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         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
1012
1013         if (atomic_read(&root->fs_info->async_delalloc_pages) <
1014             5 * 1024 * 1024 &&
1015             waitqueue_active(&root->fs_info->async_submit_wait))
1016                 wake_up(&root->fs_info->async_submit_wait);
1017
1018         if (async_cow->inode)
1019                 submit_compressed_extents(async_cow->inode, async_cow);
1020 }
1021
1022 static noinline void async_cow_free(struct btrfs_work *work)
1023 {
1024         struct async_cow *async_cow;
1025         async_cow = container_of(work, struct async_cow, work);
1026         if (async_cow->inode)
1027                 btrfs_add_delayed_iput(async_cow->inode);
1028         kfree(async_cow);
1029 }
1030
1031 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1032                                 u64 start, u64 end, int *page_started,
1033                                 unsigned long *nr_written)
1034 {
1035         struct async_cow *async_cow;
1036         struct btrfs_root *root = BTRFS_I(inode)->root;
1037         unsigned long nr_pages;
1038         u64 cur_end;
1039         int limit = 10 * 1024 * 1024;
1040
1041         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1042                          1, 0, NULL, GFP_NOFS);
1043         while (start < end) {
1044                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1045                 BUG_ON(!async_cow); /* -ENOMEM */
1046                 async_cow->inode = igrab(inode);
1047                 async_cow->root = root;
1048                 async_cow->locked_page = locked_page;
1049                 async_cow->start = start;
1050
1051                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1052                         cur_end = end;
1053                 else
1054                         cur_end = min(end, start + 512 * 1024 - 1);
1055
1056                 async_cow->end = cur_end;
1057                 INIT_LIST_HEAD(&async_cow->extents);
1058
1059                 async_cow->work.func = async_cow_start;
1060                 async_cow->work.ordered_func = async_cow_submit;
1061                 async_cow->work.ordered_free = async_cow_free;
1062                 async_cow->work.flags = 0;
1063
1064                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1065                         PAGE_CACHE_SHIFT;
1066                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1067
1068                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1069                                    &async_cow->work);
1070
1071                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1072                         wait_event(root->fs_info->async_submit_wait,
1073                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1074                             limit));
1075                 }
1076
1077                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1078                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1079                         wait_event(root->fs_info->async_submit_wait,
1080                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1081                            0));
1082                 }
1083
1084                 *nr_written += nr_pages;
1085                 start = cur_end + 1;
1086         }
1087         *page_started = 1;
1088         return 0;
1089 }
1090
1091 static noinline int csum_exist_in_range(struct btrfs_root *root,
1092                                         u64 bytenr, u64 num_bytes)
1093 {
1094         int ret;
1095         struct btrfs_ordered_sum *sums;
1096         LIST_HEAD(list);
1097
1098         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1099                                        bytenr + num_bytes - 1, &list, 0);
1100         if (ret == 0 && list_empty(&list))
1101                 return 0;
1102
1103         while (!list_empty(&list)) {
1104                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1105                 list_del(&sums->list);
1106                 kfree(sums);
1107         }
1108         return 1;
1109 }
1110
1111 /*
1112  * when nowcow writeback call back.  This checks for snapshots or COW copies
1113  * of the extents that exist in the file, and COWs the file as required.
1114  *
1115  * If no cow copies or snapshots exist, we write directly to the existing
1116  * blocks on disk
1117  */
1118 static noinline int run_delalloc_nocow(struct inode *inode,
1119                                        struct page *locked_page,
1120                               u64 start, u64 end, int *page_started, int force,
1121                               unsigned long *nr_written)
1122 {
1123         struct btrfs_root *root = BTRFS_I(inode)->root;
1124         struct btrfs_trans_handle *trans;
1125         struct extent_buffer *leaf;
1126         struct btrfs_path *path;
1127         struct btrfs_file_extent_item *fi;
1128         struct btrfs_key found_key;
1129         u64 cow_start;
1130         u64 cur_offset;
1131         u64 extent_end;
1132         u64 extent_offset;
1133         u64 disk_bytenr;
1134         u64 num_bytes;
1135         int extent_type;
1136         int ret, err;
1137         int type;
1138         int nocow;
1139         int check_prev = 1;
1140         bool nolock;
1141         u64 ino = btrfs_ino(inode);
1142
1143         path = btrfs_alloc_path();
1144         if (!path) {
1145                 extent_clear_unlock_delalloc(inode,
1146                              &BTRFS_I(inode)->io_tree,
1147                              start, end, locked_page,
1148                              EXTENT_CLEAR_UNLOCK_PAGE |
1149                              EXTENT_CLEAR_UNLOCK |
1150                              EXTENT_CLEAR_DELALLOC |
1151                              EXTENT_CLEAR_DIRTY |
1152                              EXTENT_SET_WRITEBACK |
1153                              EXTENT_END_WRITEBACK);
1154                 return -ENOMEM;
1155         }
1156
1157         nolock = btrfs_is_free_space_inode(inode);
1158
1159         if (nolock)
1160                 trans = btrfs_join_transaction_nolock(root);
1161         else
1162                 trans = btrfs_join_transaction(root);
1163
1164         if (IS_ERR(trans)) {
1165                 extent_clear_unlock_delalloc(inode,
1166                              &BTRFS_I(inode)->io_tree,
1167                              start, end, locked_page,
1168                              EXTENT_CLEAR_UNLOCK_PAGE |
1169                              EXTENT_CLEAR_UNLOCK |
1170                              EXTENT_CLEAR_DELALLOC |
1171                              EXTENT_CLEAR_DIRTY |
1172                              EXTENT_SET_WRITEBACK |
1173                              EXTENT_END_WRITEBACK);
1174                 btrfs_free_path(path);
1175                 return PTR_ERR(trans);
1176         }
1177
1178         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1179
1180         cow_start = (u64)-1;
1181         cur_offset = start;
1182         while (1) {
1183                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1184                                                cur_offset, 0);
1185                 if (ret < 0) {
1186                         btrfs_abort_transaction(trans, root, ret);
1187                         goto error;
1188                 }
1189                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1190                         leaf = path->nodes[0];
1191                         btrfs_item_key_to_cpu(leaf, &found_key,
1192                                               path->slots[0] - 1);
1193                         if (found_key.objectid == ino &&
1194                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1195                                 path->slots[0]--;
1196                 }
1197                 check_prev = 0;
1198 next_slot:
1199                 leaf = path->nodes[0];
1200                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1201                         ret = btrfs_next_leaf(root, path);
1202                         if (ret < 0) {
1203                                 btrfs_abort_transaction(trans, root, ret);
1204                                 goto error;
1205                         }
1206                         if (ret > 0)
1207                                 break;
1208                         leaf = path->nodes[0];
1209                 }
1210
1211                 nocow = 0;
1212                 disk_bytenr = 0;
1213                 num_bytes = 0;
1214                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1215
1216                 if (found_key.objectid > ino ||
1217                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1218                     found_key.offset > end)
1219                         break;
1220
1221                 if (found_key.offset > cur_offset) {
1222                         extent_end = found_key.offset;
1223                         extent_type = 0;
1224                         goto out_check;
1225                 }
1226
1227                 fi = btrfs_item_ptr(leaf, path->slots[0],
1228                                     struct btrfs_file_extent_item);
1229                 extent_type = btrfs_file_extent_type(leaf, fi);
1230
1231                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1232                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1233                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1234                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1235                         extent_end = found_key.offset +
1236                                 btrfs_file_extent_num_bytes(leaf, fi);
1237                         if (extent_end <= start) {
1238                                 path->slots[0]++;
1239                                 goto next_slot;
1240                         }
1241                         if (disk_bytenr == 0)
1242                                 goto out_check;
1243                         if (btrfs_file_extent_compression(leaf, fi) ||
1244                             btrfs_file_extent_encryption(leaf, fi) ||
1245                             btrfs_file_extent_other_encoding(leaf, fi))
1246                                 goto out_check;
1247                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1248                                 goto out_check;
1249                         if (btrfs_extent_readonly(root, disk_bytenr))
1250                                 goto out_check;
1251                         if (btrfs_cross_ref_exist(trans, root, ino,
1252                                                   found_key.offset -
1253                                                   extent_offset, disk_bytenr))
1254                                 goto out_check;
1255                         disk_bytenr += extent_offset;
1256                         disk_bytenr += cur_offset - found_key.offset;
1257                         num_bytes = min(end + 1, extent_end) - cur_offset;
1258                         /*
1259                          * force cow if csum exists in the range.
1260                          * this ensure that csum for a given extent are
1261                          * either valid or do not exist.
1262                          */
1263                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1264                                 goto out_check;
1265                         nocow = 1;
1266                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1267                         extent_end = found_key.offset +
1268                                 btrfs_file_extent_inline_len(leaf, fi);
1269                         extent_end = ALIGN(extent_end, root->sectorsize);
1270                 } else {
1271                         BUG_ON(1);
1272                 }
1273 out_check:
1274                 if (extent_end <= start) {
1275                         path->slots[0]++;
1276                         goto next_slot;
1277                 }
1278                 if (!nocow) {
1279                         if (cow_start == (u64)-1)
1280                                 cow_start = cur_offset;
1281                         cur_offset = extent_end;
1282                         if (cur_offset > end)
1283                                 break;
1284                         path->slots[0]++;
1285                         goto next_slot;
1286                 }
1287
1288                 btrfs_release_path(path);
1289                 if (cow_start != (u64)-1) {
1290                         ret = cow_file_range(inode, locked_page, cow_start,
1291                                         found_key.offset - 1, page_started,
1292                                         nr_written, 1);
1293                         if (ret) {
1294                                 btrfs_abort_transaction(trans, root, ret);
1295                                 goto error;
1296                         }
1297                         cow_start = (u64)-1;
1298                 }
1299
1300                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1301                         struct extent_map *em;
1302                         struct extent_map_tree *em_tree;
1303                         em_tree = &BTRFS_I(inode)->extent_tree;
1304                         em = alloc_extent_map();
1305                         BUG_ON(!em); /* -ENOMEM */
1306                         em->start = cur_offset;
1307                         em->orig_start = em->start;
1308                         em->len = num_bytes;
1309                         em->block_len = num_bytes;
1310                         em->block_start = disk_bytenr;
1311                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1312                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1313                         while (1) {
1314                                 write_lock(&em_tree->lock);
1315                                 ret = add_extent_mapping(em_tree, em);
1316                                 write_unlock(&em_tree->lock);
1317                                 if (ret != -EEXIST) {
1318                                         free_extent_map(em);
1319                                         break;
1320                                 }
1321                                 btrfs_drop_extent_cache(inode, em->start,
1322                                                 em->start + em->len - 1, 0);
1323                         }
1324                         type = BTRFS_ORDERED_PREALLOC;
1325                 } else {
1326                         type = BTRFS_ORDERED_NOCOW;
1327                 }
1328
1329                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1330                                                num_bytes, num_bytes, type);
1331                 BUG_ON(ret); /* -ENOMEM */
1332
1333                 if (root->root_key.objectid ==
1334                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1335                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1336                                                       num_bytes);
1337                         if (ret) {
1338                                 btrfs_abort_transaction(trans, root, ret);
1339                                 goto error;
1340                         }
1341                 }
1342
1343                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1344                                 cur_offset, cur_offset + num_bytes - 1,
1345                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1346                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1347                                 EXTENT_SET_PRIVATE2);
1348                 cur_offset = extent_end;
1349                 if (cur_offset > end)
1350                         break;
1351         }
1352         btrfs_release_path(path);
1353
1354         if (cur_offset <= end && cow_start == (u64)-1) {
1355                 cow_start = cur_offset;
1356                 cur_offset = end;
1357         }
1358
1359         if (cow_start != (u64)-1) {
1360                 ret = cow_file_range(inode, locked_page, cow_start, end,
1361                                      page_started, nr_written, 1);
1362                 if (ret) {
1363                         btrfs_abort_transaction(trans, root, ret);
1364                         goto error;
1365                 }
1366         }
1367
1368 error:
1369         if (nolock) {
1370                 err = btrfs_end_transaction_nolock(trans, root);
1371         } else {
1372                 err = btrfs_end_transaction(trans, root);
1373         }
1374         if (!ret)
1375                 ret = err;
1376
1377         if (ret && cur_offset < end)
1378                 extent_clear_unlock_delalloc(inode,
1379                              &BTRFS_I(inode)->io_tree,
1380                              cur_offset, end, locked_page,
1381                              EXTENT_CLEAR_UNLOCK_PAGE |
1382                              EXTENT_CLEAR_UNLOCK |
1383                              EXTENT_CLEAR_DELALLOC |
1384                              EXTENT_CLEAR_DIRTY |
1385                              EXTENT_SET_WRITEBACK |
1386                              EXTENT_END_WRITEBACK);
1387
1388         btrfs_free_path(path);
1389         return ret;
1390 }
1391
1392 /*
1393  * extent_io.c call back to do delayed allocation processing
1394  */
1395 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1396                               u64 start, u64 end, int *page_started,
1397                               unsigned long *nr_written)
1398 {
1399         int ret;
1400         struct btrfs_root *root = BTRFS_I(inode)->root;
1401
1402         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1403                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1404                                          page_started, 1, nr_written);
1405         } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1406                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1407                                          page_started, 0, nr_written);
1408         } else if (!btrfs_test_opt(root, COMPRESS) &&
1409                    !(BTRFS_I(inode)->force_compress) &&
1410                    !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1411                 ret = cow_file_range(inode, locked_page, start, end,
1412                                       page_started, nr_written, 1);
1413         } else {
1414                 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1415                         &BTRFS_I(inode)->runtime_flags);
1416                 ret = cow_file_range_async(inode, locked_page, start, end,
1417                                            page_started, nr_written);
1418         }
1419         return ret;
1420 }
1421
1422 static void btrfs_split_extent_hook(struct inode *inode,
1423                                     struct extent_state *orig, u64 split)
1424 {
1425         /* not delalloc, ignore it */
1426         if (!(orig->state & EXTENT_DELALLOC))
1427                 return;
1428
1429         spin_lock(&BTRFS_I(inode)->lock);
1430         BTRFS_I(inode)->outstanding_extents++;
1431         spin_unlock(&BTRFS_I(inode)->lock);
1432 }
1433
1434 /*
1435  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1436  * extents so we can keep track of new extents that are just merged onto old
1437  * extents, such as when we are doing sequential writes, so we can properly
1438  * account for the metadata space we'll need.
1439  */
1440 static void btrfs_merge_extent_hook(struct inode *inode,
1441                                     struct extent_state *new,
1442                                     struct extent_state *other)
1443 {
1444         /* not delalloc, ignore it */
1445         if (!(other->state & EXTENT_DELALLOC))
1446                 return;
1447
1448         spin_lock(&BTRFS_I(inode)->lock);
1449         BTRFS_I(inode)->outstanding_extents--;
1450         spin_unlock(&BTRFS_I(inode)->lock);
1451 }
1452
1453 /*
1454  * extent_io.c set_bit_hook, used to track delayed allocation
1455  * bytes in this file, and to maintain the list of inodes that
1456  * have pending delalloc work to be done.
1457  */
1458 static void btrfs_set_bit_hook(struct inode *inode,
1459                                struct extent_state *state, int *bits)
1460 {
1461
1462         /*
1463          * set_bit and clear bit hooks normally require _irqsave/restore
1464          * but in this case, we are only testing for the DELALLOC
1465          * bit, which is only set or cleared with irqs on
1466          */
1467         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1468                 struct btrfs_root *root = BTRFS_I(inode)->root;
1469                 u64 len = state->end + 1 - state->start;
1470                 bool do_list = !btrfs_is_free_space_inode(inode);
1471
1472                 if (*bits & EXTENT_FIRST_DELALLOC) {
1473                         *bits &= ~EXTENT_FIRST_DELALLOC;
1474                 } else {
1475                         spin_lock(&BTRFS_I(inode)->lock);
1476                         BTRFS_I(inode)->outstanding_extents++;
1477                         spin_unlock(&BTRFS_I(inode)->lock);
1478                 }
1479
1480                 spin_lock(&root->fs_info->delalloc_lock);
1481                 BTRFS_I(inode)->delalloc_bytes += len;
1482                 root->fs_info->delalloc_bytes += len;
1483                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1484                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1485                                       &root->fs_info->delalloc_inodes);
1486                 }
1487                 spin_unlock(&root->fs_info->delalloc_lock);
1488         }
1489 }
1490
1491 /*
1492  * extent_io.c clear_bit_hook, see set_bit_hook for why
1493  */
1494 static void btrfs_clear_bit_hook(struct inode *inode,
1495                                  struct extent_state *state, int *bits)
1496 {
1497         /*
1498          * set_bit and clear bit hooks normally require _irqsave/restore
1499          * but in this case, we are only testing for the DELALLOC
1500          * bit, which is only set or cleared with irqs on
1501          */
1502         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1503                 struct btrfs_root *root = BTRFS_I(inode)->root;
1504                 u64 len = state->end + 1 - state->start;
1505                 bool do_list = !btrfs_is_free_space_inode(inode);
1506
1507                 if (*bits & EXTENT_FIRST_DELALLOC) {
1508                         *bits &= ~EXTENT_FIRST_DELALLOC;
1509                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1510                         spin_lock(&BTRFS_I(inode)->lock);
1511                         BTRFS_I(inode)->outstanding_extents--;
1512                         spin_unlock(&BTRFS_I(inode)->lock);
1513                 }
1514
1515                 if (*bits & EXTENT_DO_ACCOUNTING)
1516                         btrfs_delalloc_release_metadata(inode, len);
1517
1518                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1519                     && do_list)
1520                         btrfs_free_reserved_data_space(inode, len);
1521
1522                 spin_lock(&root->fs_info->delalloc_lock);
1523                 root->fs_info->delalloc_bytes -= len;
1524                 BTRFS_I(inode)->delalloc_bytes -= len;
1525
1526                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1527                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1528                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1529                 }
1530                 spin_unlock(&root->fs_info->delalloc_lock);
1531         }
1532 }
1533
1534 /*
1535  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1536  * we don't create bios that span stripes or chunks
1537  */
1538 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1539                          size_t size, struct bio *bio,
1540                          unsigned long bio_flags)
1541 {
1542         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1543         struct btrfs_mapping_tree *map_tree;
1544         u64 logical = (u64)bio->bi_sector << 9;
1545         u64 length = 0;
1546         u64 map_length;
1547         int ret;
1548
1549         if (bio_flags & EXTENT_BIO_COMPRESSED)
1550                 return 0;
1551
1552         length = bio->bi_size;
1553         map_tree = &root->fs_info->mapping_tree;
1554         map_length = length;
1555         ret = btrfs_map_block(map_tree, READ, logical,
1556                               &map_length, NULL, 0);
1557         /* Will always return 0 or 1 with map_multi == NULL */
1558         BUG_ON(ret < 0);
1559         if (map_length < length + size)
1560                 return 1;
1561         return 0;
1562 }
1563
1564 /*
1565  * in order to insert checksums into the metadata in large chunks,
1566  * we wait until bio submission time.   All the pages in the bio are
1567  * checksummed and sums are attached onto the ordered extent record.
1568  *
1569  * At IO completion time the cums attached on the ordered extent record
1570  * are inserted into the btree
1571  */
1572 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1573                                     struct bio *bio, int mirror_num,
1574                                     unsigned long bio_flags,
1575                                     u64 bio_offset)
1576 {
1577         struct btrfs_root *root = BTRFS_I(inode)->root;
1578         int ret = 0;
1579
1580         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1581         BUG_ON(ret); /* -ENOMEM */
1582         return 0;
1583 }
1584
1585 /*
1586  * in order to insert checksums into the metadata in large chunks,
1587  * we wait until bio submission time.   All the pages in the bio are
1588  * checksummed and sums are attached onto the ordered extent record.
1589  *
1590  * At IO completion time the cums attached on the ordered extent record
1591  * are inserted into the btree
1592  */
1593 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1594                           int mirror_num, unsigned long bio_flags,
1595                           u64 bio_offset)
1596 {
1597         struct btrfs_root *root = BTRFS_I(inode)->root;
1598         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1599 }
1600
1601 /*
1602  * extent_io.c submission hook. This does the right thing for csum calculation
1603  * on write, or reading the csums from the tree before a read
1604  */
1605 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1606                           int mirror_num, unsigned long bio_flags,
1607                           u64 bio_offset)
1608 {
1609         struct btrfs_root *root = BTRFS_I(inode)->root;
1610         int ret = 0;
1611         int skip_sum;
1612         int metadata = 0;
1613
1614         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1615
1616         if (btrfs_is_free_space_inode(inode))
1617                 metadata = 2;
1618
1619         if (!(rw & REQ_WRITE)) {
1620                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1621                 if (ret)
1622                         return ret;
1623
1624                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1625                         return btrfs_submit_compressed_read(inode, bio,
1626                                                     mirror_num, bio_flags);
1627                 } else if (!skip_sum) {
1628                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1629                         if (ret)
1630                                 return ret;
1631                 }
1632                 goto mapit;
1633         } else if (!skip_sum) {
1634                 /* csum items have already been cloned */
1635                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1636                         goto mapit;
1637                 /* we're doing a write, do the async checksumming */
1638                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1639                                    inode, rw, bio, mirror_num,
1640                                    bio_flags, bio_offset,
1641                                    __btrfs_submit_bio_start,
1642                                    __btrfs_submit_bio_done);
1643         }
1644
1645 mapit:
1646         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1647 }
1648
1649 /*
1650  * given a list of ordered sums record them in the inode.  This happens
1651  * at IO completion time based on sums calculated at bio submission time.
1652  */
1653 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1654                              struct inode *inode, u64 file_offset,
1655                              struct list_head *list)
1656 {
1657         struct btrfs_ordered_sum *sum;
1658
1659         list_for_each_entry(sum, list, list) {
1660                 btrfs_csum_file_blocks(trans,
1661                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1662         }
1663         return 0;
1664 }
1665
1666 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1667                               struct extent_state **cached_state)
1668 {
1669         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1670                 WARN_ON(1);
1671         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1672                                    cached_state, GFP_NOFS);
1673 }
1674
1675 /* see btrfs_writepage_start_hook for details on why this is required */
1676 struct btrfs_writepage_fixup {
1677         struct page *page;
1678         struct btrfs_work work;
1679 };
1680
1681 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1682 {
1683         struct btrfs_writepage_fixup *fixup;
1684         struct btrfs_ordered_extent *ordered;
1685         struct extent_state *cached_state = NULL;
1686         struct page *page;
1687         struct inode *inode;
1688         u64 page_start;
1689         u64 page_end;
1690         int ret;
1691
1692         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1693         page = fixup->page;
1694 again:
1695         lock_page(page);
1696         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1697                 ClearPageChecked(page);
1698                 goto out_page;
1699         }
1700
1701         inode = page->mapping->host;
1702         page_start = page_offset(page);
1703         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1704
1705         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1706                          &cached_state);
1707
1708         /* already ordered? We're done */
1709         if (PagePrivate2(page))
1710                 goto out;
1711
1712         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1713         if (ordered) {
1714                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1715                                      page_end, &cached_state, GFP_NOFS);
1716                 unlock_page(page);
1717                 btrfs_start_ordered_extent(inode, ordered, 1);
1718                 btrfs_put_ordered_extent(ordered);
1719                 goto again;
1720         }
1721
1722         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1723         if (ret) {
1724                 mapping_set_error(page->mapping, ret);
1725                 end_extent_writepage(page, ret, page_start, page_end);
1726                 ClearPageChecked(page);
1727                 goto out;
1728          }
1729
1730         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1731         ClearPageChecked(page);
1732         set_page_dirty(page);
1733 out:
1734         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1735                              &cached_state, GFP_NOFS);
1736 out_page:
1737         unlock_page(page);
1738         page_cache_release(page);
1739         kfree(fixup);
1740 }
1741
1742 /*
1743  * There are a few paths in the higher layers of the kernel that directly
1744  * set the page dirty bit without asking the filesystem if it is a
1745  * good idea.  This causes problems because we want to make sure COW
1746  * properly happens and the data=ordered rules are followed.
1747  *
1748  * In our case any range that doesn't have the ORDERED bit set
1749  * hasn't been properly setup for IO.  We kick off an async process
1750  * to fix it up.  The async helper will wait for ordered extents, set
1751  * the delalloc bit and make it safe to write the page.
1752  */
1753 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1754 {
1755         struct inode *inode = page->mapping->host;
1756         struct btrfs_writepage_fixup *fixup;
1757         struct btrfs_root *root = BTRFS_I(inode)->root;
1758
1759         /* this page is properly in the ordered list */
1760         if (TestClearPagePrivate2(page))
1761                 return 0;
1762
1763         if (PageChecked(page))
1764                 return -EAGAIN;
1765
1766         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1767         if (!fixup)
1768                 return -EAGAIN;
1769
1770         SetPageChecked(page);
1771         page_cache_get(page);
1772         fixup->work.func = btrfs_writepage_fixup_worker;
1773         fixup->page = page;
1774         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1775         return -EBUSY;
1776 }
1777
1778 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1779                                        struct inode *inode, u64 file_pos,
1780                                        u64 disk_bytenr, u64 disk_num_bytes,
1781                                        u64 num_bytes, u64 ram_bytes,
1782                                        u8 compression, u8 encryption,
1783                                        u16 other_encoding, int extent_type)
1784 {
1785         struct btrfs_root *root = BTRFS_I(inode)->root;
1786         struct btrfs_file_extent_item *fi;
1787         struct btrfs_path *path;
1788         struct extent_buffer *leaf;
1789         struct btrfs_key ins;
1790         u64 hint;
1791         int ret;
1792
1793         path = btrfs_alloc_path();
1794         if (!path)
1795                 return -ENOMEM;
1796
1797         path->leave_spinning = 1;
1798
1799         /*
1800          * we may be replacing one extent in the tree with another.
1801          * The new extent is pinned in the extent map, and we don't want
1802          * to drop it from the cache until it is completely in the btree.
1803          *
1804          * So, tell btrfs_drop_extents to leave this extent in the cache.
1805          * the caller is expected to unpin it and allow it to be merged
1806          * with the others.
