fs/btrfs/inode.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/pagemap.h>
  25 #include <linux/highmem.h>
  26 #include <linux/time.h>
  27 #include <linux/init.h>
  28 #include <linux/string.h>
  29 #include <linux/backing-dev.h>
  30 #include <linux/mpage.h>
  31 #include <linux/swap.h>
  32 #include <linux/writeback.h>
  33 #include <linux/statfs.h>
  34 #include <linux/compat.h>
  35 #include <linux/bit_spinlock.h>
  36 #include <linux/xattr.h>
  37 #include <linux/posix_acl.h>
  38 #include <linux/falloc.h>
  39 #include <linux/slab.h>
  40 #include <linux/ratelimit.h>
  41 #include <linux/mount.h>
  42 #include "compat.h"
  43 #include "ctree.h"
  44 #include "disk-io.h"
  45 #include "transaction.h"
  46 #include "btrfs_inode.h"
  47 #include "ioctl.h"
  48 #include "print-tree.h"
  49 #include "ordered-data.h"
  50 #include "xattr.h"
  51 #include "tree-log.h"
  52 #include "volumes.h"
  53 #include "compression.h"
  54 #include "locking.h"
  55 #include "free-space-cache.h"
  56 #include "inode-map.h"
  57
  58 struct btrfs_iget_args {
  59         u64 ino;
  60         struct btrfs_root *root;
  61 };
  62
  63 static const struct inode_operations btrfs_dir_inode_operations;
  64 static const struct inode_operations btrfs_symlink_inode_operations;
  65 static const struct inode_operations btrfs_dir_ro_inode_operations;
  66 static const struct inode_operations btrfs_special_inode_operations;
  67 static const struct inode_operations btrfs_file_inode_operations;
  68 static const struct address_space_operations btrfs_aops;
  69 static const struct address_space_operations btrfs_symlink_aops;
  70 static const struct file_operations btrfs_dir_file_operations;
  71 static struct extent_io_ops btrfs_extent_io_ops;
  72
  73 static struct kmem_cache *btrfs_inode_cachep;
  74 struct kmem_cache *btrfs_trans_handle_cachep;
  75 struct kmem_cache *btrfs_transaction_cachep;
  76 struct kmem_cache *btrfs_path_cachep;
  77 struct kmem_cache *btrfs_free_space_cachep;
  78
  79 #define S_SHIFT 12
  80 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  81         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  82         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  83         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  84         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  85         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  86         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  87         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  88 };
  89
  90 static int btrfs_setsize(struct inode *inode, loff_t newsize);
  91 static int btrfs_truncate(struct inode *inode);
  92 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
  93 static noinline int cow_file_range(struct inode *inode,
  94                                    struct page *locked_page,
  95                                    u64 start, u64 end, int *page_started,
  96                                    unsigned long *nr_written, int unlock);
  97 static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
  98                                 struct btrfs_root *root, struct inode *inode);
  99
 100 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
 101                                      struct inode *inode,  struct inode *dir,
 102                                      const struct qstr *qstr)
 103 {
 104         int err;
 105
 106         err = btrfs_init_acl(trans, inode, dir);
 107         if (!err)
 108                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 109         return err;
 110 }
 111
 112 /*
 113  * this does all the hard work for inserting an inline extent into
 114  * the btree.  The caller should have done a btrfs_drop_extents so that
 115  * no overlapping inline items exist in the btree
 116  */
 117 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 118                                 struct btrfs_root *root, struct inode *inode,
 119                                 u64 start, size_t size, size_t compressed_size,
 120                                 int compress_type,
 121                                 struct page **compressed_pages)
 122 {
 123         struct btrfs_key key;
 124         struct btrfs_path *path;
 125         struct extent_buffer *leaf;
 126         struct page *page = NULL;
 127         char *kaddr;
 128         unsigned long ptr;
 129         struct btrfs_file_extent_item *ei;
 130         int err = 0;
 131         int ret;
 132         size_t cur_size = size;
 133         size_t datasize;
 134         unsigned long offset;
 135
 136         if (compressed_size && compressed_pages)
 137                 cur_size = compressed_size;
 138
 139         path = btrfs_alloc_path();
 140         if (!path)
 141                 return -ENOMEM;
 142
 143         path->leave_spinning = 1;
 144
 145         key.objectid = btrfs_ino(inode);
 146         key.offset = start;
 147         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 148         datasize = btrfs_file_extent_calc_inline_size(cur_size);
 149
 150         inode_add_bytes(inode, size);
 151         ret = btrfs_insert_empty_item(trans, root, path, &key,
 152                                       datasize);
 153         if (ret) {
 154                 err = ret;
 155                 goto fail;
 156         }
 157         leaf = path->nodes[0];
 158         ei = btrfs_item_ptr(leaf, path->slots[0],
 159                             struct btrfs_file_extent_item);
 160         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 161         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 162         btrfs_set_file_extent_encryption(leaf, ei, 0);
 163         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 164         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 165         ptr = btrfs_file_extent_inline_start(ei);
 166
 167         if (compress_type != BTRFS_COMPRESS_NONE) {
 168                 struct page *cpage;
 169                 int i = 0;
 170                 while (compressed_size > 0) {
 171                         cpage = compressed_pages[i];
 172                         cur_size = min_t(unsigned long, compressed_size,
 173                                        PAGE_CACHE_SIZE);
 174
 175                         kaddr = kmap_atomic(cpage);
 176                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 177                         kunmap_atomic(kaddr);
 178
 179                         i++;
 180                         ptr += cur_size;
 181                         compressed_size -= cur_size;
 182                 }
 183                 btrfs_set_file_extent_compression(leaf, ei,
 184                                                   compress_type);
 185         } else {
 186                 page = find_get_page(inode->i_mapping,
 187                                      start >> PAGE_CACHE_SHIFT);
 188                 btrfs_set_file_extent_compression(leaf, ei, 0);
 189                 kaddr = kmap_atomic(page);
 190                 offset = start & (PAGE_CACHE_SIZE - 1);
 191                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 192                 kunmap_atomic(kaddr);
 193                 page_cache_release(page);
 194         }
 195         btrfs_mark_buffer_dirty(leaf);
 196         btrfs_free_path(path);
 197
 198         /*
 199          * we're an inline extent, so nobody can
 200          * extend the file past i_size without locking
 201          * a page we already have locked.
 202          *
 203          * We must do any isize and inode updates
 204          * before we unlock the pages.  Otherwise we
 205          * could end up racing with unlink.
 206          */
 207         BTRFS_I(inode)->disk_i_size = inode->i_size;
 208         ret = btrfs_update_inode(trans, root, inode);
 209
 210         return ret;
 211 fail:
 212         btrfs_free_path(path);
 213         return err;
 214 }
 215
 216
 217 /*
 218  * conditionally insert an inline extent into the file.  This
 219  * does the checks required to make sure the data is small enough
 220  * to fit as an inline extent.
 221  */
 222 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 223                                  struct btrfs_root *root,
 224                                  struct inode *inode, u64 start, u64 end,
 225                                  size_t compressed_size, int compress_type,
 226                                  struct page **compressed_pages)
 227 {
 228         u64 isize = i_size_read(inode);
 229         u64 actual_end = min(end + 1, isize);
 230         u64 inline_len = actual_end - start;
 231         u64 aligned_end = (end + root->sectorsize - 1) &
 232                         ~((u64)root->sectorsize - 1);
 233         u64 hint_byte;
 234         u64 data_len = inline_len;
 235         int ret;
 236
 237         if (compressed_size)
 238                 data_len = compressed_size;
 239
 240         if (start > 0 ||
 241             actual_end >= PAGE_CACHE_SIZE ||
 242             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 243             (!compressed_size &&
 244             (actual_end & (root->sectorsize - 1)) == 0) ||
 245             end + 1 < isize ||
 246             data_len > root->fs_info->max_inline) {
 247                 return 1;
 248         }
 249
 250         ret = btrfs_drop_extents(trans, inode, start, aligned_end,
 251                                  &hint_byte, 1);
 252         if (ret)
 253                 return ret;
 254
 255         if (isize > actual_end)
 256                 inline_len = min_t(u64, isize, actual_end);
 257         ret = insert_inline_extent(trans, root, inode, start,
 258                                    inline_len, compressed_size,
 259                                    compress_type, compressed_pages);
 260         if (ret && ret != -ENOSPC) {
 261                 btrfs_abort_transaction(trans, root, ret);
 262                 return ret;
 263         } else if (ret == -ENOSPC) {
 264                 return 1;
 265         }
 266
 267         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 268         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 269         return 0;
 270 }
 271
 272 struct async_extent {
 273         u64 start;
 274         u64 ram_size;
 275         u64 compressed_size;
 276         struct page **pages;
 277         unsigned long nr_pages;
 278         int compress_type;
 279         struct list_head list;
 280 };
 281
 282 struct async_cow {
 283         struct inode *inode;
 284         struct btrfs_root *root;
 285         struct page *locked_page;
 286         u64 start;
 287         u64 end;
 288         struct list_head extents;
 289         struct btrfs_work work;
 290 };
 291
 292 static noinline int add_async_extent(struct async_cow *cow,
 293                                      u64 start, u64 ram_size,
 294                                      u64 compressed_size,
 295                                      struct page **pages,
 296                                      unsigned long nr_pages,
 297                                      int compress_type)
 298 {
 299         struct async_extent *async_extent;
 300
 301         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 302         BUG_ON(!async_extent); /* -ENOMEM */
 303         async_extent->start = start;
 304         async_extent->ram_size = ram_size;
 305         async_extent->compressed_size = compressed_size;
 306         async_extent->pages = pages;
 307         async_extent->nr_pages = nr_pages;
 308         async_extent->compress_type = compress_type;
 309         list_add_tail(&async_extent->list, &cow->extents);
 310         return 0;
 311 }
 312
 313 /*
 314  * we create compressed extents in two phases.  The first
 315  * phase compresses a range of pages that have already been
 316  * locked (both pages and state bits are locked).
 317  *
 318  * This is done inside an ordered work queue, and the compression
 319  * is spread across many cpus.  The actual IO submission is step
 320  * two, and the ordered work queue takes care of making sure that
 321  * happens in the same order things were put onto the queue by
 322  * writepages and friends.
 323  *
 324  * If this code finds it can't get good compression, it puts an
 325  * entry onto the work queue to write the uncompressed bytes.  This
 326  * makes sure that both compressed inodes and uncompressed inodes
 327  * are written in the same order that pdflush sent them down.
 328  */
 329 static noinline int compress_file_range(struct inode *inode,
 330                                         struct page *locked_page,
 331                                         u64 start, u64 end,
 332                                         struct async_cow *async_cow,
 333                                         int *num_added)
 334 {
 335         struct btrfs_root *root = BTRFS_I(inode)->root;
 336         struct btrfs_trans_handle *trans;
 337         u64 num_bytes;
 338         u64 blocksize = root->sectorsize;
 339         u64 actual_end;
 340         u64 isize = i_size_read(inode);
 341         int ret = 0;
 342         struct page **pages = NULL;
 343         unsigned long nr_pages;
 344         unsigned long nr_pages_ret = 0;
 345         unsigned long total_compressed = 0;
 346         unsigned long total_in = 0;
 347         unsigned long max_compressed = 128 * 1024;
 348         unsigned long max_uncompressed = 128 * 1024;
 349         int i;
 350         int will_compress;
 351         int compress_type = root->fs_info->compress_type;
 352
 353         /* if this is a small write inside eof, kick off a defrag */
 354         if ((end - start + 1) < 16 * 1024 &&
 355             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 356                 btrfs_add_inode_defrag(NULL, inode);
 357
 358         actual_end = min_t(u64, isize, end + 1);
 359 again:
 360         will_compress = 0;
 361         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 362         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 363
 364         /*
 365          * we don't want to send crud past the end of i_size through
 366          * compression, that's just a waste of CPU time.  So, if the
 367          * end of the file is before the start of our current
 368          * requested range of bytes, we bail out to the uncompressed
 369          * cleanup code that can deal with all of this.
 370          *
 371          * It isn't really the fastest way to fix things, but this is a
 372          * very uncommon corner.
 373          */
 374         if (actual_end <= start)
 375                 goto cleanup_and_bail_uncompressed;
 376
 377         total_compressed = actual_end - start;
 378
 379         /* we want to make sure that amount of ram required to uncompress
 380          * an extent is reasonable, so we limit the total size in ram
 381          * of a compressed extent to 128k.  This is a crucial number
 382          * because it also controls how easily we can spread reads across
 383          * cpus for decompression.
 384          *
 385          * We also want to make sure the amount of IO required to do
 386          * a random read is reasonably small, so we limit the size of
 387          * a compressed extent to 128k.
 388          */
 389         total_compressed = min(total_compressed, max_uncompressed);
 390         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 391         num_bytes = max(blocksize,  num_bytes);
 392         total_in = 0;
 393         ret = 0;
 394
 395         /*
 396          * we do compression for mount -o compress and when the
 397          * inode has not been flagged as nocompress.  This flag can
 398          * change at any time if we discover bad compression ratios.
 399          */
 400         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 401             (btrfs_test_opt(root, COMPRESS) ||
 402              (BTRFS_I(inode)->force_compress) ||
 403              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 404                 WARN_ON(pages);
 405                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 406                 if (!pages) {
 407                         /* just bail out to the uncompressed code */
 408                         goto cont;
 409                 }
 410
 411                 if (BTRFS_I(inode)->force_compress)
 412                         compress_type = BTRFS_I(inode)->force_compress;
 413
 414                 ret = btrfs_compress_pages(compress_type,
 415                                            inode->i_mapping, start,
 416                                            total_compressed, pages,
 417                                            nr_pages, &nr_pages_ret,
 418                                            &total_in,
 419                                            &total_compressed,
 420                                            max_compressed);
 421
 422                 if (!ret) {
 423                         unsigned long offset = total_compressed &
 424                                 (PAGE_CACHE_SIZE - 1);
 425                         struct page *page = pages[nr_pages_ret - 1];
 426                         char *kaddr;
 427
 428                         /* zero the tail end of the last page, we might be
 429                          * sending it down to disk
 430                          */
 431                         if (offset) {
 432                                 kaddr = kmap_atomic(page);
 433                                 memset(kaddr + offset, 0,
 434                                        PAGE_CACHE_SIZE - offset);
 435                                 kunmap_atomic(kaddr);
 436                         }
 437                         will_compress = 1;
 438                 }
 439         }
 440 cont:
 441         if (start == 0) {
 442                 trans = btrfs_join_transaction(root);
 443                 if (IS_ERR(trans)) {
 444                         ret = PTR_ERR(trans);
 445                         trans = NULL;
 446                         goto cleanup_and_out;
 447                 }
 448                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 449
 450                 /* lets try to make an inline extent */
 451                 if (ret || total_in < (actual_end - start)) {
 452                         /* we didn't compress the entire range, try
 453                          * to make an uncompressed inline extent.
 454                          */
 455                         ret = cow_file_range_inline(trans, root, inode,
 456                                                     start, end, 0, 0, NULL);
 457                 } else {
 458                         /* try making a compressed inline extent */
 459                         ret = cow_file_range_inline(trans, root, inode,
 460                                                     start, end,
 461                                                     total_compressed,
 462                                                     compress_type, pages);
 463                 }
 464                 if (ret <= 0) {
 465                         /*
 466                          * inline extent creation worked or returned error,
 467                          * we don't need to create any more async work items.
 468                          * Unlock and free up our temp pages.
