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(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 * 1024 * 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 * 1024;
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(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(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(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(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(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 dont to make sure anybody waiting knows we are done
1973          * upating 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(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(inode)
2736             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2737                 btrfs_update_root_times(trans, root);
2738
2739                 ret = btrfs_delayed_update_inode(trans, root, inode);
2740                 if (!ret)
2741                         btrfs_set_inode_last_trans(trans, inode);
2742                 return ret;
2743         }
2744
2745         return btrfs_update_inode_item(trans, root, inode);
2746 }
2747
2748 static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2749                                 struct btrfs_root *root, struct inode *inode)
2750 {
2751         int ret;
2752
2753         ret = btrfs_update_inode(trans, root, inode);
2754         if (ret == -ENOSPC)
2755                 return btrfs_update_inode_item(trans, root, inode);
2756         return ret;
2757 }
2758
2759 /*
2760  * unlink helper that gets used here in inode.c and in the tree logging
2761  * recovery code.  It remove a link in a directory with a given name, and
2762  * also drops the back refs in the inode to the directory
2763  */
2764 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2765                                 struct btrfs_root *root,
2766                                 struct inode *dir, struct inode *inode,
2767                                 const char *name, int name_len)
2768 {
2769         struct btrfs_path *path;
2770         int ret = 0;
2771         struct extent_buffer *leaf;
2772         struct btrfs_dir_item *di;
2773         struct btrfs_key key;
2774         u64 index;
2775         u64 ino = btrfs_ino(inode);
2776         u64 dir_ino = btrfs_ino(dir);
2777
2778         path = btrfs_alloc_path();
2779         if (!path) {
2780                 ret = -ENOMEM;
2781                 goto out;
2782         }
2783
2784         path->leave_spinning = 1;
2785         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2786                                     name, name_len, -1);
2787         if (IS_ERR(di)) {
2788                 ret = PTR_ERR(di);
2789                 goto err;
2790         }
2791         if (!di) {
2792                 ret = -ENOENT;
2793                 goto err;
2794         }
2795         leaf = path->nodes[0];
2796         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2797         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2798         if (ret)
2799                 goto err;
2800         btrfs_release_path(path);
2801
2802         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2803                                   dir_ino, &index);
2804         if (ret) {
2805                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2806                        "inode %llu parent %llu\n", name_len, name,
2807                        (unsigned long long)ino, (unsigned long long)dir_ino);
2808                 btrfs_abort_transaction(trans, root, ret);
2809                 goto err;
2810         }
2811
2812         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2813         if (ret) {
2814                 btrfs_abort_transaction(trans, root, ret);
2815                 goto err;
2816         }
2817
2818         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2819                                          inode, dir_ino);
2820         if (ret != 0 && ret != -ENOENT) {
2821                 btrfs_abort_transaction(trans, root, ret);
2822                 goto err;
2823         }
2824
2825         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2826                                            dir, index);
2827         if (ret == -ENOENT)
2828                 ret = 0;
2829 err:
2830         btrfs_free_path(path);
2831         if (ret)
2832                 goto out;
2833
2834         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2835         inode_inc_iversion(inode);
2836         inode_inc_iversion(dir);
2837         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2838         ret = btrfs_update_inode(trans, root, dir);
2839 out:
2840         return ret;
2841 }
2842
2843 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2844                        struct btrfs_root *root,
2845                        struct inode *dir, struct inode *inode,
2846                        const char *name, int name_len)
2847 {
2848         int ret;
2849         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2850         if (!ret) {
2851                 btrfs_drop_nlink(inode);
2852                 ret = btrfs_update_inode(trans, root, inode);
2853         }
2854         return ret;
2855 }
2856
2857
2858 /* helper to check if there is any shared block in the path */
2859 static int check_path_shared(struct btrfs_root *root,
2860                              struct btrfs_path *path)
2861 {
2862         struct extent_buffer *eb;
2863         int level;
2864         u64 refs = 1;
2865
2866         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2867                 int ret;
2868
2869                 if (!path->nodes[level])
2870                         break;
2871                 eb = path->nodes[level];
2872                 if (!btrfs_block_can_be_shared(root, eb))
2873                         continue;
2874                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2875                                                &refs, NULL);
2876                 if (refs > 1)
2877                         return 1;
2878         }
2879         return 0;
2880 }
2881
2882 /*
2883  * helper to start transaction for unlink and rmdir.
2884  *
2885  * unlink and rmdir are special in btrfs, they do not always free space.
2886  * so in enospc case, we should make sure they will free space before
2887  * allowing them to use the global metadata reservation.
2888  */
2889 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2890                                                        struct dentry *dentry)
2891 {
2892         struct btrfs_trans_handle *trans;
2893         struct btrfs_root *root = BTRFS_I(dir)->root;
2894         struct btrfs_path *path;
2895         struct btrfs_inode_ref *ref;
2896         struct btrfs_dir_item *di;
2897         struct inode *inode = dentry->d_inode;
2898         u64 index;
2899         int check_link = 1;
2900         int err = -ENOSPC;
2901         int ret;
2902         u64 ino = btrfs_ino(inode);
2903         u64 dir_ino = btrfs_ino(dir);
2904
2905         /*
2906          * 1 for the possible orphan item
2907          * 1 for the dir item
2908          * 1 for the dir index
2909          * 1 for the inode ref
2910          * 1 for the inode ref in the tree log
2911          * 2 for the dir entries in the log
2912          * 1 for the inode
2913          */
2914         trans = btrfs_start_transaction(root, 8);
2915         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2916                 return trans;
2917
2918         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2919                 return ERR_PTR(-ENOSPC);
2920
2921         /* check if there is someone else holds reference */
2922         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2923                 return ERR_PTR(-ENOSPC);
2924
2925         if (atomic_read(&inode->i_count) > 2)
2926                 return ERR_PTR(-ENOSPC);
2927
2928         if (xchg(&root->fs_info->enospc_unlink, 1))
2929                 return ERR_PTR(-ENOSPC);
2930
2931         path = btrfs_alloc_path();
2932         if (!path) {
2933                 root->fs_info->enospc_unlink = 0;
2934                 return ERR_PTR(-ENOMEM);
2935         }
2936
2937         /* 1 for the orphan item */
2938         trans = btrfs_start_transaction(root, 1);
2939         if (IS_ERR(trans)) {
2940                 btrfs_free_path(path);
2941                 root->fs_info->enospc_unlink = 0;
2942                 return trans;
2943         }
2944
2945         path->skip_locking = 1;
2946         path->search_commit_root = 1;
2947
2948         ret = btrfs_lookup_inode(trans, root, path,
2949                                 &BTRFS_I(dir)->location, 0);
2950         if (ret < 0) {
2951                 err = ret;
2952                 goto out;
2953         }
2954         if (ret == 0) {
2955                 if (check_path_shared(root, path))
2956                         goto out;
2957         } else {
2958                 check_link = 0;
2959         }
2960         btrfs_release_path(path);
2961
2962         ret = btrfs_lookup_inode(trans, root, path,
2963                                 &BTRFS_I(inode)->location, 0);
2964         if (ret < 0) {
2965                 err = ret;
2966                 goto out;
2967         }
2968         if (ret == 0) {
2969                 if (check_path_shared(root, path))
2970                         goto out;
2971         } else {
2972                 check_link = 0;
2973         }
2974         btrfs_release_path(path);
2975
2976         if (ret == 0 && S_ISREG(inode->i_mode)) {
2977                 ret = btrfs_lookup_file_extent(trans, root, path,
2978                                                ino, (u64)-1, 0);
2979                 if (ret < 0) {
2980                         err = ret;
2981                         goto out;
2982                 }
2983                 BUG_ON(ret == 0); /* Corruption */
2984                 if (check_path_shared(root, path))
2985                         goto out;
2986                 btrfs_release_path(path);
2987         }
2988
2989         if (!check_link) {
2990                 err = 0;
2991                 goto out;
2992         }
2993
2994         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2995                                 dentry->d_name.name, dentry->d_name.len, 0);
2996         if (IS_ERR(di)) {
2997                 err = PTR_ERR(di);
2998                 goto out;
2999         }
3000         if (di) {
3001                 if (check_path_shared(root, path))
3002                         goto out;
3003         } else {
3004                 err = 0;
3005                 goto out;
3006         }
3007         btrfs_release_path(path);
3008
3009         ref = btrfs_lookup_inode_ref(trans, root, path,
3010                                 dentry->d_name.name, dentry->d_name.len,
3011                                 ino, dir_ino, 0);
3012         if (IS_ERR(ref)) {
3013                 err = PTR_ERR(ref);
3014                 goto out;
3015         }
3016         BUG_ON(!ref); /* Logic error */
3017         if (check_path_shared(root, path))
3018                 goto out;
3019         index = btrfs_inode_ref_index(path->nodes[0], ref);
3020         btrfs_release_path(path);
3021
3022         /*
3023          * This is a commit root search, if we can lookup inode item and other
3024          * relative items in the commit root, it means the transaction of
3025          * dir/file creation has been committed, and the dir index item that we
3026          * delay to insert has also been inserted into the commit root. So
3027          * we needn't worry about the delayed insertion of the dir index item
3028          * here.
3029          */
3030         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
3031                                 dentry->d_name.name, dentry->d_name.len, 0);
3032         if (IS_ERR(di)) {
3033                 err = PTR_ERR(di);
3034                 goto out;
3035         }
3036         BUG_ON(ret == -ENOENT);
3037         if (check_path_shared(root, path))
3038                 goto out;
3039
3040         err = 0;
3041 out:
3042         btrfs_free_path(path);
3043         /* Migrate the orphan reservation over */
3044         if (!err)
3045                 err = btrfs_block_rsv_migrate(trans->block_rsv,
3046                                 &root->fs_info->global_block_rsv,
3047                                 trans->bytes_reserved);
3048
3049         if (err) {
3050                 btrfs_end_transaction(trans, root);
3051                 root->fs_info->enospc_unlink = 0;
3052                 return ERR_PTR(err);
3053         }
3054
3055         trans->block_rsv = &root->fs_info->global_block_rsv;
3056         return trans;
3057 }
3058
3059 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
3060                                struct btrfs_root *root)
3061 {
3062         if (trans->block_rsv == &root->fs_info->global_block_rsv) {
3063                 btrfs_block_rsv_release(root, trans->block_rsv,
3064                                         trans->bytes_reserved);
3065                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3066                 BUG_ON(!root->fs_info->enospc_unlink);
3067                 root->fs_info->enospc_unlink = 0;
3068         }
3069         btrfs_end_transaction(trans, root);
3070 }
3071
3072 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3073 {
3074         struct btrfs_root *root = BTRFS_I(dir)->root;
3075         struct btrfs_trans_handle *trans;
3076         struct inode *inode = dentry->d_inode;
3077         int ret;
3078         unsigned long nr = 0;
3079
3080         trans = __unlink_start_trans(dir, dentry);
3081         if (IS_ERR(trans))
3082                 return PTR_ERR(trans);
3083
3084         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3085
3086         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3087                                  dentry->d_name.name, dentry->d_name.len);
3088         if (ret)
3089                 goto out;
3090
3091         if (inode->i_nlink == 0) {
3092                 ret = btrfs_orphan_add(trans, inode);
3093                 if (ret)
3094                         goto out;
3095         }
3096
3097 out:
3098         nr = trans->blocks_used;
3099         __unlink_end_trans(trans, root);
3100         btrfs_btree_balance_dirty(root, nr);
3101         return ret;
3102 }
3103
3104 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3105                         struct btrfs_root *root,
3106                         struct inode *dir, u64 objectid,
3107                         const char *name, int name_len)
3108 {
3109         struct btrfs_path *path;
3110         struct extent_buffer *leaf;
3111         struct btrfs_dir_item *di;
3112         struct btrfs_key key;
3113         u64 index;
3114         int ret;
3115         u64 dir_ino = btrfs_ino(dir);
3116
3117         path = btrfs_alloc_path();
3118         if (!path)
3119                 return -ENOMEM;
3120
3121         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3122                                    name, name_len, -1);
3123         if (IS_ERR_OR_NULL(di)) {
3124                 if (!di)
3125                         ret = -ENOENT;
3126                 else
3127                         ret = PTR_ERR(di);
3128                 goto out;
3129         }
3130
3131         leaf = path->nodes[0];
3132         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3133         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3134         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3135         if (ret) {
3136                 btrfs_abort_transaction(trans, root, ret);
3137                 goto out;
3138         }
3139         btrfs_release_path(path);
3140
3141         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3142                                  objectid, root->root_key.objectid,
3143                                  dir_ino, &index, name, name_len);
3144         if (ret < 0) {
3145                 if (ret != -ENOENT) {
3146                         btrfs_abort_transaction(trans, root, ret);
3147                         goto out;
3148                 }
3149                 di = btrfs_search_dir_index_item(root, path, dir_ino,
3150                                                  name, name_len);
3151                 if (IS_ERR_OR_NULL(di)) {
3152                         if (!di)
3153                                 ret = -ENOENT;
3154                         else
3155                                 ret = PTR_ERR(di);
3156                         btrfs_abort_transaction(trans, root, ret);
3157                         goto out;
3158                 }
3159
3160                 leaf = path->nodes[0];
3161                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3162                 btrfs_release_path(path);
3163                 index = key.offset;
3164         }
3165         btrfs_release_path(path);
3166
3167         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3168         if (ret) {
3169                 btrfs_abort_transaction(trans, root, ret);
3170                 goto out;
3171         }
3172
3173         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3174         inode_inc_iversion(dir);
3175         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3176         ret = btrfs_update_inode(trans, root, dir);
3177         if (ret)
3178                 btrfs_abort_transaction(trans, root, ret);
3179 out:
3180         btrfs_free_path(path);
3181         return ret;
3182 }
3183
3184 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3185 {
3186         struct inode *inode = dentry->d_inode;
3187         int err = 0;
3188         struct btrfs_root *root = BTRFS_I(dir)->root;
3189         struct btrfs_trans_handle *trans;
3190         unsigned long nr = 0;
3191
3192         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3193             btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3194                 return -ENOTEMPTY;
3195
3196         trans = __unlink_start_trans(dir, dentry);
3197         if (IS_ERR(trans))
3198                 return PTR_ERR(trans);
3199
3200         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3201                 err = btrfs_unlink_subvol(trans, root, dir,
3202                                           BTRFS_I(inode)->location.objectid,
3203                                           dentry->d_name.name,
3204                                           dentry->d_name.len);
3205                 goto out;
3206         }
3207
3208         err = btrfs_orphan_add(trans, inode);
3209         if (err)
3210                 goto out;
3211
3212         /* now the directory is empty */
3213         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3214                                  dentry->d_name.name, dentry->d_name.len);
3215         if (!err)
3216                 btrfs_i_size_write(inode, 0);
3217 out:
3218         nr = trans->blocks_used;
3219         __unlink_end_trans(trans, root);
3220         btrfs_btree_balance_dirty(root, nr);
3221
3222         return err;
3223 }
3224
3225 /*
3226  * this can truncate away extent items, csum items and directory items.
3227  * It starts at a high offset and removes keys until it can't find
3228  * any higher than new_size
3229  *
3230  * csum items that cross the new i_size are truncated to the new size
3231  * as well.
3232  *
3233  * min_type is the minimum key type to truncate down to.  If set to 0, this
3234  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3235  */
3236 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3237                                struct btrfs_root *root,
3238                                struct inode *inode,
3239                                u64 new_size, u32 min_type)
3240 {
3241         struct btrfs_path *path;
3242         struct extent_buffer *leaf;
3243         struct btrfs_file_extent_item *fi;
3244         struct btrfs_key key;
3245         struct btrfs_key found_key;
3246         u64 extent_start = 0;
3247         u64 extent_num_bytes = 0;
3248         u64 extent_offset = 0;
3249         u64 item_end = 0;
3250         u64 mask = root->sectorsize - 1;
3251         u32 found_type = (u8)-1;
3252         int found_extent;
3253         int del_item;
3254         int pending_del_nr = 0;
3255         int pending_del_slot = 0;
3256         int extent_type = -1;
3257         int ret;
3258         int err = 0;
3259         u64 ino = btrfs_ino(inode);
3260
3261         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3262
3263         path = btrfs_alloc_path();
3264         if (!path)
3265                 return -ENOMEM;
3266         path->reada = -1;
3267
3268         if (root->ref_cows || root == root->fs_info->tree_root)
3269                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3270
3271         /*
3272          * This function is also used to drop the items in the log tree before
3273          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3274          * it is used to drop the loged items. So we shouldn't kill the delayed
3275          * items.
3276          */
3277         if (min_type == 0 && root == BTRFS_I(inode)->root)
3278                 btrfs_kill_delayed_inode_items(inode);
3279
3280         key.objectid = ino;
3281         key.offset = (u64)-1;
3282         key.type = (u8)-1;
3283
3284 search_again:
3285         path->leave_spinning = 1;
3286         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3287         if (ret < 0) {
3288                 err = ret;
3289                 goto out;
3290         }
3291
3292         if (ret > 0) {
3293                 /* there are no items in the tree for us to truncate, we're
3294                  * done
3295                  */
3296                 if (path->slots[0] == 0)
3297                         goto out;
3298                 path->slots[0]--;
3299         }
3300
3301         while (1) {
3302                 fi = NULL;
3303                 leaf = path->nodes[0];
3304                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3305                 found_type = btrfs_key_type(&found_key);
3306
3307                 if (found_key.objectid != ino)
3308                         break;
3309
3310                 if (found_type < min_type)
3311                         break;
3312
3313                 item_end = found_key.offset;
3314                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3315                         fi = btrfs_item_ptr(leaf, path->slots[0],
3316                                             struct btrfs_file_extent_item);
3317                         extent_type = btrfs_file_extent_type(leaf, fi);
3318                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3319                                 item_end +=
3320                                     btrfs_file_extent_num_bytes(leaf, fi);
3321                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3322                                 item_end += btrfs_file_extent_inline_len(leaf,
3323                                                                          fi);
3324                         }
3325                         item_end--;
3326                 }
3327                 if (found_type > min_type) {
3328                         del_item = 1;
3329                 } else {
3330                         if (item_end < new_size)
3331                                 break;
3332                         if (found_key.offset >= new_size)
3333                                 del_item = 1;
3334                         else
3335                                 del_item = 0;
3336                 }
3337                 found_extent = 0;
3338                 /* FIXME, shrink the extent if the ref count is only 1 */
3339                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3340                         goto delete;
3341
3342                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3343                         u64 num_dec;
3344                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3345                         if (!del_item) {
3346                                 u64 orig_num_bytes =
3347                                         btrfs_file_extent_num_bytes(leaf, fi);
3348                                 extent_num_bytes = new_size -
3349                                         found_key.offset + root->sectorsize - 1;
3350                                 extent_num_bytes = extent_num_bytes &
3351                                         ~((u64)root->sectorsize - 1);
3352                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3353                                                          extent_num_bytes);
3354                                 num_dec = (orig_num_bytes -
3355                                            extent_num_bytes);
3356                                 if (root->ref_cows && extent_start != 0)
3357                                         inode_sub_bytes(inode, num_dec);
3358                                 btrfs_mark_buffer_dirty(leaf);
3359                         } else {
3360                                 extent_num_bytes =
3361                                         btrfs_file_extent_disk_num_bytes(leaf,
3362                                                                          fi);
3363                                 extent_offset = found_key.offset -
3364                                         btrfs_file_extent_offset(leaf, fi);
3365
3366                                 /* FIXME blocksize != 4096 */
3367                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3368                                 if (extent_start != 0) {
3369                                         found_extent = 1;
3370                                         if (root->ref_cows)
3371                                                 inode_sub_bytes(inode, num_dec);
3372                                 }
3373                         }
3374                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3375                         /*
3376                          * we can't truncate inline items that have had
3377                          * special encodings
3378                          */
3379                         if (!del_item &&
3380                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3381                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3382                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3383                                 u32 size = new_size - found_key.offset;
3384
3385                                 if (root->ref_cows) {
3386                                         inode_sub_bytes(inode, item_end + 1 -
3387                                                         new_size);
3388                                 }
3389                                 size =
3390                                     btrfs_file_extent_calc_inline_size(size);
3391                                 btrfs_truncate_item(trans, root, path,
3392                                                     size, 1);
3393                         } else if (root->ref_cows) {
3394                                 inode_sub_bytes(inode, item_end + 1 -
3395                                                 found_key.offset);
3396                         }
3397                 }
3398 delete:
3399                 if (del_item) {
3400                         if (!pending_del_nr) {
3401                                 /* no pending yet, add ourselves */
3402                                 pending_del_slot = path->slots[0];
3403                                 pending_del_nr = 1;
3404                         } else if (pending_del_nr &&
3405                                    path->slots[0] + 1 == pending_del_slot) {
3406                                 /* hop on the pending chunk */
3407                                 pending_del_nr++;
3408                                 pending_del_slot = path->slots[0];
3409                         } else {
3410                                 BUG();
3411                         }
3412                 } else {
3413                         break;
3414                 }
3415                 if (found_extent && (root->ref_cows ||
3416                                      root == root->fs_info->tree_root)) {
3417                         btrfs_set_path_blocking(path);
3418                         ret = btrfs_free_extent(trans, root, extent_start,
3419                                                 extent_num_bytes, 0,
3420                                                 btrfs_header_owner(leaf),
3421                                                 ino, extent_offset, 0);
3422                         BUG_ON(ret);
3423                 }
3424
3425                 if (found_type == BTRFS_INODE_ITEM_KEY)
3426                         break;
3427
3428                 if (path->slots[0] == 0 ||
3429                     path->slots[0] != pending_del_slot) {
3430                         if (root->ref_cows &&
3431                             BTRFS_I(inode)->location.objectid !=
3432                                                 BTRFS_FREE_INO_OBJECTID) {
3433                                 err = -EAGAIN;
3434                                 goto out;
3435                         }
3436                         if (pending_del_nr) {
3437                                 ret = btrfs_del_items(trans, root, path,
3438                                                 pending_del_slot,
3439                                                 pending_del_nr);
3440                                 if (ret) {
3441                                         btrfs_abort_transaction(trans,
3442                                                                 root, ret);
3443                                         goto error;
3444                                 }
3445                                 pending_del_nr = 0;
3446                         }
3447                         btrfs_release_path(path);
3448                         goto search_again;
3449                 } else {
3450                         path->slots[0]--;
3451                 }
3452         }
3453 out:
3454         if (pending_del_nr) {
3455                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3456                                       pending_del_nr);
3457                 if (ret)
3458                         btrfs_abort_transaction(trans, root, ret);
3459         }
3460 error:
3461         btrfs_free_path(path);
3462         return err;
3463 }
3464
3465 /*
3466  * taken from block_truncate_page, but does cow as it zeros out
3467  * any bytes left in the last page in the file.