1807          */
1808         ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1809                                  &hint, 0);
1810         if (ret)
1811                 goto out;
1812
1813         ins.objectid = btrfs_ino(inode);
1814         ins.offset = file_pos;
1815         ins.type = BTRFS_EXTENT_DATA_KEY;
1816         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1817         if (ret)
1818                 goto out;
1819         leaf = path->nodes[0];
1820         fi = btrfs_item_ptr(leaf, path->slots[0],
1821                             struct btrfs_file_extent_item);
1822         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1823         btrfs_set_file_extent_type(leaf, fi, extent_type);
1824         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1825         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1826         btrfs_set_file_extent_offset(leaf, fi, 0);
1827         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1828         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1829         btrfs_set_file_extent_compression(leaf, fi, compression);
1830         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1831         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1832
1833         btrfs_unlock_up_safe(path, 1);
1834         btrfs_set_lock_blocking(leaf);
1835
1836         btrfs_mark_buffer_dirty(leaf);
1837
1838         inode_add_bytes(inode, num_bytes);
1839
1840         ins.objectid = disk_bytenr;
1841         ins.offset = disk_num_bytes;
1842         ins.type = BTRFS_EXTENT_ITEM_KEY;
1843         ret = btrfs_alloc_reserved_file_extent(trans, root,
1844                                         root->root_key.objectid,
1845                                         btrfs_ino(inode), file_pos, &ins);
1846 out:
1847         btrfs_free_path(path);
1848
1849         return ret;
1850 }
1851
1852 /*
1853  * helper function for btrfs_finish_ordered_io, this
1854  * just reads in some of the csum leaves to prime them into ram
1855  * before we start the transaction.  It limits the amount of btree
1856  * reads required while inside the transaction.
1857  */
1858 /* as ordered data IO finishes, this gets called so we can finish
1859  * an ordered extent if the range of bytes in the file it covers are
1860  * fully written.
1861  */
1862 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
1863 {
1864         struct inode *inode = ordered_extent->inode;
1865         struct btrfs_root *root = BTRFS_I(inode)->root;
1866         struct btrfs_trans_handle *trans = NULL;
1867         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1868         struct extent_state *cached_state = NULL;
1869         int compress_type = 0;
1870         int ret;
1871         bool nolock;
1872
1873         nolock = btrfs_is_free_space_inode(inode);
1874
1875         if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
1876                 ret = -EIO;
1877                 goto out;
1878         }
1879
1880         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1881                 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
1882                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1883                 if (!ret) {
1884                         if (nolock)
1885                                 trans = btrfs_join_transaction_nolock(root);
1886                         else
1887                                 trans = btrfs_join_transaction(root);
1888                         if (IS_ERR(trans))
1889                                 return PTR_ERR(trans);
1890                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1891                         ret = btrfs_update_inode_fallback(trans, root, inode);
1892                         if (ret) /* -ENOMEM or corruption */
1893                                 btrfs_abort_transaction(trans, root, ret);
1894                 }
1895                 goto out;
1896         }
1897
1898         lock_extent_bits(io_tree, ordered_extent->file_offset,
1899                          ordered_extent->file_offset + ordered_extent->len - 1,
1900                          0, &cached_state);
1901
1902         if (nolock)
1903                 trans = btrfs_join_transaction_nolock(root);
1904         else
1905                 trans = btrfs_join_transaction(root);
1906         if (IS_ERR(trans)) {
1907                 ret = PTR_ERR(trans);
1908                 trans = NULL;
1909                 goto out_unlock;
1910         }
1911         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1912
1913         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1914                 compress_type = ordered_extent->compress_type;
1915         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1916                 BUG_ON(compress_type);
1917                 ret = btrfs_mark_extent_written(trans, inode,
1918                                                 ordered_extent->file_offset,
1919                                                 ordered_extent->file_offset +
1920                                                 ordered_extent->len);
1921         } else {
1922                 BUG_ON(root == root->fs_info->tree_root);
1923                 ret = insert_reserved_file_extent(trans, inode,
1924                                                 ordered_extent->file_offset,
1925                                                 ordered_extent->start,
1926                                                 ordered_extent->disk_len,
1927                                                 ordered_extent->len,
1928                                                 ordered_extent->len,
1929                                                 compress_type, 0, 0,
1930                                                 BTRFS_FILE_EXTENT_REG);
1931                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1932                                    ordered_extent->file_offset,
1933                                    ordered_extent->len);
1934         }
1935
1936         if (ret < 0) {
1937                 btrfs_abort_transaction(trans, root, ret);
1938                 goto out_unlock;
1939         }
1940
1941         add_pending_csums(trans, inode, ordered_extent->file_offset,
1942                           &ordered_extent->list);
1943
1944         ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1945         if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1946                 ret = btrfs_update_inode_fallback(trans, root, inode);
1947                 if (ret) { /* -ENOMEM or corruption */
1948                         btrfs_abort_transaction(trans, root, ret);
1949                         goto out_unlock;
1950                 }
1951         }
1952         ret = 0;
1953 out_unlock:
1954         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1955                              ordered_extent->file_offset +
1956                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1957 out:
1958         if (root != root->fs_info->tree_root)
1959                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1960         if (trans) {
1961                 if (nolock)
1962                         btrfs_end_transaction_nolock(trans, root);
1963                 else
1964                         btrfs_end_transaction(trans, root);
1965         }
1966
1967         if (ret)
1968                 clear_extent_uptodate(io_tree, ordered_extent->file_offset,
1969                                       ordered_extent->file_offset +
1970                                       ordered_extent->len - 1, NULL, GFP_NOFS);
1971
1972         /*
1973          * This needs to be dont to make sure anybody waiting knows we are done
1974          * upating everything for this ordered extent.
1975          */
1976         btrfs_remove_ordered_extent(inode, ordered_extent);
1977
1978         /* once for us */
1979         btrfs_put_ordered_extent(ordered_extent);
1980         /* once for the tree */
1981         btrfs_put_ordered_extent(ordered_extent);
1982
1983         return ret;
1984 }
1985
1986 static void finish_ordered_fn(struct btrfs_work *work)
1987 {
1988         struct btrfs_ordered_extent *ordered_extent;
1989         ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
1990         btrfs_finish_ordered_io(ordered_extent);
1991 }
1992
1993 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1994                                 struct extent_state *state, int uptodate)
1995 {
1996         struct inode *inode = page->mapping->host;
1997         struct btrfs_root *root = BTRFS_I(inode)->root;
1998         struct btrfs_ordered_extent *ordered_extent = NULL;
1999         struct btrfs_workers *workers;
2000
2001         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2002
2003         ClearPagePrivate2(page);
2004         if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2005                                             end - start + 1, uptodate))
2006                 return 0;
2007
2008         ordered_extent->work.func = finish_ordered_fn;
2009         ordered_extent->work.flags = 0;
2010
2011         if (btrfs_is_free_space_inode(inode))
2012                 workers = &root->fs_info->endio_freespace_worker;
2013         else
2014                 workers = &root->fs_info->endio_write_workers;
2015         btrfs_queue_worker(workers, &ordered_extent->work);
2016
2017         return 0;
2018 }
2019
2020 /*
2021  * when reads are done, we need to check csums to verify the data is correct
2022  * if there's a match, we allow the bio to finish.  If not, the code in
2023  * extent_io.c will try to find good copies for us.
2024  */
2025 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
2026                                struct extent_state *state, int mirror)
2027 {
2028         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
2029         struct inode *inode = page->mapping->host;
2030         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2031         char *kaddr;
2032         u64 private = ~(u32)0;
2033         int ret;
2034         struct btrfs_root *root = BTRFS_I(inode)->root;
2035         u32 csum = ~(u32)0;
2036
2037         if (PageChecked(page)) {
2038                 ClearPageChecked(page);
2039                 goto good;
2040         }
2041
2042         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2043                 goto good;
2044
2045         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2046             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2047                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2048                                   GFP_NOFS);
2049                 return 0;
2050         }
2051
2052         if (state && state->start == start) {
2053                 private = state->private;
2054                 ret = 0;
2055         } else {
2056                 ret = get_state_private(io_tree, start, &private);
2057         }
2058         kaddr = kmap_atomic(page);
2059         if (ret)
2060                 goto zeroit;
2061
2062         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
2063         btrfs_csum_final(csum, (char *)&csum);
2064         if (csum != private)
2065                 goto zeroit;
2066
2067         kunmap_atomic(kaddr);
2068 good:
2069         return 0;
2070
2071 zeroit:
2072         printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
2073                        "private %llu\n",
2074                        (unsigned long long)btrfs_ino(page->mapping->host),
2075                        (unsigned long long)start, csum,
2076                        (unsigned long long)private);
2077         memset(kaddr + offset, 1, end - start + 1);
2078         flush_dcache_page(page);
2079         kunmap_atomic(kaddr);
2080         if (private == 0)
2081                 return 0;
2082         return -EIO;
2083 }
2084
2085 struct delayed_iput {
2086         struct list_head list;
2087         struct inode *inode;
2088 };
2089
2090 /* JDM: If this is fs-wide, why can't we add a pointer to
2091  * btrfs_inode instead and avoid the allocation? */
2092 void btrfs_add_delayed_iput(struct inode *inode)
2093 {
2094         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2095         struct delayed_iput *delayed;
2096
2097         if (atomic_add_unless(&inode->i_count, -1, 1))
2098                 return;
2099
2100         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2101         delayed->inode = inode;
2102
2103         spin_lock(&fs_info->delayed_iput_lock);
2104         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2105         spin_unlock(&fs_info->delayed_iput_lock);
2106 }
2107
2108 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2109 {
2110         LIST_HEAD(list);
2111         struct btrfs_fs_info *fs_info = root->fs_info;
2112         struct delayed_iput *delayed;
2113         int empty;
2114
2115         spin_lock(&fs_info->delayed_iput_lock);
2116         empty = list_empty(&fs_info->delayed_iputs);
2117         spin_unlock(&fs_info->delayed_iput_lock);
2118         if (empty)
2119                 return;
2120
2121         down_read(&root->fs_info->cleanup_work_sem);
2122         spin_lock(&fs_info->delayed_iput_lock);
2123         list_splice_init(&fs_info->delayed_iputs, &list);
2124         spin_unlock(&fs_info->delayed_iput_lock);
2125
2126         while (!list_empty(&list)) {
2127                 delayed = list_entry(list.next, struct delayed_iput, list);
2128                 list_del(&delayed->list);
2129                 iput(delayed->inode);
2130                 kfree(delayed);
2131         }
2132         up_read(&root->fs_info->cleanup_work_sem);
2133 }
2134
2135 enum btrfs_orphan_cleanup_state {
2136         ORPHAN_CLEANUP_STARTED  = 1,
2137         ORPHAN_CLEANUP_DONE     = 2,
2138 };
2139
2140 /*
2141  * This is called in transaction commit time. If there are no orphan
2142  * files in the subvolume, it removes orphan item and frees block_rsv
2143  * structure.
2144  */
2145 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2146                               struct btrfs_root *root)
2147 {
2148         struct btrfs_block_rsv *block_rsv;
2149         int ret;
2150
2151         if (atomic_read(&root->orphan_inodes) ||
2152             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2153                 return;
2154
2155         spin_lock(&root->orphan_lock);
2156         if (atomic_read(&root->orphan_inodes)) {
2157                 spin_unlock(&root->orphan_lock);
2158                 return;
2159         }
2160
2161         if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2162                 spin_unlock(&root->orphan_lock);
2163                 return;
2164         }
2165
2166         block_rsv = root->orphan_block_rsv;
2167         root->orphan_block_rsv = NULL;
2168         spin_unlock(&root->orphan_lock);
2169
2170         if (root->orphan_item_inserted &&
2171             btrfs_root_refs(&root->root_item) > 0) {
2172                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2173                                             root->root_key.objectid);
2174                 BUG_ON(ret);
2175                 root->orphan_item_inserted = 0;
2176         }
2177
2178         if (block_rsv) {
2179                 WARN_ON(block_rsv->size > 0);
2180                 btrfs_free_block_rsv(root, block_rsv);
2181         }
2182 }
2183
2184 /*
2185  * This creates an orphan entry for the given inode in case something goes
2186  * wrong in the middle of an unlink/truncate.
2187  *
2188  * NOTE: caller of this function should reserve 5 units of metadata for
2189  *       this function.
2190  */
2191 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2192 {
2193         struct btrfs_root *root = BTRFS_I(inode)->root;
2194         struct btrfs_block_rsv *block_rsv = NULL;
2195         int reserve = 0;
2196         int insert = 0;
2197         int ret;
2198
2199         if (!root->orphan_block_rsv) {
2200                 block_rsv = btrfs_alloc_block_rsv(root);
2201                 if (!block_rsv)
2202                         return -ENOMEM;
2203         }
2204
2205         spin_lock(&root->orphan_lock);
2206         if (!root->orphan_block_rsv) {
2207                 root->orphan_block_rsv = block_rsv;
2208         } else if (block_rsv) {
2209                 btrfs_free_block_rsv(root, block_rsv);
2210                 block_rsv = NULL;
2211         }
2212
2213         if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2214                               &BTRFS_I(inode)->runtime_flags)) {
2215 #if 0
2216                 /*
2217                  * For proper ENOSPC handling, we should do orphan
2218                  * cleanup when mounting. But this introduces backward
2219                  * compatibility issue.
2220                  */
2221                 if (!xchg(&root->orphan_item_inserted, 1))
2222                         insert = 2;
2223                 else
2224                         insert = 1;
2225 #endif
2226                 insert = 1;
2227                 atomic_dec(&root->orphan_inodes);
2228         }
2229
2230         if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2231                               &BTRFS_I(inode)->runtime_flags))
2232                 reserve = 1;
2233         spin_unlock(&root->orphan_lock);
2234
2235         /* grab metadata reservation from transaction handle */
2236         if (reserve) {
2237                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2238                 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
2239         }
2240
2241         /* insert an orphan item to track this unlinked/truncated file */
2242         if (insert >= 1) {
2243                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2244                 if (ret && ret != -EEXIST) {
2245                         clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2246                                   &BTRFS_I(inode)->runtime_flags);
2247                         btrfs_abort_transaction(trans, root, ret);
2248                         return ret;
2249                 }
2250                 ret = 0;
2251         }
2252
2253         /* insert an orphan item to track subvolume contains orphan files */
2254         if (insert >= 2) {
2255                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2256                                                root->root_key.objectid);
2257                 if (ret && ret != -EEXIST) {
2258                         btrfs_abort_transaction(trans, root, ret);
2259                         return ret;
2260                 }
2261         }
2262         return 0;
2263 }
2264
2265 /*
2266  * We have done the truncate/delete so we can go ahead and remove the orphan
2267  * item for this particular inode.
2268  */
2269 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2270 {
2271         struct btrfs_root *root = BTRFS_I(inode)->root;
2272         int delete_item = 0;
2273         int release_rsv = 0;
2274         int ret = 0;
2275
2276         spin_lock(&root->orphan_lock);
2277         if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2278                                &BTRFS_I(inode)->runtime_flags))
2279                 delete_item = 1;
2280
2281         if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2282                                &BTRFS_I(inode)->runtime_flags))
2283                 release_rsv = 1;
2284         spin_unlock(&root->orphan_lock);
2285
2286         if (trans && delete_item) {
2287                 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2288                 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2289         }
2290
2291         if (release_rsv) {
2292                 btrfs_orphan_release_metadata(inode);
2293                 atomic_dec(&root->orphan_inodes);
2294         }
2295
2296         return 0;
2297 }
2298
2299 /*
2300  * this cleans up any orphans that may be left on the list from the last use
2301  * of this root.
2302  */
2303 int btrfs_orphan_cleanup(struct btrfs_root *root)
2304 {
2305         struct btrfs_path *path;
2306         struct extent_buffer *leaf;
2307         struct btrfs_key key, found_key;
2308         struct btrfs_trans_handle *trans;
2309         struct inode *inode;
2310         u64 last_objectid = 0;
2311         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2312
2313         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2314                 return 0;
2315
2316         path = btrfs_alloc_path();
2317         if (!path) {
2318                 ret = -ENOMEM;
2319                 goto out;
2320         }
2321         path->reada = -1;
2322
2323         key.objectid = BTRFS_ORPHAN_OBJECTID;
2324         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2325         key.offset = (u64)-1;
2326
2327         while (1) {
2328                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2329                 if (ret < 0)
2330                         goto out;
2331
2332                 /*
2333                  * if ret == 0 means we found what we were searching for, which
2334                  * is weird, but possible, so only screw with path if we didn't
2335                  * find the key and see if we have stuff that matches
2336                  */
2337                 if (ret > 0) {
2338                         ret = 0;
2339                         if (path->slots[0] == 0)
2340                                 break;
2341                         path->slots[0]--;
2342                 }
2343
2344                 /* pull out the item */
2345                 leaf = path->nodes[0];
2346                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2347
2348                 /* make sure the item matches what we want */
2349                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2350                         break;
2351                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2352                         break;
2353
2354                 /* release the path since we're done with it */
2355                 btrfs_release_path(path);
2356
2357                 /*
2358                  * this is where we are basically btrfs_lookup, without the
2359                  * crossing root thing.  we store the inode number in the
2360                  * offset of the orphan item.
2361                  */
2362
2363                 if (found_key.offset == last_objectid) {
2364                         printk(KERN_ERR "btrfs: Error removing orphan entry, "
2365                                "stopping orphan cleanup\n");
2366                         ret = -EINVAL;
2367                         goto out;
2368                 }
2369
2370                 last_objectid = found_key.offset;
2371
2372                 found_key.objectid = found_key.offset;
2373                 found_key.type = BTRFS_INODE_ITEM_KEY;
2374                 found_key.offset = 0;
2375                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2376                 ret = PTR_RET(inode);
2377                 if (ret && ret != -ESTALE)
2378                         goto out;
2379
2380                 if (ret == -ESTALE && root == root->fs_info->tree_root) {
2381                         struct btrfs_root *dead_root;
2382                         struct btrfs_fs_info *fs_info = root->fs_info;
2383                         int is_dead_root = 0;
2384
2385                         /*
2386                          * this is an orphan in the tree root. Currently these
2387                          * could come from 2 sources:
2388                          *  a) a snapshot deletion in progress
2389                          *  b) a free space cache inode
2390                          * We need to distinguish those two, as the snapshot
2391                          * orphan must not get deleted.
2392                          * find_dead_roots already ran before us, so if this
2393                          * is a snapshot deletion, we should find the root
2394                          * in the dead_roots list
2395                          */
2396                         spin_lock(&fs_info->trans_lock);
2397                         list_for_each_entry(dead_root, &fs_info->dead_roots,
2398                                             root_list) {
2399                                 if (dead_root->root_key.objectid ==
2400                                     found_key.objectid) {
2401                                         is_dead_root = 1;
2402                                         break;
2403                                 }
2404                         }
2405                         spin_unlock(&fs_info->trans_lock);
2406                         if (is_dead_root) {
2407                                 /* prevent this orphan from being found again */
2408                                 key.offset = found_key.objectid - 1;
2409                                 continue;
2410                         }
2411                 }
2412                 /*
2413                  * Inode is already gone but the orphan item is still there,
2414                  * kill the orphan item.
2415                  */
2416                 if (ret == -ESTALE) {
2417                         trans = btrfs_start_transaction(root, 1);
2418                         if (IS_ERR(trans)) {
2419                                 ret = PTR_ERR(trans);
2420                                 goto out;
2421                         }
2422                         printk(KERN_ERR "auto deleting %Lu\n",
2423                                found_key.objectid);
2424                         ret = btrfs_del_orphan_item(trans, root,
2425                                                     found_key.objectid);
2426                         BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2427                         btrfs_end_transaction(trans, root);
2428                         continue;
2429                 }
2430
2431                 /*
2432                  * add this inode to the orphan list so btrfs_orphan_del does
2433                  * the proper thing when we hit it
2434                  */
2435                 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2436                         &BTRFS_I(inode)->runtime_flags);
2437
2438                 /* if we have links, this was a truncate, lets do that */
2439                 if (inode->i_nlink) {
2440                         if (!S_ISREG(inode->i_mode)) {
2441                                 WARN_ON(1);
2442                                 iput(inode);
2443                                 continue;
2444                         }
2445                         nr_truncate++;
2446                         ret = btrfs_truncate(inode);
2447                 } else {
2448                         nr_unlink++;
2449                 }
2450
2451                 /* this will do delete_inode and everything for us */
2452                 iput(inode);
2453                 if (ret)
2454                         goto out;
2455         }
2456         /* release the path since we're done with it */
2457         btrfs_release_path(path);
2458
2459         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2460
2461         if (root->orphan_block_rsv)
2462                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2463                                         (u64)-1);
2464
2465         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2466                 trans = btrfs_join_transaction(root);
2467                 if (!IS_ERR(trans))
2468                         btrfs_end_transaction(trans, root);
2469         }
2470
2471         if (nr_unlink)
2472                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2473         if (nr_truncate)
2474                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2475
2476 out:
2477         if (ret)
2478                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2479         btrfs_free_path(path);
2480         return ret;
2481 }
2482
2483 /*
2484  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2485  * don't find any xattrs, we know there can't be any acls.
2486  *
2487  * slot is the slot the inode is in, objectid is the objectid of the inode
2488  */
2489 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2490                                           int slot, u64 objectid)
2491 {
2492         u32 nritems = btrfs_header_nritems(leaf);
2493         struct btrfs_key found_key;
2494         int scanned = 0;
2495
2496         slot++;
2497         while (slot < nritems) {
2498                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2499
2500                 /* we found a different objectid, there must not be acls */
2501                 if (found_key.objectid != objectid)
2502                         return 0;
2503
2504                 /* we found an xattr, assume we've got an acl */
2505                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2506                         return 1;
2507
2508                 /*
2509                  * we found a key greater than an xattr key, there can't
2510                  * be any acls later on
2511                  */
2512                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2513                         return 0;
2514
2515                 slot++;
2516                 scanned++;
2517
2518                 /*
2519                  * it goes inode, inode backrefs, xattrs, extents,
2520                  * so if there are a ton of hard links to an inode there can
2521                  * be a lot of backrefs.  Don't waste time searching too hard,
2522                  * this is just an optimization
2523                  */
2524                 if (scanned >= 8)
2525                         break;
2526         }
2527         /* we hit the end of the leaf before we found an xattr or
2528          * something larger than an xattr.  We have to assume the inode
2529          * has acls
2530          */
2531         return 1;
2532 }
2533
2534 /*
2535  * read an inode from the btree into the in-memory inode
2536  */
2537 static void btrfs_read_locked_inode(struct inode *inode)
2538 {
2539         struct btrfs_path *path;
2540         struct extent_buffer *leaf;
2541         struct btrfs_inode_item *inode_item;
2542         struct btrfs_timespec *tspec;
2543         struct btrfs_root *root = BTRFS_I(inode)->root;
2544         struct btrfs_key location;
2545         int maybe_acls;
2546         u32 rdev;
2547         int ret;
2548         bool filled = false;
2549
2550         ret = btrfs_fill_inode(inode, &rdev);
2551         if (!ret)
2552                 filled = true;
2553
2554         path = btrfs_alloc_path();
2555         if (!path)
2556                 goto make_bad;
2557
2558         path->leave_spinning = 1;
2559         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2560
2561         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2562         if (ret)
2563                 goto make_bad;
2564
2565         leaf = path->nodes[0];
2566
2567         if (filled)
2568                 goto cache_acl;
2569
2570         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2571                                     struct btrfs_inode_item);
2572         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2573         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2574         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2575         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2576         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2577
2578         tspec = btrfs_inode_atime(inode_item);
2579         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2580         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2581
2582         tspec = btrfs_inode_mtime(inode_item);
2583         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2584         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2585
2586         tspec = btrfs_inode_ctime(inode_item);
2587         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2588         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2589
2590         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2591         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2592         inode->i_version = btrfs_inode_sequence(leaf, inode_item);
2593         inode->i_generation = BTRFS_I(inode)->generation;
2594         inode->i_rdev = 0;
2595         rdev = btrfs_inode_rdev(leaf, inode_item);
2596
2597         BTRFS_I(inode)->index_cnt = (u64)-1;
2598         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2599 cache_acl:
2600         /*
2601          * try to precache a NULL acl entry for files that don't have
2602          * any xattrs or acls
2603          */
2604         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2605                                            btrfs_ino(inode));
2606         if (!maybe_acls)
2607                 cache_no_acl(inode);
2608
2609         btrfs_free_path(path);
2610
2611         switch (inode->i_mode & S_IFMT) {
2612         case S_IFREG:
2613                 inode->i_mapping->a_ops = &btrfs_aops;
2614                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2615                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2616                 inode->i_fop = &btrfs_file_operations;
2617                 inode->i_op = &btrfs_file_inode_operations;
2618                 break;
2619         case S_IFDIR:
2620                 inode->i_fop = &btrfs_dir_file_operations;
2621                 if (root == root->fs_info->tree_root)
2622                         inode->i_op = &btrfs_dir_ro_inode_operations;
2623                 else
2624                         inode->i_op = &btrfs_dir_inode_operations;
2625                 break;
2626         case S_IFLNK:
2627                 inode->i_op = &btrfs_symlink_inode_operations;
2628                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2629                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2630                 break;
2631         default:
2632                 inode->i_op = &btrfs_special_inode_operations;
2633                 init_special_inode(inode, inode->i_mode, rdev);
2634                 break;
2635         }
2636
2637         btrfs_update_iflags(inode);
2638         return;
2639
2640 make_bad:
2641         btrfs_free_path(path);
2642         make_bad_inode(inode);
2643 }
2644
2645 /*
2646  * given a leaf and an inode, copy the inode fields into the leaf
2647  */
2648 static void fill_inode_item(struct btrfs_trans_handle *trans,
2649                             struct extent_buffer *leaf,
2650                             struct btrfs_inode_item *item,
2651                             struct inode *inode)
2652 {
2653         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2654         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2655         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2656         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2657         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2658
2659         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2660                                inode->i_atime.tv_sec);
2661         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2662                                 inode->i_atime.tv_nsec);
2663
2664         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2665                                inode->i_mtime.tv_sec);
2666         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2667                                 inode->i_mtime.tv_nsec);
2668
2669         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2670                                inode->i_ctime.tv_sec);
2671         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2672                                 inode->i_ctime.tv_nsec);
2673
2674         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2675         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2676         btrfs_set_inode_sequence(leaf, item, inode->i_version);
2677         btrfs_set_inode_transid(leaf, item, trans->transid);
2678         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2679         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2680         btrfs_set_inode_block_group(leaf, item, 0);
2681 }
2682
2683 /*
2684  * copy everything in the in-memory inode into the btree.