 469                          */
 470                         extent_clear_unlock_delalloc(inode,
 471                              &BTRFS_I(inode)->io_tree,
 472                              start, end, NULL,
 473                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
 474                              EXTENT_CLEAR_DELALLOC |
 475                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
 476
 477                         btrfs_end_transaction(trans, root);
 478                         goto free_pages_out;
 479                 }
 480                 btrfs_end_transaction(trans, root);
 481         }
 482
 483         if (will_compress) {
 484                 /*
 485                  * we aren't doing an inline extent round the compressed size
 486                  * up to a block size boundary so the allocator does sane
 487                  * things
 488                  */
 489                 total_compressed = (total_compressed + blocksize - 1) &
 490                         ~(blocksize - 1);
 491
 492                 /*
 493                  * one last check to make sure the compression is really a
 494                  * win, compare the page count read with the blocks on disk
 495                  */
 496                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
 497                         ~(PAGE_CACHE_SIZE - 1);
 498                 if (total_compressed >= total_in) {
 499                         will_compress = 0;
 500                 } else {
 501                         num_bytes = total_in;
 502                 }
 503         }
 504         if (!will_compress && pages) {
 505                 /*
 506                  * the compression code ran but failed to make things smaller,
 507                  * free any pages it allocated and our page pointer array
 508                  */
 509                 for (i = 0; i < nr_pages_ret; i++) {
 510                         WARN_ON(pages[i]->mapping);
 511                         page_cache_release(pages[i]);
 512                 }
 513                 kfree(pages);
 514                 pages = NULL;
 515                 total_compressed = 0;
 516                 nr_pages_ret = 0;
 517
 518                 /* flag the file so we don't compress in the future */
 519                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 520                     !(BTRFS_I(inode)->force_compress)) {
 521                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 522                 }
 523         }
 524         if (will_compress) {
 525                 *num_added += 1;
 526
 527                 /* the async work queues will take care of doing actual
 528                  * allocation on disk for these compressed pages,
 529                  * and will submit them to the elevator.
 530                  */
 531                 add_async_extent(async_cow, start, num_bytes,
 532                                  total_compressed, pages, nr_pages_ret,
 533                                  compress_type);
 534
 535                 if (start + num_bytes < end) {
 536                         start += num_bytes;
 537                         pages = NULL;
 538                         cond_resched();
 539                         goto again;
 540                 }
 541         } else {
 542 cleanup_and_bail_uncompressed:
 543                 /*
 544                  * No compression, but we still need to write the pages in
 545                  * the file we've been given so far.  redirty the locked
 546                  * page if it corresponds to our extent and set things up
 547                  * for the async work queue to run cow_file_range to do
 548                  * the normal delalloc dance
 549                  */
 550                 if (page_offset(locked_page) >= start &&
 551                     page_offset(locked_page) <= end) {
 552                         __set_page_dirty_nobuffers(locked_page);
 553                         /* unlocked later on in the async handlers */
 554                 }
 555                 add_async_extent(async_cow, start, end - start + 1,
 556                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 557                 *num_added += 1;
 558         }
 559
 560 out:
 561         return ret;
 562
 563 free_pages_out:
 564         for (i = 0; i < nr_pages_ret; i++) {
 565                 WARN_ON(pages[i]->mapping);
 566                 page_cache_release(pages[i]);
 567         }
 568         kfree(pages);
 569
 570         goto out;
 571
 572 cleanup_and_out:
 573         extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 574                                      start, end, NULL,
 575                                      EXTENT_CLEAR_UNLOCK_PAGE |
 576                                      EXTENT_CLEAR_DIRTY |
 577                                      EXTENT_CLEAR_DELALLOC |
 578                                      EXTENT_SET_WRITEBACK |
 579                                      EXTENT_END_WRITEBACK);
 580         if (!trans || IS_ERR(trans))
 581                 btrfs_error(root->fs_info, ret, "Failed to join transaction");
 582         else
 583                 btrfs_abort_transaction(trans, root, ret);
 584         goto free_pages_out;
 585 }
 586
 587 /*
 588  * phase two of compressed writeback.  This is the ordered portion
 589  * of the code, which only gets called in the order the work was
 590  * queued.  We walk all the async extents created by compress_file_range
 591  * and send them down to the disk.
 592  */
 593 static noinline int submit_compressed_extents(struct inode *inode,
 594                                               struct async_cow *async_cow)
 595 {
 596         struct async_extent *async_extent;
 597         u64 alloc_hint = 0;
 598         struct btrfs_trans_handle *trans;
 599         struct btrfs_key ins;
 600         struct extent_map *em;
 601         struct btrfs_root *root = BTRFS_I(inode)->root;
 602         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 603         struct extent_io_tree *io_tree;
 604         int ret = 0;
 605
 606         if (list_empty(&async_cow->extents))
 607                 return 0;
 608
 609
 610         while (!list_empty(&async_cow->extents)) {
 611                 async_extent = list_entry(async_cow->extents.next,
 612                                           struct async_extent, list);
 613                 list_del(&async_extent->list);
 614
 615                 io_tree = &BTRFS_I(inode)->io_tree;
 616
 617 retry:
 618                 /* did the compression code fall back to uncompressed IO? */
 619                 if (!async_extent->pages) {
 620                         int page_started = 0;
 621                         unsigned long nr_written = 0;
 622
 623                         lock_extent(io_tree, async_extent->start,
 624                                          async_extent->start +
 625                                          async_extent->ram_size - 1);
 626
 627                         /* allocate blocks */
 628                         ret = cow_file_range(inode, async_cow->locked_page,
 629                                              async_extent->start,
 630                                              async_extent->start +
 631                                              async_extent->ram_size - 1,
 632                                              &page_started, &nr_written, 0);
 633
 634                         /* JDM XXX */
 635
 636                         /*
 637                          * if page_started, cow_file_range inserted an
 638                          * inline extent and took care of all the unlocking
 639                          * and IO for us.  Otherwise, we need to submit
 640                          * all those pages down to the drive.
 641                          */
 642                         if (!page_started && !ret)
 643                                 extent_write_locked_range(io_tree,
 644                                                   inode, async_extent->start,
 645                                                   async_extent->start +
 646                                                   async_extent->ram_size - 1,
 647                                                   btrfs_get_extent,
 648                                                   WB_SYNC_ALL);
 649                         kfree(async_extent);
 650                         cond_resched();
 651                         continue;
 652                 }
 653
 654                 lock_extent(io_tree, async_extent->start,
 655                             async_extent->start + async_extent->ram_size - 1);
 656
 657                 trans = btrfs_join_transaction(root);
 658                 if (IS_ERR(trans)) {
 659                         ret = PTR_ERR(trans);
 660                 } else {
 661                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 662                         ret = btrfs_reserve_extent(trans, root,
 663                                            async_extent->compressed_size,
 664                                            async_extent->compressed_size,
 665                                            0, alloc_hint, &ins, 1);
 666                         if (ret)
 667                                 btrfs_abort_transaction(trans, root, ret);
 668                         btrfs_end_transaction(trans, root);
 669                 }
 670
 671                 if (ret) {
 672                         int i;
 673                         for (i = 0; i < async_extent->nr_pages; i++) {
 674                                 WARN_ON(async_extent->pages[i]->mapping);
 675                                 page_cache_release(async_extent->pages[i]);
 676                         }
 677                         kfree(async_extent->pages);
 678                         async_extent->nr_pages = 0;
 679                         async_extent->pages = NULL;
 680                         unlock_extent(io_tree, async_extent->start,
 681                                       async_extent->start +
 682                                       async_extent->ram_size - 1);
 683                         if (ret == -ENOSPC)
 684                                 goto retry;
 685                         goto out_free; /* JDM: Requeue? */
 686                 }
 687
 688                 /*
 689                  * here we're doing allocation and writeback of the
 690                  * compressed pages
 691                  */
 692                 btrfs_drop_extent_cache(inode, async_extent->start,
 693                                         async_extent->start +
 694                                         async_extent->ram_size - 1, 0);
 695
 696                 em = alloc_extent_map();
 697                 BUG_ON(!em); /* -ENOMEM */
 698                 em->start = async_extent->start;
 699                 em->len = async_extent->ram_size;
 700                 em->orig_start = em->start;
 701
 702                 em->block_start = ins.objectid;
 703                 em->block_len = ins.offset;
 704                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 705                 em->compress_type = async_extent->compress_type;
 706                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 707                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 708
 709                 while (1) {
 710                         write_lock(&em_tree->lock);
 711                         ret = add_extent_mapping(em_tree, em);
 712                         write_unlock(&em_tree->lock);
 713                         if (ret != -EEXIST) {
 714                                 free_extent_map(em);
 715                                 break;
 716                         }
 717                         btrfs_drop_extent_cache(inode, async_extent->start,
 718                                                 async_extent->start +
 719                                                 async_extent->ram_size - 1, 0);
 720                 }
 721
 722                 ret = btrfs_add_ordered_extent_compress(inode,
 723                                                 async_extent->start,
 724                                                 ins.objectid,
 725                                                 async_extent->ram_size,
 726                                                 ins.offset,
 727                                                 BTRFS_ORDERED_COMPRESSED,
 728                                                 async_extent->compress_type);
 729                 BUG_ON(ret); /* -ENOMEM */
 730
 731                 /*
 732                  * clear dirty, set writeback and unlock the pages.
 733                  */
 734                 extent_clear_unlock_delalloc(inode,
 735                                 &BTRFS_I(inode)->io_tree,
 736                                 async_extent->start,
 737                                 async_extent->start +
 738                                 async_extent->ram_size - 1,
 739                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 740                                 EXTENT_CLEAR_UNLOCK |
 741                                 EXTENT_CLEAR_DELALLOC |
 742                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
 743
 744                 ret = btrfs_submit_compressed_write(inode,
 745                                     async_extent->start,
 746                                     async_extent->ram_size,
 747                                     ins.objectid,
 748                                     ins.offset, async_extent->pages,
 749                                     async_extent->nr_pages);
 750
 751                 BUG_ON(ret); /* -ENOMEM */
 752                 alloc_hint = ins.objectid + ins.offset;
 753                 kfree(async_extent);
 754                 cond_resched();
 755         }
 756         ret = 0;
 757 out:
 758         return ret;
 759 out_free:
 760         kfree(async_extent);
 761         goto out;
 762 }
 763
 764 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 765                                       u64 num_bytes)
 766 {
 767         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 768         struct extent_map *em;
 769         u64 alloc_hint = 0;
 770
 771         read_lock(&em_tree->lock);
 772         em = search_extent_mapping(em_tree, start, num_bytes);
 773         if (em) {
 774                 /*
 775                  * if block start isn't an actual block number then find the
 776                  * first block in this inode and use that as a hint.  If that
 777                  * block is also bogus then just don't worry about it.
 778                  */
 779                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 780                         free_extent_map(em);
 781                         em = search_extent_mapping(em_tree, 0, 0);
 782                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 783                                 alloc_hint = em->block_start;
 784                         if (em)
 785                                 free_extent_map(em);
 786                 } else {
 787                         alloc_hint = em->block_start;
 788                         free_extent_map(em);
 789                 }
 790         }
 791         read_unlock(&em_tree->lock);
 792
 793         return alloc_hint;
 794 }
 795
 796 /*
 797  * when extent_io.c finds a delayed allocation range in the file,
 798  * the call backs end up in this code.  The basic idea is to
 799  * allocate extents on disk for the range, and create ordered data structs
 800  * in ram to track those extents.
 801  *
 802  * locked_page is the page that writepage had locked already.  We use
 803  * it to make sure we don't do extra locks or unlocks.
 804  *
 805  * *page_started is set to one if we unlock locked_page and do everything
 806  * required to start IO on it.  It may be clean and already done with
 807  * IO when we return.
