6345
6346
6347         /*
6348          * we have the page locked, so new writeback can't start,
6349          * and the dirty bit won't be cleared while we are here.
6350          *
6351          * Wait for IO on this page so that we can safely clear
6352          * the PagePrivate2 bit and do ordered accounting
6353          */
6354         wait_on_page_writeback(page);
6355
6356         tree = &BTRFS_I(page->mapping->host)->io_tree;
6357         if (offset) {
6358                 btrfs_releasepage(page, GFP_NOFS);
6359                 return;
6360         }
6361         lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6362                          GFP_NOFS);
6363         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6364                                            page_offset(page));
6365         if (ordered) {
6366                 /*
6367                  * IO on this page will never be started, so we need
6368                  * to account for any ordered extents now
6369                  */
6370                 clear_extent_bit(tree, page_start, page_end,
6371                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6372                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6373                                  &cached_state, GFP_NOFS);
6374                 /*
6375                  * whoever cleared the private bit is responsible
6376                  * for the finish_ordered_io
6377                  */
6378                 if (TestClearPagePrivate2(page)) {
6379                         btrfs_finish_ordered_io(page->mapping->host,
6380                                                 page_start, page_end);
6381                 }
6382                 btrfs_put_ordered_extent(ordered);
6383                 cached_state = NULL;
6384                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6385                                  GFP_NOFS);
6386         }
6387         clear_extent_bit(tree, page_start, page_end,
6388                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6389                  EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6390         __btrfs_releasepage(page, GFP_NOFS);
6391
6392         ClearPageChecked(page);
6393         if (PagePrivate(page)) {
6394                 ClearPagePrivate(page);
6395                 set_page_private(page, 0);
6396                 page_cache_release(page);
6397         }
6398 }
6399
6400 /*
6401  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6402  * called from a page fault handler when a page is first dirtied. Hence we must
6403  * be careful to check for EOF conditions here. We set the page up correctly
6404  * for a written page which means we get ENOSPC checking when writing into
6405  * holes and correct delalloc and unwritten extent mapping on filesystems that
6406  * support these features.
6407  *
6408  * We are not allowed to take the i_mutex here so we have to play games to
6409  * protect against truncate races as the page could now be beyond EOF.  Because
6410  * vmtruncate() writes the inode size before removing pages, once we have the
6411  * page lock we can determine safely if the page is beyond EOF. If it is not
6412  * beyond EOF, then the page is guaranteed safe against truncation until we
6413  * unlock the page.
6414  */
6415 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6416 {
6417         struct page *page = vmf->page;
6418         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6419         struct btrfs_root *root = BTRFS_I(inode)->root;
6420         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6421         struct btrfs_ordered_extent *ordered;
6422         struct extent_state *cached_state = NULL;
6423         char *kaddr;
6424         unsigned long zero_start;
6425         loff_t size;
6426         int ret;
6427         u64 page_start;
6428         u64 page_end;
6429
6430         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6431         if (ret) {
6432                 if (ret == -ENOMEM)
6433                         ret = VM_FAULT_OOM;
6434                 else /* -ENOSPC, -EIO, etc */
6435                         ret = VM_FAULT_SIGBUS;
6436                 goto out;
6437         }
6438
6439         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6440 again:
6441         lock_page(page);
6442         size = i_size_read(inode);
6443         page_start = page_offset(page);
6444         page_end = page_start + PAGE_CACHE_SIZE - 1;
6445
6446         if ((page->mapping != inode->i_mapping) ||
6447             (page_start >= size)) {
6448                 /* page got truncated out from underneath us */
6449                 goto out_unlock;
6450         }
6451         wait_on_page_writeback(page);
6452
6453         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6454                          GFP_NOFS);
6455         set_page_extent_mapped(page);
6456
6457         /*
6458          * we can't set the delalloc bits if there are pending ordered
6459          * extents.  Drop our locks and wait for them to finish
6460          */
6461         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6462         if (ordered) {
6463                 unlock_extent_cached(io_tree, page_start, page_end,
6464                                      &cached_state, GFP_NOFS);
6465                 unlock_page(page);
6466                 btrfs_start_ordered_extent(inode, ordered, 1);
6467                 btrfs_put_ordered_extent(ordered);
6468                 goto again;
6469         }
6470
6471         /*
6472          * XXX - page_mkwrite gets called every time the page is dirtied, even
6473          * if it was already dirty, so for space accounting reasons we need to
6474          * clear any delalloc bits for the range we are fixing to save.  There
6475          * is probably a better way to do this, but for now keep consistent with
6476          * prepare_pages in the normal write path.