2685  */
2686 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
2687                                 struct btrfs_root *root, struct inode *inode)
2688 {
2689         struct btrfs_inode_item *inode_item;
2690         struct btrfs_path *path;
2691         struct extent_buffer *leaf;
2692         int ret;
2693
2694         path = btrfs_alloc_path();
2695         if (!path)
2696                 return -ENOMEM;
2697
2698         path->leave_spinning = 1;
2699         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2700                                  1);
2701         if (ret) {
2702                 if (ret > 0)
2703                         ret = -ENOENT;
2704                 goto failed;
2705         }
2706
2707         btrfs_unlock_up_safe(path, 1);
2708         leaf = path->nodes[0];
2709         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2710                                     struct btrfs_inode_item);
2711
2712         fill_inode_item(trans, leaf, inode_item, inode);
2713         btrfs_mark_buffer_dirty(leaf);
2714         btrfs_set_inode_last_trans(trans, inode);
2715         ret = 0;
2716 failed:
2717         btrfs_free_path(path);
2718         return ret;
2719 }
2720
2721 /*
2722  * copy everything in the in-memory inode into the btree.
2723  */
2724 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2725                                 struct btrfs_root *root, struct inode *inode)
2726 {
2727         int ret;
2728
2729         /*
2730          * If the inode is a free space inode, we can deadlock during commit
2731          * if we put it into the delayed code.
2732          *
2733          * The data relocation inode should also be directly updated
2734          * without delay
2735          */
2736         if (!btrfs_is_free_space_inode(inode)
2737             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2738                 btrfs_update_root_times(trans, root);
2739
2740                 ret = btrfs_delayed_update_inode(trans, root, inode);
2741                 if (!ret)
2742                         btrfs_set_inode_last_trans(trans, inode);
2743                 return ret;
2744         }
2745
2746         return btrfs_update_inode_item(trans, root, inode);
2747 }
2748
2749 static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2750                                 struct btrfs_root *root, struct inode *inode)
2751 {
2752         int ret;
2753
2754         ret = btrfs_update_inode(trans, root, inode);
2755         if (ret == -ENOSPC)
2756                 return btrfs_update_inode_item(trans, root, inode);
2757         return ret;
2758 }
2759
2760 /*
2761  * unlink helper that gets used here in inode.c and in the tree logging
2762  * recovery code.  It remove a link in a directory with a given name, and
2763  * also drops the back refs in the inode to the directory
2764  */
2765 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2766                                 struct btrfs_root *root,
2767                                 struct inode *dir, struct inode *inode,
2768                                 const char *name, int name_len)
2769 {
2770         struct btrfs_path *path;
2771         int ret = 0;
2772         struct extent_buffer *leaf;
2773         struct btrfs_dir_item *di;
2774         struct btrfs_key key;
2775         u64 index;
2776         u64 ino = btrfs_ino(inode);
2777         u64 dir_ino = btrfs_ino(dir);
2778
2779         path = btrfs_alloc_path();
2780         if (!path) {
2781                 ret = -ENOMEM;
2782                 goto out;
2783         }
2784
2785         path->leave_spinning = 1;
2786         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2787                                     name, name_len, -1);
2788         if (IS_ERR(di)) {
2789                 ret = PTR_ERR(di);
2790                 goto err;
2791         }
2792         if (!di) {
2793                 ret = -ENOENT;
2794                 goto err;
2795         }
2796         leaf = path->nodes[0];
2797         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2798         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2799         if (ret)
2800                 goto err;
2801         btrfs_release_path(path);
2802
2803         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2804                                   dir_ino, &index);
2805         if (ret) {
2806                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2807                        "inode %llu parent %llu\n", name_len, name,
2808                        (unsigned long long)ino, (unsigned long long)dir_ino);
2809                 btrfs_abort_transaction(trans, root, ret);
2810                 goto err;
2811         }
2812
2813         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2814         if (ret) {
2815                 btrfs_abort_transaction(trans, root, ret);
2816                 goto err;
2817         }
2818
2819         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2820                                          inode, dir_ino);
2821         if (ret != 0 && ret != -ENOENT) {
2822                 btrfs_abort_transaction(trans, root, ret);
2823                 goto err;
2824         }
2825
2826         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2827                                            dir, index);
2828         if (ret == -ENOENT)
2829                 ret = 0;
2830 err:
2831         btrfs_free_path(path);
2832         if (ret)
2833                 goto out;
2834
2835         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2836         inode_inc_iversion(inode);
2837         inode_inc_iversion(dir);
2838         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2839         ret = btrfs_update_inode(trans, root, dir);
2840 out:
2841         return ret;
2842 }
2843
2844 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2845                        struct btrfs_root *root,
2846                        struct inode *dir, struct inode *inode,
2847                        const char *name, int name_len)
2848 {
2849         int ret;
2850         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2851         if (!ret) {
2852                 btrfs_drop_nlink(inode);
2853                 ret = btrfs_update_inode(trans, root, inode);
2854         }
2855         return ret;
2856 }
2857
2858
2859 /* helper to check if there is any shared block in the path */
2860 static int check_path_shared(struct btrfs_root *root,
2861                              struct btrfs_path *path)
2862 {
2863         struct extent_buffer *eb;
2864         int level;
2865         u64 refs = 1;
2866
2867         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2868                 int ret;
2869
2870                 if (!path->nodes[level])
2871                         break;
2872                 eb = path->nodes[level];
2873                 if (!btrfs_block_can_be_shared(root, eb))
2874                         continue;
2875                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2876                                                &refs, NULL);
2877                 if (refs > 1)
2878                         return 1;
2879         }
2880         return 0;
2881 }
2882
2883 /*
2884  * helper to start transaction for unlink and rmdir.
2885  *
2886  * unlink and rmdir are special in btrfs, they do not always free space.
2887  * so in enospc case, we should make sure they will free space before
2888  * allowing them to use the global metadata reservation.
2889  */
2890 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2891                                                        struct dentry *dentry)
2892 {
2893         struct btrfs_trans_handle *trans;
2894         struct btrfs_root *root = BTRFS_I(dir)->root;
2895         struct btrfs_path *path;
2896         struct btrfs_inode_ref *ref;
2897         struct btrfs_dir_item *di;
2898         struct inode *inode = dentry->d_inode;
2899         u64 index;
2900         int check_link = 1;
2901         int err = -ENOSPC;
2902         int ret;
2903         u64 ino = btrfs_ino(inode);
2904         u64 dir_ino = btrfs_ino(dir);
2905
2906         /*
2907          * 1 for the possible orphan item
2908          * 1 for the dir item
2909          * 1 for the dir index
2910          * 1 for the inode ref
2911          * 1 for the inode ref in the tree log
2912          * 2 for the dir entries in the log
2913          * 1 for the inode
2914          */
2915         trans = btrfs_start_transaction(root, 8);
2916         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2917                 return trans;
2918
2919         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2920                 return ERR_PTR(-ENOSPC);
2921
2922         /* check if there is someone else holds reference */
2923         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2924                 return ERR_PTR(-ENOSPC);
2925
2926         if (atomic_read(&inode->i_count) > 2)
2927                 return ERR_PTR(-ENOSPC);
2928
2929         if (xchg(&root->fs_info->enospc_unlink, 1))
2930                 return ERR_PTR(-ENOSPC);
2931
2932         path = btrfs_alloc_path();
2933         if (!path) {
2934                 root->fs_info->enospc_unlink = 0;
2935                 return ERR_PTR(-ENOMEM);
2936         }
2937
2938         /* 1 for the orphan item */
2939         trans = btrfs_start_transaction(root, 1);
2940         if (IS_ERR(trans)) {
2941                 btrfs_free_path(path);
2942                 root->fs_info->enospc_unlink = 0;
2943                 return trans;
2944         }
2945
2946         path->skip_locking = 1;
2947         path->search_commit_root = 1;
2948
2949         ret = btrfs_lookup_inode(trans, root, path,
2950                                 &BTRFS_I(dir)->location, 0);
2951         if (ret < 0) {
2952                 err = ret;
2953                 goto out;
2954         }
2955         if (ret == 0) {
2956                 if (check_path_shared(root, path))
2957                         goto out;
2958         } else {
2959                 check_link = 0;
2960         }
2961         btrfs_release_path(path);
2962
2963         ret = btrfs_lookup_inode(trans, root, path,
2964                                 &BTRFS_I(inode)->location, 0);
2965         if (ret < 0) {
2966                 err = ret;
2967                 goto out;
2968         }
2969         if (ret == 0) {
2970                 if (check_path_shared(root, path))
2971                         goto out;
2972         } else {
2973                 check_link = 0;
2974         }
2975         btrfs_release_path(path);
2976
2977         if (ret == 0 && S_ISREG(inode->i_mode)) {
2978                 ret = btrfs_lookup_file_extent(trans, root, path,
2979                                                ino, (u64)-1, 0);
2980                 if (ret < 0) {
2981                         err = ret;
2982                         goto out;
2983                 }
2984                 BUG_ON(ret == 0); /* Corruption */
2985                 if (check_path_shared(root, path))
2986                         goto out;
2987                 btrfs_release_path(path);
2988         }
2989
2990         if (!check_link) {
2991                 err = 0;
2992                 goto out;
2993         }
2994
2995         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2996                                 dentry->d_name.name, dentry->d_name.len, 0);
2997         if (IS_ERR(di)) {
2998                 err = PTR_ERR(di);
2999                 goto out;
3000         }
3001         if (di) {
3002                 if (check_path_shared(root, path))
3003                         goto out;
3004         } else {
3005                 err = 0;
3006                 goto out;
3007         }
3008         btrfs_release_path(path);
3009
3010         ref = btrfs_lookup_inode_ref(trans, root, path,
3011                                 dentry->d_name.name, dentry->d_name.len,
3012                                 ino, dir_ino, 0);
3013         if (IS_ERR(ref)) {
3014                 err = PTR_ERR(ref);
3015                 goto out;
3016         }
3017         BUG_ON(!ref); /* Logic error */
3018         if (check_path_shared(root, path))
3019                 goto out;
3020         index = btrfs_inode_ref_index(path->nodes[0], ref);
3021         btrfs_release_path(path);
3022
3023         /*
3024          * This is a commit root search, if we can lookup inode item and other
3025          * relative items in the commit root, it means the transaction of
3026          * dir/file creation has been committed, and the dir index item that we
3027          * delay to insert has also been inserted into the commit root. So
3028          * we needn't worry about the delayed insertion of the dir index item
3029          * here.
3030          */
3031         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
3032                                 dentry->d_name.name, dentry->d_name.len, 0);
3033         if (IS_ERR(di)) {
3034                 err = PTR_ERR(di);
3035                 goto out;
3036         }
3037         BUG_ON(ret == -ENOENT);
3038         if (check_path_shared(root, path))
3039                 goto out;
3040
3041         err = 0;
3042 out:
3043         btrfs_free_path(path);
3044         /* Migrate the orphan reservation over */
3045         if (!err)
3046                 err = btrfs_block_rsv_migrate(trans->block_rsv,
3047                                 &root->fs_info->global_block_rsv,
3048                                 trans->bytes_reserved);
3049
3050         if (err) {
3051                 btrfs_end_transaction(trans, root);
3052                 root->fs_info->enospc_unlink = 0;
3053                 return ERR_PTR(err);
3054         }
3055
3056         trans->block_rsv = &root->fs_info->global_block_rsv;
3057         return trans;
3058 }
3059
3060 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
3061                                struct btrfs_root *root)
3062 {
3063         if (trans->block_rsv == &root->fs_info->global_block_rsv) {
3064                 btrfs_block_rsv_release(root, trans->block_rsv,
3065                                         trans->bytes_reserved);
3066                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3067                 BUG_ON(!root->fs_info->enospc_unlink);
3068                 root->fs_info->enospc_unlink = 0;
3069         }
3070         btrfs_end_transaction(trans, root);
3071 }
3072
3073 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3074 {
3075         struct btrfs_root *root = BTRFS_I(dir)->root;
3076         struct btrfs_trans_handle *trans;
3077         struct inode *inode = dentry->d_inode;
3078         int ret;
3079         unsigned long nr = 0;
3080
3081         trans = __unlink_start_trans(dir, dentry);
3082         if (IS_ERR(trans))
3083                 return PTR_ERR(trans);
3084
3085         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3086
3087         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3088                                  dentry->d_name.name, dentry->d_name.len);
3089         if (ret)
3090                 goto out;
3091
3092         if (inode->i_nlink == 0) {
3093                 ret = btrfs_orphan_add(trans, inode);
3094                 if (ret)
3095                         goto out;
3096         }
3097
3098 out:
3099         nr = trans->blocks_used;
3100         __unlink_end_trans(trans, root);
3101         btrfs_btree_balance_dirty(root, nr);
3102         return ret;
3103 }
3104
3105 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3106                         struct btrfs_root *root,
3107                         struct inode *dir, u64 objectid,
3108                         const char *name, int name_len)
3109 {
3110         struct btrfs_path *path;
3111         struct extent_buffer *leaf;
3112         struct btrfs_dir_item *di;
3113         struct btrfs_key key;
3114         u64 index;
3115         int ret;
3116         u64 dir_ino = btrfs_ino(dir);
3117
3118         path = btrfs_alloc_path();
3119         if (!path)
3120                 return -ENOMEM;
3121
3122         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3123                                    name, name_len, -1);
3124         if (IS_ERR_OR_NULL(di)) {
3125                 if (!di)
3126                         ret = -ENOENT;
3127                 else
3128                         ret = PTR_ERR(di);
3129                 goto out;
3130         }
3131
3132         leaf = path->nodes[0];
3133         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3134         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3135         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3136         if (ret) {
3137                 btrfs_abort_transaction(trans, root, ret);
3138                 goto out;
3139         }
3140         btrfs_release_path(path);
3141
3142         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3143                                  objectid, root->root_key.objectid,
3144                                  dir_ino, &index, name, name_len);
3145         if (ret < 0) {
3146                 if (ret != -ENOENT) {
3147                         btrfs_abort_transaction(trans, root, ret);
3148                         goto out;
3149                 }
3150                 di = btrfs_search_dir_index_item(root, path, dir_ino,
3151                                                  name, name_len);
3152                 if (IS_ERR_OR_NULL(di)) {
3153                         if (!di)
3154                                 ret = -ENOENT;
3155                         else
3156                                 ret = PTR_ERR(di);
3157                         btrfs_abort_transaction(trans, root, ret);
3158                         goto out;
3159                 }
3160
3161                 leaf = path->nodes[0];
3162                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3163                 btrfs_release_path(path);
3164                 index = key.offset;
3165         }
3166         btrfs_release_path(path);
3167
3168         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3169         if (ret) {
3170                 btrfs_abort_transaction(trans, root, ret);
3171                 goto out;
3172         }
3173
3174         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3175         inode_inc_iversion(dir);
3176         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3177         ret = btrfs_update_inode(trans, root, dir);
3178         if (ret)
3179                 btrfs_abort_transaction(trans, root, ret);
3180 out:
3181         btrfs_free_path(path);
3182         return ret;
3183 }
3184
3185 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3186 {
3187         struct inode *inode = dentry->d_inode;
3188         int err = 0;
3189         struct btrfs_root *root = BTRFS_I(dir)->root;
3190         struct btrfs_trans_handle *trans;
3191         unsigned long nr = 0;
3192
3193         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3194             btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3195                 return -ENOTEMPTY;
3196
3197         trans = __unlink_start_trans(dir, dentry);
3198         if (IS_ERR(trans))
3199                 return PTR_ERR(trans);
3200
3201         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3202                 err = btrfs_unlink_subvol(trans, root, dir,
3203                                           BTRFS_I(inode)->location.objectid,
3204                                           dentry->d_name.name,
3205                                           dentry->d_name.len);
3206                 goto out;
3207         }
3208
3209         err = btrfs_orphan_add(trans, inode);
3210         if (err)
3211                 goto out;
3212
3213         /* now the directory is empty */
3214         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3215                                  dentry->d_name.name, dentry->d_name.len);
3216         if (!err)
3217                 btrfs_i_size_write(inode, 0);
3218 out:
3219         nr = trans->blocks_used;
3220         __unlink_end_trans(trans, root);
3221         btrfs_btree_balance_dirty(root, nr);
3222
3223         return err;
3224 }
3225
3226 /*
3227  * this can truncate away extent items, csum items and directory items.
3228  * It starts at a high offset and removes keys until it can't find
3229  * any higher than new_size
3230  *
3231  * csum items that cross the new i_size are truncated to the new size
3232  * as well.
3233  *
3234  * min_type is the minimum key type to truncate down to.  If set to 0, this
3235  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3236  */
3237 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3238                                struct btrfs_root *root,
3239                                struct inode *inode,
3240                                u64 new_size, u32 min_type)
3241 {
3242         struct btrfs_path *path;
3243         struct extent_buffer *leaf;
3244         struct btrfs_file_extent_item *fi;
3245         struct btrfs_key key;
3246         struct btrfs_key found_key;
3247         u64 extent_start = 0;
3248         u64 extent_num_bytes = 0;
3249         u64 extent_offset = 0;
3250         u64 item_end = 0;
3251         u64 mask = root->sectorsize - 1;
3252         u32 found_type = (u8)-1;
3253         int found_extent;
3254         int del_item;
3255         int pending_del_nr = 0;
3256         int pending_del_slot = 0;
3257         int extent_type = -1;
3258         int ret;
3259         int err = 0;
3260         u64 ino = btrfs_ino(inode);
3261
3262         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3263
3264         path = btrfs_alloc_path();
3265         if (!path)
3266                 return -ENOMEM;
3267         path->reada = -1;
3268
3269         if (root->ref_cows || root == root->fs_info->tree_root)
3270                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3271
3272         /*
3273          * This function is also used to drop the items in the log tree before
3274          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3275          * it is used to drop the loged items. So we shouldn't kill the delayed
3276          * items.
3277          */
3278         if (min_type == 0 && root == BTRFS_I(inode)->root)
3279                 btrfs_kill_delayed_inode_items(inode);
3280
3281         key.objectid = ino;
3282         key.offset = (u64)-1;
3283         key.type = (u8)-1;
3284
3285 search_again:
3286         path->leave_spinning = 1;
3287         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3288         if (ret < 0) {
3289                 err = ret;
3290                 goto out;
3291         }
3292
3293         if (ret > 0) {
3294                 /* there are no items in the tree for us to truncate, we're
3295                  * done
3296                  */
3297                 if (path->slots[0] == 0)
3298                         goto out;
3299                 path->slots[0]--;
3300         }
3301
3302         while (1) {
3303                 fi = NULL;
3304                 leaf = path->nodes[0];
3305                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3306                 found_type = btrfs_key_type(&found_key);
3307
3308                 if (found_key.objectid != ino)
3309                         break;
3310
3311                 if (found_type < min_type)
3312                         break;
3313
3314                 item_end = found_key.offset;
3315                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3316                         fi = btrfs_item_ptr(leaf, path->slots[0],
3317                                             struct btrfs_file_extent_item);
3318                         extent_type = btrfs_file_extent_type(leaf, fi);
3319                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3320                                 item_end +=
3321                                     btrfs_file_extent_num_bytes(leaf, fi);
3322                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3323                                 item_end += btrfs_file_extent_inline_len(leaf,
3324                                                                          fi);
3325                         }
3326                         item_end--;
3327                 }
3328                 if (found_type > min_type) {
3329                         del_item = 1;
3330                 } else {
3331                         if (item_end < new_size)
3332                                 break;
3333                         if (found_key.offset >= new_size)
3334                                 del_item = 1;
3335                         else
3336                                 del_item = 0;
3337                 }
3338                 found_extent = 0;
3339                 /* FIXME, shrink the extent if the ref count is only 1 */
3340                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3341                         goto delete;
3342
3343                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3344                         u64 num_dec;
3345                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3346                         if (!del_item) {
3347                                 u64 orig_num_bytes =
3348                                         btrfs_file_extent_num_bytes(leaf, fi);
3349                                 extent_num_bytes = new_size -
3350                                         found_key.offset + root->sectorsize - 1;
3351                                 extent_num_bytes = extent_num_bytes &
3352                                         ~((u64)root->sectorsize - 1);
3353                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3354                                                          extent_num_bytes);
3355                                 num_dec = (orig_num_bytes -
3356                                            extent_num_bytes);
3357                                 if (root->ref_cows && extent_start != 0)
3358                                         inode_sub_bytes(inode, num_dec);
3359                                 btrfs_mark_buffer_dirty(leaf);
3360                         } else {
3361                                 extent_num_bytes =
3362                                         btrfs_file_extent_disk_num_bytes(leaf,
3363                                                                          fi);
3364                                 extent_offset = found_key.offset -
3365                                         btrfs_file_extent_offset(leaf, fi);
3366
3367                                 /* FIXME blocksize != 4096 */
3368                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3369                                 if (extent_start != 0) {
3370                                         found_extent = 1;
3371                                         if (root->ref_cows)
3372                                                 inode_sub_bytes(inode, num_dec);
3373                                 }
3374                         }
3375                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3376                         /*
3377                          * we can't truncate inline items that have had
3378                          * special encodings
3379                          */
3380                         if (!del_item &&
3381                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3382                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3383                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3384                                 u32 size = new_size - found_key.offset;
3385
3386                                 if (root->ref_cows) {
3387                                         inode_sub_bytes(inode, item_end + 1 -
3388                                                         new_size);
3389                                 }
3390                                 size =
3391                                     btrfs_file_extent_calc_inline_size(size);
3392                                 btrfs_truncate_item(trans, root, path,
3393                                                     size, 1);
3394                         } else if (root->ref_cows) {
3395                                 inode_sub_bytes(inode, item_end + 1 -
3396                                                 found_key.offset);
3397                         }
3398                 }
3399 delete:
3400                 if (del_item) {
3401                         if (!pending_del_nr) {
3402                                 /* no pending yet, add ourselves */
3403                                 pending_del_slot = path->slots[0];
3404                                 pending_del_nr = 1;
3405                         } else if (pending_del_nr &&
3406                                    path->slots[0] + 1 == pending_del_slot) {
3407                                 /* hop on the pending chunk */
3408                                 pending_del_nr++;
3409                                 pending_del_slot = path->slots[0];
3410                         } else {
3411                                 BUG();
3412                         }
3413                 } else {
3414                         break;
3415                 }
3416                 if (found_extent && (root->ref_cows ||
3417                                      root == root->fs_info->tree_root)) {
3418                         btrfs_set_path_blocking(path);
3419                         ret = btrfs_free_extent(trans, root, extent_start,
3420                                                 extent_num_bytes, 0,
3421                                                 btrfs_header_owner(leaf),
3422                                                 ino, extent_offset, 0);
3423                         BUG_ON(ret);
3424                 }
3425
3426                 if (found_type == BTRFS_INODE_ITEM_KEY)
3427                         break;
3428
3429                 if (path->slots[0] == 0 ||
3430                     path->slots[0] != pending_del_slot) {
3431                         if (root->ref_cows &&
3432                             BTRFS_I(inode)->location.objectid !=
3433                                                 BTRFS_FREE_INO_OBJECTID) {
3434                                 err = -EAGAIN;
3435                                 goto out;
3436                         }
3437                         if (pending_del_nr) {
3438                                 ret = btrfs_del_items(trans, root, path,
3439                                                 pending_del_slot,
3440                                                 pending_del_nr);
3441                                 if (ret) {
3442                                         btrfs_abort_transaction(trans,
3443                                                                 root, ret);
3444                                         goto error;
3445                                 }
3446                                 pending_del_nr = 0;
3447                         }
3448                         btrfs_release_path(path);
3449                         goto search_again;
3450                 } else {
3451                         path->slots[0]--;
3452                 }
3453         }
3454 out:
3455         if (pending_del_nr) {
3456                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3457                                       pending_del_nr);
3458                 if (ret)
3459                         btrfs_abort_transaction(trans, root, ret);
3460         }
3461 error:
3462         btrfs_free_path(path);
3463         return err;
3464 }
3465
3466 /*
3467  * taken from block_truncate_page, but does cow as it zeros out
3468  * any bytes left in the last page in the file.