 808  */
 809 static noinline int cow_file_range(struct inode *inode,
 810                                    struct page *locked_page,
 811                                    u64 start, u64 end, int *page_started,
 812                                    unsigned long *nr_written,
 813                                    int unlock)
 814 {
 815         struct btrfs_root *root = BTRFS_I(inode)->root;
 816         struct btrfs_trans_handle *trans;
 817         u64 alloc_hint = 0;
 818         u64 num_bytes;
 819         unsigned long ram_size;
 820         u64 disk_num_bytes;
 821         u64 cur_alloc_size;
 822         u64 blocksize = root->sectorsize;
 823         struct btrfs_key ins;
 824         struct extent_map *em;
 825         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 826         int ret = 0;
 827
 828         BUG_ON(btrfs_is_free_space_inode(root, inode));
 829         trans = btrfs_join_transaction(root);
 830         if (IS_ERR(trans)) {
 831                 extent_clear_unlock_delalloc(inode,
 832                              &BTRFS_I(inode)->io_tree,
 833                              start, end, locked_page,
 834                              EXTENT_CLEAR_UNLOCK_PAGE |
 835                              EXTENT_CLEAR_UNLOCK |
 836                              EXTENT_CLEAR_DELALLOC |
 837                              EXTENT_CLEAR_DIRTY |
 838                              EXTENT_SET_WRITEBACK |
 839                              EXTENT_END_WRITEBACK);
 840                 return PTR_ERR(trans);
 841         }
 842         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 843
 844         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 845         num_bytes = max(blocksize,  num_bytes);
 846         disk_num_bytes = num_bytes;
 847         ret = 0;
 848
 849         /* if this is a small write inside eof, kick off defrag */
 850         if (num_bytes < 64 * 1024 &&
 851             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 852                 btrfs_add_inode_defrag(trans, inode);
 853
 854         if (start == 0) {
 855                 /* lets try to make an inline extent */
 856                 ret = cow_file_range_inline(trans, root, inode,
 857                                             start, end, 0, 0, NULL);
 858                 if (ret == 0) {
 859                         extent_clear_unlock_delalloc(inode,
 860                                      &BTRFS_I(inode)->io_tree,
 861                                      start, end, NULL,
 862                                      EXTENT_CLEAR_UNLOCK_PAGE |
 863                                      EXTENT_CLEAR_UNLOCK |
 864                                      EXTENT_CLEAR_DELALLOC |
 865                                      EXTENT_CLEAR_DIRTY |
 866                                      EXTENT_SET_WRITEBACK |
 867                                      EXTENT_END_WRITEBACK);
 868
 869                         *nr_written = *nr_written +
 870                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 871                         *page_started = 1;
 872                         goto out;
 873                 } else if (ret < 0) {
 874                         btrfs_abort_transaction(trans, root, ret);
 875                         goto out_unlock;
 876                 }
 877         }
 878
 879         BUG_ON(disk_num_bytes >
 880                btrfs_super_total_bytes(root->fs_info->super_copy));
 881
 882         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 883         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 884
 885         while (disk_num_bytes > 0) {
 886                 unsigned long op;
 887
 888                 cur_alloc_size = disk_num_bytes;
 889                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 890                                            root->sectorsize, 0, alloc_hint,
 891                                            &ins, 1);
 892                 if (ret < 0) {
 893                         btrfs_abort_transaction(trans, root, ret);
 894                         goto out_unlock;
 895                 }
 896
 897                 em = alloc_extent_map();
 898                 BUG_ON(!em); /* -ENOMEM */
 899                 em->start = start;
 900                 em->orig_start = em->start;
 901                 ram_size = ins.offset;
 902                 em->len = ins.offset;
 903
 904                 em->block_start = ins.objectid;
 905                 em->block_len = ins.offset;
 906                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 907                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 908
 909                 while (1) {
 910                         write_lock(&em_tree->lock);
 911                         ret = add_extent_mapping(em_tree, em);
 912                         write_unlock(&em_tree->lock);
 913                         if (ret != -EEXIST) {
 914                                 free_extent_map(em);
 915                                 break;
 916                         }
 917                         btrfs_drop_extent_cache(inode, start,
 918                                                 start + ram_size - 1, 0);
 919                 }
 920
 921                 cur_alloc_size = ins.offset;
 922                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 923                                                ram_size, cur_alloc_size, 0);
 924                 BUG_ON(ret); /* -ENOMEM */
 925
 926                 if (root->root_key.objectid ==
 927                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 928                         ret = btrfs_reloc_clone_csums(inode, start,
 929                                                       cur_alloc_size);
 930                         if (ret) {
 931                                 btrfs_abort_transaction(trans, root, ret);
 932                                 goto out_unlock;
 933                         }
 934                 }
 935
 936                 if (disk_num_bytes < cur_alloc_size)
 937                         break;
 938
 939                 /* we're not doing compressed IO, don't unlock the first
 940                  * page (which the caller expects to stay locked), don't
 941                  * clear any dirty bits and don't set any writeback bits
 942                  *
 943                  * Do set the Private2 bit so we know this page was properly
 944                  * setup for writepage
 945                  */
 946                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
 947                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
 948                         EXTENT_SET_PRIVATE2;
 949
 950                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 951                                              start, start + ram_size - 1,
 952                                              locked_page, op);
 953                 disk_num_bytes -= cur_alloc_size;
 954                 num_bytes -= cur_alloc_size;
 955                 alloc_hint = ins.objectid + ins.offset;
 956                 start += cur_alloc_size;
 957         }
 958         ret = 0;
 959 out:
 960         btrfs_end_transaction(trans, root);
 961
 962         return ret;
 963 out_unlock:
 964         extent_clear_unlock_delalloc(inode,
 965                      &BTRFS_I(inode)->io_tree,
 966                      start, end, locked_page,
 967                      EXTENT_CLEAR_UNLOCK_PAGE |
 968                      EXTENT_CLEAR_UNLOCK |
 969                      EXTENT_CLEAR_DELALLOC |
 970                      EXTENT_CLEAR_DIRTY |
 971                      EXTENT_SET_WRITEBACK |
 972                      EXTENT_END_WRITEBACK);
 973
 974         goto out;
 975 }
 976
 977 /*
 978  * work queue call back to started compression on a file and pages
 979  */
 980 static noinline void async_cow_start(struct btrfs_work *work)
 981 {
 982         struct async_cow *async_cow;
 983         int num_added = 0;
 984         async_cow = container_of(work, struct async_cow, work);
 985
 986         compress_file_range(async_cow->inode, async_cow->locked_page,
 987                             async_cow->start, async_cow->end, async_cow,
 988                             &num_added);
 989         if (num_added == 0) {
 990                 iput(async_cow->inode);
 991                 async_cow->inode = NULL;
 992         }
 993 }
 994
 995 /*
 996  * work queue call back to submit previously compressed pages
 997  */
 998 static noinline void async_cow_submit(struct btrfs_work *work)
 999 {
1000         struct async_cow *async_cow;
1001         struct btrfs_root *root;
1002         unsigned long nr_pages;
1003
1004         async_cow = container_of(work, struct async_cow, work);
1005
1006         root = async_cow->root;
1007         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1008                 PAGE_CACHE_SHIFT;
1009
1010         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
1011
1012         if (atomic_read(&root->fs_info->async_delalloc_pages) <
1013             5 * 1042 * 1024 &&
1014             waitqueue_active(&root->fs_info->async_submit_wait))
1015                 wake_up(&root->fs_info->async_submit_wait);
1016
1017         if (async_cow->inode)
1018                 submit_compressed_extents(async_cow->inode, async_cow);
1019 }
1020
1021 static noinline void async_cow_free(struct btrfs_work *work)
1022 {
1023         struct async_cow *async_cow;
1024         async_cow = container_of(work, struct async_cow, work);
1025         if (async_cow->inode)
1026                 iput(async_cow->inode);
1027         kfree(async_cow);
1028 }
1029
1030 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1031                                 u64 start, u64 end, int *page_started,
1032                                 unsigned long *nr_written)
1033 {
1034         struct async_cow *async_cow;
1035         struct btrfs_root *root = BTRFS_I(inode)->root;
1036         unsigned long nr_pages;
1037         u64 cur_end;
1038         int limit = 10 * 1024 * 1042;
1039
1040         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1041                          1, 0, NULL, GFP_NOFS);
1042         while (start < end) {
1043                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1044                 BUG_ON(!async_cow); /* -ENOMEM */
1045                 async_cow->inode = igrab(inode);
1046                 async_cow->root = root;
1047                 async_cow->locked_page = locked_page;
1048                 async_cow->start = start;
1049
1050                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1051                         cur_end = end;
1052                 else
1053                         cur_end = min(end, start + 512 * 1024 - 1);
1054
1055                 async_cow->end = cur_end;
1056                 INIT_LIST_HEAD(&async_cow->extents);
1057
1058                 async_cow->work.func = async_cow_start;
1059                 async_cow->work.ordered_func = async_cow_submit;
1060                 async_cow->work.ordered_free = async_cow_free;
1061                 async_cow->work.flags = 0;
1062
1063                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1064                         PAGE_CACHE_SHIFT;
1065                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1066
1067                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1068                                    &async_cow->work);
1069
1070                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1071                         wait_event(root->fs_info->async_submit_wait,
1072                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1073                             limit));
1074                 }
1075
1076                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1077                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1078                         wait_event(root->fs_info->async_submit_wait,
1079                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1080                            0));
1081                 }
1082
1083                 *nr_written += nr_pages;
1084                 start = cur_end + 1;
1085         }
1086         *page_started = 1;
1087         return 0;
1088 }
1089
1090 static noinline int csum_exist_in_range(struct btrfs_root *root,
1091                                         u64 bytenr, u64 num_bytes)
1092 {
1093         int ret;
1094         struct btrfs_ordered_sum *sums;
1095         LIST_HEAD(list);
1096
1097         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1098                                        bytenr + num_bytes - 1, &list, 0);
1099         if (ret == 0 && list_empty(&list))
1100                 return 0;
1101
1102         while (!list_empty(&list)) {
1103                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1104                 list_del(&sums->list);
1105                 kfree(sums);
1106         }
1107         return 1;
1108 }
1109
1110 /*
1111  * when nowcow writeback call back.  This checks for snapshots or COW copies
1112  * of the extents that exist in the file, and COWs the file as required.
1113  *
1114  * If no cow copies or snapshots exist, we write directly to the existing
1115  * blocks on disk
1116  */
1117 static noinline int run_delalloc_nocow(struct inode *inode,
1118                                        struct page *locked_page,
1119                               u64 start, u64 end, int *page_started, int force,
1120                               unsigned long *nr_written)
1121 {
1122         struct btrfs_root *root = BTRFS_I(inode)->root;
1123         struct btrfs_trans_handle *trans;
1124         struct extent_buffer *leaf;
1125         struct btrfs_path *path;
1126         struct btrfs_file_extent_item *fi;
1127         struct btrfs_key found_key;
1128         u64 cow_start;
1129         u64 cur_offset;
1130         u64 extent_end;
1131         u64 extent_offset;
1132         u64 disk_bytenr;
1133         u64 num_bytes;
1134         int extent_type;
1135         int ret, err;
1136         int type;
1137         int nocow;
1138         int check_prev = 1;
1139         bool nolock;
1140         u64 ino = btrfs_ino(inode);
1141
1142         path = btrfs_alloc_path();
1143         if (!path) {
1144                 extent_clear_unlock_delalloc(inode,
1145                              &BTRFS_I(inode)->io_tree,
1146                              start, end, locked_page,
1147                              EXTENT_CLEAR_UNLOCK_PAGE |
1148                              EXTENT_CLEAR_UNLOCK |
1149                              EXTENT_CLEAR_DELALLOC |
1150                              EXTENT_CLEAR_DIRTY |
1151                              EXTENT_SET_WRITEBACK |
1152                              EXTENT_END_WRITEBACK);
1153                 return -ENOMEM;
1154         }
1155
1156         nolock = btrfs_is_free_space_inode(root, inode);
1157
1158         if (nolock)
1159                 trans = btrfs_join_transaction_nolock(root);
1160         else
1161                 trans = btrfs_join_transaction(root);
1162
1163         if (IS_ERR(trans)) {
1164                 extent_clear_unlock_delalloc(inode,
1165                              &BTRFS_I(inode)->io_tree,
1166                              start, end, locked_page,
1167                              EXTENT_CLEAR_UNLOCK_PAGE |
1168                              EXTENT_CLEAR_UNLOCK |
1169                              EXTENT_CLEAR_DELALLOC |
1170                              EXTENT_CLEAR_DIRTY |
1171                              EXTENT_SET_WRITEBACK |
1172                              EXTENT_END_WRITEBACK);
1173                 btrfs_free_path(path);
1174                 return PTR_ERR(trans);
1175         }
1176
1177         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1178
1179         cow_start = (u64)-1;
1180         cur_offset = start;
1181         while (1) {
1182                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1183                                                cur_offset, 0);
1184                 if (ret < 0) {
1185                         btrfs_abort_transaction(trans, root, ret);
1186                         goto error;
1187                 }
1188                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1189                         leaf = path->nodes[0];
1190                         btrfs_item_key_to_cpu(leaf, &found_key,
1191                                               path->slots[0] - 1);
1192                         if (found_key.objectid == ino &&
1193                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1194                                 path->slots[0]--;
1195                 }
1196                 check_prev = 0;
1197 next_slot:
1198                 leaf = path->nodes[0];
1199                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1200                         ret = btrfs_next_leaf(root, path);
1201                         if (ret < 0) {
1202                                 btrfs_abort_transaction(trans, root, ret);
1203                                 goto error;
1204                         }
1205                         if (ret > 0)
1206                                 break;
1207                         leaf = path->nodes[0];
1208                 }
1209
1210                 nocow = 0;
1211                 disk_bytenr = 0;
1212                 num_bytes = 0;
1213                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1214
1215                 if (found_key.objectid > ino ||
1216                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1217                     found_key.offset > end)
1218                         break;
1219
1220                 if (found_key.offset > cur_offset) {
1221                         extent_end = found_key.offset;
1222                         extent_type = 0;
1223                         goto out_check;
1224                 }
1225
1226                 fi = btrfs_item_ptr(leaf, path->slots[0],
1227                                     struct btrfs_file_extent_item);
1228                 extent_type = btrfs_file_extent_type(leaf, fi);
1229
1230                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1231                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1232                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1233                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1234                         extent_end = found_key.offset +
1235                                 btrfs_file_extent_num_bytes(leaf, fi);
1236                         if (extent_end <= start) {
1237                                 path->slots[0]++;
1238                                 goto next_slot;
1239                         }
1240                         if (disk_bytenr == 0)
1241                                 goto out_check;
1242                         if (btrfs_file_extent_compression(leaf, fi) ||
1243                             btrfs_file_extent_encryption(leaf, fi) ||
1244                             btrfs_file_extent_other_encoding(leaf, fi))
1245                                 goto out_check;
1246                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1247                                 goto out_check;
1248                         if (btrfs_extent_readonly(root, disk_bytenr))
1249                                 goto out_check;
1250                         if (btrfs_cross_ref_exist(trans, root, ino,
1251                                                   found_key.offset -
1252                                                   extent_offset, disk_bytenr))
1253                                 goto out_check;
1254                         disk_bytenr += extent_offset;
1255                         disk_bytenr += cur_offset - found_key.offset;
1256                         num_bytes = min(end + 1, extent_end) - cur_offset;
1257                         /*
1258                          * force cow if csum exists in the range.
1259                          * this ensure that csum for a given extent are
1260                          * either valid or do not exist.
1261                          */
1262                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1263                                 goto out_check;
1264                         nocow = 1;
1265                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1266                         extent_end = found_key.offset +
1267                                 btrfs_file_extent_inline_len(leaf, fi);
1268                         extent_end = ALIGN(extent_end, root->sectorsize);
1269                 } else {
1270                         BUG_ON(1);
1271                 }
1272 out_check:
1273                 if (extent_end <= start) {
1274                         path->slots[0]++;
1275                         goto next_slot;
1276                 }
1277                 if (!nocow) {
1278                         if (cow_start == (u64)-1)
1279                                 cow_start = cur_offset;
1280                         cur_offset = extent_end;
1281                         if (cur_offset > end)
1282                                 break;
1283                         path->slots[0]++;
1284                         goto next_slot;
1285                 }
1286
1287                 btrfs_release_path(path);
1288                 if (cow_start != (u64)-1) {
1289                         ret = cow_file_range(inode, locked_page, cow_start,
1290                                         found_key.offset - 1, page_started,
1291                                         nr_written, 1);
1292                         if (ret) {
1293                                 btrfs_abort_transaction(trans, root, ret);
1294                                 goto error;
1295                         }
1296                         cow_start = (u64)-1;
1297                 }
1298
1299                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1300                         struct extent_map *em;
1301                         struct extent_map_tree *em_tree;
1302                         em_tree = &BTRFS_I(inode)->extent_tree;
1303                         em = alloc_extent_map();
1304                         BUG_ON(!em); /* -ENOMEM */
1305                         em->start = cur_offset;
1306                         em->orig_start = em->start;
1307                         em->len = num_bytes;
1308                         em->block_len = num_bytes;
1309                         em->block_start = disk_bytenr;
1310                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1311                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1312                         while (1) {
1313                                 write_lock(&em_tree->lock);
1314                                 ret = add_extent_mapping(em_tree, em);
1315                                 write_unlock(&em_tree->lock);
1316                                 if (ret != -EEXIST) {
1317                                         free_extent_map(em);
1318                                         break;
1319                                 }
1320                                 btrfs_drop_extent_cache(inode, em->start,
1321                                                 em->start + em->len - 1, 0);
1322                         }
1323                         type = BTRFS_ORDERED_PREALLOC;
1324                 } else {
1325                         type = BTRFS_ORDERED_NOCOW;
1326                 }
1327
1328                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1329                                                num_bytes, num_bytes, type);
1330                 BUG_ON(ret); /* -ENOMEM */
1331
1332                 if (root->root_key.objectid ==
1333                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1334                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1335                                                       num_bytes);
1336                         if (ret) {
1337                                 btrfs_abort_transaction(trans, root, ret);
1338                                 goto error;
1339                         }
1340                 }
1341
1342                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1343                                 cur_offset, cur_offset + num_bytes - 1,
1344                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1345                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1346                                 EXTENT_SET_PRIVATE2);
1347                 cur_offset = extent_end;
1348                 if (cur_offset > end)
1349                         break;
1350         }
1351         btrfs_release_path(path);
1352
1353         if (cur_offset <= end && cow_start == (u64)-1) {
1354                 cow_start = cur_offset;
1355                 cur_offset = end;
1356         }
1357
1358         if (cow_start != (u64)-1) {
1359                 ret = cow_file_range(inode, locked_page, cow_start, end,
1360                                      page_started, nr_written, 1);
1361                 if (ret) {
1362                         btrfs_abort_transaction(trans, root, ret);
1363                         goto error;
1364                 }
1365         }
1366
1367 error:
1368         if (nolock) {
1369                 err = btrfs_end_transaction_nolock(trans, root);
1370         } else {
1371                 err = btrfs_end_transaction(trans, root);
1372         }
1373         if (!ret)
1374                 ret = err;
1375
1376         if (ret && cur_offset < end)
1377                 extent_clear_unlock_delalloc(inode,
1378                              &BTRFS_I(inode)->io_tree,
1379                              cur_offset, end, locked_page,
1380                              EXTENT_CLEAR_UNLOCK_PAGE |
1381                              EXTENT_CLEAR_UNLOCK |
1382                              EXTENT_CLEAR_DELALLOC |
1383                              EXTENT_CLEAR_DIRTY |
1384                              EXTENT_SET_WRITEBACK |
1385                              EXTENT_END_WRITEBACK);
1386
1387         btrfs_free_path(path);
1388         return ret;
1389 }
1390
1391 /*
1392  * extent_io.c call back to do delayed allocation processing
1393  */
1394 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1395                               u64 start, u64 end, int *page_started,
1396                               unsigned long *nr_written)
1397 {
1398         int ret;
1399         struct btrfs_root *root = BTRFS_I(inode)->root;
1400
1401         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1402                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1403                                          page_started, 1, nr_written);
1404         else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1405                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1406                                          page_started, 0, nr_written);
1407         else if (!btrfs_test_opt(root, COMPRESS) &&
1408                  !(BTRFS_I(inode)->force_compress) &&
1409                  !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1410                 ret = cow_file_range(inode, locked_page, start, end,
1411                                       page_started, nr_written, 1);
1412         else
1413                 ret = cow_file_range_async(inode, locked_page, start, end,
1414                                            page_started, nr_written);
1415         return ret;
1416 }
1417
1418 static void btrfs_split_extent_hook(struct inode *inode,
1419                                     struct extent_state *orig, u64 split)
1420 {
1421         /* not delalloc, ignore it */
1422         if (!(orig->state & EXTENT_DELALLOC))
1423                 return;
1424
1425         spin_lock(&BTRFS_I(inode)->lock);
1426         BTRFS_I(inode)->outstanding_extents++;
1427         spin_unlock(&BTRFS_I(inode)->lock);
1428 }
1429
1430 /*
1431  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1432  * extents so we can keep track of new extents that are just merged onto old
1433  * extents, such as when we are doing sequential writes, so we can properly
1434  * account for the metadata space we'll need.