6477          */
6478         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6479                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6480                           0, 0, &cached_state, GFP_NOFS);
6481
6482         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6483                                         &cached_state);
6484         if (ret) {
6485                 unlock_extent_cached(io_tree, page_start, page_end,
6486                                      &cached_state, GFP_NOFS);
6487                 ret = VM_FAULT_SIGBUS;
6488                 goto out_unlock;
6489         }
6490         ret = 0;
6491
6492         /* page is wholly or partially inside EOF */
6493         if (page_start + PAGE_CACHE_SIZE > size)
6494                 zero_start = size & ~PAGE_CACHE_MASK;
6495         else
6496                 zero_start = PAGE_CACHE_SIZE;
6497
6498         if (zero_start != PAGE_CACHE_SIZE) {
6499                 kaddr = kmap(page);
6500                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6501                 flush_dcache_page(page);
6502                 kunmap(page);
6503         }
6504         ClearPageChecked(page);
6505         set_page_dirty(page);
6506         SetPageUptodate(page);
6507
6508         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6509         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6510
6511         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6512
6513 out_unlock:
6514         if (!ret)
6515                 return VM_FAULT_LOCKED;
6516         unlock_page(page);
6517         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6518 out:
6519         return ret;
6520 }
6521
6522 static int btrfs_truncate(struct inode *inode)
6523 {
6524         struct btrfs_root *root = BTRFS_I(inode)->root;
6525         int ret;
6526         int err = 0;
6527         struct btrfs_trans_handle *trans;
6528         unsigned long nr;
6529         u64 mask = root->sectorsize - 1;
6530
6531         ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6532         if (ret)
6533                 return ret;
6534
6535         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6536         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6537
6538         trans = btrfs_start_transaction(root, 5);
6539         if (IS_ERR(trans))
6540                 return PTR_ERR(trans);
6541
6542         btrfs_set_trans_block_group(trans, inode);
6543
6544         ret = btrfs_orphan_add(trans, inode);
6545         if (ret) {
6546                 btrfs_end_transaction(trans, root);
6547                 return ret;
6548         }
6549
6550         nr = trans->blocks_used;
6551         btrfs_end_transaction(trans, root);
6552         btrfs_btree_balance_dirty(root, nr);
6553
6554         /* Now start a transaction for the truncate */
6555         trans = btrfs_start_transaction(root, 0);
6556         if (IS_ERR(trans))
6557                 return PTR_ERR(trans);
6558         btrfs_set_trans_block_group(trans, inode);
6559         trans->block_rsv = root->orphan_block_rsv;
6560
6561         /*
6562          * setattr is responsible for setting the ordered_data_close flag,
6563          * but that is only tested during the last file release.  That
6564          * could happen well after the next commit, leaving a great big
6565          * window where new writes may get lost if someone chooses to write
6566          * to this file after truncating to zero
6567          *
6568          * The inode doesn't have any dirty data here, and so if we commit
6569          * this is a noop.  If someone immediately starts writing to the inode
6570          * it is very likely we'll catch some of their writes in this
6571          * transaction, and the commit will find this file on the ordered
6572          * data list with good things to send down.
6573          *
6574          * This is a best effort solution, there is still a window where
6575          * using truncate to replace the contents of the file will
6576          * end up with a zero length file after a crash.
6577          */
6578         if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6579                 btrfs_add_ordered_operation(trans, root, inode);
6580
6581         while (1) {
6582                 if (!trans) {
6583                         trans = btrfs_start_transaction(root, 0);
6584                         if (IS_ERR(trans))
6585                                 return PTR_ERR(trans);
6586                         btrfs_set_trans_block_group(trans, inode);
6587                         trans->block_rsv = root->orphan_block_rsv;
6588                 }
6589
6590                 ret = btrfs_block_rsv_check(trans, root,
6591                                             root->orphan_block_rsv, 0, 5);
6592                 if (ret == -EAGAIN) {
6593                         ret = btrfs_commit_transaction(trans, root);
6594                         if (ret)
6595                                 return ret;
6596                         trans = NULL;
6597                         continue;
6598                 } else if (ret) {
6599                         err = ret;
6600                         break;
6601                 }
6602
6603                 ret = btrfs_truncate_inode_items(trans, root, inode,
6604                                                  inode->i_size,
6605                                                  BTRFS_EXTENT_DATA_KEY);
6606                 if (ret != -EAGAIN) {
6607                         err = ret;
6608                         break;
6609                 }
6610
6611                 ret = btrfs_update_inode(trans, root, inode);
6612                 if (ret) {
6613                         err = ret;
6614                         break;
6615                 }
6616
6617                 nr = trans->blocks_used;
6618                 btrfs_end_transaction(trans, root);
6619                 trans = NULL;
6620                 btrfs_btree_balance_dirty(root, nr);
6621         }
6622
6623         if (ret == 0 && inode->i_nlink > 0) {
6624                 ret = btrfs_orphan_del(trans, inode);
6625                 if (ret)
6626                         err = ret;
6627         } else if (ret && inode->i_nlink > 0) {
6628                 /*
6629                  * Failed to do the truncate, remove us from the in memory
6630                  * orphan list.
6631                  */
6632                 ret = btrfs_orphan_del(NULL, inode);
6633         }
6634
6635         ret = btrfs_update_inode(trans, root, inode);
6636         if (ret && !err)
6637                 err = ret;
6638
6639         nr = trans->blocks_used;
6640         ret = btrfs_end_transaction_throttle(trans, root);
6641         if (ret && !err)
6642                 err = ret;
6643         btrfs_btree_balance_dirty(root, nr);
6644
6645         return err;
6646 }
6647
6648 /*
6649  * create a new subvolume directory/inode (helper for the ioctl).