3468  */
3469 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3470 {
3471         struct inode *inode = mapping->host;
3472         struct btrfs_root *root = BTRFS_I(inode)->root;
3473         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3474         struct btrfs_ordered_extent *ordered;
3475         struct extent_state *cached_state = NULL;
3476         char *kaddr;
3477         u32 blocksize = root->sectorsize;
3478         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3479         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3480         struct page *page;
3481         gfp_t mask = btrfs_alloc_write_mask(mapping);
3482         int ret = 0;
3483         u64 page_start;
3484         u64 page_end;
3485
3486         if ((offset & (blocksize - 1)) == 0)
3487                 goto out;
3488         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3489         if (ret)
3490                 goto out;
3491
3492         ret = -ENOMEM;
3493 again:
3494         page = find_or_create_page(mapping, index, mask);
3495         if (!page) {
3496                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3497                 goto out;
3498         }
3499
3500         page_start = page_offset(page);
3501         page_end = page_start + PAGE_CACHE_SIZE - 1;
3502
3503         if (!PageUptodate(page)) {
3504                 ret = btrfs_readpage(NULL, page);
3505                 lock_page(page);
3506                 if (page->mapping != mapping) {
3507                         unlock_page(page);
3508                         page_cache_release(page);
3509                         goto again;
3510                 }
3511                 if (!PageUptodate(page)) {
3512                         ret = -EIO;
3513                         goto out_unlock;
3514                 }
3515         }
3516         wait_on_page_writeback(page);
3517
3518         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
3519         set_page_extent_mapped(page);
3520
3521         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3522         if (ordered) {
3523                 unlock_extent_cached(io_tree, page_start, page_end,
3524                                      &cached_state, GFP_NOFS);
3525                 unlock_page(page);
3526                 page_cache_release(page);
3527                 btrfs_start_ordered_extent(inode, ordered, 1);
3528                 btrfs_put_ordered_extent(ordered);
3529                 goto again;
3530         }
3531
3532         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3533                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3534                           0, 0, &cached_state, GFP_NOFS);
3535
3536         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3537                                         &cached_state);
3538         if (ret) {
3539                 unlock_extent_cached(io_tree, page_start, page_end,
3540                                      &cached_state, GFP_NOFS);
3541                 goto out_unlock;
3542         }
3543
3544         ret = 0;
3545         if (offset != PAGE_CACHE_SIZE) {
3546                 kaddr = kmap(page);
3547                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3548                 flush_dcache_page(page);
3549                 kunmap(page);
3550         }
3551         ClearPageChecked(page);
3552         set_page_dirty(page);
3553         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3554                              GFP_NOFS);
3555
3556 out_unlock:
3557         if (ret)
3558                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3559         unlock_page(page);
3560         page_cache_release(page);
3561 out:
3562         return ret;
3563 }
3564
3565 /*
3566  * This function puts in dummy file extents for the area we're creating a hole
3567  * for.  So if we are truncating this file to a larger size we need to insert
3568  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3569  * the range between oldsize and size
3570  */
3571 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3572 {
3573         struct btrfs_trans_handle *trans;
3574         struct btrfs_root *root = BTRFS_I(inode)->root;
3575         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3576         struct extent_map *em = NULL;
3577         struct extent_state *cached_state = NULL;
3578         u64 mask = root->sectorsize - 1;
3579         u64 hole_start = (oldsize + mask) & ~mask;
3580         u64 block_end = (size + mask) & ~mask;
3581         u64 last_byte;
3582         u64 cur_offset;
3583         u64 hole_size;
3584         int err = 0;
3585
3586         if (size <= hole_start)
3587                 return 0;
3588
3589         while (1) {
3590                 struct btrfs_ordered_extent *ordered;
3591                 btrfs_wait_ordered_range(inode, hole_start,
3592                                          block_end - hole_start);
3593                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3594                                  &cached_state);
3595                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3596                 if (!ordered)
3597                         break;
3598                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3599                                      &cached_state, GFP_NOFS);
3600                 btrfs_put_ordered_extent(ordered);
3601         }
3602
3603         cur_offset = hole_start;
3604         while (1) {
3605                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3606                                 block_end - cur_offset, 0);
3607                 if (IS_ERR(em)) {
3608                         err = PTR_ERR(em);
3609                         break;
3610                 }
3611                 last_byte = min(extent_map_end(em), block_end);
3612                 last_byte = (last_byte + mask) & ~mask;
3613                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3614                         u64 hint_byte = 0;
3615                         hole_size = last_byte - cur_offset;
3616
3617                         trans = btrfs_start_transaction(root, 3);
3618                         if (IS_ERR(trans)) {
3619                                 err = PTR_ERR(trans);
3620                                 break;
3621                         }
3622
3623                         err = btrfs_drop_extents(trans, inode, cur_offset,
3624                                                  cur_offset + hole_size,
3625                                                  &hint_byte, 1);
3626                         if (err) {
3627                                 btrfs_abort_transaction(trans, root, err);
3628                                 btrfs_end_transaction(trans, root);
3629                                 break;
3630                         }
3631
3632                         err = btrfs_insert_file_extent(trans, root,
3633                                         btrfs_ino(inode), cur_offset, 0,
3634                                         0, hole_size, 0, hole_size,
3635                                         0, 0, 0);
3636                         if (err) {
3637                                 btrfs_abort_transaction(trans, root, err);
3638                                 btrfs_end_transaction(trans, root);
3639                                 break;
3640                         }
3641
3642                         btrfs_drop_extent_cache(inode, hole_start,
3643                                         last_byte - 1, 0);
3644
3645                         btrfs_update_inode(trans, root, inode);
3646                         btrfs_end_transaction(trans, root);
3647                 }
3648                 free_extent_map(em);
3649                 em = NULL;
3650                 cur_offset = last_byte;
3651                 if (cur_offset >= block_end)
3652                         break;
3653         }
3654
3655         free_extent_map(em);
3656         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3657                              GFP_NOFS);
3658         return err;
3659 }
3660
3661 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3662 {
3663         struct btrfs_root *root = BTRFS_I(inode)->root;
3664         struct btrfs_trans_handle *trans;
3665         loff_t oldsize = i_size_read(inode);
3666         int ret;
3667
3668         if (newsize == oldsize)
3669                 return 0;
3670
3671         if (newsize > oldsize) {
3672                 truncate_pagecache(inode, oldsize, newsize);
3673                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3674                 if (ret)
3675                         return ret;
3676
3677                 trans = btrfs_start_transaction(root, 1);
3678                 if (IS_ERR(trans))
3679                         return PTR_ERR(trans);
3680
3681                 i_size_write(inode, newsize);
3682                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3683                 ret = btrfs_update_inode(trans, root, inode);
3684                 btrfs_end_transaction(trans, root);
3685         } else {
3686
3687                 /*
3688                  * We're truncating a file that used to have good data down to
3689                  * zero. Make sure it gets into the ordered flush list so that
3690                  * any new writes get down to disk quickly.
3691                  */
3692                 if (newsize == 0)
3693                         set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
3694                                 &BTRFS_I(inode)->runtime_flags);
3695
3696                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3697                 truncate_setsize(inode, newsize);
3698                 ret = btrfs_truncate(inode);
3699         }
3700
3701         return ret;
3702 }
3703
3704 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3705 {
3706         struct inode *inode = dentry->d_inode;
3707         struct btrfs_root *root = BTRFS_I(inode)->root;
3708         int err;
3709
3710         if (btrfs_root_readonly(root))
3711                 return -EROFS;
3712
3713         err = inode_change_ok(inode, attr);
3714         if (err)
3715                 return err;
3716
3717         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3718                 err = btrfs_setsize(inode, attr->ia_size);
3719                 if (err)
3720                         return err;
3721         }
3722
3723         if (attr->ia_valid) {
3724                 setattr_copy(inode, attr);
3725                 inode_inc_iversion(inode);
3726                 err = btrfs_dirty_inode(inode);
3727
3728                 if (!err && attr->ia_valid & ATTR_MODE)
3729                         err = btrfs_acl_chmod(inode);
3730         }
3731
3732         return err;
3733 }
3734
3735 void btrfs_evict_inode(struct inode *inode)
3736 {
3737         struct btrfs_trans_handle *trans;
3738         struct btrfs_root *root = BTRFS_I(inode)->root;
3739         struct btrfs_block_rsv *rsv, *global_rsv;
3740         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
3741         unsigned long nr;
3742         int ret;
3743
3744         trace_btrfs_inode_evict(inode);
3745
3746         truncate_inode_pages(&inode->i_data, 0);
3747         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3748                                btrfs_is_free_space_inode(inode)))
3749                 goto no_delete;
3750
3751         if (is_bad_inode(inode)) {
3752                 btrfs_orphan_del(NULL, inode);
3753                 goto no_delete;
3754         }
3755         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3756         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3757
3758         if (root->fs_info->log_root_recovering) {
3759                 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3760                                  &BTRFS_I(inode)->runtime_flags));
3761                 goto no_delete;
3762         }
3763
3764         if (inode->i_nlink > 0) {
3765                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3766                 goto no_delete;
3767         }
3768
3769         rsv = btrfs_alloc_block_rsv(root);
3770         if (!rsv) {
3771                 btrfs_orphan_del(NULL, inode);
3772                 goto no_delete;
3773         }
3774         rsv->size = min_size;
3775         global_rsv = &root->fs_info->global_block_rsv;
3776
3777         btrfs_i_size_write(inode, 0);
3778
3779         /*
3780          * This is a bit simpler than btrfs_truncate since
3781          *
3782          * 1) We've already reserved our space for our orphan item in the
3783          *    unlink.
3784          * 2) We're going to delete the inode item, so we don't need to update
3785          *    it at all.
3786          *
3787          * So we just need to reserve some slack space in case we add bytes when
3788          * doing the truncate.
3789          */
3790         while (1) {
3791                 ret = btrfs_block_rsv_refill_noflush(root, rsv, min_size);
3792
3793                 /*
3794                  * Try and steal from the global reserve since we will
3795                  * likely not use this space anyway, we want to try as
3796                  * hard as possible to get this to work.
3797                  */
3798                 if (ret)
3799                         ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
3800
3801                 if (ret) {
3802                         printk(KERN_WARNING "Could not get space for a "
3803                                "delete, will truncate on mount %d\n", ret);
3804                         btrfs_orphan_del(NULL, inode);
3805                         btrfs_free_block_rsv(root, rsv);
3806                         goto no_delete;
3807                 }
3808
3809                 trans = btrfs_start_transaction(root, 0);
3810                 if (IS_ERR(trans)) {
3811                         btrfs_orphan_del(NULL, inode);
3812                         btrfs_free_block_rsv(root, rsv);
3813                         goto no_delete;
3814                 }
3815
3816                 trans->block_rsv = rsv;
3817
3818                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3819                 if (ret != -EAGAIN)
3820                         break;
3821
3822                 nr = trans->blocks_used;
3823                 btrfs_end_transaction(trans, root);
3824                 trans = NULL;
3825                 btrfs_btree_balance_dirty(root, nr);
3826         }
3827
3828         btrfs_free_block_rsv(root, rsv);
3829
3830         if (ret == 0) {
3831                 trans->block_rsv = root->orphan_block_rsv;
3832                 ret = btrfs_orphan_del(trans, inode);
3833                 BUG_ON(ret);
3834         }
3835
3836         trans->block_rsv = &root->fs_info->trans_block_rsv;
3837         if (!(root == root->fs_info->tree_root ||
3838               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3839                 btrfs_return_ino(root, btrfs_ino(inode));
3840
3841         nr = trans->blocks_used;
3842         btrfs_end_transaction(trans, root);
3843         btrfs_btree_balance_dirty(root, nr);
3844 no_delete:
3845         clear_inode(inode);
3846         return;
3847 }
3848
3849 /*
3850  * this returns the key found in the dir entry in the location pointer.
3851  * If no dir entries were found, location->objectid is 0.
3852  */
3853 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3854                                struct btrfs_key *location)
3855 {
3856         const char *name = dentry->d_name.name;
3857         int namelen = dentry->d_name.len;
3858         struct btrfs_dir_item *di;
3859         struct btrfs_path *path;
3860         struct btrfs_root *root = BTRFS_I(dir)->root;
3861         int ret = 0;
3862
3863         path = btrfs_alloc_path();
3864         if (!path)
3865                 return -ENOMEM;
3866
3867         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3868                                     namelen, 0);
3869         if (IS_ERR(di))
3870                 ret = PTR_ERR(di);
3871
3872         if (IS_ERR_OR_NULL(di))
3873                 goto out_err;
3874
3875         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3876 out:
3877         btrfs_free_path(path);
3878         return ret;
3879 out_err:
3880         location->objectid = 0;
3881         goto out;
3882 }
3883
3884 /*
3885  * when we hit a tree root in a directory, the btrfs part of the inode
3886  * needs to be changed to reflect the root directory of the tree root.  This
3887  * is kind of like crossing a mount point.
3888  */
3889 static int fixup_tree_root_location(struct btrfs_root *root,
3890                                     struct inode *dir,
3891                                     struct dentry *dentry,
3892                                     struct btrfs_key *location,
3893                                     struct btrfs_root **sub_root)
3894 {
3895         struct btrfs_path *path;
3896         struct btrfs_root *new_root;
3897         struct btrfs_root_ref *ref;
3898         struct extent_buffer *leaf;
3899         int ret;
3900         int err = 0;
3901
3902         path = btrfs_alloc_path();
3903         if (!path) {
3904                 err = -ENOMEM;
3905                 goto out;
3906         }
3907
3908         err = -ENOENT;
3909         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3910                                   BTRFS_I(dir)->root->root_key.objectid,
3911                                   location->objectid);
3912         if (ret) {
3913                 if (ret < 0)
3914                         err = ret;
3915                 goto out;
3916         }
3917
3918         leaf = path->nodes[0];
3919         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3920         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3921             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3922                 goto out;
3923
3924         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3925                                    (unsigned long)(ref + 1),
3926                                    dentry->d_name.len);
3927         if (ret)
3928                 goto out;
3929
3930         btrfs_release_path(path);
3931
3932         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3933         if (IS_ERR(new_root)) {
3934                 err = PTR_ERR(new_root);
3935                 goto out;
3936         }
3937
3938         if (btrfs_root_refs(&new_root->root_item) == 0) {
3939                 err = -ENOENT;
3940                 goto out;
3941         }
3942
3943         *sub_root = new_root;
3944         location->objectid = btrfs_root_dirid(&new_root->root_item);
3945         location->type = BTRFS_INODE_ITEM_KEY;
3946         location->offset = 0;
3947         err = 0;
3948 out:
3949         btrfs_free_path(path);
3950         return err;
3951 }
3952
3953 static void inode_tree_add(struct inode *inode)
3954 {
3955         struct btrfs_root *root = BTRFS_I(inode)->root;
3956         struct btrfs_inode *entry;
3957         struct rb_node **p;
3958         struct rb_node *parent;
3959         u64 ino = btrfs_ino(inode);
3960 again:
3961         p = &root->inode_tree.rb_node;
3962         parent = NULL;
3963
3964         if (inode_unhashed(inode))
3965                 return;
3966
3967         spin_lock(&root->inode_lock);
3968         while (*p) {
3969                 parent = *p;
3970                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3971
3972                 if (ino < btrfs_ino(&entry->vfs_inode))
3973                         p = &parent->rb_left;
3974                 else if (ino > btrfs_ino(&entry->vfs_inode))
3975                         p = &parent->rb_right;
3976                 else {
3977                         WARN_ON(!(entry->vfs_inode.i_state &
3978                                   (I_WILL_FREE | I_FREEING)));
3979                         rb_erase(parent, &root->inode_tree);
3980                         RB_CLEAR_NODE(parent);
3981                         spin_unlock(&root->inode_lock);
3982                         goto again;
3983                 }
3984         }
3985         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3986         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3987         spin_unlock(&root->inode_lock);
3988 }
3989
3990 static void inode_tree_del(struct inode *inode)
3991 {
3992         struct btrfs_root *root = BTRFS_I(inode)->root;
3993         int empty = 0;
3994
3995         spin_lock(&root->inode_lock);
3996         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3997                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3998                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3999                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4000         }
4001         spin_unlock(&root->inode_lock);
4002
4003         /*
4004          * Free space cache has inodes in the tree root, but the tree root has a
4005          * root_refs of 0, so this could end up dropping the tree root as a
4006          * snapshot, so we need the extra !root->fs_info->tree_root check to
4007          * make sure we don't drop it.