1435  */
1436 static void btrfs_merge_extent_hook(struct inode *inode,
1437                                     struct extent_state *new,
1438                                     struct extent_state *other)
1439 {
1440         /* not delalloc, ignore it */
1441         if (!(other->state & EXTENT_DELALLOC))
1442                 return;
1443
1444         spin_lock(&BTRFS_I(inode)->lock);
1445         BTRFS_I(inode)->outstanding_extents--;
1446         spin_unlock(&BTRFS_I(inode)->lock);
1447 }
1448
1449 /*
1450  * extent_io.c set_bit_hook, used to track delayed allocation
1451  * bytes in this file, and to maintain the list of inodes that
1452  * have pending delalloc work to be done.
1453  */
1454 static void btrfs_set_bit_hook(struct inode *inode,
1455                                struct extent_state *state, int *bits)
1456 {
1457
1458         /*
1459          * set_bit and clear bit hooks normally require _irqsave/restore
1460          * but in this case, we are only testing for the DELALLOC
1461          * bit, which is only set or cleared with irqs on
1462          */
1463         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1464                 struct btrfs_root *root = BTRFS_I(inode)->root;
1465                 u64 len = state->end + 1 - state->start;
1466                 bool do_list = !btrfs_is_free_space_inode(root, inode);
1467
1468                 if (*bits & EXTENT_FIRST_DELALLOC) {
1469                         *bits &= ~EXTENT_FIRST_DELALLOC;
1470                 } else {
1471                         spin_lock(&BTRFS_I(inode)->lock);
1472                         BTRFS_I(inode)->outstanding_extents++;
1473                         spin_unlock(&BTRFS_I(inode)->lock);
1474                 }
1475
1476                 spin_lock(&root->fs_info->delalloc_lock);
1477                 BTRFS_I(inode)->delalloc_bytes += len;
1478                 root->fs_info->delalloc_bytes += len;
1479                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1480                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1481                                       &root->fs_info->delalloc_inodes);
1482                 }
1483                 spin_unlock(&root->fs_info->delalloc_lock);
1484         }
1485 }
1486
1487 /*
1488  * extent_io.c clear_bit_hook, see set_bit_hook for why
1489  */
1490 static void btrfs_clear_bit_hook(struct inode *inode,
1491                                  struct extent_state *state, int *bits)
1492 {
1493         /*
1494          * set_bit and clear bit hooks normally require _irqsave/restore
1495          * but in this case, we are only testing for the DELALLOC
1496          * bit, which is only set or cleared with irqs on
1497          */
1498         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1499                 struct btrfs_root *root = BTRFS_I(inode)->root;
1500                 u64 len = state->end + 1 - state->start;
1501                 bool do_list = !btrfs_is_free_space_inode(root, inode);
1502
1503                 if (*bits & EXTENT_FIRST_DELALLOC) {
1504                         *bits &= ~EXTENT_FIRST_DELALLOC;
1505                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1506                         spin_lock(&BTRFS_I(inode)->lock);
1507                         BTRFS_I(inode)->outstanding_extents--;
1508                         spin_unlock(&BTRFS_I(inode)->lock);
1509                 }
1510
1511                 if (*bits & EXTENT_DO_ACCOUNTING)
1512                         btrfs_delalloc_release_metadata(inode, len);
1513
1514                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1515                     && do_list)
1516                         btrfs_free_reserved_data_space(inode, len);
1517
1518                 spin_lock(&root->fs_info->delalloc_lock);
1519                 root->fs_info->delalloc_bytes -= len;
1520                 BTRFS_I(inode)->delalloc_bytes -= len;
1521
1522                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1523                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1524                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1525                 }
1526                 spin_unlock(&root->fs_info->delalloc_lock);
1527         }
1528 }
1529
1530 /*
1531  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1532  * we don't create bios that span stripes or chunks
1533  */
1534 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1535                          size_t size, struct bio *bio,
1536                          unsigned long bio_flags)
1537 {
1538         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1539         struct btrfs_mapping_tree *map_tree;
1540         u64 logical = (u64)bio->bi_sector << 9;
1541         u64 length = 0;
1542         u64 map_length;
1543         int ret;
1544
1545         if (bio_flags & EXTENT_BIO_COMPRESSED)
1546                 return 0;
1547
1548         length = bio->bi_size;
1549         map_tree = &root->fs_info->mapping_tree;
1550         map_length = length;
1551         ret = btrfs_map_block(map_tree, READ, logical,
1552                               &map_length, NULL, 0);
1553         /* Will always return 0 or 1 with map_multi == NULL */
1554         BUG_ON(ret < 0);
1555         if (map_length < length + size)
1556                 return 1;
1557         return 0;
1558 }
1559
1560 /*
1561  * in order to insert checksums into the metadata in large chunks,
1562  * we wait until bio submission time.   All the pages in the bio are
1563  * checksummed and sums are attached onto the ordered extent record.
1564  *
1565  * At IO completion time the cums attached on the ordered extent record
1566  * are inserted into the btree
1567  */
1568 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1569                                     struct bio *bio, int mirror_num,
1570                                     unsigned long bio_flags,
1571                                     u64 bio_offset)
1572 {
1573         struct btrfs_root *root = BTRFS_I(inode)->root;
1574         int ret = 0;
1575
1576         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1577         BUG_ON(ret); /* -ENOMEM */
1578         return 0;
1579 }
1580
1581 /*
1582  * in order to insert checksums into the metadata in large chunks,
1583  * we wait until bio submission time.   All the pages in the bio are
1584  * checksummed and sums are attached onto the ordered extent record.
1585  *
1586  * At IO completion time the cums attached on the ordered extent record
1587  * are inserted into the btree
1588  */
1589 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1590                           int mirror_num, unsigned long bio_flags,
1591                           u64 bio_offset)
1592 {
1593         struct btrfs_root *root = BTRFS_I(inode)->root;
1594         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1595 }
1596
1597 /*
1598  * extent_io.c submission hook. This does the right thing for csum calculation
1599  * on write, or reading the csums from the tree before a read
1600  */
1601 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1602                           int mirror_num, unsigned long bio_flags,
1603                           u64 bio_offset)
1604 {
1605         struct btrfs_root *root = BTRFS_I(inode)->root;
1606         int ret = 0;
1607         int skip_sum;
1608         int metadata = 0;
1609
1610         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1611
1612         if (btrfs_is_free_space_inode(root, inode))
1613                 metadata = 2;
1614
1615         if (!(rw & REQ_WRITE)) {
1616                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1617                 if (ret)
1618                         return ret;
1619
1620                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1621                         return btrfs_submit_compressed_read(inode, bio,
1622                                                     mirror_num, bio_flags);
1623                 } else if (!skip_sum) {
1624                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1625                         if (ret)
1626                                 return ret;
1627                 }
1628                 goto mapit;
1629         } else if (!skip_sum) {
1630                 /* csum items have already been cloned */
1631                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1632                         goto mapit;
1633                 /* we're doing a write, do the async checksumming */
1634                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1635                                    inode, rw, bio, mirror_num,
1636                                    bio_flags, bio_offset,
1637                                    __btrfs_submit_bio_start,
1638                                    __btrfs_submit_bio_done);
1639         }
1640
1641 mapit:
1642         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1643 }
1644
1645 /*
1646  * given a list of ordered sums record them in the inode.  This happens
1647  * at IO completion time based on sums calculated at bio submission time.
1648  */
1649 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1650                              struct inode *inode, u64 file_offset,
1651                              struct list_head *list)
1652 {
1653         struct btrfs_ordered_sum *sum;
1654
1655         list_for_each_entry(sum, list, list) {
1656                 btrfs_csum_file_blocks(trans,
1657                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1658         }
1659         return 0;
1660 }
1661
1662 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1663                               struct extent_state **cached_state)
1664 {
1665         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1666                 WARN_ON(1);
1667         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1668                                    cached_state, GFP_NOFS);
1669 }
1670
1671 /* see btrfs_writepage_start_hook for details on why this is required */
1672 struct btrfs_writepage_fixup {
1673         struct page *page;
1674         struct btrfs_work work;
1675 };
1676
1677 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1678 {
1679         struct btrfs_writepage_fixup *fixup;
1680         struct btrfs_ordered_extent *ordered;
1681         struct extent_state *cached_state = NULL;
1682         struct page *page;
1683         struct inode *inode;
1684         u64 page_start;
1685         u64 page_end;
1686         int ret;
1687
1688         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1689         page = fixup->page;
1690 again:
1691         lock_page(page);
1692         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1693                 ClearPageChecked(page);
1694                 goto out_page;
1695         }
1696
1697         inode = page->mapping->host;
1698         page_start = page_offset(page);
1699         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1700
1701         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1702                          &cached_state);
1703
1704         /* already ordered? We're done */
1705         if (PagePrivate2(page))
1706                 goto out;
1707
1708         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1709         if (ordered) {
1710                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1711                                      page_end, &cached_state, GFP_NOFS);
1712                 unlock_page(page);
1713                 btrfs_start_ordered_extent(inode, ordered, 1);
1714                 btrfs_put_ordered_extent(ordered);
1715                 goto again;
1716         }
1717
1718         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1719         if (ret) {
1720                 mapping_set_error(page->mapping, ret);
1721                 end_extent_writepage(page, ret, page_start, page_end);
1722                 ClearPageChecked(page);
1723                 goto out;
1724          }
1725
1726         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1727         ClearPageChecked(page);
1728         set_page_dirty(page);
1729 out:
1730         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1731                              &cached_state, GFP_NOFS);
1732 out_page:
1733         unlock_page(page);
1734         page_cache_release(page);
1735         kfree(fixup);
1736 }
1737
1738 /*
1739  * There are a few paths in the higher layers of the kernel that directly
1740  * set the page dirty bit without asking the filesystem if it is a
1741  * good idea.  This causes problems because we want to make sure COW
1742  * properly happens and the data=ordered rules are followed.
1743  *
1744  * In our case any range that doesn't have the ORDERED bit set
1745  * hasn't been properly setup for IO.  We kick off an async process
1746  * to fix it up.  The async helper will wait for ordered extents, set
1747  * the delalloc bit and make it safe to write the page.
1748  */
1749 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1750 {
1751         struct inode *inode = page->mapping->host;
1752         struct btrfs_writepage_fixup *fixup;
1753         struct btrfs_root *root = BTRFS_I(inode)->root;
1754
1755         /* this page is properly in the ordered list */
1756         if (TestClearPagePrivate2(page))
1757                 return 0;
1758
1759         if (PageChecked(page))
1760                 return -EAGAIN;
1761
1762         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1763         if (!fixup)
1764                 return -EAGAIN;
1765
1766         SetPageChecked(page);
1767         page_cache_get(page);
1768         fixup->work.func = btrfs_writepage_fixup_worker;
1769         fixup->page = page;
1770         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1771         return -EBUSY;
1772 }
1773
1774 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1775                                        struct inode *inode, u64 file_pos,
1776                                        u64 disk_bytenr, u64 disk_num_bytes,
1777                                        u64 num_bytes, u64 ram_bytes,
1778                                        u8 compression, u8 encryption,
1779                                        u16 other_encoding, int extent_type)
1780 {
1781         struct btrfs_root *root = BTRFS_I(inode)->root;
1782         struct btrfs_file_extent_item *fi;
1783         struct btrfs_path *path;
1784         struct extent_buffer *leaf;
1785         struct btrfs_key ins;
1786         u64 hint;
1787         int ret;
1788
1789         path = btrfs_alloc_path();
1790         if (!path)
1791                 return -ENOMEM;
1792
1793         path->leave_spinning = 1;
1794
1795         /*
1796          * we may be replacing one extent in the tree with another.
1797          * The new extent is pinned in the extent map, and we don't want
1798          * to drop it from the cache until it is completely in the btree.
1799          *
1800          * So, tell btrfs_drop_extents to leave this extent in the cache.
1801          * the caller is expected to unpin it and allow it to be merged
1802          * with the others.
1803          */
1804         ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1805                                  &hint, 0);
1806         if (ret)
1807                 goto out;
1808
1809         ins.objectid = btrfs_ino(inode);
1810         ins.offset = file_pos;
1811         ins.type = BTRFS_EXTENT_DATA_KEY;
1812         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1813         if (ret)
1814                 goto out;
1815         leaf = path->nodes[0];
1816         fi = btrfs_item_ptr(leaf, path->slots[0],
1817                             struct btrfs_file_extent_item);
1818         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1819         btrfs_set_file_extent_type(leaf, fi, extent_type);
1820         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1821         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1822         btrfs_set_file_extent_offset(leaf, fi, 0);
1823         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1824         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1825         btrfs_set_file_extent_compression(leaf, fi, compression);
1826         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1827         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1828
1829         btrfs_unlock_up_safe(path, 1);
1830         btrfs_set_lock_blocking(leaf);
1831
1832         btrfs_mark_buffer_dirty(leaf);
1833
1834         inode_add_bytes(inode, num_bytes);
1835
1836         ins.objectid = disk_bytenr;
1837         ins.offset = disk_num_bytes;
1838         ins.type = BTRFS_EXTENT_ITEM_KEY;
1839         ret = btrfs_alloc_reserved_file_extent(trans, root,
1840                                         root->root_key.objectid,
1841                                         btrfs_ino(inode), file_pos, &ins);
1842 out:
1843         btrfs_free_path(path);
1844
1845         return ret;
1846 }
1847
1848 /*
1849  * helper function for btrfs_finish_ordered_io, this
1850  * just reads in some of the csum leaves to prime them into ram
1851  * before we start the transaction.  It limits the amount of btree
1852  * reads required while inside the transaction.
