fs/btrfs/inode.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/pagemap.h>
  25 #include <linux/highmem.h>
  26 #include <linux/time.h>
  27 #include <linux/init.h>
  28 #include <linux/string.h>
  29 #include <linux/backing-dev.h>
  30 #include <linux/mpage.h>
  31 #include <linux/swap.h>
  32 #include <linux/writeback.h>
  33 #include <linux/statfs.h>
  34 #include <linux/compat.h>
  35 #include <linux/bit_spinlock.h>
  36 #include <linux/xattr.h>
  37 #include <linux/posix_acl.h>
  38 #include <linux/falloc.h>
  39 #include <linux/slab.h>
  40 #include <linux/ratelimit.h>
  41 #include <linux/mount.h>
  42 #include "compat.h"
  43 #include "ctree.h"
  44 #include "disk-io.h"
  45 #include "transaction.h"
  46 #include "btrfs_inode.h"
  47 #include "ioctl.h"
  48 #include "print-tree.h"
  49 #include "ordered-data.h"
  50 #include "xattr.h"
  51 #include "tree-log.h"
  52 #include "volumes.h"
  53 #include "compression.h"
  54 #include "locking.h"
  55 #include "free-space-cache.h"
  56 #include "inode-map.h"
  57
  58 struct btrfs_iget_args {
  59         u64 ino;
  60         struct btrfs_root *root;
  61 };
  62
  63 static const struct inode_operations btrfs_dir_inode_operations;
  64 static const struct inode_operations btrfs_symlink_inode_operations;
  65 static const struct inode_operations btrfs_dir_ro_inode_operations;
  66 static const struct inode_operations btrfs_special_inode_operations;
  67 static const struct inode_operations btrfs_file_inode_operations;
  68 static const struct address_space_operations btrfs_aops;
  69 static const struct address_space_operations btrfs_symlink_aops;
  70 static const struct file_operations btrfs_dir_file_operations;
  71 static struct extent_io_ops btrfs_extent_io_ops;
  72
  73 static struct kmem_cache *btrfs_inode_cachep;
  74 struct kmem_cache *btrfs_trans_handle_cachep;
  75 struct kmem_cache *btrfs_transaction_cachep;
  76 struct kmem_cache *btrfs_path_cachep;
  77 struct kmem_cache *btrfs_free_space_cachep;
  78
  79 #define S_SHIFT 12
  80 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  81         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  82         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  83         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  84         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  85         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  86         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  87         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  88 };
  89
  90 static int btrfs_setsize(struct inode *inode, loff_t newsize);
  91 static int btrfs_truncate(struct inode *inode);
  92 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
  93 static noinline int cow_file_range(struct inode *inode,
  94                                    struct page *locked_page,
  95                                    u64 start, u64 end, int *page_started,
  96                                    unsigned long *nr_written, int unlock);
  97 static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
  98                                 struct btrfs_root *root, struct inode *inode);
  99
 100 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
 101                                      struct inode *inode,  struct inode *dir,
 102                                      const struct qstr *qstr)
 103 {
 104         int err;
 105
 106         err = btrfs_init_acl(trans, inode, dir);
 107         if (!err)
 108                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 109         return err;
 110 }
 111
 112 /*
 113  * this does all the hard work for inserting an inline extent into
 114  * the btree.  The caller should have done a btrfs_drop_extents so that
 115  * no overlapping inline items exist in the btree
 116  */
 117 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 118                                 struct btrfs_root *root, struct inode *inode,
 119                                 u64 start, size_t size, size_t compressed_size,
 120                                 int compress_type,
 121                                 struct page **compressed_pages)
 122 {
 123         struct btrfs_key key;
 124         struct btrfs_path *path;
 125         struct extent_buffer *leaf;
 126         struct page *page = NULL;
 127         char *kaddr;
 128         unsigned long ptr;
 129         struct btrfs_file_extent_item *ei;
 130         int err = 0;
 131         int ret;
 132         size_t cur_size = size;
 133         size_t datasize;
 134         unsigned long offset;
 135
 136         if (compressed_size && compressed_pages)
 137                 cur_size = compressed_size;
 138
 139         path = btrfs_alloc_path();
 140         if (!path)
 141                 return -ENOMEM;
 142
 143         path->leave_spinning = 1;
 144
 145         key.objectid = btrfs_ino(inode);
 146         key.offset = start;
 147         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 148         datasize = btrfs_file_extent_calc_inline_size(cur_size);
 149
 150         inode_add_bytes(inode, size);
 151         ret = btrfs_insert_empty_item(trans, root, path, &key,
 152                                       datasize);
 153         if (ret) {
 154                 err = ret;
 155                 goto fail;
 156         }
 157         leaf = path->nodes[0];
 158         ei = btrfs_item_ptr(leaf, path->slots[0],
 159                             struct btrfs_file_extent_item);
 160         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 161         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 162         btrfs_set_file_extent_encryption(leaf, ei, 0);
 163         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 164         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 165         ptr = btrfs_file_extent_inline_start(ei);
 166
 167         if (compress_type != BTRFS_COMPRESS_NONE) {
 168                 struct page *cpage;
 169                 int i = 0;
 170                 while (compressed_size > 0) {
 171                         cpage = compressed_pages[i];
 172                         cur_size = min_t(unsigned long, compressed_size,
 173                                        PAGE_CACHE_SIZE);
 174
 175                         kaddr = kmap_atomic(cpage);
 176                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 177                         kunmap_atomic(kaddr);
 178
 179                         i++;
 180                         ptr += cur_size;
 181                         compressed_size -= cur_size;
 182                 }
 183                 btrfs_set_file_extent_compression(leaf, ei,
 184                                                   compress_type);
 185         } else {
 186                 page = find_get_page(inode->i_mapping,
 187                                      start >> PAGE_CACHE_SHIFT);
 188                 btrfs_set_file_extent_compression(leaf, ei, 0);
 189                 kaddr = kmap_atomic(page);
 190                 offset = start & (PAGE_CACHE_SIZE - 1);
 191                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 192                 kunmap_atomic(kaddr);
 193                 page_cache_release(page);
 194         }
 195         btrfs_mark_buffer_dirty(leaf);
 196         btrfs_free_path(path);
 197
 198         /*
 199          * we're an inline extent, so nobody can
 200          * extend the file past i_size without locking
 201          * a page we already have locked.
 202          *
 203          * We must do any isize and inode updates
 204          * before we unlock the pages.  Otherwise we
 205          * could end up racing with unlink.
 206          */
 207         BTRFS_I(inode)->disk_i_size = inode->i_size;
 208         ret = btrfs_update_inode(trans, root, inode);
 209
 210         return ret;
 211 fail:
 212         btrfs_free_path(path);
 213         return err;
 214 }
 215
 216
 217 /*
 218  * conditionally insert an inline extent into the file.  This
 219  * does the checks required to make sure the data is small enough
 220  * to fit as an inline extent.
 221  */
 222 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 223                                  struct btrfs_root *root,
 224                                  struct inode *inode, u64 start, u64 end,
 225                                  size_t compressed_size, int compress_type,
 226                                  struct page **compressed_pages)
 227 {
 228         u64 isize = i_size_read(inode);
 229         u64 actual_end = min(end + 1, isize);
 230         u64 inline_len = actual_end - start;
 231         u64 aligned_end = (end + root->sectorsize - 1) &
 232                         ~((u64)root->sectorsize - 1);
 233         u64 hint_byte;
 234         u64 data_len = inline_len;
 235         int ret;
 236
 237         if (compressed_size)
 238                 data_len = compressed_size;
 239
 240         if (start > 0 ||
 241             actual_end >= PAGE_CACHE_SIZE ||
 242             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 243             (!compressed_size &&
 244             (actual_end & (root->sectorsize - 1)) == 0) ||
 245             end + 1 < isize ||
 246             data_len > root->fs_info->max_inline) {
 247                 return 1;
 248         }
 249
 250         ret = btrfs_drop_extents(trans, inode, start, aligned_end,
 251                                  &hint_byte, 1);
 252         if (ret)
 253                 return ret;
 254
 255         if (isize > actual_end)
 256                 inline_len = min_t(u64, isize, actual_end);
 257         ret = insert_inline_extent(trans, root, inode, start,
 258                                    inline_len, compressed_size,
 259                                    compress_type, compressed_pages);
 260         if (ret && ret != -ENOSPC) {
 261                 btrfs_abort_transaction(trans, root, ret);
 262                 return ret;
 263         } else if (ret == -ENOSPC) {
 264                 return 1;
 265         }
 266
 267         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 268         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 269         return 0;
 270 }
 271
 272 struct async_extent {
 273         u64 start;
 274         u64 ram_size;
 275         u64 compressed_size;
 276         struct page **pages;
 277         unsigned long nr_pages;
 278         int compress_type;
 279         struct list_head list;
 280 };
 281
 282 struct async_cow {
 283         struct inode *inode;
 284         struct btrfs_root *root;
 285         struct page *locked_page;
 286         u64 start;
 287         u64 end;
 288         struct list_head extents;
 289         struct btrfs_work work;
 290 };
 291
 292 static noinline int add_async_extent(struct async_cow *cow,
 293                                      u64 start, u64 ram_size,
 294                                      u64 compressed_size,
 295                                      struct page **pages,
 296                                      unsigned long nr_pages,
 297                                      int compress_type)
 298 {
 299         struct async_extent *async_extent;
 300
 301         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 302         BUG_ON(!async_extent); /* -ENOMEM */
 303         async_extent->start = start;
 304         async_extent->ram_size = ram_size;
 305         async_extent->compressed_size = compressed_size;
 306         async_extent->pages = pages;
 307         async_extent->nr_pages = nr_pages;
 308         async_extent->compress_type = compress_type;
 309         list_add_tail(&async_extent->list, &cow->extents);
 310         return 0;
 311 }
 312
 313 /*
 314  * we create compressed extents in two phases.  The first
 315  * phase compresses a range of pages that have already been
 316  * locked (both pages and state bits are locked).
 317  *
 318  * This is done inside an ordered work queue, and the compression
 319  * is spread across many cpus.  The actual IO submission is step
 320  * two, and the ordered work queue takes care of making sure that
 321  * happens in the same order things were put onto the queue by
 322  * writepages and friends.
 323  *
 324  * If this code finds it can't get good compression, it puts an
 325  * entry onto the work queue to write the uncompressed bytes.  This
 326  * makes sure that both compressed inodes and uncompressed inodes
 327  * are written in the same order that the flusher thread sent them
 328  * down.
 329  */
 330 static noinline int compress_file_range(struct inode *inode,
 331                                         struct page *locked_page,
 332                                         u64 start, u64 end,
 333                                         struct async_cow *async_cow,
 334                                         int *num_added)
 335 {
 336         struct btrfs_root *root = BTRFS_I(inode)->root;
 337         struct btrfs_trans_handle *trans;
 338         u64 num_bytes;
 339         u64 blocksize = root->sectorsize;
 340         u64 actual_end;
 341         u64 isize = i_size_read(inode);
 342         int ret = 0;
 343         struct page **pages = NULL;
 344         unsigned long nr_pages;
 345         unsigned long nr_pages_ret = 0;
 346         unsigned long total_compressed = 0;
 347         unsigned long total_in = 0;
 348         unsigned long max_compressed = 128 * 1024;
 349         unsigned long max_uncompressed = 128 * 1024;
 350         int i;
 351         int will_compress;
 352         int compress_type = root->fs_info->compress_type;
 353
 354         /* if this is a small write inside eof, kick off a defrag */
 355         if ((end - start + 1) < 16 * 1024 &&
 356             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 357                 btrfs_add_inode_defrag(NULL, inode);
 358
 359         actual_end = min_t(u64, isize, end + 1);
 360 again:
 361         will_compress = 0;
 362         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 363         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 364
 365         /*
 366          * we don't want to send crud past the end of i_size through
 367          * compression, that's just a waste of CPU time.  So, if the
 368          * end of the file is before the start of our current
 369          * requested range of bytes, we bail out to the uncompressed
 370          * cleanup code that can deal with all of this.
 371          *
 372          * It isn't really the fastest way to fix things, but this is a
 373          * very uncommon corner.
 374          */
 375         if (actual_end <= start)
 376                 goto cleanup_and_bail_uncompressed;
 377
 378         total_compressed = actual_end - start;
 379
 380         /* we want to make sure that amount of ram required to uncompress
 381          * an extent is reasonable, so we limit the total size in ram
 382          * of a compressed extent to 128k.  This is a crucial number
 383          * because it also controls how easily we can spread reads across
 384          * cpus for decompression.
 385          *
 386          * We also want to make sure the amount of IO required to do
 387          * a random read is reasonably small, so we limit the size of
 388          * a compressed extent to 128k.
 389          */
 390         total_compressed = min(total_compressed, max_uncompressed);
 391         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 392         num_bytes = max(blocksize,  num_bytes);
 393         total_in = 0;
 394         ret = 0;
 395
 396         /*
 397          * we do compression for mount -o compress and when the
 398          * inode has not been flagged as nocompress.  This flag can
 399          * change at any time if we discover bad compression ratios.
 400          */
 401         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 402             (btrfs_test_opt(root, COMPRESS) ||
 403              (BTRFS_I(inode)->force_compress) ||
 404              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 405                 WARN_ON(pages);
 406                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 407                 if (!pages) {
 408                         /* just bail out to the uncompressed code */
 409                         goto cont;
 410                 }
 411
 412                 if (BTRFS_I(inode)->force_compress)
 413                         compress_type = BTRFS_I(inode)->force_compress;
 414
 415                 ret = btrfs_compress_pages(compress_type,
 416                                            inode->i_mapping, start,
 417                                            total_compressed, pages,
 418                                            nr_pages, &nr_pages_ret,
 419                                            &total_in,
 420                                            &total_compressed,
 421                                            max_compressed);
 422
 423                 if (!ret) {
 424                         unsigned long offset = total_compressed &
 425                                 (PAGE_CACHE_SIZE - 1);
 426                         struct page *page = pages[nr_pages_ret - 1];
 427                         char *kaddr;
 428
 429                         /* zero the tail end of the last page, we might be
 430                          * sending it down to disk
 431                          */
 432                         if (offset) {
 433                                 kaddr = kmap_atomic(page);
 434                                 memset(kaddr + offset, 0,
 435                                        PAGE_CACHE_SIZE - offset);
 436                                 kunmap_atomic(kaddr);
 437                         }
 438                         will_compress = 1;
 439                 }
 440         }
 441 cont:
 442         if (start == 0) {
 443                 trans = btrfs_join_transaction(root);
 444                 if (IS_ERR(trans)) {
 445                         ret = PTR_ERR(trans);
 446                         trans = NULL;
 447                         goto cleanup_and_out;
 448                 }
 449                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 450
 451                 /* lets try to make an inline extent */
 452                 if (ret || total_in < (actual_end - start)) {
 453                         /* we didn't compress the entire range, try
 454                          * to make an uncompressed inline extent.
 455                          */
 456                         ret = cow_file_range_inline(trans, root, inode,
 457                                                     start, end, 0, 0, NULL);
 458                 } else {
 459                         /* try making a compressed inline extent */
 460                         ret = cow_file_range_inline(trans, root, inode,
 461                                                     start, end,
 462                                                     total_compressed,
 463                                                     compress_type, pages);
 464                 }
 465                 if (ret <= 0) {
 466                         /*
 467                          * inline extent creation worked or returned error,
 468                          * we don't need to create any more async work items.
 469                          * Unlock and free up our temp pages.
 470                          */
 471                         extent_clear_unlock_delalloc(inode,
 472                              &BTRFS_I(inode)->io_tree,
 473                              start, end, NULL,
 474                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
 475                              EXTENT_CLEAR_DELALLOC |
 476                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
 477
 478                         btrfs_end_transaction(trans, root);
 479                         goto free_pages_out;
 480                 }
 481                 btrfs_end_transaction(trans, root);
 482         }
 483
 484         if (will_compress) {
 485                 /*
 486                  * we aren't doing an inline extent round the compressed size
 487                  * up to a block size boundary so the allocator does sane
 488                  * things
 489                  */
 490                 total_compressed = (total_compressed + blocksize - 1) &
 491                         ~(blocksize - 1);
 492
 493                 /*
 494                  * one last check to make sure the compression is really a
 495                  * win, compare the page count read with the blocks on disk
 496                  */
 497                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
 498                         ~(PAGE_CACHE_SIZE - 1);
 499                 if (total_compressed >= total_in) {
 500                         will_compress = 0;
 501                 } else {
 502                         num_bytes = total_in;
 503                 }
 504         }
 505         if (!will_compress && pages) {
 506                 /*
 507                  * the compression code ran but failed to make things smaller,
 508                  * free any pages it allocated and our page pointer array
 509                  */
 510                 for (i = 0; i < nr_pages_ret; i++) {
 511                         WARN_ON(pages[i]->mapping);
 512                         page_cache_release(pages[i]);
 513                 }
 514                 kfree(pages);
 515                 pages = NULL;
 516                 total_compressed = 0;
 517                 nr_pages_ret = 0;
 518
 519                 /* flag the file so we don't compress in the future */
 520                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 521                     !(BTRFS_I(inode)->force_compress)) {
 522                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 523                 }
 524         }
 525         if (will_compress) {
 526                 *num_added += 1;
 527
 528                 /* the async work queues will take care of doing actual
 529                  * allocation on disk for these compressed pages,
 530                  * and will submit them to the elevator.
 531                  */
 532                 add_async_extent(async_cow, start, num_bytes,
 533                                  total_compressed, pages, nr_pages_ret,
 534                                  compress_type);
 535
 536                 if (start + num_bytes < end) {
 537                         start += num_bytes;
 538                         pages = NULL;
 539                         cond_resched();
 540                         goto again;
 541                 }
 542         } else {
 543 cleanup_and_bail_uncompressed:
 544                 /*
 545                  * No compression, but we still need to write the pages in
 546                  * the file we've been given so far.  redirty the locked
 547                  * page if it corresponds to our extent and set things up
 548                  * for the async work queue to run cow_file_range to do
 549                  * the normal delalloc dance
 550                  */
 551                 if (page_offset(locked_page) >= start &&
 552                     page_offset(locked_page) <= end) {
 553                         __set_page_dirty_nobuffers(locked_page);
 554                         /* unlocked later on in the async handlers */
 555                 }
 556                 add_async_extent(async_cow, start, end - start + 1,
 557                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 558                 *num_added += 1;
 559         }
 560
 561 out:
 562         return ret;
 563
 564 free_pages_out:
 565         for (i = 0; i < nr_pages_ret; i++) {
 566                 WARN_ON(pages[i]->mapping);
 567                 page_cache_release(pages[i]);
 568         }
 569         kfree(pages);
 570
 571         goto out;
 572
 573 cleanup_and_out:
 574         extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 575                                      start, end, NULL,
 576                                      EXTENT_CLEAR_UNLOCK_PAGE |
 577                                      EXTENT_CLEAR_DIRTY |
 578                                      EXTENT_CLEAR_DELALLOC |
 579                                      EXTENT_SET_WRITEBACK |
 580                                      EXTENT_END_WRITEBACK);
 581         if (!trans || IS_ERR(trans))
 582                 btrfs_error(root->fs_info, ret, "Failed to join transaction");
 583         else
 584                 btrfs_abort_transaction(trans, root, ret);
 585         goto free_pages_out;
 586 }
 587
 588 /*
 589  * phase two of compressed writeback.  This is the ordered portion
 590  * of the code, which only gets called in the order the work was
 591  * queued.  We walk all the async extents created by compress_file_range
 592  * and send them down to the disk.
 593  */
 594 static noinline int submit_compressed_extents(struct inode *inode,
 595                                               struct async_cow *async_cow)
 596 {
 597         struct async_extent *async_extent;
 598         u64 alloc_hint = 0;
 599         struct btrfs_trans_handle *trans;
 600         struct btrfs_key ins;
 601         struct extent_map *em;
 602         struct btrfs_root *root = BTRFS_I(inode)->root;
 603         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 604         struct extent_io_tree *io_tree;
 605         int ret = 0;
 606
 607         if (list_empty(&async_cow->extents))
 608                 return 0;
 609
 610
 611         while (!list_empty(&async_cow->extents)) {
 612                 async_extent = list_entry(async_cow->extents.next,
 613                                           struct async_extent, list);
 614                 list_del(&async_extent->list);
 615
 616                 io_tree = &BTRFS_I(inode)->io_tree;
 617
 618 retry:
 619                 /* did the compression code fall back to uncompressed IO? */
 620                 if (!async_extent->pages) {
 621                         int page_started = 0;
 622                         unsigned long nr_written = 0;
 623
 624                         lock_extent(io_tree, async_extent->start,
 625                                          async_extent->start +
 626                                          async_extent->ram_size - 1);
 627
 628                         /* allocate blocks */
 629                         ret = cow_file_range(inode, async_cow->locked_page,
 630                                              async_extent->start,
 631                                              async_extent->start +
 632                                              async_extent->ram_size - 1,
 633                                              &page_started, &nr_written, 0);
 634
 635                         /* JDM XXX */
 636
 637                         /*
 638                          * if page_started, cow_file_range inserted an
 639                          * inline extent and took care of all the unlocking
 640                          * and IO for us.  Otherwise, we need to submit
 641                          * all those pages down to the drive.
 642                          */
 643                         if (!page_started && !ret)
 644                                 extent_write_locked_range(io_tree,
 645                                                   inode, async_extent->start,
 646                                                   async_extent->start +
 647                                                   async_extent->ram_size - 1,
 648                                                   btrfs_get_extent,
 649                                                   WB_SYNC_ALL);
 650                         kfree(async_extent);
 651                         cond_resched();
 652                         continue;
 653                 }
 654
 655                 lock_extent(io_tree, async_extent->start,
 656                             async_extent->start + async_extent->ram_size - 1);
 657
 658                 trans = btrfs_join_transaction(root);
 659                 if (IS_ERR(trans)) {
 660                         ret = PTR_ERR(trans);
 661                 } else {
 662                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 663                         ret = btrfs_reserve_extent(trans, root,
 664                                            async_extent->compressed_size,
 665                                            async_extent->compressed_size,
 666                                            0, alloc_hint, &ins, 1);
 667                         if (ret)
 668                                 btrfs_abort_transaction(trans, root, ret);
 669                         btrfs_end_transaction(trans, root);
 670                 }
 671
 672                 if (ret) {
 673                         int i;
 674                         for (i = 0; i < async_extent->nr_pages; i++) {
 675                                 WARN_ON(async_extent->pages[i]->mapping);
 676                                 page_cache_release(async_extent->pages[i]);
 677                         }
 678                         kfree(async_extent->pages);
 679                         async_extent->nr_pages = 0;
 680                         async_extent->pages = NULL;
 681                         unlock_extent(io_tree, async_extent->start,
 682                                       async_extent->start +
 683                                       async_extent->ram_size - 1);
 684                         if (ret == -ENOSPC)
 685                                 goto retry;
 686                         goto out_free; /* JDM: Requeue? */
 687                 }
 688
 689                 /*
 690                  * here we're doing allocation and writeback of the
 691                  * compressed pages
 692                  */
 693                 btrfs_drop_extent_cache(inode, async_extent->start,
 694                                         async_extent->start +
 695                                         async_extent->ram_size - 1, 0);
 696
 697                 em = alloc_extent_map();
 698                 BUG_ON(!em); /* -ENOMEM */
 699                 em->start = async_extent->start;
 700                 em->len = async_extent->ram_size;
 701                 em->orig_start = em->start;
 702
 703                 em->block_start = ins.objectid;
 704                 em->block_len = ins.offset;
 705                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 706                 em->compress_type = async_extent->compress_type;
 707                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 708                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 709
 710                 while (1) {
 711                         write_lock(&em_tree->lock);
 712                         ret = add_extent_mapping(em_tree, em);
 713                         write_unlock(&em_tree->lock);
 714                         if (ret != -EEXIST) {
 715                                 free_extent_map(em);
 716                                 break;
 717                         }
 718                         btrfs_drop_extent_cache(inode, async_extent->start,
 719                                                 async_extent->start +
 720                                                 async_extent->ram_size - 1, 0);
 721                 }
 722
 723                 ret = btrfs_add_ordered_extent_compress(inode,
 724                                                 async_extent->start,
 725                                                 ins.objectid,
 726                                                 async_extent->ram_size,
 727                                                 ins.offset,
 728                                                 BTRFS_ORDERED_COMPRESSED,
 729                                                 async_extent->compress_type);
 730                 BUG_ON(ret); /* -ENOMEM */
 731
 732                 /*
 733                  * clear dirty, set writeback and unlock the pages.
 734                  */
 735                 extent_clear_unlock_delalloc(inode,
 736                                 &BTRFS_I(inode)->io_tree,
 737                                 async_extent->start,
 738                                 async_extent->start +
 739                                 async_extent->ram_size - 1,
 740                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 741                                 EXTENT_CLEAR_UNLOCK |
 742                                 EXTENT_CLEAR_DELALLOC |
 743                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
 744
 745                 ret = btrfs_submit_compressed_write(inode,
 746                                     async_extent->start,
 747                                     async_extent->ram_size,
 748                                     ins.objectid,
 749                                     ins.offset, async_extent->pages,
 750                                     async_extent->nr_pages);
 751
 752                 BUG_ON(ret); /* -ENOMEM */
 753                 alloc_hint = ins.objectid + ins.offset;
 754                 kfree(async_extent);
 755                 cond_resched();
 756         }
 757         ret = 0;
 758 out:
 759         return ret;
 760 out_free:
 761         kfree(async_extent);
 762         goto out;
 763 }
 764
 765 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 766                                       u64 num_bytes)
 767 {
 768         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 769         struct extent_map *em;
 770         u64 alloc_hint = 0;
 771
 772         read_lock(&em_tree->lock);
 773         em = search_extent_mapping(em_tree, start, num_bytes);
 774         if (em) {
 775                 /*
 776                  * if block start isn't an actual block number then find the
 777                  * first block in this inode and use that as a hint.  If that
 778                  * block is also bogus then just don't worry about it.
 779                  */
 780                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 781                         free_extent_map(em);
 782                         em = search_extent_mapping(em_tree, 0, 0);
 783                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 784                                 alloc_hint = em->block_start;
 785                         if (em)
 786                                 free_extent_map(em);
 787                 } else {
 788                         alloc_hint = em->block_start;
 789                         free_extent_map(em);
 790                 }
 791         }
 792         read_unlock(&em_tree->lock);
 793
 794         return alloc_hint;
 795 }
 796
 797 /*
 798  * when extent_io.c finds a delayed allocation range in the file,
 799  * the call backs end up in this code.  The basic idea is to
 800  * allocate extents on disk for the range, and create ordered data structs
 801  * in ram to track those extents.
 802  *
 803  * locked_page is the page that writepage had locked already.  We use
 804  * it to make sure we don't do extra locks or unlocks.
 805  *
 806  * *page_started is set to one if we unlock locked_page and do everything
 807  * required to start IO on it.  It may be clean and already done with
 808  * IO when we return.
