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 pdflush sent them down.
 328  */
 329 static noinline int compress_file_range(struct inode *inode,
 330                                         struct page *locked_page,
 331                                         u64 start, u64 end,
 332                                         struct async_cow *async_cow,
 333                                         int *num_added)
 334 {
 335         struct btrfs_root *root = BTRFS_I(inode)->root;
 336         struct btrfs_trans_handle *trans;
 337         u64 num_bytes;
 338         u64 blocksize = root->sectorsize;
 339         u64 actual_end;
 340         u64 isize = i_size_read(inode);
 341         int ret = 0;
 342         struct page **pages = NULL;
 343         unsigned long nr_pages;
 344         unsigned long nr_pages_ret = 0;
 345         unsigned long total_compressed = 0;
 346         unsigned long total_in = 0;
 347         unsigned long max_compressed = 128 * 1024;
 348         unsigned long max_uncompressed = 128 * 1024;
 349         int i;
 350         int will_compress;
 351         int compress_type = root->fs_info->compress_type;
 352
 353         /* if this is a small write inside eof, kick off a defrag */
 354         if ((end - start + 1) < 16 * 1024 &&
 355             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 356                 btrfs_add_inode_defrag(NULL, inode);
 357
 358         actual_end = min_t(u64, isize, end + 1);
 359 again:
 360         will_compress = 0;
 361         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 362         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 363
 364         /*
 365          * we don't want to send crud past the end of i_size through
 366          * compression, that's just a waste of CPU time.  So, if the
 367          * end of the file is before the start of our current
 368          * requested range of bytes, we bail out to the uncompressed
 369          * cleanup code that can deal with all of this.
 370          *
 371          * It isn't really the fastest way to fix things, but this is a
 372          * very uncommon corner.
 373          */
 374         if (actual_end <= start)
 375                 goto cleanup_and_bail_uncompressed;
 376
 377         total_compressed = actual_end - start;
 378
 379         /* we want to make sure that amount of ram required to uncompress
 380          * an extent is reasonable, so we limit the total size in ram
 381          * of a compressed extent to 128k.  This is a crucial number
 382          * because it also controls how easily we can spread reads across
 383          * cpus for decompression.
 384          *
 385          * We also want to make sure the amount of IO required to do
 386          * a random read is reasonably small, so we limit the size of
 387          * a compressed extent to 128k.
 388          */
 389         total_compressed = min(total_compressed, max_uncompressed);
 390         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 391         num_bytes = max(blocksize,  num_bytes);
 392         total_in = 0;
 393         ret = 0;
 394
 395         /*
 396          * we do compression for mount -o compress and when the
 397          * inode has not been flagged as nocompress.  This flag can
 398          * change at any time if we discover bad compression ratios.
 399          */
 400         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 401             (btrfs_test_opt(root, COMPRESS) ||
 402              (BTRFS_I(inode)->force_compress) ||
 403              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 404                 WARN_ON(pages);
 405                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 406                 if (!pages) {
 407                         /* just bail out to the uncompressed code */
 408                         goto cont;
 409                 }
 410
 411                 if (BTRFS_I(inode)->force_compress)
 412                         compress_type = BTRFS_I(inode)->force_compress;
 413
 414                 ret = btrfs_compress_pages(compress_type,
 415                                            inode->i_mapping, start,
 416                                            total_compressed, pages,
 417                                            nr_pages, &nr_pages_ret,
 418                                            &total_in,
 419                                            &total_compressed,
 420                                            max_compressed);
 421
 422                 if (!ret) {
 423                         unsigned long offset = total_compressed &
 424                                 (PAGE_CACHE_SIZE - 1);
 425                         struct page *page = pages[nr_pages_ret - 1];
 426                         char *kaddr;
 427
 428                         /* zero the tail end of the last page, we might be
 429                          * sending it down to disk
 430                          */
 431                         if (offset) {
 432                                 kaddr = kmap_atomic(page);
 433                                 memset(kaddr + offset, 0,
 434                                        PAGE_CACHE_SIZE - offset);
 435                                 kunmap_atomic(kaddr);
 436                         }
 437                         will_compress = 1;
 438                 }
 439         }
 440 cont:
 441         if (start == 0) {
 442                 trans = btrfs_join_transaction(root);
 443                 if (IS_ERR(trans)) {
 444                         ret = PTR_ERR(trans);
 445                         trans = NULL;
 446                         goto cleanup_and_out;
 447                 }
 448                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 449
 450                 /* lets try to make an inline extent */
 451                 if (ret || total_in < (actual_end - start)) {
 452                         /* we didn't compress the entire range, try
 453                          * to make an uncompressed inline extent.
 454                          */
 455                         ret = cow_file_range_inline(trans, root, inode,
 456                                                     start, end, 0, 0, NULL);
 457                 } else {
 458                         /* try making a compressed inline extent */
 459                         ret = cow_file_range_inline(trans, root, inode,
 460                                                     start, end,
 461                                                     total_compressed,
 462                                                     compress_type, pages);
 463                 }
 464                 if (ret <= 0) {
 465                         /*
 466                          * inline extent creation worked or returned error,
 467                          * we don't need to create any more async work items.
 468                          * Unlock and free up our temp pages.
 469                          */
 470                         extent_clear_unlock_delalloc(inode,
 471                              &BTRFS_I(inode)->io_tree,
 472                              start, end, NULL,
 473                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
 474                              EXTENT_CLEAR_DELALLOC |
 475                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
 476
 477                         btrfs_end_transaction(trans, root);
 478                         goto free_pages_out;
 479                 }
 480                 btrfs_end_transaction(trans, root);
 481         }
 482
 483         if (will_compress) {
 484                 /*
 485                  * we aren't doing an inline extent round the compressed size
 486                  * up to a block size boundary so the allocator does sane
 487                  * things
 488                  */
 489                 total_compressed = (total_compressed + blocksize - 1) &
 490                         ~(blocksize - 1);
 491
 492                 /*
 493                  * one last check to make sure the compression is really a
 494                  * win, compare the page count read with the blocks on disk
 495                  */
 496                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
 497                         ~(PAGE_CACHE_SIZE - 1);
 498                 if (total_compressed >= total_in) {
 499                         will_compress = 0;
 500                 } else {
 501                         num_bytes = total_in;
 502                 }
 503         }
 504         if (!will_compress && pages) {
 505                 /*
 506                  * the compression code ran but failed to make things smaller,
 507                  * free any pages it allocated and our page pointer array
 508                  */
 509                 for (i = 0; i < nr_pages_ret; i++) {
 510                         WARN_ON(pages[i]->mapping);
 511                         page_cache_release(pages[i]);
 512                 }
 513                 kfree(pages);
 514                 pages = NULL;
 515                 total_compressed = 0;
 516                 nr_pages_ret = 0;
 517
 518                 /* flag the file so we don't compress in the future */
 519                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 520                     !(BTRFS_I(inode)->force_compress)) {
 521                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 522                 }
 523         }
 524         if (will_compress) {
 525                 *num_added += 1;
 526
 527                 /* the async work queues will take care of doing actual
 528                  * allocation on disk for these compressed pages,
 529                  * and will submit them to the elevator.
 530                  */
 531                 add_async_extent(async_cow, start, num_bytes,
 532                                  total_compressed, pages, nr_pages_ret,
 533                                  compress_type);
 534
 535                 if (start + num_bytes < end) {
 536                         start += num_bytes;
 537                         pages = NULL;
 538                         cond_resched();
 539                         goto again;
 540                 }
 541         } else {
 542 cleanup_and_bail_uncompressed:
 543                 /*
 544                  * No compression, but we still need to write the pages in
 545                  * the file we've been given so far.  redirty the locked
 546                  * page if it corresponds to our extent and set things up
 547                  * for the async work queue to run cow_file_range to do
 548                  * the normal delalloc dance
 549                  */
 550                 if (page_offset(locked_page) >= start &&
 551                     page_offset(locked_page) <= end) {
 552                         __set_page_dirty_nobuffers(locked_page);
 553                         /* unlocked later on in the async handlers */
 554                 }
 555                 add_async_extent(async_cow, start, end - start + 1,
 556                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 557                 *num_added += 1;
 558         }
 559
 560 out:
 561         return ret;
 562
 563 free_pages_out:
 564         for (i = 0; i < nr_pages_ret; i++) {
 565                 WARN_ON(pages[i]->mapping);
 566                 page_cache_release(pages[i]);
 567         }
 568         kfree(pages);
 569
 570         goto out;
 571
 572 cleanup_and_out:
 573         extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 574                                      start, end, NULL,
 575                                      EXTENT_CLEAR_UNLOCK_PAGE |
 576                                      EXTENT_CLEAR_DIRTY |
 577                                      EXTENT_CLEAR_DELALLOC |
 578                                      EXTENT_SET_WRITEBACK |
 579                                      EXTENT_END_WRITEBACK);
 580         if (!trans || IS_ERR(trans))
 581                 btrfs_error(root->fs_info, ret, "Failed to join transaction");
 582         else
 583                 btrfs_abort_transaction(trans, root, ret);
 584         goto free_pages_out;
 585 }
 586
 587 /*
 588  * phase two of compressed writeback.  This is the ordered portion
 589  * of the code, which only gets called in the order the work was
 590  * queued.  We walk all the async extents created by compress_file_range
 591  * and send them down to the disk.
 592  */
 593 static noinline int submit_compressed_extents(struct inode *inode,
 594                                               struct async_cow *async_cow)
 595 {
 596         struct async_extent *async_extent;
 597         u64 alloc_hint = 0;
 598         struct btrfs_trans_handle *trans;
 599         struct btrfs_key ins;
 600         struct extent_map *em;
 601         struct btrfs_root *root = BTRFS_I(inode)->root;
 602         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 603         struct extent_io_tree *io_tree;
 604         int ret = 0;
 605
 606         if (list_empty(&async_cow->extents))
 607                 return 0;
 608
 609
 610         while (!list_empty(&async_cow->extents)) {
 611                 async_extent = list_entry(async_cow->extents.next,
 612                                           struct async_extent, list);
 613                 list_del(&async_extent->list);
 614
 615                 io_tree = &BTRFS_I(inode)->io_tree;
 616
 617 retry:
 618                 /* did the compression code fall back to uncompressed IO? */
 619                 if (!async_extent->pages) {
 620                         int page_started = 0;
 621                         unsigned long nr_written = 0;
 622
 623                         lock_extent(io_tree, async_extent->start,
 624                                          async_extent->start +
 625                                          async_extent->ram_size - 1);
 626
 627                         /* allocate blocks */
 628                         ret = cow_file_range(inode, async_cow->locked_page,
 629                                              async_extent->start,
 630                                              async_extent->start +
 631                                              async_extent->ram_size - 1,
 632                                              &page_started, &nr_written, 0);
 633
 634                         /* JDM XXX */
 635
 636                         /*
 637                          * if page_started, cow_file_range inserted an
 638                          * inline extent and took care of all the unlocking
 639                          * and IO for us.  Otherwise, we need to submit
 640                          * all those pages down to the drive.
 641                          */
 642                         if (!page_started && !ret)
 643                                 extent_write_locked_range(io_tree,
 644                                                   inode, async_extent->start,
 645                                                   async_extent->start +
 646                                                   async_extent->ram_size - 1,
 647                                                   btrfs_get_extent,
 648                                                   WB_SYNC_ALL);
 649                         kfree(async_extent);
 650                         cond_resched();
 651                         continue;
 652                 }
 653
 654                 lock_extent(io_tree, async_extent->start,
 655                             async_extent->start + async_extent->ram_size - 1);
 656
 657                 trans = btrfs_join_transaction(root);
 658                 if (IS_ERR(trans)) {
 659                         ret = PTR_ERR(trans);
 660                 } else {
 661                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 662                         ret = btrfs_reserve_extent(trans, root,
 663                                            async_extent->compressed_size,
 664                                            async_extent->compressed_size,
 665                                            0, alloc_hint, &ins, 1);
 666                         if (ret)
 667                                 btrfs_abort_transaction(trans, root, ret);
 668                         btrfs_end_transaction(trans, root);
 669                 }
 670
 671                 if (ret) {
 672                         int i;
 673                         for (i = 0; i < async_extent->nr_pages; i++) {
 674                                 WARN_ON(async_extent->pages[i]->mapping);
 675                                 page_cache_release(async_extent->pages[i]);
 676                         }
 677                         kfree(async_extent->pages);
 678                         async_extent->nr_pages = 0;
 679                         async_extent->pages = NULL;
 680                         unlock_extent(io_tree, async_extent->start,
 681                                       async_extent->start +
 682                                       async_extent->ram_size - 1);
 683                         if (ret == -ENOSPC)
 684                                 goto retry;
 685                         goto out_free; /* JDM: Requeue? */
 686                 }
 687
 688                 /*
 689                  * here we're doing allocation and writeback of the
 690                  * compressed pages
 691                  */
 692                 btrfs_drop_extent_cache(inode, async_extent->start,
 693                                         async_extent->start +
 694                                         async_extent->ram_size - 1, 0);
 695
 696                 em = alloc_extent_map();
 697                 BUG_ON(!em); /* -ENOMEM */
 698                 em->start = async_extent->start;
 699                 em->len = async_extent->ram_size;
 700                 em->orig_start = em->start;
 701
 702                 em->block_start = ins.objectid;
 703                 em->block_len = ins.offset;
 704                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 705                 em->compress_type = async_extent->compress_type;
 706                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 707                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 708
 709                 while (1) {
 710                         write_lock(&em_tree->lock);
 711                         ret = add_extent_mapping(em_tree, em);
 712                         write_unlock(&em_tree->lock);
 713                         if (ret != -EEXIST) {
 714                                 free_extent_map(em);
 715                                 break;
 716                         }
 717                         btrfs_drop_extent_cache(inode, async_extent->start,
 718                                                 async_extent->start +
 719                                                 async_extent->ram_size - 1, 0);
 720                 }
 721
 722                 ret = btrfs_add_ordered_extent_compress(inode,
 723                                                 async_extent->start,
 724                                                 ins.objectid,
 725                                                 async_extent->ram_size,
 726                                                 ins.offset,
 727                                                 BTRFS_ORDERED_COMPRESSED,
 728                                                 async_extent->compress_type);
 729                 BUG_ON(ret); /* -ENOMEM */
 730
 731                 /*
 732                  * clear dirty, set writeback and unlock the pages.
 733                  */
 734                 extent_clear_unlock_delalloc(inode,
 735                                 &BTRFS_I(inode)->io_tree,
 736                                 async_extent->start,
 737                                 async_extent->start +
 738                                 async_extent->ram_size - 1,
 739                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 740                                 EXTENT_CLEAR_UNLOCK |
 741                                 EXTENT_CLEAR_DELALLOC |
 742                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
 743
 744                 ret = btrfs_submit_compressed_write(inode,
 745                                     async_extent->start,
 746                                     async_extent->ram_size,
 747                                     ins.objectid,
 748                                     ins.offset, async_extent->pages,
 749                                     async_extent->nr_pages);
 750
 751                 BUG_ON(ret); /* -ENOMEM */
 752                 alloc_hint = ins.objectid + ins.offset;
 753                 kfree(async_extent);
 754                 cond_resched();
 755         }
 756         ret = 0;
 757 out:
 758         return ret;
 759 out_free:
 760         kfree(async_extent);
 761         goto out;
 762 }
 763
 764 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 765                                       u64 num_bytes)
 766 {
 767         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 768         struct extent_map *em;
 769         u64 alloc_hint = 0;
 770
 771         read_lock(&em_tree->lock);
 772         em = search_extent_mapping(em_tree, start, num_bytes);
 773         if (em) {
 774                 /*
 775                  * if block start isn't an actual block number then find the
 776                  * first block in this inode and use that as a hint.  If that
 777                  * block is also bogus then just don't worry about it.
 778                  */
 779                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 780                         free_extent_map(em);
 781                         em = search_extent_mapping(em_tree, 0, 0);
 782                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 783                                 alloc_hint = em->block_start;
 784                         if (em)
 785                                 free_extent_map(em);
 786                 } else {
 787                         alloc_hint = em->block_start;
 788                         free_extent_map(em);
 789                 }
 790         }
 791         read_unlock(&em_tree->lock);
 792
 793         return alloc_hint;
 794 }
 795
 796 /*
 797  * when extent_io.c finds a delayed allocation range in the file,
 798  * the call backs end up in this code.  The basic idea is to
 799  * allocate extents on disk for the range, and create ordered data structs
 800  * in ram to track those extents.
 801  *
 802  * locked_page is the page that writepage had locked already.  We use
 803  * it to make sure we don't do extra locks or unlocks.
 804  *
 805  * *page_started is set to one if we unlock locked_page and do everything
 806  * required to start IO on it.  It may be clean and already done with
 807  * IO when we return.
 808  */
 809 static noinline int cow_file_range(struct inode *inode,
 810                                    struct page *locked_page,
 811                                    u64 start, u64 end, int *page_started,
 812                                    unsigned long *nr_written,
 813                                    int unlock)
 814 {
 815         struct btrfs_root *root = BTRFS_I(inode)->root;
 816         struct btrfs_trans_handle *trans;
 817         u64 alloc_hint = 0;
 818         u64 num_bytes;
 819         unsigned long ram_size;
 820         u64 disk_num_bytes;
 821         u64 cur_alloc_size;
 822         u64 blocksize = root->sectorsize;
 823         struct btrfs_key ins;
 824         struct extent_map *em;
 825         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 826         int ret = 0;
 827
 828         BUG_ON(btrfs_is_free_space_inode(root, inode));
 829         trans = btrfs_join_transaction(root);
 830         if (IS_ERR(trans)) {
 831                 extent_clear_unlock_delalloc(inode,
 832                              &BTRFS_I(inode)->io_tree,
 833                              start, end, locked_page,
 834                              EXTENT_CLEAR_UNLOCK_PAGE |
 835                              EXTENT_CLEAR_UNLOCK |
 836                              EXTENT_CLEAR_DELALLOC |
 837                              EXTENT_CLEAR_DIRTY |
 838                              EXTENT_SET_WRITEBACK |
 839                              EXTENT_END_WRITEBACK);
 840                 return PTR_ERR(trans);
 841         }
 842         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 843
 844         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 845         num_bytes = max(blocksize,  num_bytes);
 846         disk_num_bytes = num_bytes;
 847         ret = 0;
 848
 849         /* if this is a small write inside eof, kick off defrag */
 850         if (num_bytes < 64 * 1024 &&
 851             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 852                 btrfs_add_inode_defrag(trans, inode);
 853
 854         if (start == 0) {
 855                 /* lets try to make an inline extent */
 856                 ret = cow_file_range_inline(trans, root, inode,
 857                                             start, end, 0, 0, NULL);
 858                 if (ret == 0) {
 859                         extent_clear_unlock_delalloc(inode,
 860                                      &BTRFS_I(inode)->io_tree,
 861                                      start, end, NULL,
 862                                      EXTENT_CLEAR_UNLOCK_PAGE |
 863                                      EXTENT_CLEAR_UNLOCK |
 864                                      EXTENT_CLEAR_DELALLOC |
 865                                      EXTENT_CLEAR_DIRTY |
 866                                      EXTENT_SET_WRITEBACK |
 867                                      EXTENT_END_WRITEBACK);
 868
 869                         *nr_written = *nr_written +
 870                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 871                         *page_started = 1;
 872                         goto out;
 873                 } else if (ret < 0) {
 874                         btrfs_abort_transaction(trans, root, ret);
 875                         goto out_unlock;
 876                 }
 877         }
 878
 879         BUG_ON(disk_num_bytes >
 880                btrfs_super_total_bytes(root->fs_info->super_copy));
 881
 882         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 883         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 884
 885         while (disk_num_bytes > 0) {
 886                 unsigned long op;
 887
 888                 cur_alloc_size = disk_num_bytes;
 889                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 890                                            root->sectorsize, 0, alloc_hint,
 891                                            &ins, 1);
 892                 if (ret < 0) {
 893                         btrfs_abort_transaction(trans, root, ret);
 894                         goto out_unlock;
 895                 }
 896
 897                 em = alloc_extent_map();
 898                 BUG_ON(!em); /* -ENOMEM */
 899                 em->start = start;
 900                 em->orig_start = em->start;
 901                 ram_size = ins.offset;
 902                 em->len = ins.offset;
 903
 904                 em->block_start = ins.objectid;
 905                 em->block_len = ins.offset;
 906                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 907                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 908
 909                 while (1) {
 910                         write_lock(&em_tree->lock);
 911                         ret = add_extent_mapping(em_tree, em);
 912                         write_unlock(&em_tree->lock);
 913                         if (ret != -EEXIST) {
 914                                 free_extent_map(em);
 915                                 break;
 916                         }
 917                         btrfs_drop_extent_cache(inode, start,
 918                                                 start + ram_size - 1, 0);
 919                 }
 920
 921                 cur_alloc_size = ins.offset;
 922                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 923                                                ram_size, cur_alloc_size, 0);
 924                 BUG_ON(ret); /* -ENOMEM */
 925
 926                 if (root->root_key.objectid ==
 927                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 928                         ret = btrfs_reloc_clone_csums(inode, start,
 929                                                       cur_alloc_size);
 930                         if (ret) {
 931                                 btrfs_abort_transaction(trans, root, ret);
 932                                 goto out_unlock;
 933                         }
 934                 }
 935
 936                 if (disk_num_bytes < cur_alloc_size)
 937                         break;
 938
 939                 /* we're not doing compressed IO, don't unlock the first
 940                  * page (which the caller expects to stay locked), don't
 941                  * clear any dirty bits and don't set any writeback bits
 942                  *
 943                  * Do set the Private2 bit so we know this page was properly
 944                  * setup for writepage
 945                  */
 946                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
 947                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
 948                         EXTENT_SET_PRIVATE2;
 949
 950                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 951                                              start, start + ram_size - 1,
 952                                              locked_page, op);
 953                 disk_num_bytes -= cur_alloc_size;
 954                 num_bytes -= cur_alloc_size;
 955                 alloc_hint = ins.objectid + ins.offset;
 956                 start += cur_alloc_size;
 957         }
 958         ret = 0;
 959 out:
 960         btrfs_end_transaction(trans, root);
 961
 962         return ret;
 963 out_unlock:
 964         extent_clear_unlock_delalloc(inode,
 965                      &BTRFS_I(inode)->io_tree,
 966                      start, end, locked_page,
 967                      EXTENT_CLEAR_UNLOCK_PAGE |
 968                      EXTENT_CLEAR_UNLOCK |
 969                      EXTENT_CLEAR_DELALLOC |
 970                      EXTENT_CLEAR_DIRTY |
 971                      EXTENT_SET_WRITEBACK |
 972                      EXTENT_END_WRITEBACK);
 973
 974         goto out;
 975 }
 976
 977 /*
 978  * work queue call back to started compression on a file and pages
 979  */
 980 static noinline void async_cow_start(struct btrfs_work *work)
 981 {
 982         struct async_cow *async_cow;
 983         int num_added = 0;
 984         async_cow = container_of(work, struct async_cow, work);
 985
 986         compress_file_range(async_cow->inode, async_cow->locked_page,
 987                             async_cow->start, async_cow->end, async_cow,
 988                             &num_added);
 989         if (num_added == 0) {
 990                 btrfs_add_delayed_iput(async_cow->inode);
 991                 async_cow->inode = NULL;
 992         }
 993 }
 994
 995 /*
 996  * work queue call back to submit previously compressed pages
 997  */
 998 static noinline void async_cow_submit(struct btrfs_work *work)
 999 {
1000         struct async_cow *async_cow;
1001         struct btrfs_root *root;
1002         unsigned long nr_pages;
1003
1004         async_cow = container_of(work, struct async_cow, work);
1005
1006         root = async_cow->root;
1007         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1008                 PAGE_CACHE_SHIFT;
1009
1010         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
1011
1012         if (atomic_read(&root->fs_info->async_delalloc_pages) <
1013             5 * 1042 * 1024 &&
1014             waitqueue_active(&root->fs_info->async_submit_wait))
1015                 wake_up(&root->fs_info->async_submit_wait);
1016
1017         if (async_cow->inode)
1018                 submit_compressed_extents(async_cow->inode, async_cow);
1019 }
1020
1021 static noinline void async_cow_free(struct btrfs_work *work)
1022 {
1023         struct async_cow *async_cow;
1024         async_cow = container_of(work, struct async_cow, work);
1025         if (async_cow->inode)
1026                 btrfs_add_delayed_iput(async_cow->inode);
1027         kfree(async_cow);
1028 }
1029
1030 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1031                                 u64 start, u64 end, int *page_started,
1032                                 unsigned long *nr_written)
1033 {
1034         struct async_cow *async_cow;
1035         struct btrfs_root *root = BTRFS_I(inode)->root;
1036         unsigned long nr_pages;
1037         u64 cur_end;
1038         int limit = 10 * 1024 * 1042;
1039
1040         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1041                          1, 0, NULL, GFP_NOFS);
1042         while (start < end) {
1043                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1044                 BUG_ON(!async_cow); /* -ENOMEM */
1045                 async_cow->inode = igrab(inode);
1046                 async_cow->root = root;
1047                 async_cow->locked_page = locked_page;
1048                 async_cow->start = start;
1049
1050                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1051                         cur_end = end;
1052                 else
1053                         cur_end = min(end, start + 512 * 1024 - 1);
1054
1055                 async_cow->end = cur_end;
1056                 INIT_LIST_HEAD(&async_cow->extents);
1057
1058                 async_cow->work.func = async_cow_start;
1059                 async_cow->work.ordered_func = async_cow_submit;
1060                 async_cow->work.ordered_free = async_cow_free;
1061                 async_cow->work.flags = 0;
1062
1063                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1064                         PAGE_CACHE_SHIFT;
1065                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1066
1067                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1068                                    &async_cow->work);
1069
1070                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1071                         wait_event(root->fs_info->async_submit_wait,
1072                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1073                             limit));
1074                 }
1075
1076                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1077                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1078                         wait_event(root->fs_info->async_submit_wait,
1079                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1080                            0));
1081                 }
1082
1083                 *nr_written += nr_pages;
1084                 start = cur_end + 1;
1085         }
1086         *page_started = 1;
1087         return 0;
1088 }
1089
1090 static noinline int csum_exist_in_range(struct btrfs_root *root,
1091                                         u64 bytenr, u64 num_bytes)
1092 {
1093         int ret;
1094         struct btrfs_ordered_sum *sums;
1095         LIST_HEAD(list);
1096
1097         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1098                                        bytenr + num_bytes - 1, &list, 0);
1099         if (ret == 0 && list_empty(&list))
1100                 return 0;
1101
1102         while (!list_empty(&list)) {
1103                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1104                 list_del(&sums->list);
1105                 kfree(sums);
1106         }
1107         return 1;
1108 }
1109
1110 /*
1111  * when nowcow writeback call back.  This checks for snapshots or COW copies
1112  * of the extents that exist in the file, and COWs the file as required.