4008          */
4009         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
4010             root != root->fs_info->tree_root) {
4011                 synchronize_srcu(&root->fs_info->subvol_srcu);
4012                 spin_lock(&root->inode_lock);
4013                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4014                 spin_unlock(&root->inode_lock);
4015                 if (empty)
4016                         btrfs_add_dead_root(root);
4017         }
4018 }
4019
4020 void btrfs_invalidate_inodes(struct btrfs_root *root)
4021 {
4022         struct rb_node *node;
4023         struct rb_node *prev;
4024         struct btrfs_inode *entry;
4025         struct inode *inode;
4026         u64 objectid = 0;
4027
4028         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4029
4030         spin_lock(&root->inode_lock);
4031 again:
4032         node = root->inode_tree.rb_node;
4033         prev = NULL;
4034         while (node) {
4035                 prev = node;
4036                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4037
4038                 if (objectid < btrfs_ino(&entry->vfs_inode))
4039                         node = node->rb_left;
4040                 else if (objectid > btrfs_ino(&entry->vfs_inode))
4041                         node = node->rb_right;
4042                 else
4043                         break;
4044         }
4045         if (!node) {
4046                 while (prev) {
4047                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
4048                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4049                                 node = prev;
4050                                 break;
4051                         }
4052                         prev = rb_next(prev);
4053                 }
4054         }
4055         while (node) {
4056                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4057                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4058                 inode = igrab(&entry->vfs_inode);
4059                 if (inode) {
4060                         spin_unlock(&root->inode_lock);
4061                         if (atomic_read(&inode->i_count) > 1)
4062                                 d_prune_aliases(inode);
4063                         /*
4064                          * btrfs_drop_inode will have it removed from
4065                          * the inode cache when its usage count
4066                          * hits zero.
4067                          */
4068                         iput(inode);
4069                         cond_resched();
4070                         spin_lock(&root->inode_lock);
4071                         goto again;
4072                 }
4073
4074                 if (cond_resched_lock(&root->inode_lock))
4075                         goto again;
4076
4077                 node = rb_next(node);
4078         }
4079         spin_unlock(&root->inode_lock);
4080 }
4081
4082 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4083 {
4084         struct btrfs_iget_args *args = p;
4085         inode->i_ino = args->ino;
4086         BTRFS_I(inode)->root = args->root;
4087         return 0;
4088 }
4089
4090 static int btrfs_find_actor(struct inode *inode, void *opaque)
4091 {
4092         struct btrfs_iget_args *args = opaque;
4093         return args->ino == btrfs_ino(inode) &&
4094                 args->root == BTRFS_I(inode)->root;
4095 }
4096
4097 static struct inode *btrfs_iget_locked(struct super_block *s,
4098                                        u64 objectid,
4099                                        struct btrfs_root *root)
4100 {
4101         struct inode *inode;
4102         struct btrfs_iget_args args;
4103         args.ino = objectid;
4104         args.root = root;
4105
4106         inode = iget5_locked(s, objectid, btrfs_find_actor,
4107                              btrfs_init_locked_inode,
4108                              (void *)&args);
4109         return inode;
4110 }
4111
4112 /* Get an inode object given its location and corresponding root.
4113  * Returns in *is_new if the inode was read from disk
4114  */
4115 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4116                          struct btrfs_root *root, int *new)
4117 {
4118         struct inode *inode;
4119
4120         inode = btrfs_iget_locked(s, location->objectid, root);
4121         if (!inode)
4122                 return ERR_PTR(-ENOMEM);
4123
4124         if (inode->i_state & I_NEW) {
4125                 BTRFS_I(inode)->root = root;
4126                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4127                 btrfs_read_locked_inode(inode);
4128                 if (!is_bad_inode(inode)) {
4129                         inode_tree_add(inode);
4130                         unlock_new_inode(inode);
4131                         if (new)
4132                                 *new = 1;
4133                 } else {
4134                         unlock_new_inode(inode);
4135                         iput(inode);
4136                         inode = ERR_PTR(-ESTALE);
4137                 }
4138         }
4139
4140         return inode;
4141 }
4142
4143 static struct inode *new_simple_dir(struct super_block *s,
4144                                     struct btrfs_key *key,
4145                                     struct btrfs_root *root)
4146 {
4147         struct inode *inode = new_inode(s);
4148
4149         if (!inode)
4150                 return ERR_PTR(-ENOMEM);
4151
4152         BTRFS_I(inode)->root = root;
4153         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4154         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4155
4156         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4157         inode->i_op = &btrfs_dir_ro_inode_operations;
4158         inode->i_fop = &simple_dir_operations;
4159         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4160         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4161
4162         return inode;
4163 }
4164
4165 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4166 {
4167         struct inode *inode;
4168         struct btrfs_root *root = BTRFS_I(dir)->root;
4169         struct btrfs_root *sub_root = root;
4170         struct btrfs_key location;
4171         int index;
4172         int ret = 0;
4173
4174         if (dentry->d_name.len > BTRFS_NAME_LEN)
4175                 return ERR_PTR(-ENAMETOOLONG);
4176
4177         if (unlikely(d_need_lookup(dentry))) {
4178                 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
4179                 kfree(dentry->d_fsdata);
4180                 dentry->d_fsdata = NULL;
4181                 /* This thing is hashed, drop it for now */
4182                 d_drop(dentry);
4183         } else {
4184                 ret = btrfs_inode_by_name(dir, dentry, &location);
4185         }
4186
4187         if (ret < 0)
4188                 return ERR_PTR(ret);
4189
4190         if (location.objectid == 0)
4191                 return NULL;
4192
4193         if (location.type == BTRFS_INODE_ITEM_KEY) {
4194                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4195                 return inode;
4196         }
4197
4198         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4199
4200         index = srcu_read_lock(&root->fs_info->subvol_srcu);
4201         ret = fixup_tree_root_location(root, dir, dentry,
4202                                        &location, &sub_root);
4203         if (ret < 0) {
4204                 if (ret != -ENOENT)
4205                         inode = ERR_PTR(ret);
4206                 else
4207                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
4208         } else {
4209                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4210         }
4211         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4212
4213         if (!IS_ERR(inode) && root != sub_root) {
4214                 down_read(&root->fs_info->cleanup_work_sem);
4215                 if (!(inode->i_sb->s_flags & MS_RDONLY))
4216                         ret = btrfs_orphan_cleanup(sub_root);
4217                 up_read(&root->fs_info->cleanup_work_sem);
4218                 if (ret)
4219                         inode = ERR_PTR(ret);
4220         }
4221
4222         return inode;
4223 }
4224
4225 static int btrfs_dentry_delete(const struct dentry *dentry)
4226 {
4227         struct btrfs_root *root;
4228         struct inode *inode = dentry->d_inode;
4229
4230         if (!inode && !IS_ROOT(dentry))
4231                 inode = dentry->d_parent->d_inode;
4232
4233         if (inode) {
4234                 root = BTRFS_I(inode)->root;
4235                 if (btrfs_root_refs(&root->root_item) == 0)
4236                         return 1;
4237
4238                 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
4239                         return 1;
4240         }
4241         return 0;
4242 }
4243
4244 static void btrfs_dentry_release(struct dentry *dentry)
4245 {
4246         if (dentry->d_fsdata)
4247                 kfree(dentry->d_fsdata);
4248 }
4249
4250 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4251                                    struct nameidata *nd)
4252 {
4253         struct dentry *ret;
4254
4255         ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4256         if (unlikely(d_need_lookup(dentry))) {
4257                 spin_lock(&dentry->d_lock);
4258                 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4259                 spin_unlock(&dentry->d_lock);
4260         }
4261         return ret;
4262 }
4263
4264 unsigned char btrfs_filetype_table[] = {
4265         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4266 };
4267
4268 static int btrfs_real_readdir(struct file *filp, void *dirent,
4269                               filldir_t filldir)
4270 {
4271         struct inode *inode = filp->f_dentry->d_inode;
4272         struct btrfs_root *root = BTRFS_I(inode)->root;
4273         struct btrfs_item *item;
4274         struct btrfs_dir_item *di;
4275         struct btrfs_key key;
4276         struct btrfs_key found_key;
4277         struct btrfs_path *path;
4278         struct list_head ins_list;
4279         struct list_head del_list;
4280         int ret;
4281         struct extent_buffer *leaf;
4282         int slot;
4283         unsigned char d_type;
4284         int over = 0;
4285         u32 di_cur;
4286         u32 di_total;
4287         u32 di_len;
4288         int key_type = BTRFS_DIR_INDEX_KEY;
4289         char tmp_name[32];
4290         char *name_ptr;
4291         int name_len;
4292         int is_curr = 0;        /* filp->f_pos points to the current index? */
4293
4294         /* FIXME, use a real flag for deciding about the key type */
4295         if (root->fs_info->tree_root == root)
4296                 key_type = BTRFS_DIR_ITEM_KEY;
4297
4298         /* special case for "." */
4299         if (filp->f_pos == 0) {
4300                 over = filldir(dirent, ".", 1,
4301                                filp->f_pos, btrfs_ino(inode), DT_DIR);
4302                 if (over)
4303                         return 0;
4304                 filp->f_pos = 1;
4305         }
4306         /* special case for .., just use the back ref */
4307         if (filp->f_pos == 1) {
4308                 u64 pino = parent_ino(filp->f_path.dentry);
4309                 over = filldir(dirent, "..", 2,
4310                                filp->f_pos, pino, DT_DIR);
4311                 if (over)
4312                         return 0;
4313                 filp->f_pos = 2;
4314         }
4315         path = btrfs_alloc_path();
4316         if (!path)
4317                 return -ENOMEM;
4318
4319         path->reada = 1;
4320
4321         if (key_type == BTRFS_DIR_INDEX_KEY) {
4322                 INIT_LIST_HEAD(&ins_list);
4323                 INIT_LIST_HEAD(&del_list);
4324                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4325         }
4326
4327         btrfs_set_key_type(&key, key_type);
4328         key.offset = filp->f_pos;
4329         key.objectid = btrfs_ino(inode);
4330
4331         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4332         if (ret < 0)
4333                 goto err;
4334
4335         while (1) {
4336                 leaf = path->nodes[0];
4337                 slot = path->slots[0];
4338                 if (slot >= btrfs_header_nritems(leaf)) {
4339                         ret = btrfs_next_leaf(root, path);
4340                         if (ret < 0)
4341                                 goto err;
4342                         else if (ret > 0)
4343                                 break;
4344                         continue;
4345                 }
4346
4347                 item = btrfs_item_nr(leaf, slot);
4348                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4349
4350                 if (found_key.objectid != key.objectid)
4351                         break;
4352                 if (btrfs_key_type(&found_key) != key_type)
4353                         break;
4354                 if (found_key.offset < filp->f_pos)
4355                         goto next;
4356                 if (key_type == BTRFS_DIR_INDEX_KEY &&
4357                     btrfs_should_delete_dir_index(&del_list,
4358                                                   found_key.offset))
4359                         goto next;
4360
4361                 filp->f_pos = found_key.offset;
4362                 is_curr = 1;
4363
4364                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4365                 di_cur = 0;
4366                 di_total = btrfs_item_size(leaf, item);
4367
4368                 while (di_cur < di_total) {
4369                         struct btrfs_key location;
4370
4371                         if (verify_dir_item(root, leaf, di))
4372                                 break;
4373
4374                         name_len = btrfs_dir_name_len(leaf, di);
4375                         if (name_len <= sizeof(tmp_name)) {
4376                                 name_ptr = tmp_name;
4377                         } else {
4378                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4379                                 if (!name_ptr) {
4380                                         ret = -ENOMEM;
4381                                         goto err;
4382                                 }
4383                         }
4384                         read_extent_buffer(leaf, name_ptr,
4385                                            (unsigned long)(di + 1), name_len);
4386
4387                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4388                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4389
4390
4391                         /* is this a reference to our own snapshot? If so
4392                          * skip it.
4393                          *
4394                          * In contrast to old kernels, we insert the snapshot's
4395                          * dir item and dir index after it has been created, so
4396                          * we won't find a reference to our own snapshot. We
4397                          * still keep the following code for backward
4398                          * compatibility.
4399                          */
4400                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4401                             location.objectid == root->root_key.objectid) {
4402                                 over = 0;
4403                                 goto skip;
4404                         }
4405                         over = filldir(dirent, name_ptr, name_len,
4406                                        found_key.offset, location.objectid,
4407                                        d_type);
4408
4409 skip:
4410                         if (name_ptr != tmp_name)
4411                                 kfree(name_ptr);
4412
4413                         if (over)
4414                                 goto nopos;
4415                         di_len = btrfs_dir_name_len(leaf, di) +
4416                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4417                         di_cur += di_len;
4418                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4419                 }
4420 next:
4421                 path->slots[0]++;
4422         }
4423
4424         if (key_type == BTRFS_DIR_INDEX_KEY) {
4425                 if (is_curr)
4426                         filp->f_pos++;
4427                 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4428                                                       &ins_list);
4429                 if (ret)
4430                         goto nopos;
4431         }
4432
4433         /* Reached end of directory/root. Bump pos past the last item. */
4434         if (key_type == BTRFS_DIR_INDEX_KEY)
4435                 /*
4436                  * 32-bit glibc will use getdents64, but then strtol -
4437                  * so the last number we can serve is this.
4438                  */
4439                 filp->f_pos = 0x7fffffff;
4440         else
4441                 filp->f_pos++;
4442 nopos:
4443         ret = 0;
4444 err:
4445         if (key_type == BTRFS_DIR_INDEX_KEY)
4446                 btrfs_put_delayed_items(&ins_list, &del_list);
4447         btrfs_free_path(path);
4448         return ret;
4449 }
4450
4451 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4452 {
4453         struct btrfs_root *root = BTRFS_I(inode)->root;
4454         struct btrfs_trans_handle *trans;
4455         int ret = 0;
4456         bool nolock = false;
4457
4458         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4459                 return 0;
4460
4461         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
4462                 nolock = true;
4463
4464         if (wbc->sync_mode == WB_SYNC_ALL) {
4465                 if (nolock)
4466                         trans = btrfs_join_transaction_nolock(root);
4467                 else
4468                         trans = btrfs_join_transaction(root);
4469                 if (IS_ERR(trans))
4470                         return PTR_ERR(trans);
4471                 if (nolock)
4472                         ret = btrfs_end_transaction_nolock(trans, root);
4473                 else
4474                         ret = btrfs_commit_transaction(trans, root);
4475         }
4476         return ret;
4477 }
4478
4479 /*
4480  * This is somewhat expensive, updating the tree every time the
4481  * inode changes.  But, it is most likely to find the inode in cache.
4482  * FIXME, needs more benchmarking...there are no reasons other than performance
4483  * to keep or drop this code.
4484  */
4485 int btrfs_dirty_inode(struct inode *inode)
4486 {
4487         struct btrfs_root *root = BTRFS_I(inode)->root;
4488         struct btrfs_trans_handle *trans;
4489         int ret;
4490
4491         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4492                 return 0;
4493
4494         trans = btrfs_join_transaction(root);
4495         if (IS_ERR(trans))
4496                 return PTR_ERR(trans);
4497
4498         ret = btrfs_update_inode(trans, root, inode);
4499         if (ret && ret == -ENOSPC) {
4500                 /* whoops, lets try again with the full transaction */
4501                 btrfs_end_transaction(trans, root);
4502                 trans = btrfs_start_transaction(root, 1);
4503                 if (IS_ERR(trans))
4504                         return PTR_ERR(trans);
4505
4506                 ret = btrfs_update_inode(trans, root, inode);
4507         }
4508         btrfs_end_transaction(trans, root);
4509         if (BTRFS_I(inode)->delayed_node)
4510                 btrfs_balance_delayed_items(root);
4511
4512         return ret;
4513 }
4514
4515 /*
4516  * This is a copy of file_update_time.  We need this so we can return error on
4517  * ENOSPC for updating the inode in the case of file write and mmap writes.
4518  */
4519 static int btrfs_update_time(struct inode *inode, struct timespec *now,
4520                              int flags)
4521 {
4522         struct btrfs_root *root = BTRFS_I(inode)->root;
4523
4524         if (btrfs_root_readonly(root))
4525                 return -EROFS;
4526
4527         if (flags & S_VERSION)
4528                 inode_inc_iversion(inode);
4529         if (flags & S_CTIME)
4530                 inode->i_ctime = *now;
4531         if (flags & S_MTIME)
4532                 inode->i_mtime = *now;
4533         if (flags & S_ATIME)
4534                 inode->i_atime = *now;
4535         return btrfs_dirty_inode(inode);
4536 }
4537
4538 /*
4539  * find the highest existing sequence number in a directory
4540  * and then set the in-memory index_cnt variable to reflect
4541  * free sequence numbers
4542  */
4543 static int btrfs_set_inode_index_count(struct inode *inode)
4544 {
4545         struct btrfs_root *root = BTRFS_I(inode)->root;
4546         struct btrfs_key key, found_key;
4547         struct btrfs_path *path;
4548         struct extent_buffer *leaf;
4549         int ret;
4550
4551         key.objectid = btrfs_ino(inode);
4552         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4553         key.offset = (u64)-1;
4554
4555         path = btrfs_alloc_path();
4556         if (!path)
4557                 return -ENOMEM;
4558
4559         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4560         if (ret < 0)
4561                 goto out;
4562         /* FIXME: we should be able to handle this */
4563         if (ret == 0)
4564                 goto out;
4565         ret = 0;
4566
4567         /*
4568          * MAGIC NUMBER EXPLANATION:
4569          * since we search a directory based on f_pos we have to start at 2
4570          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4571          * else has to start at 2
4572          */
4573         if (path->slots[0] == 0) {
4574                 BTRFS_I(inode)->index_cnt = 2;
4575                 goto out;
4576         }
4577
4578         path->slots[0]--;
4579
4580         leaf = path->nodes[0];
4581         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4582
4583         if (found_key.objectid != btrfs_ino(inode) ||
4584             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4585                 BTRFS_I(inode)->index_cnt = 2;
4586                 goto out;
4587         }
4588
4589         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4590 out:
4591         btrfs_free_path(path);
4592         return ret;
4593 }
4594
4595 /*
4596  * helper to find a free sequence number in a given directory.  This current
4597  * code is very simple, later versions will do smarter things in the btree
4598  */
4599 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4600 {
4601         int ret = 0;
4602
4603         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4604                 ret = btrfs_inode_delayed_dir_index_count(dir);
4605                 if (ret) {
4606                         ret = btrfs_set_inode_index_count(dir);
4607                         if (ret)
4608                                 return ret;
4609                 }
4610         }
4611
4612         *index = BTRFS_I(dir)->index_cnt;
4613         BTRFS_I(dir)->index_cnt++;
4614
4615         return ret;
4616 }
4617
4618 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4619                                      struct btrfs_root *root,
4620                                      struct inode *dir,
4621                                      const char *name, int name_len,
4622                                      u64 ref_objectid, u64 objectid,
4623                                      umode_t mode, u64 *index)
4624 {
4625         struct inode *inode;
4626         struct btrfs_inode_item *inode_item;
4627         struct btrfs_key *location;
4628         struct btrfs_path *path;
4629         struct btrfs_inode_ref *ref;
4630         struct btrfs_key key[2];
4631         u32 sizes[2];
4632         unsigned long ptr;
4633         int ret;
4634         int owner;
4635
4636         path = btrfs_alloc_path();
4637         if (!path)
4638                 return ERR_PTR(-ENOMEM);
4639
4640         inode = new_inode(root->fs_info->sb);
4641         if (!inode) {
4642                 btrfs_free_path(path);
4643                 return ERR_PTR(-ENOMEM);
4644         }
4645
4646         /*
4647          * we have to initialize this early, so we can reclaim the inode
4648          * number if we fail afterwards in this function.