 809  */
 810 static noinline int cow_file_range(struct inode *inode,
 811                                    struct page *locked_page,
 812                                    u64 start, u64 end, int *page_started,
 813                                    unsigned long *nr_written,
 814                                    int unlock)
 815 {
 816         struct btrfs_root *root = BTRFS_I(inode)->root;
 817         struct btrfs_trans_handle *trans;
 818         u64 alloc_hint = 0;
 819         u64 num_bytes;
 820         unsigned long ram_size;
 821         u64 disk_num_bytes;
 822         u64 cur_alloc_size;
 823         u64 blocksize = root->sectorsize;
 824         struct btrfs_key ins;
 825         struct extent_map *em;
 826         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 827         int ret = 0;
 828
 829         BUG_ON(btrfs_is_free_space_inode(inode));
 830         trans = btrfs_join_transaction(root);
 831         if (IS_ERR(trans)) {
 832                 extent_clear_unlock_delalloc(inode,
 833                              &BTRFS_I(inode)->io_tree,
 834                              start, end, locked_page,
 835                              EXTENT_CLEAR_UNLOCK_PAGE |
 836                              EXTENT_CLEAR_UNLOCK |
 837                              EXTENT_CLEAR_DELALLOC |
 838                              EXTENT_CLEAR_DIRTY |
 839                              EXTENT_SET_WRITEBACK |
 840                              EXTENT_END_WRITEBACK);
 841                 return PTR_ERR(trans);
 842         }
 843         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 844
 845         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 846         num_bytes = max(blocksize,  num_bytes);
 847         disk_num_bytes = num_bytes;
 848         ret = 0;
 849
 850         /* if this is a small write inside eof, kick off defrag */
 851         if (num_bytes < 64 * 1024 &&
 852             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 853                 btrfs_add_inode_defrag(trans, inode);
 854
 855         if (start == 0) {
 856                 /* lets try to make an inline extent */
 857                 ret = cow_file_range_inline(trans, root, inode,
 858                                             start, end, 0, 0, NULL);
 859                 if (ret == 0) {
 860                         extent_clear_unlock_delalloc(inode,
 861                                      &BTRFS_I(inode)->io_tree,
 862                                      start, end, NULL,
 863                                      EXTENT_CLEAR_UNLOCK_PAGE |
 864                                      EXTENT_CLEAR_UNLOCK |
 865                                      EXTENT_CLEAR_DELALLOC |
 866                                      EXTENT_CLEAR_DIRTY |
 867                                      EXTENT_SET_WRITEBACK |
 868                                      EXTENT_END_WRITEBACK);
 869
 870                         *nr_written = *nr_written +
 871                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 872                         *page_started = 1;
 873                         goto out;
 874                 } else if (ret < 0) {
 875                         btrfs_abort_transaction(trans, root, ret);
 876                         goto out_unlock;
 877                 }
 878         }
 879
 880         BUG_ON(disk_num_bytes >
 881                btrfs_super_total_bytes(root->fs_info->super_copy));
 882
 883         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 884         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 885
 886         while (disk_num_bytes > 0) {
 887                 unsigned long op;
 888
 889                 cur_alloc_size = disk_num_bytes;
 890                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 891                                            root->sectorsize, 0, alloc_hint,
 892                                            &ins, 1);
 893                 if (ret < 0) {
 894                         btrfs_abort_transaction(trans, root, ret);
 895                         goto out_unlock;
 896                 }
 897
 898                 em = alloc_extent_map();
 899                 BUG_ON(!em); /* -ENOMEM */
 900                 em->start = start;
 901                 em->orig_start = em->start;
 902                 ram_size = ins.offset;
 903                 em->len = ins.offset;
 904
 905                 em->block_start = ins.objectid;
 906                 em->block_len = ins.offset;
 907                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 908                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 909
 910                 while (1) {
 911                         write_lock(&em_tree->lock);
 912                         ret = add_extent_mapping(em_tree, em);
 913                         write_unlock(&em_tree->lock);
 914                         if (ret != -EEXIST) {
 915                                 free_extent_map(em);
 916                                 break;
 917                         }
 918                         btrfs_drop_extent_cache(inode, start,
 919                                                 start + ram_size - 1, 0);
 920                 }
 921
 922                 cur_alloc_size = ins.offset;
 923                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 924                                                ram_size, cur_alloc_size, 0);
 925                 BUG_ON(ret); /* -ENOMEM */
 926
 927                 if (root->root_key.objectid ==
 928                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 929                         ret = btrfs_reloc_clone_csums(inode, start,
 930                                                       cur_alloc_size);
 931                         if (ret) {
 932                                 btrfs_abort_transaction(trans, root, ret);
 933                                 goto out_unlock;
 934                         }
 935                 }
 936
 937                 if (disk_num_bytes < cur_alloc_size)
 938                         break;
 939
 940                 /* we're not doing compressed IO, don't unlock the first
 941                  * page (which the caller expects to stay locked), don't
 942                  * clear any dirty bits and don't set any writeback bits
 943                  *
 944                  * Do set the Private2 bit so we know this page was properly
 945                  * setup for writepage
 946                  */
 947                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
 948                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
 949                         EXTENT_SET_PRIVATE2;
 950
 951                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 952                                              start, start + ram_size - 1,
 953                                              locked_page, op);
 954                 disk_num_bytes -= cur_alloc_size;
 955                 num_bytes -= cur_alloc_size;
 956                 alloc_hint = ins.objectid + ins.offset;
 957                 start += cur_alloc_size;
 958         }
 959         ret = 0;
 960 out:
 961         btrfs_end_transaction(trans, root);
 962
 963         return ret;
 964 out_unlock:
 965         extent_clear_unlock_delalloc(inode,
 966                      &BTRFS_I(inode)->io_tree,
 967                      start, end, locked_page,
 968                      EXTENT_CLEAR_UNLOCK_PAGE |
 969                      EXTENT_CLEAR_UNLOCK |
 970                      EXTENT_CLEAR_DELALLOC |
 971                      EXTENT_CLEAR_DIRTY |
 972                      EXTENT_SET_WRITEBACK |
 973                      EXTENT_END_WRITEBACK);
 974
 975         goto out;
 976 }
 977
 978 /*
 979  * work queue call back to started compression on a file and pages
 980  */
 981 static noinline void async_cow_start(struct btrfs_work *work)
 982 {
 983         struct async_cow *async_cow;
 984         int num_added = 0;
 985         async_cow = container_of(work, struct async_cow, work);
 986
 987         compress_file_range(async_cow->inode, async_cow->locked_page,
 988                             async_cow->start, async_cow->end, async_cow,
 989                             &num_added);
 990         if (num_added == 0) {
 991                 btrfs_add_delayed_iput(async_cow->inode);
 992                 async_cow->inode = NULL;
 993         }
 994 }
 995
 996 /*
 997  * work queue call back to submit previously compressed pages
 998  */
 999 static noinline void async_cow_submit(struct btrfs_work *work)
1000 {
1001         struct async_cow *async_cow;
1002         struct btrfs_root *root;
1003         unsigned long nr_pages;
1004
1005         async_cow = container_of(work, struct async_cow, work);
1006
1007         root = async_cow->root;
1008         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1009                 PAGE_CACHE_SHIFT;
1010
1011         if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1012             5 * 1024 * 1024 &&
1013             waitqueue_active(&root->fs_info->async_submit_wait))
1014                 wake_up(&root->fs_info->async_submit_wait);
1015
1016         if (async_cow->inode)
1017                 submit_compressed_extents(async_cow->inode, async_cow);
1018 }
1019
1020 static noinline void async_cow_free(struct btrfs_work *work)
1021 {
1022         struct async_cow *async_cow;
1023         async_cow = container_of(work, struct async_cow, work);
1024         if (async_cow->inode)
1025                 btrfs_add_delayed_iput(async_cow->inode);
1026         kfree(async_cow);
1027 }
1028
1029 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1030                                 u64 start, u64 end, int *page_started,
1031                                 unsigned long *nr_written)
1032 {
1033         struct async_cow *async_cow;
1034         struct btrfs_root *root = BTRFS_I(inode)->root;
1035         unsigned long nr_pages;
1036         u64 cur_end;
1037         int limit = 10 * 1024 * 1024;
1038
1039         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1040                          1, 0, NULL, GFP_NOFS);
1041         while (start < end) {
1042                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1043                 BUG_ON(!async_cow); /* -ENOMEM */
1044                 async_cow->inode = igrab(inode);
1045                 async_cow->root = root;
1046                 async_cow->locked_page = locked_page;
1047                 async_cow->start = start;
1048
1049                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1050                         cur_end = end;
1051                 else
1052                         cur_end = min(end, start + 512 * 1024 - 1);
1053
1054                 async_cow->end = cur_end;
1055                 INIT_LIST_HEAD(&async_cow->extents);
1056
1057                 async_cow->work.func = async_cow_start;
1058                 async_cow->work.ordered_func = async_cow_submit;
1059                 async_cow->work.ordered_free = async_cow_free;
1060                 async_cow->work.flags = 0;
1061
1062                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1063                         PAGE_CACHE_SHIFT;
1064                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1065
1066                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1067                                    &async_cow->work);
1068
1069                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1070                         wait_event(root->fs_info->async_submit_wait,
1071                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1072                             limit));
1073                 }
1074
1075                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1076                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1077                         wait_event(root->fs_info->async_submit_wait,
1078                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1079                            0));
1080                 }
1081
1082                 *nr_written += nr_pages;
1083                 start = cur_end + 1;
1084         }
1085         *page_started = 1;
1086         return 0;
1087 }
1088
1089 static noinline int csum_exist_in_range(struct btrfs_root *root,
1090                                         u64 bytenr, u64 num_bytes)
1091 {
1092         int ret;
1093         struct btrfs_ordered_sum *sums;
1094         LIST_HEAD(list);
1095
1096         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1097                                        bytenr + num_bytes - 1, &list, 0);
1098         if (ret == 0 && list_empty(&list))
1099                 return 0;
1100
1101         while (!list_empty(&list)) {
1102                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1103                 list_del(&sums->list);
1104                 kfree(sums);
1105         }
1106         return 1;
1107 }
1108
1109 /*
1110  * when nowcow writeback call back.  This checks for snapshots or COW copies
1111  * of the extents that exist in the file, and COWs the file as required.
1112  *
1113  * If no cow copies or snapshots exist, we write directly to the existing
1114  * blocks on disk
1115  */
1116 static noinline int run_delalloc_nocow(struct inode *inode,
1117                                        struct page *locked_page,
1118                               u64 start, u64 end, int *page_started, int force,
1119                               unsigned long *nr_written)
1120 {
1121         struct btrfs_root *root = BTRFS_I(inode)->root;
1122         struct btrfs_trans_handle *trans;
1123         struct extent_buffer *leaf;
1124         struct btrfs_path *path;
1125         struct btrfs_file_extent_item *fi;
1126         struct btrfs_key found_key;
1127         u64 cow_start;
1128         u64 cur_offset;
1129         u64 extent_end;
1130         u64 extent_offset;
1131         u64 disk_bytenr;
1132         u64 num_bytes;
1133         int extent_type;
1134         int ret, err;
1135         int type;
1136         int nocow;
1137         int check_prev = 1;
1138         bool nolock;
1139         u64 ino = btrfs_ino(inode);
1140
1141         path = btrfs_alloc_path();
1142         if (!path) {
1143                 extent_clear_unlock_delalloc(inode,
1144                              &BTRFS_I(inode)->io_tree,
1145                              start, end, locked_page,
1146                              EXTENT_CLEAR_UNLOCK_PAGE |
1147                              EXTENT_CLEAR_UNLOCK |
1148                              EXTENT_CLEAR_DELALLOC |
1149                              EXTENT_CLEAR_DIRTY |
1150                              EXTENT_SET_WRITEBACK |
1151                              EXTENT_END_WRITEBACK);
1152                 return -ENOMEM;
1153         }
1154
1155         nolock = btrfs_is_free_space_inode(inode);
1156
1157         if (nolock)
1158                 trans = btrfs_join_transaction_nolock(root);
1159         else
1160                 trans = btrfs_join_transaction(root);
1161
1162         if (IS_ERR(trans)) {
1163                 extent_clear_unlock_delalloc(inode,
1164                              &BTRFS_I(inode)->io_tree,
1165                              start, end, locked_page,
1166                              EXTENT_CLEAR_UNLOCK_PAGE |
1167                              EXTENT_CLEAR_UNLOCK |
1168                              EXTENT_CLEAR_DELALLOC |
1169                              EXTENT_CLEAR_DIRTY |
1170                              EXTENT_SET_WRITEBACK |
1171                              EXTENT_END_WRITEBACK);
1172                 btrfs_free_path(path);
1173                 return PTR_ERR(trans);
1174         }
1175
1176         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1177
1178         cow_start = (u64)-1;
1179         cur_offset = start;
1180         while (1) {
1181                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1182                                                cur_offset, 0);
1183                 if (ret < 0) {
1184                         btrfs_abort_transaction(trans, root, ret);
1185                         goto error;
1186                 }
1187                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1188                         leaf = path->nodes[0];
1189                         btrfs_item_key_to_cpu(leaf, &found_key,
1190                                               path->slots[0] - 1);
1191                         if (found_key.objectid == ino &&
1192                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1193                                 path->slots[0]--;
1194                 }
1195                 check_prev = 0;
1196 next_slot:
1197                 leaf = path->nodes[0];
1198                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1199                         ret = btrfs_next_leaf(root, path);
1200                         if (ret < 0) {
1201                                 btrfs_abort_transaction(trans, root, ret);
1202                                 goto error;
1203                         }
1204                         if (ret > 0)
1205                                 break;
1206                         leaf = path->nodes[0];
1207                 }
1208
1209                 nocow = 0;
1210                 disk_bytenr = 0;
1211                 num_bytes = 0;
1212                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1213
1214                 if (found_key.objectid > ino ||
1215                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1216                     found_key.offset > end)
1217                         break;
1218
1219                 if (found_key.offset > cur_offset) {
1220                         extent_end = found_key.offset;
1221                         extent_type = 0;
1222                         goto out_check;
1223                 }
1224
1225                 fi = btrfs_item_ptr(leaf, path->slots[0],
1226                                     struct btrfs_file_extent_item);
1227                 extent_type = btrfs_file_extent_type(leaf, fi);
1228
1229                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1230                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1231                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1232                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1233                         extent_end = found_key.offset +
1234                                 btrfs_file_extent_num_bytes(leaf, fi);
1235                         if (extent_end <= start) {
1236                                 path->slots[0]++;
1237                                 goto next_slot;
1238                         }
1239                         if (disk_bytenr == 0)
1240                                 goto out_check;
1241                         if (btrfs_file_extent_compression(leaf, fi) ||
1242                             btrfs_file_extent_encryption(leaf, fi) ||
1243                             btrfs_file_extent_other_encoding(leaf, fi))
1244                                 goto out_check;
1245                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1246                                 goto out_check;
1247                         if (btrfs_extent_readonly(root, disk_bytenr))
1248                                 goto out_check;
1249                         if (btrfs_cross_ref_exist(trans, root, ino,
1250                                                   found_key.offset -
1251                                                   extent_offset, disk_bytenr))
1252                                 goto out_check;
1253                         disk_bytenr += extent_offset;
1254                         disk_bytenr += cur_offset - found_key.offset;
1255                         num_bytes = min(end + 1, extent_end) - cur_offset;
1256                         /*
1257                          * force cow if csum exists in the range.
1258                          * this ensure that csum for a given extent are
1259                          * either valid or do not exist.
1260                          */
1261                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1262                                 goto out_check;
1263                         nocow = 1;
1264                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1265                         extent_end = found_key.offset +
1266                                 btrfs_file_extent_inline_len(leaf, fi);
1267                         extent_end = ALIGN(extent_end, root->sectorsize);
1268                 } else {
1269                         BUG_ON(1);
1270                 }
1271 out_check:
1272                 if (extent_end <= start) {
1273                         path->slots[0]++;
1274                         goto next_slot;
1275                 }
1276                 if (!nocow) {
1277                         if (cow_start == (u64)-1)
1278                                 cow_start = cur_offset;
1279                         cur_offset = extent_end;
1280                         if (cur_offset > end)
1281                                 break;
1282                         path->slots[0]++;
1283                         goto next_slot;
1284                 }
1285
1286                 btrfs_release_path(path);
1287                 if (cow_start != (u64)-1) {
1288                         ret = cow_file_range(inode, locked_page, cow_start,
1289                                         found_key.offset - 1, page_started,
1290                                         nr_written, 1);
1291                         if (ret) {
1292                                 btrfs_abort_transaction(trans, root, ret);
1293                                 goto error;
1294                         }
1295                         cow_start = (u64)-1;
1296                 }
1297
1298                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1299                         struct extent_map *em;
1300                         struct extent_map_tree *em_tree;
1301                         em_tree = &BTRFS_I(inode)->extent_tree;
1302                         em = alloc_extent_map();
1303                         BUG_ON(!em); /* -ENOMEM */
1304                         em->start = cur_offset;
1305                         em->orig_start = em->start;
1306                         em->len = num_bytes;
1307                         em->block_len = num_bytes;
1308                         em->block_start = disk_bytenr;
1309                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1310                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1311                         while (1) {
1312                                 write_lock(&em_tree->lock);
1313                                 ret = add_extent_mapping(em_tree, em);
1314                                 write_unlock(&em_tree->lock);
1315                                 if (ret != -EEXIST) {
1316                                         free_extent_map(em);
1317                                         break;
1318                                 }
1319                                 btrfs_drop_extent_cache(inode, em->start,
1320                                                 em->start + em->len - 1, 0);
1321                         }
1322                         type = BTRFS_ORDERED_PREALLOC;
1323                 } else {
1324                         type = BTRFS_ORDERED_NOCOW;
1325                 }
1326
1327                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1328                                                num_bytes, num_bytes, type);
1329                 BUG_ON(ret); /* -ENOMEM */
1330
1331                 if (root->root_key.objectid ==
1332                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1333                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1334                                                       num_bytes);
1335                         if (ret) {
1336                                 btrfs_abort_transaction(trans, root, ret);
1337                                 goto error;
1338                         }
1339                 }
1340
1341                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1342                                 cur_offset, cur_offset + num_bytes - 1,
1343                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1344                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1345                                 EXTENT_SET_PRIVATE2);
1346                 cur_offset = extent_end;
1347                 if (cur_offset > end)
1348                         break;
1349         }
1350         btrfs_release_path(path);
1351
1352         if (cur_offset <= end && cow_start == (u64)-1) {
1353                 cow_start = cur_offset;
1354                 cur_offset = end;
1355         }
1356
1357         if (cow_start != (u64)-1) {
1358                 ret = cow_file_range(inode, locked_page, cow_start, end,
1359                                      page_started, nr_written, 1);
1360                 if (ret) {
1361                         btrfs_abort_transaction(trans, root, ret);
1362                         goto error;
1363                 }
1364         }
1365
1366 error:
1367         if (nolock) {
1368                 err = btrfs_end_transaction_nolock(trans, root);
1369         } else {
1370                 err = btrfs_end_transaction(trans, root);
1371         }
1372         if (!ret)
1373                 ret = err;
1374
1375         if (ret && cur_offset < end)
1376                 extent_clear_unlock_delalloc(inode,
1377                              &BTRFS_I(inode)->io_tree,
1378                              cur_offset, end, locked_page,
1379                              EXTENT_CLEAR_UNLOCK_PAGE |
1380                              EXTENT_CLEAR_UNLOCK |
1381                              EXTENT_CLEAR_DELALLOC |
1382                              EXTENT_CLEAR_DIRTY |
1383                              EXTENT_SET_WRITEBACK |
1384                              EXTENT_END_WRITEBACK);
1385
1386         btrfs_free_path(path);
1387         return ret;
1388 }
1389
1390 /*
1391  * extent_io.c call back to do delayed allocation processing
1392  */
1393 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1394                               u64 start, u64 end, int *page_started,
1395                               unsigned long *nr_written)
1396 {
1397         int ret;
1398         struct btrfs_root *root = BTRFS_I(inode)->root;
1399
1400         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1401                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1402                                          page_started, 1, nr_written);
1403         } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1404                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1405                                          page_started, 0, nr_written);
1406         } else if (!btrfs_test_opt(root, COMPRESS) &&
1407                    !(BTRFS_I(inode)->force_compress) &&
1408                    !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1409                 ret = cow_file_range(inode, locked_page, start, end,
1410                                       page_started, nr_written, 1);
1411         } else {
1412                 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1413                         &BTRFS_I(inode)->runtime_flags);
1414                 ret = cow_file_range_async(inode, locked_page, start, end,
1415                                            page_started, nr_written);
1416         }
1417         return ret;
1418 }
1419
1420 static void btrfs_split_extent_hook(struct inode *inode,
1421                                     struct extent_state *orig, u64 split)
1422 {
1423         /* not delalloc, ignore it */
1424         if (!(orig->state & EXTENT_DELALLOC))
1425                 return;
1426
1427         spin_lock(&BTRFS_I(inode)->lock);
1428         BTRFS_I(inode)->outstanding_extents++;
1429         spin_unlock(&BTRFS_I(inode)->lock);
1430 }
1431
1432 /*
1433  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1434  * extents so we can keep track of new extents that are just merged onto old
1435  * extents, such as when we are doing sequential writes, so we can properly
1436  * account for the metadata space we'll need.
1437  */
1438 static void btrfs_merge_extent_hook(struct inode *inode,
1439                                     struct extent_state *new,
1440                                     struct extent_state *other)
1441 {
1442         /* not delalloc, ignore it */
1443         if (!(other->state & EXTENT_DELALLOC))
1444                 return;
1445
1446         spin_lock(&BTRFS_I(inode)->lock);
1447         BTRFS_I(inode)->outstanding_extents--;
1448         spin_unlock(&BTRFS_I(inode)->lock);
1449 }
1450
1451 /*
1452  * extent_io.c set_bit_hook, used to track delayed allocation
1453  * bytes in this file, and to maintain the list of inodes that
1454  * have pending delalloc work to be done.
1455  */
1456 static void btrfs_set_bit_hook(struct inode *inode,
1457                                struct extent_state *state, int *bits)
1458 {
1459
1460         /*
1461          * set_bit and clear bit hooks normally require _irqsave/restore
1462          * but in this case, we are only testing for the DELALLOC
1463          * bit, which is only set or cleared with irqs on
1464          */
1465         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1466                 struct btrfs_root *root = BTRFS_I(inode)->root;
1467                 u64 len = state->end + 1 - state->start;
1468                 bool do_list = !btrfs_is_free_space_inode(inode);
1469
1470                 if (*bits & EXTENT_FIRST_DELALLOC) {
1471                         *bits &= ~EXTENT_FIRST_DELALLOC;
1472                 } else {
1473                         spin_lock(&BTRFS_I(inode)->lock);
1474                         BTRFS_I(inode)->outstanding_extents++;
1475                         spin_unlock(&BTRFS_I(inode)->lock);
1476                 }
1477
1478                 spin_lock(&root->fs_info->delalloc_lock);
1479                 BTRFS_I(inode)->delalloc_bytes += len;
1480                 root->fs_info->delalloc_bytes += len;
1481                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1482                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1483                                       &root->fs_info->delalloc_inodes);
1484                 }
1485                 spin_unlock(&root->fs_info->delalloc_lock);
1486         }
1487 }
1488
1489 /*
1490  * extent_io.c clear_bit_hook, see set_bit_hook for why
1491  */
1492 static void btrfs_clear_bit_hook(struct inode *inode,
1493                                  struct extent_state *state, int *bits)
1494 {
1495         /*
1496          * set_bit and clear bit hooks normally require _irqsave/restore
1497          * but in this case, we are only testing for the DELALLOC
1498          * bit, which is only set or cleared with irqs on
1499          */
1500         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1501                 struct btrfs_root *root = BTRFS_I(inode)->root;
1502                 u64 len = state->end + 1 - state->start;
1503                 bool do_list = !btrfs_is_free_space_inode(inode);
1504
1505                 if (*bits & EXTENT_FIRST_DELALLOC) {
1506                         *bits &= ~EXTENT_FIRST_DELALLOC;
1507                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1508                         spin_lock(&BTRFS_I(inode)->lock);
1509                         BTRFS_I(inode)->outstanding_extents--;
1510                         spin_unlock(&BTRFS_I(inode)->lock);
1511                 }
1512
1513                 if (*bits & EXTENT_DO_ACCOUNTING)
1514                         btrfs_delalloc_release_metadata(inode, len);
1515
1516                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1517                     && do_list)
1518                         btrfs_free_reserved_data_space(inode, len);
1519
1520                 spin_lock(&root->fs_info->delalloc_lock);
1521                 root->fs_info->delalloc_bytes -= len;
1522                 BTRFS_I(inode)->delalloc_bytes -= len;
1523
1524                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1525                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1526                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1527                 }
1528                 spin_unlock(&root->fs_info->delalloc_lock);
1529         }
1530 }
1531
1532 /*
1533  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1534  * we don't create bios that span stripes or chunks
1535  */
1536 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1537                          size_t size, struct bio *bio,
1538                          unsigned long bio_flags)
1539 {
1540         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1541         struct btrfs_mapping_tree *map_tree;
1542         u64 logical = (u64)bio->bi_sector << 9;
1543         u64 length = 0;
1544         u64 map_length;
1545         int ret;
1546
1547         if (bio_flags & EXTENT_BIO_COMPRESSED)
1548                 return 0;
1549
1550         length = bio->bi_size;
1551         map_tree = &root->fs_info->mapping_tree;
1552         map_length = length;
1553         ret = btrfs_map_block(map_tree, READ, logical,
1554                               &map_length, NULL, 0);
1555         /* Will always return 0 or 1 with map_multi == NULL */
1556         BUG_ON(ret < 0);
1557         if (map_length < length + size)
1558                 return 1;
1559         return 0;
1560 }
1561
1562 /*
1563  * in order to insert checksums into the metadata in large chunks,
1564  * we wait until bio submission time.   All the pages in the bio are
1565  * checksummed and sums are attached onto the ordered extent record.
1566  *
1567  * At IO completion time the cums attached on the ordered extent record
1568  * are inserted into the btree
1569  */
1570 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1571                                     struct bio *bio, int mirror_num,
1572                                     unsigned long bio_flags,
1573                                     u64 bio_offset)
1574 {
1575         struct btrfs_root *root = BTRFS_I(inode)->root;
1576         int ret = 0;
1577
1578         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1579         BUG_ON(ret); /* -ENOMEM */
1580         return 0;
1581 }
1582
1583 /*
1584  * in order to insert checksums into the metadata in large chunks,
1585  * we wait until bio submission time.   All the pages in the bio are
1586  * checksummed and sums are attached onto the ordered extent record.
1587  *
1588  * At IO completion time the cums attached on the ordered extent record
1589  * are inserted into the btree
1590  */
1591 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1592                           int mirror_num, unsigned long bio_flags,
1593                           u64 bio_offset)
1594 {
1595         struct btrfs_root *root = BTRFS_I(inode)->root;
1596         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1597 }
1598
1599 /*
1600  * extent_io.c submission hook. This does the right thing for csum calculation
1601  * on write, or reading the csums from the tree before a read
1602  */
1603 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1604                           int mirror_num, unsigned long bio_flags,
1605                           u64 bio_offset)
1606 {
1607         struct btrfs_root *root = BTRFS_I(inode)->root;
1608         int ret = 0;
1609         int skip_sum;
1610         int metadata = 0;
1611
1612         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1613
1614         if (btrfs_is_free_space_inode(inode))
1615                 metadata = 2;
1616
1617         if (!(rw & REQ_WRITE)) {
1618                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1619                 if (ret)
1620                         return ret;
1621
1622                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1623                         return btrfs_submit_compressed_read(inode, bio,
1624                                                     mirror_num, bio_flags);
1625                 } else if (!skip_sum) {
1626                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1627                         if (ret)
1628                                 return ret;
1629                 }
1630                 goto mapit;
1631         } else if (!skip_sum) {
1632                 /* csum items have already been cloned */
1633                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1634                         goto mapit;
1635                 /* we're doing a write, do the async checksumming */
1636                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1637                                    inode, rw, bio, mirror_num,
1638                                    bio_flags, bio_offset,
1639                                    __btrfs_submit_bio_start,
1640                                    __btrfs_submit_bio_done);
1641         }
1642
1643 mapit:
1644         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1645 }
1646
1647 /*
1648  * given a list of ordered sums record them in the inode.  This happens
1649  * at IO completion time based on sums calculated at bio submission time.
1650  */
1651 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1652                              struct inode *inode, u64 file_offset,
1653                              struct list_head *list)
1654 {
1655         struct btrfs_ordered_sum *sum;
1656
1657         list_for_each_entry(sum, list, list) {
1658                 btrfs_csum_file_blocks(trans,
1659                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1660         }
1661         return 0;
1662 }
1663
1664 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1665                               struct extent_state **cached_state)
1666 {
1667         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1668                 WARN_ON(1);
1669         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1670                                    cached_state, GFP_NOFS);
1671 }
1672
1673 /* see btrfs_writepage_start_hook for details on why this is required */
1674 struct btrfs_writepage_fixup {
1675         struct page *page;
1676         struct btrfs_work work;
1677 };
1678
1679 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1680 {
1681         struct btrfs_writepage_fixup *fixup;
1682         struct btrfs_ordered_extent *ordered;
1683         struct extent_state *cached_state = NULL;
1684         struct page *page;
1685         struct inode *inode;
1686         u64 page_start;
1687         u64 page_end;
1688         int ret;
1689
1690         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1691         page = fixup->page;
1692 again:
1693         lock_page(page);
1694         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1695                 ClearPageChecked(page);
1696                 goto out_page;
1697         }
1698
1699         inode = page->mapping->host;
1700         page_start = page_offset(page);
1701         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1702
1703         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1704                          &cached_state);
1705
1706         /* already ordered? We're done */
1707         if (PagePrivate2(page))
1708                 goto out;
1709
1710         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1711         if (ordered) {
1712                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1713                                      page_end, &cached_state, GFP_NOFS);
1714                 unlock_page(page);
1715                 btrfs_start_ordered_extent(inode, ordered, 1);
1716                 btrfs_put_ordered_extent(ordered);
1717                 goto again;
1718         }
1719
1720         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1721         if (ret) {
1722                 mapping_set_error(page->mapping, ret);
1723                 end_extent_writepage(page, ret, page_start, page_end);
1724                 ClearPageChecked(page);
1725                 goto out;
1726          }
1727
1728         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1729         ClearPageChecked(page);
1730         set_page_dirty(page);
1731 out:
1732         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1733                              &cached_state, GFP_NOFS);
1734 out_page:
1735         unlock_page(page);
1736         page_cache_release(page);
1737         kfree(fixup);
1738 }
1739
1740 /*
1741  * There are a few paths in the higher layers of the kernel that directly
1742  * set the page dirty bit without asking the filesystem if it is a
1743  * good idea.  This causes problems because we want to make sure COW
1744  * properly happens and the data=ordered rules are followed.
1745  *
1746  * In our case any range that doesn't have the ORDERED bit set
1747  * hasn't been properly setup for IO.  We kick off an async process
1748  * to fix it up.  The async helper will wait for ordered extents, set
1749  * the delalloc bit and make it safe to write the page.
1750  */
1751 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1752 {
1753         struct inode *inode = page->mapping->host;
1754         struct btrfs_writepage_fixup *fixup;
1755         struct btrfs_root *root = BTRFS_I(inode)->root;
1756
1757         /* this page is properly in the ordered list */
1758         if (TestClearPagePrivate2(page))
1759                 return 0;
1760
1761         if (PageChecked(page))
1762                 return -EAGAIN;
1763
1764         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1765         if (!fixup)
1766                 return -EAGAIN;
1767
1768         SetPageChecked(page);
1769         page_cache_get(page);
1770         fixup->work.func = btrfs_writepage_fixup_worker;
1771         fixup->page = page;
1772         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1773         return -EBUSY;
1774 }
1775
1776 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1777                                        struct inode *inode, u64 file_pos,
1778                                        u64 disk_bytenr, u64 disk_num_bytes,
1779                                        u64 num_bytes, u64 ram_bytes,
1780                                        u8 compression, u8 encryption,
1781                                        u16 other_encoding, int extent_type)
1782 {
1783         struct btrfs_root *root = BTRFS_I(inode)->root;
1784         struct btrfs_file_extent_item *fi;
1785         struct btrfs_path *path;
1786         struct extent_buffer *leaf;
1787         struct btrfs_key ins;
1788         u64 hint;
1789         int ret;
1790
1791         path = btrfs_alloc_path();
1792         if (!path)
1793                 return -ENOMEM;
1794
1795         path->leave_spinning = 1;
1796
1797         /*
1798          * we may be replacing one extent in the tree with another.
1799          * The new extent is pinned in the extent map, and we don't want
1800          * to drop it from the cache until it is completely in the btree.
1801          *
1802          * So, tell btrfs_drop_extents to leave this extent in the cache.
1803          * the caller is expected to unpin it and allow it to be merged
1804          * with the others.
1805          */
1806         ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1807                                  &hint, 0);
1808         if (ret)
1809                 goto out;
1810
1811         ins.objectid = btrfs_ino(inode);
1812         ins.offset = file_pos;
1813         ins.type = BTRFS_EXTENT_DATA_KEY;
1814         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1815         if (ret)
1816                 goto out;
1817         leaf = path->nodes[0];
1818         fi = btrfs_item_ptr(leaf, path->slots[0],
1819                             struct btrfs_file_extent_item);
1820         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1821         btrfs_set_file_extent_type(leaf, fi, extent_type);
1822         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1823         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1824         btrfs_set_file_extent_offset(leaf, fi, 0);
1825         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1826         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1827         btrfs_set_file_extent_compression(leaf, fi, compression);
1828         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1829         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1830
1831         btrfs_unlock_up_safe(path, 1);
1832         btrfs_set_lock_blocking(leaf);
1833
1834         btrfs_mark_buffer_dirty(leaf);
1835
1836         inode_add_bytes(inode, num_bytes);
1837
1838         ins.objectid = disk_bytenr;
1839         ins.offset = disk_num_bytes;
1840         ins.type = BTRFS_EXTENT_ITEM_KEY;
1841         ret = btrfs_alloc_reserved_file_extent(trans, root,
1842                                         root->root_key.objectid,
1843                                         btrfs_ino(inode), file_pos, &ins);
1844 out:
1845         btrfs_free_path(path);
1846
1847         return ret;
1848 }
1849
1850 /*
1851  * helper function for btrfs_finish_ordered_io, this
1852  * just reads in some of the csum leaves to prime them into ram
1853  * before we start the transaction.  It limits the amount of btree
1854  * reads required while inside the transaction.