3469  */
3470 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3471 {
3472         struct inode *inode = mapping->host;
3473         struct btrfs_root *root = BTRFS_I(inode)->root;
3474         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3475         struct btrfs_ordered_extent *ordered;
3476         struct extent_state *cached_state = NULL;
3477         char *kaddr;
3478         u32 blocksize = root->sectorsize;
3479         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3480         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3481         struct page *page;
3482         gfp_t mask = btrfs_alloc_write_mask(mapping);
3483         int ret = 0;
3484         u64 page_start;
3485         u64 page_end;
3486
3487         if ((offset & (blocksize - 1)) == 0)
3488                 goto out;
3489         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3490         if (ret)
3491                 goto out;
3492
3493         ret = -ENOMEM;
3494 again:
3495         page = find_or_create_page(mapping, index, mask);
3496         if (!page) {
3497                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3498                 goto out;
3499         }
3500
3501         page_start = page_offset(page);
3502         page_end = page_start + PAGE_CACHE_SIZE - 1;
3503
3504         if (!PageUptodate(page)) {
3505                 ret = btrfs_readpage(NULL, page);
3506                 lock_page(page);
3507                 if (page->mapping != mapping) {
3508                         unlock_page(page);
3509                         page_cache_release(page);
3510                         goto again;
3511                 }
3512                 if (!PageUptodate(page)) {
3513                         ret = -EIO;
3514                         goto out_unlock;
3515                 }
3516         }
3517         wait_on_page_writeback(page);
3518
3519         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
3520         set_page_extent_mapped(page);
3521
3522         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3523         if (ordered) {
3524                 unlock_extent_cached(io_tree, page_start, page_end,
3525                                      &cached_state, GFP_NOFS);
3526                 unlock_page(page);
3527                 page_cache_release(page);
3528                 btrfs_start_ordered_extent(inode, ordered, 1);
3529                 btrfs_put_ordered_extent(ordered);
3530                 goto again;
3531         }
3532
3533         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3534                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3535                           0, 0, &cached_state, GFP_NOFS);
3536
3537         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3538                                         &cached_state);
3539         if (ret) {
3540                 unlock_extent_cached(io_tree, page_start, page_end,
3541                                      &cached_state, GFP_NOFS);
3542                 goto out_unlock;
3543         }
3544
3545         ret = 0;
3546         if (offset != PAGE_CACHE_SIZE) {
3547                 kaddr = kmap(page);
3548                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3549                 flush_dcache_page(page);
3550                 kunmap(page);
3551         }
3552         ClearPageChecked(page);
3553         set_page_dirty(page);
3554         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3555                              GFP_NOFS);
3556
3557 out_unlock:
3558         if (ret)
3559                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3560         unlock_page(page);
3561         page_cache_release(page);
3562 out:
3563         return ret;
3564 }
3565
3566 /*
3567  * This function puts in dummy file extents for the area we're creating a hole
3568  * for.  So if we are truncating this file to a larger size we need to insert
3569  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3570  * the range between oldsize and size
3571  */
3572 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3573 {
3574         struct btrfs_trans_handle *trans;
3575         struct btrfs_root *root = BTRFS_I(inode)->root;
3576         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3577         struct extent_map *em = NULL;
3578         struct extent_state *cached_state = NULL;
3579         u64 mask = root->sectorsize - 1;
3580         u64 hole_start = (oldsize + mask) & ~mask;
3581         u64 block_end = (size + mask) & ~mask;
3582         u64 last_byte;
3583         u64 cur_offset;
3584         u64 hole_size;
3585         int err = 0;
3586
3587         if (size <= hole_start)
3588                 return 0;
3589
3590         while (1) {
3591                 struct btrfs_ordered_extent *ordered;
3592                 btrfs_wait_ordered_range(inode, hole_start,
3593                                          block_end - hole_start);
3594                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3595                                  &cached_state);
3596                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3597                 if (!ordered)
3598                         break;
3599                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3600                                      &cached_state, GFP_NOFS);
3601                 btrfs_put_ordered_extent(ordered);
3602         }
3603
3604         cur_offset = hole_start;
3605         while (1) {
3606                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3607                                 block_end - cur_offset, 0);
3608                 if (IS_ERR(em)) {
3609                         err = PTR_ERR(em);
3610                         break;
3611                 }
3612                 last_byte = min(extent_map_end(em), block_end);
3613                 last_byte = (last_byte + mask) & ~mask;
3614                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3615                         u64 hint_byte = 0;
3616                         hole_size = last_byte - cur_offset;
3617
3618                         trans = btrfs_start_transaction(root, 3);
3619                         if (IS_ERR(trans)) {
3620                                 err = PTR_ERR(trans);
3621                                 break;
3622                         }
3623
3624                         err = btrfs_drop_extents(trans, inode, cur_offset,
3625                                                  cur_offset + hole_size,
3626                                                  &hint_byte, 1);
3627                         if (err) {
3628                                 btrfs_abort_transaction(trans, root, err);
3629                                 btrfs_end_transaction(trans, root);
3630                                 break;
3631                         }
3632
3633                         err = btrfs_insert_file_extent(trans, root,
3634                                         btrfs_ino(inode), cur_offset, 0,
3635                                         0, hole_size, 0, hole_size,
3636                                         0, 0, 0);
3637                         if (err) {
3638                                 btrfs_abort_transaction(trans, root, err);
3639                                 btrfs_end_transaction(trans, root);
3640                                 break;
3641                         }
3642
3643                         btrfs_drop_extent_cache(inode, hole_start,
3644                                         last_byte - 1, 0);
3645
3646                         btrfs_update_inode(trans, root, inode);
3647                         btrfs_end_transaction(trans, root);
3648                 }
3649                 free_extent_map(em);
3650                 em = NULL;
3651                 cur_offset = last_byte;
3652                 if (cur_offset >= block_end)
3653                         break;
3654         }
3655
3656         free_extent_map(em);
3657         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3658                              GFP_NOFS);
3659         return err;
3660 }
3661
3662 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3663 {
3664         struct btrfs_root *root = BTRFS_I(inode)->root;
3665         struct btrfs_trans_handle *trans;
3666         loff_t oldsize = i_size_read(inode);
3667         int ret;
3668
3669         if (newsize == oldsize)
3670                 return 0;
3671
3672         if (newsize > oldsize) {
3673                 truncate_pagecache(inode, oldsize, newsize);
3674                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3675                 if (ret)
3676                         return ret;
3677
3678                 trans = btrfs_start_transaction(root, 1);
3679                 if (IS_ERR(trans))
3680                         return PTR_ERR(trans);
3681
3682                 i_size_write(inode, newsize);
3683                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3684                 ret = btrfs_update_inode(trans, root, inode);
3685                 btrfs_end_transaction(trans, root);
3686         } else {
3687
3688                 /*
3689                  * We're truncating a file that used to have good data down to
3690                  * zero. Make sure it gets into the ordered flush list so that
3691                  * any new writes get down to disk quickly.
3692                  */
3693                 if (newsize == 0)
3694                         set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
3695                                 &BTRFS_I(inode)->runtime_flags);
3696
3697                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3698                 truncate_setsize(inode, newsize);
3699                 ret = btrfs_truncate(inode);
3700         }
3701
3702         return ret;
3703 }
3704
3705 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3706 {
3707         struct inode *inode = dentry->d_inode;
3708         struct btrfs_root *root = BTRFS_I(inode)->root;
3709         int err;
3710
3711         if (btrfs_root_readonly(root))
3712                 return -EROFS;
3713
3714         err = inode_change_ok(inode, attr);
3715         if (err)
3716                 return err;
3717
3718         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3719                 err = btrfs_setsize(inode, attr->ia_size);
3720                 if (err)
3721                         return err;
3722         }
3723
3724         if (attr->ia_valid) {
3725                 setattr_copy(inode, attr);
3726                 inode_inc_iversion(inode);
3727                 err = btrfs_dirty_inode(inode);
3728
3729                 if (!err && attr->ia_valid & ATTR_MODE)
3730                         err = btrfs_acl_chmod(inode);
3731         }
3732
3733         return err;
3734 }
3735
3736 void btrfs_evict_inode(struct inode *inode)
3737 {
3738         struct btrfs_trans_handle *trans;
3739         struct btrfs_root *root = BTRFS_I(inode)->root;
3740         struct btrfs_block_rsv *rsv, *global_rsv;
3741         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
3742         unsigned long nr;
3743         int ret;
3744
3745         trace_btrfs_inode_evict(inode);
3746
3747         truncate_inode_pages(&inode->i_data, 0);
3748         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3749                                btrfs_is_free_space_inode(inode)))
3750                 goto no_delete;
3751
3752         if (is_bad_inode(inode)) {
3753                 btrfs_orphan_del(NULL, inode);
3754                 goto no_delete;
3755         }
3756         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3757         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3758
3759         if (root->fs_info->log_root_recovering) {
3760                 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3761                                  &BTRFS_I(inode)->runtime_flags));
3762                 goto no_delete;
3763         }
3764
3765         if (inode->i_nlink > 0) {
3766                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3767                 goto no_delete;
3768         }
3769
3770         rsv = btrfs_alloc_block_rsv(root);
3771         if (!rsv) {
3772                 btrfs_orphan_del(NULL, inode);
3773                 goto no_delete;
3774         }
3775         rsv->size = min_size;
3776         global_rsv = &root->fs_info->global_block_rsv;
3777
3778         btrfs_i_size_write(inode, 0);
3779
3780         /*
3781          * This is a bit simpler than btrfs_truncate since
3782          *
3783          * 1) We've already reserved our space for our orphan item in the
3784          *    unlink.
3785          * 2) We're going to delete the inode item, so we don't need to update
3786          *    it at all.
3787          *
3788          * So we just need to reserve some slack space in case we add bytes when
3789          * doing the truncate.
3790          */
3791         while (1) {
3792                 ret = btrfs_block_rsv_refill_noflush(root, rsv, min_size);
3793
3794                 /*
3795                  * Try and steal from the global reserve since we will
3796                  * likely not use this space anyway, we want to try as
3797                  * hard as possible to get this to work.
3798                  */
3799                 if (ret)
3800                         ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
3801
3802                 if (ret) {
3803                         printk(KERN_WARNING "Could not get space for a "
3804                                "delete, will truncate on mount %d\n", ret);
3805                         btrfs_orphan_del(NULL, inode);
3806                         btrfs_free_block_rsv(root, rsv);
3807                         goto no_delete;
3808                 }
3809
3810                 trans = btrfs_start_transaction(root, 0);
3811                 if (IS_ERR(trans)) {
3812                         btrfs_orphan_del(NULL, inode);
3813                         btrfs_free_block_rsv(root, rsv);
3814                         goto no_delete;
3815                 }
3816
3817                 trans->block_rsv = rsv;
3818
3819                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3820                 if (ret != -EAGAIN)
3821                         break;
3822
3823                 nr = trans->blocks_used;
3824                 btrfs_end_transaction(trans, root);
3825                 trans = NULL;
3826                 btrfs_btree_balance_dirty(root, nr);
3827         }
3828
3829         btrfs_free_block_rsv(root, rsv);
3830
3831         if (ret == 0) {
3832                 trans->block_rsv = root->orphan_block_rsv;
3833                 ret = btrfs_orphan_del(trans, inode);
3834                 BUG_ON(ret);
3835         }
3836
3837         trans->block_rsv = &root->fs_info->trans_block_rsv;
3838         if (!(root == root->fs_info->tree_root ||
3839               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3840                 btrfs_return_ino(root, btrfs_ino(inode));
3841
3842         nr = trans->blocks_used;
3843         btrfs_end_transaction(trans, root);
3844         btrfs_btree_balance_dirty(root, nr);
3845 no_delete:
3846         clear_inode(inode);
3847         return;
3848 }
3849
3850 /*
3851  * this returns the key found in the dir entry in the location pointer.
3852  * If no dir entries were found, location->objectid is 0.
3853  */
3854 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3855                                struct btrfs_key *location)
3856 {
3857         const char *name = dentry->d_name.name;
3858         int namelen = dentry->d_name.len;
3859         struct btrfs_dir_item *di;
3860         struct btrfs_path *path;
3861         struct btrfs_root *root = BTRFS_I(dir)->root;
3862         int ret = 0;
3863
3864         path = btrfs_alloc_path();
3865         if (!path)
3866                 return -ENOMEM;
3867
3868         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3869                                     namelen, 0);
3870         if (IS_ERR(di))
3871                 ret = PTR_ERR(di);
3872
3873         if (IS_ERR_OR_NULL(di))
3874                 goto out_err;
3875
3876         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3877 out:
3878         btrfs_free_path(path);
3879         return ret;
3880 out_err:
3881         location->objectid = 0;
3882         goto out;
3883 }
3884
3885 /*
3886  * when we hit a tree root in a directory, the btrfs part of the inode
3887  * needs to be changed to reflect the root directory of the tree root.  This
3888  * is kind of like crossing a mount point.
3889  */
3890 static int fixup_tree_root_location(struct btrfs_root *root,
3891                                     struct inode *dir,
3892                                     struct dentry *dentry,
3893                                     struct btrfs_key *location,
3894                                     struct btrfs_root **sub_root)
3895 {
3896         struct btrfs_path *path;
3897         struct btrfs_root *new_root;
3898         struct btrfs_root_ref *ref;
3899         struct extent_buffer *leaf;
3900         int ret;
3901         int err = 0;
3902
3903         path = btrfs_alloc_path();
3904         if (!path) {
3905                 err = -ENOMEM;
3906                 goto out;
3907         }
3908
3909         err = -ENOENT;
3910         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3911                                   BTRFS_I(dir)->root->root_key.objectid,
3912                                   location->objectid);
3913         if (ret) {
3914                 if (ret < 0)
3915                         err = ret;
3916                 goto out;
3917         }
3918
3919         leaf = path->nodes[0];
3920         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3921         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3922             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3923                 goto out;
3924
3925         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3926                                    (unsigned long)(ref + 1),
3927                                    dentry->d_name.len);
3928         if (ret)
3929                 goto out;
3930
3931         btrfs_release_path(path);
3932
3933         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3934         if (IS_ERR(new_root)) {
3935                 err = PTR_ERR(new_root);
3936                 goto out;
3937         }
3938
3939         if (btrfs_root_refs(&new_root->root_item) == 0) {
3940                 err = -ENOENT;
3941                 goto out;
3942         }
3943
3944         *sub_root = new_root;
3945         location->objectid = btrfs_root_dirid(&new_root->root_item);
3946         location->type = BTRFS_INODE_ITEM_KEY;
3947         location->offset = 0;
3948         err = 0;
3949 out:
3950         btrfs_free_path(path);
3951         return err;
3952 }
3953
3954 static void inode_tree_add(struct inode *inode)
3955 {
3956         struct btrfs_root *root = BTRFS_I(inode)->root;
3957         struct btrfs_inode *entry;
3958         struct rb_node **p;
3959         struct rb_node *parent;
3960         u64 ino = btrfs_ino(inode);
3961 again:
3962         p = &root->inode_tree.rb_node;
3963         parent = NULL;
3964
3965         if (inode_unhashed(inode))
3966                 return;
3967
3968         spin_lock(&root->inode_lock);
3969         while (*p) {
3970                 parent = *p;
3971                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3972
3973                 if (ino < btrfs_ino(&entry->vfs_inode))
3974                         p = &parent->rb_left;
3975                 else if (ino > btrfs_ino(&entry->vfs_inode))
3976                         p = &parent->rb_right;
3977                 else {
3978                         WARN_ON(!(entry->vfs_inode.i_state &
3979                                   (I_WILL_FREE | I_FREEING)));
3980                         rb_erase(parent, &root->inode_tree);
3981                         RB_CLEAR_NODE(parent);
3982                         spin_unlock(&root->inode_lock);
3983                         goto again;
3984                 }
3985         }
3986         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3987         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3988         spin_unlock(&root->inode_lock);
3989 }
3990
3991 static void inode_tree_del(struct inode *inode)
3992 {
3993         struct btrfs_root *root = BTRFS_I(inode)->root;
3994         int empty = 0;
3995
3996         spin_lock(&root->inode_lock);
3997         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3998                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3999                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4000                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4001         }
4002         spin_unlock(&root->inode_lock);
4003
4004         /*
4005          * Free space cache has inodes in the tree root, but the tree root has a
4006          * root_refs of 0, so this could end up dropping the tree root as a
4007          * snapshot, so we need the extra !root->fs_info->tree_root check to
4008          * make sure we don't drop it.
4009          */
4010         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
4011             root != root->fs_info->tree_root) {
4012                 synchronize_srcu(&root->fs_info->subvol_srcu);
4013                 spin_lock(&root->inode_lock);
4014                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4015                 spin_unlock(&root->inode_lock);
4016                 if (empty)
4017                         btrfs_add_dead_root(root);
4018         }
4019 }
4020
4021 void btrfs_invalidate_inodes(struct btrfs_root *root)
4022 {
4023         struct rb_node *node;
4024         struct rb_node *prev;
4025         struct btrfs_inode *entry;
4026         struct inode *inode;
4027         u64 objectid = 0;
4028
4029         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4030
4031         spin_lock(&root->inode_lock);
4032 again:
4033         node = root->inode_tree.rb_node;
4034         prev = NULL;
4035         while (node) {
4036                 prev = node;
4037                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4038
4039                 if (objectid < btrfs_ino(&entry->vfs_inode))
4040                         node = node->rb_left;
4041                 else if (objectid > btrfs_ino(&entry->vfs_inode))
4042                         node = node->rb_right;
4043                 else
4044                         break;
4045         }
4046         if (!node) {
4047                 while (prev) {
4048                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
4049                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4050                                 node = prev;
4051                                 break;
4052                         }
4053                         prev = rb_next(prev);
4054                 }
4055         }
4056         while (node) {
4057                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4058                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4059                 inode = igrab(&entry->vfs_inode);
4060                 if (inode) {
4061                         spin_unlock(&root->inode_lock);
4062                         if (atomic_read(&inode->i_count) > 1)
4063                                 d_prune_aliases(inode);
4064                         /*
4065                          * btrfs_drop_inode will have it removed from
4066                          * the inode cache when its usage count
4067                          * hits zero.
4068                          */
4069                         iput(inode);
4070                         cond_resched();
4071                         spin_lock(&root->inode_lock);
4072                         goto again;
4073                 }
4074
4075                 if (cond_resched_lock(&root->inode_lock))
4076                         goto again;
4077
4078                 node = rb_next(node);
4079         }
4080         spin_unlock(&root->inode_lock);
4081 }
4082
4083 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4084 {
4085         struct btrfs_iget_args *args = p;
4086         inode->i_ino = args->ino;
4087         BTRFS_I(inode)->root = args->root;
4088         return 0;
4089 }
4090
4091 static int btrfs_find_actor(struct inode *inode, void *opaque)
4092 {
4093         struct btrfs_iget_args *args = opaque;
4094         return args->ino == btrfs_ino(inode) &&
4095                 args->root == BTRFS_I(inode)->root;
4096 }
4097
4098 static struct inode *btrfs_iget_locked(struct super_block *s,
4099                                        u64 objectid,
4100                                        struct btrfs_root *root)
4101 {
4102         struct inode *inode;
4103         struct btrfs_iget_args args;
4104         args.ino = objectid;
4105         args.root = root;
4106
4107         inode = iget5_locked(s, objectid, btrfs_find_actor,
4108                              btrfs_init_locked_inode,
4109                              (void *)&args);
4110         return inode;
4111 }
4112
4113 /* Get an inode object given its location and corresponding root.
4114  * Returns in *is_new if the inode was read from disk
4115  */
4116 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4117                          struct btrfs_root *root, int *new)
4118 {
4119         struct inode *inode;
4120
4121         inode = btrfs_iget_locked(s, location->objectid, root);
4122         if (!inode)
4123                 return ERR_PTR(-ENOMEM);
4124
4125         if (inode->i_state & I_NEW) {
4126                 BTRFS_I(inode)->root = root;
4127                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4128                 btrfs_read_locked_inode(inode);
4129                 if (!is_bad_inode(inode)) {
4130                         inode_tree_add(inode);
4131                         unlock_new_inode(inode);
4132                         if (new)
4133                                 *new = 1;
4134                 } else {
4135                         unlock_new_inode(inode);
4136                         iput(inode);
4137                         inode = ERR_PTR(-ESTALE);
4138                 }
4139         }
4140
4141         return inode;
4142 }
4143
4144 static struct inode *new_simple_dir(struct super_block *s,
4145                                     struct btrfs_key *key,
4146                                     struct btrfs_root *root)
4147 {
4148         struct inode *inode = new_inode(s);
4149
4150         if (!inode)
4151                 return ERR_PTR(-ENOMEM);
4152
4153         BTRFS_I(inode)->root = root;
4154         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4155         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4156
4157         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4158         inode->i_op = &btrfs_dir_ro_inode_operations;
4159         inode->i_fop = &simple_dir_operations;
4160         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4161         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4162
4163         return inode;
4164 }
4165
4166 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4167 {
4168         struct inode *inode;
4169         struct btrfs_root *root = BTRFS_I(dir)->root;
4170         struct btrfs_root *sub_root = root;
4171         struct btrfs_key location;
4172         int index;
4173         int ret = 0;
4174
4175         if (dentry->d_name.len > BTRFS_NAME_LEN)
4176                 return ERR_PTR(-ENAMETOOLONG);
4177
4178         if (unlikely(d_need_lookup(dentry))) {
4179                 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
4180                 kfree(dentry->d_fsdata);
4181                 dentry->d_fsdata = NULL;
4182                 /* This thing is hashed, drop it for now */
4183                 d_drop(dentry);
4184         } else {
4185                 ret = btrfs_inode_by_name(dir, dentry, &location);
4186         }
4187
4188         if (ret < 0)
4189                 return ERR_PTR(ret);
4190
4191         if (location.objectid == 0)
4192                 return NULL;
4193
4194         if (location.type == BTRFS_INODE_ITEM_KEY) {
4195                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4196                 return inode;
4197         }
4198
4199         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4200
4201         index = srcu_read_lock(&root->fs_info->subvol_srcu);
4202         ret = fixup_tree_root_location(root, dir, dentry,
4203                                        &location, &sub_root);
4204         if (ret < 0) {
4205                 if (ret != -ENOENT)
4206                         inode = ERR_PTR(ret);
4207                 else
4208                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
4209         } else {
4210                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4211         }
4212         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4213
4214         if (!IS_ERR(inode) && root != sub_root) {
4215                 down_read(&root->fs_info->cleanup_work_sem);
4216                 if (!(inode->i_sb->s_flags & MS_RDONLY))
4217                         ret = btrfs_orphan_cleanup(sub_root);
4218                 up_read(&root->fs_info->cleanup_work_sem);
4219                 if (ret)
4220                         inode = ERR_PTR(ret);
4221         }
4222
4223         return inode;
4224 }
4225
4226 static int btrfs_dentry_delete(const struct dentry *dentry)
4227 {
4228         struct btrfs_root *root;
4229         struct inode *inode = dentry->d_inode;
4230
4231         if (!inode && !IS_ROOT(dentry))
4232                 inode = dentry->d_parent->d_inode;
4233
4234         if (inode) {
4235                 root = BTRFS_I(inode)->root;
4236                 if (btrfs_root_refs(&root->root_item) == 0)
4237                         return 1;
4238
4239                 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
4240                         return 1;
4241         }
4242         return 0;
4243 }
4244
4245 static void btrfs_dentry_release(struct dentry *dentry)
4246 {
4247         if (dentry->d_fsdata)
4248                 kfree(dentry->d_fsdata);
4249 }
4250
4251 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4252                                    unsigned int flags)
4253 {
4254         struct dentry *ret;
4255
4256         ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4257         if (unlikely(d_need_lookup(dentry))) {
4258                 spin_lock(&dentry->d_lock);
4259                 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4260                 spin_unlock(&dentry->d_lock);
4261         }
4262         return ret;
4263 }
4264
4265 unsigned char btrfs_filetype_table[] = {
4266         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4267 };
4268
4269 static int btrfs_real_readdir(struct file *filp, void *dirent,
4270                               filldir_t filldir)
4271 {
4272         struct inode *inode = filp->f_dentry->d_inode;
4273         struct btrfs_root *root = BTRFS_I(inode)->root;
4274         struct btrfs_item *item;
4275         struct btrfs_dir_item *di;
4276         struct btrfs_key key;
4277         struct btrfs_key found_key;
4278         struct btrfs_path *path;
4279         struct list_head ins_list;
4280         struct list_head del_list;
4281         int ret;
4282         struct extent_buffer *leaf;
4283         int slot;
4284         unsigned char d_type;
4285         int over = 0;
4286         u32 di_cur;
4287         u32 di_total;
4288         u32 di_len;
4289         int key_type = BTRFS_DIR_INDEX_KEY;
4290         char tmp_name[32];
4291         char *name_ptr;
4292         int name_len;
4293         int is_curr = 0;        /* filp->f_pos points to the current index? */
4294
4295         /* FIXME, use a real flag for deciding about the key type */
4296         if (root->fs_info->tree_root == root)
4297                 key_type = BTRFS_DIR_ITEM_KEY;
4298
4299         /* special case for "." */
4300         if (filp->f_pos == 0) {
4301                 over = filldir(dirent, ".", 1,
4302                                filp->f_pos, btrfs_ino(inode), DT_DIR);
4303                 if (over)
4304                         return 0;
4305                 filp->f_pos = 1;
4306         }
4307         /* special case for .., just use the back ref */
4308         if (filp->f_pos == 1) {
4309                 u64 pino = parent_ino(filp->f_path.dentry);
4310                 over = filldir(dirent, "..", 2,
4311                                filp->f_pos, pino, DT_DIR);
4312                 if (over)
4313                         return 0;
4314                 filp->f_pos = 2;
4315         }
4316         path = btrfs_alloc_path();
4317         if (!path)
4318                 return -ENOMEM;
4319
4320         path->reada = 1;
4321
4322         if (key_type == BTRFS_DIR_INDEX_KEY) {
4323                 INIT_LIST_HEAD(&ins_list);
4324                 INIT_LIST_HEAD(&del_list);
4325                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4326         }
4327
4328         btrfs_set_key_type(&key, key_type);
4329         key.offset = filp->f_pos;
4330         key.objectid = btrfs_ino(inode);
4331
4332         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4333         if (ret < 0)
4334                 goto err;
4335
4336         while (1) {
4337                 leaf = path->nodes[0];
4338                 slot = path->slots[0];
4339                 if (slot >= btrfs_header_nritems(leaf)) {
4340                         ret = btrfs_next_leaf(root, path);
4341                         if (ret < 0)
4342                                 goto err;
4343                         else if (ret > 0)
4344                                 break;
4345                         continue;
4346                 }
4347
4348                 item = btrfs_item_nr(leaf, slot);
4349                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4350
4351                 if (found_key.objectid != key.objectid)
4352                         break;
4353                 if (btrfs_key_type(&found_key) != key_type)
4354                         break;
4355                 if (found_key.offset < filp->f_pos)
4356                         goto next;
4357                 if (key_type == BTRFS_DIR_INDEX_KEY &&
4358                     btrfs_should_delete_dir_index(&del_list,
4359                                                   found_key.offset))
4360                         goto next;
4361
4362                 filp->f_pos = found_key.offset;
4363                 is_curr = 1;
4364
4365                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4366                 di_cur = 0;
4367                 di_total = btrfs_item_size(leaf, item);
4368
4369                 while (di_cur < di_total) {
4370                         struct btrfs_key location;
4371
4372                         if (verify_dir_item(root, leaf, di))
4373                                 break;
4374
4375                         name_len = btrfs_dir_name_len(leaf, di);
4376                         if (name_len <= sizeof(tmp_name)) {
4377                                 name_ptr = tmp_name;
4378                         } else {
4379                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4380                                 if (!name_ptr) {
4381                                         ret = -ENOMEM;
4382                                         goto err;
4383                                 }
4384                         }
4385                         read_extent_buffer(leaf, name_ptr,
4386                                            (unsigned long)(di + 1), name_len);
4387
4388                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4389                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4390
4391
4392                         /* is this a reference to our own snapshot? If so
4393                          * skip it.
4394                          *
4395                          * In contrast to old kernels, we insert the snapshot's
4396                          * dir item and dir index after it has been created, so
4397                          * we won't find a reference to our own snapshot. We
4398                          * still keep the following code for backward
4399                          * compatibility.
4400                          */
4401                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4402                             location.objectid == root->root_key.objectid) {
4403                                 over = 0;
4404                                 goto skip;
4405                         }
4406                         over = filldir(dirent, name_ptr, name_len,
4407                                        found_key.offset, location.objectid,
4408                                        d_type);
4409
4410 skip:
4411                         if (name_ptr != tmp_name)
4412                                 kfree(name_ptr);
4413
4414                         if (over)
4415                                 goto nopos;
4416                         di_len = btrfs_dir_name_len(leaf, di) +
4417                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4418                         di_cur += di_len;
4419                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4420                 }
4421 next:
4422                 path->slots[0]++;
4423         }
4424
4425         if (key_type == BTRFS_DIR_INDEX_KEY) {
4426                 if (is_curr)
4427                         filp->f_pos++;
4428                 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4429                                                       &ins_list);
4430                 if (ret)
4431                         goto nopos;
4432         }
4433
4434         /* Reached end of directory/root. Bump pos past the last item. */
4435         if (key_type == BTRFS_DIR_INDEX_KEY)
4436                 /*
4437                  * 32-bit glibc will use getdents64, but then strtol -
4438                  * so the last number we can serve is this.