4649          */
4650         inode->i_ino = objectid;
4651
4652         if (dir) {
4653                 trace_btrfs_inode_request(dir);
4654
4655                 ret = btrfs_set_inode_index(dir, index);
4656                 if (ret) {
4657                         btrfs_free_path(path);
4658                         iput(inode);
4659                         return ERR_PTR(ret);
4660                 }
4661         }
4662         /*
4663          * index_cnt is ignored for everything but a dir,
4664          * btrfs_get_inode_index_count has an explanation for the magic
4665          * number
4666          */
4667         BTRFS_I(inode)->index_cnt = 2;
4668         BTRFS_I(inode)->root = root;
4669         BTRFS_I(inode)->generation = trans->transid;
4670         inode->i_generation = BTRFS_I(inode)->generation;
4671
4672         if (S_ISDIR(mode))
4673                 owner = 0;
4674         else
4675                 owner = 1;
4676
4677         key[0].objectid = objectid;
4678         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4679         key[0].offset = 0;
4680
4681         key[1].objectid = objectid;
4682         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4683         key[1].offset = ref_objectid;
4684
4685         sizes[0] = sizeof(struct btrfs_inode_item);
4686         sizes[1] = name_len + sizeof(*ref);
4687
4688         path->leave_spinning = 1;
4689         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4690         if (ret != 0)
4691                 goto fail;
4692
4693         inode_init_owner(inode, dir, mode);
4694         inode_set_bytes(inode, 0);
4695         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4696         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4697                                   struct btrfs_inode_item);
4698         memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
4699                              sizeof(*inode_item));
4700         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4701
4702         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4703                              struct btrfs_inode_ref);
4704         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4705         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4706         ptr = (unsigned long)(ref + 1);
4707         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4708
4709         btrfs_mark_buffer_dirty(path->nodes[0]);
4710         btrfs_free_path(path);
4711
4712         location = &BTRFS_I(inode)->location;
4713         location->objectid = objectid;
4714         location->offset = 0;
4715         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4716
4717         btrfs_inherit_iflags(inode, dir);
4718
4719         if (S_ISREG(mode)) {
4720                 if (btrfs_test_opt(root, NODATASUM))
4721                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4722                 if (btrfs_test_opt(root, NODATACOW) ||
4723                     (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4724                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4725         }
4726
4727         insert_inode_hash(inode);
4728         inode_tree_add(inode);
4729
4730         trace_btrfs_inode_new(inode);
4731         btrfs_set_inode_last_trans(trans, inode);
4732
4733         btrfs_update_root_times(trans, root);
4734
4735         return inode;
4736 fail:
4737         if (dir)
4738                 BTRFS_I(dir)->index_cnt--;
4739         btrfs_free_path(path);
4740         iput(inode);
4741         return ERR_PTR(ret);
4742 }
4743
4744 static inline u8 btrfs_inode_type(struct inode *inode)
4745 {
4746         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4747 }
4748
4749 /*
4750  * utility function to add 'inode' into 'parent_inode' with
4751  * a give name and a given sequence number.
4752  * if 'add_backref' is true, also insert a backref from the
4753  * inode to the parent directory.
4754  */
4755 int btrfs_add_link(struct btrfs_trans_handle *trans,
4756                    struct inode *parent_inode, struct inode *inode,
4757                    const char *name, int name_len, int add_backref, u64 index)
4758 {
4759         int ret = 0;
4760         struct btrfs_key key;
4761         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4762         u64 ino = btrfs_ino(inode);
4763         u64 parent_ino = btrfs_ino(parent_inode);
4764
4765         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4766                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4767         } else {
4768                 key.objectid = ino;
4769                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4770                 key.offset = 0;
4771         }
4772
4773         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4774                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4775                                          key.objectid, root->root_key.objectid,
4776                                          parent_ino, index, name, name_len);
4777         } else if (add_backref) {
4778                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4779                                              parent_ino, index);
4780         }
4781
4782         /* Nothing to clean up yet */
4783         if (ret)
4784                 return ret;
4785
4786         ret = btrfs_insert_dir_item(trans, root, name, name_len,
4787                                     parent_inode, &key,
4788                                     btrfs_inode_type(inode), index);
4789         if (ret == -EEXIST)
4790                 goto fail_dir_item;
4791         else if (ret) {
4792                 btrfs_abort_transaction(trans, root, ret);
4793                 return ret;
4794         }
4795
4796         btrfs_i_size_write(parent_inode, parent_inode->i_size +
4797                            name_len * 2);
4798         inode_inc_iversion(parent_inode);
4799         parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4800         ret = btrfs_update_inode(trans, root, parent_inode);
4801         if (ret)
4802                 btrfs_abort_transaction(trans, root, ret);
4803         return ret;
4804
4805 fail_dir_item:
4806         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4807                 u64 local_index;
4808                 int err;
4809                 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4810                                  key.objectid, root->root_key.objectid,
4811                                  parent_ino, &local_index, name, name_len);
4812
4813         } else if (add_backref) {
4814                 u64 local_index;
4815                 int err;
4816
4817                 err = btrfs_del_inode_ref(trans, root, name, name_len,
4818                                           ino, parent_ino, &local_index);
4819         }
4820         return ret;
4821 }
4822
4823 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4824                             struct inode *dir, struct dentry *dentry,
4825                             struct inode *inode, int backref, u64 index)
4826 {
4827         int err = btrfs_add_link(trans, dir, inode,
4828                                  dentry->d_name.name, dentry->d_name.len,
4829                                  backref, index);
4830         if (err > 0)
4831                 err = -EEXIST;
4832         return err;
4833 }
4834
4835 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4836                         umode_t mode, dev_t rdev)
4837 {
4838         struct btrfs_trans_handle *trans;
4839         struct btrfs_root *root = BTRFS_I(dir)->root;
4840         struct inode *inode = NULL;
4841         int err;
4842         int drop_inode = 0;
4843         u64 objectid;
4844         unsigned long nr = 0;
4845         u64 index = 0;
4846
4847         if (!new_valid_dev(rdev))
4848                 return -EINVAL;
4849
4850         /*
4851          * 2 for inode item and ref
4852          * 2 for dir items
4853          * 1 for xattr if selinux is on
4854          */
4855         trans = btrfs_start_transaction(root, 5);
4856         if (IS_ERR(trans))
4857                 return PTR_ERR(trans);
4858
4859         err = btrfs_find_free_ino(root, &objectid);
4860         if (err)
4861                 goto out_unlock;
4862
4863         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4864                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4865                                 mode, &index);
4866         if (IS_ERR(inode)) {
4867                 err = PTR_ERR(inode);
4868                 goto out_unlock;
4869         }
4870
4871         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4872         if (err) {
4873                 drop_inode = 1;
4874                 goto out_unlock;
4875         }
4876
4877         /*
4878         * If the active LSM wants to access the inode during
4879         * d_instantiate it needs these. Smack checks to see
4880         * if the filesystem supports xattrs by looking at the
4881         * ops vector.
4882         */
4883
4884         inode->i_op = &btrfs_special_inode_operations;
4885         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4886         if (err)
4887                 drop_inode = 1;
4888         else {
4889                 init_special_inode(inode, inode->i_mode, rdev);
4890                 btrfs_update_inode(trans, root, inode);
4891                 d_instantiate(dentry, inode);
4892         }
4893 out_unlock:
4894         nr = trans->blocks_used;
4895         btrfs_end_transaction(trans, root);
4896         btrfs_btree_balance_dirty(root, nr);
4897         if (drop_inode) {
4898                 inode_dec_link_count(inode);
4899                 iput(inode);
4900         }
4901         return err;
4902 }
4903
4904 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4905                         umode_t mode, struct nameidata *nd)
4906 {
4907         struct btrfs_trans_handle *trans;
4908         struct btrfs_root *root = BTRFS_I(dir)->root;
4909         struct inode *inode = NULL;
4910         int drop_inode = 0;
4911         int err;
4912         unsigned long nr = 0;
4913         u64 objectid;
4914         u64 index = 0;
4915
4916         /*
4917          * 2 for inode item and ref
4918          * 2 for dir items
4919          * 1 for xattr if selinux is on
4920          */
4921         trans = btrfs_start_transaction(root, 5);
4922         if (IS_ERR(trans))
4923                 return PTR_ERR(trans);
4924
4925         err = btrfs_find_free_ino(root, &objectid);
4926         if (err)
4927                 goto out_unlock;
4928
4929         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4930                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4931                                 mode, &index);
4932         if (IS_ERR(inode)) {
4933                 err = PTR_ERR(inode);
4934                 goto out_unlock;
4935         }
4936
4937         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4938         if (err) {
4939                 drop_inode = 1;
4940                 goto out_unlock;
4941         }
4942
4943         /*
4944         * If the active LSM wants to access the inode during
4945         * d_instantiate it needs these. Smack checks to see
4946         * if the filesystem supports xattrs by looking at the
4947         * ops vector.
4948         */
4949         inode->i_fop = &btrfs_file_operations;
4950         inode->i_op = &btrfs_file_inode_operations;
4951
4952         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4953         if (err)
4954                 drop_inode = 1;
4955         else {
4956                 inode->i_mapping->a_ops = &btrfs_aops;
4957                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4958                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4959                 d_instantiate(dentry, inode);
4960         }
4961 out_unlock:
4962         nr = trans->blocks_used;
4963         btrfs_end_transaction(trans, root);
4964         if (drop_inode) {
4965                 inode_dec_link_count(inode);
4966                 iput(inode);
4967         }
4968         btrfs_btree_balance_dirty(root, nr);
4969         return err;
4970 }
4971
4972 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4973                       struct dentry *dentry)
4974 {
4975         struct btrfs_trans_handle *trans;
4976         struct btrfs_root *root = BTRFS_I(dir)->root;
4977         struct inode *inode = old_dentry->d_inode;
4978         u64 index;
4979         unsigned long nr = 0;
4980         int err;
4981         int drop_inode = 0;
4982
4983         /* do not allow sys_link's with other subvols of the same device */
4984         if (root->objectid != BTRFS_I(inode)->root->objectid)
4985                 return -EXDEV;
4986
4987         if (inode->i_nlink == ~0U)
4988                 return -EMLINK;
4989
4990         err = btrfs_set_inode_index(dir, &index);
4991         if (err)
4992                 goto fail;
4993
4994         /*
4995          * 2 items for inode and inode ref
4996          * 2 items for dir items
4997          * 1 item for parent inode
4998          */
4999         trans = btrfs_start_transaction(root, 5);
5000         if (IS_ERR(trans)) {
5001                 err = PTR_ERR(trans);
5002                 goto fail;
5003         }
5004
5005         btrfs_inc_nlink(inode);
5006         inode_inc_iversion(inode);
5007         inode->i_ctime = CURRENT_TIME;
5008         ihold(inode);
5009
5010         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5011
5012         if (err) {
5013                 drop_inode = 1;
5014         } else {
5015                 struct dentry *parent = dentry->d_parent;
5016                 err = btrfs_update_inode(trans, root, inode);
5017                 if (err)
5018                         goto fail;
5019                 d_instantiate(dentry, inode);
5020                 btrfs_log_new_name(trans, inode, NULL, parent);
5021         }
5022
5023         nr = trans->blocks_used;
5024         btrfs_end_transaction(trans, root);
5025 fail:
5026         if (drop_inode) {
5027                 inode_dec_link_count(inode);
5028                 iput(inode);
5029         }
5030         btrfs_btree_balance_dirty(root, nr);
5031         return err;
5032 }
5033
5034 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5035 {
5036         struct inode *inode = NULL;
5037         struct btrfs_trans_handle *trans;
5038         struct btrfs_root *root = BTRFS_I(dir)->root;
5039         int err = 0;
5040         int drop_on_err = 0;
5041         u64 objectid = 0;
5042         u64 index = 0;
5043         unsigned long nr = 1;
5044
5045         /*
5046          * 2 items for inode and ref
5047          * 2 items for dir items
5048          * 1 for xattr if selinux is on
5049          */
5050         trans = btrfs_start_transaction(root, 5);
5051         if (IS_ERR(trans))
5052                 return PTR_ERR(trans);
5053
5054         err = btrfs_find_free_ino(root, &objectid);
5055         if (err)
5056                 goto out_fail;
5057
5058         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5059                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5060                                 S_IFDIR | mode, &index);
5061         if (IS_ERR(inode)) {
5062                 err = PTR_ERR(inode);
5063                 goto out_fail;
5064         }
5065
5066         drop_on_err = 1;
5067
5068         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5069         if (err)
5070                 goto out_fail;
5071
5072         inode->i_op = &btrfs_dir_inode_operations;
5073         inode->i_fop = &btrfs_dir_file_operations;
5074
5075         btrfs_i_size_write(inode, 0);
5076         err = btrfs_update_inode(trans, root, inode);
5077         if (err)
5078                 goto out_fail;
5079
5080         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5081                              dentry->d_name.len, 0, index);
5082         if (err)
5083                 goto out_fail;
5084
5085         d_instantiate(dentry, inode);
5086         drop_on_err = 0;
5087
5088 out_fail:
5089         nr = trans->blocks_used;
5090         btrfs_end_transaction(trans, root);
5091         if (drop_on_err)
5092                 iput(inode);
5093         btrfs_btree_balance_dirty(root, nr);
5094         return err;
5095 }
5096
5097 /* helper for btfs_get_extent.  Given an existing extent in the tree,
5098  * and an extent that you want to insert, deal with overlap and insert
5099  * the new extent into the tree.
5100  */
5101 static int merge_extent_mapping(struct extent_map_tree *em_tree,
5102                                 struct extent_map *existing,
5103                                 struct extent_map *em,
5104                                 u64 map_start, u64 map_len)
5105 {
5106         u64 start_diff;
5107
5108         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5109         start_diff = map_start - em->start;
5110         em->start = map_start;
5111         em->len = map_len;
5112         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5113             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
5114                 em->block_start += start_diff;
5115                 em->block_len -= start_diff;
5116         }
5117         return add_extent_mapping(em_tree, em);
5118 }
5119
5120 static noinline int uncompress_inline(struct btrfs_path *path,
5121                                       struct inode *inode, struct page *page,
5122                                       size_t pg_offset, u64 extent_offset,
5123                                       struct btrfs_file_extent_item *item)
5124 {
5125         int ret;
5126         struct extent_buffer *leaf = path->nodes[0];
5127         char *tmp;
5128         size_t max_size;
5129         unsigned long inline_size;
5130         unsigned long ptr;
5131         int compress_type;
5132
5133         WARN_ON(pg_offset != 0);
5134         compress_type = btrfs_file_extent_compression(leaf, item);
5135         max_size = btrfs_file_extent_ram_bytes(leaf, item);
5136         inline_size = btrfs_file_extent_inline_item_len(leaf,
5137                                         btrfs_item_nr(leaf, path->slots[0]));
5138         tmp = kmalloc(inline_size, GFP_NOFS);
5139         if (!tmp)
5140                 return -ENOMEM;
5141         ptr = btrfs_file_extent_inline_start(item);
5142
5143         read_extent_buffer(leaf, tmp, ptr, inline_size);
5144
5145         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5146         ret = btrfs_decompress(compress_type, tmp, page,
5147                                extent_offset, inline_size, max_size);
5148         if (ret) {
5149                 char *kaddr = kmap_atomic(page);
5150                 unsigned long copy_size = min_t(u64,
5151                                   PAGE_CACHE_SIZE - pg_offset,
5152                                   max_size - extent_offset);
5153                 memset(kaddr + pg_offset, 0, copy_size);
5154                 kunmap_atomic(kaddr);
5155         }
5156         kfree(tmp);
5157         return 0;
5158 }
5159
5160 /*
5161  * a bit scary, this does extent mapping from logical file offset to the disk.
5162  * the ugly parts come from merging extents from the disk with the in-ram
5163  * representation.  This gets more complex because of the data=ordered code,
5164  * where the in-ram extents might be locked pending data=ordered completion.
5165  *
5166  * This also copies inline extents directly into the page.