1855  */
1856 /* as ordered data IO finishes, this gets called so we can finish
1857  * an ordered extent if the range of bytes in the file it covers are
1858  * fully written.
1859  */
1860 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
1861 {
1862         struct inode *inode = ordered_extent->inode;
1863         struct btrfs_root *root = BTRFS_I(inode)->root;
1864         struct btrfs_trans_handle *trans = NULL;
1865         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1866         struct extent_state *cached_state = NULL;
1867         int compress_type = 0;
1868         int ret;
1869         bool nolock;
1870
1871         nolock = btrfs_is_free_space_inode(inode);
1872
1873         if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
1874                 ret = -EIO;
1875                 goto out;
1876         }
1877
1878         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1879                 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
1880                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1881                 if (!ret) {
1882                         if (nolock)
1883                                 trans = btrfs_join_transaction_nolock(root);
1884                         else
1885                                 trans = btrfs_join_transaction(root);
1886                         if (IS_ERR(trans)) {
1887                                 ret = PTR_ERR(trans);
1888                                 trans = NULL;
1889                                 goto out;
1890                         }
1891                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1892                         ret = btrfs_update_inode_fallback(trans, root, inode);
1893                         if (ret) /* -ENOMEM or corruption */
1894                                 btrfs_abort_transaction(trans, root, ret);
1895                 }
1896                 goto out;
1897         }
1898
1899         lock_extent_bits(io_tree, ordered_extent->file_offset,
1900                          ordered_extent->file_offset + ordered_extent->len - 1,
1901                          0, &cached_state);
1902
1903         if (nolock)
1904                 trans = btrfs_join_transaction_nolock(root);
1905         else
1906                 trans = btrfs_join_transaction(root);
1907         if (IS_ERR(trans)) {
1908                 ret = PTR_ERR(trans);
1909                 trans = NULL;
1910                 goto out_unlock;
1911         }
1912         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1913
1914         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1915                 compress_type = ordered_extent->compress_type;
1916         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1917                 BUG_ON(compress_type);
1918                 ret = btrfs_mark_extent_written(trans, inode,
1919                                                 ordered_extent->file_offset,
1920                                                 ordered_extent->file_offset +
1921                                                 ordered_extent->len);
1922         } else {
1923                 BUG_ON(root == root->fs_info->tree_root);
1924                 ret = insert_reserved_file_extent(trans, inode,
1925                                                 ordered_extent->file_offset,
1926                                                 ordered_extent->start,
1927                                                 ordered_extent->disk_len,
1928                                                 ordered_extent->len,
1929                                                 ordered_extent->len,
1930                                                 compress_type, 0, 0,
1931                                                 BTRFS_FILE_EXTENT_REG);
1932                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1933                                    ordered_extent->file_offset,
1934                                    ordered_extent->len);
1935         }
1936
1937         if (ret < 0) {
1938                 btrfs_abort_transaction(trans, root, ret);
1939                 goto out_unlock;
1940         }
1941
1942         add_pending_csums(trans, inode, ordered_extent->file_offset,
1943                           &ordered_extent->list);
1944
1945         ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1946         if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1947                 ret = btrfs_update_inode_fallback(trans, root, inode);
1948                 if (ret) { /* -ENOMEM or corruption */
1949                         btrfs_abort_transaction(trans, root, ret);
1950                         goto out_unlock;
1951                 }
1952         }
1953         ret = 0;
1954 out_unlock:
1955         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1956                              ordered_extent->file_offset +
1957                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1958 out:
1959         if (root != root->fs_info->tree_root)
1960                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1961         if (trans) {
1962                 if (nolock)
1963                         btrfs_end_transaction_nolock(trans, root);
1964                 else
1965                         btrfs_end_transaction(trans, root);
1966         }
1967
1968         if (ret)
1969                 clear_extent_uptodate(io_tree, ordered_extent->file_offset,
1970                                       ordered_extent->file_offset +
1971                                       ordered_extent->len - 1, NULL, GFP_NOFS);
1972
1973         /*
1974          * This needs to be dont to make sure anybody waiting knows we are done
1975          * upating everything for this ordered extent.
1976          */
1977         btrfs_remove_ordered_extent(inode, ordered_extent);
1978
1979         /* once for us */
1980         btrfs_put_ordered_extent(ordered_extent);
1981         /* once for the tree */
1982         btrfs_put_ordered_extent(ordered_extent);
1983
1984         return ret;
1985 }
1986
1987 static void finish_ordered_fn(struct btrfs_work *work)
1988 {
1989         struct btrfs_ordered_extent *ordered_extent;
1990         ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
1991         btrfs_finish_ordered_io(ordered_extent);
1992 }
1993
1994 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1995                                 struct extent_state *state, int uptodate)
1996 {
1997         struct inode *inode = page->mapping->host;
1998         struct btrfs_root *root = BTRFS_I(inode)->root;
1999         struct btrfs_ordered_extent *ordered_extent = NULL;
2000         struct btrfs_workers *workers;
2001
2002         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2003
2004         ClearPagePrivate2(page);
2005         if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2006                                             end - start + 1, uptodate))
2007                 return 0;
2008
2009         ordered_extent->work.func = finish_ordered_fn;
2010         ordered_extent->work.flags = 0;
2011
2012         if (btrfs_is_free_space_inode(inode))
2013                 workers = &root->fs_info->endio_freespace_worker;
2014         else
2015                 workers = &root->fs_info->endio_write_workers;
2016         btrfs_queue_worker(workers, &ordered_extent->work);
2017
2018         return 0;
2019 }
2020
2021 /*
2022  * when reads are done, we need to check csums to verify the data is correct
2023  * if there's a match, we allow the bio to finish.  If not, the code in
2024  * extent_io.c will try to find good copies for us.
2025  */
2026 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
2027                                struct extent_state *state, int mirror)
2028 {
2029         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
2030         struct inode *inode = page->mapping->host;
2031         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2032         char *kaddr;
2033         u64 private = ~(u32)0;
2034         int ret;
2035         struct btrfs_root *root = BTRFS_I(inode)->root;
2036         u32 csum = ~(u32)0;
2037
2038         if (PageChecked(page)) {
2039                 ClearPageChecked(page);
2040                 goto good;
2041         }
2042
2043         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2044                 goto good;
2045
2046         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2047             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2048                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2049                                   GFP_NOFS);
2050                 return 0;
2051         }
2052
2053         if (state && state->start == start) {
2054                 private = state->private;
2055                 ret = 0;
2056         } else {
2057                 ret = get_state_private(io_tree, start, &private);
2058         }
2059         kaddr = kmap_atomic(page);
2060         if (ret)
2061                 goto zeroit;
2062
2063         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
2064         btrfs_csum_final(csum, (char *)&csum);
2065         if (csum != private)
2066                 goto zeroit;
2067
2068         kunmap_atomic(kaddr);
2069 good:
2070         return 0;
2071
2072 zeroit:
2073         printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
2074                        "private %llu\n",
2075                        (unsigned long long)btrfs_ino(page->mapping->host),
2076                        (unsigned long long)start, csum,
2077                        (unsigned long long)private);
2078         memset(kaddr + offset, 1, end - start + 1);
2079         flush_dcache_page(page);
2080         kunmap_atomic(kaddr);
2081         if (private == 0)
2082                 return 0;
2083         return -EIO;
2084 }
2085
2086 struct delayed_iput {
2087         struct list_head list;
2088         struct inode *inode;
2089 };
2090
2091 /* JDM: If this is fs-wide, why can't we add a pointer to
2092  * btrfs_inode instead and avoid the allocation? */
2093 void btrfs_add_delayed_iput(struct inode *inode)
2094 {
2095         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2096         struct delayed_iput *delayed;
2097
2098         if (atomic_add_unless(&inode->i_count, -1, 1))
2099                 return;
2100
2101         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2102         delayed->inode = inode;
2103
2104         spin_lock(&fs_info->delayed_iput_lock);
2105         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2106         spin_unlock(&fs_info->delayed_iput_lock);
2107 }
2108
2109 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2110 {
2111         LIST_HEAD(list);
2112         struct btrfs_fs_info *fs_info = root->fs_info;
2113         struct delayed_iput *delayed;
2114         int empty;
2115
2116         spin_lock(&fs_info->delayed_iput_lock);
2117         empty = list_empty(&fs_info->delayed_iputs);
2118         spin_unlock(&fs_info->delayed_iput_lock);
2119         if (empty)
2120                 return;
2121
2122         down_read(&root->fs_info->cleanup_work_sem);
2123         spin_lock(&fs_info->delayed_iput_lock);
2124         list_splice_init(&fs_info->delayed_iputs, &list);
2125         spin_unlock(&fs_info->delayed_iput_lock);
2126
2127         while (!list_empty(&list)) {
2128                 delayed = list_entry(list.next, struct delayed_iput, list);
2129                 list_del(&delayed->list);
2130                 iput(delayed->inode);
2131                 kfree(delayed);
2132         }
2133         up_read(&root->fs_info->cleanup_work_sem);
2134 }
2135
2136 enum btrfs_orphan_cleanup_state {
2137         ORPHAN_CLEANUP_STARTED  = 1,
2138         ORPHAN_CLEANUP_DONE     = 2,
2139 };
2140
2141 /*
2142  * This is called in transaction commit time. If there are no orphan
2143  * files in the subvolume, it removes orphan item and frees block_rsv
2144  * structure.
2145  */
2146 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2147                               struct btrfs_root *root)
2148 {
2149         struct btrfs_block_rsv *block_rsv;
2150         int ret;
2151
2152         if (atomic_read(&root->orphan_inodes) ||
2153             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2154                 return;
2155
2156         spin_lock(&root->orphan_lock);
2157         if (atomic_read(&root->orphan_inodes)) {
2158                 spin_unlock(&root->orphan_lock);
2159                 return;
2160         }
2161
2162         if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2163                 spin_unlock(&root->orphan_lock);
2164                 return;
2165         }
2166
2167         block_rsv = root->orphan_block_rsv;
2168         root->orphan_block_rsv = NULL;
2169         spin_unlock(&root->orphan_lock);
2170
2171         if (root->orphan_item_inserted &&
2172             btrfs_root_refs(&root->root_item) > 0) {
2173                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2174                                             root->root_key.objectid);
2175                 BUG_ON(ret);
2176                 root->orphan_item_inserted = 0;
2177         }
2178
2179         if (block_rsv) {
2180                 WARN_ON(block_rsv->size > 0);
2181                 btrfs_free_block_rsv(root, block_rsv);
2182         }
2183 }
2184
2185 /*
2186  * This creates an orphan entry for the given inode in case something goes
2187  * wrong in the middle of an unlink/truncate.
2188  *
2189  * NOTE: caller of this function should reserve 5 units of metadata for
2190  *       this function.
2191  */
2192 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2193 {
2194         struct btrfs_root *root = BTRFS_I(inode)->root;
2195         struct btrfs_block_rsv *block_rsv = NULL;
2196         int reserve = 0;
2197         int insert = 0;
2198         int ret;
2199
2200         if (!root->orphan_block_rsv) {
2201                 block_rsv = btrfs_alloc_block_rsv(root);
2202                 if (!block_rsv)
2203                         return -ENOMEM;
2204         }
2205
2206         spin_lock(&root->orphan_lock);
2207         if (!root->orphan_block_rsv) {
2208                 root->orphan_block_rsv = block_rsv;
2209         } else if (block_rsv) {
2210                 btrfs_free_block_rsv(root, block_rsv);
2211                 block_rsv = NULL;
2212         }
2213
2214         if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2215                               &BTRFS_I(inode)->runtime_flags)) {
2216 #if 0
2217                 /*
2218                  * For proper ENOSPC handling, we should do orphan
2219                  * cleanup when mounting. But this introduces backward
2220                  * compatibility issue.
2221                  */
2222                 if (!xchg(&root->orphan_item_inserted, 1))
2223                         insert = 2;
2224                 else
2225                         insert = 1;
2226 #endif
2227                 insert = 1;
2228                 atomic_dec(&root->orphan_inodes);
2229         }
2230
2231         if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2232                               &BTRFS_I(inode)->runtime_flags))
2233                 reserve = 1;
2234         spin_unlock(&root->orphan_lock);
2235
2236         /* grab metadata reservation from transaction handle */
2237         if (reserve) {
2238                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2239                 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
2240         }
2241
2242         /* insert an orphan item to track this unlinked/truncated file */
2243         if (insert >= 1) {
2244                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2245                 if (ret && ret != -EEXIST) {
2246                         clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2247                                   &BTRFS_I(inode)->runtime_flags);
2248                         btrfs_abort_transaction(trans, root, ret);
2249                         return ret;
2250                 }
2251                 ret = 0;
2252         }
2253
2254         /* insert an orphan item to track subvolume contains orphan files */
2255         if (insert >= 2) {
2256                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2257                                                root->root_key.objectid);
2258                 if (ret && ret != -EEXIST) {
2259                         btrfs_abort_transaction(trans, root, ret);
2260                         return ret;
2261                 }
2262         }
2263         return 0;
2264 }
2265
2266 /*
2267  * We have done the truncate/delete so we can go ahead and remove the orphan
2268  * item for this particular inode.
2269  */
2270 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2271 {
2272         struct btrfs_root *root = BTRFS_I(inode)->root;
2273         int delete_item = 0;
2274         int release_rsv = 0;
2275         int ret = 0;
2276
2277         spin_lock(&root->orphan_lock);
2278         if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2279                                &BTRFS_I(inode)->runtime_flags))
2280                 delete_item = 1;
2281
2282         if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2283                                &BTRFS_I(inode)->runtime_flags))
2284                 release_rsv = 1;
2285         spin_unlock(&root->orphan_lock);
2286
2287         if (trans && delete_item) {
2288                 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2289                 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2290         }
2291
2292         if (release_rsv) {
2293                 btrfs_orphan_release_metadata(inode);
2294                 atomic_dec(&root->orphan_inodes);
2295         }
2296
2297         return 0;
2298 }
2299
2300 /*
2301  * this cleans up any orphans that may be left on the list from the last use
2302  * of this root.
2303  */
2304 int btrfs_orphan_cleanup(struct btrfs_root *root)
2305 {
2306         struct btrfs_path *path;
2307         struct extent_buffer *leaf;
2308         struct btrfs_key key, found_key;
2309         struct btrfs_trans_handle *trans;
2310         struct inode *inode;
2311         u64 last_objectid = 0;
2312         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2313
2314         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2315                 return 0;
2316
2317         path = btrfs_alloc_path();
2318         if (!path) {
2319                 ret = -ENOMEM;
2320                 goto out;
2321         }
2322         path->reada = -1;
2323
2324         key.objectid = BTRFS_ORPHAN_OBJECTID;
2325         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2326         key.offset = (u64)-1;
2327
2328         while (1) {
2329                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2330                 if (ret < 0)
2331                         goto out;
2332
2333                 /*
2334                  * if ret == 0 means we found what we were searching for, which
2335                  * is weird, but possible, so only screw with path if we didn't
2336                  * find the key and see if we have stuff that matches
2337                  */
2338                 if (ret > 0) {
2339                         ret = 0;
2340                         if (path->slots[0] == 0)
2341                                 break;
2342                         path->slots[0]--;
2343                 }
2344
2345                 /* pull out the item */
2346                 leaf = path->nodes[0];
2347                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2348
2349                 /* make sure the item matches what we want */
2350                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2351                         break;
2352                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2353                         break;
2354
2355                 /* release the path since we're done with it */
2356                 btrfs_release_path(path);
2357
2358                 /*
2359                  * this is where we are basically btrfs_lookup, without the
2360                  * crossing root thing.  we store the inode number in the
2361                  * offset of the orphan item.
2362                  */
2363
2364                 if (found_key.offset == last_objectid) {
2365                         printk(KERN_ERR "btrfs: Error removing orphan entry, "
2366                                "stopping orphan cleanup\n");
2367                         ret = -EINVAL;
2368                         goto out;
2369                 }
2370
2371                 last_objectid = found_key.offset;
2372
2373                 found_key.objectid = found_key.offset;
2374                 found_key.type = BTRFS_INODE_ITEM_KEY;
2375                 found_key.offset = 0;
2376                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2377                 ret = PTR_RET(inode);
2378                 if (ret && ret != -ESTALE)
2379                         goto out;
2380
2381                 if (ret == -ESTALE && root == root->fs_info->tree_root) {
2382                         struct btrfs_root *dead_root;
2383                         struct btrfs_fs_info *fs_info = root->fs_info;
2384                         int is_dead_root = 0;
2385
2386                         /*
2387                          * this is an orphan in the tree root. Currently these
2388                          * could come from 2 sources:
2389                          *  a) a snapshot deletion in progress
2390                          *  b) a free space cache inode
2391                          * We need to distinguish those two, as the snapshot
2392                          * orphan must not get deleted.
2393                          * find_dead_roots already ran before us, so if this
2394                          * is a snapshot deletion, we should find the root
2395                          * in the dead_roots list
2396                          */
2397                         spin_lock(&fs_info->trans_lock);
2398                         list_for_each_entry(dead_root, &fs_info->dead_roots,
2399                                             root_list) {
2400                                 if (dead_root->root_key.objectid ==
2401                                     found_key.objectid) {
2402                                         is_dead_root = 1;
2403                                         break;
2404                                 }
2405                         }
2406                         spin_unlock(&fs_info->trans_lock);
2407                         if (is_dead_root) {
2408                                 /* prevent this orphan from being found again */
2409                                 key.offset = found_key.objectid - 1;
2410                                 continue;
2411                         }
2412                 }
2413                 /*
2414                  * Inode is already gone but the orphan item is still there,
2415                  * kill the orphan item.
2416                  */
2417                 if (ret == -ESTALE) {
2418                         trans = btrfs_start_transaction(root, 1);
2419                         if (IS_ERR(trans)) {
2420                                 ret = PTR_ERR(trans);
2421                                 goto out;
2422                         }
2423                         printk(KERN_ERR "auto deleting %Lu\n",
2424                                found_key.objectid);
2425                         ret = btrfs_del_orphan_item(trans, root,
2426                                                     found_key.objectid);
2427                         BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2428                         btrfs_end_transaction(trans, root);
2429                         continue;
2430                 }
2431
2432                 /*
2433                  * add this inode to the orphan list so btrfs_orphan_del does
2434                  * the proper thing when we hit it
2435                  */
2436                 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2437                         &BTRFS_I(inode)->runtime_flags);
2438
2439                 /* if we have links, this was a truncate, lets do that */
2440                 if (inode->i_nlink) {
2441                         if (!S_ISREG(inode->i_mode)) {
2442                                 WARN_ON(1);
2443                                 iput(inode);
2444                                 continue;
2445                         }
2446                         nr_truncate++;
2447                         ret = btrfs_truncate(inode);
2448                 } else {
2449                         nr_unlink++;
2450                 }
2451
2452                 /* this will do delete_inode and everything for us */
2453                 iput(inode);
2454                 if (ret)
2455                         goto out;
2456         }
2457         /* release the path since we're done with it */
2458         btrfs_release_path(path);
2459
2460         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2461
2462         if (root->orphan_block_rsv)
2463                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2464                                         (u64)-1);
2465
2466         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2467                 trans = btrfs_join_transaction(root);
2468                 if (!IS_ERR(trans))
2469                         btrfs_end_transaction(trans, root);
2470         }
2471
2472         if (nr_unlink)
2473                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2474         if (nr_truncate)
2475                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2476
2477 out:
2478         if (ret)
2479                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2480         btrfs_free_path(path);
2481         return ret;
2482 }
2483
2484 /*
2485  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2486  * don't find any xattrs, we know there can't be any acls.
2487  *
2488  * slot is the slot the inode is in, objectid is the objectid of the inode
2489  */
2490 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2491                                           int slot, u64 objectid)
2492 {
2493         u32 nritems = btrfs_header_nritems(leaf);
2494         struct btrfs_key found_key;
2495         int scanned = 0;
2496
2497         slot++;
2498         while (slot < nritems) {
2499                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2500
2501                 /* we found a different objectid, there must not be acls */
2502                 if (found_key.objectid != objectid)
2503                         return 0;
2504
2505                 /* we found an xattr, assume we've got an acl */
2506                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2507                         return 1;
2508
2509                 /*
2510                  * we found a key greater than an xattr key, there can't
2511                  * be any acls later on
2512                  */
2513                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2514                         return 0;
2515
2516                 slot++;
2517                 scanned++;
2518
2519                 /*
2520                  * it goes inode, inode backrefs, xattrs, extents,
2521                  * so if there are a ton of hard links to an inode there can
2522                  * be a lot of backrefs.  Don't waste time searching too hard,
2523                  * this is just an optimization
2524                  */
2525                 if (scanned >= 8)
2526                         break;
2527         }
2528         /* we hit the end of the leaf before we found an xattr or
2529          * something larger than an xattr.  We have to assume the inode
2530          * has acls
2531          */
2532         return 1;
2533 }
2534
2535 /*
2536  * read an inode from the btree into the in-memory inode
2537  */
2538 static void btrfs_read_locked_inode(struct inode *inode)
2539 {
2540         struct btrfs_path *path;
2541         struct extent_buffer *leaf;
2542         struct btrfs_inode_item *inode_item;
2543         struct btrfs_timespec *tspec;
2544         struct btrfs_root *root = BTRFS_I(inode)->root;
2545         struct btrfs_key location;
2546         int maybe_acls;
2547         u32 rdev;
2548         int ret;
2549         bool filled = false;
2550
2551         ret = btrfs_fill_inode(inode, &rdev);
2552         if (!ret)
2553                 filled = true;
2554
2555         path = btrfs_alloc_path();
2556         if (!path)
2557                 goto make_bad;
2558
2559         path->leave_spinning = 1;
2560         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2561
2562         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2563         if (ret)
2564                 goto make_bad;
2565
2566         leaf = path->nodes[0];
2567
2568         if (filled)
2569                 goto cache_acl;
2570
2571         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2572                                     struct btrfs_inode_item);
2573         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2574         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2575         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2576         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2577         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2578
2579         tspec = btrfs_inode_atime(inode_item);
2580         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2581         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2582
2583         tspec = btrfs_inode_mtime(inode_item);
2584         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2585         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2586
2587         tspec = btrfs_inode_ctime(inode_item);
2588         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2589         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2590
2591         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2592         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2593         inode->i_version = btrfs_inode_sequence(leaf, inode_item);
2594         inode->i_generation = BTRFS_I(inode)->generation;
2595         inode->i_rdev = 0;
2596         rdev = btrfs_inode_rdev(leaf, inode_item);
2597
2598         BTRFS_I(inode)->index_cnt = (u64)-1;
2599         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2600 cache_acl:
2601         /*
2602          * try to precache a NULL acl entry for files that don't have
2603          * any xattrs or acls
2604          */
2605         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2606                                            btrfs_ino(inode));
2607         if (!maybe_acls)
2608                 cache_no_acl(inode);
2609
2610         btrfs_free_path(path);
2611
2612         switch (inode->i_mode & S_IFMT) {
2613         case S_IFREG:
2614                 inode->i_mapping->a_ops = &btrfs_aops;
2615                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2616                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2617                 inode->i_fop = &btrfs_file_operations;
2618                 inode->i_op = &btrfs_file_inode_operations;
2619                 break;
2620         case S_IFDIR:
2621                 inode->i_fop = &btrfs_dir_file_operations;
2622                 if (root == root->fs_info->tree_root)
2623                         inode->i_op = &btrfs_dir_ro_inode_operations;
2624                 else
2625                         inode->i_op = &btrfs_dir_inode_operations;
2626                 break;
2627         case S_IFLNK:
2628                 inode->i_op = &btrfs_symlink_inode_operations;
2629                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2630                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2631                 break;
2632         default:
2633                 inode->i_op = &btrfs_special_inode_operations;
2634                 init_special_inode(inode, inode->i_mode, rdev);
2635                 break;
2636         }
2637
2638         btrfs_update_iflags(inode);
2639         return;
2640
2641 make_bad:
2642         btrfs_free_path(path);
2643         make_bad_inode(inode);
2644 }
2645
2646 /*
2647  * given a leaf and an inode, copy the inode fields into the leaf
2648  */
2649 static void fill_inode_item(struct btrfs_trans_handle *trans,
2650                             struct extent_buffer *leaf,
2651                             struct btrfs_inode_item *item,
2652                             struct inode *inode)
2653 {
2654         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2655         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2656         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2657         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2658         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2659
2660         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2661                                inode->i_atime.tv_sec);
2662         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2663                                 inode->i_atime.tv_nsec);
2664
2665         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2666                                inode->i_mtime.tv_sec);
2667         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2668                                 inode->i_mtime.tv_nsec);
2669
2670         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2671                                inode->i_ctime.tv_sec);
2672         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2673                                 inode->i_ctime.tv_nsec);
2674
2675         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2676         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2677         btrfs_set_inode_sequence(leaf, item, inode->i_version);
2678         btrfs_set_inode_transid(leaf, item, trans->transid);
2679         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2680         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2681         btrfs_set_inode_block_group(leaf, item, 0);
2682 }
2683
2684 /*
2685  * copy everything in the in-memory inode into the btree.
2686  */
2687 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
2688                                 struct btrfs_root *root, struct inode *inode)
2689 {
2690         struct btrfs_inode_item *inode_item;
2691         struct btrfs_path *path;
2692         struct extent_buffer *leaf;
2693         int ret;
2694
2695         path = btrfs_alloc_path();
2696         if (!path)
2697                 return -ENOMEM;
2698
2699         path->leave_spinning = 1;
2700         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2701                                  1);
2702         if (ret) {
2703                 if (ret > 0)
2704                         ret = -ENOENT;
2705                 goto failed;
2706         }
2707
2708         btrfs_unlock_up_safe(path, 1);
2709         leaf = path->nodes[0];
2710         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2711                                     struct btrfs_inode_item);
2712
2713         fill_inode_item(trans, leaf, inode_item, inode);
2714         btrfs_mark_buffer_dirty(leaf);
2715         btrfs_set_inode_last_trans(trans, inode);
2716         ret = 0;
2717 failed:
2718         btrfs_free_path(path);
2719         return ret;
2720 }
2721
2722 /*
2723  * copy everything in the in-memory inode into the btree.
2724  */
2725 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2726                                 struct btrfs_root *root, struct inode *inode)
2727 {
2728         int ret;
2729
2730         /*
2731          * If the inode is a free space inode, we can deadlock during commit
2732          * if we put it into the delayed code.
2733          *
2734          * The data relocation inode should also be directly updated
2735          * without delay
2736          */
2737         if (!btrfs_is_free_space_inode(inode)
2738             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2739                 btrfs_update_root_times(trans, root);
2740
2741                 ret = btrfs_delayed_update_inode(trans, root, inode);
2742                 if (!ret)
2743                         btrfs_set_inode_last_trans(trans, inode);
2744                 return ret;
2745         }
2746
2747         return btrfs_update_inode_item(trans, root, inode);
2748 }
2749
2750 static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2751                                 struct btrfs_root *root, struct inode *inode)
2752 {
2753         int ret;
2754
2755         ret = btrfs_update_inode(trans, root, inode);
2756         if (ret == -ENOSPC)
2757                 return btrfs_update_inode_item(trans, root, inode);
2758         return ret;
2759 }
2760
2761 /*
2762  * unlink helper that gets used here in inode.c and in the tree logging
2763  * recovery code.  It remove a link in a directory with a given name, and
2764  * also drops the back refs in the inode to the directory
2765  */
2766 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2767                                 struct btrfs_root *root,
2768                                 struct inode *dir, struct inode *inode,
2769                                 const char *name, int name_len)
2770 {
2771         struct btrfs_path *path;
2772         int ret = 0;
2773         struct extent_buffer *leaf;
2774         struct btrfs_dir_item *di;
2775         struct btrfs_key key;
2776         u64 index;
2777         u64 ino = btrfs_ino(inode);
2778         u64 dir_ino = btrfs_ino(dir);
2779
2780         path = btrfs_alloc_path();
2781         if (!path) {
2782                 ret = -ENOMEM;
2783                 goto out;
2784         }
2785
2786         path->leave_spinning = 1;
2787         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2788                                     name, name_len, -1);
2789         if (IS_ERR(di)) {
2790                 ret = PTR_ERR(di);
2791                 goto err;
2792         }
2793         if (!di) {
2794                 ret = -ENOENT;
2795                 goto err;
2796         }
2797         leaf = path->nodes[0];
2798         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2799         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2800         if (ret)
2801                 goto err;
2802         btrfs_release_path(path);
2803
2804         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2805                                   dir_ino, &index);
2806         if (ret) {
2807                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2808                        "inode %llu parent %llu\n", name_len, name,
2809                        (unsigned long long)ino, (unsigned long long)dir_ino);
2810                 btrfs_abort_transaction(trans, root, ret);
2811                 goto err;
2812         }
2813
2814         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2815         if (ret) {
2816                 btrfs_abort_transaction(trans, root, ret);
2817                 goto err;
2818         }
2819
2820         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2821                                          inode, dir_ino);
2822         if (ret != 0 && ret != -ENOENT) {
2823                 btrfs_abort_transaction(trans, root, ret);
2824                 goto err;
2825         }
2826
2827         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2828                                            dir, index);
2829         if (ret == -ENOENT)
2830                 ret = 0;
2831 err:
2832         btrfs_free_path(path);
2833         if (ret)
2834                 goto out;
2835
2836         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2837         inode_inc_iversion(inode);
2838         inode_inc_iversion(dir);
2839         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2840         ret = btrfs_update_inode(trans, root, dir);
2841 out:
2842         return ret;
2843 }
2844
2845 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2846                        struct btrfs_root *root,
2847                        struct inode *dir, struct inode *inode,
2848                        const char *name, int name_len)
2849 {
2850         int ret;
2851         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2852         if (!ret) {
2853                 btrfs_drop_nlink(inode);
2854                 ret = btrfs_update_inode(trans, root, inode);
2855         }
2856         return ret;
2857 }
2858
2859
2860 /* helper to check if there is any shared block in the path */
2861 static int check_path_shared(struct btrfs_root *root,
2862                              struct btrfs_path *path)
2863 {
2864         struct extent_buffer *eb;
2865         int level;
2866         u64 refs = 1;
2867
2868         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2869                 int ret;
2870
2871                 if (!path->nodes[level])
2872                         break;
2873                 eb = path->nodes[level];
2874                 if (!btrfs_block_can_be_shared(root, eb))
2875                         continue;
2876                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2877                                                &refs, NULL);
2878                 if (refs > 1)
2879                         return 1;
2880         }
2881         return 0;
2882 }
2883
2884 /*
2885  * helper to start transaction for unlink and rmdir.