1853  */
1854 /* as ordered data IO finishes, this gets called so we can finish
1855  * an ordered extent if the range of bytes in the file it covers are
1856  * fully written.
1857  */
1858 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
1859 {
1860         struct inode *inode = ordered_extent->inode;
1861         struct btrfs_root *root = BTRFS_I(inode)->root;
1862         struct btrfs_trans_handle *trans = NULL;
1863         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1864         struct extent_state *cached_state = NULL;
1865         int compress_type = 0;
1866         int ret;
1867         bool nolock;
1868
1869         nolock = btrfs_is_free_space_inode(root, inode);
1870
1871         if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
1872                 ret = -EIO;
1873                 goto out;
1874         }
1875
1876         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1877                 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
1878                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1879                 if (!ret) {
1880                         if (nolock)
1881                                 trans = btrfs_join_transaction_nolock(root);
1882                         else
1883                                 trans = btrfs_join_transaction(root);
1884                         if (IS_ERR(trans))
1885                                 return PTR_ERR(trans);
1886                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1887                         ret = btrfs_update_inode_fallback(trans, root, inode);
1888                         if (ret) /* -ENOMEM or corruption */
1889                                 btrfs_abort_transaction(trans, root, ret);
1890                 }
1891                 goto out;
1892         }
1893
1894         lock_extent_bits(io_tree, ordered_extent->file_offset,
1895                          ordered_extent->file_offset + ordered_extent->len - 1,
1896                          0, &cached_state);
1897
1898         if (nolock)
1899                 trans = btrfs_join_transaction_nolock(root);
1900         else
1901                 trans = btrfs_join_transaction(root);
1902         if (IS_ERR(trans)) {
1903                 ret = PTR_ERR(trans);
1904                 trans = NULL;
1905                 goto out_unlock;
1906         }
1907         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1908
1909         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1910                 compress_type = ordered_extent->compress_type;
1911         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1912                 BUG_ON(compress_type);
1913                 ret = btrfs_mark_extent_written(trans, inode,
1914                                                 ordered_extent->file_offset,
1915                                                 ordered_extent->file_offset +
1916                                                 ordered_extent->len);
1917         } else {
1918                 BUG_ON(root == root->fs_info->tree_root);
1919                 ret = insert_reserved_file_extent(trans, inode,
1920                                                 ordered_extent->file_offset,
1921                                                 ordered_extent->start,
1922                                                 ordered_extent->disk_len,
1923                                                 ordered_extent->len,
1924                                                 ordered_extent->len,
1925                                                 compress_type, 0, 0,
1926                                                 BTRFS_FILE_EXTENT_REG);
1927                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1928                                    ordered_extent->file_offset,
1929                                    ordered_extent->len);
1930         }
1931
1932         if (ret < 0) {
1933                 btrfs_abort_transaction(trans, root, ret);
1934                 goto out_unlock;
1935         }
1936
1937         add_pending_csums(trans, inode, ordered_extent->file_offset,
1938                           &ordered_extent->list);
1939
1940         ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1941         if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1942                 ret = btrfs_update_inode_fallback(trans, root, inode);
1943                 if (ret) { /* -ENOMEM or corruption */
1944                         btrfs_abort_transaction(trans, root, ret);
1945                         goto out_unlock;
1946                 }
1947         }
1948         ret = 0;
1949 out_unlock:
1950         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1951                              ordered_extent->file_offset +
1952                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1953 out:
1954         if (root != root->fs_info->tree_root)
1955                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1956         if (trans) {
1957                 if (nolock)
1958                         btrfs_end_transaction_nolock(trans, root);
1959                 else
1960                         btrfs_end_transaction(trans, root);
1961         }
1962
1963         if (ret)
1964                 clear_extent_uptodate(io_tree, ordered_extent->file_offset,
1965                                       ordered_extent->file_offset +
1966                                       ordered_extent->len - 1, NULL, GFP_NOFS);
1967
1968         /*
1969          * This needs to be dont to make sure anybody waiting knows we are done
1970          * upating everything for this ordered extent.
1971          */
1972         btrfs_remove_ordered_extent(inode, ordered_extent);
1973
1974         /* once for us */
1975         btrfs_put_ordered_extent(ordered_extent);
1976         /* once for the tree */
1977         btrfs_put_ordered_extent(ordered_extent);
1978
1979         return ret;
1980 }
1981
1982 static void finish_ordered_fn(struct btrfs_work *work)
1983 {
1984         struct btrfs_ordered_extent *ordered_extent;
1985         ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
1986         btrfs_finish_ordered_io(ordered_extent);
1987 }
1988
1989 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1990                                 struct extent_state *state, int uptodate)
1991 {
1992         struct inode *inode = page->mapping->host;
1993         struct btrfs_root *root = BTRFS_I(inode)->root;
1994         struct btrfs_ordered_extent *ordered_extent = NULL;
1995         struct btrfs_workers *workers;
1996
1997         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1998
1999         ClearPagePrivate2(page);
2000         if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2001                                             end - start + 1, uptodate))
2002                 return 0;
2003
2004         ordered_extent->work.func = finish_ordered_fn;
2005         ordered_extent->work.flags = 0;
2006
2007         if (btrfs_is_free_space_inode(root, inode))
2008                 workers = &root->fs_info->endio_freespace_worker;
2009         else
2010                 workers = &root->fs_info->endio_write_workers;
2011         btrfs_queue_worker(workers, &ordered_extent->work);
2012
2013         return 0;
2014 }
2015
2016 /*
2017  * when reads are done, we need to check csums to verify the data is correct
2018  * if there's a match, we allow the bio to finish.  If not, the code in
2019  * extent_io.c will try to find good copies for us.
2020  */
2021 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
2022                                struct extent_state *state, int mirror)
2023 {
2024         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
2025         struct inode *inode = page->mapping->host;
2026         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2027         char *kaddr;
2028         u64 private = ~(u32)0;
2029         int ret;
2030         struct btrfs_root *root = BTRFS_I(inode)->root;
2031         u32 csum = ~(u32)0;
2032
2033         if (PageChecked(page)) {
2034                 ClearPageChecked(page);
2035                 goto good;
2036         }
2037
2038         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2039                 goto good;
2040
2041         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2042             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2043                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2044                                   GFP_NOFS);
2045                 return 0;
2046         }
2047
2048         if (state && state->start == start) {
2049                 private = state->private;
2050                 ret = 0;
2051         } else {
2052                 ret = get_state_private(io_tree, start, &private);
2053         }
2054         kaddr = kmap_atomic(page);
2055         if (ret)
2056                 goto zeroit;
2057
2058         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
2059         btrfs_csum_final(csum, (char *)&csum);
2060         if (csum != private)
2061                 goto zeroit;
2062
2063         kunmap_atomic(kaddr);
2064 good:
2065         return 0;
2066
2067 zeroit:
2068         printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
2069                        "private %llu\n",
2070                        (unsigned long long)btrfs_ino(page->mapping->host),
2071                        (unsigned long long)start, csum,
2072                        (unsigned long long)private);
2073         memset(kaddr + offset, 1, end - start + 1);
2074         flush_dcache_page(page);
2075         kunmap_atomic(kaddr);
2076         if (private == 0)
2077                 return 0;
2078         return -EIO;
2079 }
2080
2081 struct delayed_iput {
2082         struct list_head list;
2083         struct inode *inode;
2084 };
2085
2086 /* JDM: If this is fs-wide, why can't we add a pointer to
2087  * btrfs_inode instead and avoid the allocation? */
2088 void btrfs_add_delayed_iput(struct inode *inode)
2089 {
2090         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2091         struct delayed_iput *delayed;
2092
2093         if (atomic_add_unless(&inode->i_count, -1, 1))
2094                 return;
2095
2096         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2097         delayed->inode = inode;
2098
2099         spin_lock(&fs_info->delayed_iput_lock);
2100         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2101         spin_unlock(&fs_info->delayed_iput_lock);
2102 }
2103
2104 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2105 {
2106         LIST_HEAD(list);
2107         struct btrfs_fs_info *fs_info = root->fs_info;
2108         struct delayed_iput *delayed;
2109         int empty;
2110
2111         spin_lock(&fs_info->delayed_iput_lock);
2112         empty = list_empty(&fs_info->delayed_iputs);
2113         spin_unlock(&fs_info->delayed_iput_lock);
2114         if (empty)
2115                 return;
2116
2117         down_read(&root->fs_info->cleanup_work_sem);
2118         spin_lock(&fs_info->delayed_iput_lock);
2119         list_splice_init(&fs_info->delayed_iputs, &list);
2120         spin_unlock(&fs_info->delayed_iput_lock);
2121
2122         while (!list_empty(&list)) {
2123                 delayed = list_entry(list.next, struct delayed_iput, list);
2124                 list_del(&delayed->list);
2125                 iput(delayed->inode);
2126                 kfree(delayed);
2127         }
2128         up_read(&root->fs_info->cleanup_work_sem);
2129 }
2130
2131 enum btrfs_orphan_cleanup_state {
2132         ORPHAN_CLEANUP_STARTED  = 1,
2133         ORPHAN_CLEANUP_DONE     = 2,
2134 };
2135
2136 /*
2137  * This is called in transaction commit time. If there are no orphan
2138  * files in the subvolume, it removes orphan item and frees block_rsv
2139  * structure.
2140  */
2141 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2142                               struct btrfs_root *root)
2143 {
2144         struct btrfs_block_rsv *block_rsv;
2145         int ret;
2146
2147         if (atomic_read(&root->orphan_inodes) ||
2148             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2149                 return;
2150
2151         spin_lock(&root->orphan_lock);
2152         if (atomic_read(&root->orphan_inodes)) {
2153                 spin_unlock(&root->orphan_lock);
2154                 return;
2155         }
2156
2157         if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2158                 spin_unlock(&root->orphan_lock);
2159                 return;
2160         }
2161
2162         block_rsv = root->orphan_block_rsv;
2163         root->orphan_block_rsv = NULL;
2164         spin_unlock(&root->orphan_lock);
2165
2166         if (root->orphan_item_inserted &&
2167             btrfs_root_refs(&root->root_item) > 0) {
2168                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2169                                             root->root_key.objectid);
2170                 BUG_ON(ret);
2171                 root->orphan_item_inserted = 0;
2172         }
2173
2174         if (block_rsv) {
2175                 WARN_ON(block_rsv->size > 0);
2176                 btrfs_free_block_rsv(root, block_rsv);
2177         }
2178 }
2179
2180 /*
2181  * This creates an orphan entry for the given inode in case something goes
2182  * wrong in the middle of an unlink/truncate.
2183  *
2184  * NOTE: caller of this function should reserve 5 units of metadata for
2185  *       this function.
2186  */
2187 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2188 {
2189         struct btrfs_root *root = BTRFS_I(inode)->root;
2190         struct btrfs_block_rsv *block_rsv = NULL;
2191         int reserve = 0;
2192         int insert = 0;
2193         int ret;
2194
2195         if (!root->orphan_block_rsv) {
2196                 block_rsv = btrfs_alloc_block_rsv(root);
2197                 if (!block_rsv)
2198                         return -ENOMEM;
2199         }
2200
2201         spin_lock(&root->orphan_lock);
2202         if (!root->orphan_block_rsv) {
2203                 root->orphan_block_rsv = block_rsv;
2204         } else if (block_rsv) {
2205                 btrfs_free_block_rsv(root, block_rsv);
2206                 block_rsv = NULL;
2207         }
2208
2209         if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2210                               &BTRFS_I(inode)->runtime_flags)) {
2211 #if 0
2212                 /*
2213                  * For proper ENOSPC handling, we should do orphan
2214                  * cleanup when mounting. But this introduces backward
2215                  * compatibility issue.
2216                  */
2217                 if (!xchg(&root->orphan_item_inserted, 1))
2218                         insert = 2;
2219                 else
2220                         insert = 1;
2221 #endif
2222                 insert = 1;
2223                 atomic_dec(&root->orphan_inodes);
2224         }
2225
2226         if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2227                               &BTRFS_I(inode)->runtime_flags))
2228                 reserve = 1;
2229         spin_unlock(&root->orphan_lock);
2230
2231         /* grab metadata reservation from transaction handle */
2232         if (reserve) {
2233                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2234                 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
2235         }
2236
2237         /* insert an orphan item to track this unlinked/truncated file */
2238         if (insert >= 1) {
2239                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2240                 if (ret && ret != -EEXIST) {
2241                         clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2242                                   &BTRFS_I(inode)->runtime_flags);
2243                         btrfs_abort_transaction(trans, root, ret);
2244                         return ret;
2245                 }
2246                 ret = 0;
2247         }
2248
2249         /* insert an orphan item to track subvolume contains orphan files */
2250         if (insert >= 2) {
2251                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2252                                                root->root_key.objectid);
2253                 if (ret && ret != -EEXIST) {
2254                         btrfs_abort_transaction(trans, root, ret);
2255                         return ret;
2256                 }
2257         }
2258         return 0;
2259 }
2260
2261 /*
2262  * We have done the truncate/delete so we can go ahead and remove the orphan
2263  * item for this particular inode.
2264  */
2265 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2266 {
2267         struct btrfs_root *root = BTRFS_I(inode)->root;
2268         int delete_item = 0;
2269         int release_rsv = 0;
2270         int ret = 0;
2271
2272         spin_lock(&root->orphan_lock);
2273         if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2274                                &BTRFS_I(inode)->runtime_flags))
2275                 delete_item = 1;
2276
2277         if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2278                                &BTRFS_I(inode)->runtime_flags))
2279                 release_rsv = 1;
2280         spin_unlock(&root->orphan_lock);
2281
2282         if (trans && delete_item) {
2283                 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2284                 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2285         }
2286
2287         if (release_rsv) {
2288                 btrfs_orphan_release_metadata(inode);
2289                 atomic_dec(&root->orphan_inodes);
2290         }
2291
2292         return 0;
2293 }
2294
2295 /*
2296  * this cleans up any orphans that may be left on the list from the last use
2297  * of this root.
2298  */
2299 int btrfs_orphan_cleanup(struct btrfs_root *root)
2300 {
2301         struct btrfs_path *path;
2302         struct extent_buffer *leaf;
2303         struct btrfs_key key, found_key;
2304         struct btrfs_trans_handle *trans;
2305         struct inode *inode;
2306         u64 last_objectid = 0;
2307         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2308
2309         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2310                 return 0;
2311
2312         path = btrfs_alloc_path();
2313         if (!path) {
2314                 ret = -ENOMEM;
2315                 goto out;
2316         }
2317         path->reada = -1;
2318
2319         key.objectid = BTRFS_ORPHAN_OBJECTID;
2320         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2321         key.offset = (u64)-1;
2322
2323         while (1) {
2324                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2325                 if (ret < 0)
2326                         goto out;
2327
2328                 /*
2329                  * if ret == 0 means we found what we were searching for, which
2330                  * is weird, but possible, so only screw with path if we didn't
2331                  * find the key and see if we have stuff that matches
2332                  */
2333                 if (ret > 0) {
2334                         ret = 0;
2335                         if (path->slots[0] == 0)
2336                                 break;
2337                         path->slots[0]--;
2338                 }
2339
2340                 /* pull out the item */
2341                 leaf = path->nodes[0];
2342                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2343
2344                 /* make sure the item matches what we want */
2345                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2346                         break;
2347                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2348                         break;
2349
2350                 /* release the path since we're done with it */
2351                 btrfs_release_path(path);
2352
2353                 /*
2354                  * this is where we are basically btrfs_lookup, without the
2355                  * crossing root thing.  we store the inode number in the
2356                  * offset of the orphan item.