5167  */
5168
5169 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5170                                     size_t pg_offset, u64 start, u64 len,
5171                                     int create)
5172 {
5173         int ret;
5174         int err = 0;
5175         u64 bytenr;
5176         u64 extent_start = 0;
5177         u64 extent_end = 0;
5178         u64 objectid = btrfs_ino(inode);
5179         u32 found_type;
5180         struct btrfs_path *path = NULL;
5181         struct btrfs_root *root = BTRFS_I(inode)->root;
5182         struct btrfs_file_extent_item *item;
5183         struct extent_buffer *leaf;
5184         struct btrfs_key found_key;
5185         struct extent_map *em = NULL;
5186         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5187         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5188         struct btrfs_trans_handle *trans = NULL;
5189         int compress_type;
5190
5191 again:
5192         read_lock(&em_tree->lock);
5193         em = lookup_extent_mapping(em_tree, start, len);
5194         if (em)
5195                 em->bdev = root->fs_info->fs_devices->latest_bdev;
5196         read_unlock(&em_tree->lock);
5197
5198         if (em) {
5199                 if (em->start > start || em->start + em->len <= start)
5200                         free_extent_map(em);
5201                 else if (em->block_start == EXTENT_MAP_INLINE && page)
5202                         free_extent_map(em);
5203                 else
5204                         goto out;
5205         }
5206         em = alloc_extent_map();
5207         if (!em) {
5208                 err = -ENOMEM;
5209                 goto out;
5210         }
5211         em->bdev = root->fs_info->fs_devices->latest_bdev;
5212         em->start = EXTENT_MAP_HOLE;
5213         em->orig_start = EXTENT_MAP_HOLE;
5214         em->len = (u64)-1;
5215         em->block_len = (u64)-1;
5216
5217         if (!path) {
5218                 path = btrfs_alloc_path();
5219                 if (!path) {
5220                         err = -ENOMEM;
5221                         goto out;
5222                 }
5223                 /*
5224                  * Chances are we'll be called again, so go ahead and do
5225                  * readahead
5226                  */
5227                 path->reada = 1;
5228         }
5229
5230         ret = btrfs_lookup_file_extent(trans, root, path,
5231                                        objectid, start, trans != NULL);
5232         if (ret < 0) {
5233                 err = ret;
5234                 goto out;
5235         }
5236
5237         if (ret != 0) {
5238                 if (path->slots[0] == 0)
5239                         goto not_found;
5240                 path->slots[0]--;
5241         }
5242
5243         leaf = path->nodes[0];
5244         item = btrfs_item_ptr(leaf, path->slots[0],
5245                               struct btrfs_file_extent_item);
5246         /* are we inside the extent that was found? */
5247         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5248         found_type = btrfs_key_type(&found_key);
5249         if (found_key.objectid != objectid ||
5250             found_type != BTRFS_EXTENT_DATA_KEY) {
5251                 goto not_found;
5252         }
5253
5254         found_type = btrfs_file_extent_type(leaf, item);
5255         extent_start = found_key.offset;
5256         compress_type = btrfs_file_extent_compression(leaf, item);
5257         if (found_type == BTRFS_FILE_EXTENT_REG ||
5258             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5259                 extent_end = extent_start +
5260                        btrfs_file_extent_num_bytes(leaf, item);
5261         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5262                 size_t size;
5263                 size = btrfs_file_extent_inline_len(leaf, item);
5264                 extent_end = (extent_start + size + root->sectorsize - 1) &
5265                         ~((u64)root->sectorsize - 1);
5266         }
5267
5268         if (start >= extent_end) {
5269                 path->slots[0]++;
5270                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5271                         ret = btrfs_next_leaf(root, path);
5272                         if (ret < 0) {
5273                                 err = ret;
5274                                 goto out;
5275                         }
5276                         if (ret > 0)
5277                                 goto not_found;
5278                         leaf = path->nodes[0];
5279                 }
5280                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5281                 if (found_key.objectid != objectid ||
5282                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5283                         goto not_found;
5284                 if (start + len <= found_key.offset)
5285                         goto not_found;
5286                 em->start = start;
5287                 em->len = found_key.offset - start;
5288                 goto not_found_em;
5289         }
5290
5291         if (found_type == BTRFS_FILE_EXTENT_REG ||
5292             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5293                 em->start = extent_start;
5294                 em->len = extent_end - extent_start;
5295                 em->orig_start = extent_start -
5296                                  btrfs_file_extent_offset(leaf, item);
5297                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5298                 if (bytenr == 0) {
5299                         em->block_start = EXTENT_MAP_HOLE;
5300                         goto insert;
5301                 }
5302                 if (compress_type != BTRFS_COMPRESS_NONE) {
5303                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5304                         em->compress_type = compress_type;
5305                         em->block_start = bytenr;
5306                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5307                                                                          item);
5308                 } else {
5309                         bytenr += btrfs_file_extent_offset(leaf, item);
5310                         em->block_start = bytenr;
5311                         em->block_len = em->len;
5312                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5313                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5314                 }
5315                 goto insert;
5316         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5317                 unsigned long ptr;
5318                 char *map;
5319                 size_t size;
5320                 size_t extent_offset;
5321                 size_t copy_size;
5322
5323                 em->block_start = EXTENT_MAP_INLINE;
5324                 if (!page || create) {
5325                         em->start = extent_start;
5326                         em->len = extent_end - extent_start;
5327                         goto out;
5328                 }
5329
5330                 size = btrfs_file_extent_inline_len(leaf, item);
5331                 extent_offset = page_offset(page) + pg_offset - extent_start;
5332                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5333                                 size - extent_offset);
5334                 em->start = extent_start + extent_offset;
5335                 em->len = (copy_size + root->sectorsize - 1) &
5336                         ~((u64)root->sectorsize - 1);
5337                 em->orig_start = EXTENT_MAP_INLINE;
5338                 if (compress_type) {
5339                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5340                         em->compress_type = compress_type;
5341                 }
5342                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5343                 if (create == 0 && !PageUptodate(page)) {
5344                         if (btrfs_file_extent_compression(leaf, item) !=
5345                             BTRFS_COMPRESS_NONE) {
5346                                 ret = uncompress_inline(path, inode, page,
5347                                                         pg_offset,
5348                                                         extent_offset, item);
5349                                 BUG_ON(ret); /* -ENOMEM */
5350                         } else {
5351                                 map = kmap(page);
5352                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5353                                                    copy_size);
5354                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5355                                         memset(map + pg_offset + copy_size, 0,
5356                                                PAGE_CACHE_SIZE - pg_offset -
5357                                                copy_size);
5358                                 }
5359                                 kunmap(page);
5360                         }
5361                         flush_dcache_page(page);
5362                 } else if (create && PageUptodate(page)) {
5363                         BUG();
5364                         if (!trans) {
5365                                 kunmap(page);
5366                                 free_extent_map(em);
5367                                 em = NULL;
5368
5369                                 btrfs_release_path(path);
5370                                 trans = btrfs_join_transaction(root);
5371
5372                                 if (IS_ERR(trans))
5373                                         return ERR_CAST(trans);
5374                                 goto again;
5375                         }
5376                         map = kmap(page);
5377                         write_extent_buffer(leaf, map + pg_offset, ptr,
5378                                             copy_size);
5379                         kunmap(page);
5380                         btrfs_mark_buffer_dirty(leaf);
5381                 }
5382                 set_extent_uptodate(io_tree, em->start,
5383                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
5384                 goto insert;
5385         } else {
5386                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5387                 WARN_ON(1);
5388         }
5389 not_found:
5390         em->start = start;
5391         em->len = len;
5392 not_found_em:
5393         em->block_start = EXTENT_MAP_HOLE;
5394         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5395 insert:
5396         btrfs_release_path(path);
5397         if (em->start > start || extent_map_end(em) <= start) {
5398                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5399                        "[%llu %llu]\n", (unsigned long long)em->start,
5400                        (unsigned long long)em->len,
5401                        (unsigned long long)start,
5402                        (unsigned long long)len);
5403                 err = -EIO;
5404                 goto out;
5405         }
5406
5407         err = 0;
5408         write_lock(&em_tree->lock);
5409         ret = add_extent_mapping(em_tree, em);
5410         /* it is possible that someone inserted the extent into the tree
5411          * while we had the lock dropped.  It is also possible that
5412          * an overlapping map exists in the tree
5413          */
5414         if (ret == -EEXIST) {
5415                 struct extent_map *existing;
5416
5417                 ret = 0;
5418
5419                 existing = lookup_extent_mapping(em_tree, start, len);
5420                 if (existing && (existing->start > start ||
5421                     existing->start + existing->len <= start)) {
5422                         free_extent_map(existing);
5423                         existing = NULL;
5424                 }
5425                 if (!existing) {
5426                         existing = lookup_extent_mapping(em_tree, em->start,
5427                                                          em->len);
5428                         if (existing) {
5429                                 err = merge_extent_mapping(em_tree, existing,
5430                                                            em, start,
5431                                                            root->sectorsize);
5432                                 free_extent_map(existing);
5433                                 if (err) {
5434                                         free_extent_map(em);
5435                                         em = NULL;
5436                                 }
5437                         } else {
5438                                 err = -EIO;
5439                                 free_extent_map(em);
5440                                 em = NULL;
5441                         }
5442                 } else {
5443                         free_extent_map(em);
5444                         em = existing;
5445                         err = 0;
5446                 }
5447         }
5448         write_unlock(&em_tree->lock);
5449 out:
5450
5451         trace_btrfs_get_extent(root, em);
5452
5453         if (path)
5454                 btrfs_free_path(path);
5455         if (trans) {
5456                 ret = btrfs_end_transaction(trans, root);
5457                 if (!err)
5458                         err = ret;
5459         }
5460         if (err) {
5461                 free_extent_map(em);
5462                 return ERR_PTR(err);
5463         }
5464         BUG_ON(!em); /* Error is always set */
5465         return em;
5466 }
5467
5468 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5469                                            size_t pg_offset, u64 start, u64 len,
5470                                            int create)
5471 {
5472         struct extent_map *em;
5473         struct extent_map *hole_em = NULL;
5474         u64 range_start = start;
5475         u64 end;
5476         u64 found;
5477         u64 found_end;
5478         int err = 0;
5479
5480         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5481         if (IS_ERR(em))
5482                 return em;
5483         if (em) {
5484                 /*
5485                  * if our em maps to a hole, there might
5486                  * actually be delalloc bytes behind it
5487                  */
5488                 if (em->block_start != EXTENT_MAP_HOLE)
5489                         return em;
5490                 else
5491                         hole_em = em;
5492         }
5493
5494         /* check to see if we've wrapped (len == -1 or similar) */
5495         end = start + len;
5496         if (end < start)
5497                 end = (u64)-1;
5498         else
5499                 end -= 1;
5500
5501         em = NULL;
5502
5503         /* ok, we didn't find anything, lets look for delalloc */
5504         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5505                                  end, len, EXTENT_DELALLOC, 1);
5506         found_end = range_start + found;
5507         if (found_end < range_start)
5508                 found_end = (u64)-1;
5509
5510         /*
5511          * we didn't find anything useful, return
5512          * the original results from get_extent()
5513          */
5514         if (range_start > end || found_end <= start) {
5515                 em = hole_em;
5516                 hole_em = NULL;
5517                 goto out;
5518         }
5519
5520         /* adjust the range_start to make sure it doesn't
5521          * go backwards from the start they passed in
5522          */
5523         range_start = max(start,range_start);
5524         found = found_end - range_start;
5525
5526         if (found > 0) {
5527                 u64 hole_start = start;
5528                 u64 hole_len = len;
5529
5530                 em = alloc_extent_map();
5531                 if (!em) {
5532                         err = -ENOMEM;
5533                         goto out;
5534                 }
5535                 /*
5536                  * when btrfs_get_extent can't find anything it
5537                  * returns one huge hole
5538                  *
5539                  * make sure what it found really fits our range, and
5540                  * adjust to make sure it is based on the start from
5541                  * the caller
5542                  */
5543                 if (hole_em) {
5544                         u64 calc_end = extent_map_end(hole_em);
5545
5546                         if (calc_end <= start || (hole_em->start > end)) {
5547                                 free_extent_map(hole_em);
5548                                 hole_em = NULL;
5549                         } else {
5550                                 hole_start = max(hole_em->start, start);
5551                                 hole_len = calc_end - hole_start;
5552                         }
5553                 }
5554                 em->bdev = NULL;
5555                 if (hole_em && range_start > hole_start) {
5556                         /* our hole starts before our delalloc, so we
5557                          * have to return just the parts of the hole
5558                          * that go until  the delalloc starts
5559                          */
5560                         em->len = min(hole_len,
5561                                       range_start - hole_start);
5562                         em->start = hole_start;
5563                         em->orig_start = hole_start;
5564                         /*
5565                          * don't adjust block start at all,
5566                          * it is fixed at EXTENT_MAP_HOLE
5567                          */
5568                         em->block_start = hole_em->block_start;
5569                         em->block_len = hole_len;
5570                 } else {
5571                         em->start = range_start;
5572                         em->len = found;
5573                         em->orig_start = range_start;
5574                         em->block_start = EXTENT_MAP_DELALLOC;
5575                         em->block_len = found;
5576                 }
5577         } else if (hole_em) {
5578                 return hole_em;
5579         }
5580 out:
5581
5582         free_extent_map(hole_em);
5583         if (err) {
5584                 free_extent_map(em);
5585                 return ERR_PTR(err);
5586         }
5587         return em;
5588 }
5589
5590 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5591                                                   struct extent_map *em,
5592                                                   u64 start, u64 len)
5593 {
5594         struct btrfs_root *root = BTRFS_I(inode)->root;
5595         struct btrfs_trans_handle *trans;
5596         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5597         struct btrfs_key ins;
5598         u64 alloc_hint;
5599         int ret;
5600         bool insert = false;
5601
5602         /*
5603          * Ok if the extent map we looked up is a hole and is for the exact
5604          * range we want, there is no reason to allocate a new one, however if
5605          * it is not right then we need to free this one and drop the cache for
5606          * our range.
5607          */
5608         if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5609             em->len != len) {
5610                 free_extent_map(em);
5611                 em = NULL;
5612                 insert = true;
5613                 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5614         }
5615
5616         trans = btrfs_join_transaction(root);
5617         if (IS_ERR(trans))
5618                 return ERR_CAST(trans);
5619
5620         if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5621                 btrfs_add_inode_defrag(trans, inode);
5622
5623         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5624
5625         alloc_hint = get_extent_allocation_hint(inode, start, len);
5626         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5627                                    alloc_hint, &ins, 1);
5628         if (ret) {
5629                 em = ERR_PTR(ret);
5630                 goto out;
5631         }
5632
5633         if (!em) {
5634                 em = alloc_extent_map();
5635                 if (!em) {
5636                         em = ERR_PTR(-ENOMEM);
5637                         goto out;
5638                 }
5639         }
5640
5641         em->start = start;
5642         em->orig_start = em->start;
5643         em->len = ins.offset;
5644
5645         em->block_start = ins.objectid;
5646         em->block_len = ins.offset;
5647         em->bdev = root->fs_info->fs_devices->latest_bdev;
5648
5649         /*
5650          * We need to do this because if we're using the original em we searched
5651          * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5652          */
5653         em->flags = 0;
5654         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5655
5656         while (insert) {
5657                 write_lock(&em_tree->lock);
5658                 ret = add_extent_mapping(em_tree, em);
5659                 write_unlock(&em_tree->lock);
5660                 if (ret != -EEXIST)
5661                         break;
5662                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5663         }
5664
5665         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5666                                            ins.offset, ins.offset, 0);
5667         if (ret) {
5668                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5669                 em = ERR_PTR(ret);
5670         }
5671 out:
5672         btrfs_end_transaction(trans, root);
5673         return em;
5674 }
5675
5676 /*
5677  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5678  * block must be cow'd
5679  */
5680 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5681                                       struct inode *inode, u64 offset, u64 len)
5682 {
5683         struct btrfs_path *path;
5684         int ret;
5685         struct extent_buffer *leaf;
5686         struct btrfs_root *root = BTRFS_I(inode)->root;
5687         struct btrfs_file_extent_item *fi;
5688         struct btrfs_key key;
5689         u64 disk_bytenr;
5690         u64 backref_offset;
5691         u64 extent_end;
5692         u64 num_bytes;
5693         int slot;
5694         int found_type;
5695
5696         path = btrfs_alloc_path();
5697         if (!path)
5698                 return -ENOMEM;
5699
5700         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5701                                        offset, 0);
5702         if (ret < 0)
5703                 goto out;
5704
5705         slot = path->slots[0];
5706         if (ret == 1) {
5707                 if (slot == 0) {
5708                         /* can't find the item, must cow */
5709                         ret = 0;
5710                         goto out;
5711                 }
5712                 slot--;
5713         }
5714         ret = 0;
5715         leaf = path->nodes[0];
5716         btrfs_item_key_to_cpu(leaf, &key, slot);
5717         if (key.objectid != btrfs_ino(inode) ||
5718             key.type != BTRFS_EXTENT_DATA_KEY) {
5719                 /* not our file or wrong item type, must cow */
5720                 goto out;
5721         }
5722
5723         if (key.offset > offset) {
5724                 /* Wrong offset, must cow */
5725                 goto out;
5726         }
5727
5728         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5729         found_type = btrfs_file_extent_type(leaf, fi);
5730         if (found_type != BTRFS_FILE_EXTENT_REG &&
5731             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5732                 /* not a regular extent, must cow */
5733                 goto out;
5734         }
5735         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5736         backref_offset = btrfs_file_extent_offset(leaf, fi);
5737
5738         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5739         if (extent_end < offset + len) {
5740                 /* extent doesn't include our full range, must cow */
5741                 goto out;
5742         }
5743
5744         if (btrfs_extent_readonly(root, disk_bytenr))
5745                 goto out;
5746
5747         /*
5748          * look for other files referencing this extent, if we
5749          * find any we must cow
5750          */
5751         if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5752                                   key.offset - backref_offset, disk_bytenr))
5753                 goto out;
5754
5755         /*
5756          * adjust disk_bytenr and num_bytes to cover just the bytes
5757          * in this extent we are about to write.  If there
5758          * are any csums in that range we have to cow in order
5759          * to keep the csums correct
5760          */
5761         disk_bytenr += backref_offset;
5762         disk_bytenr += offset - key.offset;
5763         num_bytes = min(offset + len, extent_end) - offset;
5764         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5765                                 goto out;
5766         /*
5767          * all of the above have passed, it is safe to overwrite this extent
5768          * without cow
5769          */
5770         ret = 1;
5771 out:
5772         btrfs_free_path(path);
5773         return ret;
5774 }
5775
5776 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
5777                               struct extent_state **cached_state, int writing)
5778 {
5779         struct btrfs_ordered_extent *ordered;
5780         int ret = 0;
5781
5782         while (1) {
5783                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
5784                                  0, cached_state);
5785                 /*
5786                  * We're concerned with the entire range that we're going to be
5787                  * doing DIO to, so we need to make sure theres no ordered
5788                  * extents in this range.
5789                  */
5790                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
5791                                                      lockend - lockstart + 1);
5792
5793                 /*
5794                  * We need to make sure there are no buffered pages in this
5795                  * range either, we could have raced between the invalidate in
5796                  * generic_file_direct_write and locking the extent.  The
5797                  * invalidate needs to happen so that reads after a write do not
5798                  * get stale data.
5799                  */
5800                 if (!ordered && (!writing ||
5801                     !test_range_bit(&BTRFS_I(inode)->io_tree,
5802                                     lockstart, lockend, EXTENT_UPTODATE, 0,
5803                                     *cached_state)))
5804                         break;
5805
5806                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
5807                                      cached_state, GFP_NOFS);
5808
5809                 if (ordered) {
5810                         btrfs_start_ordered_extent(inode, ordered, 1);
5811                         btrfs_put_ordered_extent(ordered);
5812                 } else {
5813                         /* Screw you mmap */
5814                         ret = filemap_write_and_wait_range(inode->i_mapping,
5815                                                            lockstart,
5816                                                            lockend);
5817                         if (ret)
5818                                 break;
5819
5820                         /*
5821                          * If we found a page that couldn't be invalidated just
5822                          * fall back to buffered.
5823                          */
5824                         ret = invalidate_inode_pages2_range(inode->i_mapping,
5825                                         lockstart >> PAGE_CACHE_SHIFT,
5826                                         lockend >> PAGE_CACHE_SHIFT);
5827                         if (ret)
5828                                 break;
5829                 }
5830
5831                 cond_resched();
5832         }
5833
5834         return ret;
5835 }
5836
5837 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5838                                    struct buffer_head *bh_result, int create)
5839 {
5840         struct extent_map *em;
5841         struct btrfs_root *root = BTRFS_I(inode)->root;
5842         struct extent_state *cached_state = NULL;
5843         u64 start = iblock << inode->i_blkbits;
5844         u64 lockstart, lockend;
5845         u64 len = bh_result->b_size;
5846         struct btrfs_trans_handle *trans;
5847         int unlock_bits = EXTENT_LOCKED;
5848         int ret;
5849
5850         lockstart = start;
5851         lockend = start + len - 1;
5852         if (create) {
5853                 ret = btrfs_delalloc_reserve_space(inode, len);
5854                 if (ret)
5855                         return ret;
5856                 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
5857         }
5858
5859         /*
5860          * If this errors out it's because we couldn't invalidate pagecache for
5861          * this range and we need to fallback to buffered.
5862          */
5863         if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
5864                 return -ENOTBLK;
5865
5866         if (create) {
5867                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
5868                                      lockend, EXTENT_DELALLOC, NULL,
5869                                      &cached_state, GFP_NOFS);
5870                 if (ret)
5871                         goto unlock_err;
5872         }
5873
5874         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5875         if (IS_ERR(em)) {
5876                 ret = PTR_ERR(em);
5877                 goto unlock_err;
5878         }
5879
5880         /*
5881          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5882          * io.  INLINE is special, and we could probably kludge it in here, but
5883          * it's still buffered so for safety lets just fall back to the generic
5884          * buffered path.
5885          *
5886          * For COMPRESSED we _have_ to read the entire extent in so we can
5887          * decompress it, so there will be buffering required no matter what we
5888          * do, so go ahead and fallback to buffered.
5889          *
5890          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5891          * to buffered IO.  Don't blame me, this is the price we pay for using
5892          * the generic code.
5893          */
5894         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5895             em->block_start == EXTENT_MAP_INLINE) {
5896                 free_extent_map(em);
5897                 ret = -ENOTBLK;
5898                 goto unlock_err;
5899         }
5900
5901         /* Just a good old fashioned hole, return */
5902         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5903                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5904                 free_extent_map(em);
5905                 ret = 0;
5906                 goto unlock_err;
5907         }
5908
5909         /*
5910          * We don't allocate a new extent in the following cases
5911          *
5912          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
5913          * existing extent.
5914          * 2) The extent is marked as PREALLOC.  We're good to go here and can
5915          * just use the extent.
5916          *
5917          */
5918         if (!create) {
5919                 len = min(len, em->len - (start - em->start));
5920                 lockstart = start + len;
5921                 goto unlock;
5922         }
5923
5924         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5925             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5926              em->block_start != EXTENT_MAP_HOLE)) {
5927                 int type;
5928                 int ret;
5929                 u64 block_start;
5930
5931                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5932                         type = BTRFS_ORDERED_PREALLOC;
5933                 else
5934                         type = BTRFS_ORDERED_NOCOW;
5935                 len = min(len, em->len - (start - em->start));
5936                 block_start = em->block_start + (start - em->start);
5937
5938                 /*
5939                  * we're not going to log anything, but we do need
5940                  * to make sure the current transaction stays open
5941                  * while we look for nocow cross refs
5942                  */
5943                 trans = btrfs_join_transaction(root);
5944                 if (IS_ERR(trans))
5945                         goto must_cow;
5946
5947                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5948                         ret = btrfs_add_ordered_extent_dio(inode, start,
5949                                            block_start, len, len, type);
5950                         btrfs_end_transaction(trans, root);
5951                         if (ret) {
5952                                 free_extent_map(em);
5953                                 goto unlock_err;
5954                         }
5955                         goto unlock;
5956                 }
5957                 btrfs_end_transaction(trans, root);
5958         }
5959 must_cow:
5960         /*
5961          * this will cow the extent, reset the len in case we changed
5962          * it above
5963          */
5964         len = bh_result->b_size;
5965         em = btrfs_new_extent_direct(inode, em, start, len);
5966         if (IS_ERR(em)) {
5967                 ret = PTR_ERR(em);
5968                 goto unlock_err;
5969         }
5970         len = min(len, em->len - (start - em->start));
5971 unlock:
5972         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5973                 inode->i_blkbits;
5974         bh_result->b_size = len;
5975         bh_result->b_bdev = em->bdev;
5976         set_buffer_mapped(bh_result);
5977         if (create) {
5978                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5979                         set_buffer_new(bh_result);
5980
5981                 /*
5982                  * Need to update the i_size under the extent lock so buffered
5983                  * readers will get the updated i_size when we unlock.