6650  */
6651 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6652                              struct btrfs_root *new_root,
6653                              u64 new_dirid, u64 alloc_hint)
6654 {
6655         struct inode *inode;
6656         int err;
6657         u64 index = 0;
6658
6659         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6660                                 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
6661         if (IS_ERR(inode))
6662                 return PTR_ERR(inode);
6663         inode->i_op = &btrfs_dir_inode_operations;
6664         inode->i_fop = &btrfs_dir_file_operations;
6665
6666         inode->i_nlink = 1;
6667         btrfs_i_size_write(inode, 0);
6668
6669         err = btrfs_update_inode(trans, new_root, inode);
6670         BUG_ON(err);
6671
6672         iput(inode);
6673         return 0;
6674 }
6675
6676 /* helper function for file defrag and space balancing.  This
6677  * forces readahead on a given range of bytes in an inode
6678  */
6679 unsigned long btrfs_force_ra(struct address_space *mapping,
6680                               struct file_ra_state *ra, struct file *file,
6681                               pgoff_t offset, pgoff_t last_index)
6682 {
6683         pgoff_t req_size = last_index - offset + 1;
6684
6685         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
6686         return offset + req_size;
6687 }
6688
6689 struct inode *btrfs_alloc_inode(struct super_block *sb)
6690 {
6691         struct btrfs_inode *ei;
6692         struct inode *inode;
6693
6694         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6695         if (!ei)
6696                 return NULL;
6697
6698         ei->root = NULL;
6699         ei->space_info = NULL;
6700         ei->generation = 0;
6701         ei->sequence = 0;
6702         ei->last_trans = 0;
6703         ei->last_sub_trans = 0;
6704         ei->logged_trans = 0;
6705         ei->delalloc_bytes = 0;
6706         ei->reserved_bytes = 0;
6707         ei->disk_i_size = 0;
6708         ei->flags = 0;
6709         ei->index_cnt = (u64)-1;
6710         ei->last_unlink_trans = 0;
6711
6712         atomic_set(&ei->outstanding_extents, 0);
6713         atomic_set(&ei->reserved_extents, 0);
6714
6715         ei->ordered_data_close = 0;
6716         ei->orphan_meta_reserved = 0;
6717         ei->dummy_inode = 0;
6718         ei->force_compress = BTRFS_COMPRESS_NONE;
6719
6720         inode = &ei->vfs_inode;
6721         extent_map_tree_init(&ei->extent_tree, GFP_NOFS);
6722         extent_io_tree_init(&ei->io_tree, &inode->i_data, GFP_NOFS);
6723         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data, GFP_NOFS);
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(atomic_read(&BTRFS_I(inode)->outstanding_extents));
6749         WARN_ON(atomic_read(&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         if (root == root->fs_info->tree_root) {
6771                 struct btrfs_block_group_cache *block_group;
6772
6773                 block_group = btrfs_lookup_block_group(root->fs_info,
6774                                                 BTRFS_I(inode)->block_group);
6775                 if (block_group && block_group->inode == inode) {
6776                         spin_lock(&block_group->lock);
6777                         block_group->inode = NULL;
6778                         spin_unlock(&block_group->lock);
6779                         btrfs_put_block_group(block_group);
6780                 } else if (block_group) {
6781                         btrfs_put_block_group(block_group);
6782                 }
6783         }
6784
6785         spin_lock(&root->orphan_lock);
6786         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6787                 printk(KERN_INFO "BTRFS: inode %lu still on the orphan list\n",
6788                        inode->i_ino);
6789                 list_del_init(&BTRFS_I(inode)->i_orphan);
6790         }
6791         spin_unlock(&root->orphan_lock);
6792
6793         while (1) {
6794                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6795                 if (!ordered)
6796                         break;
6797                 else {
6798                         printk(KERN_ERR "btrfs found ordered "
6799                                "extent %llu %llu on inode cleanup\n",
6800                                (unsigned long long)ordered->file_offset,
6801                                (unsigned long long)ordered->len);
6802                         btrfs_remove_ordered_extent(inode, ordered);
6803                         btrfs_put_ordered_extent(ordered);
6804                         btrfs_put_ordered_extent(ordered);
6805                 }
6806         }
6807         inode_tree_del(inode);
6808         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6809 free:
6810         call_rcu(&inode->i_rcu, btrfs_i_callback);
6811 }
6812
6813 int btrfs_drop_inode(struct inode *inode)
6814 {
6815         struct btrfs_root *root = BTRFS_I(inode)->root;
6816
6817         if (btrfs_root_refs(&root->root_item) == 0 &&
6818             root != root->fs_info->tree_root)
6819                 return 1;
6820         else
6821                 return generic_drop_inode(inode);
6822 }
6823
6824 static void init_once(void *foo)
6825 {
6826         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6827
6828         inode_init_once(&ei->vfs_inode);
6829 }
6830
6831 void btrfs_destroy_cachep(void)
6832 {
6833         if (btrfs_inode_cachep)
6834                 kmem_cache_destroy(btrfs_inode_cachep);