2886  *
2887  * unlink and rmdir are special in btrfs, they do not always free space.
2888  * so in enospc case, we should make sure they will free space before
2889  * allowing them to use the global metadata reservation.
2890  */
2891 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2892                                                        struct dentry *dentry)
2893 {
2894         struct btrfs_trans_handle *trans;
2895         struct btrfs_root *root = BTRFS_I(dir)->root;
2896         struct btrfs_path *path;
2897         struct btrfs_inode_ref *ref;
2898         struct btrfs_dir_item *di;
2899         struct inode *inode = dentry->d_inode;
2900         u64 index;
2901         int check_link = 1;
2902         int err = -ENOSPC;
2903         int ret;
2904         u64 ino = btrfs_ino(inode);
2905         u64 dir_ino = btrfs_ino(dir);
2906
2907         /*
2908          * 1 for the possible orphan item
2909          * 1 for the dir item
2910          * 1 for the dir index
2911          * 1 for the inode ref
2912          * 1 for the inode ref in the tree log
2913          * 2 for the dir entries in the log
2914          * 1 for the inode
2915          */
2916         trans = btrfs_start_transaction(root, 8);
2917         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2918                 return trans;
2919
2920         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2921                 return ERR_PTR(-ENOSPC);
2922
2923         /* check if there is someone else holds reference */
2924         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2925                 return ERR_PTR(-ENOSPC);
2926
2927         if (atomic_read(&inode->i_count) > 2)
2928                 return ERR_PTR(-ENOSPC);
2929
2930         if (xchg(&root->fs_info->enospc_unlink, 1))
2931                 return ERR_PTR(-ENOSPC);
2932
2933         path = btrfs_alloc_path();
2934         if (!path) {
2935                 root->fs_info->enospc_unlink = 0;
2936                 return ERR_PTR(-ENOMEM);
2937         }
2938
2939         /* 1 for the orphan item */
2940         trans = btrfs_start_transaction(root, 1);
2941         if (IS_ERR(trans)) {
2942                 btrfs_free_path(path);
2943                 root->fs_info->enospc_unlink = 0;
2944                 return trans;
2945         }
2946
2947         path->skip_locking = 1;
2948         path->search_commit_root = 1;
2949
2950         ret = btrfs_lookup_inode(trans, root, path,
2951                                 &BTRFS_I(dir)->location, 0);
2952         if (ret < 0) {
2953                 err = ret;
2954                 goto out;
2955         }
2956         if (ret == 0) {
2957                 if (check_path_shared(root, path))
2958                         goto out;
2959         } else {
2960                 check_link = 0;
2961         }
2962         btrfs_release_path(path);
2963
2964         ret = btrfs_lookup_inode(trans, root, path,
2965                                 &BTRFS_I(inode)->location, 0);
2966         if (ret < 0) {
2967                 err = ret;
2968                 goto out;
2969         }
2970         if (ret == 0) {
2971                 if (check_path_shared(root, path))
2972                         goto out;
2973         } else {
2974                 check_link = 0;
2975         }
2976         btrfs_release_path(path);
2977
2978         if (ret == 0 && S_ISREG(inode->i_mode)) {
2979                 ret = btrfs_lookup_file_extent(trans, root, path,
2980                                                ino, (u64)-1, 0);
2981                 if (ret < 0) {
2982                         err = ret;
2983                         goto out;
2984                 }
2985                 BUG_ON(ret == 0); /* Corruption */
2986                 if (check_path_shared(root, path))
2987                         goto out;
2988                 btrfs_release_path(path);
2989         }
2990
2991         if (!check_link) {
2992                 err = 0;
2993                 goto out;
2994         }
2995
2996         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2997                                 dentry->d_name.name, dentry->d_name.len, 0);
2998         if (IS_ERR(di)) {
2999                 err = PTR_ERR(di);
3000                 goto out;
3001         }
3002         if (di) {
3003                 if (check_path_shared(root, path))
3004                         goto out;
3005         } else {
3006                 err = 0;
3007                 goto out;
3008         }
3009         btrfs_release_path(path);
3010
3011         ref = btrfs_lookup_inode_ref(trans, root, path,
3012                                 dentry->d_name.name, dentry->d_name.len,
3013                                 ino, dir_ino, 0);
3014         if (IS_ERR(ref)) {
3015                 err = PTR_ERR(ref);
3016                 goto out;
3017         }
3018         BUG_ON(!ref); /* Logic error */
3019         if (check_path_shared(root, path))
3020                 goto out;
3021         index = btrfs_inode_ref_index(path->nodes[0], ref);
3022         btrfs_release_path(path);
3023
3024         /*
3025          * This is a commit root search, if we can lookup inode item and other
3026          * relative items in the commit root, it means the transaction of
3027          * dir/file creation has been committed, and the dir index item that we
3028          * delay to insert has also been inserted into the commit root. So
3029          * we needn't worry about the delayed insertion of the dir index item
3030          * here.
3031          */
3032         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
3033                                 dentry->d_name.name, dentry->d_name.len, 0);
3034         if (IS_ERR(di)) {
3035                 err = PTR_ERR(di);
3036                 goto out;
3037         }
3038         BUG_ON(ret == -ENOENT);
3039         if (check_path_shared(root, path))
3040                 goto out;
3041
3042         err = 0;
3043 out:
3044         btrfs_free_path(path);
3045         /* Migrate the orphan reservation over */
3046         if (!err)
3047                 err = btrfs_block_rsv_migrate(trans->block_rsv,
3048                                 &root->fs_info->global_block_rsv,
3049                                 trans->bytes_reserved);
3050
3051         if (err) {
3052                 btrfs_end_transaction(trans, root);
3053                 root->fs_info->enospc_unlink = 0;
3054                 return ERR_PTR(err);
3055         }
3056
3057         trans->block_rsv = &root->fs_info->global_block_rsv;
3058         return trans;
3059 }
3060
3061 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
3062                                struct btrfs_root *root)
3063 {
3064         if (trans->block_rsv == &root->fs_info->global_block_rsv) {
3065                 btrfs_block_rsv_release(root, trans->block_rsv,
3066                                         trans->bytes_reserved);
3067                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3068                 BUG_ON(!root->fs_info->enospc_unlink);
3069                 root->fs_info->enospc_unlink = 0;
3070         }
3071         btrfs_end_transaction(trans, root);
3072 }
3073
3074 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3075 {
3076         struct btrfs_root *root = BTRFS_I(dir)->root;
3077         struct btrfs_trans_handle *trans;
3078         struct inode *inode = dentry->d_inode;
3079         int ret;
3080         unsigned long nr = 0;
3081
3082         trans = __unlink_start_trans(dir, dentry);
3083         if (IS_ERR(trans))
3084                 return PTR_ERR(trans);
3085
3086         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3087
3088         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3089                                  dentry->d_name.name, dentry->d_name.len);
3090         if (ret)
3091                 goto out;
3092
3093         if (inode->i_nlink == 0) {
3094                 ret = btrfs_orphan_add(trans, inode);
3095                 if (ret)
3096                         goto out;
3097         }
3098
3099 out:
3100         nr = trans->blocks_used;
3101         __unlink_end_trans(trans, root);
3102         btrfs_btree_balance_dirty(root, nr);
3103         return ret;
3104 }
3105
3106 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3107                         struct btrfs_root *root,
3108                         struct inode *dir, u64 objectid,
3109                         const char *name, int name_len)
3110 {
3111         struct btrfs_path *path;
3112         struct extent_buffer *leaf;
3113         struct btrfs_dir_item *di;
3114         struct btrfs_key key;
3115         u64 index;
3116         int ret;
3117         u64 dir_ino = btrfs_ino(dir);
3118
3119         path = btrfs_alloc_path();
3120         if (!path)
3121                 return -ENOMEM;
3122
3123         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3124                                    name, name_len, -1);
3125         if (IS_ERR_OR_NULL(di)) {
3126                 if (!di)
3127                         ret = -ENOENT;
3128                 else
3129                         ret = PTR_ERR(di);
3130                 goto out;
3131         }
3132
3133         leaf = path->nodes[0];
3134         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3135         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3136         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3137         if (ret) {
3138                 btrfs_abort_transaction(trans, root, ret);
3139                 goto out;
3140         }
3141         btrfs_release_path(path);
3142
3143         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3144                                  objectid, root->root_key.objectid,
3145                                  dir_ino, &index, name, name_len);
3146         if (ret < 0) {
3147                 if (ret != -ENOENT) {
3148                         btrfs_abort_transaction(trans, root, ret);
3149                         goto out;
3150                 }
3151                 di = btrfs_search_dir_index_item(root, path, dir_ino,
3152                                                  name, name_len);
3153                 if (IS_ERR_OR_NULL(di)) {
3154                         if (!di)
3155                                 ret = -ENOENT;
3156                         else
3157                                 ret = PTR_ERR(di);
3158                         btrfs_abort_transaction(trans, root, ret);
3159                         goto out;
3160                 }
3161
3162                 leaf = path->nodes[0];
3163                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3164                 btrfs_release_path(path);
3165                 index = key.offset;
3166         }
3167         btrfs_release_path(path);
3168
3169         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3170         if (ret) {
3171                 btrfs_abort_transaction(trans, root, ret);
3172                 goto out;
3173         }
3174
3175         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3176         inode_inc_iversion(dir);
3177         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3178         ret = btrfs_update_inode_fallback(trans, root, dir);
3179         if (ret)
3180                 btrfs_abort_transaction(trans, root, ret);
3181 out:
3182         btrfs_free_path(path);
3183         return ret;
3184 }
3185
3186 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3187 {
3188         struct inode *inode = dentry->d_inode;
3189         int err = 0;
3190         struct btrfs_root *root = BTRFS_I(dir)->root;
3191         struct btrfs_trans_handle *trans;
3192         unsigned long nr = 0;
3193
3194         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3195             btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3196                 return -ENOTEMPTY;
3197
3198         trans = __unlink_start_trans(dir, dentry);
3199         if (IS_ERR(trans))
3200                 return PTR_ERR(trans);
3201
3202         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3203                 err = btrfs_unlink_subvol(trans, root, dir,
3204                                           BTRFS_I(inode)->location.objectid,
3205                                           dentry->d_name.name,
3206                                           dentry->d_name.len);
3207                 goto out;
3208         }
3209
3210         err = btrfs_orphan_add(trans, inode);
3211         if (err)
3212                 goto out;
3213
3214         /* now the directory is empty */
3215         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3216                                  dentry->d_name.name, dentry->d_name.len);
3217         if (!err)
3218                 btrfs_i_size_write(inode, 0);
3219 out:
3220         nr = trans->blocks_used;
3221         __unlink_end_trans(trans, root);
3222         btrfs_btree_balance_dirty(root, nr);
3223
3224         return err;
3225 }
3226
3227 /*
3228  * this can truncate away extent items, csum items and directory items.
3229  * It starts at a high offset and removes keys until it can't find
3230  * any higher than new_size
3231  *
3232  * csum items that cross the new i_size are truncated to the new size
3233  * as well.
3234  *
3235  * min_type is the minimum key type to truncate down to.  If set to 0, this
3236  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3237  */
3238 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3239                                struct btrfs_root *root,
3240                                struct inode *inode,
3241                                u64 new_size, u32 min_type)
3242 {
3243         struct btrfs_path *path;
3244         struct extent_buffer *leaf;
3245         struct btrfs_file_extent_item *fi;
3246         struct btrfs_key key;
3247         struct btrfs_key found_key;
3248         u64 extent_start = 0;
3249         u64 extent_num_bytes = 0;
3250         u64 extent_offset = 0;
3251         u64 item_end = 0;
3252         u64 mask = root->sectorsize - 1;
3253         u32 found_type = (u8)-1;
3254         int found_extent;
3255         int del_item;
3256         int pending_del_nr = 0;
3257         int pending_del_slot = 0;
3258         int extent_type = -1;
3259         int ret;
3260         int err = 0;
3261         u64 ino = btrfs_ino(inode);
3262
3263         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3264
3265         path = btrfs_alloc_path();
3266         if (!path)
3267                 return -ENOMEM;
3268         path->reada = -1;
3269
3270         if (root->ref_cows || root == root->fs_info->tree_root)
3271                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3272
3273         /*
3274          * This function is also used to drop the items in the log tree before
3275          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3276          * it is used to drop the loged items. So we shouldn't kill the delayed
3277          * items.
3278          */
3279         if (min_type == 0 && root == BTRFS_I(inode)->root)
3280                 btrfs_kill_delayed_inode_items(inode);
3281
3282         key.objectid = ino;
3283         key.offset = (u64)-1;
3284         key.type = (u8)-1;
3285
3286 search_again:
3287         path->leave_spinning = 1;
3288         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3289         if (ret < 0) {
3290                 err = ret;
3291                 goto out;
3292         }
3293
3294         if (ret > 0) {
3295                 /* there are no items in the tree for us to truncate, we're
3296                  * done
3297                  */
3298                 if (path->slots[0] == 0)
3299                         goto out;
3300                 path->slots[0]--;
3301         }
3302
3303         while (1) {
3304                 fi = NULL;
3305                 leaf = path->nodes[0];
3306                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3307                 found_type = btrfs_key_type(&found_key);
3308
3309                 if (found_key.objectid != ino)
3310                         break;
3311
3312                 if (found_type < min_type)
3313                         break;
3314
3315                 item_end = found_key.offset;
3316                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3317                         fi = btrfs_item_ptr(leaf, path->slots[0],
3318                                             struct btrfs_file_extent_item);
3319                         extent_type = btrfs_file_extent_type(leaf, fi);
3320                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3321                                 item_end +=
3322                                     btrfs_file_extent_num_bytes(leaf, fi);
3323                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3324                                 item_end += btrfs_file_extent_inline_len(leaf,
3325                                                                          fi);
3326                         }
3327                         item_end--;
3328                 }
3329                 if (found_type > min_type) {
3330                         del_item = 1;
3331                 } else {
3332                         if (item_end < new_size)
3333                                 break;
3334                         if (found_key.offset >= new_size)
3335                                 del_item = 1;
3336                         else
3337                                 del_item = 0;
3338                 }
3339                 found_extent = 0;
3340                 /* FIXME, shrink the extent if the ref count is only 1 */
3341                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3342                         goto delete;
3343
3344                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3345                         u64 num_dec;
3346                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3347                         if (!del_item) {
3348                                 u64 orig_num_bytes =
3349                                         btrfs_file_extent_num_bytes(leaf, fi);
3350                                 extent_num_bytes = new_size -
3351                                         found_key.offset + root->sectorsize - 1;
3352                                 extent_num_bytes = extent_num_bytes &
3353                                         ~((u64)root->sectorsize - 1);
3354                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3355                                                          extent_num_bytes);
3356                                 num_dec = (orig_num_bytes -
3357                                            extent_num_bytes);
3358                                 if (root->ref_cows && extent_start != 0)
3359                                         inode_sub_bytes(inode, num_dec);
3360                                 btrfs_mark_buffer_dirty(leaf);
3361                         } else {
3362                                 extent_num_bytes =
3363                                         btrfs_file_extent_disk_num_bytes(leaf,
3364                                                                          fi);
3365                                 extent_offset = found_key.offset -
3366                                         btrfs_file_extent_offset(leaf, fi);
3367
3368                                 /* FIXME blocksize != 4096 */
3369                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3370                                 if (extent_start != 0) {
3371                                         found_extent = 1;
3372                                         if (root->ref_cows)
3373                                                 inode_sub_bytes(inode, num_dec);
3374                                 }
3375                         }
3376                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3377                         /*
3378                          * we can't truncate inline items that have had
3379                          * special encodings
3380                          */
3381                         if (!del_item &&
3382                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3383                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3384                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3385                                 u32 size = new_size - found_key.offset;
3386
3387                                 if (root->ref_cows) {
3388                                         inode_sub_bytes(inode, item_end + 1 -
3389                                                         new_size);
3390                                 }
3391                                 size =
3392                                     btrfs_file_extent_calc_inline_size(size);
3393                                 btrfs_truncate_item(trans, root, path,
3394                                                     size, 1);
3395                         } else if (root->ref_cows) {
3396                                 inode_sub_bytes(inode, item_end + 1 -
3397                                                 found_key.offset);
3398                         }
3399                 }
3400 delete:
3401                 if (del_item) {
3402                         if (!pending_del_nr) {
3403                                 /* no pending yet, add ourselves */
3404                                 pending_del_slot = path->slots[0];
3405                                 pending_del_nr = 1;
3406                         } else if (pending_del_nr &&
3407                                    path->slots[0] + 1 == pending_del_slot) {
3408                                 /* hop on the pending chunk */
3409                                 pending_del_nr++;
3410                                 pending_del_slot = path->slots[0];
3411                         } else {
3412                                 BUG();
3413                         }
3414                 } else {
3415                         break;
3416                 }
3417                 if (found_extent && (root->ref_cows ||
3418                                      root == root->fs_info->tree_root)) {
3419                         btrfs_set_path_blocking(path);
3420                         ret = btrfs_free_extent(trans, root, extent_start,
3421                                                 extent_num_bytes, 0,
3422                                                 btrfs_header_owner(leaf),
3423                                                 ino, extent_offset, 0);
3424                         BUG_ON(ret);
3425                 }
3426
3427                 if (found_type == BTRFS_INODE_ITEM_KEY)
3428                         break;
3429
3430                 if (path->slots[0] == 0 ||
3431                     path->slots[0] != pending_del_slot) {
3432                         if (root->ref_cows &&
3433                             BTRFS_I(inode)->location.objectid !=
3434                                                 BTRFS_FREE_INO_OBJECTID) {
3435                                 err = -EAGAIN;
3436                                 goto out;
3437                         }
3438                         if (pending_del_nr) {
3439                                 ret = btrfs_del_items(trans, root, path,
3440                                                 pending_del_slot,
3441                                                 pending_del_nr);
3442                                 if (ret) {
3443                                         btrfs_abort_transaction(trans,
3444                                                                 root, ret);
3445                                         goto error;
3446                                 }
3447                                 pending_del_nr = 0;
3448                         }
3449                         btrfs_release_path(path);
3450                         goto search_again;
3451                 } else {
3452                         path->slots[0]--;
3453                 }
3454         }
3455 out:
3456         if (pending_del_nr) {
3457                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3458                                       pending_del_nr);
3459                 if (ret)
3460                         btrfs_abort_transaction(trans, root, ret);
3461         }
3462 error:
3463         btrfs_free_path(path);
3464         return err;
3465 }
3466
3467 /*
3468  * taken from block_truncate_page, but does cow as it zeros out
3469  * any bytes left in the last page in the file.
3470  */
3471 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3472 {
3473         struct inode *inode = mapping->host;
3474         struct btrfs_root *root = BTRFS_I(inode)->root;
3475         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3476         struct btrfs_ordered_extent *ordered;
3477         struct extent_state *cached_state = NULL;
3478         char *kaddr;
3479         u32 blocksize = root->sectorsize;
3480         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3481         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3482         struct page *page;
3483         gfp_t mask = btrfs_alloc_write_mask(mapping);
3484         int ret = 0;
3485         u64 page_start;
3486         u64 page_end;
3487
3488         if ((offset & (blocksize - 1)) == 0)
3489                 goto out;
3490         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3491         if (ret)
3492                 goto out;
3493
3494         ret = -ENOMEM;
3495 again:
3496         page = find_or_create_page(mapping, index, mask);
3497         if (!page) {
3498                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3499                 goto out;
3500         }
3501
3502         page_start = page_offset(page);
3503         page_end = page_start + PAGE_CACHE_SIZE - 1;
3504
3505         if (!PageUptodate(page)) {
3506                 ret = btrfs_readpage(NULL, page);
3507                 lock_page(page);
3508                 if (page->mapping != mapping) {
3509                         unlock_page(page);
3510                         page_cache_release(page);
3511                         goto again;
3512                 }
3513                 if (!PageUptodate(page)) {
3514                         ret = -EIO;
3515                         goto out_unlock;
3516                 }
3517         }
3518         wait_on_page_writeback(page);
3519
3520         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
3521         set_page_extent_mapped(page);
3522
3523         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3524         if (ordered) {
3525                 unlock_extent_cached(io_tree, page_start, page_end,
3526                                      &cached_state, GFP_NOFS);
3527                 unlock_page(page);
3528                 page_cache_release(page);
3529                 btrfs_start_ordered_extent(inode, ordered, 1);
3530                 btrfs_put_ordered_extent(ordered);
3531                 goto again;
3532         }
3533
3534         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3535                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3536                           0, 0, &cached_state, GFP_NOFS);
3537
3538         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3539                                         &cached_state);
3540         if (ret) {
3541                 unlock_extent_cached(io_tree, page_start, page_end,
3542                                      &cached_state, GFP_NOFS);
3543                 goto out_unlock;
3544         }
3545
3546         ret = 0;
3547         if (offset != PAGE_CACHE_SIZE) {
3548                 kaddr = kmap(page);
3549                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3550                 flush_dcache_page(page);
3551                 kunmap(page);
3552         }
3553         ClearPageChecked(page);
3554         set_page_dirty(page);
3555         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3556                              GFP_NOFS);
3557
3558 out_unlock:
3559         if (ret)
3560                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3561         unlock_page(page);
3562         page_cache_release(page);
3563 out:
3564         return ret;
3565 }
3566
3567 /*
3568  * This function puts in dummy file extents for the area we're creating a hole
3569  * for.  So if we are truncating this file to a larger size we need to insert
3570  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3571  * the range between oldsize and size
3572  */
3573 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3574 {
3575         struct btrfs_trans_handle *trans;
3576         struct btrfs_root *root = BTRFS_I(inode)->root;
3577         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3578         struct extent_map *em = NULL;
3579         struct extent_state *cached_state = NULL;
3580         u64 mask = root->sectorsize - 1;
3581         u64 hole_start = (oldsize + mask) & ~mask;
3582         u64 block_end = (size + mask) & ~mask;
3583         u64 last_byte;
3584         u64 cur_offset;
3585         u64 hole_size;
3586         int err = 0;
3587
3588         if (size <= hole_start)
3589                 return 0;
3590
3591         while (1) {
3592                 struct btrfs_ordered_extent *ordered;
3593                 btrfs_wait_ordered_range(inode, hole_start,
3594                                          block_end - hole_start);
3595                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3596                                  &cached_state);
3597                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3598                 if (!ordered)
3599                         break;
3600                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3601                                      &cached_state, GFP_NOFS);
3602                 btrfs_put_ordered_extent(ordered);
3603         }
3604
3605         cur_offset = hole_start;
3606         while (1) {
3607                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3608                                 block_end - cur_offset, 0);
3609                 if (IS_ERR(em)) {
3610                         err = PTR_ERR(em);
3611                         break;
3612                 }
3613                 last_byte = min(extent_map_end(em), block_end);
3614                 last_byte = (last_byte + mask) & ~mask;
3615                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3616                         u64 hint_byte = 0;
3617                         hole_size = last_byte - cur_offset;
3618
3619                         trans = btrfs_start_transaction(root, 3);
3620                         if (IS_ERR(trans)) {
3621                                 err = PTR_ERR(trans);
3622                                 break;
3623                         }
3624
3625                         err = btrfs_drop_extents(trans, inode, cur_offset,
3626                                                  cur_offset + hole_size,
3627                                                  &hint_byte, 1);
3628                         if (err) {
3629                                 btrfs_abort_transaction(trans, root, err);
3630                                 btrfs_end_transaction(trans, root);
3631                                 break;
3632                         }
3633
3634                         err = btrfs_insert_file_extent(trans, root,
3635                                         btrfs_ino(inode), cur_offset, 0,
3636                                         0, hole_size, 0, hole_size,
3637                                         0, 0, 0);
3638                         if (err) {
3639                                 btrfs_abort_transaction(trans, root, err);
3640                                 btrfs_end_transaction(trans, root);
3641                                 break;
3642                         }
3643
3644                         btrfs_drop_extent_cache(inode, hole_start,
3645                                         last_byte - 1, 0);
3646
3647                         btrfs_update_inode(trans, root, inode);
3648                         btrfs_end_transaction(trans, root);
3649                 }
3650                 free_extent_map(em);
3651                 em = NULL;
3652                 cur_offset = last_byte;
3653                 if (cur_offset >= block_end)
3654                         break;
3655         }
3656
3657         free_extent_map(em);
3658         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3659                              GFP_NOFS);
3660         return err;
3661 }
3662
3663 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3664 {
3665         struct btrfs_root *root = BTRFS_I(inode)->root;
3666         struct btrfs_trans_handle *trans;
3667         loff_t oldsize = i_size_read(inode);
3668         int ret;
3669
3670         if (newsize == oldsize)
3671                 return 0;
3672
3673         if (newsize > oldsize) {
3674                 truncate_pagecache(inode, oldsize, newsize);
3675                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3676                 if (ret)
3677                         return ret;
3678
3679                 trans = btrfs_start_transaction(root, 1);
3680                 if (IS_ERR(trans))
3681                         return PTR_ERR(trans);
3682
3683                 i_size_write(inode, newsize);
3684                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3685                 ret = btrfs_update_inode(trans, root, inode);
3686                 btrfs_end_transaction(trans, root);
3687         } else {
3688
3689                 /*
3690                  * We're truncating a file that used to have good data down to
3691                  * zero. Make sure it gets into the ordered flush list so that
3692                  * any new writes get down to disk quickly.
3693                  */
3694                 if (newsize == 0)
3695                         set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
3696                                 &BTRFS_I(inode)->runtime_flags);
3697
3698                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3699                 truncate_setsize(inode, newsize);
3700                 ret = btrfs_truncate(inode);
3701         }
3702
3703         return ret;
3704 }
3705
3706 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3707 {
3708         struct inode *inode = dentry->d_inode;
3709         struct btrfs_root *root = BTRFS_I(inode)->root;
3710         int err;
3711
3712         if (btrfs_root_readonly(root))
3713                 return -EROFS;
3714
3715         err = inode_change_ok(inode, attr);
3716         if (err)
3717                 return err;
3718
3719         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3720                 err = btrfs_setsize(inode, attr->ia_size);
3721                 if (err)
3722                         return err;
3723         }
3724
3725         if (attr->ia_valid) {
3726                 setattr_copy(inode, attr);
3727                 inode_inc_iversion(inode);
3728                 err = btrfs_dirty_inode(inode);
3729
3730                 if (!err && attr->ia_valid & ATTR_MODE)
3731                         err = btrfs_acl_chmod(inode);
3732         }
3733
3734         return err;
3735 }
3736
3737 void btrfs_evict_inode(struct inode *inode)
3738 {
3739         struct btrfs_trans_handle *trans;
3740         struct btrfs_root *root = BTRFS_I(inode)->root;
3741         struct btrfs_block_rsv *rsv, *global_rsv;
3742         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
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                                btrfs_is_free_space_inode(inode)))
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(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3762                                  &BTRFS_I(inode)->runtime_flags));
3763                 goto no_delete;
3764         }
3765
3766         if (inode->i_nlink > 0) {
3767                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3768                 goto no_delete;
3769         }
3770
3771         rsv = btrfs_alloc_block_rsv(root);
3772         if (!rsv) {
3773                 btrfs_orphan_del(NULL, inode);
3774                 goto no_delete;
3775         }
3776         rsv->size = min_size;
3777         global_rsv = &root->fs_info->global_block_rsv;
3778
3779         btrfs_i_size_write(inode, 0);
3780
3781         /*
3782          * This is a bit simpler than btrfs_truncate since
3783          *
3784          * 1) We've already reserved our space for our orphan item in the
3785          *    unlink.
3786          * 2) We're going to delete the inode item, so we don't need to update
3787          *    it at all.
3788          *
3789          * So we just need to reserve some slack space in case we add bytes when
3790          * doing the truncate.
3791          */
3792         while (1) {
3793                 ret = btrfs_block_rsv_refill_noflush(root, rsv, min_size);
3794
3795                 /*
3796                  * Try and steal from the global reserve since we will
3797                  * likely not use this space anyway, we want to try as
3798                  * hard as possible to get this to work.
3799                  */
3800                 if (ret)
3801                         ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
3802
3803                 if (ret) {
3804                         printk(KERN_WARNING "Could not get space for a "
3805                                "delete, will truncate on mount %d\n", ret);
3806                         btrfs_orphan_del(NULL, inode);
3807                         btrfs_free_block_rsv(root, rsv);
3808                         goto no_delete;
3809                 }
3810
3811                 trans = btrfs_start_transaction(root, 0);
3812                 if (IS_ERR(trans)) {
3813                         btrfs_orphan_del(NULL, inode);
3814                         btrfs_free_block_rsv(root, rsv);
3815                         goto no_delete;
3816                 }
3817
3818                 trans->block_rsv = rsv;
3819
3820                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3821                 if (ret != -EAGAIN)
3822                         break;
3823
3824                 nr = trans->blocks_used;
3825                 btrfs_end_transaction(trans, root);
3826                 trans = NULL;
3827                 btrfs_btree_balance_dirty(root, nr);
3828         }
3829
3830         btrfs_free_block_rsv(root, rsv);
3831
3832         if (ret == 0) {
3833                 trans->block_rsv = root->orphan_block_rsv;
3834                 ret = btrfs_orphan_del(trans, inode);
3835                 BUG_ON(ret);
3836         }
3837
3838         trans->block_rsv = &root->fs_info->trans_block_rsv;
3839         if (!(root == root->fs_info->tree_root ||
3840               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3841                 btrfs_return_ino(root, btrfs_ino(inode));
3842
3843         nr = trans->blocks_used;
3844         btrfs_end_transaction(trans, root);
3845         btrfs_btree_balance_dirty(root, nr);
3846 no_delete:
3847         clear_inode(inode);
3848         return;
3849 }
3850
3851 /*
3852  * this returns the key found in the dir entry in the location pointer.
3853  * If no dir entries were found, location->objectid is 0.
3854  */
3855 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3856                                struct btrfs_key *location)
3857 {
3858         const char *name = dentry->d_name.name;
3859         int namelen = dentry->d_name.len;
3860         struct btrfs_dir_item *di;
3861         struct btrfs_path *path;
3862         struct btrfs_root *root = BTRFS_I(dir)->root;
3863         int ret = 0;
3864
3865         path = btrfs_alloc_path();
3866         if (!path)
3867                 return -ENOMEM;
3868
3869         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3870                                     namelen, 0);
3871         if (IS_ERR(di))
3872                 ret = PTR_ERR(di);
3873
3874         if (IS_ERR_OR_NULL(di))
3875                 goto out_err;
3876
3877         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3878 out:
3879         btrfs_free_path(path);
3880         return ret;
3881 out_err:
3882         location->objectid = 0;
3883         goto out;
3884 }
3885
3886 /*
3887  * when we hit a tree root in a directory, the btrfs part of the inode
3888  * needs to be changed to reflect the root directory of the tree root.  This
3889  * is kind of like crossing a mount point.