4439                  */
4440                 filp->f_pos = 0x7fffffff;
4441         else
4442                 filp->f_pos++;
4443 nopos:
4444         ret = 0;
4445 err:
4446         if (key_type == BTRFS_DIR_INDEX_KEY)
4447                 btrfs_put_delayed_items(&ins_list, &del_list);
4448         btrfs_free_path(path);
4449         return ret;
4450 }
4451
4452 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4453 {
4454         struct btrfs_root *root = BTRFS_I(inode)->root;
4455         struct btrfs_trans_handle *trans;
4456         int ret = 0;
4457         bool nolock = false;
4458
4459         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4460                 return 0;
4461
4462         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
4463                 nolock = true;
4464
4465         if (wbc->sync_mode == WB_SYNC_ALL) {
4466                 if (nolock)
4467                         trans = btrfs_join_transaction_nolock(root);
4468                 else
4469                         trans = btrfs_join_transaction(root);
4470                 if (IS_ERR(trans))
4471                         return PTR_ERR(trans);
4472                 if (nolock)
4473                         ret = btrfs_end_transaction_nolock(trans, root);
4474                 else
4475                         ret = btrfs_commit_transaction(trans, root);
4476         }
4477         return ret;
4478 }
4479
4480 /*
4481  * This is somewhat expensive, updating the tree every time the
4482  * inode changes.  But, it is most likely to find the inode in cache.
4483  * FIXME, needs more benchmarking...there are no reasons other than performance
4484  * to keep or drop this code.
4485  */
4486 int btrfs_dirty_inode(struct inode *inode)
4487 {
4488         struct btrfs_root *root = BTRFS_I(inode)->root;
4489         struct btrfs_trans_handle *trans;
4490         int ret;
4491
4492         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4493                 return 0;
4494
4495         trans = btrfs_join_transaction(root);
4496         if (IS_ERR(trans))
4497                 return PTR_ERR(trans);
4498
4499         ret = btrfs_update_inode(trans, root, inode);
4500         if (ret && ret == -ENOSPC) {
4501                 /* whoops, lets try again with the full transaction */
4502                 btrfs_end_transaction(trans, root);
4503                 trans = btrfs_start_transaction(root, 1);
4504                 if (IS_ERR(trans))
4505                         return PTR_ERR(trans);
4506
4507                 ret = btrfs_update_inode(trans, root, inode);
4508         }
4509         btrfs_end_transaction(trans, root);
4510         if (BTRFS_I(inode)->delayed_node)
4511                 btrfs_balance_delayed_items(root);
4512
4513         return ret;
4514 }
4515
4516 /*
4517  * This is a copy of file_update_time.  We need this so we can return error on
4518  * ENOSPC for updating the inode in the case of file write and mmap writes.
4519  */
4520 static int btrfs_update_time(struct inode *inode, struct timespec *now,
4521                              int flags)
4522 {
4523         struct btrfs_root *root = BTRFS_I(inode)->root;
4524
4525         if (btrfs_root_readonly(root))
4526                 return -EROFS;
4527
4528         if (flags & S_VERSION)
4529                 inode_inc_iversion(inode);
4530         if (flags & S_CTIME)
4531                 inode->i_ctime = *now;
4532         if (flags & S_MTIME)
4533                 inode->i_mtime = *now;
4534         if (flags & S_ATIME)
4535                 inode->i_atime = *now;
4536         return btrfs_dirty_inode(inode);
4537 }
4538
4539 /*
4540  * find the highest existing sequence number in a directory
4541  * and then set the in-memory index_cnt variable to reflect
4542  * free sequence numbers
4543  */
4544 static int btrfs_set_inode_index_count(struct inode *inode)
4545 {
4546         struct btrfs_root *root = BTRFS_I(inode)->root;
4547         struct btrfs_key key, found_key;
4548         struct btrfs_path *path;
4549         struct extent_buffer *leaf;
4550         int ret;
4551
4552         key.objectid = btrfs_ino(inode);
4553         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4554         key.offset = (u64)-1;
4555
4556         path = btrfs_alloc_path();
4557         if (!path)
4558                 return -ENOMEM;
4559
4560         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4561         if (ret < 0)
4562                 goto out;
4563         /* FIXME: we should be able to handle this */
4564         if (ret == 0)
4565                 goto out;
4566         ret = 0;
4567
4568         /*
4569          * MAGIC NUMBER EXPLANATION:
4570          * since we search a directory based on f_pos we have to start at 2
4571          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4572          * else has to start at 2
4573          */
4574         if (path->slots[0] == 0) {
4575                 BTRFS_I(inode)->index_cnt = 2;
4576                 goto out;
4577         }
4578
4579         path->slots[0]--;
4580
4581         leaf = path->nodes[0];
4582         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4583
4584         if (found_key.objectid != btrfs_ino(inode) ||
4585             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4586                 BTRFS_I(inode)->index_cnt = 2;
4587                 goto out;
4588         }
4589
4590         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4591 out:
4592         btrfs_free_path(path);
4593         return ret;
4594 }
4595
4596 /*
4597  * helper to find a free sequence number in a given directory.  This current
4598  * code is very simple, later versions will do smarter things in the btree
4599  */
4600 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4601 {
4602         int ret = 0;
4603
4604         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4605                 ret = btrfs_inode_delayed_dir_index_count(dir);
4606                 if (ret) {
4607                         ret = btrfs_set_inode_index_count(dir);
4608                         if (ret)
4609                                 return ret;
4610                 }
4611         }
4612
4613         *index = BTRFS_I(dir)->index_cnt;
4614         BTRFS_I(dir)->index_cnt++;
4615
4616         return ret;
4617 }
4618
4619 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4620                                      struct btrfs_root *root,
4621                                      struct inode *dir,
4622                                      const char *name, int name_len,
4623                                      u64 ref_objectid, u64 objectid,
4624                                      umode_t mode, u64 *index)
4625 {
4626         struct inode *inode;
4627         struct btrfs_inode_item *inode_item;
4628         struct btrfs_key *location;
4629         struct btrfs_path *path;
4630         struct btrfs_inode_ref *ref;
4631         struct btrfs_key key[2];
4632         u32 sizes[2];
4633         unsigned long ptr;
4634         int ret;
4635         int owner;
4636
4637         path = btrfs_alloc_path();
4638         if (!path)
4639                 return ERR_PTR(-ENOMEM);
4640
4641         inode = new_inode(root->fs_info->sb);
4642         if (!inode) {
4643                 btrfs_free_path(path);
4644                 return ERR_PTR(-ENOMEM);
4645         }
4646
4647         /*
4648          * we have to initialize this early, so we can reclaim the inode
4649          * number if we fail afterwards in this function.
4650          */
4651         inode->i_ino = objectid;
4652
4653         if (dir) {
4654                 trace_btrfs_inode_request(dir);
4655
4656                 ret = btrfs_set_inode_index(dir, index);
4657                 if (ret) {
4658                         btrfs_free_path(path);
4659                         iput(inode);
4660                         return ERR_PTR(ret);
4661                 }
4662         }
4663         /*
4664          * index_cnt is ignored for everything but a dir,
4665          * btrfs_get_inode_index_count has an explanation for the magic
4666          * number
4667          */
4668         BTRFS_I(inode)->index_cnt = 2;
4669         BTRFS_I(inode)->root = root;
4670         BTRFS_I(inode)->generation = trans->transid;
4671         inode->i_generation = BTRFS_I(inode)->generation;
4672
4673         if (S_ISDIR(mode))
4674                 owner = 0;
4675         else
4676                 owner = 1;
4677
4678         key[0].objectid = objectid;
4679         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4680         key[0].offset = 0;
4681
4682         key[1].objectid = objectid;
4683         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4684         key[1].offset = ref_objectid;
4685
4686         sizes[0] = sizeof(struct btrfs_inode_item);
4687         sizes[1] = name_len + sizeof(*ref);
4688
4689         path->leave_spinning = 1;
4690         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4691         if (ret != 0)
4692                 goto fail;
4693
4694         inode_init_owner(inode, dir, mode);
4695         inode_set_bytes(inode, 0);
4696         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4697         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4698                                   struct btrfs_inode_item);
4699         memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
4700                              sizeof(*inode_item));
4701         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4702
4703         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4704                              struct btrfs_inode_ref);
4705         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4706         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4707         ptr = (unsigned long)(ref + 1);
4708         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4709
4710         btrfs_mark_buffer_dirty(path->nodes[0]);
4711         btrfs_free_path(path);
4712
4713         location = &BTRFS_I(inode)->location;
4714         location->objectid = objectid;
4715         location->offset = 0;
4716         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4717
4718         btrfs_inherit_iflags(inode, dir);
4719
4720         if (S_ISREG(mode)) {
4721                 if (btrfs_test_opt(root, NODATASUM))
4722                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4723                 if (btrfs_test_opt(root, NODATACOW) ||
4724                     (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4725                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4726         }
4727
4728         insert_inode_hash(inode);
4729         inode_tree_add(inode);
4730
4731         trace_btrfs_inode_new(inode);
4732         btrfs_set_inode_last_trans(trans, inode);
4733
4734         btrfs_update_root_times(trans, root);
4735
4736         return inode;
4737 fail:
4738         if (dir)
4739                 BTRFS_I(dir)->index_cnt--;
4740         btrfs_free_path(path);
4741         iput(inode);
4742         return ERR_PTR(ret);
4743 }
4744
4745 static inline u8 btrfs_inode_type(struct inode *inode)
4746 {
4747         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4748 }
4749
4750 /*
4751  * utility function to add 'inode' into 'parent_inode' with
4752  * a give name and a given sequence number.
4753  * if 'add_backref' is true, also insert a backref from the
4754  * inode to the parent directory.
4755  */
4756 int btrfs_add_link(struct btrfs_trans_handle *trans,
4757                    struct inode *parent_inode, struct inode *inode,
4758                    const char *name, int name_len, int add_backref, u64 index)
4759 {
4760         int ret = 0;
4761         struct btrfs_key key;
4762         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4763         u64 ino = btrfs_ino(inode);
4764         u64 parent_ino = btrfs_ino(parent_inode);
4765
4766         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4767                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4768         } else {
4769                 key.objectid = ino;
4770                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4771                 key.offset = 0;
4772         }
4773
4774         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4775                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4776                                          key.objectid, root->root_key.objectid,
4777                                          parent_ino, index, name, name_len);
4778         } else if (add_backref) {
4779                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4780                                              parent_ino, index);
4781         }
4782
4783         /* Nothing to clean up yet */
4784         if (ret)
4785                 return ret;
4786
4787         ret = btrfs_insert_dir_item(trans, root, name, name_len,
4788                                     parent_inode, &key,
4789                                     btrfs_inode_type(inode), index);
4790         if (ret == -EEXIST)
4791                 goto fail_dir_item;
4792         else if (ret) {
4793                 btrfs_abort_transaction(trans, root, ret);
4794                 return ret;
4795         }
4796
4797         btrfs_i_size_write(parent_inode, parent_inode->i_size +
4798                            name_len * 2);
4799         inode_inc_iversion(parent_inode);
4800         parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4801         ret = btrfs_update_inode(trans, root, parent_inode);
4802         if (ret)
4803                 btrfs_abort_transaction(trans, root, ret);
4804         return ret;
4805
4806 fail_dir_item:
4807         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4808                 u64 local_index;
4809                 int err;
4810                 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4811                                  key.objectid, root->root_key.objectid,
4812                                  parent_ino, &local_index, name, name_len);
4813
4814         } else if (add_backref) {
4815                 u64 local_index;
4816                 int err;
4817
4818                 err = btrfs_del_inode_ref(trans, root, name, name_len,
4819                                           ino, parent_ino, &local_index);
4820         }
4821         return ret;
4822 }
4823
4824 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4825                             struct inode *dir, struct dentry *dentry,
4826                             struct inode *inode, int backref, u64 index)
4827 {
4828         int err = btrfs_add_link(trans, dir, inode,
4829                                  dentry->d_name.name, dentry->d_name.len,
4830                                  backref, index);
4831         if (err > 0)
4832                 err = -EEXIST;
4833         return err;
4834 }
4835
4836 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4837                         umode_t mode, dev_t rdev)
4838 {
4839         struct btrfs_trans_handle *trans;
4840         struct btrfs_root *root = BTRFS_I(dir)->root;
4841         struct inode *inode = NULL;
4842         int err;
4843         int drop_inode = 0;
4844         u64 objectid;
4845         unsigned long nr = 0;
4846         u64 index = 0;
4847
4848         if (!new_valid_dev(rdev))
4849                 return -EINVAL;
4850
4851         /*
4852          * 2 for inode item and ref
4853          * 2 for dir items
4854          * 1 for xattr if selinux is on
4855          */
4856         trans = btrfs_start_transaction(root, 5);
4857         if (IS_ERR(trans))
4858                 return PTR_ERR(trans);
4859
4860         err = btrfs_find_free_ino(root, &objectid);
4861         if (err)
4862                 goto out_unlock;
4863
4864         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4865                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4866                                 mode, &index);
4867         if (IS_ERR(inode)) {
4868                 err = PTR_ERR(inode);
4869                 goto out_unlock;
4870         }
4871
4872         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4873         if (err) {
4874                 drop_inode = 1;
4875                 goto out_unlock;
4876         }
4877
4878         /*
4879         * If the active LSM wants to access the inode during
4880         * d_instantiate it needs these. Smack checks to see
4881         * if the filesystem supports xattrs by looking at the
4882         * ops vector.
4883         */
4884
4885         inode->i_op = &btrfs_special_inode_operations;
4886         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4887         if (err)
4888                 drop_inode = 1;
4889         else {
4890                 init_special_inode(inode, inode->i_mode, rdev);
4891                 btrfs_update_inode(trans, root, inode);
4892                 d_instantiate(dentry, inode);
4893         }
4894 out_unlock:
4895         nr = trans->blocks_used;
4896         btrfs_end_transaction(trans, root);
4897         btrfs_btree_balance_dirty(root, nr);
4898         if (drop_inode) {
4899                 inode_dec_link_count(inode);
4900                 iput(inode);
4901         }
4902         return err;
4903 }
4904
4905 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4906                         umode_t mode, bool excl)
4907 {
4908         struct btrfs_trans_handle *trans;
4909         struct btrfs_root *root = BTRFS_I(dir)->root;
4910         struct inode *inode = NULL;
4911         int drop_inode = 0;
4912         int err;
4913         unsigned long nr = 0;
4914         u64 objectid;
4915         u64 index = 0;
4916
4917         /*
4918          * 2 for inode item and ref
4919          * 2 for dir items
4920          * 1 for xattr if selinux is on
4921          */
4922         trans = btrfs_start_transaction(root, 5);
4923         if (IS_ERR(trans))
4924                 return PTR_ERR(trans);
4925
4926         err = btrfs_find_free_ino(root, &objectid);
4927         if (err)
4928                 goto out_unlock;
4929
4930         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4931                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4932                                 mode, &index);
4933         if (IS_ERR(inode)) {
4934                 err = PTR_ERR(inode);
4935                 goto out_unlock;
4936         }
4937
4938         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4939         if (err) {
4940                 drop_inode = 1;
4941                 goto out_unlock;
4942         }
4943
4944         /*
4945         * If the active LSM wants to access the inode during
4946         * d_instantiate it needs these. Smack checks to see
4947         * if the filesystem supports xattrs by looking at the
4948         * ops vector.
4949         */
4950         inode->i_fop = &btrfs_file_operations;
4951         inode->i_op = &btrfs_file_inode_operations;
4952
4953         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4954         if (err)
4955                 drop_inode = 1;
4956         else {
4957                 inode->i_mapping->a_ops = &btrfs_aops;
4958                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4959                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4960                 d_instantiate(dentry, inode);
4961         }
4962 out_unlock:
4963         nr = trans->blocks_used;
4964         btrfs_end_transaction(trans, root);
4965         if (drop_inode) {
4966                 inode_dec_link_count(inode);
4967                 iput(inode);
4968         }
4969         btrfs_btree_balance_dirty(root, nr);
4970         return err;
4971 }
4972
4973 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4974                       struct dentry *dentry)
4975 {
4976         struct btrfs_trans_handle *trans;
4977         struct btrfs_root *root = BTRFS_I(dir)->root;
4978         struct inode *inode = old_dentry->d_inode;
4979         u64 index;
4980         unsigned long nr = 0;
4981         int err;
4982         int drop_inode = 0;
4983
4984         /* do not allow sys_link's with other subvols of the same device */
4985         if (root->objectid != BTRFS_I(inode)->root->objectid)
4986                 return -EXDEV;
4987
4988         if (inode->i_nlink == ~0U)
4989                 return -EMLINK;
4990
4991         err = btrfs_set_inode_index(dir, &index);
4992         if (err)
4993                 goto fail;
4994
4995         /*
4996          * 2 items for inode and inode ref
4997          * 2 items for dir items
4998          * 1 item for parent inode
4999          */
5000         trans = btrfs_start_transaction(root, 5);
5001         if (IS_ERR(trans)) {
5002                 err = PTR_ERR(trans);
5003                 goto fail;
5004         }
5005
5006         btrfs_inc_nlink(inode);
5007         inode_inc_iversion(inode);
5008         inode->i_ctime = CURRENT_TIME;
5009         ihold(inode);
5010
5011         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5012
5013         if (err) {
5014                 drop_inode = 1;
5015         } else {
5016                 struct dentry *parent = dentry->d_parent;
5017                 err = btrfs_update_inode(trans, root, inode);
5018                 if (err)
5019                         goto fail;
5020                 d_instantiate(dentry, inode);
5021                 btrfs_log_new_name(trans, inode, NULL, parent);
5022         }
5023
5024         nr = trans->blocks_used;
5025         btrfs_end_transaction(trans, root);
5026 fail:
5027         if (drop_inode) {
5028                 inode_dec_link_count(inode);
5029                 iput(inode);
5030         }
5031         btrfs_btree_balance_dirty(root, nr);
5032         return err;
5033 }
5034
5035 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5036 {
5037         struct inode *inode = NULL;
5038         struct btrfs_trans_handle *trans;
5039         struct btrfs_root *root = BTRFS_I(dir)->root;
5040         int err = 0;
5041         int drop_on_err = 0;
5042         u64 objectid = 0;
5043         u64 index = 0;
5044         unsigned long nr = 1;
5045
5046         /*
5047          * 2 items for inode and ref
5048          * 2 items for dir items
5049          * 1 for xattr if selinux is on
5050          */
5051         trans = btrfs_start_transaction(root, 5);
5052         if (IS_ERR(trans))
5053                 return PTR_ERR(trans);
5054
5055         err = btrfs_find_free_ino(root, &objectid);
5056         if (err)
5057                 goto out_fail;
5058
5059         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5060                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5061                                 S_IFDIR | mode, &index);
5062         if (IS_ERR(inode)) {
5063                 err = PTR_ERR(inode);
5064                 goto out_fail;
5065         }
5066
5067         drop_on_err = 1;
5068
5069         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5070         if (err)
5071                 goto out_fail;
5072
5073         inode->i_op = &btrfs_dir_inode_operations;
5074         inode->i_fop = &btrfs_dir_file_operations;
5075
5076         btrfs_i_size_write(inode, 0);
5077         err = btrfs_update_inode(trans, root, inode);
5078         if (err)
5079                 goto out_fail;
5080
5081         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5082                              dentry->d_name.len, 0, index);
5083         if (err)
5084                 goto out_fail;
5085
5086         d_instantiate(dentry, inode);
5087         drop_on_err = 0;
5088
5089 out_fail:
5090         nr = trans->blocks_used;
5091         btrfs_end_transaction(trans, root);
5092         if (drop_on_err)
5093                 iput(inode);
5094         btrfs_btree_balance_dirty(root, nr);
5095         return err;
5096 }
5097
5098 /* helper for btfs_get_extent.  Given an existing extent in the tree,
5099  * and an extent that you want to insert, deal with overlap and insert
5100  * the new extent into the tree.
5101  */
5102 static int merge_extent_mapping(struct extent_map_tree *em_tree,
5103                                 struct extent_map *existing,
5104                                 struct extent_map *em,
5105                                 u64 map_start, u64 map_len)
5106 {
5107         u64 start_diff;
5108
5109         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5110         start_diff = map_start - em->start;
5111         em->start = map_start;
5112         em->len = map_len;
5113         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5114             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
5115                 em->block_start += start_diff;
5116                 em->block_len -= start_diff;
5117         }
5118         return add_extent_mapping(em_tree, em);
5119 }
5120
5121 static noinline int uncompress_inline(struct btrfs_path *path,
5122                                       struct inode *inode, struct page *page,
5123                                       size_t pg_offset, u64 extent_offset,
5124                                       struct btrfs_file_extent_item *item)
5125 {
5126         int ret;
5127         struct extent_buffer *leaf = path->nodes[0];
5128         char *tmp;
5129         size_t max_size;
5130         unsigned long inline_size;
5131         unsigned long ptr;
5132         int compress_type;
5133
5134         WARN_ON(pg_offset != 0);
5135         compress_type = btrfs_file_extent_compression(leaf, item);
5136         max_size = btrfs_file_extent_ram_bytes(leaf, item);
5137         inline_size = btrfs_file_extent_inline_item_len(leaf,
5138                                         btrfs_item_nr(leaf, path->slots[0]));
5139         tmp = kmalloc(inline_size, GFP_NOFS);
5140         if (!tmp)
5141                 return -ENOMEM;
5142         ptr = btrfs_file_extent_inline_start(item);
5143
5144         read_extent_buffer(leaf, tmp, ptr, inline_size);
5145
5146         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5147         ret = btrfs_decompress(compress_type, tmp, page,
5148                                extent_offset, inline_size, max_size);
5149         if (ret) {
5150                 char *kaddr = kmap_atomic(page);
5151                 unsigned long copy_size = min_t(u64,
5152                                   PAGE_CACHE_SIZE - pg_offset,
5153                                   max_size - extent_offset);
5154                 memset(kaddr + pg_offset, 0, copy_size);
5155                 kunmap_atomic(kaddr);
5156         }
5157         kfree(tmp);
5158         return 0;
5159 }
5160
5161 /*
5162  * a bit scary, this does extent mapping from logical file offset to the disk.
5163  * the ugly parts come from merging extents from the disk with the in-ram
5164  * representation.  This gets more complex because of the data=ordered code,
5165  * where the in-ram extents might be locked pending data=ordered completion.
5166  *
5167  * This also copies inline extents directly into the page.