6835         if (btrfs_trans_handle_cachep)
6836                 kmem_cache_destroy(btrfs_trans_handle_cachep);
6837         if (btrfs_transaction_cachep)
6838                 kmem_cache_destroy(btrfs_transaction_cachep);
6839         if (btrfs_path_cachep)
6840                 kmem_cache_destroy(btrfs_path_cachep);
6841         if (btrfs_free_space_cachep)
6842                 kmem_cache_destroy(btrfs_free_space_cachep);
6843 }
6844
6845 int btrfs_init_cachep(void)
6846 {
6847         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6848                         sizeof(struct btrfs_inode), 0,
6849                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6850         if (!btrfs_inode_cachep)
6851                 goto fail;
6852
6853         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6854                         sizeof(struct btrfs_trans_handle), 0,
6855                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6856         if (!btrfs_trans_handle_cachep)
6857                 goto fail;
6858
6859         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6860                         sizeof(struct btrfs_transaction), 0,
6861                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6862         if (!btrfs_transaction_cachep)
6863                 goto fail;
6864
6865         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6866                         sizeof(struct btrfs_path), 0,
6867                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6868         if (!btrfs_path_cachep)
6869                 goto fail;
6870
6871         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6872                         sizeof(struct btrfs_free_space), 0,
6873                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6874         if (!btrfs_free_space_cachep)
6875                 goto fail;
6876
6877         return 0;
6878 fail:
6879         btrfs_destroy_cachep();
6880         return -ENOMEM;
6881 }
6882
6883 static int btrfs_getattr(struct vfsmount *mnt,
6884                          struct dentry *dentry, struct kstat *stat)
6885 {
6886         struct inode *inode = dentry->d_inode;
6887         generic_fillattr(inode, stat);
6888         stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
6889         stat->blksize = PAGE_CACHE_SIZE;
6890         stat->blocks = (inode_get_bytes(inode) +
6891                         BTRFS_I(inode)->delalloc_bytes) >> 9;
6892         return 0;
6893 }
6894
6895 /*
6896  * If a file is moved, it will inherit the cow and compression flags of the new
6897  * directory.
6898  */
6899 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6900 {
6901         struct btrfs_inode *b_dir = BTRFS_I(dir);
6902         struct btrfs_inode *b_inode = BTRFS_I(inode);
6903
6904         if (b_dir->flags & BTRFS_INODE_NODATACOW)
6905                 b_inode->flags |= BTRFS_INODE_NODATACOW;
6906         else
6907                 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6908
6909         if (b_dir->flags & BTRFS_INODE_COMPRESS)
6910                 b_inode->flags |= BTRFS_INODE_COMPRESS;
6911         else
6912                 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6913 }
6914
6915 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6916                            struct inode *new_dir, struct dentry *new_dentry)
6917 {
6918         struct btrfs_trans_handle *trans;
6919         struct btrfs_root *root = BTRFS_I(old_dir)->root;
6920         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6921         struct inode *new_inode = new_dentry->d_inode;
6922         struct inode *old_inode = old_dentry->d_inode;
6923         struct timespec ctime = CURRENT_TIME;
6924         u64 index = 0;
6925         u64 root_objectid;
6926         int ret;
6927
6928         if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6929                 return -EPERM;
6930
6931         /* we only allow rename subvolume link between subvolumes */
6932         if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6933                 return -EXDEV;
6934
6935         if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6936             (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID))
6937                 return -ENOTEMPTY;
6938
6939         if (S_ISDIR(old_inode->i_mode) && new_inode &&
6940             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6941                 return -ENOTEMPTY;
6942         /*
6943          * we're using rename to replace one file with another.
6944          * and the replacement file is large.  Start IO on it now so
6945          * we don't add too much work to the end of the transaction
6946          */
6947         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
6948             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6949                 filemap_flush(old_inode->i_mapping);
6950
6951         /* close the racy window with snapshot create/destroy ioctl */
6952         if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
6953                 down_read(&root->fs_info->subvol_sem);
6954         /*
6955          * We want to reserve the absolute worst case amount of items.  So if
6956          * both inodes are subvols and we need to unlink them then that would
6957          * require 4 item modifications, but if they are both normal inodes it
6958          * would require 5 item modifications, so we'll assume their normal
6959          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6960          * should cover the worst case number of items we'll modify.