3890  */
3891 static int fixup_tree_root_location(struct btrfs_root *root,
3892                                     struct inode *dir,
3893                                     struct dentry *dentry,
3894                                     struct btrfs_key *location,
3895                                     struct btrfs_root **sub_root)
3896 {
3897         struct btrfs_path *path;
3898         struct btrfs_root *new_root;
3899         struct btrfs_root_ref *ref;
3900         struct extent_buffer *leaf;
3901         int ret;
3902         int err = 0;
3903
3904         path = btrfs_alloc_path();
3905         if (!path) {
3906                 err = -ENOMEM;
3907                 goto out;
3908         }
3909
3910         err = -ENOENT;
3911         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3912                                   BTRFS_I(dir)->root->root_key.objectid,
3913                                   location->objectid);
3914         if (ret) {
3915                 if (ret < 0)
3916                         err = ret;
3917                 goto out;
3918         }
3919
3920         leaf = path->nodes[0];
3921         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3922         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3923             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3924                 goto out;
3925
3926         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3927                                    (unsigned long)(ref + 1),
3928                                    dentry->d_name.len);
3929         if (ret)
3930                 goto out;
3931
3932         btrfs_release_path(path);
3933
3934         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3935         if (IS_ERR(new_root)) {
3936                 err = PTR_ERR(new_root);
3937                 goto out;
3938         }
3939
3940         if (btrfs_root_refs(&new_root->root_item) == 0) {
3941                 err = -ENOENT;
3942                 goto out;
3943         }
3944
3945         *sub_root = new_root;
3946         location->objectid = btrfs_root_dirid(&new_root->root_item);
3947         location->type = BTRFS_INODE_ITEM_KEY;
3948         location->offset = 0;
3949         err = 0;
3950 out:
3951         btrfs_free_path(path);
3952         return err;
3953 }
3954
3955 static void inode_tree_add(struct inode *inode)
3956 {
3957         struct btrfs_root *root = BTRFS_I(inode)->root;
3958         struct btrfs_inode *entry;
3959         struct rb_node **p;
3960         struct rb_node *parent;
3961         u64 ino = btrfs_ino(inode);
3962 again:
3963         p = &root->inode_tree.rb_node;
3964         parent = NULL;
3965
3966         if (inode_unhashed(inode))
3967                 return;
3968
3969         spin_lock(&root->inode_lock);
3970         while (*p) {
3971                 parent = *p;
3972                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3973
3974                 if (ino < btrfs_ino(&entry->vfs_inode))
3975                         p = &parent->rb_left;
3976                 else if (ino > btrfs_ino(&entry->vfs_inode))
3977                         p = &parent->rb_right;
3978                 else {
3979                         WARN_ON(!(entry->vfs_inode.i_state &
3980                                   (I_WILL_FREE | I_FREEING)));
3981                         rb_erase(parent, &root->inode_tree);
3982                         RB_CLEAR_NODE(parent);
3983                         spin_unlock(&root->inode_lock);
3984                         goto again;
3985                 }
3986         }
3987         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3988         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3989         spin_unlock(&root->inode_lock);
3990 }
3991
3992 static void inode_tree_del(struct inode *inode)
3993 {
3994         struct btrfs_root *root = BTRFS_I(inode)->root;
3995         int empty = 0;
3996
3997         spin_lock(&root->inode_lock);
3998         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3999                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4000                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4001                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4002         }
4003         spin_unlock(&root->inode_lock);
4004
4005         /*
4006          * Free space cache has inodes in the tree root, but the tree root has a
4007          * root_refs of 0, so this could end up dropping the tree root as a
4008          * snapshot, so we need the extra !root->fs_info->tree_root check to
4009          * make sure we don't drop it.
4010          */
4011         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
4012             root != root->fs_info->tree_root) {
4013                 synchronize_srcu(&root->fs_info->subvol_srcu);
4014                 spin_lock(&root->inode_lock);
4015                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4016                 spin_unlock(&root->inode_lock);
4017                 if (empty)
4018                         btrfs_add_dead_root(root);
4019         }
4020 }
4021
4022 void btrfs_invalidate_inodes(struct btrfs_root *root)
4023 {
4024         struct rb_node *node;
4025         struct rb_node *prev;
4026         struct btrfs_inode *entry;
4027         struct inode *inode;
4028         u64 objectid = 0;
4029
4030         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4031
4032         spin_lock(&root->inode_lock);
4033 again:
4034         node = root->inode_tree.rb_node;
4035         prev = NULL;
4036         while (node) {
4037                 prev = node;
4038                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4039
4040                 if (objectid < btrfs_ino(&entry->vfs_inode))
4041                         node = node->rb_left;
4042                 else if (objectid > btrfs_ino(&entry->vfs_inode))
4043                         node = node->rb_right;
4044                 else
4045                         break;
4046         }
4047         if (!node) {
4048                 while (prev) {
4049                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
4050                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4051                                 node = prev;
4052                                 break;
4053                         }
4054                         prev = rb_next(prev);
4055                 }
4056         }
4057         while (node) {
4058                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4059                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4060                 inode = igrab(&entry->vfs_inode);
4061                 if (inode) {
4062                         spin_unlock(&root->inode_lock);
4063                         if (atomic_read(&inode->i_count) > 1)
4064                                 d_prune_aliases(inode);
4065                         /*
4066                          * btrfs_drop_inode will have it removed from
4067                          * the inode cache when its usage count
4068                          * hits zero.
4069                          */
4070                         iput(inode);
4071                         cond_resched();
4072                         spin_lock(&root->inode_lock);
4073                         goto again;
4074                 }
4075
4076                 if (cond_resched_lock(&root->inode_lock))
4077                         goto again;
4078
4079                 node = rb_next(node);
4080         }
4081         spin_unlock(&root->inode_lock);
4082 }
4083
4084 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4085 {
4086         struct btrfs_iget_args *args = p;
4087         inode->i_ino = args->ino;
4088         BTRFS_I(inode)->root = args->root;
4089         return 0;
4090 }
4091
4092 static int btrfs_find_actor(struct inode *inode, void *opaque)
4093 {
4094         struct btrfs_iget_args *args = opaque;
4095         return args->ino == btrfs_ino(inode) &&
4096                 args->root == BTRFS_I(inode)->root;
4097 }
4098
4099 static struct inode *btrfs_iget_locked(struct super_block *s,
4100                                        u64 objectid,
4101                                        struct btrfs_root *root)
4102 {
4103         struct inode *inode;
4104         struct btrfs_iget_args args;
4105         args.ino = objectid;
4106         args.root = root;
4107
4108         inode = iget5_locked(s, objectid, btrfs_find_actor,
4109                              btrfs_init_locked_inode,
4110                              (void *)&args);
4111         return inode;
4112 }
4113
4114 /* Get an inode object given its location and corresponding root.
4115  * Returns in *is_new if the inode was read from disk
4116  */
4117 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4118                          struct btrfs_root *root, int *new)
4119 {
4120         struct inode *inode;
4121
4122         inode = btrfs_iget_locked(s, location->objectid, root);
4123         if (!inode)
4124                 return ERR_PTR(-ENOMEM);
4125
4126         if (inode->i_state & I_NEW) {
4127                 BTRFS_I(inode)->root = root;
4128                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4129                 btrfs_read_locked_inode(inode);
4130                 if (!is_bad_inode(inode)) {
4131                         inode_tree_add(inode);
4132                         unlock_new_inode(inode);
4133                         if (new)
4134                                 *new = 1;
4135                 } else {
4136                         unlock_new_inode(inode);
4137                         iput(inode);
4138                         inode = ERR_PTR(-ESTALE);
4139                 }
4140         }
4141
4142         return inode;
4143 }
4144
4145 static struct inode *new_simple_dir(struct super_block *s,
4146                                     struct btrfs_key *key,
4147                                     struct btrfs_root *root)
4148 {
4149         struct inode *inode = new_inode(s);
4150
4151         if (!inode)
4152                 return ERR_PTR(-ENOMEM);
4153
4154         BTRFS_I(inode)->root = root;
4155         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4156         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4157
4158         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4159         inode->i_op = &btrfs_dir_ro_inode_operations;
4160         inode->i_fop = &simple_dir_operations;
4161         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4162         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4163
4164         return inode;
4165 }
4166
4167 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4168 {
4169         struct inode *inode;
4170         struct btrfs_root *root = BTRFS_I(dir)->root;
4171         struct btrfs_root *sub_root = root;
4172         struct btrfs_key location;
4173         int index;
4174         int ret = 0;
4175
4176         if (dentry->d_name.len > BTRFS_NAME_LEN)
4177                 return ERR_PTR(-ENAMETOOLONG);
4178
4179         if (unlikely(d_need_lookup(dentry))) {
4180                 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
4181                 kfree(dentry->d_fsdata);
4182                 dentry->d_fsdata = NULL;
4183                 /* This thing is hashed, drop it for now */
4184                 d_drop(dentry);
4185         } else {
4186                 ret = btrfs_inode_by_name(dir, dentry, &location);
4187         }
4188
4189         if (ret < 0)
4190                 return ERR_PTR(ret);
4191
4192         if (location.objectid == 0)
4193                 return NULL;
4194
4195         if (location.type == BTRFS_INODE_ITEM_KEY) {
4196                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4197                 return inode;
4198         }
4199
4200         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4201
4202         index = srcu_read_lock(&root->fs_info->subvol_srcu);
4203         ret = fixup_tree_root_location(root, dir, dentry,
4204                                        &location, &sub_root);
4205         if (ret < 0) {
4206                 if (ret != -ENOENT)
4207                         inode = ERR_PTR(ret);
4208                 else
4209                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
4210         } else {
4211                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4212         }
4213         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4214
4215         if (!IS_ERR(inode) && root != sub_root) {
4216                 down_read(&root->fs_info->cleanup_work_sem);
4217                 if (!(inode->i_sb->s_flags & MS_RDONLY))
4218                         ret = btrfs_orphan_cleanup(sub_root);
4219                 up_read(&root->fs_info->cleanup_work_sem);
4220                 if (ret)
4221                         inode = ERR_PTR(ret);
4222         }
4223
4224         return inode;
4225 }
4226
4227 static int btrfs_dentry_delete(const struct dentry *dentry)
4228 {
4229         struct btrfs_root *root;
4230         struct inode *inode = dentry->d_inode;
4231
4232         if (!inode && !IS_ROOT(dentry))
4233                 inode = dentry->d_parent->d_inode;
4234
4235         if (inode) {
4236                 root = BTRFS_I(inode)->root;
4237                 if (btrfs_root_refs(&root->root_item) == 0)
4238                         return 1;
4239
4240                 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
4241                         return 1;
4242         }
4243         return 0;
4244 }
4245
4246 static void btrfs_dentry_release(struct dentry *dentry)
4247 {
4248         if (dentry->d_fsdata)
4249                 kfree(dentry->d_fsdata);
4250 }
4251
4252 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4253                                    unsigned int flags)
4254 {
4255         struct dentry *ret;
4256
4257         ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4258         if (unlikely(d_need_lookup(dentry))) {
4259                 spin_lock(&dentry->d_lock);
4260                 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4261                 spin_unlock(&dentry->d_lock);
4262         }
4263         return ret;
4264 }
4265
4266 unsigned char btrfs_filetype_table[] = {
4267         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4268 };
4269
4270 static int btrfs_real_readdir(struct file *filp, void *dirent,
4271                               filldir_t filldir)
4272 {
4273         struct inode *inode = filp->f_dentry->d_inode;
4274         struct btrfs_root *root = BTRFS_I(inode)->root;
4275         struct btrfs_item *item;
4276         struct btrfs_dir_item *di;
4277         struct btrfs_key key;
4278         struct btrfs_key found_key;
4279         struct btrfs_path *path;
4280         struct list_head ins_list;
4281         struct list_head del_list;
4282         int ret;
4283         struct extent_buffer *leaf;
4284         int slot;
4285         unsigned char d_type;
4286         int over = 0;
4287         u32 di_cur;
4288         u32 di_total;
4289         u32 di_len;
4290         int key_type = BTRFS_DIR_INDEX_KEY;
4291         char tmp_name[32];
4292         char *name_ptr;
4293         int name_len;
4294         int is_curr = 0;        /* filp->f_pos points to the current index? */
4295
4296         /* FIXME, use a real flag for deciding about the key type */
4297         if (root->fs_info->tree_root == root)
4298                 key_type = BTRFS_DIR_ITEM_KEY;
4299
4300         /* special case for "." */
4301         if (filp->f_pos == 0) {
4302                 over = filldir(dirent, ".", 1,
4303                                filp->f_pos, btrfs_ino(inode), DT_DIR);
4304                 if (over)
4305                         return 0;
4306                 filp->f_pos = 1;
4307         }
4308         /* special case for .., just use the back ref */
4309         if (filp->f_pos == 1) {
4310                 u64 pino = parent_ino(filp->f_path.dentry);
4311                 over = filldir(dirent, "..", 2,
4312                                filp->f_pos, pino, DT_DIR);
4313                 if (over)
4314                         return 0;
4315                 filp->f_pos = 2;
4316         }
4317         path = btrfs_alloc_path();
4318         if (!path)
4319                 return -ENOMEM;
4320
4321         path->reada = 1;
4322
4323         if (key_type == BTRFS_DIR_INDEX_KEY) {
4324                 INIT_LIST_HEAD(&ins_list);
4325                 INIT_LIST_HEAD(&del_list);
4326                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4327         }
4328
4329         btrfs_set_key_type(&key, key_type);
4330         key.offset = filp->f_pos;
4331         key.objectid = btrfs_ino(inode);
4332
4333         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4334         if (ret < 0)
4335                 goto err;
4336
4337         while (1) {
4338                 leaf = path->nodes[0];
4339                 slot = path->slots[0];
4340                 if (slot >= btrfs_header_nritems(leaf)) {
4341                         ret = btrfs_next_leaf(root, path);
4342                         if (ret < 0)
4343                                 goto err;
4344                         else if (ret > 0)
4345                                 break;
4346                         continue;
4347                 }
4348
4349                 item = btrfs_item_nr(leaf, slot);
4350                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4351
4352                 if (found_key.objectid != key.objectid)
4353                         break;
4354                 if (btrfs_key_type(&found_key) != key_type)
4355                         break;
4356                 if (found_key.offset < filp->f_pos)
4357                         goto next;
4358                 if (key_type == BTRFS_DIR_INDEX_KEY &&
4359                     btrfs_should_delete_dir_index(&del_list,
4360                                                   found_key.offset))
4361                         goto next;
4362
4363                 filp->f_pos = found_key.offset;
4364                 is_curr = 1;
4365
4366                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4367                 di_cur = 0;
4368                 di_total = btrfs_item_size(leaf, item);
4369
4370                 while (di_cur < di_total) {
4371                         struct btrfs_key location;
4372
4373                         if (verify_dir_item(root, leaf, di))
4374                                 break;
4375
4376                         name_len = btrfs_dir_name_len(leaf, di);
4377                         if (name_len <= sizeof(tmp_name)) {
4378                                 name_ptr = tmp_name;
4379                         } else {
4380                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4381                                 if (!name_ptr) {
4382                                         ret = -ENOMEM;
4383                                         goto err;
4384                                 }
4385                         }
4386                         read_extent_buffer(leaf, name_ptr,
4387                                            (unsigned long)(di + 1), name_len);
4388
4389                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4390                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4391
4392
4393                         /* is this a reference to our own snapshot? If so
4394                          * skip it.
4395                          *
4396                          * In contrast to old kernels, we insert the snapshot's
4397                          * dir item and dir index after it has been created, so
4398                          * we won't find a reference to our own snapshot. We
4399                          * still keep the following code for backward
4400                          * compatibility.
4401                          */
4402                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4403                             location.objectid == root->root_key.objectid) {
4404                                 over = 0;
4405                                 goto skip;
4406                         }
4407                         over = filldir(dirent, name_ptr, name_len,
4408                                        found_key.offset, location.objectid,
4409                                        d_type);
4410
4411 skip:
4412                         if (name_ptr != tmp_name)
4413                                 kfree(name_ptr);
4414
4415                         if (over)
4416                                 goto nopos;
4417                         di_len = btrfs_dir_name_len(leaf, di) +
4418                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4419                         di_cur += di_len;
4420                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4421                 }
4422 next:
4423                 path->slots[0]++;
4424         }
4425
4426         if (key_type == BTRFS_DIR_INDEX_KEY) {
4427                 if (is_curr)
4428                         filp->f_pos++;
4429                 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4430                                                       &ins_list);
4431                 if (ret)
4432                         goto nopos;
4433         }
4434
4435         /* Reached end of directory/root. Bump pos past the last item. */
4436         if (key_type == BTRFS_DIR_INDEX_KEY)
4437                 /*
4438                  * 32-bit glibc will use getdents64, but then strtol -
4439                  * so the last number we can serve is this.
4440                  */
4441                 filp->f_pos = 0x7fffffff;
4442         else
4443                 filp->f_pos++;
4444 nopos:
4445         ret = 0;
4446 err:
4447         if (key_type == BTRFS_DIR_INDEX_KEY)
4448                 btrfs_put_delayed_items(&ins_list, &del_list);
4449         btrfs_free_path(path);
4450         return ret;
4451 }
4452
4453 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4454 {
4455         struct btrfs_root *root = BTRFS_I(inode)->root;
4456         struct btrfs_trans_handle *trans;
4457         int ret = 0;
4458         bool nolock = false;
4459
4460         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4461                 return 0;
4462
4463         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
4464                 nolock = true;
4465
4466         if (wbc->sync_mode == WB_SYNC_ALL) {
4467                 if (nolock)
4468                         trans = btrfs_join_transaction_nolock(root);
4469                 else
4470                         trans = btrfs_join_transaction(root);
4471                 if (IS_ERR(trans))
4472                         return PTR_ERR(trans);
4473                 if (nolock)
4474                         ret = btrfs_end_transaction_nolock(trans, root);
4475                 else
4476                         ret = btrfs_commit_transaction(trans, root);
4477         }
4478         return ret;
4479 }
4480
4481 /*
4482  * This is somewhat expensive, updating the tree every time the
4483  * inode changes.  But, it is most likely to find the inode in cache.
4484  * FIXME, needs more benchmarking...there are no reasons other than performance
4485  * to keep or drop this code.
4486  */
4487 int btrfs_dirty_inode(struct inode *inode)
4488 {
4489         struct btrfs_root *root = BTRFS_I(inode)->root;
4490         struct btrfs_trans_handle *trans;
4491         int ret;
4492
4493         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4494                 return 0;
4495
4496         trans = btrfs_join_transaction(root);
4497         if (IS_ERR(trans))
4498                 return PTR_ERR(trans);
4499
4500         ret = btrfs_update_inode(trans, root, inode);
4501         if (ret && ret == -ENOSPC) {
4502                 /* whoops, lets try again with the full transaction */
4503                 btrfs_end_transaction(trans, root);
4504                 trans = btrfs_start_transaction(root, 1);
4505                 if (IS_ERR(trans))
4506                         return PTR_ERR(trans);
4507
4508                 ret = btrfs_update_inode(trans, root, inode);
4509         }
4510         btrfs_end_transaction(trans, root);
4511         if (BTRFS_I(inode)->delayed_node)
4512                 btrfs_balance_delayed_items(root);
4513
4514         return ret;
4515 }
4516
4517 /*
4518  * This is a copy of file_update_time.  We need this so we can return error on
4519  * ENOSPC for updating the inode in the case of file write and mmap writes.
4520  */
4521 static int btrfs_update_time(struct inode *inode, struct timespec *now,
4522                              int flags)
4523 {
4524         struct btrfs_root *root = BTRFS_I(inode)->root;
4525
4526         if (btrfs_root_readonly(root))
4527                 return -EROFS;
4528
4529         if (flags & S_VERSION)
4530                 inode_inc_iversion(inode);
4531         if (flags & S_CTIME)
4532                 inode->i_ctime = *now;
4533         if (flags & S_MTIME)
4534                 inode->i_mtime = *now;
4535         if (flags & S_ATIME)
4536                 inode->i_atime = *now;
4537         return btrfs_dirty_inode(inode);
4538 }
4539
4540 /*
4541  * find the highest existing sequence number in a directory
4542  * and then set the in-memory index_cnt variable to reflect
4543  * free sequence numbers
4544  */
4545 static int btrfs_set_inode_index_count(struct inode *inode)
4546 {
4547         struct btrfs_root *root = BTRFS_I(inode)->root;
4548         struct btrfs_key key, found_key;
4549         struct btrfs_path *path;
4550         struct extent_buffer *leaf;
4551         int ret;
4552
4553         key.objectid = btrfs_ino(inode);
4554         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4555         key.offset = (u64)-1;
4556
4557         path = btrfs_alloc_path();
4558         if (!path)
4559                 return -ENOMEM;
4560
4561         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4562         if (ret < 0)
4563                 goto out;
4564         /* FIXME: we should be able to handle this */
4565         if (ret == 0)
4566                 goto out;
4567         ret = 0;
4568
4569         /*
4570          * MAGIC NUMBER EXPLANATION:
4571          * since we search a directory based on f_pos we have to start at 2
4572          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4573          * else has to start at 2
4574          */
4575         if (path->slots[0] == 0) {
4576                 BTRFS_I(inode)->index_cnt = 2;
4577                 goto out;
4578         }
4579
4580         path->slots[0]--;
4581
4582         leaf = path->nodes[0];
4583         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4584
4585         if (found_key.objectid != btrfs_ino(inode) ||
4586             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4587                 BTRFS_I(inode)->index_cnt = 2;
4588                 goto out;
4589         }
4590
4591         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4592 out:
4593         btrfs_free_path(path);
4594         return ret;
4595 }
4596
4597 /*
4598  * helper to find a free sequence number in a given directory.  This current
4599  * code is very simple, later versions will do smarter things in the btree
4600  */
4601 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4602 {
4603         int ret = 0;
4604
4605         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4606                 ret = btrfs_inode_delayed_dir_index_count(dir);
4607                 if (ret) {
4608                         ret = btrfs_set_inode_index_count(dir);
4609                         if (ret)
4610                                 return ret;
4611                 }
4612         }
4613
4614         *index = BTRFS_I(dir)->index_cnt;
4615         BTRFS_I(dir)->index_cnt++;
4616
4617         return ret;
4618 }
4619
4620 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4621                                      struct btrfs_root *root,
4622                                      struct inode *dir,
4623                                      const char *name, int name_len,
4624                                      u64 ref_objectid, u64 objectid,
4625                                      umode_t mode, u64 *index)
4626 {
4627         struct inode *inode;
4628         struct btrfs_inode_item *inode_item;
4629         struct btrfs_key *location;
4630         struct btrfs_path *path;
4631         struct btrfs_inode_ref *ref;
4632         struct btrfs_key key[2];
4633         u32 sizes[2];
4634         unsigned long ptr;
4635         int ret;
4636         int owner;
4637
4638         path = btrfs_alloc_path();
4639         if (!path)
4640                 return ERR_PTR(-ENOMEM);
4641
4642         inode = new_inode(root->fs_info->sb);
4643         if (!inode) {
4644                 btrfs_free_path(path);
4645                 return ERR_PTR(-ENOMEM);
4646         }
4647
4648         /*
4649          * we have to initialize this early, so we can reclaim the inode
4650          * number if we fail afterwards in this function.
4651          */
4652         inode->i_ino = objectid;
4653
4654         if (dir) {
4655                 trace_btrfs_inode_request(dir);
4656
4657                 ret = btrfs_set_inode_index(dir, index);
4658                 if (ret) {
4659                         btrfs_free_path(path);
4660                         iput(inode);
4661                         return ERR_PTR(ret);
4662                 }
4663         }
4664         /*
4665          * index_cnt is ignored for everything but a dir,
4666          * btrfs_get_inode_index_count has an explanation for the magic
4667          * number
4668          */
4669         BTRFS_I(inode)->index_cnt = 2;
4670         BTRFS_I(inode)->root = root;
4671         BTRFS_I(inode)->generation = trans->transid;
4672         inode->i_generation = BTRFS_I(inode)->generation;
4673
4674         if (S_ISDIR(mode))
4675                 owner = 0;
4676         else
4677                 owner = 1;
4678
4679         key[0].objectid = objectid;
4680         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4681         key[0].offset = 0;
4682
4683         key[1].objectid = objectid;
4684         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4685         key[1].offset = ref_objectid;
4686
4687         sizes[0] = sizeof(struct btrfs_inode_item);
4688         sizes[1] = name_len + sizeof(*ref);
4689
4690         path->leave_spinning = 1;
4691         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4692         if (ret != 0)
4693                 goto fail;
4694
4695         inode_init_owner(inode, dir, mode);
4696         inode_set_bytes(inode, 0);
4697         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4698         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4699                                   struct btrfs_inode_item);
4700         memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
4701                              sizeof(*inode_item));
4702         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4703
4704         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4705                              struct btrfs_inode_ref);
4706         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4707         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4708         ptr = (unsigned long)(ref + 1);
4709         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4710
4711         btrfs_mark_buffer_dirty(path->nodes[0]);
4712         btrfs_free_path(path);
4713
4714         location = &BTRFS_I(inode)->location;
4715         location->objectid = objectid;
4716         location->offset = 0;
4717         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4718
4719         btrfs_inherit_iflags(inode, dir);
4720
4721         if (S_ISREG(mode)) {
4722                 if (btrfs_test_opt(root, NODATASUM))
4723                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4724                 if (btrfs_test_opt(root, NODATACOW) ||
4725                     (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4726                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4727         }
4728
4729         insert_inode_hash(inode);
4730         inode_tree_add(inode);
4731
4732         trace_btrfs_inode_new(inode);
4733         btrfs_set_inode_last_trans(trans, inode);
4734
4735         btrfs_update_root_times(trans, root);
4736
4737         return inode;
4738 fail:
4739         if (dir)
4740                 BTRFS_I(dir)->index_cnt--;
4741         btrfs_free_path(path);
4742         iput(inode);
4743         return ERR_PTR(ret);
4744 }
4745
4746 static inline u8 btrfs_inode_type(struct inode *inode)
4747 {
4748         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4749 }
4750
4751 /*
4752  * utility function to add 'inode' into 'parent_inode' with
4753  * a give name and a given sequence number.
4754  * if 'add_backref' is true, also insert a backref from the
4755  * inode to the parent directory.
4756  */
4757 int btrfs_add_link(struct btrfs_trans_handle *trans,
4758                    struct inode *parent_inode, struct inode *inode,
4759                    const char *name, int name_len, int add_backref, u64 index)
4760 {
4761         int ret = 0;
4762         struct btrfs_key key;
4763         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4764         u64 ino = btrfs_ino(inode);
4765         u64 parent_ino = btrfs_ino(parent_inode);
4766
4767         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4768                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4769         } else {
4770                 key.objectid = ino;
4771                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4772                 key.offset = 0;
4773         }
4774
4775         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4776                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4777                                          key.objectid, root->root_key.objectid,
4778                                          parent_ino, index, name, name_len);
4779         } else if (add_backref) {
4780                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4781                                              parent_ino, index);
4782         }
4783
4784         /* Nothing to clean up yet */
4785         if (ret)
4786                 return ret;
4787
4788         ret = btrfs_insert_dir_item(trans, root, name, name_len,
4789                                     parent_inode, &key,
4790                                     btrfs_inode_type(inode), index);
4791         if (ret == -EEXIST)
4792                 goto fail_dir_item;
4793         else if (ret) {
4794                 btrfs_abort_transaction(trans, root, ret);
4795                 return ret;
4796         }
4797
4798         btrfs_i_size_write(parent_inode, parent_inode->i_size +
4799                            name_len * 2);
4800         inode_inc_iversion(parent_inode);
4801         parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4802         ret = btrfs_update_inode(trans, root, parent_inode);
4803         if (ret)
4804                 btrfs_abort_transaction(trans, root, ret);
4805         return ret;
4806
4807 fail_dir_item:
4808         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4809                 u64 local_index;
4810                 int err;
4811                 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4812                                  key.objectid, root->root_key.objectid,
4813                                  parent_ino, &local_index, name, name_len);
4814
4815         } else if (add_backref) {
4816                 u64 local_index;
4817                 int err;
4818
4819                 err = btrfs_del_inode_ref(trans, root, name, name_len,
4820                                           ino, parent_ino, &local_index);
4821         }
4822         return ret;
4823 }
4824
4825 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4826                             struct inode *dir, struct dentry *dentry,
4827                             struct inode *inode, int backref, u64 index)
4828 {
4829         int err = btrfs_add_link(trans, dir, inode,
4830                                  dentry->d_name.name, dentry->d_name.len,
4831                                  backref, index);
4832         if (err > 0)
4833                 err = -EEXIST;
4834         return err;
4835 }
4836
4837 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4838                         umode_t mode, dev_t rdev)
4839 {
4840         struct btrfs_trans_handle *trans;
4841         struct btrfs_root *root = BTRFS_I(dir)->root;
4842         struct inode *inode = NULL;
4843         int err;
4844         int drop_inode = 0;
4845         u64 objectid;
4846         unsigned long nr = 0;
4847         u64 index = 0;
4848
4849         if (!new_valid_dev(rdev))
4850                 return -EINVAL;
4851
4852         /*
4853          * 2 for inode item and ref
4854          * 2 for dir items
4855          * 1 for xattr if selinux is on
4856          */
4857         trans = btrfs_start_transaction(root, 5);
4858         if (IS_ERR(trans))
4859                 return PTR_ERR(trans);
4860
4861         err = btrfs_find_free_ino(root, &objectid);
4862         if (err)
4863                 goto out_unlock;
4864
4865         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4866                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4867                                 mode, &index);
4868         if (IS_ERR(inode)) {
4869                 err = PTR_ERR(inode);
4870                 goto out_unlock;
4871         }
4872
4873         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4874         if (err) {
4875                 drop_inode = 1;
4876                 goto out_unlock;
4877         }
4878
4879         /*
4880         * If the active LSM wants to access the inode during
4881         * d_instantiate it needs these. Smack checks to see
4882         * if the filesystem supports xattrs by looking at the
4883         * ops vector.