2357                  */
2358
2359                 if (found_key.offset == last_objectid) {
2360                         printk(KERN_ERR "btrfs: Error removing orphan entry, "
2361                                "stopping orphan cleanup\n");
2362                         ret = -EINVAL;
2363                         goto out;
2364                 }
2365
2366                 last_objectid = found_key.offset;
2367
2368                 found_key.objectid = found_key.offset;
2369                 found_key.type = BTRFS_INODE_ITEM_KEY;
2370                 found_key.offset = 0;
2371                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2372                 ret = PTR_RET(inode);
2373                 if (ret && ret != -ESTALE)
2374                         goto out;
2375
2376                 if (ret == -ESTALE && root == root->fs_info->tree_root) {
2377                         struct btrfs_root *dead_root;
2378                         struct btrfs_fs_info *fs_info = root->fs_info;
2379                         int is_dead_root = 0;
2380
2381                         /*
2382                          * this is an orphan in the tree root. Currently these
2383                          * could come from 2 sources:
2384                          *  a) a snapshot deletion in progress
2385                          *  b) a free space cache inode
2386                          * We need to distinguish those two, as the snapshot
2387                          * orphan must not get deleted.
2388                          * find_dead_roots already ran before us, so if this
2389                          * is a snapshot deletion, we should find the root
2390                          * in the dead_roots list
2391                          */
2392                         spin_lock(&fs_info->trans_lock);
2393                         list_for_each_entry(dead_root, &fs_info->dead_roots,
2394                                             root_list) {
2395                                 if (dead_root->root_key.objectid ==
2396                                     found_key.objectid) {
2397                                         is_dead_root = 1;
2398                                         break;
2399                                 }
2400                         }
2401                         spin_unlock(&fs_info->trans_lock);
2402                         if (is_dead_root) {
2403                                 /* prevent this orphan from being found again */
2404                                 key.offset = found_key.objectid - 1;
2405                                 continue;
2406                         }
2407                 }
2408                 /*
2409                  * Inode is already gone but the orphan item is still there,
2410                  * kill the orphan item.
2411                  */
2412                 if (ret == -ESTALE) {
2413                         trans = btrfs_start_transaction(root, 1);
2414                         if (IS_ERR(trans)) {
2415                                 ret = PTR_ERR(trans);
2416                                 goto out;
2417                         }
2418                         printk(KERN_ERR "auto deleting %Lu\n",
2419                                found_key.objectid);
2420                         ret = btrfs_del_orphan_item(trans, root,
2421                                                     found_key.objectid);
2422                         BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2423                         btrfs_end_transaction(trans, root);
2424                         continue;
2425                 }
2426
2427                 /*
2428                  * add this inode to the orphan list so btrfs_orphan_del does
2429                  * the proper thing when we hit it
2430                  */
2431                 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2432                         &BTRFS_I(inode)->runtime_flags);
2433
2434                 /* if we have links, this was a truncate, lets do that */
2435                 if (inode->i_nlink) {
2436                         if (!S_ISREG(inode->i_mode)) {
2437                                 WARN_ON(1);
2438                                 iput(inode);
2439                                 continue;
2440                         }
2441                         nr_truncate++;
2442                         ret = btrfs_truncate(inode);
2443                 } else {
2444                         nr_unlink++;
2445                 }
2446
2447                 /* this will do delete_inode and everything for us */
2448                 iput(inode);
2449                 if (ret)
2450                         goto out;
2451         }
2452         /* release the path since we're done with it */
2453         btrfs_release_path(path);
2454
2455         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2456
2457         if (root->orphan_block_rsv)
2458                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2459                                         (u64)-1);
2460
2461         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2462                 trans = btrfs_join_transaction(root);
2463                 if (!IS_ERR(trans))
2464                         btrfs_end_transaction(trans, root);
2465         }
2466
2467         if (nr_unlink)
2468                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2469         if (nr_truncate)
2470                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2471
2472 out:
2473         if (ret)
2474                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2475         btrfs_free_path(path);
2476         return ret;
2477 }
2478
2479 /*
2480  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2481  * don't find any xattrs, we know there can't be any acls.
2482  *
2483  * slot is the slot the inode is in, objectid is the objectid of the inode
2484  */
2485 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2486                                           int slot, u64 objectid)
2487 {
2488         u32 nritems = btrfs_header_nritems(leaf);
2489         struct btrfs_key found_key;
2490         int scanned = 0;
2491
2492         slot++;
2493         while (slot < nritems) {
2494                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2495
2496                 /* we found a different objectid, there must not be acls */
2497                 if (found_key.objectid != objectid)
2498                         return 0;
2499
2500                 /* we found an xattr, assume we've got an acl */
2501                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2502                         return 1;
2503
2504                 /*
2505                  * we found a key greater than an xattr key, there can't
2506                  * be any acls later on
2507                  */
2508                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2509                         return 0;
2510
2511                 slot++;
2512                 scanned++;
2513
2514                 /*
2515                  * it goes inode, inode backrefs, xattrs, extents,
2516                  * so if there are a ton of hard links to an inode there can
2517                  * be a lot of backrefs.  Don't waste time searching too hard,
2518                  * this is just an optimization
2519                  */
2520                 if (scanned >= 8)
2521                         break;
2522         }
2523         /* we hit the end of the leaf before we found an xattr or
2524          * something larger than an xattr.  We have to assume the inode
2525          * has acls
2526          */
2527         return 1;
2528 }
2529
2530 /*
2531  * read an inode from the btree into the in-memory inode
2532  */
2533 static void btrfs_read_locked_inode(struct inode *inode)
2534 {
2535         struct btrfs_path *path;
2536         struct extent_buffer *leaf;
2537         struct btrfs_inode_item *inode_item;
2538         struct btrfs_timespec *tspec;
2539         struct btrfs_root *root = BTRFS_I(inode)->root;
2540         struct btrfs_key location;
2541         int maybe_acls;
2542         u32 rdev;
2543         int ret;
2544         bool filled = false;
2545
2546         ret = btrfs_fill_inode(inode, &rdev);
2547         if (!ret)
2548                 filled = true;
2549
2550         path = btrfs_alloc_path();
2551         if (!path)
2552                 goto make_bad;
2553
2554         path->leave_spinning = 1;
2555         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2556
2557         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2558         if (ret)
2559                 goto make_bad;
2560
2561         leaf = path->nodes[0];
2562
2563         if (filled)
2564                 goto cache_acl;
2565
2566         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2567                                     struct btrfs_inode_item);
2568         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2569         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2570         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2571         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2572         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2573
2574         tspec = btrfs_inode_atime(inode_item);
2575         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2576         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2577
2578         tspec = btrfs_inode_mtime(inode_item);
2579         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2580         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2581
2582         tspec = btrfs_inode_ctime(inode_item);
2583         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2584         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2585
2586         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2587         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2588         inode->i_version = btrfs_inode_sequence(leaf, inode_item);
2589         inode->i_generation = BTRFS_I(inode)->generation;
2590         inode->i_rdev = 0;
2591         rdev = btrfs_inode_rdev(leaf, inode_item);
2592
2593         BTRFS_I(inode)->index_cnt = (u64)-1;
2594         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2595 cache_acl:
2596         /*
2597          * try to precache a NULL acl entry for files that don't have
2598          * any xattrs or acls
2599          */
2600         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2601                                            btrfs_ino(inode));
2602         if (!maybe_acls)
2603                 cache_no_acl(inode);
2604
2605         btrfs_free_path(path);
2606
2607         switch (inode->i_mode & S_IFMT) {
2608         case S_IFREG:
2609                 inode->i_mapping->a_ops = &btrfs_aops;
2610                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2611                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2612                 inode->i_fop = &btrfs_file_operations;
2613                 inode->i_op = &btrfs_file_inode_operations;
2614                 break;
2615         case S_IFDIR:
2616                 inode->i_fop = &btrfs_dir_file_operations;
2617                 if (root == root->fs_info->tree_root)
2618                         inode->i_op = &btrfs_dir_ro_inode_operations;
2619                 else
2620                         inode->i_op = &btrfs_dir_inode_operations;
2621                 break;
2622         case S_IFLNK:
2623                 inode->i_op = &btrfs_symlink_inode_operations;
2624                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2625                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2626                 break;
2627         default:
2628                 inode->i_op = &btrfs_special_inode_operations;
2629                 init_special_inode(inode, inode->i_mode, rdev);
2630                 break;
2631         }
2632
2633         btrfs_update_iflags(inode);
2634         return;
2635
2636 make_bad:
2637         btrfs_free_path(path);
2638         make_bad_inode(inode);
2639 }
2640
2641 /*
2642  * given a leaf and an inode, copy the inode fields into the leaf
2643  */
2644 static void fill_inode_item(struct btrfs_trans_handle *trans,
2645                             struct extent_buffer *leaf,
2646                             struct btrfs_inode_item *item,
2647                             struct inode *inode)
2648 {
2649         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2650         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2651         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2652         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2653         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2654
2655         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2656                                inode->i_atime.tv_sec);
2657         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2658                                 inode->i_atime.tv_nsec);
2659
2660         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2661                                inode->i_mtime.tv_sec);
2662         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2663                                 inode->i_mtime.tv_nsec);
2664
2665         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2666                                inode->i_ctime.tv_sec);
2667         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2668                                 inode->i_ctime.tv_nsec);
2669
2670         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2671         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2672         btrfs_set_inode_sequence(leaf, item, inode->i_version);
2673         btrfs_set_inode_transid(leaf, item, trans->transid);
2674         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2675         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2676         btrfs_set_inode_block_group(leaf, item, 0);
2677 }
2678
2679 /*
2680  * copy everything in the in-memory inode into the btree.
2681  */
2682 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
2683                                 struct btrfs_root *root, struct inode *inode)
2684 {
2685         struct btrfs_inode_item *inode_item;
2686         struct btrfs_path *path;
2687         struct extent_buffer *leaf;
2688         int ret;
2689
2690         path = btrfs_alloc_path();
2691         if (!path)
2692                 return -ENOMEM;
2693
2694         path->leave_spinning = 1;
2695         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2696                                  1);
2697         if (ret) {
2698                 if (ret > 0)
2699                         ret = -ENOENT;
2700                 goto failed;
2701         }
2702
2703         btrfs_unlock_up_safe(path, 1);
2704         leaf = path->nodes[0];
2705         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2706                                     struct btrfs_inode_item);
2707
2708         fill_inode_item(trans, leaf, inode_item, inode);
2709         btrfs_mark_buffer_dirty(leaf);
2710         btrfs_set_inode_last_trans(trans, inode);
2711         ret = 0;
2712 failed:
2713         btrfs_free_path(path);
2714         return ret;
2715 }
2716
2717 /*
2718  * copy everything in the in-memory inode into the btree.
2719  */
2720 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2721                                 struct btrfs_root *root, struct inode *inode)
2722 {
2723         int ret;
2724
2725         /*
2726          * If the inode is a free space inode, we can deadlock during commit
2727          * if we put it into the delayed code.
2728          *
2729          * The data relocation inode should also be directly updated
2730          * without delay
2731          */
2732         if (!btrfs_is_free_space_inode(root, inode)
2733             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2734                 ret = btrfs_delayed_update_inode(trans, root, inode);
2735                 if (!ret)
2736                         btrfs_set_inode_last_trans(trans, inode);
2737                 return ret;
2738         }
2739
2740         return btrfs_update_inode_item(trans, root, inode);
2741 }
2742
2743 static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2744                                 struct btrfs_root *root, struct inode *inode)
2745 {
2746         int ret;
2747
2748         ret = btrfs_update_inode(trans, root, inode);
2749         if (ret == -ENOSPC)
2750                 return btrfs_update_inode_item(trans, root, inode);
2751         return ret;
2752 }
2753
2754 /*
2755  * unlink helper that gets used here in inode.c and in the tree logging
2756  * recovery code.  It remove a link in a directory with a given name, and
2757  * also drops the back refs in the inode to the directory
2758  */
2759 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2760                                 struct btrfs_root *root,
2761                                 struct inode *dir, struct inode *inode,
2762                                 const char *name, int name_len)
2763 {
2764         struct btrfs_path *path;
2765         int ret = 0;
2766         struct extent_buffer *leaf;
2767         struct btrfs_dir_item *di;
2768         struct btrfs_key key;
2769         u64 index;
2770         u64 ino = btrfs_ino(inode);
2771         u64 dir_ino = btrfs_ino(dir);
2772
2773         path = btrfs_alloc_path();
2774         if (!path) {
2775                 ret = -ENOMEM;
2776                 goto out;
2777         }
2778
2779         path->leave_spinning = 1;
2780         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2781                                     name, name_len, -1);
2782         if (IS_ERR(di)) {
2783                 ret = PTR_ERR(di);
2784                 goto err;
2785         }
2786         if (!di) {
2787                 ret = -ENOENT;
2788                 goto err;
2789         }
2790         leaf = path->nodes[0];
2791         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2792         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2793         if (ret)
2794                 goto err;
2795         btrfs_release_path(path);
2796
2797         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2798                                   dir_ino, &index);
2799         if (ret) {
2800                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2801                        "inode %llu parent %llu\n", name_len, name,
2802                        (unsigned long long)ino, (unsigned long long)dir_ino);
2803                 btrfs_abort_transaction(trans, root, ret);
2804                 goto err;
2805         }
2806
2807         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2808         if (ret) {
2809                 btrfs_abort_transaction(trans, root, ret);
2810                 goto err;
2811         }
2812
2813         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2814                                          inode, dir_ino);
2815         if (ret != 0 && ret != -ENOENT) {
2816                 btrfs_abort_transaction(trans, root, ret);
2817                 goto err;
2818         }
2819
2820         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2821                                            dir, index);
2822         if (ret == -ENOENT)
2823                 ret = 0;
2824 err:
2825         btrfs_free_path(path);
2826         if (ret)
2827                 goto out;
2828
2829         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2830         inode_inc_iversion(inode);
2831         inode_inc_iversion(dir);
2832         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2833         btrfs_update_inode(trans, root, dir);
2834 out:
2835         return ret;
2836 }
2837
2838 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2839                        struct btrfs_root *root,
2840                        struct inode *dir, struct inode *inode,
2841                        const char *name, int name_len)
2842 {
2843         int ret;
2844         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2845         if (!ret) {
2846                 btrfs_drop_nlink(inode);
2847                 ret = btrfs_update_inode(trans, root, inode);
2848         }
2849         return ret;
2850 }
2851
2852
2853 /* helper to check if there is any shared block in the path */
2854 static int check_path_shared(struct btrfs_root *root,
2855                              struct btrfs_path *path)
2856 {
2857         struct extent_buffer *eb;
2858         int level;
2859         u64 refs = 1;
2860
2861         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2862                 int ret;
2863
2864                 if (!path->nodes[level])
2865                         break;
2866                 eb = path->nodes[level];
2867                 if (!btrfs_block_can_be_shared(root, eb))
2868                         continue;
2869                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2870                                                &refs, NULL);
2871                 if (refs > 1)
2872                         return 1;
2873         }
2874         return 0;
2875 }
2876
2877 /*
2878  * helper to start transaction for unlink and rmdir.
2879  *
2880  * unlink and rmdir are special in btrfs, they do not always free space.
2881  * so in enospc case, we should make sure they will free space before
2882  * allowing them to use the global metadata reservation.