5168  */
5169
5170 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5171                                     size_t pg_offset, u64 start, u64 len,
5172                                     int create)
5173 {
5174         int ret;
5175         int err = 0;
5176         u64 bytenr;
5177         u64 extent_start = 0;
5178         u64 extent_end = 0;
5179         u64 objectid = btrfs_ino(inode);
5180         u32 found_type;
5181         struct btrfs_path *path = NULL;
5182         struct btrfs_root *root = BTRFS_I(inode)->root;
5183         struct btrfs_file_extent_item *item;
5184         struct extent_buffer *leaf;
5185         struct btrfs_key found_key;
5186         struct extent_map *em = NULL;
5187         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5188         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5189         struct btrfs_trans_handle *trans = NULL;
5190         int compress_type;
5191
5192 again:
5193         read_lock(&em_tree->lock);
5194         em = lookup_extent_mapping(em_tree, start, len);
5195         if (em)
5196                 em->bdev = root->fs_info->fs_devices->latest_bdev;
5197         read_unlock(&em_tree->lock);
5198
5199         if (em) {
5200                 if (em->start > start || em->start + em->len <= start)
5201                         free_extent_map(em);
5202                 else if (em->block_start == EXTENT_MAP_INLINE && page)
5203                         free_extent_map(em);
5204                 else
5205                         goto out;
5206         }
5207         em = alloc_extent_map();
5208         if (!em) {
5209                 err = -ENOMEM;
5210                 goto out;
5211         }
5212         em->bdev = root->fs_info->fs_devices->latest_bdev;
5213         em->start = EXTENT_MAP_HOLE;
5214         em->orig_start = EXTENT_MAP_HOLE;
5215         em->len = (u64)-1;
5216         em->block_len = (u64)-1;
5217
5218         if (!path) {
5219                 path = btrfs_alloc_path();
5220                 if (!path) {
5221                         err = -ENOMEM;
5222                         goto out;
5223                 }
5224                 /*
5225                  * Chances are we'll be called again, so go ahead and do
5226                  * readahead
5227                  */
5228                 path->reada = 1;
5229         }
5230
5231         ret = btrfs_lookup_file_extent(trans, root, path,
5232                                        objectid, start, trans != NULL);
5233         if (ret < 0) {
5234                 err = ret;
5235                 goto out;
5236         }
5237
5238         if (ret != 0) {
5239                 if (path->slots[0] == 0)
5240                         goto not_found;
5241                 path->slots[0]--;
5242         }
5243
5244         leaf = path->nodes[0];
5245         item = btrfs_item_ptr(leaf, path->slots[0],
5246                               struct btrfs_file_extent_item);
5247         /* are we inside the extent that was found? */
5248         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5249         found_type = btrfs_key_type(&found_key);
5250         if (found_key.objectid != objectid ||
5251             found_type != BTRFS_EXTENT_DATA_KEY) {
5252                 goto not_found;
5253         }
5254
5255         found_type = btrfs_file_extent_type(leaf, item);
5256         extent_start = found_key.offset;
5257         compress_type = btrfs_file_extent_compression(leaf, item);
5258         if (found_type == BTRFS_FILE_EXTENT_REG ||
5259             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5260                 extent_end = extent_start +
5261                        btrfs_file_extent_num_bytes(leaf, item);
5262         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5263                 size_t size;
5264                 size = btrfs_file_extent_inline_len(leaf, item);
5265                 extent_end = (extent_start + size + root->sectorsize - 1) &
5266                         ~((u64)root->sectorsize - 1);
5267         }
5268
5269         if (start >= extent_end) {
5270                 path->slots[0]++;
5271                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5272                         ret = btrfs_next_leaf(root, path);
5273                         if (ret < 0) {
5274                                 err = ret;
5275                                 goto out;
5276                         }
5277                         if (ret > 0)
5278                                 goto not_found;
5279                         leaf = path->nodes[0];
5280                 }
5281                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5282                 if (found_key.objectid != objectid ||
5283                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5284                         goto not_found;
5285                 if (start + len <= found_key.offset)
5286                         goto not_found;
5287                 em->start = start;
5288                 em->len = found_key.offset - start;
5289                 goto not_found_em;
5290         }
5291
5292         if (found_type == BTRFS_FILE_EXTENT_REG ||
5293             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5294                 em->start = extent_start;
5295                 em->len = extent_end - extent_start;
5296                 em->orig_start = extent_start -
5297                                  btrfs_file_extent_offset(leaf, item);
5298                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5299                 if (bytenr == 0) {
5300                         em->block_start = EXTENT_MAP_HOLE;
5301                         goto insert;
5302                 }
5303                 if (compress_type != BTRFS_COMPRESS_NONE) {
5304                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5305                         em->compress_type = compress_type;
5306                         em->block_start = bytenr;
5307                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5308                                                                          item);
5309                 } else {
5310                         bytenr += btrfs_file_extent_offset(leaf, item);
5311                         em->block_start = bytenr;
5312                         em->block_len = em->len;
5313                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5314                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5315                 }
5316                 goto insert;
5317         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5318                 unsigned long ptr;
5319                 char *map;
5320                 size_t size;
5321                 size_t extent_offset;
5322                 size_t copy_size;
5323
5324                 em->block_start = EXTENT_MAP_INLINE;
5325                 if (!page || create) {
5326                         em->start = extent_start;
5327                         em->len = extent_end - extent_start;
5328                         goto out;
5329                 }
5330
5331                 size = btrfs_file_extent_inline_len(leaf, item);
5332                 extent_offset = page_offset(page) + pg_offset - extent_start;
5333                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5334                                 size - extent_offset);
5335                 em->start = extent_start + extent_offset;
5336                 em->len = (copy_size + root->sectorsize - 1) &
5337                         ~((u64)root->sectorsize - 1);
5338                 em->orig_start = EXTENT_MAP_INLINE;
5339                 if (compress_type) {
5340                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5341                         em->compress_type = compress_type;
5342                 }
5343                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5344                 if (create == 0 && !PageUptodate(page)) {
5345                         if (btrfs_file_extent_compression(leaf, item) !=
5346                             BTRFS_COMPRESS_NONE) {
5347                                 ret = uncompress_inline(path, inode, page,
5348                                                         pg_offset,
5349                                                         extent_offset, item);
5350                                 BUG_ON(ret); /* -ENOMEM */
5351                         } else {
5352                                 map = kmap(page);
5353                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5354                                                    copy_size);
5355                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5356                                         memset(map + pg_offset + copy_size, 0,
5357                                                PAGE_CACHE_SIZE - pg_offset -
5358                                                copy_size);
5359                                 }
5360                                 kunmap(page);
5361                         }
5362                         flush_dcache_page(page);
5363                 } else if (create && PageUptodate(page)) {
5364                         BUG();
5365                         if (!trans) {
5366                                 kunmap(page);
5367                                 free_extent_map(em);
5368                                 em = NULL;
5369
5370                                 btrfs_release_path(path);
5371                                 trans = btrfs_join_transaction(root);
5372
5373                                 if (IS_ERR(trans))
5374                                         return ERR_CAST(trans);
5375                                 goto again;
5376                         }
5377                         map = kmap(page);
5378                         write_extent_buffer(leaf, map + pg_offset, ptr,
5379                                             copy_size);
5380                         kunmap(page);
5381                         btrfs_mark_buffer_dirty(leaf);
5382                 }
5383                 set_extent_uptodate(io_tree, em->start,
5384                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
5385                 goto insert;
5386         } else {
5387                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5388                 WARN_ON(1);
5389         }
5390 not_found:
5391         em->start = start;
5392         em->len = len;
5393 not_found_em:
5394         em->block_start = EXTENT_MAP_HOLE;
5395         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5396 insert:
5397         btrfs_release_path(path);
5398         if (em->start > start || extent_map_end(em) <= start) {
5399                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5400                        "[%llu %llu]\n", (unsigned long long)em->start,
5401                        (unsigned long long)em->len,
5402                        (unsigned long long)start,
5403                        (unsigned long long)len);
5404                 err = -EIO;
5405                 goto out;
5406         }
5407
5408         err = 0;
5409         write_lock(&em_tree->lock);
5410         ret = add_extent_mapping(em_tree, em);
5411         /* it is possible that someone inserted the extent into the tree
5412          * while we had the lock dropped.  It is also possible that
5413          * an overlapping map exists in the tree
5414          */
5415         if (ret == -EEXIST) {
5416                 struct extent_map *existing;
5417
5418                 ret = 0;
5419
5420                 existing = lookup_extent_mapping(em_tree, start, len);
5421                 if (existing && (existing->start > start ||
5422                     existing->start + existing->len <= start)) {
5423                         free_extent_map(existing);
5424                         existing = NULL;
5425                 }
5426                 if (!existing) {
5427                         existing = lookup_extent_mapping(em_tree, em->start,
5428                                                          em->len);
5429                         if (existing) {
5430                                 err = merge_extent_mapping(em_tree, existing,
5431                                                            em, start,
5432                                                            root->sectorsize);
5433                                 free_extent_map(existing);
5434                                 if (err) {
5435                                         free_extent_map(em);
5436                                         em = NULL;
5437                                 }
5438                         } else {
5439                                 err = -EIO;
5440                                 free_extent_map(em);
5441                                 em = NULL;
5442                         }
5443                 } else {
5444                         free_extent_map(em);
5445                         em = existing;
5446                         err = 0;
5447                 }
5448         }
5449         write_unlock(&em_tree->lock);
5450 out:
5451
5452         trace_btrfs_get_extent(root, em);
5453
5454         if (path)
5455                 btrfs_free_path(path);
5456         if (trans) {
5457                 ret = btrfs_end_transaction(trans, root);
5458                 if (!err)
5459                         err = ret;
5460         }
5461         if (err) {
5462                 free_extent_map(em);
5463                 return ERR_PTR(err);
5464         }
5465         BUG_ON(!em); /* Error is always set */
5466         return em;
5467 }
5468
5469 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5470                                            size_t pg_offset, u64 start, u64 len,
5471                                            int create)
5472 {
5473         struct extent_map *em;
5474         struct extent_map *hole_em = NULL;
5475         u64 range_start = start;
5476         u64 end;
5477         u64 found;
5478         u64 found_end;
5479         int err = 0;
5480
5481         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5482         if (IS_ERR(em))
5483                 return em;
5484         if (em) {
5485                 /*
5486                  * if our em maps to a hole, there might
5487                  * actually be delalloc bytes behind it
5488                  */
5489                 if (em->block_start != EXTENT_MAP_HOLE)
5490                         return em;
5491                 else
5492                         hole_em = em;
5493         }
5494
5495         /* check to see if we've wrapped (len == -1 or similar) */
5496         end = start + len;
5497         if (end < start)
5498                 end = (u64)-1;
5499         else
5500                 end -= 1;
5501
5502         em = NULL;
5503
5504         /* ok, we didn't find anything, lets look for delalloc */
5505         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5506                                  end, len, EXTENT_DELALLOC, 1);
5507         found_end = range_start + found;
5508         if (found_end < range_start)
5509                 found_end = (u64)-1;
5510
5511         /*
5512          * we didn't find anything useful, return
5513          * the original results from get_extent()
5514          */
5515         if (range_start > end || found_end <= start) {
5516                 em = hole_em;
5517                 hole_em = NULL;
5518                 goto out;
5519         }
5520
5521         /* adjust the range_start to make sure it doesn't
5522          * go backwards from the start they passed in
5523          */
5524         range_start = max(start,range_start);
5525         found = found_end - range_start;
5526
5527         if (found > 0) {
5528                 u64 hole_start = start;
5529                 u64 hole_len = len;
5530
5531                 em = alloc_extent_map();
5532                 if (!em) {
5533                         err = -ENOMEM;
5534                         goto out;
5535                 }
5536                 /*
5537                  * when btrfs_get_extent can't find anything it
5538                  * returns one huge hole
5539                  *
5540                  * make sure what it found really fits our range, and
5541                  * adjust to make sure it is based on the start from
5542                  * the caller
5543                  */
5544                 if (hole_em) {
5545                         u64 calc_end = extent_map_end(hole_em);
5546
5547                         if (calc_end <= start || (hole_em->start > end)) {
5548                                 free_extent_map(hole_em);
5549                                 hole_em = NULL;
5550                         } else {
5551                                 hole_start = max(hole_em->start, start);
5552                                 hole_len = calc_end - hole_start;
5553                         }
5554                 }
5555                 em->bdev = NULL;
5556                 if (hole_em && range_start > hole_start) {
5557                         /* our hole starts before our delalloc, so we
5558                          * have to return just the parts of the hole
5559                          * that go until  the delalloc starts
5560                          */
5561                         em->len = min(hole_len,
5562                                       range_start - hole_start);
5563                         em->start = hole_start;
5564                         em->orig_start = hole_start;
5565                         /*
5566                          * don't adjust block start at all,
5567                          * it is fixed at EXTENT_MAP_HOLE
5568                          */
5569                         em->block_start = hole_em->block_start;
5570                         em->block_len = hole_len;
5571                 } else {
5572                         em->start = range_start;
5573                         em->len = found;
5574                         em->orig_start = range_start;
5575                         em->block_start = EXTENT_MAP_DELALLOC;
5576                         em->block_len = found;
5577                 }
5578         } else if (hole_em) {
5579                 return hole_em;
5580         }
5581 out:
5582
5583         free_extent_map(hole_em);
5584         if (err) {
5585                 free_extent_map(em);
5586                 return ERR_PTR(err);
5587         }
5588         return em;
5589 }
5590
5591 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5592                                                   struct extent_map *em,
5593                                                   u64 start, u64 len)
5594 {
5595         struct btrfs_root *root = BTRFS_I(inode)->root;
5596         struct btrfs_trans_handle *trans;
5597         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5598         struct btrfs_key ins;
5599         u64 alloc_hint;
5600         int ret;
5601         bool insert = false;
5602
5603         /*
5604          * Ok if the extent map we looked up is a hole and is for the exact
5605          * range we want, there is no reason to allocate a new one, however if
5606          * it is not right then we need to free this one and drop the cache for
5607          * our range.
5608          */
5609         if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5610             em->len != len) {
5611                 free_extent_map(em);
5612                 em = NULL;
5613                 insert = true;
5614                 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5615         }
5616
5617         trans = btrfs_join_transaction(root);
5618         if (IS_ERR(trans))
5619                 return ERR_CAST(trans);
5620
5621         if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5622                 btrfs_add_inode_defrag(trans, inode);
5623
5624         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5625
5626         alloc_hint = get_extent_allocation_hint(inode, start, len);
5627         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5628                                    alloc_hint, &ins, 1);
5629         if (ret) {
5630                 em = ERR_PTR(ret);
5631                 goto out;
5632         }
5633
5634         if (!em) {
5635                 em = alloc_extent_map();
5636                 if (!em) {
5637                         em = ERR_PTR(-ENOMEM);
5638                         goto out;
5639                 }
5640         }
5641
5642         em->start = start;
5643         em->orig_start = em->start;
5644         em->len = ins.offset;
5645
5646         em->block_start = ins.objectid;
5647         em->block_len = ins.offset;
5648         em->bdev = root->fs_info->fs_devices->latest_bdev;
5649
5650         /*
5651          * We need to do this because if we're using the original em we searched
5652          * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5653          */
5654         em->flags = 0;
5655         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5656
5657         while (insert) {
5658                 write_lock(&em_tree->lock);
5659                 ret = add_extent_mapping(em_tree, em);
5660                 write_unlock(&em_tree->lock);
5661                 if (ret != -EEXIST)
5662                         break;
5663                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5664         }
5665
5666         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5667                                            ins.offset, ins.offset, 0);
5668         if (ret) {
5669                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5670                 em = ERR_PTR(ret);
5671         }
5672 out:
5673         btrfs_end_transaction(trans, root);
5674         return em;
5675 }
5676
5677 /*
5678  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5679  * block must be cow'd
5680  */
5681 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5682                                       struct inode *inode, u64 offset, u64 len)
5683 {
5684         struct btrfs_path *path;
5685         int ret;
5686         struct extent_buffer *leaf;
5687         struct btrfs_root *root = BTRFS_I(inode)->root;
5688         struct btrfs_file_extent_item *fi;
5689         struct btrfs_key key;
5690         u64 disk_bytenr;
5691         u64 backref_offset;
5692         u64 extent_end;
5693         u64 num_bytes;
5694         int slot;
5695         int found_type;
5696
5697         path = btrfs_alloc_path();
5698         if (!path)
5699                 return -ENOMEM;
5700
5701         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5702                                        offset, 0);
5703         if (ret < 0)
5704                 goto out;
5705
5706         slot = path->slots[0];
5707         if (ret == 1) {
5708                 if (slot == 0) {
5709                         /* can't find the item, must cow */
5710                         ret = 0;
5711                         goto out;
5712                 }
5713                 slot--;
5714         }
5715         ret = 0;
5716         leaf = path->nodes[0];
5717         btrfs_item_key_to_cpu(leaf, &key, slot);
5718         if (key.objectid != btrfs_ino(inode) ||
5719             key.type != BTRFS_EXTENT_DATA_KEY) {
5720                 /* not our file or wrong item type, must cow */
5721                 goto out;
5722         }
5723
5724         if (key.offset > offset) {
5725                 /* Wrong offset, must cow */
5726                 goto out;
5727         }
5728
5729         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5730         found_type = btrfs_file_extent_type(leaf, fi);
5731         if (found_type != BTRFS_FILE_EXTENT_REG &&
5732             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5733                 /* not a regular extent, must cow */
5734                 goto out;
5735         }
5736         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5737         backref_offset = btrfs_file_extent_offset(leaf, fi);
5738
5739         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5740         if (extent_end < offset + len) {
5741                 /* extent doesn't include our full range, must cow */
5742                 goto out;
5743         }
5744
5745         if (btrfs_extent_readonly(root, disk_bytenr))
5746                 goto out;
5747
5748         /*
5749          * look for other files referencing this extent, if we
5750          * find any we must cow
5751          */
5752         if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5753                                   key.offset - backref_offset, disk_bytenr))
5754                 goto out;
5755
5756         /*
5757          * adjust disk_bytenr and num_bytes to cover just the bytes
5758          * in this extent we are about to write.  If there
5759          * are any csums in that range we have to cow in order
5760          * to keep the csums correct
5761          */
5762         disk_bytenr += backref_offset;
5763         disk_bytenr += offset - key.offset;
5764         num_bytes = min(offset + len, extent_end) - offset;
5765         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5766                                 goto out;
5767         /*
5768          * all of the above have passed, it is safe to overwrite this extent
5769          * without cow
5770          */
5771         ret = 1;
5772 out:
5773         btrfs_free_path(path);
5774         return ret;
5775 }
5776
5777 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5778                                    struct buffer_head *bh_result, int create)
5779 {
5780         struct extent_map *em;
5781         struct btrfs_root *root = BTRFS_I(inode)->root;
5782         u64 start = iblock << inode->i_blkbits;
5783         u64 len = bh_result->b_size;
5784         struct btrfs_trans_handle *trans;
5785
5786         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5787         if (IS_ERR(em))
5788                 return PTR_ERR(em);
5789
5790         /*
5791          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5792          * io.  INLINE is special, and we could probably kludge it in here, but
5793          * it's still buffered so for safety lets just fall back to the generic
5794          * buffered path.
5795          *
5796          * For COMPRESSED we _have_ to read the entire extent in so we can
5797          * decompress it, so there will be buffering required no matter what we
5798          * do, so go ahead and fallback to buffered.
5799          *
5800          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5801          * to buffered IO.  Don't blame me, this is the price we pay for using
5802          * the generic code.
5803          */
5804         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5805             em->block_start == EXTENT_MAP_INLINE) {
5806                 free_extent_map(em);
5807                 return -ENOTBLK;
5808         }
5809
5810         /* Just a good old fashioned hole, return */
5811         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5812                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5813                 free_extent_map(em);
5814                 /* DIO will do one hole at a time, so just unlock a sector */
5815                 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5816                               start + root->sectorsize - 1);
5817                 return 0;
5818         }
5819
5820         /*
5821          * We don't allocate a new extent in the following cases
5822          *
5823          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
5824          * existing extent.
5825          * 2) The extent is marked as PREALLOC.  We're good to go here and can
5826          * just use the extent.
5827          *
5828          */
5829         if (!create) {
5830                 len = em->len - (start - em->start);
5831                 goto map;
5832         }
5833
5834         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5835             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5836              em->block_start != EXTENT_MAP_HOLE)) {
5837                 int type;
5838                 int ret;
5839                 u64 block_start;
5840
5841                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5842                         type = BTRFS_ORDERED_PREALLOC;
5843                 else
5844                         type = BTRFS_ORDERED_NOCOW;
5845                 len = min(len, em->len - (start - em->start));
5846                 block_start = em->block_start + (start - em->start);
5847
5848                 /*
5849                  * we're not going to log anything, but we do need
5850                  * to make sure the current transaction stays open
5851                  * while we look for nocow cross refs
5852                  */
5853                 trans = btrfs_join_transaction(root);
5854                 if (IS_ERR(trans))
5855                         goto must_cow;
5856
5857                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5858                         ret = btrfs_add_ordered_extent_dio(inode, start,
5859                                            block_start, len, len, type);
5860                         btrfs_end_transaction(trans, root);
5861                         if (ret) {
5862                                 free_extent_map(em);
5863                                 return ret;
5864                         }
5865                         goto unlock;
5866                 }
5867                 btrfs_end_transaction(trans, root);
5868         }
5869 must_cow:
5870         /*
5871          * this will cow the extent, reset the len in case we changed
5872          * it above
5873          */
5874         len = bh_result->b_size;
5875         em = btrfs_new_extent_direct(inode, em, start, len);
5876         if (IS_ERR(em))
5877                 return PTR_ERR(em);
5878         len = min(len, em->len - (start - em->start));
5879 unlock:
5880         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5881                           EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5882                           0, NULL, GFP_NOFS);
5883 map:
5884         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5885                 inode->i_blkbits;
5886         bh_result->b_size = len;
5887         bh_result->b_bdev = em->bdev;
5888         set_buffer_mapped(bh_result);
5889         if (create) {
5890                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5891                         set_buffer_new(bh_result);
5892
5893                 /*
5894                  * Need to update the i_size under the extent lock so buffered
5895                  * readers will get the updated i_size when we unlock.
5896                  */
5897                 if (start + len > i_size_read(inode))
5898                         i_size_write(inode, start + len);
5899         }
5900
5901         free_extent_map(em);
5902
5903         return 0;
5904 }
5905
5906 struct btrfs_dio_private {
5907         struct inode *inode;
5908         u64 logical_offset;
5909         u64 disk_bytenr;
5910         u64 bytes;
5911         u32 *csums;
5912         void *private;
5913
5914         /* number of bios pending for this dio */
5915         atomic_t pending_bios;
5916
5917         /* IO errors */
5918         int errors;
5919
5920         struct bio *orig_bio;
5921 };
5922
5923 static void btrfs_endio_direct_read(struct bio *bio, int err)
5924 {
5925         struct btrfs_dio_private *dip = bio->bi_private;
5926         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5927         struct bio_vec *bvec = bio->bi_io_vec;
5928         struct inode *inode = dip->inode;
5929         struct btrfs_root *root = BTRFS_I(inode)->root;
5930         u64 start;
5931         u32 *private = dip->csums;
5932
5933         start = dip->logical_offset;
5934         do {
5935                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5936                         struct page *page = bvec->bv_page;
5937                         char *kaddr;
5938                         u32 csum = ~(u32)0;
5939                         unsigned long flags;
5940
5941                         local_irq_save(flags);
5942                         kaddr = kmap_atomic(page);
5943                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5944                                                csum, bvec->bv_len);
5945                         btrfs_csum_final(csum, (char *)&csum);
5946                         kunmap_atomic(kaddr);
5947                         local_irq_restore(flags);
5948
5949                         flush_dcache_page(bvec->bv_page);
5950                         if (csum != *private) {
5951                                 printk(KERN_ERR "btrfs csum failed ino %llu off"
5952                                       " %llu csum %u private %u\n",
5953                                       (unsigned long long)btrfs_ino(inode),
5954                                       (unsigned long long)start,
5955                                       csum, *private);
5956                                 err = -EIO;
5957                         }
5958                 }
5959
5960                 start += bvec->bv_len;
5961                 private++;
5962                 bvec++;
5963         } while (bvec <= bvec_end);
5964
5965         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5966                       dip->logical_offset + dip->bytes - 1);
5967         bio->bi_private = dip->private;
5968
5969         kfree(dip->csums);
5970         kfree(dip);
5971
5972         /* If we had a csum failure make sure to clear the uptodate flag */
5973         if (err)
5974                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5975         dio_end_io(bio, err);
5976 }
5977
5978 static void btrfs_endio_direct_write(struct bio *bio, int err)
5979 {
5980         struct btrfs_dio_private *dip = bio->bi_private;
5981         struct inode *inode = dip->inode;
5982         struct btrfs_root *root = BTRFS_I(inode)->root;
5983         struct btrfs_ordered_extent *ordered = NULL;
5984         u64 ordered_offset = dip->logical_offset;
5985         u64 ordered_bytes = dip->bytes;
5986         int ret;
5987
5988         if (err)
5989                 goto out_done;
5990 again:
5991         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5992                                                    &ordered_offset,
5993                                                    ordered_bytes, !err);
5994         if (!ret)
5995                 goto out_test;
5996
5997         ordered->work.func = finish_ordered_fn;
5998         ordered->work.flags = 0;
5999         btrfs_queue_worker(&root->fs_info->endio_write_workers,
6000                            &ordered->work);
6001 out_test:
6002         /*
6003          * our bio might span multiple ordered extents.  If we haven't
6004          * completed the accounting for the whole dio, go back and try again
6005          */
6006         if (ordered_offset < dip->logical_offset + dip->bytes) {
6007                 ordered_bytes = dip->logical_offset + dip->bytes -
6008                         ordered_offset;
6009                 ordered = NULL;
6010                 goto again;
6011         }
6012 out_done:
6013         bio->bi_private = dip->private;
6014
6015         kfree(dip);
6016
6017         /* If we had an error make sure to clear the uptodate flag */
6018         if (err)
6019                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
6020         dio_end_io(bio, err);
6021 }
6022
6023 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
6024                                     struct bio *bio, int mirror_num,
6025                                     unsigned long bio_flags, u64 offset)
6026 {
6027         int ret;
6028         struct btrfs_root *root = BTRFS_I(inode)->root;
6029         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
6030         BUG_ON(ret); /* -ENOMEM */
6031         return 0;
6032 }
6033
6034 static void btrfs_end_dio_bio(struct bio *bio, int err)
6035 {
6036         struct btrfs_dio_private *dip = bio->bi_private;
6037
6038         if (err) {
6039                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
6040                       "sector %#Lx len %u err no %d\n",
6041                       (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
6042                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
6043                 dip->errors = 1;
6044
6045                 /*
6046                  * before atomic variable goto zero, we must make sure
6047                  * dip->errors is perceived to be set.
6048                  */
6049                 smp_mb__before_atomic_dec();
6050         }
6051
6052         /* if there are more bios still pending for this dio, just exit */
6053         if (!atomic_dec_and_test(&dip->pending_bios))
6054                 goto out;
6055
6056         if (dip->errors)
6057                 bio_io_error(dip->orig_bio);
6058         else {
6059                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
6060                 bio_endio(dip->orig_bio, 0);
6061         }
6062 out:
6063         bio_put(bio);
6064 }
6065
6066 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
6067                                        u64 first_sector, gfp_t gfp_flags)
6068 {
6069         int nr_vecs = bio_get_nr_vecs(bdev);
6070         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
6071 }
6072
6073 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
6074                                          int rw, u64 file_offset, int skip_sum,
6075                                          u32 *csums, int async_submit)
6076 {
6077         int write = rw & REQ_WRITE;
6078         struct btrfs_root *root = BTRFS_I(inode)->root;
6079         int ret;
6080
6081         bio_get(bio);
6082
6083         if (!write) {
6084                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
6085                 if (ret)
6086                         goto err;
6087         }
6088
6089         if (skip_sum)
6090                 goto map;
6091
6092         if (write && async_submit) {
6093                 ret = btrfs_wq_submit_bio(root->fs_info,
6094                                    inode, rw, bio, 0, 0,
6095                                    file_offset,
6096                                    __btrfs_submit_bio_start_direct_io,
6097                                    __btrfs_submit_bio_done);
6098                 goto err;
6099         } else if (write) {
6100                 /*
6101                  * If we aren't doing async submit, calculate the csum of the
6102                  * bio now.
6103                  */
6104                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
6105                 if (ret)
6106                         goto err;
6107         } else if (!skip_sum) {
6108                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
6109                                           file_offset, csums);
6110                 if (ret)
6111                         goto err;
6112         }
6113
6114 map:
6115         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
6116 err:
6117         bio_put(bio);
6118         return ret;
6119 }
6120
6121 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
6122                                     int skip_sum)
6123 {
6124         struct inode *inode = dip->inode;
6125         struct btrfs_root *root = BTRFS_I(inode)->root;
6126         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6127         struct bio *bio;
6128         struct bio *orig_bio = dip->orig_bio;
6129         struct bio_vec *bvec = orig_bio->bi_io_vec;
6130         u64 start_sector = orig_bio->bi_sector;
6131         u64 file_offset = dip->logical_offset;
6132         u64 submit_len = 0;
6133         u64 map_length;
6134         int nr_pages = 0;
6135         u32 *csums = dip->csums;
6136         int ret = 0;
6137         int async_submit = 0;
6138         int write = rw & REQ_WRITE;
6139
6140         map_length = orig_bio->bi_size;
6141         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6142                               &map_length, NULL, 0);
6143         if (ret) {
6144                 bio_put(orig_bio);
6145                 return -EIO;
6146         }
6147
6148         if (map_length >= orig_bio->bi_size) {
6149                 bio = orig_bio;
6150                 goto submit;
6151         }
6152
6153         async_submit = 1;
6154         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
6155         if (!bio)
6156                 return -ENOMEM;
6157         bio->bi_private = dip;
6158         bio->bi_end_io = btrfs_end_dio_bio;
6159         atomic_inc(&dip->pending_bios);
6160
6161         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6162                 if (unlikely(map_length < submit_len + bvec->bv_len ||
6163                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6164                                  bvec->bv_offset) < bvec->bv_len)) {
6165                         /*
6166                          * inc the count before we submit the bio so
6167                          * we know the end IO handler won't happen before
6168                          * we inc the count. Otherwise, the dip might get freed
6169                          * before we're done setting it up
6170                          */
6171                         atomic_inc(&dip->pending_bios);
6172                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
6173                                                      file_offset, skip_sum,
6174                                                      csums, async_submit);
6175                         if (ret) {
6176                                 bio_put(bio);
6177                                 atomic_dec(&dip->pending_bios);
6178                                 goto out_err;
6179                         }
6180
6181                         /* Write's use the ordered csums */
6182                         if (!write && !skip_sum)
6183                                 csums = csums + nr_pages;
6184                         start_sector += submit_len >> 9;
6185                         file_offset += submit_len;
6186
6187                         submit_len = 0;
6188                         nr_pages = 0;
6189
6190                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6191                                                   start_sector, GFP_NOFS);
6192                         if (!bio)
6193                                 goto out_err;
6194                         bio->bi_private = dip;
6195                         bio->bi_end_io = btrfs_end_dio_bio;
6196
6197                         map_length = orig_bio->bi_size;
6198                         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6199                                               &map_length, NULL, 0);
6200                         if (ret) {
6201                                 bio_put(bio);
6202                                 goto out_err;
6203                         }
6204                 } else {
6205                         submit_len += bvec->bv_len;
6206                         nr_pages ++;
6207                         bvec++;
6208                 }
6209         }
6210
6211 submit:
6212         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6213                                      csums, async_submit);
6214         if (!ret)
6215                 return 0;
6216
6217         bio_put(bio);
6218 out_err:
6219         dip->errors = 1;
6220         /*
6221          * before atomic variable goto zero, we must
6222          * make sure dip->errors is perceived to be set.