5984                  */
5985                 if (start + len > i_size_read(inode))
5986                         i_size_write(inode, start + len);
5987         }
5988
5989         /*
5990          * In the case of write we need to clear and unlock the entire range,
5991          * in the case of read we need to unlock only the end area that we
5992          * aren't using if there is any left over space.
5993          */
5994         if (lockstart < lockend)
5995                 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
5996                                  unlock_bits, 1, 0, &cached_state, GFP_NOFS);
5997         else
5998                 free_extent_state(cached_state);
5999
6000         free_extent_map(em);
6001
6002         return 0;
6003
6004 unlock_err:
6005         if (create)
6006                 unlock_bits |= EXTENT_DO_ACCOUNTING;
6007
6008         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6009                          unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6010         return ret;
6011 }
6012
6013 struct btrfs_dio_private {
6014         struct inode *inode;
6015         u64 logical_offset;
6016         u64 disk_bytenr;
6017         u64 bytes;
6018         u32 *csums;
6019         void *private;
6020
6021         /* number of bios pending for this dio */
6022         atomic_t pending_bios;
6023
6024         /* IO errors */
6025         int errors;
6026
6027         struct bio *orig_bio;
6028 };
6029
6030 static void btrfs_endio_direct_read(struct bio *bio, int err)
6031 {
6032         struct btrfs_dio_private *dip = bio->bi_private;
6033         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6034         struct bio_vec *bvec = bio->bi_io_vec;
6035         struct inode *inode = dip->inode;
6036         struct btrfs_root *root = BTRFS_I(inode)->root;
6037         u64 start;
6038         u32 *private = dip->csums;
6039
6040         start = dip->logical_offset;
6041         do {
6042                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6043                         struct page *page = bvec->bv_page;
6044                         char *kaddr;
6045                         u32 csum = ~(u32)0;
6046                         unsigned long flags;
6047
6048                         local_irq_save(flags);
6049                         kaddr = kmap_atomic(page);
6050                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
6051                                                csum, bvec->bv_len);
6052                         btrfs_csum_final(csum, (char *)&csum);
6053                         kunmap_atomic(kaddr);
6054                         local_irq_restore(flags);
6055
6056                         flush_dcache_page(bvec->bv_page);
6057                         if (csum != *private) {
6058                                 printk(KERN_ERR "btrfs csum failed ino %llu off"
6059                                       " %llu csum %u private %u\n",
6060                                       (unsigned long long)btrfs_ino(inode),
6061                                       (unsigned long long)start,
6062                                       csum, *private);
6063                                 err = -EIO;
6064                         }
6065                 }
6066
6067                 start += bvec->bv_len;
6068                 private++;
6069                 bvec++;
6070         } while (bvec <= bvec_end);
6071
6072         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
6073                       dip->logical_offset + dip->bytes - 1);
6074         bio->bi_private = dip->private;
6075
6076         kfree(dip->csums);
6077         kfree(dip);
6078
6079         /* If we had a csum failure make sure to clear the uptodate flag */
6080         if (err)
6081                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
6082         dio_end_io(bio, err);
6083 }
6084
6085 static void btrfs_endio_direct_write(struct bio *bio, int err)
6086 {
6087         struct btrfs_dio_private *dip = bio->bi_private;
6088         struct inode *inode = dip->inode;
6089         struct btrfs_root *root = BTRFS_I(inode)->root;
6090         struct btrfs_ordered_extent *ordered = NULL;
6091         u64 ordered_offset = dip->logical_offset;
6092         u64 ordered_bytes = dip->bytes;
6093         int ret;
6094
6095         if (err)
6096                 goto out_done;
6097 again:
6098         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
6099                                                    &ordered_offset,
6100                                                    ordered_bytes, !err);
6101         if (!ret)
6102                 goto out_test;
6103
6104         ordered->work.func = finish_ordered_fn;
6105         ordered->work.flags = 0;
6106         btrfs_queue_worker(&root->fs_info->endio_write_workers,
6107                            &ordered->work);
6108 out_test:
6109         /*
6110          * our bio might span multiple ordered extents.  If we haven't
6111          * completed the accounting for the whole dio, go back and try again
6112          */
6113         if (ordered_offset < dip->logical_offset + dip->bytes) {
6114                 ordered_bytes = dip->logical_offset + dip->bytes -
6115                         ordered_offset;
6116                 ordered = NULL;
6117                 goto again;
6118         }
6119 out_done:
6120         bio->bi_private = dip->private;
6121
6122         kfree(dip);
6123
6124         /* If we had an error make sure to clear the uptodate flag */
6125         if (err)
6126                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
6127         dio_end_io(bio, err);
6128 }
6129
6130 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
6131                                     struct bio *bio, int mirror_num,
6132                                     unsigned long bio_flags, u64 offset)
6133 {
6134         int ret;
6135         struct btrfs_root *root = BTRFS_I(inode)->root;
6136         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
6137         BUG_ON(ret); /* -ENOMEM */
6138         return 0;
6139 }
6140
6141 static void btrfs_end_dio_bio(struct bio *bio, int err)
6142 {
6143         struct btrfs_dio_private *dip = bio->bi_private;
6144
6145         if (err) {
6146                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
6147                       "sector %#Lx len %u err no %d\n",
6148                       (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
6149                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
6150                 dip->errors = 1;
6151
6152                 /*
6153                  * before atomic variable goto zero, we must make sure
6154                  * dip->errors is perceived to be set.
6155                  */
6156                 smp_mb__before_atomic_dec();
6157         }
6158
6159         /* if there are more bios still pending for this dio, just exit */
6160         if (!atomic_dec_and_test(&dip->pending_bios))
6161                 goto out;
6162
6163         if (dip->errors)
6164                 bio_io_error(dip->orig_bio);
6165         else {
6166                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
6167                 bio_endio(dip->orig_bio, 0);
6168         }
6169 out:
6170         bio_put(bio);
6171 }
6172
6173 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
6174                                        u64 first_sector, gfp_t gfp_flags)
6175 {
6176         int nr_vecs = bio_get_nr_vecs(bdev);
6177         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
6178 }
6179
6180 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
6181                                          int rw, u64 file_offset, int skip_sum,
6182                                          u32 *csums, int async_submit)
6183 {
6184         int write = rw & REQ_WRITE;
6185         struct btrfs_root *root = BTRFS_I(inode)->root;
6186         int ret;
6187
6188         bio_get(bio);
6189
6190         if (!write) {
6191                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
6192                 if (ret)
6193                         goto err;
6194         }
6195
6196         if (skip_sum)
6197                 goto map;
6198
6199         if (write && async_submit) {
6200                 ret = btrfs_wq_submit_bio(root->fs_info,
6201                                    inode, rw, bio, 0, 0,
6202                                    file_offset,
6203                                    __btrfs_submit_bio_start_direct_io,
6204                                    __btrfs_submit_bio_done);
6205                 goto err;
6206         } else if (write) {
6207                 /*
6208                  * If we aren't doing async submit, calculate the csum of the
6209                  * bio now.
6210                  */
6211                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
6212                 if (ret)
6213                         goto err;
6214         } else if (!skip_sum) {
6215                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
6216                                           file_offset, csums);
6217                 if (ret)
6218                         goto err;
6219         }
6220
6221 map:
6222         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
6223 err:
6224         bio_put(bio);
6225         return ret;
6226 }
6227
6228 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
6229                                     int skip_sum)
6230 {
6231         struct inode *inode = dip->inode;
6232         struct btrfs_root *root = BTRFS_I(inode)->root;
6233         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6234         struct bio *bio;
6235         struct bio *orig_bio = dip->orig_bio;
6236         struct bio_vec *bvec = orig_bio->bi_io_vec;
6237         u64 start_sector = orig_bio->bi_sector;
6238         u64 file_offset = dip->logical_offset;
6239         u64 submit_len = 0;
6240         u64 map_length;
6241         int nr_pages = 0;
6242         u32 *csums = dip->csums;
6243         int ret = 0;
6244         int async_submit = 0;
6245         int write = rw & REQ_WRITE;
6246
6247         map_length = orig_bio->bi_size;
6248         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6249                               &map_length, NULL, 0);
6250         if (ret) {
6251                 bio_put(orig_bio);
6252                 return -EIO;
6253         }
6254
6255         if (map_length >= orig_bio->bi_size) {
6256                 bio = orig_bio;
6257                 goto submit;
6258         }
6259
6260         async_submit = 1;
6261         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
6262         if (!bio)
6263                 return -ENOMEM;
6264         bio->bi_private = dip;
6265         bio->bi_end_io = btrfs_end_dio_bio;
6266         atomic_inc(&dip->pending_bios);
6267
6268         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6269                 if (unlikely(map_length < submit_len + bvec->bv_len ||
6270                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6271                                  bvec->bv_offset) < bvec->bv_len)) {
6272                         /*
6273                          * inc the count before we submit the bio so
6274                          * we know the end IO handler won't happen before
6275                          * we inc the count. Otherwise, the dip might get freed
6276                          * before we're done setting it up
6277                          */
6278                         atomic_inc(&dip->pending_bios);
6279                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
6280                                                      file_offset, skip_sum,
6281                                                      csums, async_submit);
6282                         if (ret) {
6283                                 bio_put(bio);
6284                                 atomic_dec(&dip->pending_bios);
6285                                 goto out_err;
6286                         }
6287
6288                         /* Write's use the ordered csums */
6289                         if (!write && !skip_sum)
6290                                 csums = csums + nr_pages;
6291                         start_sector += submit_len >> 9;
6292                         file_offset += submit_len;
6293
6294                         submit_len = 0;
6295                         nr_pages = 0;
6296
6297                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6298                                                   start_sector, GFP_NOFS);
6299                         if (!bio)
6300                                 goto out_err;
6301                         bio->bi_private = dip;
6302                         bio->bi_end_io = btrfs_end_dio_bio;
6303
6304                         map_length = orig_bio->bi_size;
6305                         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6306                                               &map_length, NULL, 0);
6307                         if (ret) {
6308                                 bio_put(bio);
6309                                 goto out_err;
6310                         }
6311                 } else {
6312                         submit_len += bvec->bv_len;
6313                         nr_pages ++;
6314                         bvec++;
6315                 }
6316         }
6317
6318 submit:
6319         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6320                                      csums, async_submit);
6321         if (!ret)
6322                 return 0;
6323
6324         bio_put(bio);
6325 out_err:
6326         dip->errors = 1;
6327         /*
6328          * before atomic variable goto zero, we must
6329          * make sure dip->errors is perceived to be set.
6330          */
6331         smp_mb__before_atomic_dec();
6332         if (atomic_dec_and_test(&dip->pending_bios))
6333                 bio_io_error(dip->orig_bio);
6334
6335         /* bio_end_io() will handle error, so we needn't return it */
6336         return 0;
6337 }
6338
6339 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6340                                 loff_t file_offset)
6341 {
6342         struct btrfs_root *root = BTRFS_I(inode)->root;
6343         struct btrfs_dio_private *dip;
6344         struct bio_vec *bvec = bio->bi_io_vec;
6345         int skip_sum;
6346         int write = rw & REQ_WRITE;
6347         int ret = 0;
6348
6349         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6350
6351         dip = kmalloc(sizeof(*dip), GFP_NOFS);
6352         if (!dip) {
6353                 ret = -ENOMEM;
6354                 goto free_ordered;
6355         }
6356         dip->csums = NULL;
6357
6358         /* Write's use the ordered csum stuff, so we don't need dip->csums */
6359         if (!write && !skip_sum) {
6360                 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6361                 if (!dip->csums) {
6362                         kfree(dip);
6363                         ret = -ENOMEM;
6364                         goto free_ordered;
6365                 }
6366         }
6367
6368         dip->private = bio->bi_private;
6369         dip->inode = inode;
6370         dip->logical_offset = file_offset;
6371
6372         dip->bytes = 0;
6373         do {
6374                 dip->bytes += bvec->bv_len;
6375                 bvec++;
6376         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6377
6378         dip->disk_bytenr = (u64)bio->bi_sector << 9;
6379         bio->bi_private = dip;
6380         dip->errors = 0;
6381         dip->orig_bio = bio;
6382         atomic_set(&dip->pending_bios, 0);
6383
6384         if (write)
6385                 bio->bi_end_io = btrfs_endio_direct_write;
6386         else
6387                 bio->bi_end_io = btrfs_endio_direct_read;
6388
6389         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6390         if (!ret)
6391                 return;
6392 free_ordered:
6393         /*
6394          * If this is a write, we need to clean up the reserved space and kill
6395          * the ordered extent.
6396          */
6397         if (write) {
6398                 struct btrfs_ordered_extent *ordered;
6399                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6400                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6401                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6402                         btrfs_free_reserved_extent(root, ordered->start,
6403                                                    ordered->disk_len);
6404                 btrfs_put_ordered_extent(ordered);
6405                 btrfs_put_ordered_extent(ordered);
6406         }
6407         bio_endio(bio, ret);
6408 }
6409
6410 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6411                         const struct iovec *iov, loff_t offset,
6412                         unsigned long nr_segs)
6413 {
6414         int seg;
6415         int i;
6416         size_t size;
6417         unsigned long addr;
6418         unsigned blocksize_mask = root->sectorsize - 1;
6419         ssize_t retval = -EINVAL;
6420         loff_t end = offset;
6421
6422         if (offset & blocksize_mask)
6423                 goto out;
6424
6425         /* Check the memory alignment.  Blocks cannot straddle pages */
6426         for (seg = 0; seg < nr_segs; seg++) {
6427                 addr = (unsigned long)iov[seg].iov_base;
6428                 size = iov[seg].iov_len;
6429                 end += size;
6430                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6431                         goto out;
6432
6433                 /* If this is a write we don't need to check anymore */
6434                 if (rw & WRITE)
6435                         continue;
6436
6437                 /*
6438                  * Check to make sure we don't have duplicate iov_base's in this
6439                  * iovec, if so return EINVAL, otherwise we'll get csum errors
6440                  * when reading back.
6441                  */
6442                 for (i = seg + 1; i < nr_segs; i++) {
6443                         if (iov[seg].iov_base == iov[i].iov_base)
6444                                 goto out;
6445                 }
6446         }
6447         retval = 0;
6448 out:
6449         return retval;
6450 }
6451
6452 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6453                         const struct iovec *iov, loff_t offset,
6454                         unsigned long nr_segs)
6455 {
6456         struct file *file = iocb->ki_filp;
6457         struct inode *inode = file->f_mapping->host;
6458
6459         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6460                             offset, nr_segs))
6461                 return 0;
6462
6463         return __blockdev_direct_IO(rw, iocb, inode,
6464                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6465                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6466                    btrfs_submit_direct, 0);
6467 }
6468
6469 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6470                 __u64 start, __u64 len)
6471 {
6472         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6473 }
6474
6475 int btrfs_readpage(struct file *file, struct page *page)
6476 {
6477         struct extent_io_tree *tree;
6478         tree = &BTRFS_I(page->mapping->host)->io_tree;
6479         return extent_read_full_page(tree, page, btrfs_get_extent, 0);
6480 }
6481
6482 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6483 {
6484         struct extent_io_tree *tree;
6485
6486
6487         if (current->flags & PF_MEMALLOC) {
6488                 redirty_page_for_writepage(wbc, page);
6489                 unlock_page(page);
6490                 return 0;
6491         }
6492         tree = &BTRFS_I(page->mapping->host)->io_tree;
6493         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6494 }
6495
6496 int btrfs_writepages(struct address_space *mapping,
6497                      struct writeback_control *wbc)
6498 {
6499         struct extent_io_tree *tree;
6500
6501         tree = &BTRFS_I(mapping->host)->io_tree;
6502         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6503 }
6504
6505 static int
6506 btrfs_readpages(struct file *file, struct address_space *mapping,
6507                 struct list_head *pages, unsigned nr_pages)
6508 {
6509         struct extent_io_tree *tree;
6510         tree = &BTRFS_I(mapping->host)->io_tree;
6511         return extent_readpages(tree, mapping, pages, nr_pages,
6512                                 btrfs_get_extent);
6513 }
6514 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6515 {
6516         struct extent_io_tree *tree;
6517         struct extent_map_tree *map;
6518         int ret;
6519
6520         tree = &BTRFS_I(page->mapping->host)->io_tree;
6521         map = &BTRFS_I(page->mapping->host)->extent_tree;
6522         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6523         if (ret == 1) {
6524                 ClearPagePrivate(page);
6525                 set_page_private(page, 0);
6526                 page_cache_release(page);
6527         }
6528         return ret;
6529 }
6530
6531 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6532 {
6533         if (PageWriteback(page) || PageDirty(page))
6534                 return 0;
6535         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6536 }
6537
6538 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6539 {
6540         struct inode *inode = page->mapping->host;
6541         struct extent_io_tree *tree;
6542         struct btrfs_ordered_extent *ordered;
6543         struct extent_state *cached_state = NULL;
6544         u64 page_start = page_offset(page);
6545         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6546
6547         /*
6548          * we have the page locked, so new writeback can't start,
6549          * and the dirty bit won't be cleared while we are here.
6550          *
6551          * Wait for IO on this page so that we can safely clear
6552          * the PagePrivate2 bit and do ordered accounting
6553          */
6554         wait_on_page_writeback(page);
6555
6556         tree = &BTRFS_I(inode)->io_tree;
6557         if (offset) {
6558                 btrfs_releasepage(page, GFP_NOFS);
6559                 return;
6560         }
6561         lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6562         ordered = btrfs_lookup_ordered_extent(inode,
6563                                            page_offset(page));
6564         if (ordered) {
6565                 /*
6566                  * IO on this page will never be started, so we need
6567                  * to account for any ordered extents now
6568                  */
6569                 clear_extent_bit(tree, page_start, page_end,
6570                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6571                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6572                                  &cached_state, GFP_NOFS);
6573                 /*
6574                  * whoever cleared the private bit is responsible
6575                  * for the finish_ordered_io
6576                  */
6577                 if (TestClearPagePrivate2(page) &&
6578                     btrfs_dec_test_ordered_pending(inode, &ordered, page_start,
6579                                                    PAGE_CACHE_SIZE, 1)) {
6580                         btrfs_finish_ordered_io(ordered);
6581                 }
6582                 btrfs_put_ordered_extent(ordered);
6583                 cached_state = NULL;
6584                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6585         }
6586         clear_extent_bit(tree, page_start, page_end,
6587                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6588                  EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6589         __btrfs_releasepage(page, GFP_NOFS);
6590
6591         ClearPageChecked(page);
6592         if (PagePrivate(page)) {
6593                 ClearPagePrivate(page);
6594                 set_page_private(page, 0);
6595                 page_cache_release(page);
6596         }
6597 }
6598
6599 /*
6600  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6601  * called from a page fault handler when a page is first dirtied. Hence we must
6602  * be careful to check for EOF conditions here. We set the page up correctly
6603  * for a written page which means we get ENOSPC checking when writing into
6604  * holes and correct delalloc and unwritten extent mapping on filesystems that
6605  * support these features.