2883  */
2884 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2885                                                        struct dentry *dentry)
2886 {
2887         struct btrfs_trans_handle *trans;
2888         struct btrfs_root *root = BTRFS_I(dir)->root;
2889         struct btrfs_path *path;
2890         struct btrfs_inode_ref *ref;
2891         struct btrfs_dir_item *di;
2892         struct inode *inode = dentry->d_inode;
2893         u64 index;
2894         int check_link = 1;
2895         int err = -ENOSPC;
2896         int ret;
2897         u64 ino = btrfs_ino(inode);
2898         u64 dir_ino = btrfs_ino(dir);
2899
2900         /*
2901          * 1 for the possible orphan item
2902          * 1 for the dir item
2903          * 1 for the dir index
2904          * 1 for the inode ref
2905          * 1 for the inode ref in the tree log
2906          * 2 for the dir entries in the log
2907          * 1 for the inode
2908          */
2909         trans = btrfs_start_transaction(root, 8);
2910         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2911                 return trans;
2912
2913         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2914                 return ERR_PTR(-ENOSPC);
2915
2916         /* check if there is someone else holds reference */
2917         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2918                 return ERR_PTR(-ENOSPC);
2919
2920         if (atomic_read(&inode->i_count) > 2)
2921                 return ERR_PTR(-ENOSPC);
2922
2923         if (xchg(&root->fs_info->enospc_unlink, 1))
2924                 return ERR_PTR(-ENOSPC);
2925
2926         path = btrfs_alloc_path();
2927         if (!path) {
2928                 root->fs_info->enospc_unlink = 0;
2929                 return ERR_PTR(-ENOMEM);
2930         }
2931
2932         /* 1 for the orphan item */
2933         trans = btrfs_start_transaction(root, 1);
2934         if (IS_ERR(trans)) {
2935                 btrfs_free_path(path);
2936                 root->fs_info->enospc_unlink = 0;
2937                 return trans;
2938         }
2939
2940         path->skip_locking = 1;
2941         path->search_commit_root = 1;
2942
2943         ret = btrfs_lookup_inode(trans, root, path,
2944                                 &BTRFS_I(dir)->location, 0);
2945         if (ret < 0) {
2946                 err = ret;
2947                 goto out;
2948         }
2949         if (ret == 0) {
2950                 if (check_path_shared(root, path))
2951                         goto out;
2952         } else {
2953                 check_link = 0;
2954         }
2955         btrfs_release_path(path);
2956
2957         ret = btrfs_lookup_inode(trans, root, path,
2958                                 &BTRFS_I(inode)->location, 0);
2959         if (ret < 0) {
2960                 err = ret;
2961                 goto out;
2962         }
2963         if (ret == 0) {
2964                 if (check_path_shared(root, path))
2965                         goto out;
2966         } else {
2967                 check_link = 0;
2968         }
2969         btrfs_release_path(path);
2970
2971         if (ret == 0 && S_ISREG(inode->i_mode)) {
2972                 ret = btrfs_lookup_file_extent(trans, root, path,
2973                                                ino, (u64)-1, 0);
2974                 if (ret < 0) {
2975                         err = ret;
2976                         goto out;
2977                 }
2978                 BUG_ON(ret == 0); /* Corruption */
2979                 if (check_path_shared(root, path))
2980                         goto out;
2981                 btrfs_release_path(path);
2982         }
2983
2984         if (!check_link) {
2985                 err = 0;
2986                 goto out;
2987         }
2988
2989         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2990                                 dentry->d_name.name, dentry->d_name.len, 0);
2991         if (IS_ERR(di)) {
2992                 err = PTR_ERR(di);
2993                 goto out;
2994         }
2995         if (di) {
2996                 if (check_path_shared(root, path))
2997                         goto out;
2998         } else {
2999                 err = 0;
3000                 goto out;
3001         }
3002         btrfs_release_path(path);
3003
3004         ref = btrfs_lookup_inode_ref(trans, root, path,
3005                                 dentry->d_name.name, dentry->d_name.len,
3006                                 ino, dir_ino, 0);
3007         if (IS_ERR(ref)) {
3008                 err = PTR_ERR(ref);
3009                 goto out;
3010         }
3011         BUG_ON(!ref); /* Logic error */
3012         if (check_path_shared(root, path))
3013                 goto out;
3014         index = btrfs_inode_ref_index(path->nodes[0], ref);
3015         btrfs_release_path(path);
3016
3017         /*
3018          * This is a commit root search, if we can lookup inode item and other
3019          * relative items in the commit root, it means the transaction of
3020          * dir/file creation has been committed, and the dir index item that we
3021          * delay to insert has also been inserted into the commit root. So
3022          * we needn't worry about the delayed insertion of the dir index item
3023          * here.
3024          */
3025         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
3026                                 dentry->d_name.name, dentry->d_name.len, 0);
3027         if (IS_ERR(di)) {
3028                 err = PTR_ERR(di);
3029                 goto out;
3030         }
3031         BUG_ON(ret == -ENOENT);
3032         if (check_path_shared(root, path))
3033                 goto out;
3034
3035         err = 0;
3036 out:
3037         btrfs_free_path(path);
3038         /* Migrate the orphan reservation over */
3039         if (!err)
3040                 err = btrfs_block_rsv_migrate(trans->block_rsv,
3041                                 &root->fs_info->global_block_rsv,
3042                                 trans->bytes_reserved);
3043
3044         if (err) {
3045                 btrfs_end_transaction(trans, root);
3046                 root->fs_info->enospc_unlink = 0;
3047                 return ERR_PTR(err);
3048         }
3049
3050         trans->block_rsv = &root->fs_info->global_block_rsv;
3051         return trans;
3052 }
3053
3054 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
3055                                struct btrfs_root *root)
3056 {
3057         if (trans->block_rsv == &root->fs_info->global_block_rsv) {
3058                 btrfs_block_rsv_release(root, trans->block_rsv,
3059                                         trans->bytes_reserved);
3060                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3061                 BUG_ON(!root->fs_info->enospc_unlink);
3062                 root->fs_info->enospc_unlink = 0;
3063         }
3064         btrfs_end_transaction(trans, root);
3065 }
3066
3067 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3068 {
3069         struct btrfs_root *root = BTRFS_I(dir)->root;
3070         struct btrfs_trans_handle *trans;
3071         struct inode *inode = dentry->d_inode;
3072         int ret;
3073         unsigned long nr = 0;
3074
3075         trans = __unlink_start_trans(dir, dentry);
3076         if (IS_ERR(trans))
3077                 return PTR_ERR(trans);
3078
3079         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3080
3081         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3082                                  dentry->d_name.name, dentry->d_name.len);
3083         if (ret)
3084                 goto out;
3085
3086         if (inode->i_nlink == 0) {
3087                 ret = btrfs_orphan_add(trans, inode);
3088                 if (ret)
3089                         goto out;
3090         }
3091
3092 out:
3093         nr = trans->blocks_used;
3094         __unlink_end_trans(trans, root);
3095         btrfs_btree_balance_dirty(root, nr);
3096         return ret;
3097 }
3098
3099 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3100                         struct btrfs_root *root,
3101                         struct inode *dir, u64 objectid,
3102                         const char *name, int name_len)
3103 {
3104         struct btrfs_path *path;
3105         struct extent_buffer *leaf;
3106         struct btrfs_dir_item *di;
3107         struct btrfs_key key;
3108         u64 index;
3109         int ret;
3110         u64 dir_ino = btrfs_ino(dir);
3111
3112         path = btrfs_alloc_path();
3113         if (!path)
3114                 return -ENOMEM;
3115
3116         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3117                                    name, name_len, -1);
3118         if (IS_ERR_OR_NULL(di)) {
3119                 if (!di)
3120                         ret = -ENOENT;
3121                 else
3122                         ret = PTR_ERR(di);
3123                 goto out;
3124         }
3125
3126         leaf = path->nodes[0];
3127         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3128         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3129         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3130         if (ret) {
3131                 btrfs_abort_transaction(trans, root, ret);
3132                 goto out;
3133         }
3134         btrfs_release_path(path);
3135
3136         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3137                                  objectid, root->root_key.objectid,
3138                                  dir_ino, &index, name, name_len);
3139         if (ret < 0) {
3140                 if (ret != -ENOENT) {
3141                         btrfs_abort_transaction(trans, root, ret);
3142                         goto out;
3143                 }
3144                 di = btrfs_search_dir_index_item(root, path, dir_ino,
3145                                                  name, name_len);
3146                 if (IS_ERR_OR_NULL(di)) {
3147                         if (!di)
3148                                 ret = -ENOENT;
3149                         else
3150                                 ret = PTR_ERR(di);
3151                         btrfs_abort_transaction(trans, root, ret);
3152                         goto out;
3153                 }
3154
3155                 leaf = path->nodes[0];
3156                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3157                 btrfs_release_path(path);
3158                 index = key.offset;
3159         }
3160         btrfs_release_path(path);
3161
3162         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3163         if (ret) {
3164                 btrfs_abort_transaction(trans, root, ret);
3165                 goto out;
3166         }
3167
3168         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3169         inode_inc_iversion(dir);
3170         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3171         ret = btrfs_update_inode(trans, root, dir);
3172         if (ret)
3173                 btrfs_abort_transaction(trans, root, ret);
3174 out:
3175         btrfs_free_path(path);
3176         return ret;
3177 }
3178
3179 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3180 {
3181         struct inode *inode = dentry->d_inode;
3182         int err = 0;
3183         struct btrfs_root *root = BTRFS_I(dir)->root;
3184         struct btrfs_trans_handle *trans;
3185         unsigned long nr = 0;
3186
3187         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3188             btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3189                 return -ENOTEMPTY;
3190
3191         trans = __unlink_start_trans(dir, dentry);
3192         if (IS_ERR(trans))
3193                 return PTR_ERR(trans);
3194
3195         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3196                 err = btrfs_unlink_subvol(trans, root, dir,
3197                                           BTRFS_I(inode)->location.objectid,
3198                                           dentry->d_name.name,
3199                                           dentry->d_name.len);
3200                 goto out;
3201         }
3202
3203         err = btrfs_orphan_add(trans, inode);
3204         if (err)
3205                 goto out;
3206
3207         /* now the directory is empty */
3208         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3209                                  dentry->d_name.name, dentry->d_name.len);
3210         if (!err)
3211                 btrfs_i_size_write(inode, 0);
3212 out:
3213         nr = trans->blocks_used;
3214         __unlink_end_trans(trans, root);
3215         btrfs_btree_balance_dirty(root, nr);
3216
3217         return err;
3218 }
3219
3220 /*
3221  * this can truncate away extent items, csum items and directory items.
3222  * It starts at a high offset and removes keys until it can't find
3223  * any higher than new_size
3224  *
3225  * csum items that cross the new i_size are truncated to the new size
3226  * as well.
3227  *
3228  * min_type is the minimum key type to truncate down to.  If set to 0, this
3229  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3230  */
3231 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3232                                struct btrfs_root *root,
3233                                struct inode *inode,
3234                                u64 new_size, u32 min_type)
3235 {
3236         struct btrfs_path *path;
3237         struct extent_buffer *leaf;
3238         struct btrfs_file_extent_item *fi;
3239         struct btrfs_key key;
3240         struct btrfs_key found_key;
3241         u64 extent_start = 0;
3242         u64 extent_num_bytes = 0;
3243         u64 extent_offset = 0;
3244         u64 item_end = 0;
3245         u64 mask = root->sectorsize - 1;
3246         u32 found_type = (u8)-1;
3247         int found_extent;
3248         int del_item;
3249         int pending_del_nr = 0;
3250         int pending_del_slot = 0;
3251         int extent_type = -1;
3252         int ret;
3253         int err = 0;
3254         u64 ino = btrfs_ino(inode);
3255
3256         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3257
3258         path = btrfs_alloc_path();
3259         if (!path)
3260                 return -ENOMEM;
3261         path->reada = -1;
3262
3263         if (root->ref_cows || root == root->fs_info->tree_root)
3264                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3265
3266         /*
3267          * This function is also used to drop the items in the log tree before
3268          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3269          * it is used to drop the loged items. So we shouldn't kill the delayed
3270          * items.
3271          */
3272         if (min_type == 0 && root == BTRFS_I(inode)->root)
3273                 btrfs_kill_delayed_inode_items(inode);
3274
3275         key.objectid = ino;
3276         key.offset = (u64)-1;
3277         key.type = (u8)-1;
3278
3279 search_again:
3280         path->leave_spinning = 1;
3281         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3282         if (ret < 0) {
3283                 err = ret;
3284                 goto out;
3285         }
3286
3287         if (ret > 0) {
3288                 /* there are no items in the tree for us to truncate, we're
3289                  * done
3290                  */
3291                 if (path->slots[0] == 0)
3292                         goto out;
3293                 path->slots[0]--;
3294         }
3295
3296         while (1) {
3297                 fi = NULL;
3298                 leaf = path->nodes[0];
3299                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3300                 found_type = btrfs_key_type(&found_key);
3301
3302                 if (found_key.objectid != ino)
3303                         break;
3304
3305                 if (found_type < min_type)
3306                         break;
3307
3308                 item_end = found_key.offset;
3309                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3310                         fi = btrfs_item_ptr(leaf, path->slots[0],
3311                                             struct btrfs_file_extent_item);
3312                         extent_type = btrfs_file_extent_type(leaf, fi);
3313                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3314                                 item_end +=
3315                                     btrfs_file_extent_num_bytes(leaf, fi);
3316                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3317                                 item_end += btrfs_file_extent_inline_len(leaf,
3318                                                                          fi);
3319                         }
3320                         item_end--;
3321                 }
3322                 if (found_type > min_type) {
3323                         del_item = 1;
3324                 } else {
3325                         if (item_end < new_size)
3326                                 break;
3327                         if (found_key.offset >= new_size)
3328                                 del_item = 1;
3329                         else
3330                                 del_item = 0;
3331                 }
3332                 found_extent = 0;
3333                 /* FIXME, shrink the extent if the ref count is only 1 */
3334                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3335                         goto delete;
3336
3337                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3338                         u64 num_dec;
3339                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3340                         if (!del_item) {
3341                                 u64 orig_num_bytes =
3342                                         btrfs_file_extent_num_bytes(leaf, fi);
3343                                 extent_num_bytes = new_size -
3344                                         found_key.offset + root->sectorsize - 1;
3345                                 extent_num_bytes = extent_num_bytes &
3346                                         ~((u64)root->sectorsize - 1);
3347                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3348                                                          extent_num_bytes);
3349                                 num_dec = (orig_num_bytes -
3350                                            extent_num_bytes);
3351                                 if (root->ref_cows && extent_start != 0)
3352                                         inode_sub_bytes(inode, num_dec);
3353                                 btrfs_mark_buffer_dirty(leaf);
3354                         } else {
3355                                 extent_num_bytes =
3356                                         btrfs_file_extent_disk_num_bytes(leaf,
3357                                                                          fi);
3358                                 extent_offset = found_key.offset -
3359                                         btrfs_file_extent_offset(leaf, fi);
3360
3361                                 /* FIXME blocksize != 4096 */
3362                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3363                                 if (extent_start != 0) {
3364                                         found_extent = 1;
3365                                         if (root->ref_cows)
3366                                                 inode_sub_bytes(inode, num_dec);
3367                                 }
3368                         }
3369                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3370                         /*
3371                          * we can't truncate inline items that have had
3372                          * special encodings
3373                          */
3374                         if (!del_item &&
3375                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3376                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3377                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3378                                 u32 size = new_size - found_key.offset;
3379
3380                                 if (root->ref_cows) {
3381                                         inode_sub_bytes(inode, item_end + 1 -
3382                                                         new_size);
3383                                 }
3384                                 size =
3385                                     btrfs_file_extent_calc_inline_size(size);
3386                                 btrfs_truncate_item(trans, root, path,
3387                                                     size, 1);
3388                         } else if (root->ref_cows) {
3389                                 inode_sub_bytes(inode, item_end + 1 -
3390                                                 found_key.offset);
3391                         }
3392                 }
3393 delete:
3394                 if (del_item) {
3395                         if (!pending_del_nr) {
3396                                 /* no pending yet, add ourselves */
3397                                 pending_del_slot = path->slots[0];
3398                                 pending_del_nr = 1;
3399                         } else if (pending_del_nr &&
3400                                    path->slots[0] + 1 == pending_del_slot) {
3401                                 /* hop on the pending chunk */
3402                                 pending_del_nr++;
3403                                 pending_del_slot = path->slots[0];
3404                         } else {
3405                                 BUG();
3406                         }
3407                 } else {
3408                         break;
3409                 }
3410                 if (found_extent && (root->ref_cows ||
3411                                      root == root->fs_info->tree_root)) {
3412                         btrfs_set_path_blocking(path);
3413                         ret = btrfs_free_extent(trans, root, extent_start,
3414                                                 extent_num_bytes, 0,
3415                                                 btrfs_header_owner(leaf),
3416                                                 ino, extent_offset, 0);
3417                         BUG_ON(ret);
3418                 }
3419
3420                 if (found_type == BTRFS_INODE_ITEM_KEY)
3421                         break;
3422
3423                 if (path->slots[0] == 0 ||
3424                     path->slots[0] != pending_del_slot) {
3425                         if (root->ref_cows &&
3426                             BTRFS_I(inode)->location.objectid !=
3427                                                 BTRFS_FREE_INO_OBJECTID) {
3428                                 err = -EAGAIN;
3429                                 goto out;
3430                         }
3431                         if (pending_del_nr) {
3432                                 ret = btrfs_del_items(trans, root, path,
3433                                                 pending_del_slot,
3434                                                 pending_del_nr);
3435                                 if (ret) {
3436                                         btrfs_abort_transaction(trans,
3437                                                                 root, ret);
3438                                         goto error;
3439                                 }
3440                                 pending_del_nr = 0;
3441                         }
3442                         btrfs_release_path(path);
3443                         goto search_again;
3444                 } else {
3445                         path->slots[0]--;
3446                 }
3447         }
3448 out:
3449         if (pending_del_nr) {
3450                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3451                                       pending_del_nr);
3452                 if (ret)
3453                         btrfs_abort_transaction(trans, root, ret);
3454         }
3455 error:
3456         btrfs_free_path(path);
3457         return err;
3458 }
3459
3460 /*
3461  * taken from block_truncate_page, but does cow as it zeros out
3462  * any bytes left in the last page in the file.