1113  *
1114  * If no cow copies or snapshots exist, we write directly to the existing
1115  * blocks on disk
1116  */
1117 static noinline int run_delalloc_nocow(struct inode *inode,
1118                                        struct page *locked_page,
1119                               u64 start, u64 end, int *page_started, int force,
1120                               unsigned long *nr_written)
1121 {
1122         struct btrfs_root *root = BTRFS_I(inode)->root;
1123         struct btrfs_trans_handle *trans;
1124         struct extent_buffer *leaf;
1125         struct btrfs_path *path;
1126         struct btrfs_file_extent_item *fi;
1127         struct btrfs_key found_key;
1128         u64 cow_start;
1129         u64 cur_offset;
1130         u64 extent_end;
1131         u64 extent_offset;
1132         u64 disk_bytenr;
1133         u64 num_bytes;
1134         int extent_type;
1135         int ret, err;
1136         int type;
1137         int nocow;
1138         int check_prev = 1;
1139         bool nolock;
1140         u64 ino = btrfs_ino(inode);
1141
1142         path = btrfs_alloc_path();
1143         if (!path) {
1144                 extent_clear_unlock_delalloc(inode,
1145                              &BTRFS_I(inode)->io_tree,
1146                              start, end, locked_page,
1147                              EXTENT_CLEAR_UNLOCK_PAGE |
1148                              EXTENT_CLEAR_UNLOCK |
1149                              EXTENT_CLEAR_DELALLOC |
1150                              EXTENT_CLEAR_DIRTY |
1151                              EXTENT_SET_WRITEBACK |
1152                              EXTENT_END_WRITEBACK);
1153                 return -ENOMEM;
1154         }
1155
1156         nolock = btrfs_is_free_space_inode(root, inode);
1157
1158         if (nolock)
1159                 trans = btrfs_join_transaction_nolock(root);
1160         else
1161                 trans = btrfs_join_transaction(root);
1162
1163         if (IS_ERR(trans)) {
1164                 extent_clear_unlock_delalloc(inode,
1165                              &BTRFS_I(inode)->io_tree,
1166                              start, end, locked_page,
1167                              EXTENT_CLEAR_UNLOCK_PAGE |
1168                              EXTENT_CLEAR_UNLOCK |
1169                              EXTENT_CLEAR_DELALLOC |
1170                              EXTENT_CLEAR_DIRTY |
1171                              EXTENT_SET_WRITEBACK |
1172                              EXTENT_END_WRITEBACK);
1173                 btrfs_free_path(path);
1174                 return PTR_ERR(trans);
1175         }
1176
1177         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1178
1179         cow_start = (u64)-1;
1180         cur_offset = start;
1181         while (1) {
1182                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1183                                                cur_offset, 0);
1184                 if (ret < 0) {
1185                         btrfs_abort_transaction(trans, root, ret);
1186                         goto error;
1187                 }
1188                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1189                         leaf = path->nodes[0];
1190                         btrfs_item_key_to_cpu(leaf, &found_key,
1191                                               path->slots[0] - 1);
1192                         if (found_key.objectid == ino &&
1193                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1194                                 path->slots[0]--;
1195                 }
1196                 check_prev = 0;
1197 next_slot:
1198                 leaf = path->nodes[0];
1199                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1200                         ret = btrfs_next_leaf(root, path);
1201                         if (ret < 0) {
1202                                 btrfs_abort_transaction(trans, root, ret);
1203                                 goto error;
1204                         }
1205                         if (ret > 0)
1206                                 break;
1207                         leaf = path->nodes[0];
1208                 }
1209
1210                 nocow = 0;
1211                 disk_bytenr = 0;
1212                 num_bytes = 0;
1213                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1214
1215                 if (found_key.objectid > ino ||
1216                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1217                     found_key.offset > end)
1218                         break;
1219
1220                 if (found_key.offset > cur_offset) {
1221                         extent_end = found_key.offset;
1222                         extent_type = 0;
1223                         goto out_check;
1224                 }
1225
1226                 fi = btrfs_item_ptr(leaf, path->slots[0],
1227                                     struct btrfs_file_extent_item);
1228                 extent_type = btrfs_file_extent_type(leaf, fi);
1229
1230                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1231                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1232                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1233                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1234                         extent_end = found_key.offset +
1235                                 btrfs_file_extent_num_bytes(leaf, fi);
1236                         if (extent_end <= start) {
1237                                 path->slots[0]++;
1238                                 goto next_slot;
1239                         }
1240                         if (disk_bytenr == 0)
1241                                 goto out_check;
1242                         if (btrfs_file_extent_compression(leaf, fi) ||
1243                             btrfs_file_extent_encryption(leaf, fi) ||
1244                             btrfs_file_extent_other_encoding(leaf, fi))
1245                                 goto out_check;
1246                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1247                                 goto out_check;
1248                         if (btrfs_extent_readonly(root, disk_bytenr))
1249                                 goto out_check;
1250                         if (btrfs_cross_ref_exist(trans, root, ino,
1251                                                   found_key.offset -
1252                                                   extent_offset, disk_bytenr))
1253                                 goto out_check;
1254                         disk_bytenr += extent_offset;
1255                         disk_bytenr += cur_offset - found_key.offset;
1256                         num_bytes = min(end + 1, extent_end) - cur_offset;
1257                         /*
1258                          * force cow if csum exists in the range.
1259                          * this ensure that csum for a given extent are
1260                          * either valid or do not exist.
1261                          */
1262                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1263                                 goto out_check;
1264                         nocow = 1;
1265                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1266                         extent_end = found_key.offset +
1267                                 btrfs_file_extent_inline_len(leaf, fi);
1268                         extent_end = ALIGN(extent_end, root->sectorsize);
1269                 } else {
1270                         BUG_ON(1);
1271                 }
1272 out_check:
1273                 if (extent_end <= start) {
1274                         path->slots[0]++;
1275                         goto next_slot;
1276                 }
1277                 if (!nocow) {
1278                         if (cow_start == (u64)-1)
1279                                 cow_start = cur_offset;
1280                         cur_offset = extent_end;
1281                         if (cur_offset > end)
1282                                 break;
1283                         path->slots[0]++;
1284                         goto next_slot;
1285                 }
1286
1287                 btrfs_release_path(path);
1288                 if (cow_start != (u64)-1) {
1289                         ret = cow_file_range(inode, locked_page, cow_start,
1290                                         found_key.offset - 1, page_started,
1291                                         nr_written, 1);
1292                         if (ret) {
1293                                 btrfs_abort_transaction(trans, root, ret);
1294                                 goto error;
1295                         }
1296                         cow_start = (u64)-1;
1297                 }
1298
1299                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1300                         struct extent_map *em;
1301                         struct extent_map_tree *em_tree;
1302                         em_tree = &BTRFS_I(inode)->extent_tree;
1303                         em = alloc_extent_map();
1304                         BUG_ON(!em); /* -ENOMEM */
1305                         em->start = cur_offset;
1306                         em->orig_start = em->start;
1307                         em->len = num_bytes;
1308                         em->block_len = num_bytes;
1309                         em->block_start = disk_bytenr;
1310                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1311                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1312                         while (1) {
1313                                 write_lock(&em_tree->lock);
1314                                 ret = add_extent_mapping(em_tree, em);
1315                                 write_unlock(&em_tree->lock);
1316                                 if (ret != -EEXIST) {
1317                                         free_extent_map(em);
1318                                         break;
1319                                 }
1320                                 btrfs_drop_extent_cache(inode, em->start,
1321                                                 em->start + em->len - 1, 0);
1322                         }
1323                         type = BTRFS_ORDERED_PREALLOC;
1324                 } else {
1325                         type = BTRFS_ORDERED_NOCOW;
1326                 }
1327
1328                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1329                                                num_bytes, num_bytes, type);
1330                 BUG_ON(ret); /* -ENOMEM */
1331
1332                 if (root->root_key.objectid ==
1333                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1334                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1335                                                       num_bytes);
1336                         if (ret) {
1337                                 btrfs_abort_transaction(trans, root, ret);
1338                                 goto error;
1339                         }
1340                 }
1341
1342                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1343                                 cur_offset, cur_offset + num_bytes - 1,
1344                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1345                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1346                                 EXTENT_SET_PRIVATE2);
1347                 cur_offset = extent_end;
1348                 if (cur_offset > end)
1349                         break;
1350         }
1351         btrfs_release_path(path);
1352
1353         if (cur_offset <= end && cow_start == (u64)-1) {
1354                 cow_start = cur_offset;
1355                 cur_offset = end;
1356         }
1357
1358         if (cow_start != (u64)-1) {
1359                 ret = cow_file_range(inode, locked_page, cow_start, end,
1360                                      page_started, nr_written, 1);
1361                 if (ret) {
1362                         btrfs_abort_transaction(trans, root, ret);
1363                         goto error;
1364                 }
1365         }
1366
1367 error:
1368         if (nolock) {
1369                 err = btrfs_end_transaction_nolock(trans, root);
1370         } else {
1371                 err = btrfs_end_transaction(trans, root);
1372         }
1373         if (!ret)
1374                 ret = err;
1375
1376         if (ret && cur_offset < end)
1377                 extent_clear_unlock_delalloc(inode,
1378                              &BTRFS_I(inode)->io_tree,
1379                              cur_offset, end, locked_page,
1380                              EXTENT_CLEAR_UNLOCK_PAGE |
1381                              EXTENT_CLEAR_UNLOCK |
1382                              EXTENT_CLEAR_DELALLOC |
1383                              EXTENT_CLEAR_DIRTY |
1384                              EXTENT_SET_WRITEBACK |
1385                              EXTENT_END_WRITEBACK);
1386
1387         btrfs_free_path(path);
1388         return ret;
1389 }
1390
1391 /*
1392  * extent_io.c call back to do delayed allocation processing
1393  */
1394 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1395                               u64 start, u64 end, int *page_started,
1396                               unsigned long *nr_written)
1397 {
1398         int ret;
1399         struct btrfs_root *root = BTRFS_I(inode)->root;
1400
1401         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1402                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1403                                          page_started, 1, nr_written);
1404         } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1405                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1406                                          page_started, 0, nr_written);
1407         } else if (!btrfs_test_opt(root, COMPRESS) &&
1408                    !(BTRFS_I(inode)->force_compress) &&
1409                    !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1410                 ret = cow_file_range(inode, locked_page, start, end,
1411                                       page_started, nr_written, 1);
1412         } else {
1413                 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1414                         &BTRFS_I(inode)->runtime_flags);
1415                 ret = cow_file_range_async(inode, locked_page, start, end,
1416                                            page_started, nr_written);
1417         }
1418         return ret;
1419 }
1420
1421 static void btrfs_split_extent_hook(struct inode *inode,
1422                                     struct extent_state *orig, u64 split)
1423 {
1424         /* not delalloc, ignore it */
1425         if (!(orig->state & EXTENT_DELALLOC))
1426                 return;
1427
1428         spin_lock(&BTRFS_I(inode)->lock);
1429         BTRFS_I(inode)->outstanding_extents++;
1430         spin_unlock(&BTRFS_I(inode)->lock);
1431 }
1432
1433 /*
1434  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1435  * extents so we can keep track of new extents that are just merged onto old
1436  * extents, such as when we are doing sequential writes, so we can properly
1437  * account for the metadata space we'll need.
1438  */
1439 static void btrfs_merge_extent_hook(struct inode *inode,
1440                                     struct extent_state *new,
1441                                     struct extent_state *other)
1442 {
1443         /* not delalloc, ignore it */
1444         if (!(other->state & EXTENT_DELALLOC))
1445                 return;
1446
1447         spin_lock(&BTRFS_I(inode)->lock);
1448         BTRFS_I(inode)->outstanding_extents--;
1449         spin_unlock(&BTRFS_I(inode)->lock);
1450 }
1451
1452 /*
1453  * extent_io.c set_bit_hook, used to track delayed allocation
1454  * bytes in this file, and to maintain the list of inodes that
1455  * have pending delalloc work to be done.
1456  */
1457 static void btrfs_set_bit_hook(struct inode *inode,
1458                                struct extent_state *state, int *bits)
1459 {
1460
1461         /*
1462          * set_bit and clear bit hooks normally require _irqsave/restore
1463          * but in this case, we are only testing for the DELALLOC
1464          * bit, which is only set or cleared with irqs on
1465          */
1466         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1467                 struct btrfs_root *root = BTRFS_I(inode)->root;
1468                 u64 len = state->end + 1 - state->start;
1469                 bool do_list = !btrfs_is_free_space_inode(root, inode);
1470
1471                 if (*bits & EXTENT_FIRST_DELALLOC) {
1472                         *bits &= ~EXTENT_FIRST_DELALLOC;
1473                 } else {
1474                         spin_lock(&BTRFS_I(inode)->lock);
1475                         BTRFS_I(inode)->outstanding_extents++;
1476                         spin_unlock(&BTRFS_I(inode)->lock);
1477                 }
1478
1479                 spin_lock(&root->fs_info->delalloc_lock);
1480                 BTRFS_I(inode)->delalloc_bytes += len;
1481                 root->fs_info->delalloc_bytes += len;
1482                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1483                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1484                                       &root->fs_info->delalloc_inodes);
1485                 }
1486                 spin_unlock(&root->fs_info->delalloc_lock);
1487         }
1488 }
1489
1490 /*
1491  * extent_io.c clear_bit_hook, see set_bit_hook for why
1492  */
1493 static void btrfs_clear_bit_hook(struct inode *inode,
1494                                  struct extent_state *state, int *bits)
1495 {
1496         /*
1497          * set_bit and clear bit hooks normally require _irqsave/restore
1498          * but in this case, we are only testing for the DELALLOC
1499          * bit, which is only set or cleared with irqs on
1500          */
1501         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1502                 struct btrfs_root *root = BTRFS_I(inode)->root;
1503                 u64 len = state->end + 1 - state->start;
1504                 bool do_list = !btrfs_is_free_space_inode(root, inode);
1505
1506                 if (*bits & EXTENT_FIRST_DELALLOC) {
1507                         *bits &= ~EXTENT_FIRST_DELALLOC;
1508                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1509                         spin_lock(&BTRFS_I(inode)->lock);
1510                         BTRFS_I(inode)->outstanding_extents--;
1511                         spin_unlock(&BTRFS_I(inode)->lock);
1512                 }
1513
1514                 if (*bits & EXTENT_DO_ACCOUNTING)
1515                         btrfs_delalloc_release_metadata(inode, len);
1516
1517                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1518                     && do_list)
1519                         btrfs_free_reserved_data_space(inode, len);
1520
1521                 spin_lock(&root->fs_info->delalloc_lock);
1522                 root->fs_info->delalloc_bytes -= len;
1523                 BTRFS_I(inode)->delalloc_bytes -= len;
1524
1525                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1526                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1527                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1528                 }
1529                 spin_unlock(&root->fs_info->delalloc_lock);
1530         }
1531 }
1532
1533 /*
1534  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1535  * we don't create bios that span stripes or chunks
1536  */
1537 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1538                          size_t size, struct bio *bio,
1539                          unsigned long bio_flags)
1540 {
1541         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1542         struct btrfs_mapping_tree *map_tree;
1543         u64 logical = (u64)bio->bi_sector << 9;
1544         u64 length = 0;
1545         u64 map_length;
1546         int ret;
1547
1548         if (bio_flags & EXTENT_BIO_COMPRESSED)
1549                 return 0;
1550
1551         length = bio->bi_size;
1552         map_tree = &root->fs_info->mapping_tree;
1553         map_length = length;
1554         ret = btrfs_map_block(map_tree, READ, logical,
1555                               &map_length, NULL, 0);
1556         /* Will always return 0 or 1 with map_multi == NULL */
1557         BUG_ON(ret < 0);
1558         if (map_length < length + size)
1559                 return 1;
1560         return 0;
1561 }
1562
1563 /*
1564  * in order to insert checksums into the metadata in large chunks,
1565  * we wait until bio submission time.   All the pages in the bio are
1566  * checksummed and sums are attached onto the ordered extent record.
1567  *
1568  * At IO completion time the cums attached on the ordered extent record
1569  * are inserted into the btree
1570  */
1571 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1572                                     struct bio *bio, int mirror_num,
1573                                     unsigned long bio_flags,
1574                                     u64 bio_offset)
1575 {
1576         struct btrfs_root *root = BTRFS_I(inode)->root;
1577         int ret = 0;
1578
1579         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1580         BUG_ON(ret); /* -ENOMEM */
1581         return 0;
1582 }
1583
1584 /*
1585  * in order to insert checksums into the metadata in large chunks,
1586  * we wait until bio submission time.   All the pages in the bio are
1587  * checksummed and sums are attached onto the ordered extent record.
1588  *
1589  * At IO completion time the cums attached on the ordered extent record
1590  * are inserted into the btree
1591  */
1592 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1593                           int mirror_num, unsigned long bio_flags,
1594                           u64 bio_offset)
1595 {
1596         struct btrfs_root *root = BTRFS_I(inode)->root;
1597         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1598 }
1599
1600 /*
1601  * extent_io.c submission hook. This does the right thing for csum calculation
1602  * on write, or reading the csums from the tree before a read
1603  */
1604 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1605                           int mirror_num, unsigned long bio_flags,
1606                           u64 bio_offset)
1607 {
1608         struct btrfs_root *root = BTRFS_I(inode)->root;
1609         int ret = 0;
1610         int skip_sum;
1611         int metadata = 0;
1612
1613         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1614
1615         if (btrfs_is_free_space_inode(root, inode))
1616                 metadata = 2;
1617
1618         if (!(rw & REQ_WRITE)) {
1619                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1620                 if (ret)
1621                         return ret;
1622
1623                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1624                         return btrfs_submit_compressed_read(inode, bio,
1625                                                     mirror_num, bio_flags);
1626                 } else if (!skip_sum) {
1627                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1628                         if (ret)
1629                                 return ret;
1630                 }
1631                 goto mapit;
1632         } else if (!skip_sum) {
1633                 /* csum items have already been cloned */
1634                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1635                         goto mapit;
1636                 /* we're doing a write, do the async checksumming */
1637                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1638                                    inode, rw, bio, mirror_num,
1639                                    bio_flags, bio_offset,
1640                                    __btrfs_submit_bio_start,
1641                                    __btrfs_submit_bio_done);
1642         }
1643
1644 mapit:
1645         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1646 }
1647
1648 /*
1649  * given a list of ordered sums record them in the inode.  This happens
1650  * at IO completion time based on sums calculated at bio submission time.
1651  */
1652 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1653                              struct inode *inode, u64 file_offset,
1654                              struct list_head *list)
1655 {
1656         struct btrfs_ordered_sum *sum;
1657
1658         list_for_each_entry(sum, list, list) {
1659                 btrfs_csum_file_blocks(trans,
1660                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1661         }
1662         return 0;
1663 }
1664
1665 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1666                               struct extent_state **cached_state)
1667 {
1668         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1669                 WARN_ON(1);
1670         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1671                                    cached_state, GFP_NOFS);
1672 }
1673
1674 /* see btrfs_writepage_start_hook for details on why this is required */
1675 struct btrfs_writepage_fixup {
1676         struct page *page;
1677         struct btrfs_work work;
1678 };
1679
1680 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1681 {
1682         struct btrfs_writepage_fixup *fixup;
1683         struct btrfs_ordered_extent *ordered;
1684         struct extent_state *cached_state = NULL;
1685         struct page *page;
1686         struct inode *inode;
1687         u64 page_start;
1688         u64 page_end;
1689         int ret;
1690
1691         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1692         page = fixup->page;
1693 again:
1694         lock_page(page);
1695         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1696                 ClearPageChecked(page);
1697                 goto out_page;
1698         }
1699
1700         inode = page->mapping->host;
1701         page_start = page_offset(page);
1702         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1703
1704         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1705                          &cached_state);
1706
1707         /* already ordered? We're done */
1708         if (PagePrivate2(page))
1709                 goto out;
1710
1711         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1712         if (ordered) {
1713                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1714                                      page_end, &cached_state, GFP_NOFS);
1715                 unlock_page(page);
1716                 btrfs_start_ordered_extent(inode, ordered, 1);
1717                 btrfs_put_ordered_extent(ordered);
1718                 goto again;
1719         }
1720
1721         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1722         if (ret) {
1723                 mapping_set_error(page->mapping, ret);
1724                 end_extent_writepage(page, ret, page_start, page_end);
1725                 ClearPageChecked(page);
1726                 goto out;
1727          }
1728
1729         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1730         ClearPageChecked(page);
1731         set_page_dirty(page);
1732 out:
1733         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1734                              &cached_state, GFP_NOFS);
1735 out_page:
1736         unlock_page(page);
1737         page_cache_release(page);
1738         kfree(fixup);
1739 }
1740
1741 /*
1742  * There are a few paths in the higher layers of the kernel that directly
1743  * set the page dirty bit without asking the filesystem if it is a
1744  * good idea.  This causes problems because we want to make sure COW
1745  * properly happens and the data=ordered rules are followed.
1746  *
1747  * In our case any range that doesn't have the ORDERED bit set
1748  * hasn't been properly setup for IO.  We kick off an async process
1749  * to fix it up.  The async helper will wait for ordered extents, set
1750  * the delalloc bit and make it safe to write the page.
1751  */
1752 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1753 {
1754         struct inode *inode = page->mapping->host;
1755         struct btrfs_writepage_fixup *fixup;
1756         struct btrfs_root *root = BTRFS_I(inode)->root;
1757
1758         /* this page is properly in the ordered list */
1759         if (TestClearPagePrivate2(page))
1760                 return 0;
1761
1762         if (PageChecked(page))
1763                 return -EAGAIN;
1764
1765         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1766         if (!fixup)
1767                 return -EAGAIN;
1768
1769         SetPageChecked(page);
1770         page_cache_get(page);
1771         fixup->work.func = btrfs_writepage_fixup_worker;
1772         fixup->page = page;
1773         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1774         return -EBUSY;
1775 }
1776
1777 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1778                                        struct inode *inode, u64 file_pos,
1779                                        u64 disk_bytenr, u64 disk_num_bytes,
1780                                        u64 num_bytes, u64 ram_bytes,
1781                                        u8 compression, u8 encryption,
1782                                        u16 other_encoding, int extent_type)
1783 {
1784         struct btrfs_root *root = BTRFS_I(inode)->root;
1785         struct btrfs_file_extent_item *fi;
1786         struct btrfs_path *path;
1787         struct extent_buffer *leaf;
1788         struct btrfs_key ins;
1789         u64 hint;
1790         int ret;
1791
1792         path = btrfs_alloc_path();
1793         if (!path)
1794                 return -ENOMEM;
1795
1796         path->leave_spinning = 1;
1797
1798         /*
1799          * we may be replacing one extent in the tree with another.
1800          * The new extent is pinned in the extent map, and we don't want
1801          * to drop it from the cache until it is completely in the btree.
1802          *
1803          * So, tell btrfs_drop_extents to leave this extent in the cache.
1804          * the caller is expected to unpin it and allow it to be merged
1805          * with the others.
1806          */
1807         ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1808                                  &hint, 0);
1809         if (ret)
1810                 goto out;
1811
1812         ins.objectid = btrfs_ino(inode);
1813         ins.offset = file_pos;
1814         ins.type = BTRFS_EXTENT_DATA_KEY;
1815         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1816         if (ret)
1817                 goto out;
1818         leaf = path->nodes[0];
1819         fi = btrfs_item_ptr(leaf, path->slots[0],
1820                             struct btrfs_file_extent_item);
1821         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1822         btrfs_set_file_extent_type(leaf, fi, extent_type);
1823         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1824         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1825         btrfs_set_file_extent_offset(leaf, fi, 0);
1826         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1827         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1828         btrfs_set_file_extent_compression(leaf, fi, compression);
1829         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1830         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1831
1832         btrfs_unlock_up_safe(path, 1);
1833         btrfs_set_lock_blocking(leaf);
1834
1835         btrfs_mark_buffer_dirty(leaf);
1836
1837         inode_add_bytes(inode, num_bytes);
1838
1839         ins.objectid = disk_bytenr;
1840         ins.offset = disk_num_bytes;
1841         ins.type = BTRFS_EXTENT_ITEM_KEY;
1842         ret = btrfs_alloc_reserved_file_extent(trans, root,
1843                                         root->root_key.objectid,
1844                                         btrfs_ino(inode), file_pos, &ins);
1845 out:
1846         btrfs_free_path(path);
1847
1848         return ret;
1849 }
1850
1851 /*
1852  * helper function for btrfs_finish_ordered_io, this
1853  * just reads in some of the csum leaves to prime them into ram
1854  * before we start the transaction.  It limits the amount of btree
1855  * reads required while inside the transaction.
1856  */
1857 /* as ordered data IO finishes, this gets called so we can finish
1858  * an ordered extent if the range of bytes in the file it covers are
1859  * fully written.
1860  */
1861 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
1862 {
1863         struct inode *inode = ordered_extent->inode;
1864         struct btrfs_root *root = BTRFS_I(inode)->root;
1865         struct btrfs_trans_handle *trans = NULL;
1866         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1867         struct extent_state *cached_state = NULL;
1868         int compress_type = 0;
1869         int ret;
1870         bool nolock;
1871
1872         nolock = btrfs_is_free_space_inode(root, inode);
1873
1874         if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
1875                 ret = -EIO;
1876                 goto out;
1877         }
1878
1879         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1880                 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
1881                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1882                 if (!ret) {
1883                         if (nolock)
1884                                 trans = btrfs_join_transaction_nolock(root);
1885                         else
1886                                 trans = btrfs_join_transaction(root);
1887                         if (IS_ERR(trans))
1888                                 return PTR_ERR(trans);
1889                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1890                         ret = btrfs_update_inode_fallback(trans, root, inode);
1891                         if (ret) /* -ENOMEM or corruption */
1892                                 btrfs_abort_transaction(trans, root, ret);
1893                 }
1894                 goto out;
1895         }
1896
1897         lock_extent_bits(io_tree, ordered_extent->file_offset,
1898                          ordered_extent->file_offset + ordered_extent->len - 1,
1899                          0, &cached_state);
1900
1901         if (nolock)
1902                 trans = btrfs_join_transaction_nolock(root);
1903         else
1904                 trans = btrfs_join_transaction(root);
1905         if (IS_ERR(trans)) {
1906                 ret = PTR_ERR(trans);
1907                 trans = NULL;
1908                 goto out_unlock;
1909         }
1910         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1911
1912         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1913                 compress_type = ordered_extent->compress_type;
1914         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1915                 BUG_ON(compress_type);
1916                 ret = btrfs_mark_extent_written(trans, inode,
1917                                                 ordered_extent->file_offset,
1918                                                 ordered_extent->file_offset +
1919                                                 ordered_extent->len);
1920         } else {
1921                 BUG_ON(root == root->fs_info->tree_root);
1922                 ret = insert_reserved_file_extent(trans, inode,
1923                                                 ordered_extent->file_offset,
1924                                                 ordered_extent->start,
1925                                                 ordered_extent->disk_len,
1926                                                 ordered_extent->len,
1927                                                 ordered_extent->len,
1928                                                 compress_type, 0, 0,
1929                                                 BTRFS_FILE_EXTENT_REG);
1930                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1931                                    ordered_extent->file_offset,
1932                                    ordered_extent->len);
1933         }
1934
1935         if (ret < 0) {
1936                 btrfs_abort_transaction(trans, root, ret);
1937                 goto out_unlock;
1938         }
1939
1940         add_pending_csums(trans, inode, ordered_extent->file_offset,
1941                           &ordered_extent->list);
1942
1943         ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1944         if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1945                 ret = btrfs_update_inode_fallback(trans, root, inode);
1946                 if (ret) { /* -ENOMEM or corruption */
1947                         btrfs_abort_transaction(trans, root, ret);
1948                         goto out_unlock;
1949                 }
1950         }
1951         ret = 0;
1952 out_unlock:
1953         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1954                              ordered_extent->file_offset +
1955                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1956 out:
1957         if (root != root->fs_info->tree_root)
1958                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1959         if (trans) {
1960                 if (nolock)
1961                         btrfs_end_transaction_nolock(trans, root);
1962                 else
1963                         btrfs_end_transaction(trans, root);
1964         }
1965
1966         if (ret)
1967                 clear_extent_uptodate(io_tree, ordered_extent->file_offset,
1968                                       ordered_extent->file_offset +
1969                                       ordered_extent->len - 1, NULL, GFP_NOFS);
1970
1971         /*
1972          * This needs to be done to make sure anybody waiting knows we are done
1973          * updating everything for this ordered extent.
1974          */
1975         btrfs_remove_ordered_extent(inode, ordered_extent);
1976
1977         /* once for us */
1978         btrfs_put_ordered_extent(ordered_extent);
1979         /* once for the tree */
1980         btrfs_put_ordered_extent(ordered_extent);
1981
1982         return ret;
1983 }
1984
1985 static void finish_ordered_fn(struct btrfs_work *work)
1986 {
1987         struct btrfs_ordered_extent *ordered_extent;
1988         ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
1989         btrfs_finish_ordered_io(ordered_extent);
1990 }
1991
1992 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1993                                 struct extent_state *state, int uptodate)
1994 {
1995         struct inode *inode = page->mapping->host;
1996         struct btrfs_root *root = BTRFS_I(inode)->root;
1997         struct btrfs_ordered_extent *ordered_extent = NULL;
1998         struct btrfs_workers *workers;
1999
2000         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2001
2002         ClearPagePrivate2(page);
2003         if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2004                                             end - start + 1, uptodate))
2005                 return 0;
2006
2007         ordered_extent->work.func = finish_ordered_fn;
2008         ordered_extent->work.flags = 0;
2009
2010         if (btrfs_is_free_space_inode(root, inode))
2011                 workers = &root->fs_info->endio_freespace_worker;
2012         else
2013                 workers = &root->fs_info->endio_write_workers;
2014         btrfs_queue_worker(workers, &ordered_extent->work);
2015
2016         return 0;
2017 }
2018
2019 /*
2020  * when reads are done, we need to check csums to verify the data is correct
2021  * if there's a match, we allow the bio to finish.  If not, the code in
2022  * extent_io.c will try to find good copies for us.
2023  */
2024 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
2025                                struct extent_state *state, int mirror)
2026 {
2027         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
2028         struct inode *inode = page->mapping->host;
2029         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2030         char *kaddr;
2031         u64 private = ~(u32)0;
2032         int ret;
2033         struct btrfs_root *root = BTRFS_I(inode)->root;
2034         u32 csum = ~(u32)0;
2035
2036         if (PageChecked(page)) {
2037                 ClearPageChecked(page);
2038                 goto good;
2039         }
2040
2041         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2042                 goto good;
2043
2044         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2045             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2046                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2047                                   GFP_NOFS);
2048                 return 0;
2049         }
2050
2051         if (state && state->start == start) {
2052                 private = state->private;
2053                 ret = 0;
2054         } else {
2055                 ret = get_state_private(io_tree, start, &private);
2056         }
2057         kaddr = kmap_atomic(page);
2058         if (ret)
2059                 goto zeroit;
2060
2061         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
2062         btrfs_csum_final(csum, (char *)&csum);
2063         if (csum != private)
2064                 goto zeroit;
2065
2066         kunmap_atomic(kaddr);
2067 good:
2068         return 0;
2069
2070 zeroit:
2071         printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
2072                        "private %llu\n",
2073                        (unsigned long long)btrfs_ino(page->mapping->host),
2074                        (unsigned long long)start, csum,
2075                        (unsigned long long)private);
2076         memset(kaddr + offset, 1, end - start + 1);
2077         flush_dcache_page(page);
2078         kunmap_atomic(kaddr);
2079         if (private == 0)
2080                 return 0;
2081         return -EIO;
2082 }
2083
2084 struct delayed_iput {
2085         struct list_head list;
2086         struct inode *inode;
2087 };
2088
2089 /* JDM: If this is fs-wide, why can't we add a pointer to
2090  * btrfs_inode instead and avoid the allocation? */
2091 void btrfs_add_delayed_iput(struct inode *inode)
2092 {
2093         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2094         struct delayed_iput *delayed;
2095
2096         if (atomic_add_unless(&inode->i_count, -1, 1))
2097                 return;
2098
2099         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2100         delayed->inode = inode;
2101
2102         spin_lock(&fs_info->delayed_iput_lock);
2103         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2104         spin_unlock(&fs_info->delayed_iput_lock);
2105 }
2106
2107 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2108 {
2109         LIST_HEAD(list);
2110         struct btrfs_fs_info *fs_info = root->fs_info;
2111         struct delayed_iput *delayed;
2112         int empty;
2113
2114         spin_lock(&fs_info->delayed_iput_lock);
2115         empty = list_empty(&fs_info->delayed_iputs);
2116         spin_unlock(&fs_info->delayed_iput_lock);
2117         if (empty)
2118                 return;
2119
2120         down_read(&root->fs_info->cleanup_work_sem);
2121         spin_lock(&fs_info->delayed_iput_lock);
2122         list_splice_init(&fs_info->delayed_iputs, &list);
2123         spin_unlock(&fs_info->delayed_iput_lock);
2124
2125         while (!list_empty(&list)) {
2126                 delayed = list_entry(list.next, struct delayed_iput, list);
2127                 list_del(&delayed->list);
2128                 iput(delayed->inode);
2129                 kfree(delayed);
2130         }
2131         up_read(&root->fs_info->cleanup_work_sem);
2132 }
2133
2134 enum btrfs_orphan_cleanup_state {
2135         ORPHAN_CLEANUP_STARTED  = 1,
2136         ORPHAN_CLEANUP_DONE     = 2,
2137 };
2138
2139 /*
2140  * This is called in transaction commit time. If there are no orphan
2141  * files in the subvolume, it removes orphan item and frees block_rsv
2142  * structure.