6223          */
6224         smp_mb__before_atomic_dec();
6225         if (atomic_dec_and_test(&dip->pending_bios))
6226                 bio_io_error(dip->orig_bio);
6227
6228         /* bio_end_io() will handle error, so we needn't return it */
6229         return 0;
6230 }
6231
6232 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6233                                 loff_t file_offset)
6234 {
6235         struct btrfs_root *root = BTRFS_I(inode)->root;
6236         struct btrfs_dio_private *dip;
6237         struct bio_vec *bvec = bio->bi_io_vec;
6238         int skip_sum;
6239         int write = rw & REQ_WRITE;
6240         int ret = 0;
6241
6242         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6243
6244         dip = kmalloc(sizeof(*dip), GFP_NOFS);
6245         if (!dip) {
6246                 ret = -ENOMEM;
6247                 goto free_ordered;
6248         }
6249         dip->csums = NULL;
6250
6251         /* Write's use the ordered csum stuff, so we don't need dip->csums */
6252         if (!write && !skip_sum) {
6253                 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6254                 if (!dip->csums) {
6255                         kfree(dip);
6256                         ret = -ENOMEM;
6257                         goto free_ordered;
6258                 }
6259         }
6260
6261         dip->private = bio->bi_private;
6262         dip->inode = inode;
6263         dip->logical_offset = file_offset;
6264
6265         dip->bytes = 0;
6266         do {
6267                 dip->bytes += bvec->bv_len;
6268                 bvec++;
6269         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6270
6271         dip->disk_bytenr = (u64)bio->bi_sector << 9;
6272         bio->bi_private = dip;
6273         dip->errors = 0;
6274         dip->orig_bio = bio;
6275         atomic_set(&dip->pending_bios, 0);
6276
6277         if (write)
6278                 bio->bi_end_io = btrfs_endio_direct_write;
6279         else
6280                 bio->bi_end_io = btrfs_endio_direct_read;
6281
6282         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6283         if (!ret)
6284                 return;
6285 free_ordered:
6286         /*
6287          * If this is a write, we need to clean up the reserved space and kill
6288          * the ordered extent.
6289          */
6290         if (write) {
6291                 struct btrfs_ordered_extent *ordered;
6292                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6293                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6294                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6295                         btrfs_free_reserved_extent(root, ordered->start,
6296                                                    ordered->disk_len);
6297                 btrfs_put_ordered_extent(ordered);
6298                 btrfs_put_ordered_extent(ordered);
6299         }
6300         bio_endio(bio, ret);
6301 }
6302
6303 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6304                         const struct iovec *iov, loff_t offset,
6305                         unsigned long nr_segs)
6306 {
6307         int seg;
6308         int i;
6309         size_t size;
6310         unsigned long addr;
6311         unsigned blocksize_mask = root->sectorsize - 1;
6312         ssize_t retval = -EINVAL;
6313         loff_t end = offset;
6314
6315         if (offset & blocksize_mask)
6316                 goto out;
6317
6318         /* Check the memory alignment.  Blocks cannot straddle pages */
6319         for (seg = 0; seg < nr_segs; seg++) {
6320                 addr = (unsigned long)iov[seg].iov_base;
6321                 size = iov[seg].iov_len;
6322                 end += size;
6323                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6324                         goto out;
6325
6326                 /* If this is a write we don't need to check anymore */
6327                 if (rw & WRITE)
6328                         continue;
6329
6330                 /*
6331                  * Check to make sure we don't have duplicate iov_base's in this
6332                  * iovec, if so return EINVAL, otherwise we'll get csum errors
6333                  * when reading back.
6334                  */
6335                 for (i = seg + 1; i < nr_segs; i++) {
6336                         if (iov[seg].iov_base == iov[i].iov_base)
6337                                 goto out;
6338                 }
6339         }
6340         retval = 0;
6341 out:
6342         return retval;
6343 }
6344 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6345                         const struct iovec *iov, loff_t offset,
6346                         unsigned long nr_segs)
6347 {
6348         struct file *file = iocb->ki_filp;
6349         struct inode *inode = file->f_mapping->host;
6350         struct btrfs_ordered_extent *ordered;
6351         struct extent_state *cached_state = NULL;
6352         u64 lockstart, lockend;
6353         ssize_t ret;
6354         int writing = rw & WRITE;
6355         int write_bits = 0;
6356         size_t count = iov_length(iov, nr_segs);
6357
6358         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6359                             offset, nr_segs)) {
6360                 return 0;
6361         }
6362
6363         lockstart = offset;
6364         lockend = offset + count - 1;
6365
6366         if (writing) {
6367                 ret = btrfs_delalloc_reserve_space(inode, count);
6368                 if (ret)
6369                         goto out;
6370         }
6371
6372         while (1) {
6373                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6374                                  0, &cached_state);
6375                 /*
6376                  * We're concerned with the entire range that we're going to be
6377                  * doing DIO to, so we need to make sure theres no ordered
6378                  * extents in this range.
6379                  */
6380                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6381                                                      lockend - lockstart + 1);
6382
6383                 /*
6384                  * We need to make sure there are no buffered pages in this
6385                  * range either, we could have raced between the invalidate in
6386                  * generic_file_direct_write and locking the extent.  The
6387                  * invalidate needs to happen so that reads after a write do not
6388                  * get stale data.
6389                  */
6390                 if (!ordered && (!writing ||
6391                     !test_range_bit(&BTRFS_I(inode)->io_tree,
6392                                     lockstart, lockend, EXTENT_UPTODATE, 0,
6393                                     cached_state)))
6394                         break;
6395
6396                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6397                                      &cached_state, GFP_NOFS);
6398
6399                 if (ordered) {
6400                         btrfs_start_ordered_extent(inode, ordered, 1);
6401                         btrfs_put_ordered_extent(ordered);
6402                 } else {
6403                         /* Screw you mmap */
6404                         ret = filemap_write_and_wait_range(file->f_mapping,
6405                                                            lockstart,
6406                                                            lockend);
6407                         if (ret)
6408                                 goto out;
6409
6410                         /*
6411                          * If we found a page that couldn't be invalidated just
6412                          * fall back to buffered.
6413                          */
6414                         ret = invalidate_inode_pages2_range(file->f_mapping,
6415                                         lockstart >> PAGE_CACHE_SHIFT,
6416                                         lockend >> PAGE_CACHE_SHIFT);
6417                         if (ret) {
6418                                 if (ret == -EBUSY)
6419                                         ret = 0;
6420                                 goto out;
6421                         }
6422                 }
6423
6424                 cond_resched();
6425         }
6426
6427         /*
6428          * we don't use btrfs_set_extent_delalloc because we don't want
6429          * the dirty or uptodate bits
6430          */
6431         if (writing) {
6432                 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6433                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6434                                      EXTENT_DELALLOC, NULL, &cached_state,
6435                                      GFP_NOFS);
6436                 if (ret) {
6437                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6438                                          lockend, EXTENT_LOCKED | write_bits,
6439                                          1, 0, &cached_state, GFP_NOFS);
6440                         goto out;
6441                 }
6442         }
6443
6444         free_extent_state(cached_state);
6445         cached_state = NULL;
6446
6447         ret = __blockdev_direct_IO(rw, iocb, inode,
6448                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6449                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6450                    btrfs_submit_direct, 0);
6451
6452         if (ret < 0 && ret != -EIOCBQUEUED) {
6453                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6454                               offset + iov_length(iov, nr_segs) - 1,
6455                               EXTENT_LOCKED | write_bits, 1, 0,
6456                               &cached_state, GFP_NOFS);
6457         } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6458                 /*
6459                  * We're falling back to buffered, unlock the section we didn't
6460                  * do IO on.
6461                  */
6462                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6463                               offset + iov_length(iov, nr_segs) - 1,
6464                               EXTENT_LOCKED | write_bits, 1, 0,
6465                               &cached_state, GFP_NOFS);
6466         }
6467 out:
6468         free_extent_state(cached_state);
6469         return ret;
6470 }
6471
6472 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6473                 __u64 start, __u64 len)
6474 {
6475         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6476 }
6477
6478 int btrfs_readpage(struct file *file, struct page *page)
6479 {
6480         struct extent_io_tree *tree;
6481         tree = &BTRFS_I(page->mapping->host)->io_tree;
6482         return extent_read_full_page(tree, page, btrfs_get_extent, 0);
6483 }
6484
6485 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6486 {
6487         struct extent_io_tree *tree;
6488
6489
6490         if (current->flags & PF_MEMALLOC) {
6491                 redirty_page_for_writepage(wbc, page);
6492                 unlock_page(page);
6493                 return 0;
6494         }
6495         tree = &BTRFS_I(page->mapping->host)->io_tree;
6496         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6497 }
6498
6499 int btrfs_writepages(struct address_space *mapping,
6500                      struct writeback_control *wbc)
6501 {
6502         struct extent_io_tree *tree;
6503
6504         tree = &BTRFS_I(mapping->host)->io_tree;
6505         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6506 }
6507
6508 static int
6509 btrfs_readpages(struct file *file, struct address_space *mapping,
6510                 struct list_head *pages, unsigned nr_pages)
6511 {
6512         struct extent_io_tree *tree;
6513         tree = &BTRFS_I(mapping->host)->io_tree;
6514         return extent_readpages(tree, mapping, pages, nr_pages,
6515                                 btrfs_get_extent);
6516 }
6517 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6518 {
6519         struct extent_io_tree *tree;
6520         struct extent_map_tree *map;
6521         int ret;
6522
6523         tree = &BTRFS_I(page->mapping->host)->io_tree;
6524         map = &BTRFS_I(page->mapping->host)->extent_tree;
6525         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6526         if (ret == 1) {
6527                 ClearPagePrivate(page);
6528                 set_page_private(page, 0);
6529                 page_cache_release(page);
6530         }
6531         return ret;
6532 }
6533
6534 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6535 {
6536         if (PageWriteback(page) || PageDirty(page))
6537                 return 0;
6538         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6539 }
6540
6541 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6542 {
6543         struct inode *inode = page->mapping->host;
6544         struct extent_io_tree *tree;
6545         struct btrfs_ordered_extent *ordered;
6546         struct extent_state *cached_state = NULL;
6547         u64 page_start = page_offset(page);
6548         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6549
6550         /*
6551          * we have the page locked, so new writeback can't start,
6552          * and the dirty bit won't be cleared while we are here.
6553          *
6554          * Wait for IO on this page so that we can safely clear
6555          * the PagePrivate2 bit and do ordered accounting
6556          */
6557         wait_on_page_writeback(page);
6558
6559         tree = &BTRFS_I(inode)->io_tree;
6560         if (offset) {
6561                 btrfs_releasepage(page, GFP_NOFS);
6562                 return;
6563         }
6564         lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6565         ordered = btrfs_lookup_ordered_extent(inode,
6566                                            page_offset(page));
6567         if (ordered) {
6568                 /*
6569                  * IO on this page will never be started, so we need
6570                  * to account for any ordered extents now
6571                  */
6572                 clear_extent_bit(tree, page_start, page_end,
6573                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6574                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6575                                  &cached_state, GFP_NOFS);
6576                 /*
6577                  * whoever cleared the private bit is responsible
6578                  * for the finish_ordered_io
6579                  */
6580                 if (TestClearPagePrivate2(page) &&
6581                     btrfs_dec_test_ordered_pending(inode, &ordered, page_start,
6582                                                    PAGE_CACHE_SIZE, 1)) {
6583                         btrfs_finish_ordered_io(ordered);
6584                 }
6585                 btrfs_put_ordered_extent(ordered);
6586                 cached_state = NULL;
6587                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6588         }
6589         clear_extent_bit(tree, page_start, page_end,
6590                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6591                  EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6592         __btrfs_releasepage(page, GFP_NOFS);
6593
6594         ClearPageChecked(page);
6595         if (PagePrivate(page)) {
6596                 ClearPagePrivate(page);
6597                 set_page_private(page, 0);
6598                 page_cache_release(page);
6599         }
6600 }
6601
6602 /*
6603  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6604  * called from a page fault handler when a page is first dirtied. Hence we must
6605  * be careful to check for EOF conditions here. We set the page up correctly
6606  * for a written page which means we get ENOSPC checking when writing into
6607  * holes and correct delalloc and unwritten extent mapping on filesystems that
6608  * support these features.
6609  *
6610  * We are not allowed to take the i_mutex here so we have to play games to
6611  * protect against truncate races as the page could now be beyond EOF.  Because
6612  * vmtruncate() writes the inode size before removing pages, once we have the
6613  * page lock we can determine safely if the page is beyond EOF. If it is not
6614  * beyond EOF, then the page is guaranteed safe against truncation until we
6615  * unlock the page.
6616  */
6617 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6618 {
6619         struct page *page = vmf->page;
6620         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6621         struct btrfs_root *root = BTRFS_I(inode)->root;
6622         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6623         struct btrfs_ordered_extent *ordered;
6624         struct extent_state *cached_state = NULL;
6625         char *kaddr;
6626         unsigned long zero_start;
6627         loff_t size;
6628         int ret;
6629         int reserved = 0;
6630         u64 page_start;
6631         u64 page_end;
6632
6633         sb_start_pagefault(inode->i_sb);
6634         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6635         if (!ret) {
6636                 ret = file_update_time(vma->vm_file);
6637                 reserved = 1;
6638         }
6639         if (ret) {
6640                 if (ret == -ENOMEM)
6641                         ret = VM_FAULT_OOM;
6642                 else /* -ENOSPC, -EIO, etc */
6643                         ret = VM_FAULT_SIGBUS;
6644                 if (reserved)
6645                         goto out;
6646                 goto out_noreserve;
6647         }
6648
6649         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6650 again:
6651         lock_page(page);
6652         size = i_size_read(inode);
6653         page_start = page_offset(page);
6654         page_end = page_start + PAGE_CACHE_SIZE - 1;
6655
6656         if ((page->mapping != inode->i_mapping) ||
6657             (page_start >= size)) {
6658                 /* page got truncated out from underneath us */
6659                 goto out_unlock;
6660         }
6661         wait_on_page_writeback(page);
6662
6663         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
6664         set_page_extent_mapped(page);
6665
6666         /*
6667          * we can't set the delalloc bits if there are pending ordered
6668          * extents.  Drop our locks and wait for them to finish
6669          */
6670         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6671         if (ordered) {
6672                 unlock_extent_cached(io_tree, page_start, page_end,
6673                                      &cached_state, GFP_NOFS);
6674                 unlock_page(page);
6675                 btrfs_start_ordered_extent(inode, ordered, 1);
6676                 btrfs_put_ordered_extent(ordered);
6677                 goto again;
6678         }
6679
6680         /*
6681          * XXX - page_mkwrite gets called every time the page is dirtied, even
6682          * if it was already dirty, so for space accounting reasons we need to
6683          * clear any delalloc bits for the range we are fixing to save.  There
6684          * is probably a better way to do this, but for now keep consistent with
6685          * prepare_pages in the normal write path.
6686          */
6687         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6688                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6689                           0, 0, &cached_state, GFP_NOFS);
6690
6691         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6692                                         &cached_state);
6693         if (ret) {
6694                 unlock_extent_cached(io_tree, page_start, page_end,
6695                                      &cached_state, GFP_NOFS);
6696                 ret = VM_FAULT_SIGBUS;
6697                 goto out_unlock;
6698         }
6699         ret = 0;
6700
6701         /* page is wholly or partially inside EOF */
6702         if (page_start + PAGE_CACHE_SIZE > size)
6703                 zero_start = size & ~PAGE_CACHE_MASK;
6704         else
6705                 zero_start = PAGE_CACHE_SIZE;
6706
6707         if (zero_start != PAGE_CACHE_SIZE) {
6708                 kaddr = kmap(page);
6709                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6710                 flush_dcache_page(page);
6711                 kunmap(page);
6712         }
6713         ClearPageChecked(page);
6714         set_page_dirty(page);
6715         SetPageUptodate(page);
6716
6717         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6718         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6719
6720         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6721
6722 out_unlock:
6723         if (!ret) {
6724                 sb_end_pagefault(inode->i_sb);
6725                 return VM_FAULT_LOCKED;
6726         }
6727         unlock_page(page);
6728 out:
6729         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6730 out_noreserve:
6731         sb_end_pagefault(inode->i_sb);
6732         return ret;
6733 }
6734
6735 static int btrfs_truncate(struct inode *inode)
6736 {
6737         struct btrfs_root *root = BTRFS_I(inode)->root;
6738         struct btrfs_block_rsv *rsv;
6739         int ret;
6740         int err = 0;
6741         struct btrfs_trans_handle *trans;
6742         unsigned long nr;
6743         u64 mask = root->sectorsize - 1;
6744         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
6745
6746         ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6747         if (ret)
6748                 return ret;
6749
6750         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6751         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6752
6753         /*
6754          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
6755          * 3 things going on here
6756          *
6757          * 1) We need to reserve space for our orphan item and the space to
6758          * delete our orphan item.  Lord knows we don't want to have a dangling
6759          * orphan item because we didn't reserve space to remove it.
6760          *
6761          * 2) We need to reserve space to update our inode.
6762          *
6763          * 3) We need to have something to cache all the space that is going to
6764          * be free'd up by the truncate operation, but also have some slack
6765          * space reserved in case it uses space during the truncate (thank you
6766          * very much snapshotting).
6767          *
6768          * And we need these to all be seperate.  The fact is we can use alot of
6769          * space doing the truncate, and we have no earthly idea how much space
6770          * we will use, so we need the truncate reservation to be seperate so it
6771          * doesn't end up using space reserved for updating the inode or
6772          * removing the orphan item.  We also need to be able to stop the
6773          * transaction and start a new one, which means we need to be able to
6774          * update the inode several times, and we have no idea of knowing how
6775          * many times that will be, so we can't just reserve 1 item for the
6776          * entirety of the opration, so that has to be done seperately as well.
6777          * Then there is the orphan item, which does indeed need to be held on
6778          * to for the whole operation, and we need nobody to touch this reserved
6779          * space except the orphan code.
6780          *
6781          * So that leaves us with
6782          *
6783          * 1) root->orphan_block_rsv - for the orphan deletion.
6784          * 2) rsv - for the truncate reservation, which we will steal from the
6785          * transaction reservation.
6786          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6787          * updating the inode.
6788          */
6789         rsv = btrfs_alloc_block_rsv(root);
6790         if (!rsv)
6791                 return -ENOMEM;
6792         rsv->size = min_size;
6793
6794         /*
6795          * 1 for the truncate slack space
6796          * 1 for the orphan item we're going to add
6797          * 1 for the orphan item deletion
6798          * 1 for updating the inode.
6799          */
6800         trans = btrfs_start_transaction(root, 4);
6801         if (IS_ERR(trans)) {
6802                 err = PTR_ERR(trans);
6803                 goto out;
6804         }
6805
6806         /* Migrate the slack space for the truncate to our reserve */
6807         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6808                                       min_size);
6809         BUG_ON(ret);
6810
6811         ret = btrfs_orphan_add(trans, inode);
6812         if (ret) {
6813                 btrfs_end_transaction(trans, root);
6814                 goto out;
6815         }
6816
6817         /*
6818          * setattr is responsible for setting the ordered_data_close flag,
6819          * but that is only tested during the last file release.  That
6820          * could happen well after the next commit, leaving a great big
6821          * window where new writes may get lost if someone chooses to write
6822          * to this file after truncating to zero
6823          *
6824          * The inode doesn't have any dirty data here, and so if we commit
6825          * this is a noop.  If someone immediately starts writing to the inode
6826          * it is very likely we'll catch some of their writes in this
6827          * transaction, and the commit will find this file on the ordered
6828          * data list with good things to send down.
6829          *
6830          * This is a best effort solution, there is still a window where
6831          * using truncate to replace the contents of the file will
6832          * end up with a zero length file after a crash.
6833          */
6834         if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
6835                                            &BTRFS_I(inode)->runtime_flags))
6836                 btrfs_add_ordered_operation(trans, root, inode);
6837
6838         while (1) {
6839                 ret = btrfs_block_rsv_refill(root, rsv, min_size);
6840                 if (ret) {
6841                         /*
6842                          * This can only happen with the original transaction we
6843                          * started above, every other time we shouldn't have a
6844                          * transaction started yet.
6845                          */
6846                         if (ret == -EAGAIN)
6847                                 goto end_trans;
6848                         err = ret;
6849                         break;
6850                 }
6851
6852                 if (!trans) {
6853                         /* Just need the 1 for updating the inode */
6854                         trans = btrfs_start_transaction(root, 1);
6855                         if (IS_ERR(trans)) {
6856                                 ret = err = PTR_ERR(trans);
6857                                 trans = NULL;
6858                                 break;
6859                         }
6860                 }
6861
6862                 trans->block_rsv = rsv;
6863
6864                 ret = btrfs_truncate_inode_items(trans, root, inode,
6865                                                  inode->i_size,
6866                                                  BTRFS_EXTENT_DATA_KEY);
6867                 if (ret != -EAGAIN) {
6868                         err = ret;
6869                         break;
6870                 }
6871
6872                 trans->block_rsv = &root->fs_info->trans_block_rsv;
6873                 ret = btrfs_update_inode(trans, root, inode);
6874                 if (ret) {
6875                         err = ret;
6876                         break;
6877                 }
6878 end_trans:
6879                 nr = trans->blocks_used;
6880                 btrfs_end_transaction(trans, root);
6881                 trans = NULL;
6882                 btrfs_btree_balance_dirty(root, nr);
6883         }
6884
6885         if (ret == 0 && inode->i_nlink > 0) {
6886                 trans->block_rsv = root->orphan_block_rsv;
6887                 ret = btrfs_orphan_del(trans, inode);
6888                 if (ret)
6889                         err = ret;
6890         } else if (ret && inode->i_nlink > 0) {
6891                 /*
6892                  * Failed to do the truncate, remove us from the in memory
6893                  * orphan list.
6894                  */
6895                 ret = btrfs_orphan_del(NULL, inode);
6896         }
6897
6898         if (trans) {
6899                 trans->block_rsv = &root->fs_info->trans_block_rsv;
6900                 ret = btrfs_update_inode(trans, root, inode);
6901                 if (ret && !err)
6902                         err = ret;
6903
6904                 nr = trans->blocks_used;
6905                 ret = btrfs_end_transaction(trans, root);
6906                 btrfs_btree_balance_dirty(root, nr);
6907         }
6908
6909 out:
6910         btrfs_free_block_rsv(root, rsv);
6911
6912         if (ret && !err)
6913                 err = ret;
6914
6915         return err;
6916 }
6917
6918 /*
6919  * create a new subvolume directory/inode (helper for the ioctl).
6920  */
6921 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6922                              struct btrfs_root *new_root, u64 new_dirid)
6923 {
6924         struct inode *inode;
6925         int err;
6926         u64 index = 0;
6927
6928         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
6929                                 new_dirid, new_dirid,
6930                                 S_IFDIR | (~current_umask() & S_IRWXUGO),
6931                                 &index);
6932         if (IS_ERR(inode))
6933                 return PTR_ERR(inode);
6934         inode->i_op = &btrfs_dir_inode_operations;
6935         inode->i_fop = &btrfs_dir_file_operations;
6936
6937         set_nlink(inode, 1);
6938         btrfs_i_size_write(inode, 0);
6939
6940         err = btrfs_update_inode(trans, new_root, inode);
6941
6942         iput(inode);
6943         return err;
6944 }
6945
6946 struct inode *btrfs_alloc_inode(struct super_block *sb)
6947 {
6948         struct btrfs_inode *ei;
6949         struct inode *inode;
6950
6951         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6952         if (!ei)
6953                 return NULL;
6954
6955         ei->root = NULL;
6956         ei->generation = 0;
6957         ei->last_trans = 0;
6958         ei->last_sub_trans = 0;
6959         ei->logged_trans = 0;
6960         ei->delalloc_bytes = 0;
6961         ei->disk_i_size = 0;
6962         ei->flags = 0;
6963         ei->csum_bytes = 0;
6964         ei->index_cnt = (u64)-1;
6965         ei->last_unlink_trans = 0;
6966
6967         spin_lock_init(&ei->lock);
6968         ei->outstanding_extents = 0;
6969         ei->reserved_extents = 0;
6970
6971         ei->runtime_flags = 0;
6972         ei->force_compress = BTRFS_COMPRESS_NONE;
6973
6974         ei->delayed_node = NULL;
6975
6976         inode = &ei->vfs_inode;
6977         extent_map_tree_init(&ei->extent_tree);
6978         extent_io_tree_init(&ei->io_tree, &inode->i_data);
6979         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
6980         ei->io_tree.track_uptodate = 1;
6981         ei->io_failure_tree.track_uptodate = 1;
6982         mutex_init(&ei->log_mutex);
6983         mutex_init(&ei->delalloc_mutex);
6984         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6985         INIT_LIST_HEAD(&ei->delalloc_inodes);
6986         INIT_LIST_HEAD(&ei->ordered_operations);
6987         RB_CLEAR_NODE(&ei->rb_node);
6988
6989         return inode;
6990 }
6991
6992 static void btrfs_i_callback(struct rcu_head *head)
6993 {
6994         struct inode *inode = container_of(head, struct inode, i_rcu);
6995         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6996 }
6997
6998 void btrfs_destroy_inode(struct inode *inode)
6999 {
7000         struct btrfs_ordered_extent *ordered;
7001         struct btrfs_root *root = BTRFS_I(inode)->root;
7002
7003         WARN_ON(!hlist_empty(&inode->i_dentry));
7004         WARN_ON(inode->i_data.nrpages);
7005         WARN_ON(BTRFS_I(inode)->outstanding_extents);
7006         WARN_ON(BTRFS_I(inode)->reserved_extents);
7007         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
7008         WARN_ON(BTRFS_I(inode)->csum_bytes);
7009
7010         /*
7011          * This can happen where we create an inode, but somebody else also
7012          * created the same inode and we need to destroy the one we already
7013          * created.