6606  *
6607  * We are not allowed to take the i_mutex here so we have to play games to
6608  * protect against truncate races as the page could now be beyond EOF.  Because
6609  * vmtruncate() writes the inode size before removing pages, once we have the
6610  * page lock we can determine safely if the page is beyond EOF. If it is not
6611  * beyond EOF, then the page is guaranteed safe against truncation until we
6612  * unlock the page.
6613  */
6614 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6615 {
6616         struct page *page = vmf->page;
6617         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6618         struct btrfs_root *root = BTRFS_I(inode)->root;
6619         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6620         struct btrfs_ordered_extent *ordered;
6621         struct extent_state *cached_state = NULL;
6622         char *kaddr;
6623         unsigned long zero_start;
6624         loff_t size;
6625         int ret;
6626         int reserved = 0;
6627         u64 page_start;
6628         u64 page_end;
6629
6630         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6631         if (!ret) {
6632                 ret = file_update_time(vma->vm_file);
6633                 reserved = 1;
6634         }
6635         if (ret) {
6636                 if (ret == -ENOMEM)
6637                         ret = VM_FAULT_OOM;
6638                 else /* -ENOSPC, -EIO, etc */
6639                         ret = VM_FAULT_SIGBUS;
6640                 if (reserved)
6641                         goto out;
6642                 goto out_noreserve;
6643         }
6644
6645         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6646 again:
6647         lock_page(page);
6648         size = i_size_read(inode);
6649         page_start = page_offset(page);
6650         page_end = page_start + PAGE_CACHE_SIZE - 1;
6651
6652         if ((page->mapping != inode->i_mapping) ||
6653             (page_start >= size)) {
6654                 /* page got truncated out from underneath us */
6655                 goto out_unlock;
6656         }
6657         wait_on_page_writeback(page);
6658
6659         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
6660         set_page_extent_mapped(page);
6661
6662         /*
6663          * we can't set the delalloc bits if there are pending ordered
6664          * extents.  Drop our locks and wait for them to finish
6665          */
6666         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6667         if (ordered) {
6668                 unlock_extent_cached(io_tree, page_start, page_end,
6669                                      &cached_state, GFP_NOFS);
6670                 unlock_page(page);
6671                 btrfs_start_ordered_extent(inode, ordered, 1);
6672                 btrfs_put_ordered_extent(ordered);
6673                 goto again;
6674         }
6675
6676         /*
6677          * XXX - page_mkwrite gets called every time the page is dirtied, even
6678          * if it was already dirty, so for space accounting reasons we need to
6679          * clear any delalloc bits for the range we are fixing to save.  There
6680          * is probably a better way to do this, but for now keep consistent with
6681          * prepare_pages in the normal write path.
6682          */
6683         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6684                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6685                           0, 0, &cached_state, GFP_NOFS);
6686
6687         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6688                                         &cached_state);
6689         if (ret) {
6690                 unlock_extent_cached(io_tree, page_start, page_end,
6691                                      &cached_state, GFP_NOFS);
6692                 ret = VM_FAULT_SIGBUS;
6693                 goto out_unlock;
6694         }
6695         ret = 0;
6696
6697         /* page is wholly or partially inside EOF */
6698         if (page_start + PAGE_CACHE_SIZE > size)
6699                 zero_start = size & ~PAGE_CACHE_MASK;
6700         else
6701                 zero_start = PAGE_CACHE_SIZE;
6702
6703         if (zero_start != PAGE_CACHE_SIZE) {
6704                 kaddr = kmap(page);
6705                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6706                 flush_dcache_page(page);
6707                 kunmap(page);
6708         }
6709         ClearPageChecked(page);
6710         set_page_dirty(page);
6711         SetPageUptodate(page);
6712
6713         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6714         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6715
6716         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6717
6718 out_unlock:
6719         if (!ret)
6720                 return VM_FAULT_LOCKED;
6721         unlock_page(page);
6722 out:
6723         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6724 out_noreserve:
6725         return ret;
6726 }
6727
6728 static int btrfs_truncate(struct inode *inode)
6729 {
6730         struct btrfs_root *root = BTRFS_I(inode)->root;
6731         struct btrfs_block_rsv *rsv;
6732         int ret;
6733         int err = 0;
6734         struct btrfs_trans_handle *trans;
6735         unsigned long nr;
6736         u64 mask = root->sectorsize - 1;
6737         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
6738
6739         ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6740         if (ret)
6741                 return ret;
6742
6743         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6744         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6745
6746         /*
6747          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
6748          * 3 things going on here
6749          *
6750          * 1) We need to reserve space for our orphan item and the space to
6751          * delete our orphan item.  Lord knows we don't want to have a dangling
6752          * orphan item because we didn't reserve space to remove it.
6753          *
6754          * 2) We need to reserve space to update our inode.
6755          *
6756          * 3) We need to have something to cache all the space that is going to
6757          * be free'd up by the truncate operation, but also have some slack
6758          * space reserved in case it uses space during the truncate (thank you
6759          * very much snapshotting).
6760          *
6761          * And we need these to all be seperate.  The fact is we can use alot of
6762          * space doing the truncate, and we have no earthly idea how much space
6763          * we will use, so we need the truncate reservation to be seperate so it
6764          * doesn't end up using space reserved for updating the inode or
6765          * removing the orphan item.  We also need to be able to stop the
6766          * transaction and start a new one, which means we need to be able to
6767          * update the inode several times, and we have no idea of knowing how
6768          * many times that will be, so we can't just reserve 1 item for the
6769          * entirety of the opration, so that has to be done seperately as well.
6770          * Then there is the orphan item, which does indeed need to be held on
6771          * to for the whole operation, and we need nobody to touch this reserved
6772          * space except the orphan code.
6773          *
6774          * So that leaves us with
6775          *
6776          * 1) root->orphan_block_rsv - for the orphan deletion.
6777          * 2) rsv - for the truncate reservation, which we will steal from the
6778          * transaction reservation.
6779          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6780          * updating the inode.
6781          */
6782         rsv = btrfs_alloc_block_rsv(root);
6783         if (!rsv)
6784                 return -ENOMEM;
6785         rsv->size = min_size;
6786
6787         /*
6788          * 1 for the truncate slack space
6789          * 1 for the orphan item we're going to add
6790          * 1 for the orphan item deletion
6791          * 1 for updating the inode.
6792          */
6793         trans = btrfs_start_transaction(root, 4);
6794         if (IS_ERR(trans)) {
6795                 err = PTR_ERR(trans);
6796                 goto out;
6797         }
6798
6799         /* Migrate the slack space for the truncate to our reserve */
6800         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6801                                       min_size);
6802         BUG_ON(ret);
6803
6804         ret = btrfs_orphan_add(trans, inode);
6805         if (ret) {
6806                 btrfs_end_transaction(trans, root);
6807                 goto out;
6808         }
6809
6810         /*
6811          * setattr is responsible for setting the ordered_data_close flag,
6812          * but that is only tested during the last file release.  That
6813          * could happen well after the next commit, leaving a great big
6814          * window where new writes may get lost if someone chooses to write
6815          * to this file after truncating to zero
6816          *
6817          * The inode doesn't have any dirty data here, and so if we commit
6818          * this is a noop.  If someone immediately starts writing to the inode
6819          * it is very likely we'll catch some of their writes in this
6820          * transaction, and the commit will find this file on the ordered
6821          * data list with good things to send down.
6822          *
6823          * This is a best effort solution, there is still a window where
6824          * using truncate to replace the contents of the file will
6825          * end up with a zero length file after a crash.
6826          */
6827         if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
6828                                            &BTRFS_I(inode)->runtime_flags))
6829                 btrfs_add_ordered_operation(trans, root, inode);
6830
6831         while (1) {
6832                 ret = btrfs_block_rsv_refill(root, rsv, min_size);
6833                 if (ret) {
6834                         /*
6835                          * This can only happen with the original transaction we
6836                          * started above, every other time we shouldn't have a
6837                          * transaction started yet.
6838                          */
6839                         if (ret == -EAGAIN)
6840                                 goto end_trans;
6841                         err = ret;
6842                         break;
6843                 }
6844
6845                 if (!trans) {
6846                         /* Just need the 1 for updating the inode */
6847                         trans = btrfs_start_transaction(root, 1);
6848                         if (IS_ERR(trans)) {
6849                                 ret = err = PTR_ERR(trans);
6850                                 trans = NULL;
6851                                 break;
6852                         }
6853                 }
6854
6855                 trans->block_rsv = rsv;
6856
6857                 ret = btrfs_truncate_inode_items(trans, root, inode,
6858                                                  inode->i_size,
6859                                                  BTRFS_EXTENT_DATA_KEY);
6860                 if (ret != -EAGAIN) {
6861                         err = ret;
6862                         break;
6863                 }
6864
6865                 trans->block_rsv = &root->fs_info->trans_block_rsv;
6866                 ret = btrfs_update_inode(trans, root, inode);
6867                 if (ret) {
6868                         err = ret;
6869                         break;
6870                 }
6871 end_trans:
6872                 nr = trans->blocks_used;
6873                 btrfs_end_transaction(trans, root);
6874                 trans = NULL;
6875                 btrfs_btree_balance_dirty(root, nr);
6876         }
6877
6878         if (ret == 0 && inode->i_nlink > 0) {
6879                 trans->block_rsv = root->orphan_block_rsv;
6880                 ret = btrfs_orphan_del(trans, inode);
6881                 if (ret)
6882                         err = ret;
6883         } else if (ret && inode->i_nlink > 0) {
6884                 /*
6885                  * Failed to do the truncate, remove us from the in memory
6886                  * orphan list.
6887                  */
6888                 ret = btrfs_orphan_del(NULL, inode);
6889         }
6890
6891         if (trans) {
6892                 trans->block_rsv = &root->fs_info->trans_block_rsv;
6893                 ret = btrfs_update_inode(trans, root, inode);
6894                 if (ret && !err)
6895                         err = ret;
6896
6897                 nr = trans->blocks_used;
6898                 ret = btrfs_end_transaction(trans, root);
6899                 btrfs_btree_balance_dirty(root, nr);
6900         }
6901
6902 out:
6903         btrfs_free_block_rsv(root, rsv);
6904
6905         if (ret && !err)
6906                 err = ret;
6907
6908         return err;
6909 }
6910
6911 /*
6912  * create a new subvolume directory/inode (helper for the ioctl).
6913  */
6914 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6915                              struct btrfs_root *new_root, u64 new_dirid)
6916 {
6917         struct inode *inode;
6918         int err;
6919         u64 index = 0;
6920
6921         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
6922                                 new_dirid, new_dirid,
6923                                 S_IFDIR | (~current_umask() & S_IRWXUGO),
6924                                 &index);
6925         if (IS_ERR(inode))
6926                 return PTR_ERR(inode);
6927         inode->i_op = &btrfs_dir_inode_operations;
6928         inode->i_fop = &btrfs_dir_file_operations;
6929
6930         set_nlink(inode, 1);
6931         btrfs_i_size_write(inode, 0);
6932
6933         err = btrfs_update_inode(trans, new_root, inode);
6934
6935         iput(inode);
6936         return err;
6937 }
6938
6939 struct inode *btrfs_alloc_inode(struct super_block *sb)
6940 {
6941         struct btrfs_inode *ei;
6942         struct inode *inode;
6943
6944         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6945         if (!ei)
6946                 return NULL;
6947
6948         ei->root = NULL;
6949         ei->generation = 0;
6950         ei->last_trans = 0;
6951         ei->last_sub_trans = 0;
6952         ei->logged_trans = 0;
6953         ei->delalloc_bytes = 0;
6954         ei->disk_i_size = 0;
6955         ei->flags = 0;
6956         ei->csum_bytes = 0;
6957         ei->index_cnt = (u64)-1;
6958         ei->last_unlink_trans = 0;
6959
6960         spin_lock_init(&ei->lock);
6961         ei->outstanding_extents = 0;
6962         ei->reserved_extents = 0;
6963
6964         ei->runtime_flags = 0;
6965         ei->force_compress = BTRFS_COMPRESS_NONE;
6966
6967         ei->delayed_node = NULL;
6968
6969         inode = &ei->vfs_inode;
6970         extent_map_tree_init(&ei->extent_tree);
6971         extent_io_tree_init(&ei->io_tree, &inode->i_data);
6972         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
6973         ei->io_tree.track_uptodate = 1;
6974         ei->io_failure_tree.track_uptodate = 1;
6975         mutex_init(&ei->log_mutex);
6976         mutex_init(&ei->delalloc_mutex);
6977         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6978         INIT_LIST_HEAD(&ei->delalloc_inodes);
6979         INIT_LIST_HEAD(&ei->ordered_operations);
6980         RB_CLEAR_NODE(&ei->rb_node);
6981
6982         return inode;
6983 }
6984
6985 static void btrfs_i_callback(struct rcu_head *head)
6986 {
6987         struct inode *inode = container_of(head, struct inode, i_rcu);
6988         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6989 }
6990
6991 void btrfs_destroy_inode(struct inode *inode)
6992 {
6993         struct btrfs_ordered_extent *ordered;
6994         struct btrfs_root *root = BTRFS_I(inode)->root;
6995
6996         WARN_ON(!list_empty(&inode->i_dentry));
6997         WARN_ON(inode->i_data.nrpages);
6998         WARN_ON(BTRFS_I(inode)->outstanding_extents);
6999         WARN_ON(BTRFS_I(inode)->reserved_extents);
7000         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
7001         WARN_ON(BTRFS_I(inode)->csum_bytes);
7002
7003         /*
7004          * This can happen where we create an inode, but somebody else also
7005          * created the same inode and we need to destroy the one we already
7006          * created.
7007          */
7008         if (!root)
7009                 goto free;
7010
7011         /*
7012          * Make sure we're properly removed from the ordered operation
7013          * lists.
7014          */
7015         smp_mb();
7016         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
7017                 spin_lock(&root->fs_info->ordered_extent_lock);
7018                 list_del_init(&BTRFS_I(inode)->ordered_operations);
7019                 spin_unlock(&root->fs_info->ordered_extent_lock);
7020         }
7021
7022         if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
7023                      &BTRFS_I(inode)->runtime_flags)) {
7024                 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
7025                        (unsigned long long)btrfs_ino(inode));
7026                 atomic_dec(&root->orphan_inodes);
7027         }
7028
7029         while (1) {
7030                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7031                 if (!ordered)
7032                         break;
7033                 else {
7034                         printk(KERN_ERR "btrfs found ordered "
7035                                "extent %llu %llu on inode cleanup\n",
7036                                (unsigned long long)ordered->file_offset,
7037                                (unsigned long long)ordered->len);
7038                         btrfs_remove_ordered_extent(inode, ordered);
7039                         btrfs_put_ordered_extent(ordered);
7040                         btrfs_put_ordered_extent(ordered);
7041                 }
7042         }
7043         inode_tree_del(inode);
7044         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
7045 free:
7046         btrfs_remove_delayed_node(inode);
7047         call_rcu(&inode->i_rcu, btrfs_i_callback);
7048 }
7049
7050 int btrfs_drop_inode(struct inode *inode)
7051 {
7052         struct btrfs_root *root = BTRFS_I(inode)->root;
7053
7054         if (btrfs_root_refs(&root->root_item) == 0 &&
7055             !btrfs_is_free_space_inode(inode))
7056                 return 1;
7057         else
7058                 return generic_drop_inode(inode);
7059 }
7060
7061 static void init_once(void *foo)
7062 {
7063         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
7064
7065         inode_init_once(&ei->vfs_inode);
7066 }
7067
7068 void btrfs_destroy_cachep(void)
7069 {
7070         if (btrfs_inode_cachep)
7071                 kmem_cache_destroy(btrfs_inode_cachep);
7072         if (btrfs_trans_handle_cachep)
7073                 kmem_cache_destroy(btrfs_trans_handle_cachep);
7074         if (btrfs_transaction_cachep)
7075                 kmem_cache_destroy(btrfs_transaction_cachep);
7076         if (btrfs_path_cachep)
7077                 kmem_cache_destroy(btrfs_path_cachep);
7078         if (btrfs_free_space_cachep)
7079                 kmem_cache_destroy(btrfs_free_space_cachep);
7080 }
7081
7082 int btrfs_init_cachep(void)
7083 {
7084         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
7085                         sizeof(struct btrfs_inode), 0,
7086                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
7087         if (!btrfs_inode_cachep)
7088                 goto fail;
7089
7090         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
7091                         sizeof(struct btrfs_trans_handle), 0,
7092                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7093         if (!btrfs_trans_handle_cachep)
7094                 goto fail;
7095
7096         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
7097                         sizeof(struct btrfs_transaction), 0,
7098                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7099         if (!btrfs_transaction_cachep)
7100                 goto fail;
7101
7102         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
7103                         sizeof(struct btrfs_path), 0,
7104                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7105         if (!btrfs_path_cachep)
7106                 goto fail;
7107
7108         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
7109                         sizeof(struct btrfs_free_space), 0,
7110                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7111         if (!btrfs_free_space_cachep)
7112                 goto fail;
7113
7114         return 0;
7115 fail:
7116         btrfs_destroy_cachep();
7117         return -ENOMEM;
7118 }
7119
7120 static int btrfs_getattr(struct vfsmount *mnt,
7121                          struct dentry *dentry, struct kstat *stat)
7122 {
7123         struct inode *inode = dentry->d_inode;
7124         u32 blocksize = inode->i_sb->s_blocksize;
7125
7126         generic_fillattr(inode, stat);
7127         stat->dev = BTRFS_I(inode)->root->anon_dev;
7128         stat->blksize = PAGE_CACHE_SIZE;
7129         stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
7130                 ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9;
7131         return 0;
7132 }
7133
7134 /*
7135  * If a file is moved, it will inherit the cow and compression flags of the new
7136  * directory.
7137  */
7138 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
7139 {
7140         struct btrfs_inode *b_dir = BTRFS_I(dir);
7141         struct btrfs_inode *b_inode = BTRFS_I(inode);
7142
7143         if (b_dir->flags & BTRFS_INODE_NODATACOW)
7144                 b_inode->flags |= BTRFS_INODE_NODATACOW;
7145         else
7146                 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
7147
7148         if (b_dir->flags & BTRFS_INODE_COMPRESS) {
7149                 b_inode->flags |= BTRFS_INODE_COMPRESS;
7150                 b_inode->flags &= ~BTRFS_INODE_NOCOMPRESS;
7151         } else {
7152                 b_inode->flags &= ~(BTRFS_INODE_COMPRESS |
7153                                     BTRFS_INODE_NOCOMPRESS);
7154         }
7155 }
7156
7157 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
7158                            struct inode *new_dir, struct dentry *new_dentry)
7159 {
7160         struct btrfs_trans_handle *trans;
7161         struct btrfs_root *root = BTRFS_I(old_dir)->root;
7162         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
7163         struct inode *new_inode = new_dentry->d_inode;
7164         struct inode *old_inode = old_dentry->d_inode;
7165         struct timespec ctime = CURRENT_TIME;
7166         u64 index = 0;
7167         u64 root_objectid;
7168         int ret;
7169         u64 old_ino = btrfs_ino(old_inode);
7170
7171         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
7172                 return -EPERM;
7173
7174         /* we only allow rename subvolume link between subvolumes */
7175         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
7176                 return -EXDEV;
7177
7178         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
7179             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
7180                 return -ENOTEMPTY;
7181
7182         if (S_ISDIR(old_inode->i_mode) && new_inode &&
7183             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
7184                 return -ENOTEMPTY;
7185         /*
7186          * we're using rename to replace one file with another.