4884         */
4885
4886         inode->i_op = &btrfs_special_inode_operations;
4887         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4888         if (err)
4889                 drop_inode = 1;
4890         else {
4891                 init_special_inode(inode, inode->i_mode, rdev);
4892                 btrfs_update_inode(trans, root, inode);
4893                 d_instantiate(dentry, inode);
4894         }
4895 out_unlock:
4896         nr = trans->blocks_used;
4897         btrfs_end_transaction(trans, root);
4898         btrfs_btree_balance_dirty(root, nr);
4899         if (drop_inode) {
4900                 inode_dec_link_count(inode);
4901                 iput(inode);
4902         }
4903         return err;
4904 }
4905
4906 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4907                         umode_t mode, bool excl)
4908 {
4909         struct btrfs_trans_handle *trans;
4910         struct btrfs_root *root = BTRFS_I(dir)->root;
4911         struct inode *inode = NULL;
4912         int drop_inode = 0;
4913         int err;
4914         unsigned long nr = 0;
4915         u64 objectid;
4916         u64 index = 0;
4917
4918         /*
4919          * 2 for inode item and ref
4920          * 2 for dir items
4921          * 1 for xattr if selinux is on
4922          */
4923         trans = btrfs_start_transaction(root, 5);
4924         if (IS_ERR(trans))
4925                 return PTR_ERR(trans);
4926
4927         err = btrfs_find_free_ino(root, &objectid);
4928         if (err)
4929                 goto out_unlock;
4930
4931         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4932                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4933                                 mode, &index);
4934         if (IS_ERR(inode)) {
4935                 err = PTR_ERR(inode);
4936                 goto out_unlock;
4937         }
4938
4939         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4940         if (err) {
4941                 drop_inode = 1;
4942                 goto out_unlock;
4943         }
4944
4945         /*
4946         * If the active LSM wants to access the inode during
4947         * d_instantiate it needs these. Smack checks to see
4948         * if the filesystem supports xattrs by looking at the
4949         * ops vector.
4950         */
4951         inode->i_fop = &btrfs_file_operations;
4952         inode->i_op = &btrfs_file_inode_operations;
4953
4954         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4955         if (err)
4956                 drop_inode = 1;
4957         else {
4958                 inode->i_mapping->a_ops = &btrfs_aops;
4959                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4960                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4961                 d_instantiate(dentry, inode);
4962         }
4963 out_unlock:
4964         nr = trans->blocks_used;
4965         btrfs_end_transaction(trans, root);
4966         if (drop_inode) {
4967                 inode_dec_link_count(inode);
4968                 iput(inode);
4969         }
4970         btrfs_btree_balance_dirty(root, nr);
4971         return err;
4972 }
4973
4974 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4975                       struct dentry *dentry)
4976 {
4977         struct btrfs_trans_handle *trans;
4978         struct btrfs_root *root = BTRFS_I(dir)->root;
4979         struct inode *inode = old_dentry->d_inode;
4980         u64 index;
4981         unsigned long nr = 0;
4982         int err;
4983         int drop_inode = 0;
4984
4985         /* do not allow sys_link's with other subvols of the same device */
4986         if (root->objectid != BTRFS_I(inode)->root->objectid)
4987                 return -EXDEV;
4988
4989         if (inode->i_nlink == ~0U)
4990                 return -EMLINK;
4991
4992         err = btrfs_set_inode_index(dir, &index);
4993         if (err)
4994                 goto fail;
4995
4996         /*
4997          * 2 items for inode and inode ref
4998          * 2 items for dir items
4999          * 1 item for parent inode
5000          */
5001         trans = btrfs_start_transaction(root, 5);
5002         if (IS_ERR(trans)) {
5003                 err = PTR_ERR(trans);
5004                 goto fail;
5005         }
5006
5007         btrfs_inc_nlink(inode);
5008         inode_inc_iversion(inode);
5009         inode->i_ctime = CURRENT_TIME;
5010         ihold(inode);
5011
5012         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5013
5014         if (err) {
5015                 drop_inode = 1;
5016         } else {
5017                 struct dentry *parent = dentry->d_parent;
5018                 err = btrfs_update_inode(trans, root, inode);
5019                 if (err)
5020                         goto fail;
5021                 d_instantiate(dentry, inode);
5022                 btrfs_log_new_name(trans, inode, NULL, parent);
5023         }
5024
5025         nr = trans->blocks_used;
5026         btrfs_end_transaction(trans, root);
5027 fail:
5028         if (drop_inode) {
5029                 inode_dec_link_count(inode);
5030                 iput(inode);
5031         }
5032         btrfs_btree_balance_dirty(root, nr);
5033         return err;
5034 }
5035
5036 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5037 {
5038         struct inode *inode = NULL;
5039         struct btrfs_trans_handle *trans;
5040         struct btrfs_root *root = BTRFS_I(dir)->root;
5041         int err = 0;
5042         int drop_on_err = 0;
5043         u64 objectid = 0;
5044         u64 index = 0;
5045         unsigned long nr = 1;
5046
5047         /*
5048          * 2 items for inode and ref
5049          * 2 items for dir items
5050          * 1 for xattr if selinux is on
5051          */
5052         trans = btrfs_start_transaction(root, 5);
5053         if (IS_ERR(trans))
5054                 return PTR_ERR(trans);
5055
5056         err = btrfs_find_free_ino(root, &objectid);
5057         if (err)
5058                 goto out_fail;
5059
5060         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5061                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5062                                 S_IFDIR | mode, &index);
5063         if (IS_ERR(inode)) {
5064                 err = PTR_ERR(inode);
5065                 goto out_fail;
5066         }
5067
5068         drop_on_err = 1;
5069
5070         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5071         if (err)
5072                 goto out_fail;
5073
5074         inode->i_op = &btrfs_dir_inode_operations;
5075         inode->i_fop = &btrfs_dir_file_operations;
5076
5077         btrfs_i_size_write(inode, 0);
5078         err = btrfs_update_inode(trans, root, inode);
5079         if (err)
5080                 goto out_fail;
5081
5082         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5083                              dentry->d_name.len, 0, index);
5084         if (err)
5085                 goto out_fail;
5086
5087         d_instantiate(dentry, inode);
5088         drop_on_err = 0;
5089
5090 out_fail:
5091         nr = trans->blocks_used;
5092         btrfs_end_transaction(trans, root);
5093         if (drop_on_err)
5094                 iput(inode);
5095         btrfs_btree_balance_dirty(root, nr);
5096         return err;
5097 }
5098
5099 /* helper for btfs_get_extent.  Given an existing extent in the tree,
5100  * and an extent that you want to insert, deal with overlap and insert
5101  * the new extent into the tree.
5102  */
5103 static int merge_extent_mapping(struct extent_map_tree *em_tree,
5104                                 struct extent_map *existing,
5105                                 struct extent_map *em,
5106                                 u64 map_start, u64 map_len)
5107 {
5108         u64 start_diff;
5109
5110         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5111         start_diff = map_start - em->start;
5112         em->start = map_start;
5113         em->len = map_len;
5114         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5115             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
5116                 em->block_start += start_diff;
5117                 em->block_len -= start_diff;
5118         }
5119         return add_extent_mapping(em_tree, em);
5120 }
5121
5122 static noinline int uncompress_inline(struct btrfs_path *path,
5123                                       struct inode *inode, struct page *page,
5124                                       size_t pg_offset, u64 extent_offset,
5125                                       struct btrfs_file_extent_item *item)
5126 {
5127         int ret;
5128         struct extent_buffer *leaf = path->nodes[0];
5129         char *tmp;
5130         size_t max_size;
5131         unsigned long inline_size;
5132         unsigned long ptr;
5133         int compress_type;
5134
5135         WARN_ON(pg_offset != 0);
5136         compress_type = btrfs_file_extent_compression(leaf, item);
5137         max_size = btrfs_file_extent_ram_bytes(leaf, item);
5138         inline_size = btrfs_file_extent_inline_item_len(leaf,
5139                                         btrfs_item_nr(leaf, path->slots[0]));
5140         tmp = kmalloc(inline_size, GFP_NOFS);
5141         if (!tmp)
5142                 return -ENOMEM;
5143         ptr = btrfs_file_extent_inline_start(item);
5144
5145         read_extent_buffer(leaf, tmp, ptr, inline_size);
5146
5147         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5148         ret = btrfs_decompress(compress_type, tmp, page,
5149                                extent_offset, inline_size, max_size);
5150         if (ret) {
5151                 char *kaddr = kmap_atomic(page);
5152                 unsigned long copy_size = min_t(u64,
5153                                   PAGE_CACHE_SIZE - pg_offset,
5154                                   max_size - extent_offset);
5155                 memset(kaddr + pg_offset, 0, copy_size);
5156                 kunmap_atomic(kaddr);
5157         }
5158         kfree(tmp);
5159         return 0;
5160 }
5161
5162 /*
5163  * a bit scary, this does extent mapping from logical file offset to the disk.
5164  * the ugly parts come from merging extents from the disk with the in-ram
5165  * representation.  This gets more complex because of the data=ordered code,
5166  * where the in-ram extents might be locked pending data=ordered completion.
5167  *
5168  * This also copies inline extents directly into the page.
5169  */
5170
5171 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5172                                     size_t pg_offset, u64 start, u64 len,
5173                                     int create)
5174 {
5175         int ret;
5176         int err = 0;
5177         u64 bytenr;
5178         u64 extent_start = 0;
5179         u64 extent_end = 0;
5180         u64 objectid = btrfs_ino(inode);
5181         u32 found_type;
5182         struct btrfs_path *path = NULL;
5183         struct btrfs_root *root = BTRFS_I(inode)->root;
5184         struct btrfs_file_extent_item *item;
5185         struct extent_buffer *leaf;
5186         struct btrfs_key found_key;
5187         struct extent_map *em = NULL;
5188         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5189         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5190         struct btrfs_trans_handle *trans = NULL;
5191         int compress_type;
5192
5193 again:
5194         read_lock(&em_tree->lock);
5195         em = lookup_extent_mapping(em_tree, start, len);
5196         if (em)
5197                 em->bdev = root->fs_info->fs_devices->latest_bdev;
5198         read_unlock(&em_tree->lock);
5199
5200         if (em) {
5201                 if (em->start > start || em->start + em->len <= start)
5202                         free_extent_map(em);
5203                 else if (em->block_start == EXTENT_MAP_INLINE && page)
5204                         free_extent_map(em);
5205                 else
5206                         goto out;
5207         }
5208         em = alloc_extent_map();
5209         if (!em) {
5210                 err = -ENOMEM;
5211                 goto out;
5212         }
5213         em->bdev = root->fs_info->fs_devices->latest_bdev;
5214         em->start = EXTENT_MAP_HOLE;
5215         em->orig_start = EXTENT_MAP_HOLE;
5216         em->len = (u64)-1;
5217         em->block_len = (u64)-1;
5218
5219         if (!path) {
5220                 path = btrfs_alloc_path();
5221                 if (!path) {
5222                         err = -ENOMEM;
5223                         goto out;
5224                 }
5225                 /*
5226                  * Chances are we'll be called again, so go ahead and do
5227                  * readahead
5228                  */
5229                 path->reada = 1;
5230         }
5231
5232         ret = btrfs_lookup_file_extent(trans, root, path,
5233                                        objectid, start, trans != NULL);
5234         if (ret < 0) {
5235                 err = ret;
5236                 goto out;
5237         }
5238
5239         if (ret != 0) {
5240                 if (path->slots[0] == 0)
5241                         goto not_found;
5242                 path->slots[0]--;
5243         }
5244
5245         leaf = path->nodes[0];
5246         item = btrfs_item_ptr(leaf, path->slots[0],
5247                               struct btrfs_file_extent_item);
5248         /* are we inside the extent that was found? */
5249         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5250         found_type = btrfs_key_type(&found_key);
5251         if (found_key.objectid != objectid ||
5252             found_type != BTRFS_EXTENT_DATA_KEY) {
5253                 goto not_found;
5254         }
5255
5256         found_type = btrfs_file_extent_type(leaf, item);
5257         extent_start = found_key.offset;
5258         compress_type = btrfs_file_extent_compression(leaf, item);
5259         if (found_type == BTRFS_FILE_EXTENT_REG ||
5260             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5261                 extent_end = extent_start +
5262                        btrfs_file_extent_num_bytes(leaf, item);
5263         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5264                 size_t size;
5265                 size = btrfs_file_extent_inline_len(leaf, item);
5266                 extent_end = (extent_start + size + root->sectorsize - 1) &
5267                         ~((u64)root->sectorsize - 1);
5268         }
5269
5270         if (start >= extent_end) {
5271                 path->slots[0]++;
5272                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5273                         ret = btrfs_next_leaf(root, path);
5274                         if (ret < 0) {
5275                                 err = ret;
5276                                 goto out;
5277                         }
5278                         if (ret > 0)
5279                                 goto not_found;
5280                         leaf = path->nodes[0];
5281                 }
5282                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5283                 if (found_key.objectid != objectid ||
5284                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5285                         goto not_found;
5286                 if (start + len <= found_key.offset)
5287                         goto not_found;
5288                 em->start = start;
5289                 em->len = found_key.offset - start;
5290                 goto not_found_em;
5291         }
5292
5293         if (found_type == BTRFS_FILE_EXTENT_REG ||
5294             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5295                 em->start = extent_start;
5296                 em->len = extent_end - extent_start;
5297                 em->orig_start = extent_start -
5298                                  btrfs_file_extent_offset(leaf, item);
5299                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5300                 if (bytenr == 0) {
5301                         em->block_start = EXTENT_MAP_HOLE;
5302                         goto insert;
5303                 }
5304                 if (compress_type != BTRFS_COMPRESS_NONE) {
5305                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5306                         em->compress_type = compress_type;
5307                         em->block_start = bytenr;
5308                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5309                                                                          item);
5310                 } else {
5311                         bytenr += btrfs_file_extent_offset(leaf, item);
5312                         em->block_start = bytenr;
5313                         em->block_len = em->len;
5314                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5315                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5316                 }
5317                 goto insert;
5318         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5319                 unsigned long ptr;
5320                 char *map;
5321                 size_t size;
5322                 size_t extent_offset;
5323                 size_t copy_size;
5324
5325                 em->block_start = EXTENT_MAP_INLINE;
5326                 if (!page || create) {
5327                         em->start = extent_start;
5328                         em->len = extent_end - extent_start;
5329                         goto out;
5330                 }
5331
5332                 size = btrfs_file_extent_inline_len(leaf, item);
5333                 extent_offset = page_offset(page) + pg_offset - extent_start;
5334                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5335                                 size - extent_offset);
5336                 em->start = extent_start + extent_offset;
5337                 em->len = (copy_size + root->sectorsize - 1) &
5338                         ~((u64)root->sectorsize - 1);
5339                 em->orig_start = EXTENT_MAP_INLINE;
5340                 if (compress_type) {
5341                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5342                         em->compress_type = compress_type;
5343                 }
5344                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5345                 if (create == 0 && !PageUptodate(page)) {
5346                         if (btrfs_file_extent_compression(leaf, item) !=
5347                             BTRFS_COMPRESS_NONE) {
5348                                 ret = uncompress_inline(path, inode, page,
5349                                                         pg_offset,
5350                                                         extent_offset, item);
5351                                 BUG_ON(ret); /* -ENOMEM */
5352                         } else {
5353                                 map = kmap(page);
5354                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5355                                                    copy_size);
5356                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5357                                         memset(map + pg_offset + copy_size, 0,
5358                                                PAGE_CACHE_SIZE - pg_offset -
5359                                                copy_size);
5360                                 }
5361                                 kunmap(page);
5362                         }
5363                         flush_dcache_page(page);
5364                 } else if (create && PageUptodate(page)) {
5365                         BUG();
5366                         if (!trans) {
5367                                 kunmap(page);
5368                                 free_extent_map(em);
5369                                 em = NULL;
5370
5371                                 btrfs_release_path(path);
5372                                 trans = btrfs_join_transaction(root);
5373
5374                                 if (IS_ERR(trans))
5375                                         return ERR_CAST(trans);
5376                                 goto again;
5377                         }
5378                         map = kmap(page);
5379                         write_extent_buffer(leaf, map + pg_offset, ptr,
5380                                             copy_size);
5381                         kunmap(page);
5382                         btrfs_mark_buffer_dirty(leaf);
5383                 }
5384                 set_extent_uptodate(io_tree, em->start,
5385                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
5386                 goto insert;
5387         } else {
5388                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5389                 WARN_ON(1);
5390         }
5391 not_found:
5392         em->start = start;
5393         em->len = len;
5394 not_found_em:
5395         em->block_start = EXTENT_MAP_HOLE;
5396         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5397 insert:
5398         btrfs_release_path(path);
5399         if (em->start > start || extent_map_end(em) <= start) {
5400                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5401                        "[%llu %llu]\n", (unsigned long long)em->start,
5402                        (unsigned long long)em->len,
5403                        (unsigned long long)start,
5404                        (unsigned long long)len);
5405                 err = -EIO;
5406                 goto out;
5407         }
5408
5409         err = 0;
5410         write_lock(&em_tree->lock);
5411         ret = add_extent_mapping(em_tree, em);
5412         /* it is possible that someone inserted the extent into the tree
5413          * while we had the lock dropped.  It is also possible that
5414          * an overlapping map exists in the tree
5415          */
5416         if (ret == -EEXIST) {
5417                 struct extent_map *existing;
5418
5419                 ret = 0;
5420
5421                 existing = lookup_extent_mapping(em_tree, start, len);
5422                 if (existing && (existing->start > start ||
5423                     existing->start + existing->len <= start)) {
5424                         free_extent_map(existing);
5425                         existing = NULL;
5426                 }
5427                 if (!existing) {
5428                         existing = lookup_extent_mapping(em_tree, em->start,
5429                                                          em->len);
5430                         if (existing) {
5431                                 err = merge_extent_mapping(em_tree, existing,
5432                                                            em, start,
5433                                                            root->sectorsize);
5434                                 free_extent_map(existing);
5435                                 if (err) {
5436                                         free_extent_map(em);
5437                                         em = NULL;
5438                                 }
5439                         } else {
5440                                 err = -EIO;
5441                                 free_extent_map(em);
5442                                 em = NULL;
5443                         }
5444                 } else {
5445                         free_extent_map(em);
5446                         em = existing;
5447                         err = 0;
5448                 }
5449         }
5450         write_unlock(&em_tree->lock);
5451 out:
5452
5453         trace_btrfs_get_extent(root, em);
5454
5455         if (path)
5456                 btrfs_free_path(path);
5457         if (trans) {
5458                 ret = btrfs_end_transaction(trans, root);
5459                 if (!err)
5460                         err = ret;
5461         }
5462         if (err) {
5463                 free_extent_map(em);
5464                 return ERR_PTR(err);
5465         }
5466         BUG_ON(!em); /* Error is always set */
5467         return em;
5468 }
5469
5470 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5471                                            size_t pg_offset, u64 start, u64 len,
5472                                            int create)
5473 {
5474         struct extent_map *em;
5475         struct extent_map *hole_em = NULL;
5476         u64 range_start = start;
5477         u64 end;
5478         u64 found;
5479         u64 found_end;
5480         int err = 0;
5481
5482         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5483         if (IS_ERR(em))
5484                 return em;
5485         if (em) {
5486                 /*
5487                  * if our em maps to a hole, there might
5488                  * actually be delalloc bytes behind it
5489                  */
5490                 if (em->block_start != EXTENT_MAP_HOLE)
5491                         return em;
5492                 else
5493                         hole_em = em;
5494         }
5495
5496         /* check to see if we've wrapped (len == -1 or similar) */
5497         end = start + len;
5498         if (end < start)
5499                 end = (u64)-1;
5500         else
5501                 end -= 1;
5502
5503         em = NULL;
5504
5505         /* ok, we didn't find anything, lets look for delalloc */
5506         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5507                                  end, len, EXTENT_DELALLOC, 1);
5508         found_end = range_start + found;
5509         if (found_end < range_start)
5510                 found_end = (u64)-1;
5511
5512         /*
5513          * we didn't find anything useful, return
5514          * the original results from get_extent()
5515          */
5516         if (range_start > end || found_end <= start) {
5517                 em = hole_em;
5518                 hole_em = NULL;
5519                 goto out;
5520         }
5521
5522         /* adjust the range_start to make sure it doesn't
5523          * go backwards from the start they passed in
5524          */
5525         range_start = max(start,range_start);
5526         found = found_end - range_start;
5527
5528         if (found > 0) {
5529                 u64 hole_start = start;
5530                 u64 hole_len = len;
5531
5532                 em = alloc_extent_map();
5533                 if (!em) {
5534                         err = -ENOMEM;
5535                         goto out;
5536                 }
5537                 /*
5538                  * when btrfs_get_extent can't find anything it
5539                  * returns one huge hole
5540                  *
5541                  * make sure what it found really fits our range, and
5542                  * adjust to make sure it is based on the start from
5543                  * the caller
5544                  */
5545                 if (hole_em) {
5546                         u64 calc_end = extent_map_end(hole_em);
5547
5548                         if (calc_end <= start || (hole_em->start > end)) {
5549                                 free_extent_map(hole_em);
5550                                 hole_em = NULL;
5551                         } else {
5552                                 hole_start = max(hole_em->start, start);
5553                                 hole_len = calc_end - hole_start;
5554                         }
5555                 }
5556                 em->bdev = NULL;
5557                 if (hole_em && range_start > hole_start) {
5558                         /* our hole starts before our delalloc, so we
5559                          * have to return just the parts of the hole
5560                          * that go until  the delalloc starts
5561                          */
5562                         em->len = min(hole_len,
5563                                       range_start - hole_start);
5564                         em->start = hole_start;
5565                         em->orig_start = hole_start;
5566                         /*
5567                          * don't adjust block start at all,
5568                          * it is fixed at EXTENT_MAP_HOLE
5569                          */
5570                         em->block_start = hole_em->block_start;
5571                         em->block_len = hole_len;
5572                 } else {
5573                         em->start = range_start;
5574                         em->len = found;
5575                         em->orig_start = range_start;
5576                         em->block_start = EXTENT_MAP_DELALLOC;
5577                         em->block_len = found;
5578                 }
5579         } else if (hole_em) {
5580                 return hole_em;
5581         }
5582 out:
5583
5584         free_extent_map(hole_em);
5585         if (err) {
5586                 free_extent_map(em);
5587                 return ERR_PTR(err);
5588         }
5589         return em;
5590 }
5591
5592 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5593                                                   struct extent_map *em,
5594                                                   u64 start, u64 len)
5595 {
5596         struct btrfs_root *root = BTRFS_I(inode)->root;
5597         struct btrfs_trans_handle *trans;
5598         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5599         struct btrfs_key ins;
5600         u64 alloc_hint;
5601         int ret;
5602         bool insert = false;
5603
5604         /*
5605          * Ok if the extent map we looked up is a hole and is for the exact
5606          * range we want, there is no reason to allocate a new one, however if
5607          * it is not right then we need to free this one and drop the cache for
5608          * our range.
5609          */
5610         if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5611             em->len != len) {
5612                 free_extent_map(em);
5613                 em = NULL;
5614                 insert = true;
5615                 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5616         }
5617
5618         trans = btrfs_join_transaction(root);
5619         if (IS_ERR(trans))
5620                 return ERR_CAST(trans);
5621
5622         if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5623                 btrfs_add_inode_defrag(trans, inode);
5624
5625         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5626
5627         alloc_hint = get_extent_allocation_hint(inode, start, len);
5628         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5629                                    alloc_hint, &ins, 1);
5630         if (ret) {
5631                 em = ERR_PTR(ret);
5632                 goto out;
5633         }
5634
5635         if (!em) {
5636                 em = alloc_extent_map();
5637                 if (!em) {
5638                         em = ERR_PTR(-ENOMEM);
5639                         goto out;
5640                 }
5641         }
5642
5643         em->start = start;
5644         em->orig_start = em->start;
5645         em->len = ins.offset;
5646
5647         em->block_start = ins.objectid;
5648         em->block_len = ins.offset;
5649         em->bdev = root->fs_info->fs_devices->latest_bdev;
5650
5651         /*
5652          * We need to do this because if we're using the original em we searched
5653          * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5654          */
5655         em->flags = 0;
5656         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5657
5658         while (insert) {
5659                 write_lock(&em_tree->lock);
5660                 ret = add_extent_mapping(em_tree, em);
5661                 write_unlock(&em_tree->lock);
5662                 if (ret != -EEXIST)
5663                         break;
5664                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5665         }
5666
5667         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5668                                            ins.offset, ins.offset, 0);
5669         if (ret) {
5670                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5671                 em = ERR_PTR(ret);
5672         }
5673 out:
5674         btrfs_end_transaction(trans, root);
5675         return em;
5676 }
5677
5678 /*
5679  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5680  * block must be cow'd
5681  */
5682 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5683                                       struct inode *inode, u64 offset, u64 len)
5684 {
5685         struct btrfs_path *path;
5686         int ret;
5687         struct extent_buffer *leaf;
5688         struct btrfs_root *root = BTRFS_I(inode)->root;
5689         struct btrfs_file_extent_item *fi;
5690         struct btrfs_key key;
5691         u64 disk_bytenr;
5692         u64 backref_offset;
5693         u64 extent_end;
5694         u64 num_bytes;
5695         int slot;
5696         int found_type;
5697
5698         path = btrfs_alloc_path();
5699         if (!path)
5700                 return -ENOMEM;
5701
5702         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5703                                        offset, 0);
5704         if (ret < 0)
5705                 goto out;
5706
5707         slot = path->slots[0];
5708         if (ret == 1) {
5709                 if (slot == 0) {
5710                         /* can't find the item, must cow */
5711                         ret = 0;
5712                         goto out;
5713                 }
5714                 slot--;
5715         }
5716         ret = 0;
5717         leaf = path->nodes[0];
5718         btrfs_item_key_to_cpu(leaf, &key, slot);
5719         if (key.objectid != btrfs_ino(inode) ||
5720             key.type != BTRFS_EXTENT_DATA_KEY) {
5721                 /* not our file or wrong item type, must cow */
5722                 goto out;
5723         }
5724
5725         if (key.offset > offset) {
5726                 /* Wrong offset, must cow */
5727                 goto out;
5728         }
5729
5730         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5731         found_type = btrfs_file_extent_type(leaf, fi);
5732         if (found_type != BTRFS_FILE_EXTENT_REG &&
5733             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5734                 /* not a regular extent, must cow */
5735                 goto out;
5736         }
5737         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5738         backref_offset = btrfs_file_extent_offset(leaf, fi);
5739
5740         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5741         if (extent_end < offset + len) {
5742                 /* extent doesn't include our full range, must cow */
5743                 goto out;
5744         }
5745
5746         if (btrfs_extent_readonly(root, disk_bytenr))
5747                 goto out;
5748
5749         /*
5750          * look for other files referencing this extent, if we
5751          * find any we must cow
5752          */
5753         if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5754                                   key.offset - backref_offset, disk_bytenr))
5755                 goto out;
5756
5757         /*
5758          * adjust disk_bytenr and num_bytes to cover just the bytes
5759          * in this extent we are about to write.  If there
5760          * are any csums in that range we have to cow in order
5761          * to keep the csums correct
5762          */
5763         disk_bytenr += backref_offset;
5764         disk_bytenr += offset - key.offset;
5765         num_bytes = min(offset + len, extent_end) - offset;
5766         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5767                                 goto out;
5768         /*
5769          * all of the above have passed, it is safe to overwrite this extent
5770          * without cow
5771          */
5772         ret = 1;
5773 out:
5774         btrfs_free_path(path);
5775         return ret;
5776 }
5777
5778 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
5779                               struct extent_state **cached_state, int writing)
5780 {
5781         struct btrfs_ordered_extent *ordered;
5782         int ret = 0;
5783
5784         while (1) {
5785                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
5786                                  0, cached_state);
5787                 /*
5788                  * We're concerned with the entire range that we're going to be
5789                  * doing DIO to, so we need to make sure theres no ordered
5790                  * extents in this range.
5791                  */
5792                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
5793                                                      lockend - lockstart + 1);
5794
5795                 /*
5796                  * We need to make sure there are no buffered pages in this
5797                  * range either, we could have raced between the invalidate in
5798                  * generic_file_direct_write and locking the extent.  The
5799                  * invalidate needs to happen so that reads after a write do not
5800                  * get stale data.
5801                  */
5802                 if (!ordered && (!writing ||
5803                     !test_range_bit(&BTRFS_I(inode)->io_tree,
5804                                     lockstart, lockend, EXTENT_UPTODATE, 0,
5805                                     *cached_state)))
5806                         break;
5807
5808                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
5809                                      cached_state, GFP_NOFS);
5810
5811                 if (ordered) {
5812                         btrfs_start_ordered_extent(inode, ordered, 1);
5813                         btrfs_put_ordered_extent(ordered);
5814                 } else {
5815                         /* Screw you mmap */
5816                         ret = filemap_write_and_wait_range(inode->i_mapping,
5817                                                            lockstart,
5818                                                            lockend);
5819                         if (ret)
5820                                 break;
5821
5822                         /*
5823                          * If we found a page that couldn't be invalidated just
5824                          * fall back to buffered.
5825                          */
5826                         ret = invalidate_inode_pages2_range(inode->i_mapping,
5827                                         lockstart >> PAGE_CACHE_SHIFT,
5828                                         lockend >> PAGE_CACHE_SHIFT);
5829                         if (ret)
5830                                 break;
5831                 }
5832
5833                 cond_resched();
5834         }
5835
5836         return ret;
5837 }
5838
5839 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5840                                    struct buffer_head *bh_result, int create)
5841 {
5842         struct extent_map *em;
5843         struct btrfs_root *root = BTRFS_I(inode)->root;
5844         struct extent_state *cached_state = NULL;
5845         u64 start = iblock << inode->i_blkbits;
5846         u64 lockstart, lockend;
5847         u64 len = bh_result->b_size;
5848         struct btrfs_trans_handle *trans;
5849         int unlock_bits = EXTENT_LOCKED;
5850         int ret;
5851
5852         if (create) {
5853                 ret = btrfs_delalloc_reserve_space(inode, len);
5854                 if (ret)
5855                         return ret;
5856                 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
5857         } else {
5858                 len = min_t(u64, len, root->sectorsize);
5859         }
5860
5861         lockstart = start;
5862         lockend = start + len - 1;
5863
5864         /*
5865          * If this errors out it's because we couldn't invalidate pagecache for
5866          * this range and we need to fallback to buffered.
5867          */
5868         if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
5869                 return -ENOTBLK;
5870
5871         if (create) {
5872                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
5873                                      lockend, EXTENT_DELALLOC, NULL,
5874                                      &cached_state, GFP_NOFS);
5875                 if (ret)
5876                         goto unlock_err;
5877         }
5878
5879         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5880         if (IS_ERR(em)) {
5881                 ret = PTR_ERR(em);
5882                 goto unlock_err;
5883         }
5884
5885         /*
5886          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5887          * io.  INLINE is special, and we could probably kludge it in here, but
5888          * it's still buffered so for safety lets just fall back to the generic
5889          * buffered path.
5890          *
5891          * For COMPRESSED we _have_ to read the entire extent in so we can
5892          * decompress it, so there will be buffering required no matter what we
5893          * do, so go ahead and fallback to buffered.