2143  */
2144 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2145                               struct btrfs_root *root)
2146 {
2147         struct btrfs_block_rsv *block_rsv;
2148         int ret;
2149
2150         if (atomic_read(&root->orphan_inodes) ||
2151             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2152                 return;
2153
2154         spin_lock(&root->orphan_lock);
2155         if (atomic_read(&root->orphan_inodes)) {
2156                 spin_unlock(&root->orphan_lock);
2157                 return;
2158         }
2159
2160         if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2161                 spin_unlock(&root->orphan_lock);
2162                 return;
2163         }
2164
2165         block_rsv = root->orphan_block_rsv;
2166         root->orphan_block_rsv = NULL;
2167         spin_unlock(&root->orphan_lock);
2168
2169         if (root->orphan_item_inserted &&
2170             btrfs_root_refs(&root->root_item) > 0) {
2171                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2172                                             root->root_key.objectid);
2173                 BUG_ON(ret);
2174                 root->orphan_item_inserted = 0;
2175         }
2176
2177         if (block_rsv) {
2178                 WARN_ON(block_rsv->size > 0);
2179                 btrfs_free_block_rsv(root, block_rsv);
2180         }
2181 }
2182
2183 /*
2184  * This creates an orphan entry for the given inode in case something goes
2185  * wrong in the middle of an unlink/truncate.
2186  *
2187  * NOTE: caller of this function should reserve 5 units of metadata for
2188  *       this function.
2189  */
2190 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2191 {
2192         struct btrfs_root *root = BTRFS_I(inode)->root;
2193         struct btrfs_block_rsv *block_rsv = NULL;
2194         int reserve = 0;
2195         int insert = 0;
2196         int ret;
2197
2198         if (!root->orphan_block_rsv) {
2199                 block_rsv = btrfs_alloc_block_rsv(root);
2200                 if (!block_rsv)
2201                         return -ENOMEM;
2202         }
2203
2204         spin_lock(&root->orphan_lock);
2205         if (!root->orphan_block_rsv) {
2206                 root->orphan_block_rsv = block_rsv;
2207         } else if (block_rsv) {
2208                 btrfs_free_block_rsv(root, block_rsv);
2209                 block_rsv = NULL;
2210         }
2211
2212         if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2213                               &BTRFS_I(inode)->runtime_flags)) {
2214 #if 0
2215                 /*
2216                  * For proper ENOSPC handling, we should do orphan
2217                  * cleanup when mounting. But this introduces backward
2218                  * compatibility issue.
2219                  */
2220                 if (!xchg(&root->orphan_item_inserted, 1))
2221                         insert = 2;
2222                 else
2223                         insert = 1;
2224 #endif
2225                 insert = 1;
2226                 atomic_dec(&root->orphan_inodes);
2227         }
2228
2229         if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2230                               &BTRFS_I(inode)->runtime_flags))
2231                 reserve = 1;
2232         spin_unlock(&root->orphan_lock);
2233
2234         /* grab metadata reservation from transaction handle */
2235         if (reserve) {
2236                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2237                 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
2238         }
2239
2240         /* insert an orphan item to track this unlinked/truncated file */
2241         if (insert >= 1) {
2242                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2243                 if (ret && ret != -EEXIST) {
2244                         clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2245                                   &BTRFS_I(inode)->runtime_flags);
2246                         btrfs_abort_transaction(trans, root, ret);
2247                         return ret;
2248                 }
2249                 ret = 0;
2250         }
2251
2252         /* insert an orphan item to track subvolume contains orphan files */
2253         if (insert >= 2) {
2254                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2255                                                root->root_key.objectid);
2256                 if (ret && ret != -EEXIST) {
2257                         btrfs_abort_transaction(trans, root, ret);
2258                         return ret;
2259                 }
2260         }
2261         return 0;
2262 }
2263
2264 /*
2265  * We have done the truncate/delete so we can go ahead and remove the orphan
2266  * item for this particular inode.
2267  */
2268 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2269 {
2270         struct btrfs_root *root = BTRFS_I(inode)->root;
2271         int delete_item = 0;
2272         int release_rsv = 0;
2273         int ret = 0;
2274
2275         spin_lock(&root->orphan_lock);
2276         if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2277                                &BTRFS_I(inode)->runtime_flags))
2278                 delete_item = 1;
2279
2280         if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2281                                &BTRFS_I(inode)->runtime_flags))
2282                 release_rsv = 1;
2283         spin_unlock(&root->orphan_lock);
2284
2285         if (trans && delete_item) {
2286                 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2287                 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2288         }
2289
2290         if (release_rsv) {
2291                 btrfs_orphan_release_metadata(inode);
2292                 atomic_dec(&root->orphan_inodes);
2293         }
2294
2295         return 0;
2296 }
2297
2298 /*
2299  * this cleans up any orphans that may be left on the list from the last use
2300  * of this root.
2301  */
2302 int btrfs_orphan_cleanup(struct btrfs_root *root)
2303 {
2304         struct btrfs_path *path;
2305         struct extent_buffer *leaf;
2306         struct btrfs_key key, found_key;
2307         struct btrfs_trans_handle *trans;
2308         struct inode *inode;
2309         u64 last_objectid = 0;
2310         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2311
2312         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2313                 return 0;
2314
2315         path = btrfs_alloc_path();
2316         if (!path) {
2317                 ret = -ENOMEM;
2318                 goto out;
2319         }
2320         path->reada = -1;
2321
2322         key.objectid = BTRFS_ORPHAN_OBJECTID;
2323         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2324         key.offset = (u64)-1;
2325
2326         while (1) {
2327                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2328                 if (ret < 0)
2329                         goto out;
2330
2331                 /*
2332                  * if ret == 0 means we found what we were searching for, which
2333                  * is weird, but possible, so only screw with path if we didn't
2334                  * find the key and see if we have stuff that matches
2335                  */
2336                 if (ret > 0) {
2337                         ret = 0;
2338                         if (path->slots[0] == 0)
2339                                 break;
2340                         path->slots[0]--;
2341                 }
2342
2343                 /* pull out the item */
2344                 leaf = path->nodes[0];
2345                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2346
2347                 /* make sure the item matches what we want */
2348                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2349                         break;
2350                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2351                         break;
2352
2353                 /* release the path since we're done with it */
2354                 btrfs_release_path(path);
2355
2356                 /*
2357                  * this is where we are basically btrfs_lookup, without the
2358                  * crossing root thing.  we store the inode number in the
2359                  * offset of the orphan item.
2360                  */
2361
2362                 if (found_key.offset == last_objectid) {
2363                         printk(KERN_ERR "btrfs: Error removing orphan entry, "
2364                                "stopping orphan cleanup\n");
2365                         ret = -EINVAL;
2366                         goto out;
2367                 }
2368
2369                 last_objectid = found_key.offset;
2370
2371                 found_key.objectid = found_key.offset;
2372                 found_key.type = BTRFS_INODE_ITEM_KEY;
2373                 found_key.offset = 0;
2374                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2375                 ret = PTR_RET(inode);
2376                 if (ret && ret != -ESTALE)
2377                         goto out;
2378
2379                 if (ret == -ESTALE && root == root->fs_info->tree_root) {
2380                         struct btrfs_root *dead_root;
2381                         struct btrfs_fs_info *fs_info = root->fs_info;
2382                         int is_dead_root = 0;
2383
2384                         /*
2385                          * this is an orphan in the tree root. Currently these
2386                          * could come from 2 sources:
2387                          *  a) a snapshot deletion in progress
2388                          *  b) a free space cache inode
2389                          * We need to distinguish those two, as the snapshot
2390                          * orphan must not get deleted.
2391                          * find_dead_roots already ran before us, so if this
2392                          * is a snapshot deletion, we should find the root
2393                          * in the dead_roots list
2394                          */
2395                         spin_lock(&fs_info->trans_lock);
2396                         list_for_each_entry(dead_root, &fs_info->dead_roots,
2397                                             root_list) {
2398                                 if (dead_root->root_key.objectid ==
2399                                     found_key.objectid) {
2400                                         is_dead_root = 1;
2401                                         break;
2402                                 }
2403                         }
2404                         spin_unlock(&fs_info->trans_lock);
2405                         if (is_dead_root) {
2406                                 /* prevent this orphan from being found again */
2407                                 key.offset = found_key.objectid - 1;
2408                                 continue;
2409                         }
2410                 }
2411                 /*
2412                  * Inode is already gone but the orphan item is still there,
2413                  * kill the orphan item.
2414                  */
2415                 if (ret == -ESTALE) {
2416                         trans = btrfs_start_transaction(root, 1);
2417                         if (IS_ERR(trans)) {
2418                                 ret = PTR_ERR(trans);
2419                                 goto out;
2420                         }
2421                         printk(KERN_ERR "auto deleting %Lu\n",
2422                                found_key.objectid);
2423                         ret = btrfs_del_orphan_item(trans, root,
2424                                                     found_key.objectid);
2425                         BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2426                         btrfs_end_transaction(trans, root);
2427                         continue;
2428                 }
2429
2430                 /*
2431                  * add this inode to the orphan list so btrfs_orphan_del does
2432                  * the proper thing when we hit it
2433                  */
2434                 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2435                         &BTRFS_I(inode)->runtime_flags);
2436
2437                 /* if we have links, this was a truncate, lets do that */
2438                 if (inode->i_nlink) {
2439                         if (!S_ISREG(inode->i_mode)) {
2440                                 WARN_ON(1);
2441                                 iput(inode);
2442                                 continue;
2443                         }
2444                         nr_truncate++;
2445                         ret = btrfs_truncate(inode);
2446                 } else {
2447                         nr_unlink++;
2448                 }
2449
2450                 /* this will do delete_inode and everything for us */
2451                 iput(inode);
2452                 if (ret)
2453                         goto out;
2454         }
2455         /* release the path since we're done with it */
2456         btrfs_release_path(path);
2457
2458         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2459
2460         if (root->orphan_block_rsv)
2461                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2462                                         (u64)-1);
2463
2464         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2465                 trans = btrfs_join_transaction(root);
2466                 if (!IS_ERR(trans))
2467                         btrfs_end_transaction(trans, root);
2468         }
2469
2470         if (nr_unlink)
2471                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2472         if (nr_truncate)
2473                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2474
2475 out:
2476         if (ret)
2477                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2478         btrfs_free_path(path);
2479         return ret;
2480 }
2481
2482 /*
2483  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2484  * don't find any xattrs, we know there can't be any acls.
2485  *
2486  * slot is the slot the inode is in, objectid is the objectid of the inode
2487  */
2488 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2489                                           int slot, u64 objectid)
2490 {
2491         u32 nritems = btrfs_header_nritems(leaf);
2492         struct btrfs_key found_key;
2493         int scanned = 0;
2494
2495         slot++;
2496         while (slot < nritems) {
2497                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2498
2499                 /* we found a different objectid, there must not be acls */
2500                 if (found_key.objectid != objectid)
2501                         return 0;
2502
2503                 /* we found an xattr, assume we've got an acl */
2504                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2505                         return 1;
2506
2507                 /*
2508                  * we found a key greater than an xattr key, there can't
2509                  * be any acls later on
2510                  */
2511                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2512                         return 0;
2513
2514                 slot++;
2515                 scanned++;
2516
2517                 /*
2518                  * it goes inode, inode backrefs, xattrs, extents,
2519                  * so if there are a ton of hard links to an inode there can
2520                  * be a lot of backrefs.  Don't waste time searching too hard,
2521                  * this is just an optimization
2522                  */
2523                 if (scanned >= 8)
2524                         break;
2525         }
2526         /* we hit the end of the leaf before we found an xattr or
2527          * something larger than an xattr.  We have to assume the inode
2528          * has acls
2529          */
2530         return 1;
2531 }
2532
2533 /*
2534  * read an inode from the btree into the in-memory inode
2535  */
2536 static void btrfs_read_locked_inode(struct inode *inode)
2537 {
2538         struct btrfs_path *path;
2539         struct extent_buffer *leaf;
2540         struct btrfs_inode_item *inode_item;
2541         struct btrfs_timespec *tspec;
2542         struct btrfs_root *root = BTRFS_I(inode)->root;
2543         struct btrfs_key location;
2544         int maybe_acls;
2545         u32 rdev;
2546         int ret;
2547         bool filled = false;
2548
2549         ret = btrfs_fill_inode(inode, &rdev);
2550         if (!ret)
2551                 filled = true;
2552
2553         path = btrfs_alloc_path();
2554         if (!path)
2555                 goto make_bad;
2556
2557         path->leave_spinning = 1;
2558         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2559
2560         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2561         if (ret)
2562                 goto make_bad;
2563
2564         leaf = path->nodes[0];
2565
2566         if (filled)
2567                 goto cache_acl;
2568
2569         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2570                                     struct btrfs_inode_item);
2571         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2572         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2573         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2574         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2575         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2576
2577         tspec = btrfs_inode_atime(inode_item);
2578         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2579         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2580
2581         tspec = btrfs_inode_mtime(inode_item);
2582         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2583         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2584
2585         tspec = btrfs_inode_ctime(inode_item);
2586         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2587         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2588
2589         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2590         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2591         inode->i_version = btrfs_inode_sequence(leaf, inode_item);
2592         inode->i_generation = BTRFS_I(inode)->generation;
2593         inode->i_rdev = 0;
2594         rdev = btrfs_inode_rdev(leaf, inode_item);
2595
2596         BTRFS_I(inode)->index_cnt = (u64)-1;
2597         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2598 cache_acl:
2599         /*
2600          * try to precache a NULL acl entry for files that don't have
2601          * any xattrs or acls
2602          */
2603         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2604                                            btrfs_ino(inode));
2605         if (!maybe_acls)
2606                 cache_no_acl(inode);
2607
2608         btrfs_free_path(path);
2609
2610         switch (inode->i_mode & S_IFMT) {
2611         case S_IFREG:
2612                 inode->i_mapping->a_ops = &btrfs_aops;
2613                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2614                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2615                 inode->i_fop = &btrfs_file_operations;
2616                 inode->i_op = &btrfs_file_inode_operations;
2617                 break;
2618         case S_IFDIR:
2619                 inode->i_fop = &btrfs_dir_file_operations;
2620                 if (root == root->fs_info->tree_root)
2621                         inode->i_op = &btrfs_dir_ro_inode_operations;
2622                 else
2623                         inode->i_op = &btrfs_dir_inode_operations;
2624                 break;
2625         case S_IFLNK:
2626                 inode->i_op = &btrfs_symlink_inode_operations;
2627                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2628                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2629                 break;
2630         default:
2631                 inode->i_op = &btrfs_special_inode_operations;
2632                 init_special_inode(inode, inode->i_mode, rdev);
2633                 break;
2634         }
2635
2636         btrfs_update_iflags(inode);
2637         return;
2638
2639 make_bad:
2640         btrfs_free_path(path);
2641         make_bad_inode(inode);
2642 }
2643
2644 /*
2645  * given a leaf and an inode, copy the inode fields into the leaf
2646  */
2647 static void fill_inode_item(struct btrfs_trans_handle *trans,
2648                             struct extent_buffer *leaf,
2649                             struct btrfs_inode_item *item,
2650                             struct inode *inode)
2651 {
2652         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2653         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2654         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2655         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2656         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2657
2658         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2659                                inode->i_atime.tv_sec);
2660         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2661                                 inode->i_atime.tv_nsec);
2662
2663         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2664                                inode->i_mtime.tv_sec);
2665         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2666                                 inode->i_mtime.tv_nsec);
2667
2668         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2669                                inode->i_ctime.tv_sec);
2670         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2671                                 inode->i_ctime.tv_nsec);
2672
2673         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2674         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2675         btrfs_set_inode_sequence(leaf, item, inode->i_version);
2676         btrfs_set_inode_transid(leaf, item, trans->transid);
2677         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2678         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2679         btrfs_set_inode_block_group(leaf, item, 0);
2680 }
2681
2682 /*
2683  * copy everything in the in-memory inode into the btree.
2684  */
2685 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
2686                                 struct btrfs_root *root, struct inode *inode)
2687 {
2688         struct btrfs_inode_item *inode_item;
2689         struct btrfs_path *path;
2690         struct extent_buffer *leaf;
2691         int ret;
2692
2693         path = btrfs_alloc_path();
2694         if (!path)
2695                 return -ENOMEM;
2696
2697         path->leave_spinning = 1;
2698         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2699                                  1);
2700         if (ret) {
2701                 if (ret > 0)
2702                         ret = -ENOENT;
2703                 goto failed;
2704         }
2705
2706         btrfs_unlock_up_safe(path, 1);
2707         leaf = path->nodes[0];
2708         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2709                                     struct btrfs_inode_item);
2710
2711         fill_inode_item(trans, leaf, inode_item, inode);
2712         btrfs_mark_buffer_dirty(leaf);
2713         btrfs_set_inode_last_trans(trans, inode);
2714         ret = 0;
2715 failed:
2716         btrfs_free_path(path);
2717         return ret;
2718 }
2719
2720 /*
2721  * copy everything in the in-memory inode into the btree.
2722  */
2723 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2724                                 struct btrfs_root *root, struct inode *inode)
2725 {
2726         int ret;
2727
2728         /*
2729          * If the inode is a free space inode, we can deadlock during commit
2730          * if we put it into the delayed code.
2731          *
2732          * The data relocation inode should also be directly updated
2733          * without delay
2734          */
2735         if (!btrfs_is_free_space_inode(root, inode)
2736             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2737                 ret = btrfs_delayed_update_inode(trans, root, inode);
2738                 if (!ret)
2739                         btrfs_set_inode_last_trans(trans, inode);
2740                 return ret;
2741         }
2742
2743         return btrfs_update_inode_item(trans, root, inode);
2744 }
2745
2746 static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2747                                 struct btrfs_root *root, struct inode *inode)
2748 {
2749         int ret;
2750
2751         ret = btrfs_update_inode(trans, root, inode);
2752         if (ret == -ENOSPC)
2753                 return btrfs_update_inode_item(trans, root, inode);
2754         return ret;
2755 }
2756
2757 /*
2758  * unlink helper that gets used here in inode.c and in the tree logging
2759  * recovery code.  It remove a link in a directory with a given name, and
2760  * also drops the back refs in the inode to the directory
2761  */
2762 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2763                                 struct btrfs_root *root,
2764                                 struct inode *dir, struct inode *inode,
2765                                 const char *name, int name_len)
2766 {
2767         struct btrfs_path *path;
2768         int ret = 0;
2769         struct extent_buffer *leaf;
2770         struct btrfs_dir_item *di;
2771         struct btrfs_key key;
2772         u64 index;
2773         u64 ino = btrfs_ino(inode);
2774         u64 dir_ino = btrfs_ino(dir);
2775
2776         path = btrfs_alloc_path();
2777         if (!path) {
2778                 ret = -ENOMEM;
2779                 goto out;
2780         }
2781
2782         path->leave_spinning = 1;
2783         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2784                                     name, name_len, -1);
2785         if (IS_ERR(di)) {
2786                 ret = PTR_ERR(di);
2787                 goto err;
2788         }
2789         if (!di) {
2790                 ret = -ENOENT;
2791                 goto err;
2792         }
2793         leaf = path->nodes[0];
2794         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2795         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2796         if (ret)
2797                 goto err;
2798         btrfs_release_path(path);
2799
2800         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2801                                   dir_ino, &index);
2802         if (ret) {
2803                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2804                        "inode %llu parent %llu\n", name_len, name,
2805                        (unsigned long long)ino, (unsigned long long)dir_ino);
2806                 btrfs_abort_transaction(trans, root, ret);
2807                 goto err;
2808         }
2809
2810         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2811         if (ret) {
2812                 btrfs_abort_transaction(trans, root, ret);
2813                 goto err;
2814         }
2815
2816         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2817                                          inode, dir_ino);
2818         if (ret != 0 && ret != -ENOENT) {
2819                 btrfs_abort_transaction(trans, root, ret);
2820                 goto err;
2821         }
2822
2823         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2824                                            dir, index);
2825         if (ret == -ENOENT)
2826                 ret = 0;
2827 err:
2828         btrfs_free_path(path);
2829         if (ret)
2830                 goto out;
2831
2832         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2833         inode_inc_iversion(inode);
2834         inode_inc_iversion(dir);
2835         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2836         btrfs_update_inode(trans, root, dir);
2837 out:
2838         return ret;
2839 }
2840
2841 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2842                        struct btrfs_root *root,
2843                        struct inode *dir, struct inode *inode,
2844                        const char *name, int name_len)
2845 {
2846         int ret;
2847         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2848         if (!ret) {
2849                 btrfs_drop_nlink(inode);
2850                 ret = btrfs_update_inode(trans, root, inode);
2851         }
2852         return ret;
2853 }
2854
2855
2856 /* helper to check if there is any shared block in the path */
2857 static int check_path_shared(struct btrfs_root *root,
2858                              struct btrfs_path *path)
2859 {
2860         struct extent_buffer *eb;
2861         int level;
2862         u64 refs = 1;
2863
2864         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2865                 int ret;
2866
2867                 if (!path->nodes[level])
2868                         break;
2869                 eb = path->nodes[level];
2870                 if (!btrfs_block_can_be_shared(root, eb))
2871                         continue;
2872                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2873                                                &refs, NULL);
2874                 if (refs > 1)
2875                         return 1;
2876         }
2877         return 0;
2878 }
2879
2880 /*
2881  * helper to start transaction for unlink and rmdir.
2882  *
2883  * unlink and rmdir are special in btrfs, they do not always free space.
2884  * so in enospc case, we should make sure they will free space before
2885  * allowing them to use the global metadata reservation.
2886  */
2887 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2888                                                        struct dentry *dentry)
2889 {
2890         struct btrfs_trans_handle *trans;
2891         struct btrfs_root *root = BTRFS_I(dir)->root;
2892         struct btrfs_path *path;
2893         struct btrfs_inode_ref *ref;
2894         struct btrfs_dir_item *di;
2895         struct inode *inode = dentry->d_inode;
2896         u64 index;
2897         int check_link = 1;
2898         int err = -ENOSPC;
2899         int ret;
2900         u64 ino = btrfs_ino(inode);
2901         u64 dir_ino = btrfs_ino(dir);
2902
2903         /*
2904          * 1 for the possible orphan item
2905          * 1 for the dir item
2906          * 1 for the dir index
2907          * 1 for the inode ref
2908          * 1 for the inode ref in the tree log
2909          * 2 for the dir entries in the log
2910          * 1 for the inode
2911          */
2912         trans = btrfs_start_transaction(root, 8);
2913         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2914                 return trans;
2915
2916         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2917                 return ERR_PTR(-ENOSPC);
2918
2919         /* check if there is someone else holds reference */
2920         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2921                 return ERR_PTR(-ENOSPC);
2922
2923         if (atomic_read(&inode->i_count) > 2)
2924                 return ERR_PTR(-ENOSPC);
2925
2926         if (xchg(&root->fs_info->enospc_unlink, 1))
2927                 return ERR_PTR(-ENOSPC);
2928
2929         path = btrfs_alloc_path();
2930         if (!path) {
2931                 root->fs_info->enospc_unlink = 0;
2932                 return ERR_PTR(-ENOMEM);
2933         }
2934
2935         /* 1 for the orphan item */
2936         trans = btrfs_start_transaction(root, 1);
2937         if (IS_ERR(trans)) {
2938                 btrfs_free_path(path);
2939                 root->fs_info->enospc_unlink = 0;
2940                 return trans;
2941         }
2942
2943         path->skip_locking = 1;
2944         path->search_commit_root = 1;
2945
2946         ret = btrfs_lookup_inode(trans, root, path,
2947                                 &BTRFS_I(dir)->location, 0);
2948         if (ret < 0) {
2949                 err = ret;
2950                 goto out;
2951         }
2952         if (ret == 0) {
2953                 if (check_path_shared(root, path))
2954                         goto out;
2955         } else {
2956                 check_link = 0;
2957         }
2958         btrfs_release_path(path);
2959
2960         ret = btrfs_lookup_inode(trans, root, path,
2961                                 &BTRFS_I(inode)->location, 0);
2962         if (ret < 0) {
2963                 err = ret;
2964                 goto out;
2965         }
2966         if (ret == 0) {
2967                 if (check_path_shared(root, path))
2968                         goto out;
2969         } else {
2970                 check_link = 0;
2971         }
2972         btrfs_release_path(path);
2973
2974         if (ret == 0 && S_ISREG(inode->i_mode)) {
2975                 ret = btrfs_lookup_file_extent(trans, root, path,
2976                                                ino, (u64)-1, 0);
2977                 if (ret < 0) {
2978                         err = ret;
2979                         goto out;
2980                 }
2981                 BUG_ON(ret == 0); /* Corruption */
2982                 if (check_path_shared(root, path))
2983                         goto out;
2984                 btrfs_release_path(path);
2985         }
2986
2987         if (!check_link) {
2988                 err = 0;
2989                 goto out;
2990         }
2991
2992         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2993                                 dentry->d_name.name, dentry->d_name.len, 0);
2994         if (IS_ERR(di)) {
2995                 err = PTR_ERR(di);
2996                 goto out;
2997         }
2998         if (di) {
2999                 if (check_path_shared(root, path))
3000                         goto out;
3001         } else {
3002                 err = 0;
3003                 goto out;
3004         }
3005         btrfs_release_path(path);
3006
3007         ref = btrfs_lookup_inode_ref(trans, root, path,
3008                                 dentry->d_name.name, dentry->d_name.len,
3009                                 ino, dir_ino, 0);
3010         if (IS_ERR(ref)) {
3011                 err = PTR_ERR(ref);
3012                 goto out;
3013         }
3014         BUG_ON(!ref); /* Logic error */
3015         if (check_path_shared(root, path))
3016                 goto out;
3017         index = btrfs_inode_ref_index(path->nodes[0], ref);
3018         btrfs_release_path(path);
3019
3020         /*
3021          * This is a commit root search, if we can lookup inode item and other
3022          * relative items in the commit root, it means the transaction of
3023          * dir/file creation has been committed, and the dir index item that we
3024          * delay to insert has also been inserted into the commit root. So
3025          * we needn't worry about the delayed insertion of the dir index item
3026          * here.