6961          */
6962         trans = btrfs_start_transaction(root, 20);
6963         if (IS_ERR(trans))
6964                 return PTR_ERR(trans);
6965
6966         btrfs_set_trans_block_group(trans, new_dir);
6967
6968         if (dest != root)
6969                 btrfs_record_root_in_trans(trans, dest);
6970
6971         ret = btrfs_set_inode_index(new_dir, &index);
6972         if (ret)
6973                 goto out_fail;
6974
6975         if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6976                 /* force full log commit if subvolume involved. */
6977                 root->fs_info->last_trans_log_full_commit = trans->transid;
6978         } else {
6979                 ret = btrfs_insert_inode_ref(trans, dest,
6980                                              new_dentry->d_name.name,
6981                                              new_dentry->d_name.len,
6982                                              old_inode->i_ino,
6983                                              new_dir->i_ino, index);
6984                 if (ret)
6985                         goto out_fail;
6986                 /*
6987                  * this is an ugly little race, but the rename is required
6988                  * to make sure that if we crash, the inode is either at the
6989                  * old name or the new one.  pinning the log transaction lets
6990                  * us make sure we don't allow a log commit to come in after
6991                  * we unlink the name but before we add the new name back in.
6992                  */
6993                 btrfs_pin_log_trans(root);
6994         }
6995         /*
6996          * make sure the inode gets flushed if it is replacing
6997          * something.
6998          */
6999         if (new_inode && new_inode->i_size &&
7000             old_inode && S_ISREG(old_inode->i_mode)) {
7001                 btrfs_add_ordered_operation(trans, root, old_inode);
7002         }
7003
7004         old_dir->i_ctime = old_dir->i_mtime = ctime;
7005         new_dir->i_ctime = new_dir->i_mtime = ctime;
7006         old_inode->i_ctime = ctime;
7007
7008         if (old_dentry->d_parent != new_dentry->d_parent)
7009                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7010
7011         if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7012                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7013                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7014                                         old_dentry->d_name.name,
7015                                         old_dentry->d_name.len);
7016         } else {
7017                 ret = __btrfs_unlink_inode(trans, root, old_dir,
7018                                         old_dentry->d_inode,
7019                                         old_dentry->d_name.name,
7020                                         old_dentry->d_name.len);
7021                 if (!ret)
7022                         ret = btrfs_update_inode(trans, root, old_inode);
7023         }
7024         BUG_ON(ret);
7025
7026         if (new_inode) {
7027                 new_inode->i_ctime = CURRENT_TIME;
7028                 if (unlikely(new_inode->i_ino ==
7029                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7030                         root_objectid = BTRFS_I(new_inode)->location.objectid;
7031                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
7032                                                 root_objectid,
7033                                                 new_dentry->d_name.name,
7034                                                 new_dentry->d_name.len);
7035                         BUG_ON(new_inode->i_nlink == 0);
7036                 } else {
7037                         ret = btrfs_unlink_inode(trans, dest, new_dir,
7038                                                  new_dentry->d_inode,
7039                                                  new_dentry->d_name.name,
7040                                                  new_dentry->d_name.len);
7041                 }
7042                 BUG_ON(ret);
7043                 if (new_inode->i_nlink == 0) {
7044                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7045                         BUG_ON(ret);
7046                 }
7047         }
7048
7049         fixup_inode_flags(new_dir, old_inode);
7050
7051         ret = btrfs_add_link(trans, new_dir, old_inode,
7052                              new_dentry->d_name.name,
7053                              new_dentry->d_name.len, 0, index);
7054         BUG_ON(ret);
7055
7056         if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
7057                 struct dentry *parent = dget_parent(new_dentry);
7058                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7059                 dput(parent);
7060                 btrfs_end_log_trans(root);
7061         }
7062 out_fail:
7063         btrfs_end_transaction_throttle(trans, root);
7064
7065         if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
7066                 up_read(&root->fs_info->subvol_sem);
7067
7068         return ret;
7069 }
7070
7071 /*
7072  * some fairly slow code that needs optimization. This walks the list
7073  * of all the inodes with pending delalloc and forces them to disk.