5894          *
5895          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5896          * to buffered IO.  Don't blame me, this is the price we pay for using
5897          * the generic code.
5898          */
5899         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5900             em->block_start == EXTENT_MAP_INLINE) {
5901                 free_extent_map(em);
5902                 ret = -ENOTBLK;
5903                 goto unlock_err;
5904         }
5905
5906         /* Just a good old fashioned hole, return */
5907         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5908                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5909                 free_extent_map(em);
5910                 ret = 0;
5911                 goto unlock_err;
5912         }
5913
5914         /*
5915          * We don't allocate a new extent in the following cases
5916          *
5917          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
5918          * existing extent.
5919          * 2) The extent is marked as PREALLOC.  We're good to go here and can
5920          * just use the extent.
5921          *
5922          */
5923         if (!create) {
5924                 len = min(len, em->len - (start - em->start));
5925                 lockstart = start + len;
5926                 goto unlock;
5927         }
5928
5929         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5930             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5931              em->block_start != EXTENT_MAP_HOLE)) {
5932                 int type;
5933                 int ret;
5934                 u64 block_start;
5935
5936                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5937                         type = BTRFS_ORDERED_PREALLOC;
5938                 else
5939                         type = BTRFS_ORDERED_NOCOW;
5940                 len = min(len, em->len - (start - em->start));
5941                 block_start = em->block_start + (start - em->start);
5942
5943                 /*
5944                  * we're not going to log anything, but we do need
5945                  * to make sure the current transaction stays open
5946                  * while we look for nocow cross refs
5947                  */
5948                 trans = btrfs_join_transaction(root);
5949                 if (IS_ERR(trans))
5950                         goto must_cow;
5951
5952                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5953                         ret = btrfs_add_ordered_extent_dio(inode, start,
5954                                            block_start, len, len, type);
5955                         btrfs_end_transaction(trans, root);
5956                         if (ret) {
5957                                 free_extent_map(em);
5958                                 goto unlock_err;
5959                         }
5960                         goto unlock;
5961                 }
5962                 btrfs_end_transaction(trans, root);
5963         }
5964 must_cow:
5965         /*
5966          * this will cow the extent, reset the len in case we changed
5967          * it above
5968          */
5969         len = bh_result->b_size;
5970         em = btrfs_new_extent_direct(inode, em, start, len);
5971         if (IS_ERR(em)) {
5972                 ret = PTR_ERR(em);
5973                 goto unlock_err;
5974         }
5975         len = min(len, em->len - (start - em->start));
5976 unlock:
5977         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5978                 inode->i_blkbits;
5979         bh_result->b_size = len;
5980         bh_result->b_bdev = em->bdev;
5981         set_buffer_mapped(bh_result);
5982         if (create) {
5983                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5984                         set_buffer_new(bh_result);
5985
5986                 /*
5987                  * Need to update the i_size under the extent lock so buffered
5988                  * readers will get the updated i_size when we unlock.
5989                  */
5990                 if (start + len > i_size_read(inode))
5991                         i_size_write(inode, start + len);
5992         }
5993
5994         /*
5995          * In the case of write we need to clear and unlock the entire range,
5996          * in the case of read we need to unlock only the end area that we
5997          * aren't using if there is any left over space.
5998          */
5999         if (lockstart < lockend) {
6000                 if (create && len < lockend - lockstart) {
6001                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6002                                          lockstart + len - 1, unlock_bits, 1, 0,
6003                                          &cached_state, GFP_NOFS);
6004                         /*
6005                          * Beside unlock, we also need to cleanup reserved space
6006                          * for the left range by attaching EXTENT_DO_ACCOUNTING.
6007                          */
6008                         clear_extent_bit(&BTRFS_I(inode)->io_tree,
6009                                          lockstart + len, lockend,
6010                                          unlock_bits | EXTENT_DO_ACCOUNTING,
6011                                          1, 0, NULL, GFP_NOFS);
6012                 } else {
6013                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6014                                          lockend, unlock_bits, 1, 0,
6015                                          &cached_state, GFP_NOFS);
6016                 }
6017         } else {
6018                 free_extent_state(cached_state);
6019         }
6020
6021         free_extent_map(em);
6022
6023         return 0;
6024
6025 unlock_err:
6026         if (create)
6027                 unlock_bits |= EXTENT_DO_ACCOUNTING;
6028
6029         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6030                          unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6031         return ret;
6032 }
6033
6034 struct btrfs_dio_private {
6035         struct inode *inode;
6036         u64 logical_offset;
6037         u64 disk_bytenr;
6038         u64 bytes;
6039         void *private;
6040
6041         /* number of bios pending for this dio */
6042         atomic_t pending_bios;
6043
6044         /* IO errors */
6045         int errors;
6046
6047         struct bio *orig_bio;
6048 };
6049
6050 static void btrfs_endio_direct_read(struct bio *bio, int err)
6051 {
6052         struct btrfs_dio_private *dip = bio->bi_private;
6053         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6054         struct bio_vec *bvec = bio->bi_io_vec;
6055         struct inode *inode = dip->inode;
6056         struct btrfs_root *root = BTRFS_I(inode)->root;
6057         u64 start;
6058
6059         start = dip->logical_offset;
6060         do {
6061                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6062                         struct page *page = bvec->bv_page;
6063                         char *kaddr;
6064                         u32 csum = ~(u32)0;
6065                         u64 private = ~(u32)0;
6066                         unsigned long flags;
6067
6068                         if (get_state_private(&BTRFS_I(inode)->io_tree,
6069                                               start, &private))
6070                                 goto failed;
6071                         local_irq_save(flags);
6072                         kaddr = kmap_atomic(page);
6073                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
6074                                                csum, bvec->bv_len);
6075                         btrfs_csum_final(csum, (char *)&csum);
6076                         kunmap_atomic(kaddr);
6077                         local_irq_restore(flags);
6078
6079                         flush_dcache_page(bvec->bv_page);
6080                         if (csum != private) {
6081 failed:
6082                                 printk(KERN_ERR "btrfs csum failed ino %llu off"
6083                                       " %llu csum %u private %u\n",
6084                                       (unsigned long long)btrfs_ino(inode),
6085                                       (unsigned long long)start,
6086                                       csum, (unsigned)private);
6087                                 err = -EIO;
6088                         }
6089                 }
6090
6091                 start += bvec->bv_len;
6092                 bvec++;
6093         } while (bvec <= bvec_end);
6094
6095         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
6096                       dip->logical_offset + dip->bytes - 1);
6097         bio->bi_private = dip->private;
6098
6099         kfree(dip);
6100
6101         /* If we had a csum failure make sure to clear the uptodate flag */
6102         if (err)
6103                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
6104         dio_end_io(bio, err);
6105 }
6106
6107 static void btrfs_endio_direct_write(struct bio *bio, int err)
6108 {
6109         struct btrfs_dio_private *dip = bio->bi_private;
6110         struct inode *inode = dip->inode;
6111         struct btrfs_root *root = BTRFS_I(inode)->root;
6112         struct btrfs_ordered_extent *ordered = NULL;
6113         u64 ordered_offset = dip->logical_offset;
6114         u64 ordered_bytes = dip->bytes;
6115         int ret;
6116
6117         if (err)
6118                 goto out_done;
6119 again:
6120         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
6121                                                    &ordered_offset,
6122                                                    ordered_bytes, !err);
6123         if (!ret)
6124                 goto out_test;
6125
6126         ordered->work.func = finish_ordered_fn;
6127         ordered->work.flags = 0;
6128         btrfs_queue_worker(&root->fs_info->endio_write_workers,
6129                            &ordered->work);
6130 out_test:
6131         /*
6132          * our bio might span multiple ordered extents.  If we haven't
6133          * completed the accounting for the whole dio, go back and try again
6134          */
6135         if (ordered_offset < dip->logical_offset + dip->bytes) {
6136                 ordered_bytes = dip->logical_offset + dip->bytes -
6137                         ordered_offset;
6138                 ordered = NULL;
6139                 goto again;
6140         }
6141 out_done:
6142         bio->bi_private = dip->private;
6143
6144         kfree(dip);
6145
6146         /* If we had an error make sure to clear the uptodate flag */
6147         if (err)
6148                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
6149         dio_end_io(bio, err);
6150 }
6151
6152 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
6153                                     struct bio *bio, int mirror_num,
6154                                     unsigned long bio_flags, u64 offset)
6155 {
6156         int ret;
6157         struct btrfs_root *root = BTRFS_I(inode)->root;
6158         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
6159         BUG_ON(ret); /* -ENOMEM */
6160         return 0;
6161 }
6162
6163 static void btrfs_end_dio_bio(struct bio *bio, int err)
6164 {
6165         struct btrfs_dio_private *dip = bio->bi_private;
6166
6167         if (err) {
6168                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
6169                       "sector %#Lx len %u err no %d\n",
6170                       (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
6171                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
6172                 dip->errors = 1;
6173
6174                 /*
6175                  * before atomic variable goto zero, we must make sure
6176                  * dip->errors is perceived to be set.
6177                  */
6178                 smp_mb__before_atomic_dec();
6179         }
6180
6181         /* if there are more bios still pending for this dio, just exit */
6182         if (!atomic_dec_and_test(&dip->pending_bios))
6183                 goto out;
6184
6185         if (dip->errors)
6186                 bio_io_error(dip->orig_bio);
6187         else {
6188                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
6189                 bio_endio(dip->orig_bio, 0);
6190         }
6191 out:
6192         bio_put(bio);
6193 }
6194
6195 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
6196                                        u64 first_sector, gfp_t gfp_flags)
6197 {
6198         int nr_vecs = bio_get_nr_vecs(bdev);
6199         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
6200 }
6201
6202 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
6203                                          int rw, u64 file_offset, int skip_sum,
6204                                          int async_submit)
6205 {
6206         int write = rw & REQ_WRITE;
6207         struct btrfs_root *root = BTRFS_I(inode)->root;
6208         int ret;
6209
6210         bio_get(bio);
6211
6212         if (!write) {
6213                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
6214                 if (ret)
6215                         goto err;
6216         }
6217
6218         if (skip_sum)
6219                 goto map;
6220
6221         if (write && async_submit) {
6222                 ret = btrfs_wq_submit_bio(root->fs_info,
6223                                    inode, rw, bio, 0, 0,
6224                                    file_offset,
6225                                    __btrfs_submit_bio_start_direct_io,
6226                                    __btrfs_submit_bio_done);
6227                 goto err;
6228         } else if (write) {
6229                 /*
6230                  * If we aren't doing async submit, calculate the csum of the
6231                  * bio now.
6232                  */
6233                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
6234                 if (ret)
6235                         goto err;
6236         } else if (!skip_sum) {
6237                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio, file_offset);
6238                 if (ret)
6239                         goto err;
6240         }
6241
6242 map:
6243         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
6244 err:
6245         bio_put(bio);
6246         return ret;
6247 }
6248
6249 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
6250                                     int skip_sum)
6251 {
6252         struct inode *inode = dip->inode;
6253         struct btrfs_root *root = BTRFS_I(inode)->root;
6254         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6255         struct bio *bio;
6256         struct bio *orig_bio = dip->orig_bio;
6257         struct bio_vec *bvec = orig_bio->bi_io_vec;
6258         u64 start_sector = orig_bio->bi_sector;
6259         u64 file_offset = dip->logical_offset;
6260         u64 submit_len = 0;
6261         u64 map_length;
6262         int nr_pages = 0;
6263         int ret = 0;
6264         int async_submit = 0;
6265
6266         map_length = orig_bio->bi_size;
6267         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6268                               &map_length, NULL, 0);
6269         if (ret) {
6270                 bio_put(orig_bio);
6271                 return -EIO;
6272         }
6273
6274         if (map_length >= orig_bio->bi_size) {
6275                 bio = orig_bio;
6276                 goto submit;
6277         }
6278
6279         async_submit = 1;
6280         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
6281         if (!bio)
6282                 return -ENOMEM;
6283         bio->bi_private = dip;
6284         bio->bi_end_io = btrfs_end_dio_bio;
6285         atomic_inc(&dip->pending_bios);
6286
6287         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6288                 if (unlikely(map_length < submit_len + bvec->bv_len ||
6289                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6290                                  bvec->bv_offset) < bvec->bv_len)) {
6291                         /*
6292                          * inc the count before we submit the bio so
6293                          * we know the end IO handler won't happen before
6294                          * we inc the count. Otherwise, the dip might get freed
6295                          * before we're done setting it up
6296                          */
6297                         atomic_inc(&dip->pending_bios);
6298                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
6299                                                      file_offset, skip_sum,
6300                                                      async_submit);
6301                         if (ret) {
6302                                 bio_put(bio);
6303                                 atomic_dec(&dip->pending_bios);
6304                                 goto out_err;
6305                         }
6306
6307                         start_sector += submit_len >> 9;
6308                         file_offset += submit_len;
6309
6310                         submit_len = 0;
6311                         nr_pages = 0;
6312
6313                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6314                                                   start_sector, GFP_NOFS);
6315                         if (!bio)
6316                                 goto out_err;
6317                         bio->bi_private = dip;
6318                         bio->bi_end_io = btrfs_end_dio_bio;
6319
6320                         map_length = orig_bio->bi_size;
6321                         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6322                                               &map_length, NULL, 0);
6323                         if (ret) {
6324                                 bio_put(bio);
6325                                 goto out_err;
6326                         }
6327                 } else {
6328                         submit_len += bvec->bv_len;
6329                         nr_pages ++;
6330                         bvec++;
6331                 }
6332         }
6333
6334 submit:
6335         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6336                                      async_submit);
6337         if (!ret)
6338                 return 0;
6339
6340         bio_put(bio);
6341 out_err:
6342         dip->errors = 1;
6343         /*
6344          * before atomic variable goto zero, we must
6345          * make sure dip->errors is perceived to be set.
6346          */
6347         smp_mb__before_atomic_dec();
6348         if (atomic_dec_and_test(&dip->pending_bios))
6349                 bio_io_error(dip->orig_bio);
6350
6351         /* bio_end_io() will handle error, so we needn't return it */
6352         return 0;
6353 }
6354
6355 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6356                                 loff_t file_offset)
6357 {
6358         struct btrfs_root *root = BTRFS_I(inode)->root;
6359         struct btrfs_dio_private *dip;
6360         struct bio_vec *bvec = bio->bi_io_vec;
6361         int skip_sum;
6362         int write = rw & REQ_WRITE;
6363         int ret = 0;
6364
6365         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6366
6367         dip = kmalloc(sizeof(*dip), GFP_NOFS);
6368         if (!dip) {
6369                 ret = -ENOMEM;
6370                 goto free_ordered;
6371         }
6372
6373         dip->private = bio->bi_private;
6374         dip->inode = inode;
6375         dip->logical_offset = file_offset;
6376
6377         dip->bytes = 0;
6378         do {
6379                 dip->bytes += bvec->bv_len;
6380                 bvec++;
6381         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6382
6383         dip->disk_bytenr = (u64)bio->bi_sector << 9;
6384         bio->bi_private = dip;
6385         dip->errors = 0;
6386         dip->orig_bio = bio;
6387         atomic_set(&dip->pending_bios, 0);
6388
6389         if (write)
6390                 bio->bi_end_io = btrfs_endio_direct_write;
6391         else
6392                 bio->bi_end_io = btrfs_endio_direct_read;
6393
6394         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6395         if (!ret)
6396                 return;
6397 free_ordered:
6398         /*
6399          * If this is a write, we need to clean up the reserved space and kill
6400          * the ordered extent.
6401          */
6402         if (write) {
6403                 struct btrfs_ordered_extent *ordered;
6404                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6405                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6406                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6407                         btrfs_free_reserved_extent(root, ordered->start,
6408                                                    ordered->disk_len);
6409                 btrfs_put_ordered_extent(ordered);
6410                 btrfs_put_ordered_extent(ordered);
6411         }
6412         bio_endio(bio, ret);
6413 }
6414
6415 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6416                         const struct iovec *iov, loff_t offset,
6417                         unsigned long nr_segs)
6418 {
6419         int seg;
6420         int i;
6421         size_t size;
6422         unsigned long addr;
6423         unsigned blocksize_mask = root->sectorsize - 1;
6424         ssize_t retval = -EINVAL;
6425         loff_t end = offset;
6426
6427         if (offset & blocksize_mask)
6428                 goto out;
6429
6430         /* Check the memory alignment.  Blocks cannot straddle pages */
6431         for (seg = 0; seg < nr_segs; seg++) {
6432                 addr = (unsigned long)iov[seg].iov_base;
6433                 size = iov[seg].iov_len;
6434                 end += size;
6435                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6436                         goto out;
6437
6438                 /* If this is a write we don't need to check anymore */
6439                 if (rw & WRITE)
6440                         continue;
6441
6442                 /*
6443                  * Check to make sure we don't have duplicate iov_base's in this
6444                  * iovec, if so return EINVAL, otherwise we'll get csum errors
6445                  * when reading back.
6446                  */
6447                 for (i = seg + 1; i < nr_segs; i++) {
6448                         if (iov[seg].iov_base == iov[i].iov_base)
6449                                 goto out;
6450                 }
6451         }
6452         retval = 0;
6453 out:
6454         return retval;
6455 }
6456
6457 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6458                         const struct iovec *iov, loff_t offset,
6459                         unsigned long nr_segs)
6460 {
6461         struct file *file = iocb->ki_filp;
6462         struct inode *inode = file->f_mapping->host;
6463
6464         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6465                             offset, nr_segs))
6466                 return 0;
6467
6468         return __blockdev_direct_IO(rw, iocb, inode,
6469                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6470                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6471                    btrfs_submit_direct, 0);
6472 }
6473
6474 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6475                 __u64 start, __u64 len)
6476 {
6477         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6478 }
6479
6480 int btrfs_readpage(struct file *file, struct page *page)
6481 {
6482         struct extent_io_tree *tree;
6483         tree = &BTRFS_I(page->mapping->host)->io_tree;
6484         return extent_read_full_page(tree, page, btrfs_get_extent, 0);
6485 }
6486
6487 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6488 {
6489         struct extent_io_tree *tree;
6490
6491
6492         if (current->flags & PF_MEMALLOC) {
6493                 redirty_page_for_writepage(wbc, page);
6494                 unlock_page(page);
6495                 return 0;
6496         }
6497         tree = &BTRFS_I(page->mapping->host)->io_tree;
6498         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6499 }
6500
6501 int btrfs_writepages(struct address_space *mapping,
6502                      struct writeback_control *wbc)
6503 {
6504         struct extent_io_tree *tree;
6505
6506         tree = &BTRFS_I(mapping->host)->io_tree;
6507         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6508 }
6509
6510 static int
6511 btrfs_readpages(struct file *file, struct address_space *mapping,
6512                 struct list_head *pages, unsigned nr_pages)
6513 {
6514         struct extent_io_tree *tree;
6515         tree = &BTRFS_I(mapping->host)->io_tree;
6516         return extent_readpages(tree, mapping, pages, nr_pages,
6517                                 btrfs_get_extent);
6518 }
6519 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6520 {
6521         struct extent_io_tree *tree;
6522         struct extent_map_tree *map;
6523         int ret;
6524
6525         tree = &BTRFS_I(page->mapping->host)->io_tree;
6526         map = &BTRFS_I(page->mapping->host)->extent_tree;
6527         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6528         if (ret == 1) {
6529                 ClearPagePrivate(page);
6530                 set_page_private(page, 0);
6531                 page_cache_release(page);
6532         }
6533         return ret;
6534 }
6535
6536 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6537 {
6538         if (PageWriteback(page) || PageDirty(page))
6539                 return 0;
6540         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6541 }
6542
6543 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6544 {
6545         struct inode *inode = page->mapping->host;
6546         struct extent_io_tree *tree;
6547         struct btrfs_ordered_extent *ordered;
6548         struct extent_state *cached_state = NULL;
6549         u64 page_start = page_offset(page);
6550         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6551
6552         /*
6553          * we have the page locked, so new writeback can't start,
6554          * and the dirty bit won't be cleared while we are here.
6555          *
6556          * Wait for IO on this page so that we can safely clear
6557          * the PagePrivate2 bit and do ordered accounting
6558          */
6559         wait_on_page_writeback(page);
6560
6561         tree = &BTRFS_I(inode)->io_tree;
6562         if (offset) {
6563                 btrfs_releasepage(page, GFP_NOFS);
6564                 return;
6565         }
6566         lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6567         ordered = btrfs_lookup_ordered_extent(inode,
6568                                            page_offset(page));
6569         if (ordered) {
6570                 /*
6571                  * IO on this page will never be started, so we need
6572                  * to account for any ordered extents now
6573                  */
6574                 clear_extent_bit(tree, page_start, page_end,
6575                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6576                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6577                                  &cached_state, GFP_NOFS);
6578                 /*
6579                  * whoever cleared the private bit is responsible
6580                  * for the finish_ordered_io
6581                  */
6582                 if (TestClearPagePrivate2(page) &&
6583                     btrfs_dec_test_ordered_pending(inode, &ordered, page_start,
6584                                                    PAGE_CACHE_SIZE, 1)) {
6585                         btrfs_finish_ordered_io(ordered);
6586                 }
6587                 btrfs_put_ordered_extent(ordered);
6588                 cached_state = NULL;
6589                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6590         }
6591         clear_extent_bit(tree, page_start, page_end,
6592                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6593                  EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6594         __btrfs_releasepage(page, GFP_NOFS);
6595
6596         ClearPageChecked(page);
6597         if (PagePrivate(page)) {
6598                 ClearPagePrivate(page);
6599                 set_page_private(page, 0);
6600                 page_cache_release(page);
6601         }
6602 }
6603
6604 /*
6605  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6606  * called from a page fault handler when a page is first dirtied. Hence we must
6607  * be careful to check for EOF conditions here. We set the page up correctly
6608  * for a written page which means we get ENOSPC checking when writing into
6609  * holes and correct delalloc and unwritten extent mapping on filesystems that
6610  * support these features.
6611  *
6612  * We are not allowed to take the i_mutex here so we have to play games to
6613  * protect against truncate races as the page could now be beyond EOF.  Because
6614  * vmtruncate() writes the inode size before removing pages, once we have the
6615  * page lock we can determine safely if the page is beyond EOF. If it is not
6616  * beyond EOF, then the page is guaranteed safe against truncation until we
6617  * unlock the page.
6618  */
6619 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6620 {
6621         struct page *page = vmf->page;
6622         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6623         struct btrfs_root *root = BTRFS_I(inode)->root;
6624         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6625         struct btrfs_ordered_extent *ordered;
6626         struct extent_state *cached_state = NULL;
6627         char *kaddr;
6628         unsigned long zero_start;
6629         loff_t size;
6630         int ret;
6631         int reserved = 0;
6632         u64 page_start;
6633         u64 page_end;
6634
6635         sb_start_pagefault(inode->i_sb);
6636         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6637         if (!ret) {
6638                 ret = file_update_time(vma->vm_file);
6639                 reserved = 1;
6640         }
6641         if (ret) {
6642                 if (ret == -ENOMEM)
6643                         ret = VM_FAULT_OOM;
6644                 else /* -ENOSPC, -EIO, etc */
6645                         ret = VM_FAULT_SIGBUS;
6646                 if (reserved)
6647                         goto out;
6648                 goto out_noreserve;
6649         }
6650
6651         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6652 again:
6653         lock_page(page);
6654         size = i_size_read(inode);
6655         page_start = page_offset(page);
6656         page_end = page_start + PAGE_CACHE_SIZE - 1;
6657
6658         if ((page->mapping != inode->i_mapping) ||
6659             (page_start >= size)) {
6660                 /* page got truncated out from underneath us */
6661                 goto out_unlock;
6662         }
6663         wait_on_page_writeback(page);
6664
6665         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
6666         set_page_extent_mapped(page);
6667
6668         /*
6669          * we can't set the delalloc bits if there are pending ordered
6670          * extents.  Drop our locks and wait for them to finish
6671          */
6672         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6673         if (ordered) {
6674                 unlock_extent_cached(io_tree, page_start, page_end,
6675                                      &cached_state, GFP_NOFS);
6676                 unlock_page(page);
6677                 btrfs_start_ordered_extent(inode, ordered, 1);
6678                 btrfs_put_ordered_extent(ordered);
6679                 goto again;
6680         }
6681
6682         /*
6683          * XXX - page_mkwrite gets called every time the page is dirtied, even
6684          * if it was already dirty, so for space accounting reasons we need to
6685          * clear any delalloc bits for the range we are fixing to save.  There
6686          * is probably a better way to do this, but for now keep consistent with
6687          * prepare_pages in the normal write path.
6688          */
6689         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6690                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6691                           0, 0, &cached_state, GFP_NOFS);
6692
6693         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6694                                         &cached_state);
6695         if (ret) {
6696                 unlock_extent_cached(io_tree, page_start, page_end,
6697                                      &cached_state, GFP_NOFS);
6698                 ret = VM_FAULT_SIGBUS;
6699                 goto out_unlock;
6700         }
6701         ret = 0;
6702
6703         /* page is wholly or partially inside EOF */
6704         if (page_start + PAGE_CACHE_SIZE > size)
6705                 zero_start = size & ~PAGE_CACHE_MASK;
6706         else
6707                 zero_start = PAGE_CACHE_SIZE;
6708
6709         if (zero_start != PAGE_CACHE_SIZE) {
6710                 kaddr = kmap(page);
6711                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6712                 flush_dcache_page(page);
6713                 kunmap(page);
6714         }
6715         ClearPageChecked(page);
6716         set_page_dirty(page);
6717         SetPageUptodate(page);
6718
6719         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6720         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6721
6722         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6723
6724 out_unlock:
6725         if (!ret) {
6726                 sb_end_pagefault(inode->i_sb);
6727                 return VM_FAULT_LOCKED;
6728         }
6729         unlock_page(page);
6730 out:
6731         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6732 out_noreserve:
6733         sb_end_pagefault(inode->i_sb);
6734         return ret;
6735 }
6736
6737 static int btrfs_truncate(struct inode *inode)
6738 {
6739         struct btrfs_root *root = BTRFS_I(inode)->root;
6740         struct btrfs_block_rsv *rsv;
6741         int ret;
6742         int err = 0;
6743         struct btrfs_trans_handle *trans;
6744         unsigned long nr;
6745         u64 mask = root->sectorsize - 1;
6746         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
6747
6748         ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6749         if (ret)
6750                 return ret;
6751
6752         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6753         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6754
6755         /*
6756          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
6757          * 3 things going on here
6758          *
6759          * 1) We need to reserve space for our orphan item and the space to
6760          * delete our orphan item.  Lord knows we don't want to have a dangling
6761          * orphan item because we didn't reserve space to remove it.
6762          *
6763          * 2) We need to reserve space to update our inode.
6764          *
6765          * 3) We need to have something to cache all the space that is going to
6766          * be free'd up by the truncate operation, but also have some slack
6767          * space reserved in case it uses space during the truncate (thank you
6768          * very much snapshotting).
6769          *
6770          * And we need these to all be seperate.  The fact is we can use alot of
6771          * space doing the truncate, and we have no earthly idea how much space
6772          * we will use, so we need the truncate reservation to be seperate so it
6773          * doesn't end up using space reserved for updating the inode or
6774          * removing the orphan item.  We also need to be able to stop the
6775          * transaction and start a new one, which means we need to be able to
6776          * update the inode several times, and we have no idea of knowing how
6777          * many times that will be, so we can't just reserve 1 item for the
6778          * entirety of the opration, so that has to be done seperately as well.
6779          * Then there is the orphan item, which does indeed need to be held on
6780          * to for the whole operation, and we need nobody to touch this reserved
6781          * space except the orphan code.
6782          *
6783          * So that leaves us with
6784          *
6785          * 1) root->orphan_block_rsv - for the orphan deletion.
6786          * 2) rsv - for the truncate reservation, which we will steal from the
6787          * transaction reservation.
6788          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6789          * updating the inode.
6790          */
6791         rsv = btrfs_alloc_block_rsv(root);
6792         if (!rsv)
6793                 return -ENOMEM;
6794         rsv->size = min_size;
6795
6796         /*
6797          * 1 for the truncate slack space
6798          * 1 for the orphan item we're going to add
6799          * 1 for the orphan item deletion
6800          * 1 for updating the inode.
6801          */
6802         trans = btrfs_start_transaction(root, 4);
6803         if (IS_ERR(trans)) {
6804                 err = PTR_ERR(trans);
6805                 goto out;
6806         }
6807
6808         /* Migrate the slack space for the truncate to our reserve */
6809         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6810                                       min_size);
6811         BUG_ON(ret);
6812
6813         ret = btrfs_orphan_add(trans, inode);
6814         if (ret) {
6815                 btrfs_end_transaction(trans, root);
6816                 goto out;
6817         }
6818
6819         /*
6820          * setattr is responsible for setting the ordered_data_close flag,
6821          * but that is only tested during the last file release.  That
6822          * could happen well after the next commit, leaving a great big
6823          * window where new writes may get lost if someone chooses to write
6824          * to this file after truncating to zero
6825          *
6826          * The inode doesn't have any dirty data here, and so if we commit
6827          * this is a noop.  If someone immediately starts writing to the inode
6828          * it is very likely we'll catch some of their writes in this
6829          * transaction, and the commit will find this file on the ordered
6830          * data list with good things to send down.
6831          *
6832          * This is a best effort solution, there is still a window where
6833          * using truncate to replace the contents of the file will
6834          * end up with a zero length file after a crash.
6835          */
6836         if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
6837                                            &BTRFS_I(inode)->runtime_flags))
6838                 btrfs_add_ordered_operation(trans, root, inode);
6839
6840         while (1) {
6841                 ret = btrfs_block_rsv_refill(root, rsv, min_size);
6842                 if (ret) {
6843                         /*
6844                          * This can only happen with the original transaction we
6845                          * started above, every other time we shouldn't have a
6846                          * transaction started yet.