3027          */
3028         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
3029                                 dentry->d_name.name, dentry->d_name.len, 0);
3030         if (IS_ERR(di)) {
3031                 err = PTR_ERR(di);
3032                 goto out;
3033         }
3034         BUG_ON(ret == -ENOENT);
3035         if (check_path_shared(root, path))
3036                 goto out;
3037
3038         err = 0;
3039 out:
3040         btrfs_free_path(path);
3041         /* Migrate the orphan reservation over */
3042         if (!err)
3043                 err = btrfs_block_rsv_migrate(trans->block_rsv,
3044                                 &root->fs_info->global_block_rsv,
3045                                 trans->bytes_reserved);
3046
3047         if (err) {
3048                 btrfs_end_transaction(trans, root);
3049                 root->fs_info->enospc_unlink = 0;
3050                 return ERR_PTR(err);
3051         }
3052
3053         trans->block_rsv = &root->fs_info->global_block_rsv;
3054         return trans;
3055 }
3056
3057 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
3058                                struct btrfs_root *root)
3059 {
3060         if (trans->block_rsv == &root->fs_info->global_block_rsv) {
3061                 btrfs_block_rsv_release(root, trans->block_rsv,
3062                                         trans->bytes_reserved);
3063                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3064                 BUG_ON(!root->fs_info->enospc_unlink);
3065                 root->fs_info->enospc_unlink = 0;
3066         }
3067         btrfs_end_transaction(trans, root);
3068 }
3069
3070 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3071 {
3072         struct btrfs_root *root = BTRFS_I(dir)->root;
3073         struct btrfs_trans_handle *trans;
3074         struct inode *inode = dentry->d_inode;
3075         int ret;
3076         unsigned long nr = 0;
3077
3078         trans = __unlink_start_trans(dir, dentry);
3079         if (IS_ERR(trans))
3080                 return PTR_ERR(trans);
3081
3082         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3083
3084         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3085                                  dentry->d_name.name, dentry->d_name.len);
3086         if (ret)
3087                 goto out;
3088
3089         if (inode->i_nlink == 0) {
3090                 ret = btrfs_orphan_add(trans, inode);
3091                 if (ret)
3092                         goto out;
3093         }
3094
3095 out:
3096         nr = trans->blocks_used;
3097         __unlink_end_trans(trans, root);
3098         btrfs_btree_balance_dirty(root, nr);
3099         return ret;
3100 }
3101
3102 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3103                         struct btrfs_root *root,
3104                         struct inode *dir, u64 objectid,
3105                         const char *name, int name_len)
3106 {
3107         struct btrfs_path *path;
3108         struct extent_buffer *leaf;
3109         struct btrfs_dir_item *di;
3110         struct btrfs_key key;
3111         u64 index;
3112         int ret;
3113         u64 dir_ino = btrfs_ino(dir);
3114
3115         path = btrfs_alloc_path();
3116         if (!path)
3117                 return -ENOMEM;
3118
3119         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3120                                    name, name_len, -1);
3121         if (IS_ERR_OR_NULL(di)) {
3122                 if (!di)
3123                         ret = -ENOENT;
3124                 else
3125                         ret = PTR_ERR(di);
3126                 goto out;
3127         }
3128
3129         leaf = path->nodes[0];
3130         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3131         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3132         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3133         if (ret) {
3134                 btrfs_abort_transaction(trans, root, ret);
3135                 goto out;
3136         }
3137         btrfs_release_path(path);
3138
3139         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3140                                  objectid, root->root_key.objectid,
3141                                  dir_ino, &index, name, name_len);
3142         if (ret < 0) {
3143                 if (ret != -ENOENT) {
3144                         btrfs_abort_transaction(trans, root, ret);
3145                         goto out;
3146                 }
3147                 di = btrfs_search_dir_index_item(root, path, dir_ino,
3148                                                  name, name_len);
3149                 if (IS_ERR_OR_NULL(di)) {
3150                         if (!di)
3151                                 ret = -ENOENT;
3152                         else
3153                                 ret = PTR_ERR(di);
3154                         btrfs_abort_transaction(trans, root, ret);
3155                         goto out;
3156                 }
3157
3158                 leaf = path->nodes[0];
3159                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3160                 btrfs_release_path(path);
3161                 index = key.offset;
3162         }
3163         btrfs_release_path(path);
3164
3165         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3166         if (ret) {
3167                 btrfs_abort_transaction(trans, root, ret);
3168                 goto out;
3169         }
3170
3171         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3172         inode_inc_iversion(dir);
3173         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3174         ret = btrfs_update_inode(trans, root, dir);
3175         if (ret)
3176                 btrfs_abort_transaction(trans, root, ret);
3177 out:
3178         btrfs_free_path(path);
3179         return ret;
3180 }
3181
3182 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3183 {
3184         struct inode *inode = dentry->d_inode;
3185         int err = 0;
3186         struct btrfs_root *root = BTRFS_I(dir)->root;
3187         struct btrfs_trans_handle *trans;
3188         unsigned long nr = 0;
3189
3190         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3191             btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3192                 return -ENOTEMPTY;
3193
3194         trans = __unlink_start_trans(dir, dentry);
3195         if (IS_ERR(trans))
3196                 return PTR_ERR(trans);
3197
3198         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3199                 err = btrfs_unlink_subvol(trans, root, dir,
3200                                           BTRFS_I(inode)->location.objectid,
3201                                           dentry->d_name.name,
3202                                           dentry->d_name.len);
3203                 goto out;
3204         }
3205
3206         err = btrfs_orphan_add(trans, inode);
3207         if (err)
3208                 goto out;
3209
3210         /* now the directory is empty */
3211         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3212                                  dentry->d_name.name, dentry->d_name.len);
3213         if (!err)
3214                 btrfs_i_size_write(inode, 0);
3215 out:
3216         nr = trans->blocks_used;
3217         __unlink_end_trans(trans, root);
3218         btrfs_btree_balance_dirty(root, nr);
3219
3220         return err;
3221 }
3222
3223 /*
3224  * this can truncate away extent items, csum items and directory items.
3225  * It starts at a high offset and removes keys until it can't find
3226  * any higher than new_size
3227  *
3228  * csum items that cross the new i_size are truncated to the new size
3229  * as well.
3230  *
3231  * min_type is the minimum key type to truncate down to.  If set to 0, this
3232  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3233  */
3234 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3235                                struct btrfs_root *root,
3236                                struct inode *inode,
3237                                u64 new_size, u32 min_type)
3238 {
3239         struct btrfs_path *path;
3240         struct extent_buffer *leaf;
3241         struct btrfs_file_extent_item *fi;
3242         struct btrfs_key key;
3243         struct btrfs_key found_key;
3244         u64 extent_start = 0;
3245         u64 extent_num_bytes = 0;
3246         u64 extent_offset = 0;
3247         u64 item_end = 0;
3248         u64 mask = root->sectorsize - 1;
3249         u32 found_type = (u8)-1;
3250         int found_extent;
3251         int del_item;
3252         int pending_del_nr = 0;
3253         int pending_del_slot = 0;
3254         int extent_type = -1;
3255         int ret;
3256         int err = 0;
3257         u64 ino = btrfs_ino(inode);
3258
3259         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3260
3261         path = btrfs_alloc_path();
3262         if (!path)
3263                 return -ENOMEM;
3264         path->reada = -1;
3265
3266         if (root->ref_cows || root == root->fs_info->tree_root)
3267                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3268
3269         /*
3270          * This function is also used to drop the items in the log tree before
3271          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3272          * it is used to drop the loged items. So we shouldn't kill the delayed
3273          * items.
3274          */
3275         if (min_type == 0 && root == BTRFS_I(inode)->root)
3276                 btrfs_kill_delayed_inode_items(inode);
3277
3278         key.objectid = ino;
3279         key.offset = (u64)-1;
3280         key.type = (u8)-1;
3281
3282 search_again:
3283         path->leave_spinning = 1;
3284         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3285         if (ret < 0) {
3286                 err = ret;
3287                 goto out;
3288         }
3289
3290         if (ret > 0) {
3291                 /* there are no items in the tree for us to truncate, we're
3292                  * done
3293                  */
3294                 if (path->slots[0] == 0)
3295                         goto out;
3296                 path->slots[0]--;
3297         }
3298
3299         while (1) {
3300                 fi = NULL;
3301                 leaf = path->nodes[0];
3302                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3303                 found_type = btrfs_key_type(&found_key);
3304
3305                 if (found_key.objectid != ino)
3306                         break;
3307
3308                 if (found_type < min_type)
3309                         break;
3310
3311                 item_end = found_key.offset;
3312                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3313                         fi = btrfs_item_ptr(leaf, path->slots[0],
3314                                             struct btrfs_file_extent_item);
3315                         extent_type = btrfs_file_extent_type(leaf, fi);
3316                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3317                                 item_end +=
3318                                     btrfs_file_extent_num_bytes(leaf, fi);
3319                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3320                                 item_end += btrfs_file_extent_inline_len(leaf,
3321                                                                          fi);
3322                         }
3323                         item_end--;
3324                 }
3325                 if (found_type > min_type) {
3326                         del_item = 1;
3327                 } else {
3328                         if (item_end < new_size)
3329                                 break;
3330                         if (found_key.offset >= new_size)
3331                                 del_item = 1;
3332                         else
3333                                 del_item = 0;
3334                 }
3335                 found_extent = 0;
3336                 /* FIXME, shrink the extent if the ref count is only 1 */
3337                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3338                         goto delete;
3339
3340                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3341                         u64 num_dec;
3342                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3343                         if (!del_item) {
3344                                 u64 orig_num_bytes =
3345                                         btrfs_file_extent_num_bytes(leaf, fi);
3346                                 extent_num_bytes = new_size -
3347                                         found_key.offset + root->sectorsize - 1;
3348                                 extent_num_bytes = extent_num_bytes &
3349                                         ~((u64)root->sectorsize - 1);
3350                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3351                                                          extent_num_bytes);
3352                                 num_dec = (orig_num_bytes -
3353                                            extent_num_bytes);
3354                                 if (root->ref_cows && extent_start != 0)
3355                                         inode_sub_bytes(inode, num_dec);
3356                                 btrfs_mark_buffer_dirty(leaf);
3357                         } else {
3358                                 extent_num_bytes =
3359                                         btrfs_file_extent_disk_num_bytes(leaf,
3360                                                                          fi);
3361                                 extent_offset = found_key.offset -
3362                                         btrfs_file_extent_offset(leaf, fi);
3363
3364                                 /* FIXME blocksize != 4096 */
3365                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3366                                 if (extent_start != 0) {
3367                                         found_extent = 1;
3368                                         if (root->ref_cows)
3369                                                 inode_sub_bytes(inode, num_dec);
3370                                 }
3371                         }
3372                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3373                         /*
3374                          * we can't truncate inline items that have had
3375                          * special encodings
3376                          */
3377                         if (!del_item &&
3378                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3379                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3380                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3381                                 u32 size = new_size - found_key.offset;
3382
3383                                 if (root->ref_cows) {
3384                                         inode_sub_bytes(inode, item_end + 1 -
3385                                                         new_size);
3386                                 }
3387                                 size =
3388                                     btrfs_file_extent_calc_inline_size(size);
3389                                 btrfs_truncate_item(trans, root, path,
3390                                                     size, 1);
3391                         } else if (root->ref_cows) {
3392                                 inode_sub_bytes(inode, item_end + 1 -
3393                                                 found_key.offset);
3394                         }
3395                 }
3396 delete:
3397                 if (del_item) {
3398                         if (!pending_del_nr) {
3399                                 /* no pending yet, add ourselves */
3400                                 pending_del_slot = path->slots[0];
3401                                 pending_del_nr = 1;
3402                         } else if (pending_del_nr &&
3403                                    path->slots[0] + 1 == pending_del_slot) {
3404                                 /* hop on the pending chunk */
3405                                 pending_del_nr++;
3406                                 pending_del_slot = path->slots[0];
3407                         } else {
3408                                 BUG();
3409                         }
3410                 } else {
3411                         break;
3412                 }
3413                 if (found_extent && (root->ref_cows ||
3414                                      root == root->fs_info->tree_root)) {
3415                         btrfs_set_path_blocking(path);
3416                         ret = btrfs_free_extent(trans, root, extent_start,
3417                                                 extent_num_bytes, 0,
3418                                                 btrfs_header_owner(leaf),
3419                                                 ino, extent_offset, 0);
3420                         BUG_ON(ret);
3421                 }
3422
3423                 if (found_type == BTRFS_INODE_ITEM_KEY)
3424                         break;
3425
3426                 if (path->slots[0] == 0 ||
3427                     path->slots[0] != pending_del_slot) {
3428                         if (root->ref_cows &&
3429                             BTRFS_I(inode)->location.objectid !=
3430                                                 BTRFS_FREE_INO_OBJECTID) {
3431                                 err = -EAGAIN;
3432                                 goto out;
3433                         }
3434                         if (pending_del_nr) {
3435                                 ret = btrfs_del_items(trans, root, path,
3436                                                 pending_del_slot,
3437                                                 pending_del_nr);
3438                                 if (ret) {
3439                                         btrfs_abort_transaction(trans,
3440                                                                 root, ret);
3441                                         goto error;
3442                                 }
3443                                 pending_del_nr = 0;
3444                         }
3445                         btrfs_release_path(path);
3446                         goto search_again;
3447                 } else {
3448                         path->slots[0]--;
3449                 }
3450         }
3451 out:
3452         if (pending_del_nr) {
3453                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3454                                       pending_del_nr);
3455                 if (ret)
3456                         btrfs_abort_transaction(trans, root, ret);
3457         }
3458 error:
3459         btrfs_free_path(path);
3460         return err;
3461 }
3462
3463 /*
3464  * taken from block_truncate_page, but does cow as it zeros out
3465  * any bytes left in the last page in the file.
3466  */
3467 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3468 {
3469         struct inode *inode = mapping->host;
3470         struct btrfs_root *root = BTRFS_I(inode)->root;
3471         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3472         struct btrfs_ordered_extent *ordered;
3473         struct extent_state *cached_state = NULL;
3474         char *kaddr;
3475         u32 blocksize = root->sectorsize;
3476         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3477         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3478         struct page *page;
3479         gfp_t mask = btrfs_alloc_write_mask(mapping);
3480         int ret = 0;
3481         u64 page_start;
3482         u64 page_end;
3483
3484         if ((offset & (blocksize - 1)) == 0)
3485                 goto out;
3486         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3487         if (ret)
3488                 goto out;
3489
3490         ret = -ENOMEM;
3491 again:
3492         page = find_or_create_page(mapping, index, mask);
3493         if (!page) {
3494                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3495                 goto out;
3496         }
3497
3498         page_start = page_offset(page);
3499         page_end = page_start + PAGE_CACHE_SIZE - 1;
3500
3501         if (!PageUptodate(page)) {
3502                 ret = btrfs_readpage(NULL, page);
3503                 lock_page(page);
3504                 if (page->mapping != mapping) {
3505                         unlock_page(page);
3506                         page_cache_release(page);
3507                         goto again;
3508                 }
3509                 if (!PageUptodate(page)) {
3510                         ret = -EIO;
3511                         goto out_unlock;
3512                 }
3513         }
3514         wait_on_page_writeback(page);
3515
3516         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
3517         set_page_extent_mapped(page);
3518
3519         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3520         if (ordered) {
3521                 unlock_extent_cached(io_tree, page_start, page_end,
3522                                      &cached_state, GFP_NOFS);
3523                 unlock_page(page);
3524                 page_cache_release(page);
3525                 btrfs_start_ordered_extent(inode, ordered, 1);
3526                 btrfs_put_ordered_extent(ordered);
3527                 goto again;
3528         }
3529
3530         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3531                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3532                           0, 0, &cached_state, GFP_NOFS);
3533
3534         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3535                                         &cached_state);
3536         if (ret) {
3537                 unlock_extent_cached(io_tree, page_start, page_end,
3538                                      &cached_state, GFP_NOFS);
3539                 goto out_unlock;
3540         }
3541
3542         ret = 0;
3543         if (offset != PAGE_CACHE_SIZE) {
3544                 kaddr = kmap(page);
3545                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3546                 flush_dcache_page(page);
3547                 kunmap(page);
3548         }
3549         ClearPageChecked(page);
3550         set_page_dirty(page);
3551         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3552                              GFP_NOFS);
3553
3554 out_unlock:
3555         if (ret)
3556                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3557         unlock_page(page);
3558         page_cache_release(page);
3559 out:
3560         return ret;
3561 }
3562
3563 /*
3564  * This function puts in dummy file extents for the area we're creating a hole
3565  * for.  So if we are truncating this file to a larger size we need to insert
3566  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3567  * the range between oldsize and size
3568  */
3569 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3570 {
3571         struct btrfs_trans_handle *trans;
3572         struct btrfs_root *root = BTRFS_I(inode)->root;
3573         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3574         struct extent_map *em = NULL;
3575         struct extent_state *cached_state = NULL;
3576         u64 mask = root->sectorsize - 1;
3577         u64 hole_start = (oldsize + mask) & ~mask;
3578         u64 block_end = (size + mask) & ~mask;
3579         u64 last_byte;
3580         u64 cur_offset;
3581         u64 hole_size;
3582         int err = 0;
3583
3584         if (size <= hole_start)
3585                 return 0;
3586
3587         while (1) {
3588                 struct btrfs_ordered_extent *ordered;
3589                 btrfs_wait_ordered_range(inode, hole_start,
3590                                          block_end - hole_start);
3591                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3592                                  &cached_state);
3593                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3594                 if (!ordered)
3595                         break;
3596                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3597                                      &cached_state, GFP_NOFS);
3598                 btrfs_put_ordered_extent(ordered);
3599         }
3600
3601         cur_offset = hole_start;
3602         while (1) {
3603                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3604                                 block_end - cur_offset, 0);
3605                 if (IS_ERR(em)) {
3606                         err = PTR_ERR(em);
3607                         break;
3608                 }
3609                 last_byte = min(extent_map_end(em), block_end);
3610                 last_byte = (last_byte + mask) & ~mask;
3611                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3612                         u64 hint_byte = 0;
3613                         hole_size = last_byte - cur_offset;
3614
3615                         trans = btrfs_start_transaction(root, 3);
3616                         if (IS_ERR(trans)) {
3617                                 err = PTR_ERR(trans);
3618                                 break;
3619                         }
3620
3621                         err = btrfs_drop_extents(trans, inode, cur_offset,
3622                                                  cur_offset + hole_size,
3623                                                  &hint_byte, 1);
3624                         if (err) {
3625                                 btrfs_abort_transaction(trans, root, err);
3626                                 btrfs_end_transaction(trans, root);
3627                                 break;
3628                         }
3629
3630                         err = btrfs_insert_file_extent(trans, root,
3631                                         btrfs_ino(inode), cur_offset, 0,
3632                                         0, hole_size, 0, hole_size,
3633                                         0, 0, 0);
3634                         if (err) {
3635                                 btrfs_abort_transaction(trans, root, err);
3636                                 btrfs_end_transaction(trans, root);
3637                                 break;
3638                         }
3639
3640                         btrfs_drop_extent_cache(inode, hole_start,
3641                                         last_byte - 1, 0);
3642
3643                         btrfs_update_inode(trans, root, inode);
3644                         btrfs_end_transaction(trans, root);
3645                 }
3646                 free_extent_map(em);
3647                 em = NULL;
3648                 cur_offset = last_byte;
3649                 if (cur_offset >= block_end)
3650                         break;
3651         }
3652
3653         free_extent_map(em);
3654         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3655                              GFP_NOFS);
3656         return err;
3657 }
3658
3659 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3660 {
3661         struct btrfs_root *root = BTRFS_I(inode)->root;
3662         struct btrfs_trans_handle *trans;
3663         loff_t oldsize = i_size_read(inode);
3664         int ret;
3665
3666         if (newsize == oldsize)
3667                 return 0;
3668
3669         if (newsize > oldsize) {
3670                 truncate_pagecache(inode, oldsize, newsize);
3671                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3672                 if (ret)
3673                         return ret;
3674
3675                 trans = btrfs_start_transaction(root, 1);
3676                 if (IS_ERR(trans))
3677                         return PTR_ERR(trans);
3678
3679                 i_size_write(inode, newsize);
3680                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3681                 ret = btrfs_update_inode(trans, root, inode);
3682                 btrfs_end_transaction(trans, root);
3683         } else {
3684
3685                 /*
3686                  * We're truncating a file that used to have good data down to
3687                  * zero. Make sure it gets into the ordered flush list so that
3688                  * any new writes get down to disk quickly.
3689                  */
3690                 if (newsize == 0)
3691                         set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
3692                                 &BTRFS_I(inode)->runtime_flags);
3693
3694                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3695                 truncate_setsize(inode, newsize);
3696                 ret = btrfs_truncate(inode);
3697         }
3698
3699         return ret;
3700 }
3701
3702 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3703 {
3704         struct inode *inode = dentry->d_inode;
3705         struct btrfs_root *root = BTRFS_I(inode)->root;
3706         int err;
3707
3708         if (btrfs_root_readonly(root))
3709                 return -EROFS;
3710
3711         err = inode_change_ok(inode, attr);
3712         if (err)
3713                 return err;
3714
3715         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3716                 err = btrfs_setsize(inode, attr->ia_size);
3717                 if (err)
3718                         return err;
3719         }
3720
3721         if (attr->ia_valid) {
3722                 setattr_copy(inode, attr);
3723                 inode_inc_iversion(inode);
3724                 err = btrfs_dirty_inode(inode);
3725
3726                 if (!err && attr->ia_valid & ATTR_MODE)
3727                         err = btrfs_acl_chmod(inode);
3728         }
3729
3730         return err;
3731 }
3732
3733 void btrfs_evict_inode(struct inode *inode)
3734 {
3735         struct btrfs_trans_handle *trans;
3736         struct btrfs_root *root = BTRFS_I(inode)->root;
3737         struct btrfs_block_rsv *rsv, *global_rsv;
3738         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
3739         unsigned long nr;
3740         int ret;
3741
3742         trace_btrfs_inode_evict(inode);
3743
3744         truncate_inode_pages(&inode->i_data, 0);
3745         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3746                                btrfs_is_free_space_inode(root, inode)))
3747                 goto no_delete;
3748
3749         if (is_bad_inode(inode)) {
3750                 btrfs_orphan_del(NULL, inode);
3751                 goto no_delete;
3752         }
3753         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3754         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3755
3756         if (root->fs_info->log_root_recovering) {
3757                 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3758                                  &BTRFS_I(inode)->runtime_flags));
3759                 goto no_delete;
3760         }
3761
3762         if (inode->i_nlink > 0) {
3763                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3764                 goto no_delete;
3765         }
3766
3767         rsv = btrfs_alloc_block_rsv(root);
3768         if (!rsv) {
3769                 btrfs_orphan_del(NULL, inode);
3770                 goto no_delete;
3771         }
3772         rsv->size = min_size;
3773         global_rsv = &root->fs_info->global_block_rsv;
3774
3775         btrfs_i_size_write(inode, 0);
3776
3777         /*
3778          * This is a bit simpler than btrfs_truncate since
3779          *
3780          * 1) We've already reserved our space for our orphan item in the
3781          *    unlink.
3782          * 2) We're going to delete the inode item, so we don't need to update
3783          *    it at all.
3784          *
3785          * So we just need to reserve some slack space in case we add bytes when
3786          * doing the truncate.
3787          */
3788         while (1) {
3789                 ret = btrfs_block_rsv_refill_noflush(root, rsv, min_size);
3790
3791                 /*
3792                  * Try and steal from the global reserve since we will
3793                  * likely not use this space anyway, we want to try as
3794                  * hard as possible to get this to work.
3795                  */
3796                 if (ret)
3797                         ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
3798
3799                 if (ret) {
3800                         printk(KERN_WARNING "Could not get space for a "
3801                                "delete, will truncate on mount %d\n", ret);
3802                         btrfs_orphan_del(NULL, inode);
3803                         btrfs_free_block_rsv(root, rsv);
3804                         goto no_delete;
3805                 }
3806
3807                 trans = btrfs_start_transaction(root, 0);
3808                 if (IS_ERR(trans)) {
3809                         btrfs_orphan_del(NULL, inode);
3810                         btrfs_free_block_rsv(root, rsv);
3811                         goto no_delete;
3812                 }
3813
3814                 trans->block_rsv = rsv;
3815
3816                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3817                 if (ret != -EAGAIN)
3818                         break;
3819
3820                 nr = trans->blocks_used;
3821                 btrfs_end_transaction(trans, root);
3822                 trans = NULL;
3823                 btrfs_btree_balance_dirty(root, nr);
3824         }
3825
3826         btrfs_free_block_rsv(root, rsv);
3827
3828         if (ret == 0) {
3829                 trans->block_rsv = root->orphan_block_rsv;
3830                 ret = btrfs_orphan_del(trans, inode);
3831                 BUG_ON(ret);
3832         }
3833
3834         trans->block_rsv = &root->fs_info->trans_block_rsv;
3835         if (!(root == root->fs_info->tree_root ||
3836               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3837                 btrfs_return_ino(root, btrfs_ino(inode));
3838
3839         nr = trans->blocks_used;
3840         btrfs_end_transaction(trans, root);
3841         btrfs_btree_balance_dirty(root, nr);
3842 no_delete:
3843         clear_inode(inode);
3844         return;
3845 }
3846
3847 /*
3848  * this returns the key found in the dir entry in the location pointer.
3849  * If no dir entries were found, location->objectid is 0.
3850  */
3851 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3852                                struct btrfs_key *location)
3853 {
3854         const char *name = dentry->d_name.name;
3855         int namelen = dentry->d_name.len;
3856         struct btrfs_dir_item *di;
3857         struct btrfs_path *path;
3858         struct btrfs_root *root = BTRFS_I(dir)->root;
3859         int ret = 0;
3860
3861         path = btrfs_alloc_path();
3862         if (!path)
3863                 return -ENOMEM;
3864
3865         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3866                                     namelen, 0);
3867         if (IS_ERR(di))
3868                 ret = PTR_ERR(di);
3869
3870         if (IS_ERR_OR_NULL(di))
3871                 goto out_err;
3872
3873         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3874 out:
3875         btrfs_free_path(path);
3876         return ret;
3877 out_err:
3878         location->objectid = 0;
3879         goto out;
3880 }
3881
3882 /*
3883  * when we hit a tree root in a directory, the btrfs part of the inode
3884  * needs to be changed to reflect the root directory of the tree root.  This
3885  * is kind of like crossing a mount point.
3886  */
3887 static int fixup_tree_root_location(struct btrfs_root *root,
3888                                     struct inode *dir,
3889                                     struct dentry *dentry,
3890                                     struct btrfs_key *location,
3891                                     struct btrfs_root **sub_root)
3892 {
3893         struct btrfs_path *path;
3894         struct btrfs_root *new_root;
3895         struct btrfs_root_ref *ref;
3896         struct extent_buffer *leaf;
3897         int ret;
3898         int err = 0;
3899
3900         path = btrfs_alloc_path();
3901         if (!path) {
3902                 err = -ENOMEM;
3903                 goto out;
3904         }
3905
3906         err = -ENOENT;
3907         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3908                                   BTRFS_I(dir)->root->root_key.objectid,
3909                                   location->objectid);
3910         if (ret) {
3911                 if (ret < 0)
3912                         err = ret;
3913                 goto out;
3914         }
3915
3916         leaf = path->nodes[0];
3917         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3918         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3919             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3920                 goto out;
3921
3922         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3923                                    (unsigned long)(ref + 1),
3924                                    dentry->d_name.len);
3925         if (ret)
3926                 goto out;
3927
3928         btrfs_release_path(path);
3929
3930         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3931         if (IS_ERR(new_root)) {
3932                 err = PTR_ERR(new_root);
3933                 goto out;
3934         }
3935
3936         if (btrfs_root_refs(&new_root->root_item) == 0) {
3937                 err = -ENOENT;
3938                 goto out;
3939         }
3940
3941         *sub_root = new_root;
3942         location->objectid = btrfs_root_dirid(&new_root->root_item);
3943         location->type = BTRFS_INODE_ITEM_KEY;
3944         location->offset = 0;
3945         err = 0;
3946 out:
3947         btrfs_free_path(path);
3948         return err;
3949 }
3950
3951 static void inode_tree_add(struct inode *inode)
3952 {
3953         struct btrfs_root *root = BTRFS_I(inode)->root;
3954         struct btrfs_inode *entry;
3955         struct rb_node **p;
3956         struct rb_node *parent;
3957         u64 ino = btrfs_ino(inode);
3958 again:
3959         p = &root->inode_tree.rb_node;
3960         parent = NULL;
3961
3962         if (inode_unhashed(inode))
3963                 return;
3964
3965         spin_lock(&root->inode_lock);
3966         while (*p) {
3967                 parent = *p;
3968                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3969
3970                 if (ino < btrfs_ino(&entry->vfs_inode))
3971                         p = &parent->rb_left;
3972                 else if (ino > btrfs_ino(&entry->vfs_inode))
3973                         p = &parent->rb_right;
3974                 else {
3975                         WARN_ON(!(entry->vfs_inode.i_state &
3976                                   (I_WILL_FREE | I_FREEING)));
3977                         rb_erase(parent, &root->inode_tree);
3978                         RB_CLEAR_NODE(parent);
3979                         spin_unlock(&root->inode_lock);
3980                         goto again;
3981                 }
3982         }
3983         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3984         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3985         spin_unlock(&root->inode_lock);
3986 }
3987
3988 static void inode_tree_del(struct inode *inode)
3989 {
3990         struct btrfs_root *root = BTRFS_I(inode)->root;
3991         int empty = 0;
3992
3993         spin_lock(&root->inode_lock);
3994         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3995                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3996                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3997                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3998         }
3999         spin_unlock(&root->inode_lock);
4000
4001         /*
4002          * Free space cache has inodes in the tree root, but the tree root has a
4003          * root_refs of 0, so this could end up dropping the tree root as a
4004          * snapshot, so we need the extra !root->fs_info->tree_root check to
4005          * make sure we don't drop it.
4006          */
4007         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
4008             root != root->fs_info->tree_root) {
4009                 synchronize_srcu(&root->fs_info->subvol_srcu);
4010                 spin_lock(&root->inode_lock);
4011                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4012                 spin_unlock(&root->inode_lock);
4013                 if (empty)
4014                         btrfs_add_dead_root(root);
4015         }
4016 }
4017
4018 void btrfs_invalidate_inodes(struct btrfs_root *root)
4019 {
4020         struct rb_node *node;
4021         struct rb_node *prev;
4022         struct btrfs_inode *entry;
4023         struct inode *inode;
4024         u64 objectid = 0;
4025
4026         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4027
4028         spin_lock(&root->inode_lock);
4029 again:
4030         node = root->inode_tree.rb_node;
4031         prev = NULL;
4032         while (node) {
4033                 prev = node;
4034                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4035
4036                 if (objectid < btrfs_ino(&entry->vfs_inode))
4037                         node = node->rb_left;
4038                 else if (objectid > btrfs_ino(&entry->vfs_inode))
4039                         node = node->rb_right;
4040                 else
4041                         break;
4042         }
4043         if (!node) {
4044                 while (prev) {
4045                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
4046                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4047                                 node = prev;
4048                                 break;
4049                         }
4050                         prev = rb_next(prev);
4051                 }
4052         }
4053         while (node) {
4054                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4055                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4056                 inode = igrab(&entry->vfs_inode);
4057                 if (inode) {
4058                         spin_unlock(&root->inode_lock);
4059                         if (atomic_read(&inode->i_count) > 1)
4060                                 d_prune_aliases(inode);
4061                         /*
4062                          * btrfs_drop_inode will have it removed from
4063                          * the inode cache when its usage count
4064                          * hits zero.