7074  */
7075 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7076 {
7077         struct list_head *head = &root->fs_info->delalloc_inodes;
7078         struct btrfs_inode *binode;
7079         struct inode *inode;
7080
7081         if (root->fs_info->sb->s_flags & MS_RDONLY)
7082                 return -EROFS;
7083
7084         spin_lock(&root->fs_info->delalloc_lock);
7085         while (!list_empty(head)) {
7086                 binode = list_entry(head->next, struct btrfs_inode,
7087                                     delalloc_inodes);
7088                 inode = igrab(&binode->vfs_inode);
7089                 if (!inode)
7090                         list_del_init(&binode->delalloc_inodes);
7091                 spin_unlock(&root->fs_info->delalloc_lock);
7092                 if (inode) {
7093                         filemap_flush(inode->i_mapping);
7094                         if (delay_iput)
7095                                 btrfs_add_delayed_iput(inode);
7096                         else
7097                                 iput(inode);
7098                 }
7099                 cond_resched();
7100                 spin_lock(&root->fs_info->delalloc_lock);
7101         }
7102         spin_unlock(&root->fs_info->delalloc_lock);
7103
7104         /* the filemap_flush will queue IO into the worker threads, but
7105          * we have to make sure the IO is actually started and that
7106          * ordered extents get created before we return
7107          */
7108         atomic_inc(&root->fs_info->async_submit_draining);
7109         while (atomic_read(&root->fs_info->nr_async_submits) ||
7110               atomic_read(&root->fs_info->async_delalloc_pages)) {
7111                 wait_event(root->fs_info->async_submit_wait,
7112                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7113                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7114         }
7115         atomic_dec(&root->fs_info->async_submit_draining);
7116         return 0;
7117 }
7118
7119 int btrfs_start_one_delalloc_inode(struct btrfs_root *root, int delay_iput,
7120                                    int sync)
7121 {
7122         struct btrfs_inode *binode;
7123         struct inode *inode = NULL;
7124
7125         spin_lock(&root->fs_info->delalloc_lock);
7126         while (!list_empty(&root->fs_info->delalloc_inodes)) {
7127                 binode = list_entry(root->fs_info->delalloc_inodes.next,
7128                                     struct btrfs_inode, delalloc_inodes);
7129                 inode = igrab(&binode->vfs_inode);
7130                 if (inode) {
7131                         list_move_tail(&binode->delalloc_inodes,
7132                                        &root->fs_info->delalloc_inodes);
7133                         break;
7134                 }
7135
7136                 list_del_init(&binode->delalloc_inodes);
7137                 cond_resched_lock(&root->fs_info->delalloc_lock);
7138         }
7139         spin_unlock(&root->fs_info->delalloc_lock);
7140
7141         if (inode) {
7142                 if (sync) {
7143                         filemap_write_and_wait(inode->i_mapping);
7144                         /*
7145                          * We have to do this because compression doesn't
7146                          * actually set PG_writeback until it submits the pages
7147                          * for IO, which happens in an async thread, so we could
7148                          * race and not actually wait for any writeback pages
7149                          * because they've not been submitted yet.  Technically
7150                          * this could still be the case for the ordered stuff
7151                          * since the async thread may not have started to do its
7152                          * work yet.  If this becomes the case then we need to
7153                          * figure out a way to make sure that in writepage we
7154                          * wait for any async pages to be submitted before
7155                          * returning so that fdatawait does what its supposed to
7156                          * do.
7157                          */
7158                         btrfs_wait_ordered_range(inode, 0, (u64)-1);
7159                 } else {
7160                         filemap_flush(inode->i_mapping);
7161                 }
7162                 if (delay_iput)
7163                         btrfs_add_delayed_iput(inode);
7164                 else
7165                         iput(inode);
7166                 return 1;
7167         }
7168         return 0;
7169 }
7170
7171 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7172                          const char *symname)
7173 {
7174         struct btrfs_trans_handle *trans;
7175         struct btrfs_root *root = BTRFS_I(dir)->root;
7176         struct btrfs_path *path;
7177         struct btrfs_key key;
7178         struct inode *inode = NULL;
7179         int err;
7180         int drop_inode = 0;
7181         u64 objectid;
7182         u64 index = 0 ;
7183         int name_len;
7184         int datasize;
7185         unsigned long ptr;
7186         struct btrfs_file_extent_item *ei;
7187         struct extent_buffer *leaf;
7188         unsigned long nr = 0;
7189
7190         name_len = strlen(symname) + 1;
7191         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7192                 return -ENAMETOOLONG;
7193
7194         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
7195         if (err)
7196                 return err;
7197         /*
7198          * 2 items for inode item and ref
7199          * 2 items for dir items
7200          * 1 item for xattr if selinux is on
7201          */
7202         trans = btrfs_start_transaction(root, 5);
7203         if (IS_ERR(trans))
7204                 return PTR_ERR(trans);
7205
7206         btrfs_set_trans_block_group(trans, dir);
7207
7208         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7209                                 dentry->d_name.len, dir->i_ino, objectid,
7210                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
7211                                 &index);
7212         err = PTR_ERR(inode);
7213         if (IS_ERR(inode))