6847                          */
6848                         if (ret == -EAGAIN)
6849                                 goto end_trans;
6850                         err = ret;
6851                         break;
6852                 }
6853
6854                 if (!trans) {
6855                         /* Just need the 1 for updating the inode */
6856                         trans = btrfs_start_transaction(root, 1);
6857                         if (IS_ERR(trans)) {
6858                                 ret = err = PTR_ERR(trans);
6859                                 trans = NULL;
6860                                 break;
6861                         }
6862                 }
6863
6864                 trans->block_rsv = rsv;
6865
6866                 ret = btrfs_truncate_inode_items(trans, root, inode,
6867                                                  inode->i_size,
6868                                                  BTRFS_EXTENT_DATA_KEY);
6869                 if (ret != -EAGAIN) {
6870                         err = ret;
6871                         break;
6872                 }
6873
6874                 trans->block_rsv = &root->fs_info->trans_block_rsv;
6875                 ret = btrfs_update_inode(trans, root, inode);
6876                 if (ret) {
6877                         err = ret;
6878                         break;
6879                 }
6880 end_trans:
6881                 nr = trans->blocks_used;
6882                 btrfs_end_transaction(trans, root);
6883                 trans = NULL;
6884                 btrfs_btree_balance_dirty(root, nr);
6885         }
6886
6887         if (ret == 0 && inode->i_nlink > 0) {
6888                 trans->block_rsv = root->orphan_block_rsv;
6889                 ret = btrfs_orphan_del(trans, inode);
6890                 if (ret)
6891                         err = ret;
6892         } else if (ret && inode->i_nlink > 0) {
6893                 /*
6894                  * Failed to do the truncate, remove us from the in memory
6895                  * orphan list.
6896                  */
6897                 ret = btrfs_orphan_del(NULL, inode);
6898         }
6899
6900         if (trans) {
6901                 trans->block_rsv = &root->fs_info->trans_block_rsv;
6902                 ret = btrfs_update_inode(trans, root, inode);
6903                 if (ret && !err)
6904                         err = ret;
6905
6906                 nr = trans->blocks_used;
6907                 ret = btrfs_end_transaction(trans, root);
6908                 btrfs_btree_balance_dirty(root, nr);
6909         }
6910
6911 out:
6912         btrfs_free_block_rsv(root, rsv);
6913
6914         if (ret && !err)
6915                 err = ret;
6916
6917         return err;
6918 }
6919
6920 /*
6921  * create a new subvolume directory/inode (helper for the ioctl).
6922  */
6923 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6924                              struct btrfs_root *new_root, u64 new_dirid)
6925 {
6926         struct inode *inode;
6927         int err;
6928         u64 index = 0;
6929
6930         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
6931                                 new_dirid, new_dirid,
6932                                 S_IFDIR | (~current_umask() & S_IRWXUGO),
6933                                 &index);
6934         if (IS_ERR(inode))
6935                 return PTR_ERR(inode);
6936         inode->i_op = &btrfs_dir_inode_operations;
6937         inode->i_fop = &btrfs_dir_file_operations;
6938
6939         set_nlink(inode, 1);
6940         btrfs_i_size_write(inode, 0);
6941
6942         err = btrfs_update_inode(trans, new_root, inode);
6943
6944         iput(inode);
6945         return err;
6946 }
6947
6948 struct inode *btrfs_alloc_inode(struct super_block *sb)
6949 {
6950         struct btrfs_inode *ei;
6951         struct inode *inode;
6952
6953         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6954         if (!ei)
6955                 return NULL;
6956
6957         ei->root = NULL;
6958         ei->generation = 0;
6959         ei->last_trans = 0;
6960         ei->last_sub_trans = 0;
6961         ei->logged_trans = 0;
6962         ei->delalloc_bytes = 0;
6963         ei->disk_i_size = 0;
6964         ei->flags = 0;
6965         ei->csum_bytes = 0;
6966         ei->index_cnt = (u64)-1;
6967         ei->last_unlink_trans = 0;
6968
6969         spin_lock_init(&ei->lock);
6970         ei->outstanding_extents = 0;
6971         ei->reserved_extents = 0;
6972
6973         ei->runtime_flags = 0;
6974         ei->force_compress = BTRFS_COMPRESS_NONE;
6975
6976         ei->delayed_node = NULL;
6977
6978         inode = &ei->vfs_inode;
6979         extent_map_tree_init(&ei->extent_tree);
6980         extent_io_tree_init(&ei->io_tree, &inode->i_data);
6981         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
6982         ei->io_tree.track_uptodate = 1;
6983         ei->io_failure_tree.track_uptodate = 1;
6984         mutex_init(&ei->log_mutex);
6985         mutex_init(&ei->delalloc_mutex);
6986         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6987         INIT_LIST_HEAD(&ei->delalloc_inodes);
6988         INIT_LIST_HEAD(&ei->ordered_operations);
6989         RB_CLEAR_NODE(&ei->rb_node);
6990
6991         return inode;
6992 }
6993
6994 static void btrfs_i_callback(struct rcu_head *head)
6995 {
6996         struct inode *inode = container_of(head, struct inode, i_rcu);
6997         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6998 }
6999
7000 void btrfs_destroy_inode(struct inode *inode)
7001 {
7002         struct btrfs_ordered_extent *ordered;
7003         struct btrfs_root *root = BTRFS_I(inode)->root;
7004
7005         WARN_ON(!hlist_empty(&inode->i_dentry));
7006         WARN_ON(inode->i_data.nrpages);
7007         WARN_ON(BTRFS_I(inode)->outstanding_extents);
7008         WARN_ON(BTRFS_I(inode)->reserved_extents);
7009         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
7010         WARN_ON(BTRFS_I(inode)->csum_bytes);
7011
7012         /*
7013          * This can happen where we create an inode, but somebody else also
7014          * created the same inode and we need to destroy the one we already
7015          * created.
7016          */
7017         if (!root)
7018                 goto free;
7019
7020         /*
7021          * Make sure we're properly removed from the ordered operation
7022          * lists.
7023          */
7024         smp_mb();
7025         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
7026                 spin_lock(&root->fs_info->ordered_extent_lock);
7027                 list_del_init(&BTRFS_I(inode)->ordered_operations);
7028                 spin_unlock(&root->fs_info->ordered_extent_lock);
7029         }
7030
7031         if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
7032                      &BTRFS_I(inode)->runtime_flags)) {
7033                 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
7034                        (unsigned long long)btrfs_ino(inode));
7035                 atomic_dec(&root->orphan_inodes);
7036         }
7037
7038         while (1) {
7039                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7040                 if (!ordered)
7041                         break;
7042                 else {
7043                         printk(KERN_ERR "btrfs found ordered "
7044                                "extent %llu %llu on inode cleanup\n",
7045                                (unsigned long long)ordered->file_offset,
7046                                (unsigned long long)ordered->len);
7047                         btrfs_remove_ordered_extent(inode, ordered);
7048                         btrfs_put_ordered_extent(ordered);
7049                         btrfs_put_ordered_extent(ordered);
7050                 }
7051         }
7052         inode_tree_del(inode);
7053         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
7054 free:
7055         btrfs_remove_delayed_node(inode);
7056         call_rcu(&inode->i_rcu, btrfs_i_callback);
7057 }
7058
7059 int btrfs_drop_inode(struct inode *inode)
7060 {
7061         struct btrfs_root *root = BTRFS_I(inode)->root;
7062
7063         if (btrfs_root_refs(&root->root_item) == 0 &&
7064             !btrfs_is_free_space_inode(inode))
7065                 return 1;
7066         else
7067                 return generic_drop_inode(inode);
7068 }
7069
7070 static void init_once(void *foo)
7071 {
7072         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
7073
7074         inode_init_once(&ei->vfs_inode);
7075 }
7076
7077 void btrfs_destroy_cachep(void)
7078 {
7079         if (btrfs_inode_cachep)
7080                 kmem_cache_destroy(btrfs_inode_cachep);
7081         if (btrfs_trans_handle_cachep)
7082                 kmem_cache_destroy(btrfs_trans_handle_cachep);
7083         if (btrfs_transaction_cachep)
7084                 kmem_cache_destroy(btrfs_transaction_cachep);
7085         if (btrfs_path_cachep)
7086                 kmem_cache_destroy(btrfs_path_cachep);
7087         if (btrfs_free_space_cachep)
7088                 kmem_cache_destroy(btrfs_free_space_cachep);
7089 }
7090
7091 int btrfs_init_cachep(void)
7092 {
7093         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
7094                         sizeof(struct btrfs_inode), 0,
7095                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
7096         if (!btrfs_inode_cachep)
7097                 goto fail;
7098
7099         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
7100                         sizeof(struct btrfs_trans_handle), 0,
7101                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7102         if (!btrfs_trans_handle_cachep)
7103                 goto fail;
7104
7105         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
7106                         sizeof(struct btrfs_transaction), 0,
7107                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7108         if (!btrfs_transaction_cachep)
7109                 goto fail;
7110
7111         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
7112                         sizeof(struct btrfs_path), 0,
7113                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7114         if (!btrfs_path_cachep)
7115                 goto fail;
7116
7117         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
7118                         sizeof(struct btrfs_free_space), 0,
7119                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7120         if (!btrfs_free_space_cachep)
7121                 goto fail;
7122
7123         return 0;
7124 fail:
7125         btrfs_destroy_cachep();
7126         return -ENOMEM;
7127 }
7128
7129 static int btrfs_getattr(struct vfsmount *mnt,
7130                          struct dentry *dentry, struct kstat *stat)
7131 {
7132         struct inode *inode = dentry->d_inode;
7133         u32 blocksize = inode->i_sb->s_blocksize;
7134
7135         generic_fillattr(inode, stat);
7136         stat->dev = BTRFS_I(inode)->root->anon_dev;
7137         stat->blksize = PAGE_CACHE_SIZE;
7138         stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
7139                 ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9;
7140         return 0;
7141 }
7142
7143 /*
7144  * If a file is moved, it will inherit the cow and compression flags of the new
7145  * directory.
7146  */
7147 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
7148 {
7149         struct btrfs_inode *b_dir = BTRFS_I(dir);
7150         struct btrfs_inode *b_inode = BTRFS_I(inode);
7151
7152         if (b_dir->flags & BTRFS_INODE_NODATACOW)
7153                 b_inode->flags |= BTRFS_INODE_NODATACOW;
7154         else
7155                 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
7156
7157         if (b_dir->flags & BTRFS_INODE_COMPRESS) {
7158                 b_inode->flags |= BTRFS_INODE_COMPRESS;
7159                 b_inode->flags &= ~BTRFS_INODE_NOCOMPRESS;
7160         } else {
7161                 b_inode->flags &= ~(BTRFS_INODE_COMPRESS |
7162                                     BTRFS_INODE_NOCOMPRESS);
7163         }
7164 }
7165
7166 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
7167                            struct inode *new_dir, struct dentry *new_dentry)
7168 {
7169         struct btrfs_trans_handle *trans;
7170         struct btrfs_root *root = BTRFS_I(old_dir)->root;
7171         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
7172         struct inode *new_inode = new_dentry->d_inode;
7173         struct inode *old_inode = old_dentry->d_inode;
7174         struct timespec ctime = CURRENT_TIME;
7175         u64 index = 0;
7176         u64 root_objectid;
7177         int ret;
7178         u64 old_ino = btrfs_ino(old_inode);
7179
7180         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
7181                 return -EPERM;
7182
7183         /* we only allow rename subvolume link between subvolumes */
7184         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
7185                 return -EXDEV;
7186
7187         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
7188             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
7189                 return -ENOTEMPTY;
7190
7191         if (S_ISDIR(old_inode->i_mode) && new_inode &&
7192             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
7193                 return -ENOTEMPTY;
7194         /*
7195          * we're using rename to replace one file with another.
7196          * and the replacement file is large.  Start IO on it now so
7197          * we don't add too much work to the end of the transaction
7198          */
7199         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
7200             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
7201                 filemap_flush(old_inode->i_mapping);
7202
7203         /* close the racy window with snapshot create/destroy ioctl */
7204         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7205                 down_read(&root->fs_info->subvol_sem);
7206         /*
7207          * We want to reserve the absolute worst case amount of items.  So if
7208          * both inodes are subvols and we need to unlink them then that would
7209          * require 4 item modifications, but if they are both normal inodes it
7210          * would require 5 item modifications, so we'll assume their normal
7211          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7212          * should cover the worst case number of items we'll modify.
7213          */
7214         trans = btrfs_start_transaction(root, 20);
7215         if (IS_ERR(trans)) {
7216                 ret = PTR_ERR(trans);
7217                 goto out_notrans;
7218         }
7219
7220         if (dest != root)
7221                 btrfs_record_root_in_trans(trans, dest);
7222
7223         ret = btrfs_set_inode_index(new_dir, &index);
7224         if (ret)
7225                 goto out_fail;
7226
7227         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7228                 /* force full log commit if subvolume involved. */
7229                 root->fs_info->last_trans_log_full_commit = trans->transid;
7230         } else {
7231                 ret = btrfs_insert_inode_ref(trans, dest,
7232                                              new_dentry->d_name.name,
7233                                              new_dentry->d_name.len,
7234                                              old_ino,
7235                                              btrfs_ino(new_dir), index);
7236                 if (ret)
7237                         goto out_fail;
7238                 /*
7239                  * this is an ugly little race, but the rename is required
7240                  * to make sure that if we crash, the inode is either at the
7241                  * old name or the new one.  pinning the log transaction lets
7242                  * us make sure we don't allow a log commit to come in after
7243                  * we unlink the name but before we add the new name back in.
7244                  */
7245                 btrfs_pin_log_trans(root);
7246         }
7247         /*
7248          * make sure the inode gets flushed if it is replacing
7249          * something.
7250          */
7251         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7252                 btrfs_add_ordered_operation(trans, root, old_inode);
7253
7254         inode_inc_iversion(old_dir);
7255         inode_inc_iversion(new_dir);
7256         inode_inc_iversion(old_inode);
7257         old_dir->i_ctime = old_dir->i_mtime = ctime;
7258         new_dir->i_ctime = new_dir->i_mtime = ctime;
7259         old_inode->i_ctime = ctime;
7260
7261         if (old_dentry->d_parent != new_dentry->d_parent)
7262                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7263
7264         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7265                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7266                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7267                                         old_dentry->d_name.name,
7268                                         old_dentry->d_name.len);
7269         } else {
7270                 ret = __btrfs_unlink_inode(trans, root, old_dir,
7271                                         old_dentry->d_inode,
7272                                         old_dentry->d_name.name,
7273                                         old_dentry->d_name.len);
7274                 if (!ret)
7275                         ret = btrfs_update_inode(trans, root, old_inode);
7276         }
7277         if (ret) {
7278                 btrfs_abort_transaction(trans, root, ret);
7279                 goto out_fail;
7280         }
7281
7282         if (new_inode) {
7283                 inode_inc_iversion(new_inode);
7284                 new_inode->i_ctime = CURRENT_TIME;
7285                 if (unlikely(btrfs_ino(new_inode) ==
7286                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7287                         root_objectid = BTRFS_I(new_inode)->location.objectid;
7288                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
7289                                                 root_objectid,
7290                                                 new_dentry->d_name.name,
7291                                                 new_dentry->d_name.len);
7292                         BUG_ON(new_inode->i_nlink == 0);
7293                 } else {
7294                         ret = btrfs_unlink_inode(trans, dest, new_dir,
7295                                                  new_dentry->d_inode,
7296                                                  new_dentry->d_name.name,
7297                                                  new_dentry->d_name.len);
7298                 }
7299                 if (!ret && new_inode->i_nlink == 0) {
7300                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7301                         BUG_ON(ret);
7302                 }
7303                 if (ret) {
7304                         btrfs_abort_transaction(trans, root, ret);
7305                         goto out_fail;
7306                 }
7307         }
7308
7309         fixup_inode_flags(new_dir, old_inode);
7310
7311         ret = btrfs_add_link(trans, new_dir, old_inode,
7312                              new_dentry->d_name.name,
7313                              new_dentry->d_name.len, 0, index);
7314         if (ret) {
7315                 btrfs_abort_transaction(trans, root, ret);
7316                 goto out_fail;
7317         }
7318
7319         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7320                 struct dentry *parent = new_dentry->d_parent;
7321                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7322                 btrfs_end_log_trans(root);
7323         }
7324 out_fail:
7325         btrfs_end_transaction(trans, root);
7326 out_notrans:
7327         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7328                 up_read(&root->fs_info->subvol_sem);
7329
7330         return ret;
7331 }
7332
7333 /*
7334  * some fairly slow code that needs optimization. This walks the list
7335  * of all the inodes with pending delalloc and forces them to disk.
7336  */
7337 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7338 {
7339         struct list_head *head = &root->fs_info->delalloc_inodes;
7340         struct btrfs_inode *binode;
7341         struct inode *inode;
7342
7343         if (root->fs_info->sb->s_flags & MS_RDONLY)
7344                 return -EROFS;
7345
7346         spin_lock(&root->fs_info->delalloc_lock);
7347         while (!list_empty(head)) {
7348                 binode = list_entry(head->next, struct btrfs_inode,
7349                                     delalloc_inodes);
7350                 inode = igrab(&binode->vfs_inode);
7351                 if (!inode)
7352                         list_del_init(&binode->delalloc_inodes);
7353                 spin_unlock(&root->fs_info->delalloc_lock);
7354                 if (inode) {
7355                         filemap_flush(inode->i_mapping);
7356                         if (delay_iput)
7357                                 btrfs_add_delayed_iput(inode);
7358                         else
7359                                 iput(inode);
7360                 }
7361                 cond_resched();
7362                 spin_lock(&root->fs_info->delalloc_lock);
7363         }
7364         spin_unlock(&root->fs_info->delalloc_lock);
7365
7366         /* the filemap_flush will queue IO into the worker threads, but
7367          * we have to make sure the IO is actually started and that
7368          * ordered extents get created before we return
7369          */
7370         atomic_inc(&root->fs_info->async_submit_draining);
7371         while (atomic_read(&root->fs_info->nr_async_submits) ||
7372               atomic_read(&root->fs_info->async_delalloc_pages)) {
7373                 wait_event(root->fs_info->async_submit_wait,
7374                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7375                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7376         }
7377         atomic_dec(&root->fs_info->async_submit_draining);
7378         return 0;
7379 }
7380
7381 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7382                          const char *symname)
7383 {
7384         struct btrfs_trans_handle *trans;
7385         struct btrfs_root *root = BTRFS_I(dir)->root;
7386         struct btrfs_path *path;
7387         struct btrfs_key key;
7388         struct inode *inode = NULL;
7389         int err;
7390         int drop_inode = 0;
7391         u64 objectid;
7392         u64 index = 0 ;
7393         int name_len;
7394         int datasize;
7395         unsigned long ptr;
7396         struct btrfs_file_extent_item *ei;
7397         struct extent_buffer *leaf;
7398         unsigned long nr = 0;
7399
7400         name_len = strlen(symname) + 1;
7401         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7402                 return -ENAMETOOLONG;
7403
7404         /*
7405          * 2 items for inode item and ref
7406          * 2 items for dir items
7407          * 1 item for xattr if selinux is on
7408          */
7409         trans = btrfs_start_transaction(root, 5);
7410         if (IS_ERR(trans))
7411                 return PTR_ERR(trans);
7412
7413         err = btrfs_find_free_ino(root, &objectid);
7414         if (err)
7415                 goto out_unlock;
7416
7417         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7418                                 dentry->d_name.len, btrfs_ino(dir), objectid,
7419                                 S_IFLNK|S_IRWXUGO, &index);
7420         if (IS_ERR(inode)) {
7421                 err = PTR_ERR(inode);
7422                 goto out_unlock;
7423         }
7424
7425         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7426         if (err) {
7427                 drop_inode = 1;
7428                 goto out_unlock;
7429         }
7430
7431         /*
7432         * If the active LSM wants to access the inode during
7433         * d_instantiate it needs these. Smack checks to see
7434         * if the filesystem supports xattrs by looking at the
7435         * ops vector.
7436         */
7437         inode->i_fop = &btrfs_file_operations;
7438         inode->i_op = &btrfs_file_inode_operations;
7439
7440         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7441         if (err)
7442                 drop_inode = 1;
7443         else {
7444                 inode->i_mapping->a_ops = &btrfs_aops;
7445                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7446                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7447         }
7448         if (drop_inode)
7449                 goto out_unlock;
7450
7451         path = btrfs_alloc_path();
7452         if (!path) {
7453                 err = -ENOMEM;
7454                 drop_inode = 1;
7455                 goto out_unlock;
7456         }
7457         key.objectid = btrfs_ino(inode);
7458         key.offset = 0;
7459         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7460         datasize = btrfs_file_extent_calc_inline_size(name_len);
7461         err = btrfs_insert_empty_item(trans, root, path, &key,
7462                                       datasize);
7463         if (err) {
7464                 drop_inode = 1;
7465                 btrfs_free_path(path);
7466                 goto out_unlock;
7467         }
7468         leaf = path->nodes[0];
7469         ei = btrfs_item_ptr(leaf, path->slots[0],
7470                             struct btrfs_file_extent_item);
7471         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7472         btrfs_set_file_extent_type(leaf, ei,
7473                                    BTRFS_FILE_EXTENT_INLINE);
7474         btrfs_set_file_extent_encryption(leaf, ei, 0);
7475         btrfs_set_file_extent_compression(leaf, ei, 0);
7476         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7477         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7478
7479         ptr = btrfs_file_extent_inline_start(ei);
7480         write_extent_buffer(leaf, symname, ptr, name_len);
7481         btrfs_mark_buffer_dirty(leaf);
7482         btrfs_free_path(path);
7483
7484         inode->i_op = &btrfs_symlink_inode_operations;
7485         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7486         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7487         inode_set_bytes(inode, name_len);
7488         btrfs_i_size_write(inode, name_len - 1);
7489         err = btrfs_update_inode(trans, root, inode);
7490         if (err)
7491                 drop_inode = 1;
7492
7493 out_unlock:
7494         if (!err)
7495                 d_instantiate(dentry, inode);
7496         nr = trans->blocks_used;
7497         btrfs_end_transaction(trans, root);
7498         if (drop_inode) {
7499                 inode_dec_link_count(inode);
7500                 iput(inode);
7501         }
7502         btrfs_btree_balance_dirty(root, nr);
7503         return err;
7504 }
7505
7506 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7507                                        u64 start, u64 num_bytes, u64 min_size,
7508                                        loff_t actual_len, u64 *alloc_hint,
7509                                        struct btrfs_trans_handle *trans)
7510 {
7511         struct btrfs_root *root = BTRFS_I(inode)->root;
7512         struct btrfs_key ins;
7513         u64 cur_offset = start;
7514         u64 i_size;
7515         int ret = 0;
7516         bool own_trans = true;
7517
7518         if (trans)
7519                 own_trans = false;
7520         while (num_bytes > 0) {
7521                 if (own_trans) {
7522                         trans = btrfs_start_transaction(root, 3);
7523                         if (IS_ERR(trans)) {
7524                                 ret = PTR_ERR(trans);
7525                                 break;
7526                         }
7527                 }
7528
7529                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7530                                            0, *alloc_hint, &ins, 1);
7531                 if (ret) {
7532                         if (own_trans)
7533                                 btrfs_end_transaction(trans, root);
7534                         break;
7535                 }
7536
7537                 ret = insert_reserved_file_extent(trans, inode,
7538                                                   cur_offset, ins.objectid,
7539                                                   ins.offset, ins.offset,
7540                                                   ins.offset, 0, 0, 0,
7541                                                   BTRFS_FILE_EXTENT_PREALLOC);
7542                 if (ret) {
7543                         btrfs_abort_transaction(trans, root, ret);
7544                         if (own_trans)
7545                                 btrfs_end_transaction(trans, root);
7546                         break;
7547                 }
7548                 btrfs_drop_extent_cache(inode, cur_offset,
7549                                         cur_offset + ins.offset -1, 0);
7550
7551                 num_bytes -= ins.offset;
7552                 cur_offset += ins.offset;
7553                 *alloc_hint = ins.objectid + ins.offset;
7554
7555                 inode_inc_iversion(inode);
7556                 inode->i_ctime = CURRENT_TIME;
7557                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7558                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7559                     (actual_len > inode->i_size) &&
7560                     (cur_offset > inode->i_size)) {
7561                         if (cur_offset > actual_len)
7562                                 i_size = actual_len;
7563                         else
7564                                 i_size = cur_offset;
7565                         i_size_write(inode, i_size);
7566                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7567                 }
7568
7569                 ret = btrfs_update_inode(trans, root, inode);
7570
7571                 if (ret) {
7572                         btrfs_abort_transaction(trans, root, ret);
7573                         if (own_trans)
7574                                 btrfs_end_transaction(trans, root);
7575                         break;
7576                 }
7577
7578                 if (own_trans)
7579                         btrfs_end_transaction(trans, root);
7580         }
7581         return ret;
7582 }
7583
7584 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7585                               u64 start, u64 num_bytes, u64 min_size,
7586                               loff_t actual_len, u64 *alloc_hint)
7587 {
7588         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7589                                            min_size, actual_len, alloc_hint,
7590                                            NULL);
7591 }
7592
7593 int btrfs_prealloc_file_range_trans(struct inode *inode,
7594                                     struct btrfs_trans_handle *trans, int mode,
7595                                     u64 start, u64 num_bytes, u64 min_size,
7596                                     loff_t actual_len, u64 *alloc_hint)
7597 {
7598         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7599                                            min_size, actual_len, alloc_hint, trans);
7600 }
7601
7602 static int btrfs_set_page_dirty(struct page *page)
7603 {
7604         return __set_page_dirty_nobuffers(page);
7605 }
7606
7607 static int btrfs_permission(struct inode *inode, int mask)
7608 {
7609         struct btrfs_root *root = BTRFS_I(inode)->root;
7610         umode_t mode = inode->i_mode;
7611
7612         if (mask & MAY_WRITE &&
7613             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7614                 if (btrfs_root_readonly(root))
7615                         return -EROFS;
7616                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7617                         return -EACCES;
7618         }
7619         return generic_permission(inode, mask);
7620 }
7621
7622 static const struct inode_operations btrfs_dir_inode_operations = {
7623         .getattr        = btrfs_getattr,
7624         .lookup         = btrfs_lookup,
7625         .create         = btrfs_create,
7626         .unlink         = btrfs_unlink,
7627         .link           = btrfs_link,
7628         .mkdir          = btrfs_mkdir,
7629         .rmdir          = btrfs_rmdir,
7630         .rename         = btrfs_rename,
7631         .symlink        = btrfs_symlink,
7632         .setattr        = btrfs_setattr,
7633         .mknod          = btrfs_mknod,
7634         .setxattr       = btrfs_setxattr,
7635         .getxattr       = btrfs_getxattr,
7636         .listxattr      = btrfs_listxattr,
7637         .removexattr    = btrfs_removexattr,
7638         .permission     = btrfs_permission,
7639         .get_acl        = btrfs_get_acl,
7640 };
7641 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7642         .lookup         = btrfs_lookup,
7643         .permission     = btrfs_permission,
7644         .get_acl        = btrfs_get_acl,
7645 };
7646
7647 static const struct file_operations btrfs_dir_file_operations = {
7648         .llseek         = generic_file_llseek,
7649         .read           = generic_read_dir,
7650         .readdir        = btrfs_real_readdir,
7651         .unlocked_ioctl = btrfs_ioctl,
7652 #ifdef CONFIG_COMPAT
7653         .compat_ioctl   = btrfs_ioctl,
7654 #endif
7655         .release        = btrfs_release_file,
7656         .fsync          = btrfs_sync_file,
7657 };
7658
7659 static struct extent_io_ops btrfs_extent_io_ops = {
7660         .fill_delalloc = run_delalloc_range,
7661         .submit_bio_hook = btrfs_submit_bio_hook,
7662         .merge_bio_hook = btrfs_merge_bio_hook,
7663         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7664         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7665         .writepage_start_hook = btrfs_writepage_start_hook,
7666         .set_bit_hook = btrfs_set_bit_hook,
7667         .clear_bit_hook = btrfs_clear_bit_hook,
7668         .merge_extent_hook = btrfs_merge_extent_hook,
7669         .split_extent_hook = btrfs_split_extent_hook,
7670 };
7671
7672 /*
7673  * btrfs doesn't support the bmap operation because swapfiles
7674  * use bmap to make a mapping of extents in the file.  They assume
7675  * these extents won't change over the life of the file and they
7676  * use the bmap result to do IO directly to the drive.
7677  *
7678  * the btrfs bmap call would return logical addresses that aren't
7679  * suitable for IO and they also will change frequently as COW
7680  * operations happen.  So, swapfile + btrfs == corruption.
7681  *
7682  * For now we're avoiding this by dropping bmap.
7683  */
7684 static const struct address_space_operations btrfs_aops = {
7685         .readpage       = btrfs_readpage,
7686         .writepage      = btrfs_writepage,
7687         .writepages     = btrfs_writepages,
7688         .readpages      = btrfs_readpages,
7689         .direct_IO      = btrfs_direct_IO,
7690         .invalidatepage = btrfs_invalidatepage,
7691         .releasepage    = btrfs_releasepage,
7692         .set_page_dirty = btrfs_set_page_dirty,
7693         .error_remove_page = generic_error_remove_page,
7694 };
7695
7696 static const struct address_space_operations btrfs_symlink_aops = {
7697         .readpage       = btrfs_readpage,
7698         .writepage      = btrfs_writepage,
7699         .invalidatepage = btrfs_invalidatepage,
7700         .releasepage    = btrfs_releasepage,
7701 };
7702
7703 static const struct inode_operations btrfs_file_inode_operations = {
7704         .getattr        = btrfs_getattr,
7705         .setattr        = btrfs_setattr,
7706         .setxattr       = btrfs_setxattr,
7707         .getxattr       = btrfs_getxattr,
7708         .listxattr      = btrfs_listxattr,
7709         .removexattr    = btrfs_removexattr,
7710         .permission     = btrfs_permission,
7711         .fiemap         = btrfs_fiemap,
7712         .get_acl        = btrfs_get_acl,
7713         .update_time    = btrfs_update_time,
7714 };
7715 static const struct inode_operations btrfs_special_inode_operations = {
7716         .getattr        = btrfs_getattr,
7717         .setattr        = btrfs_setattr,
7718         .permission     = btrfs_permission,
7719         .setxattr       = btrfs_setxattr,
7720         .getxattr       = btrfs_getxattr,
7721         .listxattr      = btrfs_listxattr,
7722         .removexattr    = btrfs_removexattr,
7723         .get_acl        = btrfs_get_acl,
7724         .update_time    = btrfs_update_time,
7725 };
7726 static const struct inode_operations btrfs_symlink_inode_operations = {
7727         .readlink       = generic_readlink,
7728         .follow_link    = page_follow_link_light,
7729         .put_link       = page_put_link,
7730         .getattr        = btrfs_getattr,
7731         .setattr        = btrfs_setattr,
7732         .permission     = btrfs_permission,
7733         .setxattr       = btrfs_setxattr,
7734         .getxattr       = btrfs_getxattr,
7735         .listxattr      = btrfs_listxattr,
7736         .removexattr    = btrfs_removexattr,
7737         .get_acl        = btrfs_get_acl,
7738         .update_time    = btrfs_update_time,
7739 };
7740
7741 const struct dentry_operations btrfs_dentry_operations = {
7742         .d_delete       = btrfs_dentry_delete,
7743         .d_release      = btrfs_dentry_release,
7744 };