4065                          */
4066                         iput(inode);
4067                         cond_resched();
4068                         spin_lock(&root->inode_lock);
4069                         goto again;
4070                 }
4071
4072                 if (cond_resched_lock(&root->inode_lock))
4073                         goto again;
4074
4075                 node = rb_next(node);
4076         }
4077         spin_unlock(&root->inode_lock);
4078 }
4079
4080 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4081 {
4082         struct btrfs_iget_args *args = p;
4083         inode->i_ino = args->ino;
4084         BTRFS_I(inode)->root = args->root;
4085         btrfs_set_inode_space_info(args->root, inode);
4086         return 0;
4087 }
4088
4089 static int btrfs_find_actor(struct inode *inode, void *opaque)
4090 {
4091         struct btrfs_iget_args *args = opaque;
4092         return args->ino == btrfs_ino(inode) &&
4093                 args->root == BTRFS_I(inode)->root;
4094 }
4095
4096 static struct inode *btrfs_iget_locked(struct super_block *s,
4097                                        u64 objectid,
4098                                        struct btrfs_root *root)
4099 {
4100         struct inode *inode;
4101         struct btrfs_iget_args args;
4102         args.ino = objectid;
4103         args.root = root;
4104
4105         inode = iget5_locked(s, objectid, btrfs_find_actor,
4106                              btrfs_init_locked_inode,
4107                              (void *)&args);
4108         return inode;
4109 }
4110
4111 /* Get an inode object given its location and corresponding root.
4112  * Returns in *is_new if the inode was read from disk
4113  */
4114 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4115                          struct btrfs_root *root, int *new)
4116 {
4117         struct inode *inode;
4118
4119         inode = btrfs_iget_locked(s, location->objectid, root);
4120         if (!inode)
4121                 return ERR_PTR(-ENOMEM);
4122
4123         if (inode->i_state & I_NEW) {
4124                 BTRFS_I(inode)->root = root;
4125                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4126                 btrfs_read_locked_inode(inode);
4127                 if (!is_bad_inode(inode)) {
4128                         inode_tree_add(inode);
4129                         unlock_new_inode(inode);
4130                         if (new)
4131                                 *new = 1;
4132                 } else {
4133                         unlock_new_inode(inode);
4134                         iput(inode);
4135                         inode = ERR_PTR(-ESTALE);
4136                 }
4137         }
4138
4139         return inode;
4140 }
4141
4142 static struct inode *new_simple_dir(struct super_block *s,
4143                                     struct btrfs_key *key,
4144                                     struct btrfs_root *root)
4145 {
4146         struct inode *inode = new_inode(s);
4147
4148         if (!inode)
4149                 return ERR_PTR(-ENOMEM);
4150
4151         BTRFS_I(inode)->root = root;
4152         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4153         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4154
4155         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4156         inode->i_op = &btrfs_dir_ro_inode_operations;
4157         inode->i_fop = &simple_dir_operations;
4158         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4159         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4160
4161         return inode;
4162 }
4163
4164 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4165 {
4166         struct inode *inode;
4167         struct btrfs_root *root = BTRFS_I(dir)->root;
4168         struct btrfs_root *sub_root = root;
4169         struct btrfs_key location;
4170         int index;
4171         int ret = 0;
4172
4173         if (dentry->d_name.len > BTRFS_NAME_LEN)
4174                 return ERR_PTR(-ENAMETOOLONG);
4175
4176         if (unlikely(d_need_lookup(dentry))) {
4177                 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
4178                 kfree(dentry->d_fsdata);
4179                 dentry->d_fsdata = NULL;
4180                 /* This thing is hashed, drop it for now */
4181                 d_drop(dentry);
4182         } else {
4183                 ret = btrfs_inode_by_name(dir, dentry, &location);
4184         }
4185
4186         if (ret < 0)
4187                 return ERR_PTR(ret);
4188
4189         if (location.objectid == 0)
4190                 return NULL;
4191
4192         if (location.type == BTRFS_INODE_ITEM_KEY) {
4193                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4194                 return inode;
4195         }
4196
4197         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4198
4199         index = srcu_read_lock(&root->fs_info->subvol_srcu);
4200         ret = fixup_tree_root_location(root, dir, dentry,
4201                                        &location, &sub_root);
4202         if (ret < 0) {
4203                 if (ret != -ENOENT)
4204                         inode = ERR_PTR(ret);
4205                 else
4206                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
4207         } else {
4208                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4209         }
4210         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4211
4212         if (!IS_ERR(inode) && root != sub_root) {
4213                 down_read(&root->fs_info->cleanup_work_sem);
4214                 if (!(inode->i_sb->s_flags & MS_RDONLY))
4215                         ret = btrfs_orphan_cleanup(sub_root);
4216                 up_read(&root->fs_info->cleanup_work_sem);
4217                 if (ret)
4218                         inode = ERR_PTR(ret);
4219         }
4220
4221         return inode;
4222 }
4223
4224 static int btrfs_dentry_delete(const struct dentry *dentry)
4225 {
4226         struct btrfs_root *root;
4227         struct inode *inode = dentry->d_inode;
4228
4229         if (!inode && !IS_ROOT(dentry))
4230                 inode = dentry->d_parent->d_inode;
4231
4232         if (inode) {
4233                 root = BTRFS_I(inode)->root;
4234                 if (btrfs_root_refs(&root->root_item) == 0)
4235                         return 1;
4236
4237                 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
4238                         return 1;
4239         }
4240         return 0;
4241 }
4242
4243 static void btrfs_dentry_release(struct dentry *dentry)
4244 {
4245         if (dentry->d_fsdata)
4246                 kfree(dentry->d_fsdata);
4247 }
4248
4249 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4250                                    unsigned int flags)
4251 {
4252         struct dentry *ret;
4253
4254         ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4255         if (unlikely(d_need_lookup(dentry))) {
4256                 spin_lock(&dentry->d_lock);
4257                 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4258                 spin_unlock(&dentry->d_lock);
4259         }
4260         return ret;
4261 }
4262
4263 unsigned char btrfs_filetype_table[] = {
4264         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4265 };
4266
4267 static int btrfs_real_readdir(struct file *filp, void *dirent,
4268                               filldir_t filldir)
4269 {
4270         struct inode *inode = filp->f_dentry->d_inode;
4271         struct btrfs_root *root = BTRFS_I(inode)->root;
4272         struct btrfs_item *item;
4273         struct btrfs_dir_item *di;
4274         struct btrfs_key key;
4275         struct btrfs_key found_key;
4276         struct btrfs_path *path;
4277         struct list_head ins_list;
4278         struct list_head del_list;
4279         int ret;
4280         struct extent_buffer *leaf;
4281         int slot;
4282         unsigned char d_type;
4283         int over = 0;
4284         u32 di_cur;
4285         u32 di_total;
4286         u32 di_len;
4287         int key_type = BTRFS_DIR_INDEX_KEY;
4288         char tmp_name[32];
4289         char *name_ptr;
4290         int name_len;
4291         int is_curr = 0;        /* filp->f_pos points to the current index? */
4292
4293         /* FIXME, use a real flag for deciding about the key type */
4294         if (root->fs_info->tree_root == root)
4295                 key_type = BTRFS_DIR_ITEM_KEY;
4296
4297         /* special case for "." */
4298         if (filp->f_pos == 0) {
4299                 over = filldir(dirent, ".", 1,
4300                                filp->f_pos, btrfs_ino(inode), DT_DIR);
4301                 if (over)
4302                         return 0;
4303                 filp->f_pos = 1;
4304         }
4305         /* special case for .., just use the back ref */
4306         if (filp->f_pos == 1) {
4307                 u64 pino = parent_ino(filp->f_path.dentry);
4308                 over = filldir(dirent, "..", 2,
4309                                filp->f_pos, pino, DT_DIR);
4310                 if (over)
4311                         return 0;
4312                 filp->f_pos = 2;
4313         }
4314         path = btrfs_alloc_path();
4315         if (!path)
4316                 return -ENOMEM;
4317
4318         path->reada = 1;
4319
4320         if (key_type == BTRFS_DIR_INDEX_KEY) {
4321                 INIT_LIST_HEAD(&ins_list);
4322                 INIT_LIST_HEAD(&del_list);
4323                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4324         }
4325
4326         btrfs_set_key_type(&key, key_type);
4327         key.offset = filp->f_pos;
4328         key.objectid = btrfs_ino(inode);
4329
4330         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4331         if (ret < 0)
4332                 goto err;
4333
4334         while (1) {
4335                 leaf = path->nodes[0];
4336                 slot = path->slots[0];
4337                 if (slot >= btrfs_header_nritems(leaf)) {
4338                         ret = btrfs_next_leaf(root, path);
4339                         if (ret < 0)
4340                                 goto err;
4341                         else if (ret > 0)
4342                                 break;
4343                         continue;
4344                 }
4345
4346                 item = btrfs_item_nr(leaf, slot);
4347                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4348
4349                 if (found_key.objectid != key.objectid)
4350                         break;
4351                 if (btrfs_key_type(&found_key) != key_type)
4352                         break;
4353                 if (found_key.offset < filp->f_pos)
4354                         goto next;
4355                 if (key_type == BTRFS_DIR_INDEX_KEY &&
4356                     btrfs_should_delete_dir_index(&del_list,
4357                                                   found_key.offset))
4358                         goto next;
4359
4360                 filp->f_pos = found_key.offset;
4361                 is_curr = 1;
4362
4363                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4364                 di_cur = 0;
4365                 di_total = btrfs_item_size(leaf, item);
4366
4367                 while (di_cur < di_total) {
4368                         struct btrfs_key location;
4369
4370                         if (verify_dir_item(root, leaf, di))
4371                                 break;
4372
4373                         name_len = btrfs_dir_name_len(leaf, di);
4374                         if (name_len <= sizeof(tmp_name)) {
4375                                 name_ptr = tmp_name;
4376                         } else {
4377                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4378                                 if (!name_ptr) {
4379                                         ret = -ENOMEM;
4380                                         goto err;
4381                                 }
4382                         }
4383                         read_extent_buffer(leaf, name_ptr,
4384                                            (unsigned long)(di + 1), name_len);
4385
4386                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4387                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4388
4389
4390                         /* is this a reference to our own snapshot? If so
4391                          * skip it.
4392                          *
4393                          * In contrast to old kernels, we insert the snapshot's
4394                          * dir item and dir index after it has been created, so
4395                          * we won't find a reference to our own snapshot. We
4396                          * still keep the following code for backward
4397                          * compatibility.
4398                          */
4399                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4400                             location.objectid == root->root_key.objectid) {
4401                                 over = 0;
4402                                 goto skip;
4403                         }
4404                         over = filldir(dirent, name_ptr, name_len,
4405                                        found_key.offset, location.objectid,
4406                                        d_type);
4407
4408 skip:
4409                         if (name_ptr != tmp_name)
4410                                 kfree(name_ptr);
4411
4412                         if (over)
4413                                 goto nopos;
4414                         di_len = btrfs_dir_name_len(leaf, di) +
4415                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4416                         di_cur += di_len;
4417                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4418                 }
4419 next:
4420                 path->slots[0]++;
4421         }
4422
4423         if (key_type == BTRFS_DIR_INDEX_KEY) {
4424                 if (is_curr)
4425                         filp->f_pos++;
4426                 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4427                                                       &ins_list);
4428                 if (ret)
4429                         goto nopos;
4430         }
4431
4432         /* Reached end of directory/root. Bump pos past the last item. */
4433         if (key_type == BTRFS_DIR_INDEX_KEY)
4434                 /*
4435                  * 32-bit glibc will use getdents64, but then strtol -
4436                  * so the last number we can serve is this.
4437                  */
4438                 filp->f_pos = 0x7fffffff;
4439         else
4440                 filp->f_pos++;
4441 nopos:
4442         ret = 0;
4443 err:
4444         if (key_type == BTRFS_DIR_INDEX_KEY)
4445                 btrfs_put_delayed_items(&ins_list, &del_list);
4446         btrfs_free_path(path);
4447         return ret;
4448 }
4449
4450 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4451 {
4452         struct btrfs_root *root = BTRFS_I(inode)->root;
4453         struct btrfs_trans_handle *trans;
4454         int ret = 0;
4455         bool nolock = false;
4456
4457         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4458                 return 0;
4459
4460         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(root, inode))
4461                 nolock = true;
4462
4463         if (wbc->sync_mode == WB_SYNC_ALL) {
4464                 if (nolock)
4465                         trans = btrfs_join_transaction_nolock(root);
4466                 else
4467                         trans = btrfs_join_transaction(root);
4468                 if (IS_ERR(trans))
4469                         return PTR_ERR(trans);
4470                 if (nolock)
4471                         ret = btrfs_end_transaction_nolock(trans, root);
4472                 else
4473                         ret = btrfs_commit_transaction(trans, root);
4474         }
4475         return ret;
4476 }
4477
4478 /*
4479  * This is somewhat expensive, updating the tree every time the
4480  * inode changes.  But, it is most likely to find the inode in cache.
4481  * FIXME, needs more benchmarking...there are no reasons other than performance
4482  * to keep or drop this code.
4483  */
4484 int btrfs_dirty_inode(struct inode *inode)
4485 {
4486         struct btrfs_root *root = BTRFS_I(inode)->root;
4487         struct btrfs_trans_handle *trans;
4488         int ret;
4489
4490         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4491                 return 0;
4492
4493         trans = btrfs_join_transaction(root);
4494         if (IS_ERR(trans))
4495                 return PTR_ERR(trans);
4496
4497         ret = btrfs_update_inode(trans, root, inode);
4498         if (ret && ret == -ENOSPC) {
4499                 /* whoops, lets try again with the full transaction */
4500                 btrfs_end_transaction(trans, root);
4501                 trans = btrfs_start_transaction(root, 1);
4502                 if (IS_ERR(trans))
4503                         return PTR_ERR(trans);
4504
4505                 ret = btrfs_update_inode(trans, root, inode);
4506         }
4507         btrfs_end_transaction(trans, root);
4508         if (BTRFS_I(inode)->delayed_node)
4509                 btrfs_balance_delayed_items(root);
4510
4511         return ret;
4512 }
4513
4514 /*
4515  * This is a copy of file_update_time.  We need this so we can return error on
4516  * ENOSPC for updating the inode in the case of file write and mmap writes.
4517  */
4518 static int btrfs_update_time(struct inode *inode, struct timespec *now,
4519                              int flags)
4520 {
4521         if (flags & S_VERSION)
4522                 inode_inc_iversion(inode);
4523         if (flags & S_CTIME)
4524                 inode->i_ctime = *now;
4525         if (flags & S_MTIME)
4526                 inode->i_mtime = *now;
4527         if (flags & S_ATIME)
4528                 inode->i_atime = *now;
4529         return btrfs_dirty_inode(inode);
4530 }
4531
4532 /*
4533  * find the highest existing sequence number in a directory
4534  * and then set the in-memory index_cnt variable to reflect
4535  * free sequence numbers
4536  */
4537 static int btrfs_set_inode_index_count(struct inode *inode)
4538 {
4539         struct btrfs_root *root = BTRFS_I(inode)->root;
4540         struct btrfs_key key, found_key;
4541         struct btrfs_path *path;
4542         struct extent_buffer *leaf;
4543         int ret;
4544
4545         key.objectid = btrfs_ino(inode);
4546         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4547         key.offset = (u64)-1;
4548
4549         path = btrfs_alloc_path();
4550         if (!path)
4551                 return -ENOMEM;
4552
4553         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4554         if (ret < 0)
4555                 goto out;
4556         /* FIXME: we should be able to handle this */
4557         if (ret == 0)
4558                 goto out;
4559         ret = 0;
4560
4561         /*
4562          * MAGIC NUMBER EXPLANATION:
4563          * since we search a directory based on f_pos we have to start at 2
4564          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4565          * else has to start at 2
4566          */
4567         if (path->slots[0] == 0) {
4568                 BTRFS_I(inode)->index_cnt = 2;
4569                 goto out;
4570         }
4571
4572         path->slots[0]--;
4573
4574         leaf = path->nodes[0];
4575         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4576
4577         if (found_key.objectid != btrfs_ino(inode) ||
4578             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4579                 BTRFS_I(inode)->index_cnt = 2;
4580                 goto out;
4581         }
4582
4583         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4584 out:
4585         btrfs_free_path(path);
4586         return ret;
4587 }
4588
4589 /*
4590  * helper to find a free sequence number in a given directory.  This current
4591  * code is very simple, later versions will do smarter things in the btree
4592  */
4593 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4594 {
4595         int ret = 0;
4596
4597         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4598                 ret = btrfs_inode_delayed_dir_index_count(dir);
4599                 if (ret) {
4600                         ret = btrfs_set_inode_index_count(dir);
4601                         if (ret)
4602                                 return ret;
4603                 }
4604         }
4605
4606         *index = BTRFS_I(dir)->index_cnt;
4607         BTRFS_I(dir)->index_cnt++;
4608
4609         return ret;
4610 }
4611
4612 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4613                                      struct btrfs_root *root,
4614                                      struct inode *dir,
4615                                      const char *name, int name_len,
4616                                      u64 ref_objectid, u64 objectid,
4617                                      umode_t mode, u64 *index)
4618 {
4619         struct inode *inode;
4620         struct btrfs_inode_item *inode_item;
4621         struct btrfs_key *location;
4622         struct btrfs_path *path;
4623         struct btrfs_inode_ref *ref;
4624         struct btrfs_key key[2];
4625         u32 sizes[2];
4626         unsigned long ptr;
4627         int ret;
4628         int owner;
4629
4630         path = btrfs_alloc_path();
4631         if (!path)
4632                 return ERR_PTR(-ENOMEM);
4633
4634         inode = new_inode(root->fs_info->sb);
4635         if (!inode) {
4636                 btrfs_free_path(path);
4637                 return ERR_PTR(-ENOMEM);
4638         }
4639
4640         /*
4641          * we have to initialize this early, so we can reclaim the inode
4642          * number if we fail afterwards in this function.
4643          */
4644         inode->i_ino = objectid;
4645
4646         if (dir) {
4647                 trace_btrfs_inode_request(dir);
4648
4649                 ret = btrfs_set_inode_index(dir, index);
4650                 if (ret) {
4651                         btrfs_free_path(path);
4652                         iput(inode);
4653                         return ERR_PTR(ret);
4654                 }
4655         }
4656         /*
4657          * index_cnt is ignored for everything but a dir,
4658          * btrfs_get_inode_index_count has an explanation for the magic
4659          * number
4660          */
4661         BTRFS_I(inode)->index_cnt = 2;
4662         BTRFS_I(inode)->root = root;
4663         BTRFS_I(inode)->generation = trans->transid;
4664         inode->i_generation = BTRFS_I(inode)->generation;
4665         btrfs_set_inode_space_info(root, inode);
4666
4667         if (S_ISDIR(mode))
4668                 owner = 0;
4669         else
4670                 owner = 1;
4671
4672         key[0].objectid = objectid;
4673         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4674         key[0].offset = 0;
4675
4676         key[1].objectid = objectid;
4677         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4678         key[1].offset = ref_objectid;
4679
4680         sizes[0] = sizeof(struct btrfs_inode_item);
4681         sizes[1] = name_len + sizeof(*ref);
4682
4683         path->leave_spinning = 1;
4684         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4685         if (ret != 0)
4686                 goto fail;
4687
4688         inode_init_owner(inode, dir, mode);
4689         inode_set_bytes(inode, 0);
4690         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4691         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4692                                   struct btrfs_inode_item);
4693         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4694
4695         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4696                              struct btrfs_inode_ref);
4697         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4698         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4699         ptr = (unsigned long)(ref + 1);
4700         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4701
4702         btrfs_mark_buffer_dirty(path->nodes[0]);
4703         btrfs_free_path(path);
4704
4705         location = &BTRFS_I(inode)->location;
4706         location->objectid = objectid;
4707         location->offset = 0;
4708         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4709
4710         btrfs_inherit_iflags(inode, dir);
4711
4712         if (S_ISREG(mode)) {
4713                 if (btrfs_test_opt(root, NODATASUM))
4714                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4715                 if (btrfs_test_opt(root, NODATACOW) ||
4716                     (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4717                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4718         }
4719
4720         insert_inode_hash(inode);
4721         inode_tree_add(inode);
4722
4723         trace_btrfs_inode_new(inode);
4724         btrfs_set_inode_last_trans(trans, inode);
4725
4726         return inode;
4727 fail:
4728         if (dir)
4729                 BTRFS_I(dir)->index_cnt--;
4730         btrfs_free_path(path);
4731         iput(inode);
4732         return ERR_PTR(ret);
4733 }
4734
4735 static inline u8 btrfs_inode_type(struct inode *inode)
4736 {
4737         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4738 }
4739
4740 /*
4741  * utility function to add 'inode' into 'parent_inode' with
4742  * a give name and a given sequence number.
4743  * if 'add_backref' is true, also insert a backref from the
4744  * inode to the parent directory.
4745  */
4746 int btrfs_add_link(struct btrfs_trans_handle *trans,
4747                    struct inode *parent_inode, struct inode *inode,
4748                    const char *name, int name_len, int add_backref, u64 index)
4749 {
4750         int ret = 0;
4751         struct btrfs_key key;
4752         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4753         u64 ino = btrfs_ino(inode);
4754         u64 parent_ino = btrfs_ino(parent_inode);
4755
4756         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4757                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4758         } else {
4759                 key.objectid = ino;
4760                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4761                 key.offset = 0;
4762         }
4763
4764         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4765                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4766                                          key.objectid, root->root_key.objectid,
4767                                          parent_ino, index, name, name_len);
4768         } else if (add_backref) {
4769                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4770                                              parent_ino, index);
4771         }
4772
4773         /* Nothing to clean up yet */
4774         if (ret)
4775                 return ret;
4776
4777         ret = btrfs_insert_dir_item(trans, root, name, name_len,
4778                                     parent_inode, &key,
4779                                     btrfs_inode_type(inode), index);
4780         if (ret == -EEXIST)
4781                 goto fail_dir_item;
4782         else if (ret) {
4783                 btrfs_abort_transaction(trans, root, ret);
4784                 return ret;
4785         }
4786
4787         btrfs_i_size_write(parent_inode, parent_inode->i_size +
4788                            name_len * 2);
4789         inode_inc_iversion(parent_inode);
4790         parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4791         ret = btrfs_update_inode(trans, root, parent_inode);
4792         if (ret)
4793                 btrfs_abort_transaction(trans, root, ret);
4794         return ret;
4795
4796 fail_dir_item:
4797         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4798                 u64 local_index;
4799                 int err;
4800                 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4801                                  key.objectid, root->root_key.objectid,
4802                                  parent_ino, &local_index, name, name_len);
4803
4804         } else if (add_backref) {
4805                 u64 local_index;
4806                 int err;
4807
4808                 err = btrfs_del_inode_ref(trans, root, name, name_len,
4809                                           ino, parent_ino, &local_index);
4810         }
4811         return ret;
4812 }
4813
4814 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4815                             struct inode *dir, struct dentry *dentry,
4816                             struct inode *inode, int backref, u64 index)
4817 {
4818         int err = btrfs_add_link(trans, dir, inode,
4819                                  dentry->d_name.name, dentry->d_name.len,
4820                                  backref, index);
4821         if (err > 0)
4822                 err = -EEXIST;
4823         return err;
4824 }
4825
4826 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4827                         umode_t mode, dev_t rdev)
4828 {
4829         struct btrfs_trans_handle *trans;
4830         struct btrfs_root *root = BTRFS_I(dir)->root;
4831         struct inode *inode = NULL;
4832         int err;
4833         int drop_inode = 0;
4834         u64 objectid;
4835         unsigned long nr = 0;
4836         u64 index = 0;
4837
4838         if (!new_valid_dev(rdev))
4839                 return -EINVAL;
4840
4841         /*
4842          * 2 for inode item and ref
4843          * 2 for dir items
4844          * 1 for xattr if selinux is on
4845          */
4846         trans = btrfs_start_transaction(root, 5);
4847         if (IS_ERR(trans))
4848                 return PTR_ERR(trans);
4849
4850         err = btrfs_find_free_ino(root, &objectid);
4851         if (err)
4852                 goto out_unlock;
4853
4854         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4855                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4856                                 mode, &index);
4857         if (IS_ERR(inode)) {
4858                 err = PTR_ERR(inode);
4859                 goto out_unlock;
4860         }
4861
4862         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4863         if (err) {
4864                 drop_inode = 1;
4865                 goto out_unlock;
4866         }
4867
4868         /*
4869         * If the active LSM wants to access the inode during
4870         * d_instantiate it needs these. Smack checks to see
4871         * if the filesystem supports xattrs by looking at the
4872         * ops vector.
4873         */
4874
4875         inode->i_op = &btrfs_special_inode_operations;
4876         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4877         if (err)
4878                 drop_inode = 1;
4879         else {
4880                 init_special_inode(inode, inode->i_mode, rdev);
4881                 btrfs_update_inode(trans, root, inode);
4882                 d_instantiate(dentry, inode);
4883         }
4884 out_unlock:
4885         nr = trans->blocks_used;
4886         btrfs_end_transaction(trans, root);
4887         btrfs_btree_balance_dirty(root, nr);
4888         if (drop_inode) {
4889                 inode_dec_link_count(inode);
4890                 iput(inode);
4891         }
4892         return err;
4893 }
4894
4895 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4896                         umode_t mode, bool excl)
4897 {
4898         struct btrfs_trans_handle *trans;
4899         struct btrfs_root *root = BTRFS_I(dir)->root;
4900         struct inode *inode = NULL;
4901         int drop_inode = 0;
4902         int err;
4903         unsigned long nr = 0;
4904         u64 objectid;
4905         u64 index = 0;
4906
4907         /*
4908          * 2 for inode item and ref
4909          * 2 for dir items
4910          * 1 for xattr if selinux is on
4911          */
4912         trans = btrfs_start_transaction(root, 5);
4913         if (IS_ERR(trans))
4914                 return PTR_ERR(trans);
4915
4916         err = btrfs_find_free_ino(root, &objectid);
4917         if (err)
4918                 goto out_unlock;
4919
4920         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4921                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4922                                 mode, &index);
4923         if (IS_ERR(inode)) {
4924                 err = PTR_ERR(inode);
4925                 goto out_unlock;
4926         }
4927
4928         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4929         if (err) {
4930                 drop_inode = 1;
4931                 goto out_unlock;
4932         }
4933
4934         /*
4935         * If the active LSM wants to access the inode during
4936         * d_instantiate it needs these. Smack checks to see
4937         * if the filesystem supports xattrs by looking at the
4938         * ops vector.