7214                 goto out_unlock;
7215
7216         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7217         if (err) {
7218                 drop_inode = 1;
7219                 goto out_unlock;
7220         }
7221
7222         btrfs_set_trans_block_group(trans, inode);
7223         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7224         if (err)
7225                 drop_inode = 1;
7226         else {
7227                 inode->i_mapping->a_ops = &btrfs_aops;
7228                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7229                 inode->i_fop = &btrfs_file_operations;
7230                 inode->i_op = &btrfs_file_inode_operations;
7231                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7232         }
7233         btrfs_update_inode_block_group(trans, inode);
7234         btrfs_update_inode_block_group(trans, dir);
7235         if (drop_inode)
7236                 goto out_unlock;
7237
7238         path = btrfs_alloc_path();
7239         BUG_ON(!path);
7240         key.objectid = inode->i_ino;
7241         key.offset = 0;
7242         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7243         datasize = btrfs_file_extent_calc_inline_size(name_len);
7244         err = btrfs_insert_empty_item(trans, root, path, &key,
7245                                       datasize);
7246         if (err) {
7247                 drop_inode = 1;
7248                 goto out_unlock;
7249         }
7250         leaf = path->nodes[0];
7251         ei = btrfs_item_ptr(leaf, path->slots[0],
7252                             struct btrfs_file_extent_item);
7253         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7254         btrfs_set_file_extent_type(leaf, ei,
7255                                    BTRFS_FILE_EXTENT_INLINE);
7256         btrfs_set_file_extent_encryption(leaf, ei, 0);
7257         btrfs_set_file_extent_compression(leaf, ei, 0);
7258         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7259         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7260
7261         ptr = btrfs_file_extent_inline_start(ei);
7262         write_extent_buffer(leaf, symname, ptr, name_len);
7263         btrfs_mark_buffer_dirty(leaf);
7264         btrfs_free_path(path);
7265
7266         inode->i_op = &btrfs_symlink_inode_operations;
7267         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7268         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7269         inode_set_bytes(inode, name_len);
7270         btrfs_i_size_write(inode, name_len - 1);
7271         err = btrfs_update_inode(trans, root, inode);
7272         if (err)
7273                 drop_inode = 1;
7274
7275 out_unlock:
7276         nr = trans->blocks_used;
7277         btrfs_end_transaction_throttle(trans, root);
7278         if (drop_inode) {
7279                 inode_dec_link_count(inode);
7280                 iput(inode);
7281         }
7282         btrfs_btree_balance_dirty(root, nr);
7283         return err;
7284 }
7285
7286 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7287                                        u64 start, u64 num_bytes, u64 min_size,
7288                                        loff_t actual_len, u64 *alloc_hint,
7289                                        struct btrfs_trans_handle *trans)
7290 {
7291         struct btrfs_root *root = BTRFS_I(inode)->root;
7292         struct btrfs_key ins;
7293         u64 cur_offset = start;
7294         u64 i_size;
7295         int ret = 0;
7296         bool own_trans = true;
7297
7298         if (trans)
7299                 own_trans = false;
7300         while (num_bytes > 0) {
7301                 if (own_trans) {
7302                         trans = btrfs_start_transaction(root, 3);
7303                         if (IS_ERR(trans)) {
7304                                 ret = PTR_ERR(trans);
7305                                 break;
7306                         }
7307                 }
7308
7309                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7310                                            0, *alloc_hint, (u64)-1, &ins, 1);
7311                 if (ret) {
7312                         if (own_trans)
7313                                 btrfs_end_transaction(trans, root);
7314                         break;
7315                 }
7316
7317                 ret = insert_reserved_file_extent(trans, inode,
7318                                                   cur_offset, ins.objectid,
7319                                                   ins.offset, ins.offset,
7320                                                   ins.offset, 0, 0, 0,
7321                                                   BTRFS_FILE_EXTENT_PREALLOC);
7322                 BUG_ON(ret);
7323                 btrfs_drop_extent_cache(inode, cur_offset,
7324                                         cur_offset + ins.offset -1, 0);
7325
7326                 num_bytes -= ins.offset;
7327                 cur_offset += ins.offset;
7328                 *alloc_hint = ins.objectid + ins.offset;
7329
7330                 inode->i_ctime = CURRENT_TIME;
7331                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7332                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7333                     (actual_len > inode->i_size) &&
7334                     (cur_offset > inode->i_size)) {
7335                         if (cur_offset > actual_len)
7336                                 i_size = actual_len;
7337                         else
7338                                 i_size = cur_offset;
7339                         i_size_write(inode, i_size);
7340                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7341                 }
7342
7343                 ret = btrfs_update_inode(trans, root, inode);
7344                 BUG_ON(ret);
7345
7346                 if (own_trans)
7347                         btrfs_end_transaction(trans, root);
7348         }
7349         return ret;
7350 }
7351
7352 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7353                               u64 start, u64 num_bytes, u64 min_size,
7354                               loff_t actual_len, u64 *alloc_hint)
7355 {