7014          */
7015         if (!root)
7016                 goto free;
7017
7018         /*
7019          * Make sure we're properly removed from the ordered operation
7020          * lists.
7021          */
7022         smp_mb();
7023         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
7024                 spin_lock(&root->fs_info->ordered_extent_lock);
7025                 list_del_init(&BTRFS_I(inode)->ordered_operations);
7026                 spin_unlock(&root->fs_info->ordered_extent_lock);
7027         }
7028
7029         if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
7030                      &BTRFS_I(inode)->runtime_flags)) {
7031                 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
7032                        (unsigned long long)btrfs_ino(inode));
7033                 atomic_dec(&root->orphan_inodes);
7034         }
7035
7036         while (1) {
7037                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7038                 if (!ordered)
7039                         break;
7040                 else {
7041                         printk(KERN_ERR "btrfs found ordered "
7042                                "extent %llu %llu on inode cleanup\n",
7043                                (unsigned long long)ordered->file_offset,
7044                                (unsigned long long)ordered->len);
7045                         btrfs_remove_ordered_extent(inode, ordered);
7046                         btrfs_put_ordered_extent(ordered);
7047                         btrfs_put_ordered_extent(ordered);
7048                 }
7049         }
7050         inode_tree_del(inode);
7051         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
7052 free:
7053         btrfs_remove_delayed_node(inode);
7054         call_rcu(&inode->i_rcu, btrfs_i_callback);
7055 }
7056
7057 int btrfs_drop_inode(struct inode *inode)
7058 {
7059         struct btrfs_root *root = BTRFS_I(inode)->root;
7060
7061         if (btrfs_root_refs(&root->root_item) == 0 &&
7062             !btrfs_is_free_space_inode(inode))
7063                 return 1;
7064         else
7065                 return generic_drop_inode(inode);
7066 }
7067
7068 static void init_once(void *foo)
7069 {
7070         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
7071
7072         inode_init_once(&ei->vfs_inode);
7073 }
7074
7075 void btrfs_destroy_cachep(void)
7076 {
7077         if (btrfs_inode_cachep)
7078                 kmem_cache_destroy(btrfs_inode_cachep);
7079         if (btrfs_trans_handle_cachep)
7080                 kmem_cache_destroy(btrfs_trans_handle_cachep);
7081         if (btrfs_transaction_cachep)
7082                 kmem_cache_destroy(btrfs_transaction_cachep);
7083         if (btrfs_path_cachep)
7084                 kmem_cache_destroy(btrfs_path_cachep);
7085         if (btrfs_free_space_cachep)
7086                 kmem_cache_destroy(btrfs_free_space_cachep);
7087 }
7088
7089 int btrfs_init_cachep(void)
7090 {
7091         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
7092                         sizeof(struct btrfs_inode), 0,
7093                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
7094         if (!btrfs_inode_cachep)
7095                 goto fail;
7096
7097         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
7098                         sizeof(struct btrfs_trans_handle), 0,
7099                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7100         if (!btrfs_trans_handle_cachep)
7101                 goto fail;
7102
7103         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
7104                         sizeof(struct btrfs_transaction), 0,
7105                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7106         if (!btrfs_transaction_cachep)
7107                 goto fail;
7108
7109         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
7110                         sizeof(struct btrfs_path), 0,
7111                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7112         if (!btrfs_path_cachep)
7113                 goto fail;
7114
7115         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
7116                         sizeof(struct btrfs_free_space), 0,
7117                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7118         if (!btrfs_free_space_cachep)
7119                 goto fail;
7120
7121         return 0;
7122 fail:
7123         btrfs_destroy_cachep();
7124         return -ENOMEM;
7125 }
7126
7127 static int btrfs_getattr(struct vfsmount *mnt,
7128                          struct dentry *dentry, struct kstat *stat)
7129 {
7130         struct inode *inode = dentry->d_inode;
7131         u32 blocksize = inode->i_sb->s_blocksize;
7132
7133         generic_fillattr(inode, stat);
7134         stat->dev = BTRFS_I(inode)->root->anon_dev;
7135         stat->blksize = PAGE_CACHE_SIZE;
7136         stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
7137                 ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9;
7138         return 0;
7139 }
7140
7141 /*
7142  * If a file is moved, it will inherit the cow and compression flags of the new
7143  * directory.
7144  */
7145 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
7146 {
7147         struct btrfs_inode *b_dir = BTRFS_I(dir);
7148         struct btrfs_inode *b_inode = BTRFS_I(inode);
7149
7150         if (b_dir->flags & BTRFS_INODE_NODATACOW)
7151                 b_inode->flags |= BTRFS_INODE_NODATACOW;
7152         else
7153                 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
7154
7155         if (b_dir->flags & BTRFS_INODE_COMPRESS) {
7156                 b_inode->flags |= BTRFS_INODE_COMPRESS;
7157                 b_inode->flags &= ~BTRFS_INODE_NOCOMPRESS;
7158         } else {
7159                 b_inode->flags &= ~(BTRFS_INODE_COMPRESS |
7160                                     BTRFS_INODE_NOCOMPRESS);
7161         }
7162 }
7163
7164 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
7165                            struct inode *new_dir, struct dentry *new_dentry)
7166 {
7167         struct btrfs_trans_handle *trans;
7168         struct btrfs_root *root = BTRFS_I(old_dir)->root;
7169         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
7170         struct inode *new_inode = new_dentry->d_inode;
7171         struct inode *old_inode = old_dentry->d_inode;
7172         struct timespec ctime = CURRENT_TIME;
7173         u64 index = 0;
7174         u64 root_objectid;
7175         int ret;
7176         u64 old_ino = btrfs_ino(old_inode);
7177
7178         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
7179                 return -EPERM;
7180
7181         /* we only allow rename subvolume link between subvolumes */
7182         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
7183                 return -EXDEV;
7184
7185         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
7186             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
7187                 return -ENOTEMPTY;
7188
7189         if (S_ISDIR(old_inode->i_mode) && new_inode &&
7190             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
7191                 return -ENOTEMPTY;
7192         /*
7193          * we're using rename to replace one file with another.
7194          * and the replacement file is large.  Start IO on it now so
7195          * we don't add too much work to the end of the transaction
7196          */
7197         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
7198             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
7199                 filemap_flush(old_inode->i_mapping);
7200
7201         /* close the racy window with snapshot create/destroy ioctl */
7202         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7203                 down_read(&root->fs_info->subvol_sem);
7204         /*
7205          * We want to reserve the absolute worst case amount of items.  So if
7206          * both inodes are subvols and we need to unlink them then that would
7207          * require 4 item modifications, but if they are both normal inodes it
7208          * would require 5 item modifications, so we'll assume their normal
7209          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7210          * should cover the worst case number of items we'll modify.
7211          */
7212         trans = btrfs_start_transaction(root, 20);
7213         if (IS_ERR(trans)) {
7214                 ret = PTR_ERR(trans);
7215                 goto out_notrans;
7216         }
7217
7218         if (dest != root)
7219                 btrfs_record_root_in_trans(trans, dest);
7220
7221         ret = btrfs_set_inode_index(new_dir, &index);
7222         if (ret)
7223                 goto out_fail;
7224
7225         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7226                 /* force full log commit if subvolume involved. */
7227                 root->fs_info->last_trans_log_full_commit = trans->transid;
7228         } else {
7229                 ret = btrfs_insert_inode_ref(trans, dest,
7230                                              new_dentry->d_name.name,
7231                                              new_dentry->d_name.len,
7232                                              old_ino,
7233                                              btrfs_ino(new_dir), index);
7234                 if (ret)
7235                         goto out_fail;
7236                 /*
7237                  * this is an ugly little race, but the rename is required
7238                  * to make sure that if we crash, the inode is either at the
7239                  * old name or the new one.  pinning the log transaction lets
7240                  * us make sure we don't allow a log commit to come in after
7241                  * we unlink the name but before we add the new name back in.
7242                  */
7243                 btrfs_pin_log_trans(root);
7244         }
7245         /*
7246          * make sure the inode gets flushed if it is replacing
7247          * something.
7248          */
7249         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7250                 btrfs_add_ordered_operation(trans, root, old_inode);
7251
7252         inode_inc_iversion(old_dir);
7253         inode_inc_iversion(new_dir);
7254         inode_inc_iversion(old_inode);
7255         old_dir->i_ctime = old_dir->i_mtime = ctime;
7256         new_dir->i_ctime = new_dir->i_mtime = ctime;
7257         old_inode->i_ctime = ctime;
7258
7259         if (old_dentry->d_parent != new_dentry->d_parent)
7260                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7261
7262         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7263                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7264                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7265                                         old_dentry->d_name.name,
7266                                         old_dentry->d_name.len);
7267         } else {
7268                 ret = __btrfs_unlink_inode(trans, root, old_dir,
7269                                         old_dentry->d_inode,
7270                                         old_dentry->d_name.name,
7271                                         old_dentry->d_name.len);
7272                 if (!ret)
7273                         ret = btrfs_update_inode(trans, root, old_inode);
7274         }
7275         if (ret) {
7276                 btrfs_abort_transaction(trans, root, ret);
7277                 goto out_fail;
7278         }
7279
7280         if (new_inode) {
7281                 inode_inc_iversion(new_inode);
7282                 new_inode->i_ctime = CURRENT_TIME;
7283                 if (unlikely(btrfs_ino(new_inode) ==
7284                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7285                         root_objectid = BTRFS_I(new_inode)->location.objectid;
7286                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
7287                                                 root_objectid,
7288                                                 new_dentry->d_name.name,
7289                                                 new_dentry->d_name.len);
7290                         BUG_ON(new_inode->i_nlink == 0);
7291                 } else {
7292                         ret = btrfs_unlink_inode(trans, dest, new_dir,
7293                                                  new_dentry->d_inode,
7294                                                  new_dentry->d_name.name,
7295                                                  new_dentry->d_name.len);
7296                 }
7297                 if (!ret && new_inode->i_nlink == 0) {
7298                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7299                         BUG_ON(ret);
7300                 }
7301                 if (ret) {
7302                         btrfs_abort_transaction(trans, root, ret);
7303                         goto out_fail;
7304                 }
7305         }
7306
7307         fixup_inode_flags(new_dir, old_inode);
7308
7309         ret = btrfs_add_link(trans, new_dir, old_inode,
7310                              new_dentry->d_name.name,
7311                              new_dentry->d_name.len, 0, index);
7312         if (ret) {
7313                 btrfs_abort_transaction(trans, root, ret);
7314                 goto out_fail;
7315         }
7316
7317         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7318                 struct dentry *parent = new_dentry->d_parent;
7319                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7320                 btrfs_end_log_trans(root);
7321         }
7322 out_fail:
7323         btrfs_end_transaction(trans, root);
7324 out_notrans:
7325         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7326                 up_read(&root->fs_info->subvol_sem);
7327
7328         return ret;
7329 }
7330
7331 /*
7332  * some fairly slow code that needs optimization. This walks the list
7333  * of all the inodes with pending delalloc and forces them to disk.
7334  */
7335 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7336 {
7337         struct list_head *head = &root->fs_info->delalloc_inodes;
7338         struct btrfs_inode *binode;
7339         struct inode *inode;
7340
7341         if (root->fs_info->sb->s_flags & MS_RDONLY)
7342                 return -EROFS;
7343
7344         spin_lock(&root->fs_info->delalloc_lock);
7345         while (!list_empty(head)) {
7346                 binode = list_entry(head->next, struct btrfs_inode,
7347                                     delalloc_inodes);
7348                 inode = igrab(&binode->vfs_inode);
7349                 if (!inode)
7350                         list_del_init(&binode->delalloc_inodes);
7351                 spin_unlock(&root->fs_info->delalloc_lock);
7352                 if (inode) {
7353                         filemap_flush(inode->i_mapping);
7354                         if (delay_iput)
7355                                 btrfs_add_delayed_iput(inode);
7356                         else
7357                                 iput(inode);
7358                 }
7359                 cond_resched();
7360                 spin_lock(&root->fs_info->delalloc_lock);
7361         }
7362         spin_unlock(&root->fs_info->delalloc_lock);
7363
7364         /* the filemap_flush will queue IO into the worker threads, but
7365          * we have to make sure the IO is actually started and that
7366          * ordered extents get created before we return
7367          */
7368         atomic_inc(&root->fs_info->async_submit_draining);
7369         while (atomic_read(&root->fs_info->nr_async_submits) ||
7370               atomic_read(&root->fs_info->async_delalloc_pages)) {
7371                 wait_event(root->fs_info->async_submit_wait,
7372                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7373                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7374         }
7375         atomic_dec(&root->fs_info->async_submit_draining);
7376         return 0;
7377 }
7378
7379 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7380                          const char *symname)
7381 {
7382         struct btrfs_trans_handle *trans;
7383         struct btrfs_root *root = BTRFS_I(dir)->root;
7384         struct btrfs_path *path;
7385         struct btrfs_key key;
7386         struct inode *inode = NULL;
7387         int err;
7388         int drop_inode = 0;
7389         u64 objectid;
7390         u64 index = 0 ;
7391         int name_len;
7392         int datasize;
7393         unsigned long ptr;
7394         struct btrfs_file_extent_item *ei;
7395         struct extent_buffer *leaf;
7396         unsigned long nr = 0;
7397
7398         name_len = strlen(symname) + 1;
7399         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7400                 return -ENAMETOOLONG;
7401
7402         /*
7403          * 2 items for inode item and ref
7404          * 2 items for dir items
7405          * 1 item for xattr if selinux is on
7406          */
7407         trans = btrfs_start_transaction(root, 5);
7408         if (IS_ERR(trans))
7409                 return PTR_ERR(trans);
7410
7411         err = btrfs_find_free_ino(root, &objectid);
7412         if (err)
7413                 goto out_unlock;
7414
7415         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7416                                 dentry->d_name.len, btrfs_ino(dir), objectid,
7417                                 S_IFLNK|S_IRWXUGO, &index);
7418         if (IS_ERR(inode)) {
7419                 err = PTR_ERR(inode);
7420                 goto out_unlock;
7421         }
7422
7423         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7424         if (err) {
7425                 drop_inode = 1;
7426                 goto out_unlock;
7427         }
7428
7429         /*
7430         * If the active LSM wants to access the inode during
7431         * d_instantiate it needs these. Smack checks to see
7432         * if the filesystem supports xattrs by looking at the
7433         * ops vector.
7434         */
7435         inode->i_fop = &btrfs_file_operations;
7436         inode->i_op = &btrfs_file_inode_operations;
7437
7438         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7439         if (err)
7440                 drop_inode = 1;
7441         else {
7442                 inode->i_mapping->a_ops = &btrfs_aops;
7443                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7444                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7445         }
7446         if (drop_inode)
7447                 goto out_unlock;
7448
7449         path = btrfs_alloc_path();
7450         if (!path) {
7451                 err = -ENOMEM;
7452                 drop_inode = 1;
7453                 goto out_unlock;
7454         }
7455         key.objectid = btrfs_ino(inode);
7456         key.offset = 0;
7457         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7458         datasize = btrfs_file_extent_calc_inline_size(name_len);
7459         err = btrfs_insert_empty_item(trans, root, path, &key,
7460                                       datasize);
7461         if (err) {
7462                 drop_inode = 1;
7463                 btrfs_free_path(path);
7464                 goto out_unlock;
7465         }
7466         leaf = path->nodes[0];
7467         ei = btrfs_item_ptr(leaf, path->slots[0],
7468                             struct btrfs_file_extent_item);
7469         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7470         btrfs_set_file_extent_type(leaf, ei,
7471                                    BTRFS_FILE_EXTENT_INLINE);
7472         btrfs_set_file_extent_encryption(leaf, ei, 0);
7473         btrfs_set_file_extent_compression(leaf, ei, 0);
7474         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7475         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7476
7477         ptr = btrfs_file_extent_inline_start(ei);
7478         write_extent_buffer(leaf, symname, ptr, name_len);
7479         btrfs_mark_buffer_dirty(leaf);
7480         btrfs_free_path(path);
7481
7482         inode->i_op = &btrfs_symlink_inode_operations;
7483         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7484         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7485         inode_set_bytes(inode, name_len);
7486         btrfs_i_size_write(inode, name_len - 1);
7487         err = btrfs_update_inode(trans, root, inode);
7488         if (err)
7489                 drop_inode = 1;
7490
7491 out_unlock:
7492         if (!err)
7493                 d_instantiate(dentry, inode);
7494         nr = trans->blocks_used;
7495         btrfs_end_transaction(trans, root);
7496         if (drop_inode) {
7497                 inode_dec_link_count(inode);
7498                 iput(inode);
7499         }
7500         btrfs_btree_balance_dirty(root, nr);
7501         return err;
7502 }
7503
7504 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7505                                        u64 start, u64 num_bytes, u64 min_size,
7506                                        loff_t actual_len, u64 *alloc_hint,
7507                                        struct btrfs_trans_handle *trans)
7508 {
7509         struct btrfs_root *root = BTRFS_I(inode)->root;
7510         struct btrfs_key ins;
7511         u64 cur_offset = start;
7512         u64 i_size;
7513         int ret = 0;
7514         bool own_trans = true;
7515
7516         if (trans)
7517                 own_trans = false;
7518         while (num_bytes > 0) {
7519                 if (own_trans) {
7520                         trans = btrfs_start_transaction(root, 3);
7521                         if (IS_ERR(trans)) {
7522                                 ret = PTR_ERR(trans);
7523                                 break;
7524                         }
7525                 }
7526
7527                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7528                                            0, *alloc_hint, &ins, 1);
7529                 if (ret) {
7530                         if (own_trans)
7531                                 btrfs_end_transaction(trans, root);
7532                         break;
7533                 }
7534
7535                 ret = insert_reserved_file_extent(trans, inode,
7536                                                   cur_offset, ins.objectid,
7537                                                   ins.offset, ins.offset,
7538                                                   ins.offset, 0, 0, 0,
7539                                                   BTRFS_FILE_EXTENT_PREALLOC);
7540                 if (ret) {
7541                         btrfs_abort_transaction(trans, root, ret);
7542                         if (own_trans)
7543                                 btrfs_end_transaction(trans, root);
7544                         break;
7545                 }
7546                 btrfs_drop_extent_cache(inode, cur_offset,
7547                                         cur_offset + ins.offset -1, 0);
7548
7549                 num_bytes -= ins.offset;
7550                 cur_offset += ins.offset;
7551                 *alloc_hint = ins.objectid + ins.offset;
7552
7553                 inode_inc_iversion(inode);
7554                 inode->i_ctime = CURRENT_TIME;
7555                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7556                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7557                     (actual_len > inode->i_size) &&
7558                     (cur_offset > inode->i_size)) {
7559                         if (cur_offset > actual_len)
7560                                 i_size = actual_len;
7561                         else
7562                                 i_size = cur_offset;
7563                         i_size_write(inode, i_size);
7564                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7565                 }
7566
7567                 ret = btrfs_update_inode(trans, root, inode);
7568
7569                 if (ret) {
7570                         btrfs_abort_transaction(trans, root, ret);
7571                         if (own_trans)
7572                                 btrfs_end_transaction(trans, root);
7573                         break;
7574                 }
7575
7576                 if (own_trans)
7577                         btrfs_end_transaction(trans, root);
7578         }
7579         return ret;
7580 }
7581
7582 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7583                               u64 start, u64 num_bytes, u64 min_size,
7584                               loff_t actual_len, u64 *alloc_hint)
7585 {
7586         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7587                                            min_size, actual_len, alloc_hint,
7588                                            NULL);
7589 }
7590
7591 int btrfs_prealloc_file_range_trans(struct inode *inode,
7592                                     struct btrfs_trans_handle *trans, int mode,
7593                                     u64 start, u64 num_bytes, u64 min_size,
7594                                     loff_t actual_len, u64 *alloc_hint)
7595 {
7596         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7597                                            min_size, actual_len, alloc_hint, trans);
7598 }
7599
7600 static int btrfs_set_page_dirty(struct page *page)
7601 {
7602         return __set_page_dirty_nobuffers(page);
7603 }
7604
7605 static int btrfs_permission(struct inode *inode, int mask)
7606 {
7607         struct btrfs_root *root = BTRFS_I(inode)->root;
7608         umode_t mode = inode->i_mode;
7609
7610         if (mask & MAY_WRITE &&
7611             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7612                 if (btrfs_root_readonly(root))
7613                         return -EROFS;
7614                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7615                         return -EACCES;
7616         }
7617         return generic_permission(inode, mask);
7618 }
7619
7620 static const struct inode_operations btrfs_dir_inode_operations = {
7621         .getattr        = btrfs_getattr,
7622         .lookup         = btrfs_lookup,
7623         .create         = btrfs_create,
7624         .unlink         = btrfs_unlink,
7625         .link           = btrfs_link,
7626         .mkdir          = btrfs_mkdir,
7627         .rmdir          = btrfs_rmdir,
7628         .rename         = btrfs_rename,
7629         .symlink        = btrfs_symlink,
7630         .setattr        = btrfs_setattr,
7631         .mknod          = btrfs_mknod,
7632         .setxattr       = btrfs_setxattr,
7633         .getxattr       = btrfs_getxattr,
7634         .listxattr      = btrfs_listxattr,
7635         .removexattr    = btrfs_removexattr,
7636         .permission     = btrfs_permission,
7637         .get_acl        = btrfs_get_acl,
7638 };
7639 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7640         .lookup         = btrfs_lookup,
7641         .permission     = btrfs_permission,
7642         .get_acl        = btrfs_get_acl,
7643 };
7644
7645 static const struct file_operations btrfs_dir_file_operations = {
7646         .llseek         = generic_file_llseek,
7647         .read           = generic_read_dir,
7648         .readdir        = btrfs_real_readdir,
7649         .unlocked_ioctl = btrfs_ioctl,
7650 #ifdef CONFIG_COMPAT
7651         .compat_ioctl   = btrfs_ioctl,
7652 #endif
7653         .release        = btrfs_release_file,
7654         .fsync          = btrfs_sync_file,
7655 };
7656
7657 static struct extent_io_ops btrfs_extent_io_ops = {
7658         .fill_delalloc = run_delalloc_range,
7659         .submit_bio_hook = btrfs_submit_bio_hook,
7660         .merge_bio_hook = btrfs_merge_bio_hook,
7661         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7662         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7663         .writepage_start_hook = btrfs_writepage_start_hook,
7664         .set_bit_hook = btrfs_set_bit_hook,
7665         .clear_bit_hook = btrfs_clear_bit_hook,
7666         .merge_extent_hook = btrfs_merge_extent_hook,
7667         .split_extent_hook = btrfs_split_extent_hook,
7668 };
7669
7670 /*
7671  * btrfs doesn't support the bmap operation because swapfiles
7672  * use bmap to make a mapping of extents in the file.  They assume
7673  * these extents won't change over the life of the file and they
7674  * use the bmap result to do IO directly to the drive.
7675  *
7676  * the btrfs bmap call would return logical addresses that aren't
7677  * suitable for IO and they also will change frequently as COW
7678  * operations happen.  So, swapfile + btrfs == corruption.
7679  *
7680  * For now we're avoiding this by dropping bmap.
7681  */
7682 static const struct address_space_operations btrfs_aops = {
7683         .readpage       = btrfs_readpage,
7684         .writepage      = btrfs_writepage,
7685         .writepages     = btrfs_writepages,
7686         .readpages      = btrfs_readpages,
7687         .direct_IO      = btrfs_direct_IO,
7688         .invalidatepage = btrfs_invalidatepage,
7689         .releasepage    = btrfs_releasepage,
7690         .set_page_dirty = btrfs_set_page_dirty,
7691         .error_remove_page = generic_error_remove_page,
7692 };
7693
7694 static const struct address_space_operations btrfs_symlink_aops = {
7695         .readpage       = btrfs_readpage,
7696         .writepage      = btrfs_writepage,
7697         .invalidatepage = btrfs_invalidatepage,
7698         .releasepage    = btrfs_releasepage,
7699 };
7700
7701 static const struct inode_operations btrfs_file_inode_operations = {
7702         .getattr        = btrfs_getattr,
7703         .setattr        = btrfs_setattr,
7704         .setxattr       = btrfs_setxattr,
7705         .getxattr       = btrfs_getxattr,
7706         .listxattr      = btrfs_listxattr,
7707         .removexattr    = btrfs_removexattr,
7708         .permission     = btrfs_permission,
7709         .fiemap         = btrfs_fiemap,
7710         .get_acl        = btrfs_get_acl,
7711         .update_time    = btrfs_update_time,
7712 };
7713 static const struct inode_operations btrfs_special_inode_operations = {
7714         .getattr        = btrfs_getattr,
7715         .setattr        = btrfs_setattr,
7716         .permission     = btrfs_permission,
7717         .setxattr       = btrfs_setxattr,
7718         .getxattr       = btrfs_getxattr,
7719         .listxattr      = btrfs_listxattr,
7720         .removexattr    = btrfs_removexattr,
7721         .get_acl        = btrfs_get_acl,
7722         .update_time    = btrfs_update_time,
7723 };
7724 static const struct inode_operations btrfs_symlink_inode_operations = {
7725         .readlink       = generic_readlink,
7726         .follow_link    = page_follow_link_light,
7727         .put_link       = page_put_link,
7728         .getattr        = btrfs_getattr,
7729         .setattr        = btrfs_setattr,
7730         .permission     = btrfs_permission,
7731         .setxattr       = btrfs_setxattr,
7732         .getxattr       = btrfs_getxattr,
7733         .listxattr      = btrfs_listxattr,
7734         .removexattr    = btrfs_removexattr,
7735         .get_acl        = btrfs_get_acl,
7736         .update_time    = btrfs_update_time,
7737 };
7738
7739 const struct dentry_operations btrfs_dentry_operations = {
7740         .d_delete       = btrfs_dentry_delete,
7741         .d_release      = btrfs_dentry_release,
7742 };