4939         */
4940         inode->i_fop = &btrfs_file_operations;
4941         inode->i_op = &btrfs_file_inode_operations;
4942
4943         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4944         if (err)
4945                 drop_inode = 1;
4946         else {
4947                 inode->i_mapping->a_ops = &btrfs_aops;
4948                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4949                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4950                 d_instantiate(dentry, inode);
4951         }
4952 out_unlock:
4953         nr = trans->blocks_used;
4954         btrfs_end_transaction(trans, root);
4955         if (drop_inode) {
4956                 inode_dec_link_count(inode);
4957                 iput(inode);
4958         }
4959         btrfs_btree_balance_dirty(root, nr);
4960         return err;
4961 }
4962
4963 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4964                       struct dentry *dentry)
4965 {
4966         struct btrfs_trans_handle *trans;
4967         struct btrfs_root *root = BTRFS_I(dir)->root;
4968         struct inode *inode = old_dentry->d_inode;
4969         u64 index;
4970         unsigned long nr = 0;
4971         int err;
4972         int drop_inode = 0;
4973
4974         /* do not allow sys_link's with other subvols of the same device */
4975         if (root->objectid != BTRFS_I(inode)->root->objectid)
4976                 return -EXDEV;
4977
4978         if (inode->i_nlink == ~0U)
4979                 return -EMLINK;
4980
4981         err = btrfs_set_inode_index(dir, &index);
4982         if (err)
4983                 goto fail;
4984
4985         /*
4986          * 2 items for inode and inode ref
4987          * 2 items for dir items
4988          * 1 item for parent inode
4989          */
4990         trans = btrfs_start_transaction(root, 5);
4991         if (IS_ERR(trans)) {
4992                 err = PTR_ERR(trans);
4993                 goto fail;
4994         }
4995
4996         btrfs_inc_nlink(inode);
4997         inode_inc_iversion(inode);
4998         inode->i_ctime = CURRENT_TIME;
4999         ihold(inode);
5000
5001         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5002
5003         if (err) {
5004                 drop_inode = 1;
5005         } else {
5006                 struct dentry *parent = dentry->d_parent;
5007                 err = btrfs_update_inode(trans, root, inode);
5008                 if (err)
5009                         goto fail;
5010                 d_instantiate(dentry, inode);
5011                 btrfs_log_new_name(trans, inode, NULL, parent);
5012         }
5013
5014         nr = trans->blocks_used;
5015         btrfs_end_transaction(trans, root);
5016 fail:
5017         if (drop_inode) {
5018                 inode_dec_link_count(inode);
5019                 iput(inode);
5020         }
5021         btrfs_btree_balance_dirty(root, nr);
5022         return err;
5023 }
5024
5025 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5026 {
5027         struct inode *inode = NULL;
5028         struct btrfs_trans_handle *trans;
5029         struct btrfs_root *root = BTRFS_I(dir)->root;
5030         int err = 0;
5031         int drop_on_err = 0;
5032         u64 objectid = 0;
5033         u64 index = 0;
5034         unsigned long nr = 1;
5035
5036         /*
5037          * 2 items for inode and ref
5038          * 2 items for dir items
5039          * 1 for xattr if selinux is on
5040          */
5041         trans = btrfs_start_transaction(root, 5);
5042         if (IS_ERR(trans))
5043                 return PTR_ERR(trans);
5044
5045         err = btrfs_find_free_ino(root, &objectid);
5046         if (err)
5047                 goto out_fail;
5048
5049         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5050                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5051                                 S_IFDIR | mode, &index);
5052         if (IS_ERR(inode)) {
5053                 err = PTR_ERR(inode);
5054                 goto out_fail;
5055         }
5056
5057         drop_on_err = 1;
5058
5059         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5060         if (err)
5061                 goto out_fail;
5062
5063         inode->i_op = &btrfs_dir_inode_operations;
5064         inode->i_fop = &btrfs_dir_file_operations;
5065
5066         btrfs_i_size_write(inode, 0);
5067         err = btrfs_update_inode(trans, root, inode);
5068         if (err)
5069                 goto out_fail;
5070
5071         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5072                              dentry->d_name.len, 0, index);
5073         if (err)
5074                 goto out_fail;
5075
5076         d_instantiate(dentry, inode);
5077         drop_on_err = 0;
5078
5079 out_fail:
5080         nr = trans->blocks_used;
5081         btrfs_end_transaction(trans, root);
5082         if (drop_on_err)
5083                 iput(inode);
5084         btrfs_btree_balance_dirty(root, nr);
5085         return err;
5086 }
5087
5088 /* helper for btfs_get_extent.  Given an existing extent in the tree,
5089  * and an extent that you want to insert, deal with overlap and insert
5090  * the new extent into the tree.
5091  */
5092 static int merge_extent_mapping(struct extent_map_tree *em_tree,
5093                                 struct extent_map *existing,
5094                                 struct extent_map *em,
5095                                 u64 map_start, u64 map_len)
5096 {
5097         u64 start_diff;
5098
5099         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5100         start_diff = map_start - em->start;
5101         em->start = map_start;
5102         em->len = map_len;
5103         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5104             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
5105                 em->block_start += start_diff;
5106                 em->block_len -= start_diff;
5107         }
5108         return add_extent_mapping(em_tree, em);
5109 }
5110
5111 static noinline int uncompress_inline(struct btrfs_path *path,
5112                                       struct inode *inode, struct page *page,
5113                                       size_t pg_offset, u64 extent_offset,
5114                                       struct btrfs_file_extent_item *item)
5115 {
5116         int ret;
5117         struct extent_buffer *leaf = path->nodes[0];
5118         char *tmp;
5119         size_t max_size;
5120         unsigned long inline_size;
5121         unsigned long ptr;
5122         int compress_type;
5123
5124         WARN_ON(pg_offset != 0);
5125         compress_type = btrfs_file_extent_compression(leaf, item);
5126         max_size = btrfs_file_extent_ram_bytes(leaf, item);
5127         inline_size = btrfs_file_extent_inline_item_len(leaf,
5128                                         btrfs_item_nr(leaf, path->slots[0]));
5129         tmp = kmalloc(inline_size, GFP_NOFS);
5130         if (!tmp)
5131                 return -ENOMEM;
5132         ptr = btrfs_file_extent_inline_start(item);
5133
5134         read_extent_buffer(leaf, tmp, ptr, inline_size);
5135
5136         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5137         ret = btrfs_decompress(compress_type, tmp, page,
5138                                extent_offset, inline_size, max_size);
5139         if (ret) {
5140                 char *kaddr = kmap_atomic(page);
5141                 unsigned long copy_size = min_t(u64,
5142                                   PAGE_CACHE_SIZE - pg_offset,
5143                                   max_size - extent_offset);
5144                 memset(kaddr + pg_offset, 0, copy_size);
5145                 kunmap_atomic(kaddr);
5146         }
5147         kfree(tmp);
5148         return 0;
5149 }
5150
5151 /*
5152  * a bit scary, this does extent mapping from logical file offset to the disk.
5153  * the ugly parts come from merging extents from the disk with the in-ram
5154  * representation.  This gets more complex because of the data=ordered code,
5155  * where the in-ram extents might be locked pending data=ordered completion.
5156  *
5157  * This also copies inline extents directly into the page.
5158  */
5159
5160 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5161                                     size_t pg_offset, u64 start, u64 len,
5162                                     int create)
5163 {
5164         int ret;
5165         int err = 0;
5166         u64 bytenr;
5167         u64 extent_start = 0;
5168         u64 extent_end = 0;
5169         u64 objectid = btrfs_ino(inode);
5170         u32 found_type;
5171         struct btrfs_path *path = NULL;
5172         struct btrfs_root *root = BTRFS_I(inode)->root;
5173         struct btrfs_file_extent_item *item;
5174         struct extent_buffer *leaf;
5175         struct btrfs_key found_key;
5176         struct extent_map *em = NULL;
5177         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5178         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5179         struct btrfs_trans_handle *trans = NULL;
5180         int compress_type;
5181
5182 again:
5183         read_lock(&em_tree->lock);
5184         em = lookup_extent_mapping(em_tree, start, len);
5185         if (em)
5186                 em->bdev = root->fs_info->fs_devices->latest_bdev;
5187         read_unlock(&em_tree->lock);
5188
5189         if (em) {
5190                 if (em->start > start || em->start + em->len <= start)
5191                         free_extent_map(em);
5192                 else if (em->block_start == EXTENT_MAP_INLINE && page)
5193                         free_extent_map(em);
5194                 else
5195                         goto out;
5196         }
5197         em = alloc_extent_map();
5198         if (!em) {
5199                 err = -ENOMEM;
5200                 goto out;
5201         }
5202         em->bdev = root->fs_info->fs_devices->latest_bdev;
5203         em->start = EXTENT_MAP_HOLE;
5204         em->orig_start = EXTENT_MAP_HOLE;
5205         em->len = (u64)-1;
5206         em->block_len = (u64)-1;
5207
5208         if (!path) {
5209                 path = btrfs_alloc_path();
5210                 if (!path) {
5211                         err = -ENOMEM;
5212                         goto out;
5213                 }
5214                 /*
5215                  * Chances are we'll be called again, so go ahead and do
5216                  * readahead
5217                  */
5218                 path->reada = 1;
5219         }
5220
5221         ret = btrfs_lookup_file_extent(trans, root, path,
5222                                        objectid, start, trans != NULL);
5223         if (ret < 0) {
5224                 err = ret;
5225                 goto out;
5226         }
5227
5228         if (ret != 0) {
5229                 if (path->slots[0] == 0)
5230                         goto not_found;
5231                 path->slots[0]--;
5232         }
5233
5234         leaf = path->nodes[0];
5235         item = btrfs_item_ptr(leaf, path->slots[0],
5236                               struct btrfs_file_extent_item);
5237         /* are we inside the extent that was found? */
5238         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5239         found_type = btrfs_key_type(&found_key);
5240         if (found_key.objectid != objectid ||
5241             found_type != BTRFS_EXTENT_DATA_KEY) {
5242                 goto not_found;
5243         }
5244
5245         found_type = btrfs_file_extent_type(leaf, item);
5246         extent_start = found_key.offset;
5247         compress_type = btrfs_file_extent_compression(leaf, item);
5248         if (found_type == BTRFS_FILE_EXTENT_REG ||
5249             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5250                 extent_end = extent_start +
5251                        btrfs_file_extent_num_bytes(leaf, item);
5252         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5253                 size_t size;
5254                 size = btrfs_file_extent_inline_len(leaf, item);
5255                 extent_end = (extent_start + size + root->sectorsize - 1) &
5256                         ~((u64)root->sectorsize - 1);
5257         }
5258
5259         if (start >= extent_end) {
5260                 path->slots[0]++;
5261                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5262                         ret = btrfs_next_leaf(root, path);
5263                         if (ret < 0) {
5264                                 err = ret;
5265                                 goto out;
5266                         }
5267                         if (ret > 0)
5268                                 goto not_found;
5269                         leaf = path->nodes[0];
5270                 }
5271                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5272                 if (found_key.objectid != objectid ||
5273                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5274                         goto not_found;
5275                 if (start + len <= found_key.offset)
5276                         goto not_found;
5277                 em->start = start;
5278                 em->len = found_key.offset - start;
5279                 goto not_found_em;
5280         }
5281
5282         if (found_type == BTRFS_FILE_EXTENT_REG ||
5283             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5284                 em->start = extent_start;
5285                 em->len = extent_end - extent_start;
5286                 em->orig_start = extent_start -
5287                                  btrfs_file_extent_offset(leaf, item);
5288                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5289                 if (bytenr == 0) {
5290                         em->block_start = EXTENT_MAP_HOLE;
5291                         goto insert;
5292                 }
5293                 if (compress_type != BTRFS_COMPRESS_NONE) {
5294                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5295                         em->compress_type = compress_type;
5296                         em->block_start = bytenr;
5297                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5298                                                                          item);
5299                 } else {
5300                         bytenr += btrfs_file_extent_offset(leaf, item);
5301                         em->block_start = bytenr;
5302                         em->block_len = em->len;
5303                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5304                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5305                 }
5306                 goto insert;
5307         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5308                 unsigned long ptr;
5309                 char *map;
5310                 size_t size;
5311                 size_t extent_offset;
5312                 size_t copy_size;
5313
5314                 em->block_start = EXTENT_MAP_INLINE;
5315                 if (!page || create) {
5316                         em->start = extent_start;
5317                         em->len = extent_end - extent_start;
5318                         goto out;
5319                 }
5320
5321                 size = btrfs_file_extent_inline_len(leaf, item);
5322                 extent_offset = page_offset(page) + pg_offset - extent_start;
5323                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5324                                 size - extent_offset);
5325                 em->start = extent_start + extent_offset;
5326                 em->len = (copy_size + root->sectorsize - 1) &
5327                         ~((u64)root->sectorsize - 1);
5328                 em->orig_start = EXTENT_MAP_INLINE;
5329                 if (compress_type) {
5330                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5331                         em->compress_type = compress_type;
5332                 }
5333                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5334                 if (create == 0 && !PageUptodate(page)) {
5335                         if (btrfs_file_extent_compression(leaf, item) !=
5336                             BTRFS_COMPRESS_NONE) {
5337                                 ret = uncompress_inline(path, inode, page,
5338                                                         pg_offset,
5339                                                         extent_offset, item);
5340                                 BUG_ON(ret); /* -ENOMEM */
5341                         } else {
5342                                 map = kmap(page);
5343                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5344                                                    copy_size);
5345                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5346                                         memset(map + pg_offset + copy_size, 0,
5347                                                PAGE_CACHE_SIZE - pg_offset -
5348                                                copy_size);
5349                                 }
5350                                 kunmap(page);
5351                         }
5352                         flush_dcache_page(page);
5353                 } else if (create && PageUptodate(page)) {
5354                         BUG();
5355                         if (!trans) {
5356                                 kunmap(page);
5357                                 free_extent_map(em);
5358                                 em = NULL;
5359
5360                                 btrfs_release_path(path);
5361                                 trans = btrfs_join_transaction(root);
5362
5363                                 if (IS_ERR(trans))
5364                                         return ERR_CAST(trans);
5365                                 goto again;
5366                         }
5367                         map = kmap(page);
5368                         write_extent_buffer(leaf, map + pg_offset, ptr,
5369                                             copy_size);
5370                         kunmap(page);
5371                         btrfs_mark_buffer_dirty(leaf);
5372                 }
5373                 set_extent_uptodate(io_tree, em->start,
5374                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
5375                 goto insert;
5376         } else {
5377                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5378                 WARN_ON(1);
5379         }
5380 not_found:
5381         em->start = start;
5382         em->len = len;
5383 not_found_em:
5384         em->block_start = EXTENT_MAP_HOLE;
5385         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5386 insert:
5387         btrfs_release_path(path);
5388         if (em->start > start || extent_map_end(em) <= start) {
5389                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5390                        "[%llu %llu]\n", (unsigned long long)em->start,
5391                        (unsigned long long)em->len,
5392                        (unsigned long long)start,
5393                        (unsigned long long)len);
5394                 err = -EIO;
5395                 goto out;
5396         }
5397
5398         err = 0;
5399         write_lock(&em_tree->lock);
5400         ret = add_extent_mapping(em_tree, em);
5401         /* it is possible that someone inserted the extent into the tree
5402          * while we had the lock dropped.  It is also possible that
5403          * an overlapping map exists in the tree
5404          */
5405         if (ret == -EEXIST) {
5406                 struct extent_map *existing;
5407
5408                 ret = 0;
5409
5410                 existing = lookup_extent_mapping(em_tree, start, len);
5411                 if (existing && (existing->start > start ||
5412                     existing->start + existing->len <= start)) {
5413                         free_extent_map(existing);
5414                         existing = NULL;
5415                 }
5416                 if (!existing) {
5417                         existing = lookup_extent_mapping(em_tree, em->start,
5418                                                          em->len);
5419                         if (existing) {
5420                                 err = merge_extent_mapping(em_tree, existing,
5421                                                            em, start,
5422                                                            root->sectorsize);
5423                                 free_extent_map(existing);
5424                                 if (err) {
5425                                         free_extent_map(em);
5426                                         em = NULL;
5427                                 }
5428                         } else {
5429                                 err = -EIO;
5430                                 free_extent_map(em);
5431                                 em = NULL;
5432                         }
5433                 } else {
5434                         free_extent_map(em);
5435                         em = existing;
5436                         err = 0;
5437                 }
5438         }
5439         write_unlock(&em_tree->lock);
5440 out:
5441
5442         trace_btrfs_get_extent(root, em);
5443
5444         if (path)
5445                 btrfs_free_path(path);
5446         if (trans) {
5447                 ret = btrfs_end_transaction(trans, root);
5448                 if (!err)
5449                         err = ret;
5450         }
5451         if (err) {
5452                 free_extent_map(em);
5453                 return ERR_PTR(err);
5454         }
5455         BUG_ON(!em); /* Error is always set */
5456         return em;
5457 }
5458
5459 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5460                                            size_t pg_offset, u64 start, u64 len,
5461                                            int create)
5462 {
5463         struct extent_map *em;
5464         struct extent_map *hole_em = NULL;
5465         u64 range_start = start;
5466         u64 end;
5467         u64 found;
5468         u64 found_end;
5469         int err = 0;
5470
5471         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5472         if (IS_ERR(em))
5473                 return em;
5474         if (em) {
5475                 /*
5476                  * if our em maps to a hole, there might
5477                  * actually be delalloc bytes behind it
5478                  */
5479                 if (em->block_start != EXTENT_MAP_HOLE)
5480                         return em;
5481                 else
5482                         hole_em = em;
5483         }
5484
5485         /* check to see if we've wrapped (len == -1 or similar) */
5486         end = start + len;
5487         if (end < start)
5488                 end = (u64)-1;
5489         else
5490                 end -= 1;
5491
5492         em = NULL;
5493
5494         /* ok, we didn't find anything, lets look for delalloc */
5495         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5496                                  end, len, EXTENT_DELALLOC, 1);
5497         found_end = range_start + found;
5498         if (found_end < range_start)
5499                 found_end = (u64)-1;
5500
5501         /*
5502          * we didn't find anything useful, return
5503          * the original results from get_extent()
5504          */
5505         if (range_start > end || found_end <= start) {
5506                 em = hole_em;
5507                 hole_em = NULL;
5508                 goto out;
5509         }
5510
5511         /* adjust the range_start to make sure it doesn't
5512          * go backwards from the start they passed in
5513          */
5514         range_start = max(start,range_start);
5515         found = found_end - range_start;
5516
5517         if (found > 0) {
5518                 u64 hole_start = start;
5519                 u64 hole_len = len;
5520
5521                 em = alloc_extent_map();
5522                 if (!em) {
5523                         err = -ENOMEM;
5524                         goto out;
5525                 }
5526                 /*
5527                  * when btrfs_get_extent can't find anything it
5528                  * returns one huge hole
5529                  *
5530                  * make sure what it found really fits our range, and
5531                  * adjust to make sure it is based on the start from
5532                  * the caller
5533                  */
5534                 if (hole_em) {
5535                         u64 calc_end = extent_map_end(hole_em);
5536
5537                         if (calc_end <= start || (hole_em->start > end)) {
5538                                 free_extent_map(hole_em);
5539                                 hole_em = NULL;
5540                         } else {
5541                                 hole_start = max(hole_em->start, start);
5542                                 hole_len = calc_end - hole_start;
5543                         }
5544                 }
5545                 em->bdev = NULL;
5546                 if (hole_em && range_start > hole_start) {
5547                         /* our hole starts before our delalloc, so we
5548                          * have to return just the parts of the hole
5549                          * that go until  the delalloc starts
5550                          */
5551                         em->len = min(hole_len,
5552                                       range_start - hole_start);
5553                         em->start = hole_start;
5554                         em->orig_start = hole_start;
5555                         /*
5556                          * don't adjust block start at all,
5557                          * it is fixed at EXTENT_MAP_HOLE
5558                          */
5559                         em->block_start = hole_em->block_start;
5560                         em->block_len = hole_len;
5561                 } else {
5562                         em->start = range_start;
5563                         em->len = found;
5564                         em->orig_start = range_start;
5565                         em->block_start = EXTENT_MAP_DELALLOC;
5566                         em->block_len = found;
5567                 }
5568         } else if (hole_em) {
5569                 return hole_em;
5570         }
5571 out:
5572
5573         free_extent_map(hole_em);
5574         if (err) {
5575                 free_extent_map(em);
5576                 return ERR_PTR(err);
5577         }
5578         return em;
5579 }
5580
5581 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5582                                                   struct extent_map *em,
5583                                                   u64 start, u64 len)
5584 {
5585         struct btrfs_root *root = BTRFS_I(inode)->root;
5586         struct btrfs_trans_handle *trans;
5587         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5588         struct btrfs_key ins;
5589         u64 alloc_hint;
5590         int ret;
5591         bool insert = false;
5592
5593         /*
5594          * Ok if the extent map we looked up is a hole and is for the exact
5595          * range we want, there is no reason to allocate a new one, however if
5596          * it is not right then we need to free this one and drop the cache for
5597          * our range.
5598          */
5599         if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5600             em->len != len) {
5601                 free_extent_map(em);
5602                 em = NULL;
5603                 insert = true;
5604                 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5605         }
5606
5607         trans = btrfs_join_transaction(root);
5608         if (IS_ERR(trans))
5609                 return ERR_CAST(trans);
5610
5611         if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5612                 btrfs_add_inode_defrag(trans, inode);
5613
5614         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5615
5616         alloc_hint = get_extent_allocation_hint(inode, start, len);
5617         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5618                                    alloc_hint, &ins, 1);
5619         if (ret) {
5620                 em = ERR_PTR(ret);
5621                 goto out;
5622         }
5623
5624         if (!em) {
5625                 em = alloc_extent_map();
5626                 if (!em) {
5627                         em = ERR_PTR(-ENOMEM);
5628                         goto out;
5629                 }
5630         }
5631
5632         em->start = start;
5633         em->orig_start = em->start;
5634         em->len = ins.offset;
5635
5636         em->block_start = ins.objectid;
5637         em->block_len = ins.offset;
5638         em->bdev = root->fs_info->fs_devices->latest_bdev;
5639
5640         /*
5641          * We need to do this because if we're using the original em we searched
5642          * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5643          */
5644         em->flags = 0;
5645         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5646
5647         while (insert) {
5648                 write_lock(&em_tree->lock);
5649                 ret = add_extent_mapping(em_tree, em);
5650                 write_unlock(&em_tree->lock);
5651                 if (ret != -EEXIST)
5652                         break;
5653                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5654         }
5655
5656         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5657                                            ins.offset, ins.offset, 0);
5658         if (ret) {
5659                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5660                 em = ERR_PTR(ret);
5661         }
5662 out:
5663         btrfs_end_transaction(trans, root);
5664         return em;
5665 }
5666
5667 /*
5668  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5669  * block must be cow'd
5670  */
5671 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5672                                       struct inode *inode, u64 offset, u64 len)
5673 {
5674         struct btrfs_path *path;
5675         int ret;
5676         struct extent_buffer *leaf;
5677         struct btrfs_root *root = BTRFS_I(inode)->root;
5678         struct btrfs_file_extent_item *fi;
5679         struct btrfs_key key;
5680         u64 disk_bytenr;
5681         u64 backref_offset;
5682         u64 extent_end;
5683         u64 num_bytes;
5684         int slot;
5685         int found_type;
5686
5687         path = btrfs_alloc_path();
5688         if (!path)
5689                 return -ENOMEM;
5690
5691         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5692                                        offset, 0);
5693         if (ret < 0)
5694                 goto out;
5695
5696         slot = path->slots[0];
5697         if (ret == 1) {
5698                 if (slot == 0) {
5699                         /* can't find the item, must cow */
5700                         ret = 0;
5701                         goto out;
5702                 }
5703                 slot--;
5704         }
5705         ret = 0;
5706         leaf = path->nodes[0];
5707         btrfs_item_key_to_cpu(leaf, &key, slot);
5708         if (key.objectid != btrfs_ino(inode) ||
5709             key.type != BTRFS_EXTENT_DATA_KEY) {
5710                 /* not our file or wrong item type, must cow */
5711                 goto out;
5712         }
5713
5714         if (key.offset > offset) {
5715                 /* Wrong offset, must cow */
5716                 goto out;
5717         }
5718
5719         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5720         found_type = btrfs_file_extent_type(leaf, fi);
5721         if (found_type != BTRFS_FILE_EXTENT_REG &&
5722             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5723                 /* not a regular extent, must cow */
5724                 goto out;
5725         }
5726         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5727         backref_offset = btrfs_file_extent_offset(leaf, fi);
5728
5729         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5730         if (extent_end < offset + len) {
5731                 /* extent doesn't include our full range, must cow */
5732                 goto out;
5733         }
5734
5735         if (btrfs_extent_readonly(root, disk_bytenr))
5736                 goto out;
5737
5738         /*
5739          * look for other files referencing this extent, if we
5740          * find any we must cow
5741          */
5742         if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5743                                   key.offset - backref_offset, disk_bytenr))
5744                 goto out;
5745
5746         /*
5747          * adjust disk_bytenr and num_bytes to cover just the bytes
5748          * in this extent we are about to write.  If there
5749          * are any csums in that range we have to cow in order
5750          * to keep the csums correct
5751          */
5752         disk_bytenr += backref_offset;
5753         disk_bytenr += offset - key.offset;
5754         num_bytes = min(offset + len, extent_end) - offset;
5755         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5756                                 goto out;
5757         /*
5758          * all of the above have passed, it is safe to overwrite this extent
5759          * without cow
5760          */
5761         ret = 1;
5762 out:
5763         btrfs_free_path(path);
5764         return ret;
5765 }
5766
5767 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5768                                    struct buffer_head *bh_result, int create)
5769 {
5770         struct extent_map *em;
5771         struct btrfs_root *root = BTRFS_I(inode)->root;
5772         u64 start = iblock << inode->i_blkbits;
5773         u64 len = bh_result->b_size;
5774         struct btrfs_trans_handle *trans;
5775
5776         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5777         if (IS_ERR(em))
5778                 return PTR_ERR(em);
5779
5780         /*
5781          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5782          * io.  INLINE is special, and we could probably kludge it in here, but
5783          * it's still buffered so for safety lets just fall back to the generic
5784          * buffered path.
5785          *
5786          * For COMPRESSED we _have_ to read the entire extent in so we can
5787          * decompress it, so there will be buffering required no matter what we
5788          * do, so go ahead and fallback to buffered.
5789          *
5790          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5791          * to buffered IO.  Don't blame me, this is the price we pay for using
5792          * the generic code.
5793          */
5794         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5795             em->block_start == EXTENT_MAP_INLINE) {
5796                 free_extent_map(em);
5797                 return -ENOTBLK;
5798         }
5799
5800         /* Just a good old fashioned hole, return */
5801         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5802                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5803                 free_extent_map(em);
5804                 /* DIO will do one hole at a time, so just unlock a sector */
5805                 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5806                               start + root->sectorsize - 1);
5807                 return 0;
5808         }
5809
5810         /*
5811          * We don't allocate a new extent in the following cases
5812          *
5813          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
5814          * existing extent.
5815          * 2) The extent is marked as PREALLOC.  We're good to go here and can
5816          * just use the extent.
5817          *
5818          */
5819         if (!create) {
5820                 len = em->len - (start - em->start);
5821                 goto map;
5822         }
5823
5824         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5825             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5826              em->block_start != EXTENT_MAP_HOLE)) {
5827                 int type;
5828                 int ret;
5829                 u64 block_start;
5830
5831                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5832                         type = BTRFS_ORDERED_PREALLOC;
5833                 else
5834                         type = BTRFS_ORDERED_NOCOW;
5835                 len = min(len, em->len - (start - em->start));
5836                 block_start = em->block_start + (start - em->start);
5837
5838                 /*
5839                  * we're not going to log anything, but we do need
5840                  * to make sure the current transaction stays open
5841                  * while we look for nocow cross refs
5842                  */
5843                 trans = btrfs_join_transaction(root);
5844                 if (IS_ERR(trans))
5845                         goto must_cow;
5846
5847                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5848                         ret = btrfs_add_ordered_extent_dio(inode, start,
5849                                            block_start, len, len, type);
5850                         btrfs_end_transaction(trans, root);
5851                         if (ret) {
5852                                 free_extent_map(em);
5853                                 return ret;
5854                         }
5855                         goto unlock;
5856                 }
5857                 btrfs_end_transaction(trans, root);
5858         }
5859 must_cow:
5860         /*
5861          * this will cow the extent, reset the len in case we changed
5862          * it above
5863          */
5864         len = bh_result->b_size;
5865         em = btrfs_new_extent_direct(inode, em, start, len);
5866         if (IS_ERR(em))
5867                 return PTR_ERR(em);
5868         len = min(len, em->len - (start - em->start));
5869 unlock:
5870         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5871                           EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5872                           0, NULL, GFP_NOFS);
5873 map:
5874         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5875                 inode->i_blkbits;
5876         bh_result->b_size = len;
5877         bh_result->b_bdev = em->bdev;
5878         set_buffer_mapped(bh_result);
5879         if (create) {
5880                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5881                         set_buffer_new(bh_result);
5882
5883                 /*
5884                  * Need to update the i_size under the extent lock so buffered
5885                  * readers will get the updated i_size when we unlock.
5886                  */
5887                 if (start + len > i_size_read(inode))
5888                         i_size_write(inode, start + len);
5889         }
5890
5891         free_extent_map(em);
5892
5893         return 0;
5894 }
5895
5896 struct btrfs_dio_private {
5897         struct inode *inode;
5898         u64 logical_offset;
5899         u64 disk_bytenr;
5900         u64 bytes;
5901         u32 *csums;
5902         void *private;
5903
5904         /* number of bios pending for this dio */
5905         atomic_t pending_bios;
5906
5907         /* IO errors */
5908         int errors;
5909
5910         struct bio *orig_bio;
5911 };
5912
5913 static void btrfs_endio_direct_read(struct bio *bio, int err)
5914 {
5915         struct btrfs_dio_private *dip = bio->bi_private;
5916         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5917         struct bio_vec *bvec = bio->bi_io_vec;
5918         struct inode *inode = dip->inode;
5919         struct btrfs_root *root = BTRFS_I(inode)->root;
5920         u64 start;
5921         u32 *private = dip->csums;
5922
5923         start = dip->logical_offset;
5924         do {
5925                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5926                         struct page *page = bvec->bv_page;
5927                         char *kaddr;
5928                         u32 csum = ~(u32)0;
5929                         unsigned long flags;
5930
5931                         local_irq_save(flags);
5932                         kaddr = kmap_atomic(page);
5933                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5934                                                csum, bvec->bv_len);
5935                         btrfs_csum_final(csum, (char *)&csum);
5936                         kunmap_atomic(kaddr);
5937                         local_irq_restore(flags);
5938
5939                         flush_dcache_page(bvec->bv_page);
5940                         if (csum != *private) {
5941                                 printk(KERN_ERR "btrfs csum failed ino %llu off"
5942                                       " %llu csum %u private %u\n",
5943                                       (unsigned long long)btrfs_ino(inode),
5944                                       (unsigned long long)start,
5945                                       csum, *private);
5946                                 err = -EIO;
5947                         }
5948                 }
5949
5950                 start += bvec->bv_len;
5951                 private++;
5952                 bvec++;
5953         } while (bvec <= bvec_end);
5954
5955         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5956                       dip->logical_offset + dip->bytes - 1);
5957         bio->bi_private = dip->private;
5958
5959         kfree(dip->csums);
5960         kfree(dip);
5961
5962         /* If we had a csum failure make sure to clear the uptodate flag */
5963         if (err)
5964                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5965         dio_end_io(bio, err);
5966 }
5967
5968 static void btrfs_endio_direct_write(struct bio *bio, int err)
5969 {
5970         struct btrfs_dio_private *dip = bio->bi_private;
5971         struct inode *inode = dip->inode;
5972         struct btrfs_root *root = BTRFS_I(inode)->root;
5973         struct btrfs_ordered_extent *ordered = NULL;
5974         u64 ordered_offset = dip->logical_offset;
5975         u64 ordered_bytes = dip->bytes;
5976         int ret;
5977
5978         if (err)
5979                 goto out_done;
5980 again:
5981         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5982                                                    &ordered_offset,
5983                                                    ordered_bytes, !err);
5984         if (!ret)
5985                 goto out_test;
5986
5987         ordered->work.func = finish_ordered_fn;
5988         ordered->work.flags = 0;
5989         btrfs_queue_worker(&root->fs_info->endio_write_workers,
5990                            &ordered->work);
5991 out_test:
5992         /*
5993          * our bio might span multiple ordered extents.  If we haven't
5994          * completed the accounting for the whole dio, go back and try again
5995          */
5996         if (ordered_offset < dip->logical_offset + dip->bytes) {
5997                 ordered_bytes = dip->logical_offset + dip->bytes -
5998                         ordered_offset;
5999                 ordered = NULL;
6000                 goto again;
6001         }
6002 out_done:
6003         bio->bi_private = dip->private;
6004
6005         kfree(dip);
6006
6007         /* If we had an error make sure to clear the uptodate flag */
6008         if (err)
6009                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
6010         dio_end_io(bio, err);
6011 }
6012
6013 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
6014                                     struct bio *bio, int mirror_num,
6015                                     unsigned long bio_flags, u64 offset)
6016 {
6017         int ret;
6018         struct btrfs_root *root = BTRFS_I(inode)->root;
6019         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
6020         BUG_ON(ret); /* -ENOMEM */
6021         return 0;
6022 }
6023
6024 static void btrfs_end_dio_bio(struct bio *bio, int err)
6025 {
6026         struct btrfs_dio_private *dip = bio->bi_private;
6027
6028         if (err) {
6029                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
6030                       "sector %#Lx len %u err no %d\n",
6031                       (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
6032                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
6033                 dip->errors = 1;
6034
6035                 /*
6036                  * before atomic variable goto zero, we must make sure
6037                  * dip->errors is perceived to be set.