7356         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7357                                            min_size, actual_len, alloc_hint,
7358                                            NULL);
7359 }
7360
7361 int btrfs_prealloc_file_range_trans(struct inode *inode,
7362                                     struct btrfs_trans_handle *trans, int mode,
7363                                     u64 start, u64 num_bytes, u64 min_size,
7364                                     loff_t actual_len, u64 *alloc_hint)
7365 {
7366         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7367                                            min_size, actual_len, alloc_hint, trans);
7368 }
7369
7370 static int btrfs_set_page_dirty(struct page *page)
7371 {
7372         return __set_page_dirty_nobuffers(page);
7373 }
7374
7375 static int btrfs_permission(struct inode *inode, int mask, unsigned int flags)
7376 {
7377         struct btrfs_root *root = BTRFS_I(inode)->root;
7378
7379         if (btrfs_root_readonly(root) && (mask & MAY_WRITE))
7380                 return -EROFS;
7381         if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
7382                 return -EACCES;
7383         return generic_permission(inode, mask, flags, btrfs_check_acl);
7384 }
7385
7386 static const struct inode_operations btrfs_dir_inode_operations = {
7387         .getattr        = btrfs_getattr,
7388         .lookup         = btrfs_lookup,
7389         .create         = btrfs_create,
7390         .unlink         = btrfs_unlink,
7391         .link           = btrfs_link,
7392         .mkdir          = btrfs_mkdir,
7393         .rmdir          = btrfs_rmdir,
7394         .rename         = btrfs_rename,
7395         .symlink        = btrfs_symlink,
7396         .setattr        = btrfs_setattr,
7397         .mknod          = btrfs_mknod,
7398         .setxattr       = btrfs_setxattr,
7399         .getxattr       = btrfs_getxattr,
7400         .listxattr      = btrfs_listxattr,
7401         .removexattr    = btrfs_removexattr,
7402         .permission     = btrfs_permission,
7403 };
7404 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7405         .lookup         = btrfs_lookup,
7406         .permission     = btrfs_permission,
7407 };
7408
7409 static const struct file_operations btrfs_dir_file_operations = {
7410         .llseek         = generic_file_llseek,
7411         .read           = generic_read_dir,
7412         .readdir        = btrfs_real_readdir,
7413         .unlocked_ioctl = btrfs_ioctl,
7414 #ifdef CONFIG_COMPAT
7415         .compat_ioctl   = btrfs_ioctl,
7416 #endif
7417         .release        = btrfs_release_file,
7418         .fsync          = btrfs_sync_file,
7419 };
7420
7421 static struct extent_io_ops btrfs_extent_io_ops = {
7422         .fill_delalloc = run_delalloc_range,
7423         .submit_bio_hook = btrfs_submit_bio_hook,
7424         .merge_bio_hook = btrfs_merge_bio_hook,
7425         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7426         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7427         .writepage_start_hook = btrfs_writepage_start_hook,
7428         .readpage_io_failed_hook = btrfs_io_failed_hook,
7429         .set_bit_hook = btrfs_set_bit_hook,
7430         .clear_bit_hook = btrfs_clear_bit_hook,
7431         .merge_extent_hook = btrfs_merge_extent_hook,
7432         .split_extent_hook = btrfs_split_extent_hook,
7433 };
7434
7435 /*
7436  * btrfs doesn't support the bmap operation because swapfiles
7437  * use bmap to make a mapping of extents in the file.  They assume
7438  * these extents won't change over the life of the file and they
7439  * use the bmap result to do IO directly to the drive.
7440  *
7441  * the btrfs bmap call would return logical addresses that aren't
7442  * suitable for IO and they also will change frequently as COW
7443  * operations happen.  So, swapfile + btrfs == corruption.
7444  *
7445  * For now we're avoiding this by dropping bmap.
7446  */
7447 static const struct address_space_operations btrfs_aops = {
7448         .readpage       = btrfs_readpage,
7449         .writepage      = btrfs_writepage,
7450         .writepages     = btrfs_writepages,
7451         .readpages      = btrfs_readpages,
7452         .direct_IO      = btrfs_direct_IO,
7453         .invalidatepage = btrfs_invalidatepage,
7454         .releasepage    = btrfs_releasepage,
7455         .set_page_dirty = btrfs_set_page_dirty,
7456         .error_remove_page = generic_error_remove_page,
7457 };
7458
7459 static const struct address_space_operations btrfs_symlink_aops = {
7460         .readpage       = btrfs_readpage,
7461         .writepage      = btrfs_writepage,
7462         .invalidatepage = btrfs_invalidatepage,
7463         .releasepage    = btrfs_releasepage,
7464 };
7465
7466 static const struct inode_operations btrfs_file_inode_operations = {
7467         .getattr        = btrfs_getattr,
7468         .setattr        = btrfs_setattr,
7469         .setxattr       = btrfs_setxattr,
7470         .getxattr       = btrfs_getxattr,
7471         .listxattr      = btrfs_listxattr,
7472         .removexattr    = btrfs_removexattr,
7473         .permission     = btrfs_permission,
7474         .fiemap         = btrfs_fiemap,
7475 };
7476 static const struct inode_operations btrfs_special_inode_operations = {
7477         .getattr        = btrfs_getattr,
7478         .setattr        = btrfs_setattr,
7479         .permission     = btrfs_permission,
7480         .setxattr       = btrfs_setxattr,
7481         .getxattr       = btrfs_getxattr,
7482         .listxattr      = btrfs_listxattr,
7483         .removexattr    = btrfs_removexattr,
7484 };
7485 static const struct inode_operations btrfs_symlink_inode_operations = {
7486         .readlink       = generic_readlink,
7487         .follow_link    = page_follow_link_light,
7488         .put_link       = page_put_link,
7489         .getattr        = btrfs_getattr,
7490         .permission     = btrfs_permission,
7491         .setxattr       = btrfs_setxattr,
7492         .getxattr       = btrfs_getxattr,
7493         .listxattr      = btrfs_listxattr,
7494         .removexattr    = btrfs_removexattr,
7495 };
7496
7497 const struct dentry_operations btrfs_dentry_operations = {
7498         .d_delete       = btrfs_dentry_delete,
7499 };