7187          * and the replacement file is large.  Start IO on it now so
7188          * we don't add too much work to the end of the transaction
7189          */
7190         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
7191             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
7192                 filemap_flush(old_inode->i_mapping);
7193
7194         /* close the racy window with snapshot create/destroy ioctl */
7195         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7196                 down_read(&root->fs_info->subvol_sem);
7197         /*
7198          * We want to reserve the absolute worst case amount of items.  So if
7199          * both inodes are subvols and we need to unlink them then that would
7200          * require 4 item modifications, but if they are both normal inodes it
7201          * would require 5 item modifications, so we'll assume their normal
7202          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7203          * should cover the worst case number of items we'll modify.
7204          */
7205         trans = btrfs_start_transaction(root, 20);
7206         if (IS_ERR(trans)) {
7207                 ret = PTR_ERR(trans);
7208                 goto out_notrans;
7209         }
7210
7211         if (dest != root)
7212                 btrfs_record_root_in_trans(trans, dest);
7213
7214         ret = btrfs_set_inode_index(new_dir, &index);
7215         if (ret)
7216                 goto out_fail;
7217
7218         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7219                 /* force full log commit if subvolume involved. */
7220                 root->fs_info->last_trans_log_full_commit = trans->transid;
7221         } else {
7222                 ret = btrfs_insert_inode_ref(trans, dest,
7223                                              new_dentry->d_name.name,
7224                                              new_dentry->d_name.len,
7225                                              old_ino,
7226                                              btrfs_ino(new_dir), index);
7227                 if (ret)
7228                         goto out_fail;
7229                 /*
7230                  * this is an ugly little race, but the rename is required
7231                  * to make sure that if we crash, the inode is either at the
7232                  * old name or the new one.  pinning the log transaction lets
7233                  * us make sure we don't allow a log commit to come in after
7234                  * we unlink the name but before we add the new name back in.
7235                  */
7236                 btrfs_pin_log_trans(root);
7237         }
7238         /*
7239          * make sure the inode gets flushed if it is replacing
7240          * something.
7241          */
7242         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7243                 btrfs_add_ordered_operation(trans, root, old_inode);
7244
7245         inode_inc_iversion(old_dir);
7246         inode_inc_iversion(new_dir);
7247         inode_inc_iversion(old_inode);
7248         old_dir->i_ctime = old_dir->i_mtime = ctime;
7249         new_dir->i_ctime = new_dir->i_mtime = ctime;
7250         old_inode->i_ctime = ctime;
7251
7252         if (old_dentry->d_parent != new_dentry->d_parent)
7253                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7254
7255         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7256                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7257                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7258                                         old_dentry->d_name.name,
7259                                         old_dentry->d_name.len);
7260         } else {
7261                 ret = __btrfs_unlink_inode(trans, root, old_dir,
7262                                         old_dentry->d_inode,
7263                                         old_dentry->d_name.name,
7264                                         old_dentry->d_name.len);
7265                 if (!ret)
7266                         ret = btrfs_update_inode(trans, root, old_inode);
7267         }
7268         if (ret) {
7269                 btrfs_abort_transaction(trans, root, ret);
7270                 goto out_fail;
7271         }
7272
7273         if (new_inode) {
7274                 inode_inc_iversion(new_inode);
7275                 new_inode->i_ctime = CURRENT_TIME;
7276                 if (unlikely(btrfs_ino(new_inode) ==
7277                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7278                         root_objectid = BTRFS_I(new_inode)->location.objectid;
7279                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
7280                                                 root_objectid,
7281                                                 new_dentry->d_name.name,
7282                                                 new_dentry->d_name.len);
7283                         BUG_ON(new_inode->i_nlink == 0);
7284                 } else {
7285                         ret = btrfs_unlink_inode(trans, dest, new_dir,
7286                                                  new_dentry->d_inode,
7287                                                  new_dentry->d_name.name,
7288                                                  new_dentry->d_name.len);
7289                 }
7290                 if (!ret && new_inode->i_nlink == 0) {
7291                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7292                         BUG_ON(ret);
7293                 }
7294                 if (ret) {
7295                         btrfs_abort_transaction(trans, root, ret);
7296                         goto out_fail;
7297                 }
7298         }
7299
7300         fixup_inode_flags(new_dir, old_inode);
7301
7302         ret = btrfs_add_link(trans, new_dir, old_inode,
7303                              new_dentry->d_name.name,
7304                              new_dentry->d_name.len, 0, index);
7305         if (ret) {
7306                 btrfs_abort_transaction(trans, root, ret);
7307                 goto out_fail;
7308         }
7309
7310         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7311                 struct dentry *parent = new_dentry->d_parent;
7312                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7313                 btrfs_end_log_trans(root);
7314         }
7315 out_fail:
7316         btrfs_end_transaction(trans, root);
7317 out_notrans:
7318         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7319                 up_read(&root->fs_info->subvol_sem);
7320
7321         return ret;
7322 }
7323
7324 /*
7325  * some fairly slow code that needs optimization. This walks the list
7326  * of all the inodes with pending delalloc and forces them to disk.
7327  */
7328 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7329 {
7330         struct list_head *head = &root->fs_info->delalloc_inodes;
7331         struct btrfs_inode *binode;
7332         struct inode *inode;
7333
7334         if (root->fs_info->sb->s_flags & MS_RDONLY)
7335                 return -EROFS;
7336
7337         spin_lock(&root->fs_info->delalloc_lock);
7338         while (!list_empty(head)) {
7339                 binode = list_entry(head->next, struct btrfs_inode,
7340                                     delalloc_inodes);
7341                 inode = igrab(&binode->vfs_inode);
7342                 if (!inode)
7343                         list_del_init(&binode->delalloc_inodes);
7344                 spin_unlock(&root->fs_info->delalloc_lock);
7345                 if (inode) {
7346                         filemap_flush(inode->i_mapping);
7347                         if (delay_iput)
7348                                 btrfs_add_delayed_iput(inode);
7349                         else
7350                                 iput(inode);
7351                 }
7352                 cond_resched();
7353                 spin_lock(&root->fs_info->delalloc_lock);
7354         }
7355         spin_unlock(&root->fs_info->delalloc_lock);
7356
7357         /* the filemap_flush will queue IO into the worker threads, but
7358          * we have to make sure the IO is actually started and that
7359          * ordered extents get created before we return
7360          */
7361         atomic_inc(&root->fs_info->async_submit_draining);
7362         while (atomic_read(&root->fs_info->nr_async_submits) ||
7363               atomic_read(&root->fs_info->async_delalloc_pages)) {
7364                 wait_event(root->fs_info->async_submit_wait,
7365                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7366                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7367         }
7368         atomic_dec(&root->fs_info->async_submit_draining);
7369         return 0;
7370 }
7371
7372 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7373                          const char *symname)
7374 {
7375         struct btrfs_trans_handle *trans;
7376         struct btrfs_root *root = BTRFS_I(dir)->root;
7377         struct btrfs_path *path;
7378         struct btrfs_key key;
7379         struct inode *inode = NULL;
7380         int err;
7381         int drop_inode = 0;
7382         u64 objectid;
7383         u64 index = 0 ;
7384         int name_len;
7385         int datasize;
7386         unsigned long ptr;
7387         struct btrfs_file_extent_item *ei;
7388         struct extent_buffer *leaf;
7389         unsigned long nr = 0;
7390
7391         name_len = strlen(symname) + 1;
7392         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7393                 return -ENAMETOOLONG;
7394
7395         /*
7396          * 2 items for inode item and ref
7397          * 2 items for dir items
7398          * 1 item for xattr if selinux is on
7399          */
7400         trans = btrfs_start_transaction(root, 5);
7401         if (IS_ERR(trans))
7402                 return PTR_ERR(trans);
7403
7404         err = btrfs_find_free_ino(root, &objectid);
7405         if (err)
7406                 goto out_unlock;
7407
7408         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7409                                 dentry->d_name.len, btrfs_ino(dir), objectid,
7410                                 S_IFLNK|S_IRWXUGO, &index);
7411         if (IS_ERR(inode)) {
7412                 err = PTR_ERR(inode);
7413                 goto out_unlock;
7414         }
7415
7416         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7417         if (err) {
7418                 drop_inode = 1;
7419                 goto out_unlock;
7420         }
7421
7422         /*
7423         * If the active LSM wants to access the inode during
7424         * d_instantiate it needs these. Smack checks to see
7425         * if the filesystem supports xattrs by looking at the
7426         * ops vector.
7427         */
7428         inode->i_fop = &btrfs_file_operations;
7429         inode->i_op = &btrfs_file_inode_operations;
7430
7431         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7432         if (err)
7433                 drop_inode = 1;
7434         else {
7435                 inode->i_mapping->a_ops = &btrfs_aops;
7436                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7437                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7438         }
7439         if (drop_inode)
7440                 goto out_unlock;
7441
7442         path = btrfs_alloc_path();
7443         if (!path) {
7444                 err = -ENOMEM;
7445                 drop_inode = 1;
7446                 goto out_unlock;
7447         }
7448         key.objectid = btrfs_ino(inode);
7449         key.offset = 0;
7450         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7451         datasize = btrfs_file_extent_calc_inline_size(name_len);
7452         err = btrfs_insert_empty_item(trans, root, path, &key,
7453                                       datasize);
7454         if (err) {
7455                 drop_inode = 1;
7456                 btrfs_free_path(path);
7457                 goto out_unlock;
7458         }
7459         leaf = path->nodes[0];
7460         ei = btrfs_item_ptr(leaf, path->slots[0],
7461                             struct btrfs_file_extent_item);
7462         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7463         btrfs_set_file_extent_type(leaf, ei,
7464                                    BTRFS_FILE_EXTENT_INLINE);
7465         btrfs_set_file_extent_encryption(leaf, ei, 0);
7466         btrfs_set_file_extent_compression(leaf, ei, 0);
7467         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7468         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7469
7470         ptr = btrfs_file_extent_inline_start(ei);
7471         write_extent_buffer(leaf, symname, ptr, name_len);
7472         btrfs_mark_buffer_dirty(leaf);
7473         btrfs_free_path(path);
7474
7475         inode->i_op = &btrfs_symlink_inode_operations;
7476         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7477         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7478         inode_set_bytes(inode, name_len);
7479         btrfs_i_size_write(inode, name_len - 1);
7480         err = btrfs_update_inode(trans, root, inode);
7481         if (err)
7482                 drop_inode = 1;
7483
7484 out_unlock:
7485         if (!err)
7486                 d_instantiate(dentry, inode);
7487         nr = trans->blocks_used;
7488         btrfs_end_transaction(trans, root);
7489         if (drop_inode) {
7490                 inode_dec_link_count(inode);
7491                 iput(inode);
7492         }
7493         btrfs_btree_balance_dirty(root, nr);
7494         return err;
7495 }
7496
7497 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7498                                        u64 start, u64 num_bytes, u64 min_size,
7499                                        loff_t actual_len, u64 *alloc_hint,
7500                                        struct btrfs_trans_handle *trans)
7501 {
7502         struct btrfs_root *root = BTRFS_I(inode)->root;
7503         struct btrfs_key ins;
7504         u64 cur_offset = start;
7505         u64 i_size;
7506         int ret = 0;
7507         bool own_trans = true;
7508
7509         if (trans)
7510                 own_trans = false;
7511         while (num_bytes > 0) {
7512                 if (own_trans) {
7513                         trans = btrfs_start_transaction(root, 3);
7514                         if (IS_ERR(trans)) {
7515                                 ret = PTR_ERR(trans);
7516                                 break;
7517                         }
7518                 }
7519
7520                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7521                                            0, *alloc_hint, &ins, 1);
7522                 if (ret) {
7523                         if (own_trans)
7524                                 btrfs_end_transaction(trans, root);
7525                         break;
7526                 }
7527
7528                 ret = insert_reserved_file_extent(trans, inode,
7529                                                   cur_offset, ins.objectid,
7530                                                   ins.offset, ins.offset,
7531                                                   ins.offset, 0, 0, 0,
7532                                                   BTRFS_FILE_EXTENT_PREALLOC);
7533                 if (ret) {
7534                         btrfs_abort_transaction(trans, root, ret);
7535                         if (own_trans)
7536                                 btrfs_end_transaction(trans, root);
7537                         break;
7538                 }
7539                 btrfs_drop_extent_cache(inode, cur_offset,
7540                                         cur_offset + ins.offset -1, 0);
7541
7542                 num_bytes -= ins.offset;
7543                 cur_offset += ins.offset;
7544                 *alloc_hint = ins.objectid + ins.offset;
7545
7546                 inode_inc_iversion(inode);
7547                 inode->i_ctime = CURRENT_TIME;
7548                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7549                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7550                     (actual_len > inode->i_size) &&
7551                     (cur_offset > inode->i_size)) {
7552                         if (cur_offset > actual_len)
7553                                 i_size = actual_len;
7554                         else
7555                                 i_size = cur_offset;
7556                         i_size_write(inode, i_size);
7557                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7558                 }
7559
7560                 ret = btrfs_update_inode(trans, root, inode);
7561
7562                 if (ret) {
7563                         btrfs_abort_transaction(trans, root, ret);
7564                         if (own_trans)
7565                                 btrfs_end_transaction(trans, root);
7566                         break;
7567                 }
7568
7569                 if (own_trans)
7570                         btrfs_end_transaction(trans, root);
7571         }
7572         return ret;
7573 }
7574
7575 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7576                               u64 start, u64 num_bytes, u64 min_size,
7577                               loff_t actual_len, u64 *alloc_hint)
7578 {
7579         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7580                                            min_size, actual_len, alloc_hint,
7581                                            NULL);
7582 }
7583
7584 int btrfs_prealloc_file_range_trans(struct inode *inode,
7585                                     struct btrfs_trans_handle *trans, int mode,
7586                                     u64 start, u64 num_bytes, u64 min_size,
7587                                     loff_t actual_len, u64 *alloc_hint)
7588 {
7589         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7590                                            min_size, actual_len, alloc_hint, trans);
7591 }
7592
7593 static int btrfs_set_page_dirty(struct page *page)
7594 {
7595         return __set_page_dirty_nobuffers(page);
7596 }
7597
7598 static int btrfs_permission(struct inode *inode, int mask)
7599 {
7600         struct btrfs_root *root = BTRFS_I(inode)->root;
7601         umode_t mode = inode->i_mode;
7602
7603         if (mask & MAY_WRITE &&
7604             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7605                 if (btrfs_root_readonly(root))
7606                         return -EROFS;
7607                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7608                         return -EACCES;
7609         }
7610         return generic_permission(inode, mask);
7611 }
7612
7613 static const struct inode_operations btrfs_dir_inode_operations = {
7614         .getattr        = btrfs_getattr,
7615         .lookup         = btrfs_lookup,
7616         .create         = btrfs_create,
7617         .unlink         = btrfs_unlink,
7618         .link           = btrfs_link,
7619         .mkdir          = btrfs_mkdir,
7620         .rmdir          = btrfs_rmdir,
7621         .rename         = btrfs_rename,
7622         .symlink        = btrfs_symlink,
7623         .setattr        = btrfs_setattr,
7624         .mknod          = btrfs_mknod,
7625         .setxattr       = btrfs_setxattr,
7626         .getxattr       = btrfs_getxattr,
7627         .listxattr      = btrfs_listxattr,
7628         .removexattr    = btrfs_removexattr,
7629         .permission     = btrfs_permission,
7630         .get_acl        = btrfs_get_acl,
7631 };
7632 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7633         .lookup         = btrfs_lookup,
7634         .permission     = btrfs_permission,
7635         .get_acl        = btrfs_get_acl,
7636 };
7637
7638 static const struct file_operations btrfs_dir_file_operations = {
7639         .llseek         = generic_file_llseek,
7640         .read           = generic_read_dir,
7641         .readdir        = btrfs_real_readdir,
7642         .unlocked_ioctl = btrfs_ioctl,
7643 #ifdef CONFIG_COMPAT
7644         .compat_ioctl   = btrfs_ioctl,
7645 #endif
7646         .release        = btrfs_release_file,
7647         .fsync          = btrfs_sync_file,
7648 };
7649
7650 static struct extent_io_ops btrfs_extent_io_ops = {
7651         .fill_delalloc = run_delalloc_range,
7652         .submit_bio_hook = btrfs_submit_bio_hook,
7653         .merge_bio_hook = btrfs_merge_bio_hook,
7654         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7655         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7656         .writepage_start_hook = btrfs_writepage_start_hook,
7657         .set_bit_hook = btrfs_set_bit_hook,
7658         .clear_bit_hook = btrfs_clear_bit_hook,
7659         .merge_extent_hook = btrfs_merge_extent_hook,
7660         .split_extent_hook = btrfs_split_extent_hook,
7661 };
7662
7663 /*
7664  * btrfs doesn't support the bmap operation because swapfiles
7665  * use bmap to make a mapping of extents in the file.  They assume
7666  * these extents won't change over the life of the file and they
7667  * use the bmap result to do IO directly to the drive.
7668  *
7669  * the btrfs bmap call would return logical addresses that aren't
7670  * suitable for IO and they also will change frequently as COW
7671  * operations happen.  So, swapfile + btrfs == corruption.
7672  *
7673  * For now we're avoiding this by dropping bmap.
7674  */
7675 static const struct address_space_operations btrfs_aops = {
7676         .readpage       = btrfs_readpage,
7677         .writepage      = btrfs_writepage,
7678         .writepages     = btrfs_writepages,
7679         .readpages      = btrfs_readpages,
7680         .direct_IO      = btrfs_direct_IO,
7681         .invalidatepage = btrfs_invalidatepage,
7682         .releasepage    = btrfs_releasepage,
7683         .set_page_dirty = btrfs_set_page_dirty,
7684         .error_remove_page = generic_error_remove_page,
7685 };
7686
7687 static const struct address_space_operations btrfs_symlink_aops = {
7688         .readpage       = btrfs_readpage,
7689         .writepage      = btrfs_writepage,
7690         .invalidatepage = btrfs_invalidatepage,
7691         .releasepage    = btrfs_releasepage,
7692 };
7693
7694 static const struct inode_operations btrfs_file_inode_operations = {
7695         .getattr        = btrfs_getattr,
7696         .setattr        = btrfs_setattr,
7697         .setxattr       = btrfs_setxattr,
7698         .getxattr       = btrfs_getxattr,
7699         .listxattr      = btrfs_listxattr,
7700         .removexattr    = btrfs_removexattr,
7701         .permission     = btrfs_permission,
7702         .fiemap         = btrfs_fiemap,
7703         .get_acl        = btrfs_get_acl,
7704         .update_time    = btrfs_update_time,
7705 };
7706 static const struct inode_operations btrfs_special_inode_operations = {
7707         .getattr        = btrfs_getattr,
7708         .setattr        = btrfs_setattr,
7709         .permission     = btrfs_permission,
7710         .setxattr       = btrfs_setxattr,
7711         .getxattr       = btrfs_getxattr,
7712         .listxattr      = btrfs_listxattr,
7713         .removexattr    = btrfs_removexattr,
7714         .get_acl        = btrfs_get_acl,
7715         .update_time    = btrfs_update_time,
7716 };
7717 static const struct inode_operations btrfs_symlink_inode_operations = {
7718         .readlink       = generic_readlink,
7719         .follow_link    = page_follow_link_light,
7720         .put_link       = page_put_link,
7721         .getattr        = btrfs_getattr,
7722         .setattr        = btrfs_setattr,
7723         .permission     = btrfs_permission,
7724         .setxattr       = btrfs_setxattr,
7725         .getxattr       = btrfs_getxattr,
7726         .listxattr      = btrfs_listxattr,
7727         .removexattr    = btrfs_removexattr,
7728         .get_acl        = btrfs_get_acl,
7729         .update_time    = btrfs_update_time,
7730 };
7731
7732 const struct dentry_operations btrfs_dentry_operations = {
7733         .d_delete       = btrfs_dentry_delete,
7734         .d_release      = btrfs_dentry_release,
7735 };