6038                  */
6039                 smp_mb__before_atomic_dec();
6040         }
6041
6042         /* if there are more bios still pending for this dio, just exit */
6043         if (!atomic_dec_and_test(&dip->pending_bios))
6044                 goto out;
6045
6046         if (dip->errors)
6047                 bio_io_error(dip->orig_bio);
6048         else {
6049                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
6050                 bio_endio(dip->orig_bio, 0);
6051         }
6052 out:
6053         bio_put(bio);
6054 }
6055
6056 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
6057                                        u64 first_sector, gfp_t gfp_flags)
6058 {
6059         int nr_vecs = bio_get_nr_vecs(bdev);
6060         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
6061 }
6062
6063 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
6064                                          int rw, u64 file_offset, int skip_sum,
6065                                          u32 *csums, int async_submit)
6066 {
6067         int write = rw & REQ_WRITE;
6068         struct btrfs_root *root = BTRFS_I(inode)->root;
6069         int ret;
6070
6071         bio_get(bio);
6072
6073         if (!write) {
6074                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
6075                 if (ret)
6076                         goto err;
6077         }
6078
6079         if (skip_sum)
6080                 goto map;
6081
6082         if (write && async_submit) {
6083                 ret = btrfs_wq_submit_bio(root->fs_info,
6084                                    inode, rw, bio, 0, 0,
6085                                    file_offset,
6086                                    __btrfs_submit_bio_start_direct_io,
6087                                    __btrfs_submit_bio_done);
6088                 goto err;
6089         } else if (write) {
6090                 /*
6091                  * If we aren't doing async submit, calculate the csum of the
6092                  * bio now.
6093                  */
6094                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
6095                 if (ret)
6096                         goto err;
6097         } else if (!skip_sum) {
6098                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
6099                                           file_offset, csums);
6100                 if (ret)
6101                         goto err;
6102         }
6103
6104 map:
6105         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
6106 err:
6107         bio_put(bio);
6108         return ret;
6109 }
6110
6111 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
6112                                     int skip_sum)
6113 {
6114         struct inode *inode = dip->inode;
6115         struct btrfs_root *root = BTRFS_I(inode)->root;
6116         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6117         struct bio *bio;
6118         struct bio *orig_bio = dip->orig_bio;
6119         struct bio_vec *bvec = orig_bio->bi_io_vec;
6120         u64 start_sector = orig_bio->bi_sector;
6121         u64 file_offset = dip->logical_offset;
6122         u64 submit_len = 0;
6123         u64 map_length;
6124         int nr_pages = 0;
6125         u32 *csums = dip->csums;
6126         int ret = 0;
6127         int async_submit = 0;
6128         int write = rw & REQ_WRITE;
6129
6130         map_length = orig_bio->bi_size;
6131         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6132                               &map_length, NULL, 0);
6133         if (ret) {
6134                 bio_put(orig_bio);
6135                 return -EIO;
6136         }
6137
6138         if (map_length >= orig_bio->bi_size) {
6139                 bio = orig_bio;
6140                 goto submit;
6141         }
6142
6143         async_submit = 1;
6144         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
6145         if (!bio)
6146                 return -ENOMEM;
6147         bio->bi_private = dip;
6148         bio->bi_end_io = btrfs_end_dio_bio;
6149         atomic_inc(&dip->pending_bios);
6150
6151         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6152                 if (unlikely(map_length < submit_len + bvec->bv_len ||
6153                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6154                                  bvec->bv_offset) < bvec->bv_len)) {
6155                         /*
6156                          * inc the count before we submit the bio so
6157                          * we know the end IO handler won't happen before
6158                          * we inc the count. Otherwise, the dip might get freed
6159                          * before we're done setting it up
6160                          */
6161                         atomic_inc(&dip->pending_bios);
6162                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
6163                                                      file_offset, skip_sum,
6164                                                      csums, async_submit);
6165                         if (ret) {
6166                                 bio_put(bio);
6167                                 atomic_dec(&dip->pending_bios);
6168                                 goto out_err;
6169                         }
6170
6171                         /* Write's use the ordered csums */
6172                         if (!write && !skip_sum)
6173                                 csums = csums + nr_pages;
6174                         start_sector += submit_len >> 9;
6175                         file_offset += submit_len;
6176
6177                         submit_len = 0;
6178                         nr_pages = 0;
6179
6180                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6181                                                   start_sector, GFP_NOFS);
6182                         if (!bio)
6183                                 goto out_err;
6184                         bio->bi_private = dip;
6185                         bio->bi_end_io = btrfs_end_dio_bio;
6186
6187                         map_length = orig_bio->bi_size;
6188                         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6189                                               &map_length, NULL, 0);
6190                         if (ret) {
6191                                 bio_put(bio);
6192                                 goto out_err;
6193                         }
6194                 } else {
6195                         submit_len += bvec->bv_len;
6196                         nr_pages ++;
6197                         bvec++;
6198                 }
6199         }
6200
6201 submit:
6202         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6203                                      csums, async_submit);
6204         if (!ret)
6205                 return 0;
6206
6207         bio_put(bio);
6208 out_err:
6209         dip->errors = 1;
6210         /*
6211          * before atomic variable goto zero, we must
6212          * make sure dip->errors is perceived to be set.
6213          */
6214         smp_mb__before_atomic_dec();
6215         if (atomic_dec_and_test(&dip->pending_bios))
6216                 bio_io_error(dip->orig_bio);
6217
6218         /* bio_end_io() will handle error, so we needn't return it */
6219         return 0;
6220 }
6221
6222 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6223                                 loff_t file_offset)
6224 {
6225         struct btrfs_root *root = BTRFS_I(inode)->root;
6226         struct btrfs_dio_private *dip;
6227         struct bio_vec *bvec = bio->bi_io_vec;
6228         int skip_sum;
6229         int write = rw & REQ_WRITE;
6230         int ret = 0;
6231
6232         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6233
6234         dip = kmalloc(sizeof(*dip), GFP_NOFS);
6235         if (!dip) {
6236                 ret = -ENOMEM;
6237                 goto free_ordered;
6238         }
6239         dip->csums = NULL;
6240
6241         /* Write's use the ordered csum stuff, so we don't need dip->csums */
6242         if (!write && !skip_sum) {
6243                 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6244                 if (!dip->csums) {
6245                         kfree(dip);
6246                         ret = -ENOMEM;
6247                         goto free_ordered;
6248                 }
6249         }
6250
6251         dip->private = bio->bi_private;
6252         dip->inode = inode;
6253         dip->logical_offset = file_offset;
6254
6255         dip->bytes = 0;
6256         do {
6257                 dip->bytes += bvec->bv_len;
6258                 bvec++;
6259         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6260
6261         dip->disk_bytenr = (u64)bio->bi_sector << 9;
6262         bio->bi_private = dip;
6263         dip->errors = 0;
6264         dip->orig_bio = bio;
6265         atomic_set(&dip->pending_bios, 0);
6266
6267         if (write)
6268                 bio->bi_end_io = btrfs_endio_direct_write;
6269         else
6270                 bio->bi_end_io = btrfs_endio_direct_read;
6271
6272         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6273         if (!ret)
6274                 return;
6275 free_ordered:
6276         /*
6277          * If this is a write, we need to clean up the reserved space and kill
6278          * the ordered extent.
6279          */
6280         if (write) {
6281                 struct btrfs_ordered_extent *ordered;
6282                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6283                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6284                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6285                         btrfs_free_reserved_extent(root, ordered->start,
6286                                                    ordered->disk_len);
6287                 btrfs_put_ordered_extent(ordered);
6288                 btrfs_put_ordered_extent(ordered);
6289         }
6290         bio_endio(bio, ret);
6291 }
6292
6293 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6294                         const struct iovec *iov, loff_t offset,
6295                         unsigned long nr_segs)
6296 {
6297         int seg;
6298         int i;
6299         size_t size;
6300         unsigned long addr;
6301         unsigned blocksize_mask = root->sectorsize - 1;
6302         ssize_t retval = -EINVAL;
6303         loff_t end = offset;
6304
6305         if (offset & blocksize_mask)
6306                 goto out;
6307
6308         /* Check the memory alignment.  Blocks cannot straddle pages */
6309         for (seg = 0; seg < nr_segs; seg++) {
6310                 addr = (unsigned long)iov[seg].iov_base;
6311                 size = iov[seg].iov_len;
6312                 end += size;
6313                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6314                         goto out;
6315
6316                 /* If this is a write we don't need to check anymore */
6317                 if (rw & WRITE)
6318                         continue;
6319
6320                 /*
6321                  * Check to make sure we don't have duplicate iov_base's in this
6322                  * iovec, if so return EINVAL, otherwise we'll get csum errors
6323                  * when reading back.
6324                  */
6325                 for (i = seg + 1; i < nr_segs; i++) {
6326                         if (iov[seg].iov_base == iov[i].iov_base)
6327                                 goto out;
6328                 }
6329         }
6330         retval = 0;
6331 out:
6332         return retval;
6333 }
6334 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6335                         const struct iovec *iov, loff_t offset,
6336                         unsigned long nr_segs)
6337 {
6338         struct file *file = iocb->ki_filp;
6339         struct inode *inode = file->f_mapping->host;
6340         struct btrfs_ordered_extent *ordered;
6341         struct extent_state *cached_state = NULL;
6342         u64 lockstart, lockend;
6343         ssize_t ret;
6344         int writing = rw & WRITE;
6345         int write_bits = 0;
6346         size_t count = iov_length(iov, nr_segs);
6347
6348         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6349                             offset, nr_segs)) {
6350                 return 0;
6351         }
6352
6353         lockstart = offset;
6354         lockend = offset + count - 1;
6355
6356         if (writing) {
6357                 ret = btrfs_delalloc_reserve_space(inode, count);
6358                 if (ret)
6359                         goto out;
6360         }
6361
6362         while (1) {
6363                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6364                                  0, &cached_state);
6365                 /*
6366                  * We're concerned with the entire range that we're going to be
6367                  * doing DIO to, so we need to make sure theres no ordered
6368                  * extents in this range.
6369                  */
6370                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6371                                                      lockend - lockstart + 1);
6372
6373                 /*
6374                  * We need to make sure there are no buffered pages in this
6375                  * range either, we could have raced between the invalidate in
6376                  * generic_file_direct_write and locking the extent.  The
6377                  * invalidate needs to happen so that reads after a write do not
6378                  * get stale data.
6379                  */
6380                 if (!ordered && (!writing ||
6381                     !test_range_bit(&BTRFS_I(inode)->io_tree,
6382                                     lockstart, lockend, EXTENT_UPTODATE, 0,
6383                                     cached_state)))
6384                         break;
6385
6386                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6387                                      &cached_state, GFP_NOFS);
6388
6389                 if (ordered) {
6390                         btrfs_start_ordered_extent(inode, ordered, 1);
6391                         btrfs_put_ordered_extent(ordered);
6392                 } else {
6393                         /* Screw you mmap */
6394                         ret = filemap_write_and_wait_range(file->f_mapping,
6395                                                            lockstart,
6396                                                            lockend);
6397                         if (ret)
6398                                 goto out;
6399
6400                         /*
6401                          * If we found a page that couldn't be invalidated just
6402                          * fall back to buffered.
6403                          */
6404                         ret = invalidate_inode_pages2_range(file->f_mapping,
6405                                         lockstart >> PAGE_CACHE_SHIFT,
6406                                         lockend >> PAGE_CACHE_SHIFT);
6407                         if (ret) {
6408                                 if (ret == -EBUSY)
6409                                         ret = 0;
6410                                 goto out;
6411                         }
6412                 }
6413
6414                 cond_resched();
6415         }
6416
6417         /*
6418          * we don't use btrfs_set_extent_delalloc because we don't want
6419          * the dirty or uptodate bits
6420          */
6421         if (writing) {
6422                 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6423                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6424                                      EXTENT_DELALLOC, NULL, &cached_state,
6425                                      GFP_NOFS);
6426                 if (ret) {
6427                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6428                                          lockend, EXTENT_LOCKED | write_bits,
6429                                          1, 0, &cached_state, GFP_NOFS);
6430                         goto out;
6431                 }
6432         }
6433
6434         free_extent_state(cached_state);
6435         cached_state = NULL;
6436
6437         ret = __blockdev_direct_IO(rw, iocb, inode,
6438                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6439                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6440                    btrfs_submit_direct, 0);
6441
6442         if (ret < 0 && ret != -EIOCBQUEUED) {
6443                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6444                               offset + iov_length(iov, nr_segs) - 1,
6445                               EXTENT_LOCKED | write_bits, 1, 0,
6446                               &cached_state, GFP_NOFS);
6447         } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6448                 /*
6449                  * We're falling back to buffered, unlock the section we didn't
6450                  * do IO on.
6451                  */
6452                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6453                               offset + iov_length(iov, nr_segs) - 1,
6454                               EXTENT_LOCKED | write_bits, 1, 0,
6455                               &cached_state, GFP_NOFS);
6456         }
6457 out:
6458         free_extent_state(cached_state);
6459         return ret;
6460 }
6461
6462 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6463                 __u64 start, __u64 len)
6464 {
6465         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6466 }
6467
6468 int btrfs_readpage(struct file *file, struct page *page)
6469 {
6470         struct extent_io_tree *tree;
6471         tree = &BTRFS_I(page->mapping->host)->io_tree;
6472         return extent_read_full_page(tree, page, btrfs_get_extent, 0);
6473 }
6474
6475 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6476 {
6477         struct extent_io_tree *tree;
6478
6479
6480         if (current->flags & PF_MEMALLOC) {
6481                 redirty_page_for_writepage(wbc, page);
6482                 unlock_page(page);
6483                 return 0;
6484         }
6485         tree = &BTRFS_I(page->mapping->host)->io_tree;
6486         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6487 }
6488
6489 int btrfs_writepages(struct address_space *mapping,
6490                      struct writeback_control *wbc)
6491 {
6492         struct extent_io_tree *tree;
6493
6494         tree = &BTRFS_I(mapping->host)->io_tree;
6495         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6496 }
6497
6498 static int
6499 btrfs_readpages(struct file *file, struct address_space *mapping,
6500                 struct list_head *pages, unsigned nr_pages)
6501 {
6502         struct extent_io_tree *tree;
6503         tree = &BTRFS_I(mapping->host)->io_tree;
6504         return extent_readpages(tree, mapping, pages, nr_pages,
6505                                 btrfs_get_extent);
6506 }
6507 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6508 {
6509         struct extent_io_tree *tree;
6510         struct extent_map_tree *map;
6511         int ret;
6512
6513         tree = &BTRFS_I(page->mapping->host)->io_tree;
6514         map = &BTRFS_I(page->mapping->host)->extent_tree;
6515         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6516         if (ret == 1) {
6517                 ClearPagePrivate(page);
6518                 set_page_private(page, 0);
6519                 page_cache_release(page);
6520         }
6521         return ret;
6522 }
6523
6524 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6525 {
6526         if (PageWriteback(page) || PageDirty(page))
6527                 return 0;
6528         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6529 }
6530
6531 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6532 {
6533         struct inode *inode = page->mapping->host;
6534         struct extent_io_tree *tree;
6535         struct btrfs_ordered_extent *ordered;
6536         struct extent_state *cached_state = NULL;
6537         u64 page_start = page_offset(page);
6538         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6539
6540         /*
6541          * we have the page locked, so new writeback can't start,
6542          * and the dirty bit won't be cleared while we are here.
6543          *
6544          * Wait for IO on this page so that we can safely clear
6545          * the PagePrivate2 bit and do ordered accounting
6546          */
6547         wait_on_page_writeback(page);
6548
6549         tree = &BTRFS_I(inode)->io_tree;
6550         if (offset) {
6551                 btrfs_releasepage(page, GFP_NOFS);
6552                 return;
6553         }
6554         lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6555         ordered = btrfs_lookup_ordered_extent(inode,
6556                                            page_offset(page));
6557         if (ordered) {
6558                 /*
6559                  * IO on this page will never be started, so we need
6560                  * to account for any ordered extents now
6561                  */
6562                 clear_extent_bit(tree, page_start, page_end,
6563                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6564                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6565                                  &cached_state, GFP_NOFS);
6566                 /*
6567                  * whoever cleared the private bit is responsible
6568                  * for the finish_ordered_io
6569                  */
6570                 if (TestClearPagePrivate2(page) &&
6571                     btrfs_dec_test_ordered_pending(inode, &ordered, page_start,
6572                                                    PAGE_CACHE_SIZE, 1)) {
6573                         btrfs_finish_ordered_io(ordered);
6574                 }
6575                 btrfs_put_ordered_extent(ordered);
6576                 cached_state = NULL;
6577                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6578         }
6579         clear_extent_bit(tree, page_start, page_end,
6580                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6581                  EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6582         __btrfs_releasepage(page, GFP_NOFS);
6583
6584         ClearPageChecked(page);
6585         if (PagePrivate(page)) {
6586                 ClearPagePrivate(page);
6587                 set_page_private(page, 0);
6588                 page_cache_release(page);
6589         }
6590 }
6591
6592 /*
6593  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6594  * called from a page fault handler when a page is first dirtied. Hence we must
6595  * be careful to check for EOF conditions here. We set the page up correctly
6596  * for a written page which means we get ENOSPC checking when writing into
6597  * holes and correct delalloc and unwritten extent mapping on filesystems that
6598  * support these features.
6599  *
6600  * We are not allowed to take the i_mutex here so we have to play games to
6601  * protect against truncate races as the page could now be beyond EOF.  Because
6602  * vmtruncate() writes the inode size before removing pages, once we have the
6603  * page lock we can determine safely if the page is beyond EOF. If it is not
6604  * beyond EOF, then the page is guaranteed safe against truncation until we
6605  * unlock the page.
6606  */
6607 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6608 {
6609         struct page *page = vmf->page;
6610         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6611         struct btrfs_root *root = BTRFS_I(inode)->root;
6612         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6613         struct btrfs_ordered_extent *ordered;
6614         struct extent_state *cached_state = NULL;
6615         char *kaddr;
6616         unsigned long zero_start;
6617         loff_t size;
6618         int ret;
6619         int reserved = 0;
6620         u64 page_start;
6621         u64 page_end;
6622
6623         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6624         if (!ret) {
6625                 ret = file_update_time(vma->vm_file);
6626                 reserved = 1;
6627         }
6628         if (ret) {
6629                 if (ret == -ENOMEM)
6630                         ret = VM_FAULT_OOM;
6631                 else /* -ENOSPC, -EIO, etc */
6632                         ret = VM_FAULT_SIGBUS;
6633                 if (reserved)
6634                         goto out;
6635                 goto out_noreserve;
6636         }
6637
6638         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6639 again:
6640         lock_page(page);
6641         size = i_size_read(inode);
6642         page_start = page_offset(page);
6643         page_end = page_start + PAGE_CACHE_SIZE - 1;
6644
6645         if ((page->mapping != inode->i_mapping) ||
6646             (page_start >= size)) {
6647                 /* page got truncated out from underneath us */
6648                 goto out_unlock;
6649         }
6650         wait_on_page_writeback(page);
6651
6652         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
6653         set_page_extent_mapped(page);
6654
6655         /*
6656          * we can't set the delalloc bits if there are pending ordered
6657          * extents.  Drop our locks and wait for them to finish
6658          */
6659         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6660         if (ordered) {
6661                 unlock_extent_cached(io_tree, page_start, page_end,
6662                                      &cached_state, GFP_NOFS);
6663                 unlock_page(page);
6664                 btrfs_start_ordered_extent(inode, ordered, 1);
6665                 btrfs_put_ordered_extent(ordered);
6666                 goto again;
6667         }
6668
6669         /*
6670          * XXX - page_mkwrite gets called every time the page is dirtied, even
6671          * if it was already dirty, so for space accounting reasons we need to
6672          * clear any delalloc bits for the range we are fixing to save.  There
6673          * is probably a better way to do this, but for now keep consistent with
6674          * prepare_pages in the normal write path.
6675          */
6676         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6677                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6678                           0, 0, &cached_state, GFP_NOFS);
6679
6680         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6681                                         &cached_state);
6682         if (ret) {
6683                 unlock_extent_cached(io_tree, page_start, page_end,
6684                                      &cached_state, GFP_NOFS);
6685                 ret = VM_FAULT_SIGBUS;
6686                 goto out_unlock;
6687         }
6688         ret = 0;
6689
6690         /* page is wholly or partially inside EOF */
6691         if (page_start + PAGE_CACHE_SIZE > size)
6692                 zero_start = size & ~PAGE_CACHE_MASK;
6693         else
6694                 zero_start = PAGE_CACHE_SIZE;
6695
6696         if (zero_start != PAGE_CACHE_SIZE) {
6697                 kaddr = kmap(page);
6698                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6699                 flush_dcache_page(page);
6700                 kunmap(page);
6701         }
6702         ClearPageChecked(page);
6703         set_page_dirty(page);
6704         SetPageUptodate(page);
6705
6706         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6707         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6708
6709         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6710
6711 out_unlock:
6712         if (!ret)
6713                 return VM_FAULT_LOCKED;
6714         unlock_page(page);
6715 out:
6716         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6717 out_noreserve:
6718         return ret;
6719 }
6720
6721 static int btrfs_truncate(struct inode *inode)
6722 {
6723         struct btrfs_root *root = BTRFS_I(inode)->root;
6724         struct btrfs_block_rsv *rsv;
6725         int ret;
6726         int err = 0;
6727         struct btrfs_trans_handle *trans;
6728         unsigned long nr;
6729         u64 mask = root->sectorsize - 1;
6730         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
6731
6732         ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6733         if (ret)
6734                 return ret;
6735
6736         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6737         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6738
6739         /*
6740          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
6741          * 3 things going on here
6742          *
6743          * 1) We need to reserve space for our orphan item and the space to
6744          * delete our orphan item.  Lord knows we don't want to have a dangling
6745          * orphan item because we didn't reserve space to remove it.
6746          *
6747          * 2) We need to reserve space to update our inode.
6748          *
6749          * 3) We need to have something to cache all the space that is going to
6750          * be free'd up by the truncate operation, but also have some slack
6751          * space reserved in case it uses space during the truncate (thank you
6752          * very much snapshotting).
6753          *
6754          * And we need these to all be seperate.  The fact is we can use alot of
6755          * space doing the truncate, and we have no earthly idea how much space
6756          * we will use, so we need the truncate reservation to be seperate so it
6757          * doesn't end up using space reserved for updating the inode or
6758          * removing the orphan item.  We also need to be able to stop the
6759          * transaction and start a new one, which means we need to be able to
6760          * update the inode several times, and we have no idea of knowing how
6761          * many times that will be, so we can't just reserve 1 item for the
6762          * entirety of the opration, so that has to be done seperately as well.
6763          * Then there is the orphan item, which does indeed need to be held on
6764          * to for the whole operation, and we need nobody to touch this reserved
6765          * space except the orphan code.
6766          *
6767          * So that leaves us with
6768          *
6769          * 1) root->orphan_block_rsv - for the orphan deletion.
6770          * 2) rsv - for the truncate reservation, which we will steal from the
6771          * transaction reservation.
6772          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6773          * updating the inode.
6774          */
6775         rsv = btrfs_alloc_block_rsv(root);
6776         if (!rsv)
6777                 return -ENOMEM;
6778         rsv->size = min_size;
6779
6780         /*
6781          * 1 for the truncate slack space
6782          * 1 for the orphan item we're going to add
6783          * 1 for the orphan item deletion
6784          * 1 for updating the inode.
6785          */
6786         trans = btrfs_start_transaction(root, 4);
6787         if (IS_ERR(trans)) {
6788                 err = PTR_ERR(trans);
6789                 goto out;
6790         }
6791
6792         /* Migrate the slack space for the truncate to our reserve */
6793         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6794                                       min_size);
6795         BUG_ON(ret);
6796
6797         ret = btrfs_orphan_add(trans, inode);
6798         if (ret) {
6799                 btrfs_end_transaction(trans, root);
6800                 goto out;
6801         }
6802
6803         /*
6804          * setattr is responsible for setting the ordered_data_close flag,
6805          * but that is only tested during the last file release.  That
6806          * could happen well after the next commit, leaving a great big
6807          * window where new writes may get lost if someone chooses to write
6808          * to this file after truncating to zero
6809          *
6810          * The inode doesn't have any dirty data here, and so if we commit
6811          * this is a noop.  If someone immediately starts writing to the inode
6812          * it is very likely we'll catch some of their writes in this
6813          * transaction, and the commit will find this file on the ordered
6814          * data list with good things to send down.
6815          *
6816          * This is a best effort solution, there is still a window where
6817          * using truncate to replace the contents of the file will
6818          * end up with a zero length file after a crash.
6819          */
6820         if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
6821                                            &BTRFS_I(inode)->runtime_flags))
6822                 btrfs_add_ordered_operation(trans, root, inode);
6823
6824         while (1) {
6825                 ret = btrfs_block_rsv_refill(root, rsv, min_size);
6826                 if (ret) {
6827                         /*
6828                          * This can only happen with the original transaction we
6829                          * started above, every other time we shouldn't have a
6830                          * transaction started yet.
6831                          */
6832                         if (ret == -EAGAIN)
6833                                 goto end_trans;
6834                         err = ret;
6835                         break;
6836                 }
6837
6838                 if (!trans) {
6839                         /* Just need the 1 for updating the inode */
6840                         trans = btrfs_start_transaction(root, 1);
6841                         if (IS_ERR(trans)) {
6842                                 ret = err = PTR_ERR(trans);
6843                                 trans = NULL;
6844                                 break;
6845                         }
6846                 }
6847
6848                 trans->block_rsv = rsv;
6849
6850                 ret = btrfs_truncate_inode_items(trans, root, inode,
6851                                                  inode->i_size,
6852                                                  BTRFS_EXTENT_DATA_KEY);
6853                 if (ret != -EAGAIN) {
6854                         err = ret;
6855                         break;
6856                 }
6857
6858                 trans->block_rsv = &root->fs_info->trans_block_rsv;
6859                 ret = btrfs_update_inode(trans, root, inode);
6860                 if (ret) {
6861                         err = ret;
6862                         break;
6863                 }
6864 end_trans:
6865                 nr = trans->blocks_used;
6866                 btrfs_end_transaction(trans, root);
6867                 trans = NULL;
6868                 btrfs_btree_balance_dirty(root, nr);
6869         }
6870
6871         if (ret == 0 && inode->i_nlink > 0) {
6872                 trans->block_rsv = root->orphan_block_rsv;
6873                 ret = btrfs_orphan_del(trans, inode);
6874                 if (ret)
6875                         err = ret;
6876         } else if (ret && inode->i_nlink > 0) {
6877                 /*
6878                  * Failed to do the truncate, remove us from the in memory
6879                  * orphan list.
6880                  */
6881                 ret = btrfs_orphan_del(NULL, inode);
6882         }
6883
6884         if (trans) {
6885                 trans->block_rsv = &root->fs_info->trans_block_rsv;
6886                 ret = btrfs_update_inode(trans, root, inode);
6887                 if (ret && !err)
6888                         err = ret;
6889
6890                 nr = trans->blocks_used;
6891                 ret = btrfs_end_transaction(trans, root);
6892                 btrfs_btree_balance_dirty(root, nr);
6893         }
6894
6895 out:
6896         btrfs_free_block_rsv(root, rsv);
6897
6898         if (ret && !err)
6899                 err = ret;
6900
6901         return err;
6902 }
6903
6904 /*
6905  * create a new subvolume directory/inode (helper for the ioctl).
6906  */
6907 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6908                              struct btrfs_root *new_root, u64 new_dirid)
6909 {
6910         struct inode *inode;
6911         int err;
6912         u64 index = 0;
6913
6914         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
6915                                 new_dirid, new_dirid,
6916                                 S_IFDIR | (~current_umask() & S_IRWXUGO),
6917                                 &index);
6918         if (IS_ERR(inode))
6919                 return PTR_ERR(inode);
6920         inode->i_op = &btrfs_dir_inode_operations;
6921         inode->i_fop = &btrfs_dir_file_operations;
6922
6923         set_nlink(inode, 1);
6924         btrfs_i_size_write(inode, 0);
6925
6926         err = btrfs_update_inode(trans, new_root, inode);
6927
6928         iput(inode);
6929         return err;
6930 }
6931
6932 struct inode *btrfs_alloc_inode(struct super_block *sb)
6933 {
6934         struct btrfs_inode *ei;
6935         struct inode *inode;
6936
6937         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6938         if (!ei)
6939                 return NULL;
6940
6941         ei->root = NULL;
6942         ei->space_info = NULL;
6943         ei->generation = 0;
6944         ei->last_trans = 0;
6945         ei->last_sub_trans = 0;
6946         ei->logged_trans = 0;
6947         ei->delalloc_bytes = 0;
6948         ei->disk_i_size = 0;
6949         ei->flags = 0;
6950         ei->csum_bytes = 0;
6951         ei->index_cnt = (u64)-1;
6952         ei->last_unlink_trans = 0;
6953
6954         spin_lock_init(&ei->lock);
6955         ei->outstanding_extents = 0;
6956         ei->reserved_extents = 0;
6957
6958         ei->runtime_flags = 0;
6959         ei->force_compress = BTRFS_COMPRESS_NONE;
6960
6961         ei->delayed_node = NULL;
6962
6963         inode = &ei->vfs_inode;
6964         extent_map_tree_init(&ei->extent_tree);
6965         extent_io_tree_init(&ei->io_tree, &inode->i_data);
6966         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
6967         ei->io_tree.track_uptodate = 1;
6968         ei->io_failure_tree.track_uptodate = 1;
6969         mutex_init(&ei->log_mutex);
6970         mutex_init(&ei->delalloc_mutex);
6971         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6972         INIT_LIST_HEAD(&ei->delalloc_inodes);
6973         INIT_LIST_HEAD(&ei->ordered_operations);
6974         RB_CLEAR_NODE(&ei->rb_node);
6975
6976         return inode;
6977 }
6978
6979 static void btrfs_i_callback(struct rcu_head *head)
6980 {
6981         struct inode *inode = container_of(head, struct inode, i_rcu);
6982         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6983 }
6984
6985 void btrfs_destroy_inode(struct inode *inode)
6986 {
6987         struct btrfs_ordered_extent *ordered;
6988         struct btrfs_root *root = BTRFS_I(inode)->root;
6989
6990         WARN_ON(!hlist_empty(&inode->i_dentry));
6991         WARN_ON(inode->i_data.nrpages);
6992         WARN_ON(BTRFS_I(inode)->outstanding_extents);
6993         WARN_ON(BTRFS_I(inode)->reserved_extents);
6994         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
6995         WARN_ON(BTRFS_I(inode)->csum_bytes);
6996
6997         /*
6998          * This can happen where we create an inode, but somebody else also
6999          * created the same inode and we need to destroy the one we already
7000          * created.
7001          */
7002         if (!root)
7003                 goto free;
7004
7005         /*
7006          * Make sure we're properly removed from the ordered operation
7007          * lists.
7008          */
7009         smp_mb();
7010         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
7011                 spin_lock(&root->fs_info->ordered_extent_lock);
7012                 list_del_init(&BTRFS_I(inode)->ordered_operations);
7013                 spin_unlock(&root->fs_info->ordered_extent_lock);
7014         }
7015
7016         if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
7017                      &BTRFS_I(inode)->runtime_flags)) {
7018                 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
7019                        (unsigned long long)btrfs_ino(inode));
7020                 atomic_dec(&root->orphan_inodes);
7021         }
7022
7023         while (1) {
7024                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7025                 if (!ordered)
7026                         break;
7027                 else {
7028                         printk(KERN_ERR "btrfs found ordered "
7029                                "extent %llu %llu on inode cleanup\n",
7030                                (unsigned long long)ordered->file_offset,
7031                                (unsigned long long)ordered->len);
7032                         btrfs_remove_ordered_extent(inode, ordered);
7033                         btrfs_put_ordered_extent(ordered);
7034                         btrfs_put_ordered_extent(ordered);
7035                 }
7036         }
7037         inode_tree_del(inode);
7038         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
7039 free:
7040         btrfs_remove_delayed_node(inode);
7041         call_rcu(&inode->i_rcu, btrfs_i_callback);
7042 }
7043
7044 int btrfs_drop_inode(struct inode *inode)
7045 {
7046         struct btrfs_root *root = BTRFS_I(inode)->root;
7047
7048         if (btrfs_root_refs(&root->root_item) == 0 &&
7049             !btrfs_is_free_space_inode(root, inode))
7050                 return 1;
7051         else
7052                 return generic_drop_inode(inode);
7053 }
7054
7055 static void init_once(void *foo)
7056 {
7057         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
7058
7059         inode_init_once(&ei->vfs_inode);
7060 }
7061
7062 void btrfs_destroy_cachep(void)
7063 {
7064         if (btrfs_inode_cachep)
7065                 kmem_cache_destroy(btrfs_inode_cachep);
7066         if (btrfs_trans_handle_cachep)
7067                 kmem_cache_destroy(btrfs_trans_handle_cachep);
7068         if (btrfs_transaction_cachep)
7069                 kmem_cache_destroy(btrfs_transaction_cachep);
7070         if (btrfs_path_cachep)
7071                 kmem_cache_destroy(btrfs_path_cachep);
7072         if (btrfs_free_space_cachep)
7073                 kmem_cache_destroy(btrfs_free_space_cachep);
7074 }
7075
7076 int btrfs_init_cachep(void)
7077 {
7078         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
7079                         sizeof(struct btrfs_inode), 0,
7080                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
7081         if (!btrfs_inode_cachep)
7082                 goto fail;
7083
7084         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
7085                         sizeof(struct btrfs_trans_handle), 0,
7086                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7087         if (!btrfs_trans_handle_cachep)
7088                 goto fail;
7089
7090         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
7091                         sizeof(struct btrfs_transaction), 0,
7092                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7093         if (!btrfs_transaction_cachep)
7094                 goto fail;
7095
7096         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
7097                         sizeof(struct btrfs_path), 0,
7098                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7099         if (!btrfs_path_cachep)
7100                 goto fail;
7101
7102         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
7103                         sizeof(struct btrfs_free_space), 0,
7104                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7105         if (!btrfs_free_space_cachep)
7106                 goto fail;
7107
7108         return 0;
7109 fail:
7110         btrfs_destroy_cachep();
7111         return -ENOMEM;
7112 }
7113
7114 static int btrfs_getattr(struct vfsmount *mnt,
7115                          struct dentry *dentry, struct kstat *stat)
7116 {
7117         struct inode *inode = dentry->d_inode;
7118         u32 blocksize = inode->i_sb->s_blocksize;
7119
7120         generic_fillattr(inode, stat);
7121         stat->dev = BTRFS_I(inode)->root->anon_dev;
7122         stat->blksize = PAGE_CACHE_SIZE;
7123         stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
7124                 ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9;
7125         return 0;
7126 }
7127
7128 /*
7129  * If a file is moved, it will inherit the cow and compression flags of the new
7130  * directory.
7131  */
7132 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
7133 {
7134         struct btrfs_inode *b_dir = BTRFS_I(dir);
7135         struct btrfs_inode *b_inode = BTRFS_I(inode);
7136
7137         if (b_dir->flags & BTRFS_INODE_NODATACOW)
7138                 b_inode->flags |= BTRFS_INODE_NODATACOW;
7139         else
7140                 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
7141
7142         if (b_dir->flags & BTRFS_INODE_COMPRESS) {
7143                 b_inode->flags |= BTRFS_INODE_COMPRESS;
7144                 b_inode->flags &= ~BTRFS_INODE_NOCOMPRESS;
7145         } else {
7146                 b_inode->flags &= ~(BTRFS_INODE_COMPRESS |
7147                                     BTRFS_INODE_NOCOMPRESS);
7148         }
7149 }
7150
7151 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
7152                            struct inode *new_dir, struct dentry *new_dentry)
7153 {
7154         struct btrfs_trans_handle *trans;
7155         struct btrfs_root *root = BTRFS_I(old_dir)->root;
7156         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
7157         struct inode *new_inode = new_dentry->d_inode;
7158         struct inode *old_inode = old_dentry->d_inode;
7159         struct timespec ctime = CURRENT_TIME;
7160         u64 index = 0;
7161         u64 root_objectid;
7162         int ret;
7163         u64 old_ino = btrfs_ino(old_inode);
7164
7165         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
7166                 return -EPERM;
7167
7168         /* we only allow rename subvolume link between subvolumes */
7169         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
7170                 return -EXDEV;
7171
7172         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
7173             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
7174                 return -ENOTEMPTY;
7175
7176         if (S_ISDIR(old_inode->i_mode) && new_inode &&
7177             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
7178                 return -ENOTEMPTY;
7179         /*
7180          * we're using rename to replace one file with another.
7181          * and the replacement file is large.  Start IO on it now so
7182          * we don't add too much work to the end of the transaction
7183          */
7184         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
7185             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
7186                 filemap_flush(old_inode->i_mapping);
7187
7188         /* close the racy window with snapshot create/destroy ioctl */
7189         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7190                 down_read(&root->fs_info->subvol_sem);
7191         /*
7192          * We want to reserve the absolute worst case amount of items.  So if
7193          * both inodes are subvols and we need to unlink them then that would
7194          * require 4 item modifications, but if they are both normal inodes it
7195          * would require 5 item modifications, so we'll assume their normal
7196          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7197          * should cover the worst case number of items we'll modify.
7198          */
7199         trans = btrfs_start_transaction(root, 20);
7200         if (IS_ERR(trans)) {
7201                 ret = PTR_ERR(trans);
7202                 goto out_notrans;
7203         }
7204
7205         if (dest != root)
7206                 btrfs_record_root_in_trans(trans, dest);
7207
7208         ret = btrfs_set_inode_index(new_dir, &index);
7209         if (ret)
7210                 goto out_fail;
7211
7212         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7213                 /* force full log commit if subvolume involved. */
7214                 root->fs_info->last_trans_log_full_commit = trans->transid;
7215         } else {
7216                 ret = btrfs_insert_inode_ref(trans, dest,
7217                                              new_dentry->d_name.name,
7218                                              new_dentry->d_name.len,
7219                                              old_ino,
7220                                              btrfs_ino(new_dir), index);
7221                 if (ret)
7222                         goto out_fail;
7223                 /*
7224                  * this is an ugly little race, but the rename is required
7225                  * to make sure that if we crash, the inode is either at the
7226                  * old name or the new one.  pinning the log transaction lets
7227                  * us make sure we don't allow a log commit to come in after
7228                  * we unlink the name but before we add the new name back in.
7229                  */
7230                 btrfs_pin_log_trans(root);
7231         }
7232         /*
7233          * make sure the inode gets flushed if it is replacing
7234          * something.
7235          */
7236         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7237                 btrfs_add_ordered_operation(trans, root, old_inode);
7238
7239         inode_inc_iversion(old_dir);
7240         inode_inc_iversion(new_dir);
7241         inode_inc_iversion(old_inode);
7242         old_dir->i_ctime = old_dir->i_mtime = ctime;
7243         new_dir->i_ctime = new_dir->i_mtime = ctime;
7244         old_inode->i_ctime = ctime;
7245
7246         if (old_dentry->d_parent != new_dentry->d_parent)
7247                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7248
7249         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7250                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7251                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7252                                         old_dentry->d_name.name,
7253                                         old_dentry->d_name.len);
7254         } else {
7255                 ret = __btrfs_unlink_inode(trans, root, old_dir,
7256                                         old_dentry->d_inode,
7257                                         old_dentry->d_name.name,
7258                                         old_dentry->d_name.len);
7259                 if (!ret)
7260                         ret = btrfs_update_inode(trans, root, old_inode);
7261         }
7262         if (ret) {
7263                 btrfs_abort_transaction(trans, root, ret);
7264                 goto out_fail;
7265         }
7266
7267         if (new_inode) {
7268                 inode_inc_iversion(new_inode);
7269                 new_inode->i_ctime = CURRENT_TIME;
7270                 if (unlikely(btrfs_ino(new_inode) ==
7271                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7272                         root_objectid = BTRFS_I(new_inode)->location.objectid;
7273                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
7274                                                 root_objectid,
7275                                                 new_dentry->d_name.name,
7276                                                 new_dentry->d_name.len);
7277                         BUG_ON(new_inode->i_nlink == 0);
7278                 } else {
7279                         ret = btrfs_unlink_inode(trans, dest, new_dir,
7280                                                  new_dentry->d_inode,
7281                                                  new_dentry->d_name.name,
7282                                                  new_dentry->d_name.len);
7283                 }
7284                 if (!ret && new_inode->i_nlink == 0) {
7285                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7286                         BUG_ON(ret);
7287                 }
7288                 if (ret) {
7289                         btrfs_abort_transaction(trans, root, ret);
7290                         goto out_fail;
7291                 }
7292         }
7293
7294         fixup_inode_flags(new_dir, old_inode);
7295
7296         ret = btrfs_add_link(trans, new_dir, old_inode,
7297                              new_dentry->d_name.name,
7298                              new_dentry->d_name.len, 0, index);
7299         if (ret) {
7300                 btrfs_abort_transaction(trans, root, ret);
7301                 goto out_fail;
7302         }
7303
7304         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7305                 struct dentry *parent = new_dentry->d_parent;
7306                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7307                 btrfs_end_log_trans(root);
7308         }
7309 out_fail:
7310         btrfs_end_transaction(trans, root);
7311 out_notrans:
7312         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7313                 up_read(&root->fs_info->subvol_sem);
7314
7315         return ret;
7316 }
7317
7318 /*
7319  * some fairly slow code that needs optimization. This walks the list
7320  * of all the inodes with pending delalloc and forces them to disk.
7321  */
7322 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7323 {
7324         struct list_head *head = &root->fs_info->delalloc_inodes;
7325         struct btrfs_inode *binode;
7326         struct inode *inode;
7327
7328         if (root->fs_info->sb->s_flags & MS_RDONLY)
7329                 return -EROFS;
7330
7331         spin_lock(&root->fs_info->delalloc_lock);
7332         while (!list_empty(head)) {
7333                 binode = list_entry(head->next, struct btrfs_inode,
7334                                     delalloc_inodes);
7335                 inode = igrab(&binode->vfs_inode);
7336                 if (!inode)
7337                         list_del_init(&binode->delalloc_inodes);
7338                 spin_unlock(&root->fs_info->delalloc_lock);
7339                 if (inode) {
7340                         filemap_flush(inode->i_mapping);
7341                         if (delay_iput)
7342                                 btrfs_add_delayed_iput(inode);
7343                         else
7344                                 iput(inode);
7345                 }
7346                 cond_resched();
7347                 spin_lock(&root->fs_info->delalloc_lock);
7348         }
7349         spin_unlock(&root->fs_info->delalloc_lock);
7350
7351         /* the filemap_flush will queue IO into the worker threads, but
7352          * we have to make sure the IO is actually started and that
7353          * ordered extents get created before we return
7354          */
7355         atomic_inc(&root->fs_info->async_submit_draining);
7356         while (atomic_read(&root->fs_info->nr_async_submits) ||
7357               atomic_read(&root->fs_info->async_delalloc_pages)) {
7358                 wait_event(root->fs_info->async_submit_wait,
7359                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7360                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7361         }
7362         atomic_dec(&root->fs_info->async_submit_draining);
7363         return 0;
7364 }
7365
7366 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7367                          const char *symname)
7368 {
7369         struct btrfs_trans_handle *trans;
7370         struct btrfs_root *root = BTRFS_I(dir)->root;
7371         struct btrfs_path *path;
7372         struct btrfs_key key;
7373         struct inode *inode = NULL;
7374         int err;
7375         int drop_inode = 0;
7376         u64 objectid;
7377         u64 index = 0 ;
7378         int name_len;
7379         int datasize;
7380         unsigned long ptr;
7381         struct btrfs_file_extent_item *ei;
7382         struct extent_buffer *leaf;
7383         unsigned long nr = 0;
7384
7385         name_len = strlen(symname) + 1;
7386         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7387                 return -ENAMETOOLONG;
7388
7389         /*
7390          * 2 items for inode item and ref
7391          * 2 items for dir items
7392          * 1 item for xattr if selinux is on
7393          */
7394         trans = btrfs_start_transaction(root, 5);
7395         if (IS_ERR(trans))
7396                 return PTR_ERR(trans);
7397
7398         err = btrfs_find_free_ino(root, &objectid);
7399         if (err)
7400                 goto out_unlock;
7401
7402         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7403                                 dentry->d_name.len, btrfs_ino(dir), objectid,
7404                                 S_IFLNK|S_IRWXUGO, &index);
7405         if (IS_ERR(inode)) {
7406                 err = PTR_ERR(inode);
7407                 goto out_unlock;
7408         }
7409
7410         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7411         if (err) {
7412                 drop_inode = 1;
7413                 goto out_unlock;
7414         }
7415
7416         /*
7417         * If the active LSM wants to access the inode during
7418         * d_instantiate it needs these. Smack checks to see
7419         * if the filesystem supports xattrs by looking at the
7420         * ops vector.
7421         */
7422         inode->i_fop = &btrfs_file_operations;
7423         inode->i_op = &btrfs_file_inode_operations;
7424
7425         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7426         if (err)
7427                 drop_inode = 1;
7428         else {
7429                 inode->i_mapping->a_ops = &btrfs_aops;
7430                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7431                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7432         }
7433         if (drop_inode)
7434                 goto out_unlock;
7435
7436         path = btrfs_alloc_path();
7437         if (!path) {
7438                 err = -ENOMEM;
7439                 drop_inode = 1;
7440                 goto out_unlock;
7441         }
7442         key.objectid = btrfs_ino(inode);
7443         key.offset = 0;
7444         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7445         datasize = btrfs_file_extent_calc_inline_size(name_len);
7446         err = btrfs_insert_empty_item(trans, root, path, &key,
7447                                       datasize);
7448         if (err) {
7449                 drop_inode = 1;
7450                 btrfs_free_path(path);
7451                 goto out_unlock;
7452         }
7453         leaf = path->nodes[0];
7454         ei = btrfs_item_ptr(leaf, path->slots[0],
7455                             struct btrfs_file_extent_item);
7456         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7457         btrfs_set_file_extent_type(leaf, ei,
7458                                    BTRFS_FILE_EXTENT_INLINE);
7459         btrfs_set_file_extent_encryption(leaf, ei, 0);
7460         btrfs_set_file_extent_compression(leaf, ei, 0);
7461         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7462         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7463
7464         ptr = btrfs_file_extent_inline_start(ei);
7465         write_extent_buffer(leaf, symname, ptr, name_len);
7466         btrfs_mark_buffer_dirty(leaf);
7467         btrfs_free_path(path);
7468
7469         inode->i_op = &btrfs_symlink_inode_operations;
7470         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7471         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7472         inode_set_bytes(inode, name_len);
7473         btrfs_i_size_write(inode, name_len - 1);
7474         err = btrfs_update_inode(trans, root, inode);
7475         if (err)
7476                 drop_inode = 1;
7477
7478 out_unlock:
7479         if (!err)
7480                 d_instantiate(dentry, inode);
7481         nr = trans->blocks_used;
7482         btrfs_end_transaction(trans, root);
7483         if (drop_inode) {
7484                 inode_dec_link_count(inode);
7485                 iput(inode);
7486         }
7487         btrfs_btree_balance_dirty(root, nr);
7488         return err;
7489 }
7490
7491 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7492                                        u64 start, u64 num_bytes, u64 min_size,
7493                                        loff_t actual_len, u64 *alloc_hint,
7494                                        struct btrfs_trans_handle *trans)
7495 {
7496         struct btrfs_root *root = BTRFS_I(inode)->root;
7497         struct btrfs_key ins;
7498         u64 cur_offset = start;
7499         u64 i_size;
7500         int ret = 0;
7501         bool own_trans = true;
7502
7503         if (trans)
7504                 own_trans = false;
7505         while (num_bytes > 0) {
7506                 if (own_trans) {
7507                         trans = btrfs_start_transaction(root, 3);
7508                         if (IS_ERR(trans)) {
7509                                 ret = PTR_ERR(trans);
7510                                 break;
7511                         }
7512                 }
7513
7514                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7515                                            0, *alloc_hint, &ins, 1);
7516                 if (ret) {
7517                         if (own_trans)
7518                                 btrfs_end_transaction(trans, root);
7519                         break;
7520                 }
7521
7522                 ret = insert_reserved_file_extent(trans, inode,
7523                                                   cur_offset, ins.objectid,
7524                                                   ins.offset, ins.offset,
7525                                                   ins.offset, 0, 0, 0,
7526                                                   BTRFS_FILE_EXTENT_PREALLOC);
7527                 if (ret) {
7528                         btrfs_abort_transaction(trans, root, ret);
7529                         if (own_trans)
7530                                 btrfs_end_transaction(trans, root);
7531                         break;
7532                 }
7533                 btrfs_drop_extent_cache(inode, cur_offset,
7534                                         cur_offset + ins.offset -1, 0);
7535
7536                 num_bytes -= ins.offset;
7537                 cur_offset += ins.offset;
7538                 *alloc_hint = ins.objectid + ins.offset;
7539
7540                 inode_inc_iversion(inode);
7541                 inode->i_ctime = CURRENT_TIME;
7542                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7543                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7544                     (actual_len > inode->i_size) &&
7545                     (cur_offset > inode->i_size)) {
7546                         if (cur_offset > actual_len)
7547                                 i_size = actual_len;
7548                         else
7549                                 i_size = cur_offset;
7550                         i_size_write(inode, i_size);
7551                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7552                 }
7553
7554                 ret = btrfs_update_inode(trans, root, inode);
7555
7556                 if (ret) {
7557                         btrfs_abort_transaction(trans, root, ret);
7558                         if (own_trans)
7559                                 btrfs_end_transaction(trans, root);
7560                         break;
7561                 }
7562
7563                 if (own_trans)
7564                         btrfs_end_transaction(trans, root);
7565         }
7566         return ret;
7567 }
7568
7569 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7570                               u64 start, u64 num_bytes, u64 min_size,
7571                               loff_t actual_len, u64 *alloc_hint)
7572 {
7573         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7574                                            min_size, actual_len, alloc_hint,
7575                                            NULL);
7576 }
7577
7578 int btrfs_prealloc_file_range_trans(struct inode *inode,
7579                                     struct btrfs_trans_handle *trans, int mode,
7580                                     u64 start, u64 num_bytes, u64 min_size,
7581                                     loff_t actual_len, u64 *alloc_hint)
7582 {
7583         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7584                                            min_size, actual_len, alloc_hint, trans);
7585 }
7586
7587 static int btrfs_set_page_dirty(struct page *page)
7588 {
7589         return __set_page_dirty_nobuffers(page);
7590 }
7591
7592 static int btrfs_permission(struct inode *inode, int mask)
7593 {
7594         struct btrfs_root *root = BTRFS_I(inode)->root;
7595         umode_t mode = inode->i_mode;
7596
7597         if (mask & MAY_WRITE &&
7598             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7599                 if (btrfs_root_readonly(root))
7600                         return -EROFS;
7601                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7602                         return -EACCES;
7603         }
7604         return generic_permission(inode, mask);
7605 }
7606
7607 static const struct inode_operations btrfs_dir_inode_operations = {
7608         .getattr        = btrfs_getattr,
7609         .lookup         = btrfs_lookup,
7610         .create         = btrfs_create,
7611         .unlink         = btrfs_unlink,
7612         .link           = btrfs_link,
7613         .mkdir          = btrfs_mkdir,
7614         .rmdir          = btrfs_rmdir,
7615         .rename         = btrfs_rename,
7616         .symlink        = btrfs_symlink,
7617         .setattr        = btrfs_setattr,
7618         .mknod          = btrfs_mknod,
7619         .setxattr       = btrfs_setxattr,
7620         .getxattr       = btrfs_getxattr,
7621         .listxattr      = btrfs_listxattr,
7622         .removexattr    = btrfs_removexattr,
7623         .permission     = btrfs_permission,
7624         .get_acl        = btrfs_get_acl,
7625 };
7626 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7627         .lookup         = btrfs_lookup,
7628         .permission     = btrfs_permission,
7629         .get_acl        = btrfs_get_acl,
7630 };
7631
7632 static const struct file_operations btrfs_dir_file_operations = {
7633         .llseek         = generic_file_llseek,
7634         .read           = generic_read_dir,
7635         .readdir        = btrfs_real_readdir,
7636         .unlocked_ioctl = btrfs_ioctl,
7637 #ifdef CONFIG_COMPAT
7638         .compat_ioctl   = btrfs_ioctl,
7639 #endif
7640         .release        = btrfs_release_file,
7641         .fsync          = btrfs_sync_file,
7642 };
7643
7644 static struct extent_io_ops btrfs_extent_io_ops = {
7645         .fill_delalloc = run_delalloc_range,
7646         .submit_bio_hook = btrfs_submit_bio_hook,
7647         .merge_bio_hook = btrfs_merge_bio_hook,
7648         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7649         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7650         .writepage_start_hook = btrfs_writepage_start_hook,
7651         .set_bit_hook = btrfs_set_bit_hook,
7652         .clear_bit_hook = btrfs_clear_bit_hook,
7653         .merge_extent_hook = btrfs_merge_extent_hook,
7654         .split_extent_hook = btrfs_split_extent_hook,
7655 };
7656
7657 /*
7658  * btrfs doesn't support the bmap operation because swapfiles
7659  * use bmap to make a mapping of extents in the file.  They assume
7660  * these extents won't change over the life of the file and they
7661  * use the bmap result to do IO directly to the drive.
7662  *
7663  * the btrfs bmap call would return logical addresses that aren't
7664  * suitable for IO and they also will change frequently as COW
7665  * operations happen.  So, swapfile + btrfs == corruption.
7666  *
7667  * For now we're avoiding this by dropping bmap.
7668  */
7669 static const struct address_space_operations btrfs_aops = {
7670         .readpage       = btrfs_readpage,
7671         .writepage      = btrfs_writepage,
7672         .writepages     = btrfs_writepages,
7673         .readpages      = btrfs_readpages,
7674         .direct_IO      = btrfs_direct_IO,
7675         .invalidatepage = btrfs_invalidatepage,
7676         .releasepage    = btrfs_releasepage,
7677         .set_page_dirty = btrfs_set_page_dirty,
7678         .error_remove_page = generic_error_remove_page,
7679 };
7680
7681 static const struct address_space_operations btrfs_symlink_aops = {
7682         .readpage       = btrfs_readpage,
7683         .writepage      = btrfs_writepage,
7684         .invalidatepage = btrfs_invalidatepage,
7685         .releasepage    = btrfs_releasepage,
7686 };
7687
7688 static const struct inode_operations btrfs_file_inode_operations = {
7689         .getattr        = btrfs_getattr,
7690         .setattr        = btrfs_setattr,
7691         .setxattr       = btrfs_setxattr,
7692         .getxattr       = btrfs_getxattr,
7693         .listxattr      = btrfs_listxattr,
7694         .removexattr    = btrfs_removexattr,
7695         .permission     = btrfs_permission,
7696         .fiemap         = btrfs_fiemap,
7697         .get_acl        = btrfs_get_acl,
7698         .update_time    = btrfs_update_time,
7699 };
7700 static const struct inode_operations btrfs_special_inode_operations = {
7701         .getattr        = btrfs_getattr,
7702         .setattr        = btrfs_setattr,
7703         .permission     = btrfs_permission,
7704         .setxattr       = btrfs_setxattr,
7705         .getxattr       = btrfs_getxattr,
7706         .listxattr      = btrfs_listxattr,
7707         .removexattr    = btrfs_removexattr,
7708         .get_acl        = btrfs_get_acl,
7709         .update_time    = btrfs_update_time,
7710 };
7711 static const struct inode_operations btrfs_symlink_inode_operations = {
7712         .readlink       = generic_readlink,
7713         .follow_link    = page_follow_link_light,
7714         .put_link       = page_put_link,
7715         .getattr        = btrfs_getattr,
7716         .setattr        = btrfs_setattr,
7717         .permission     = btrfs_permission,
7718         .setxattr       = btrfs_setxattr,
7719         .getxattr       = btrfs_getxattr,
7720         .listxattr      = btrfs_listxattr,
7721         .removexattr    = btrfs_removexattr,
7722         .get_acl        = btrfs_get_acl,
7723         .update_time    = btrfs_update_time,
7724 };
7725
7726 const struct dentry_operations btrfs_dentry_operations = {
7727         .d_delete       = btrfs_dentry_delete,
7728         .d_release      = btrfs_dentry_release,
7729 };