3463  */
3464 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3465 {
3466         struct inode *inode = mapping->host;
3467         struct btrfs_root *root = BTRFS_I(inode)->root;
3468         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3469         struct btrfs_ordered_extent *ordered;
3470         struct extent_state *cached_state = NULL;
3471         char *kaddr;
3472         u32 blocksize = root->sectorsize;
3473         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3474         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3475         struct page *page;
3476         gfp_t mask = btrfs_alloc_write_mask(mapping);
3477         int ret = 0;
3478         u64 page_start;
3479         u64 page_end;
3480
3481         if ((offset & (blocksize - 1)) == 0)
3482                 goto out;
3483         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3484         if (ret)
3485                 goto out;
3486
3487         ret = -ENOMEM;
3488 again:
3489         page = find_or_create_page(mapping, index, mask);
3490         if (!page) {
3491                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3492                 goto out;
3493         }
3494
3495         page_start = page_offset(page);
3496         page_end = page_start + PAGE_CACHE_SIZE - 1;
3497
3498         if (!PageUptodate(page)) {
3499                 ret = btrfs_readpage(NULL, page);
3500                 lock_page(page);
3501                 if (page->mapping != mapping) {
3502                         unlock_page(page);
3503                         page_cache_release(page);
3504                         goto again;
3505                 }
3506                 if (!PageUptodate(page)) {
3507                         ret = -EIO;
3508                         goto out_unlock;
3509                 }
3510         }
3511         wait_on_page_writeback(page);
3512
3513         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
3514         set_page_extent_mapped(page);
3515
3516         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3517         if (ordered) {
3518                 unlock_extent_cached(io_tree, page_start, page_end,
3519                                      &cached_state, GFP_NOFS);
3520                 unlock_page(page);
3521                 page_cache_release(page);
3522                 btrfs_start_ordered_extent(inode, ordered, 1);
3523                 btrfs_put_ordered_extent(ordered);
3524                 goto again;
3525         }
3526
3527         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3528                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3529                           0, 0, &cached_state, GFP_NOFS);
3530
3531         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3532                                         &cached_state);
3533         if (ret) {
3534                 unlock_extent_cached(io_tree, page_start, page_end,
3535                                      &cached_state, GFP_NOFS);
3536                 goto out_unlock;
3537         }
3538
3539         ret = 0;
3540         if (offset != PAGE_CACHE_SIZE) {
3541                 kaddr = kmap(page);
3542                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3543                 flush_dcache_page(page);
3544                 kunmap(page);
3545         }
3546         ClearPageChecked(page);
3547         set_page_dirty(page);
3548         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3549                              GFP_NOFS);
3550
3551 out_unlock:
3552         if (ret)
3553                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3554         unlock_page(page);
3555         page_cache_release(page);
3556 out:
3557         return ret;
3558 }
3559
3560 /*
3561  * This function puts in dummy file extents for the area we're creating a hole
3562  * for.  So if we are truncating this file to a larger size we need to insert
3563  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3564  * the range between oldsize and size
3565  */
3566 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3567 {
3568         struct btrfs_trans_handle *trans;
3569         struct btrfs_root *root = BTRFS_I(inode)->root;
3570         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3571         struct extent_map *em = NULL;
3572         struct extent_state *cached_state = NULL;
3573         u64 mask = root->sectorsize - 1;
3574         u64 hole_start = (oldsize + mask) & ~mask;
3575         u64 block_end = (size + mask) & ~mask;
3576         u64 last_byte;
3577         u64 cur_offset;
3578         u64 hole_size;
3579         int err = 0;
3580
3581         if (size <= hole_start)
3582                 return 0;
3583
3584         while (1) {
3585                 struct btrfs_ordered_extent *ordered;
3586                 btrfs_wait_ordered_range(inode, hole_start,
3587                                          block_end - hole_start);
3588                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3589                                  &cached_state);
3590                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3591                 if (!ordered)
3592                         break;
3593                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3594                                      &cached_state, GFP_NOFS);
3595                 btrfs_put_ordered_extent(ordered);
3596         }
3597
3598         cur_offset = hole_start;
3599         while (1) {
3600                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3601                                 block_end - cur_offset, 0);
3602                 if (IS_ERR(em)) {
3603                         err = PTR_ERR(em);
3604                         break;
3605                 }
3606                 last_byte = min(extent_map_end(em), block_end);
3607                 last_byte = (last_byte + mask) & ~mask;
3608                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3609                         u64 hint_byte = 0;
3610                         hole_size = last_byte - cur_offset;
3611
3612                         trans = btrfs_start_transaction(root, 3);
3613                         if (IS_ERR(trans)) {
3614                                 err = PTR_ERR(trans);
3615                                 break;
3616                         }
3617
3618                         err = btrfs_drop_extents(trans, inode, cur_offset,
3619                                                  cur_offset + hole_size,
3620                                                  &hint_byte, 1);
3621                         if (err) {
3622                                 btrfs_abort_transaction(trans, root, err);
3623                                 btrfs_end_transaction(trans, root);
3624                                 break;
3625                         }
3626
3627                         err = btrfs_insert_file_extent(trans, root,
3628                                         btrfs_ino(inode), cur_offset, 0,
3629                                         0, hole_size, 0, hole_size,
3630                                         0, 0, 0);
3631                         if (err) {
3632                                 btrfs_abort_transaction(trans, root, err);
3633                                 btrfs_end_transaction(trans, root);
3634                                 break;
3635                         }
3636
3637                         btrfs_drop_extent_cache(inode, hole_start,
3638                                         last_byte - 1, 0);
3639
3640                         btrfs_update_inode(trans, root, inode);
3641                         btrfs_end_transaction(trans, root);
3642                 }
3643                 free_extent_map(em);
3644                 em = NULL;
3645                 cur_offset = last_byte;
3646                 if (cur_offset >= block_end)
3647                         break;
3648         }
3649
3650         free_extent_map(em);
3651         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3652                              GFP_NOFS);
3653         return err;
3654 }
3655
3656 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3657 {
3658         struct btrfs_root *root = BTRFS_I(inode)->root;
3659         struct btrfs_trans_handle *trans;
3660         loff_t oldsize = i_size_read(inode);
3661         int ret;
3662
3663         if (newsize == oldsize)
3664                 return 0;
3665
3666         if (newsize > oldsize) {
3667                 truncate_pagecache(inode, oldsize, newsize);
3668                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3669                 if (ret)
3670                         return ret;
3671
3672                 trans = btrfs_start_transaction(root, 1);
3673                 if (IS_ERR(trans))
3674                         return PTR_ERR(trans);
3675
3676                 i_size_write(inode, newsize);
3677                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3678                 ret = btrfs_update_inode(trans, root, inode);
3679                 btrfs_end_transaction(trans, root);
3680         } else {
3681
3682                 /*
3683                  * We're truncating a file that used to have good data down to
3684                  * zero. Make sure it gets into the ordered flush list so that
3685                  * any new writes get down to disk quickly.
3686                  */
3687                 if (newsize == 0)
3688                         set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
3689                                 &BTRFS_I(inode)->runtime_flags);
3690
3691                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3692                 truncate_setsize(inode, newsize);
3693                 ret = btrfs_truncate(inode);
3694         }
3695
3696         return ret;
3697 }
3698
3699 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3700 {
3701         struct inode *inode = dentry->d_inode;
3702         struct btrfs_root *root = BTRFS_I(inode)->root;
3703         int err;
3704
3705         if (btrfs_root_readonly(root))
3706                 return -EROFS;
3707
3708         err = inode_change_ok(inode, attr);
3709         if (err)
3710                 return err;
3711
3712         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3713                 err = btrfs_setsize(inode, attr->ia_size);
3714                 if (err)
3715                         return err;
3716         }
3717
3718         if (attr->ia_valid) {
3719                 setattr_copy(inode, attr);
3720                 inode_inc_iversion(inode);
3721                 err = btrfs_dirty_inode(inode);
3722
3723                 if (!err && attr->ia_valid & ATTR_MODE)
3724                         err = btrfs_acl_chmod(inode);
3725         }
3726
3727         return err;
3728 }
3729
3730 void btrfs_evict_inode(struct inode *inode)
3731 {
3732         struct btrfs_trans_handle *trans;
3733         struct btrfs_root *root = BTRFS_I(inode)->root;
3734         struct btrfs_block_rsv *rsv, *global_rsv;
3735         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
3736         unsigned long nr;
3737         int ret;
3738
3739         trace_btrfs_inode_evict(inode);
3740
3741         truncate_inode_pages(&inode->i_data, 0);
3742         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3743                                btrfs_is_free_space_inode(root, inode)))
3744                 goto no_delete;
3745
3746         if (is_bad_inode(inode)) {
3747                 btrfs_orphan_del(NULL, inode);
3748                 goto no_delete;
3749         }
3750         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3751         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3752
3753         if (root->fs_info->log_root_recovering) {
3754                 BUG_ON(!test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3755                                  &BTRFS_I(inode)->runtime_flags));
3756                 goto no_delete;
3757         }
3758
3759         if (inode->i_nlink > 0) {
3760                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3761                 goto no_delete;
3762         }
3763
3764         rsv = btrfs_alloc_block_rsv(root);
3765         if (!rsv) {
3766                 btrfs_orphan_del(NULL, inode);
3767                 goto no_delete;
3768         }
3769         rsv->size = min_size;
3770         global_rsv = &root->fs_info->global_block_rsv;
3771
3772         btrfs_i_size_write(inode, 0);
3773
3774         /*
3775          * This is a bit simpler than btrfs_truncate since
3776          *
3777          * 1) We've already reserved our space for our orphan item in the
3778          *    unlink.
3779          * 2) We're going to delete the inode item, so we don't need to update
3780          *    it at all.
3781          *
3782          * So we just need to reserve some slack space in case we add bytes when
3783          * doing the truncate.
3784          */
3785         while (1) {
3786                 ret = btrfs_block_rsv_refill_noflush(root, rsv, min_size);
3787
3788                 /*
3789                  * Try and steal from the global reserve since we will
3790                  * likely not use this space anyway, we want to try as
3791                  * hard as possible to get this to work.
3792                  */
3793                 if (ret)
3794                         ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
3795
3796                 if (ret) {
3797                         printk(KERN_WARNING "Could not get space for a "
3798                                "delete, will truncate on mount %d\n", ret);
3799                         btrfs_orphan_del(NULL, inode);
3800                         btrfs_free_block_rsv(root, rsv);
3801                         goto no_delete;
3802                 }
3803
3804                 trans = btrfs_start_transaction(root, 0);
3805                 if (IS_ERR(trans)) {
3806                         btrfs_orphan_del(NULL, inode);
3807                         btrfs_free_block_rsv(root, rsv);
3808                         goto no_delete;
3809                 }
3810
3811                 trans->block_rsv = rsv;
3812
3813                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3814                 if (ret != -EAGAIN)
3815                         break;
3816
3817                 nr = trans->blocks_used;
3818                 btrfs_end_transaction(trans, root);
3819                 trans = NULL;
3820                 btrfs_btree_balance_dirty(root, nr);
3821         }
3822
3823         btrfs_free_block_rsv(root, rsv);
3824
3825         if (ret == 0) {
3826                 trans->block_rsv = root->orphan_block_rsv;
3827                 ret = btrfs_orphan_del(trans, inode);
3828                 BUG_ON(ret);
3829         }
3830
3831         trans->block_rsv = &root->fs_info->trans_block_rsv;
3832         if (!(root == root->fs_info->tree_root ||
3833               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3834                 btrfs_return_ino(root, btrfs_ino(inode));
3835
3836         nr = trans->blocks_used;
3837         btrfs_end_transaction(trans, root);
3838         btrfs_btree_balance_dirty(root, nr);
3839 no_delete:
3840         end_writeback(inode);
3841         return;
3842 }
3843
3844 /*
3845  * this returns the key found in the dir entry in the location pointer.
3846  * If no dir entries were found, location->objectid is 0.
3847  */
3848 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3849                                struct btrfs_key *location)
3850 {
3851         const char *name = dentry->d_name.name;
3852         int namelen = dentry->d_name.len;
3853         struct btrfs_dir_item *di;
3854         struct btrfs_path *path;
3855         struct btrfs_root *root = BTRFS_I(dir)->root;
3856         int ret = 0;
3857
3858         path = btrfs_alloc_path();
3859         if (!path)
3860                 return -ENOMEM;
3861
3862         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3863                                     namelen, 0);
3864         if (IS_ERR(di))
3865                 ret = PTR_ERR(di);
3866
3867         if (IS_ERR_OR_NULL(di))
3868                 goto out_err;
3869
3870         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3871 out:
3872         btrfs_free_path(path);
3873         return ret;
3874 out_err:
3875         location->objectid = 0;
3876         goto out;
3877 }
3878
3879 /*
3880  * when we hit a tree root in a directory, the btrfs part of the inode
3881  * needs to be changed to reflect the root directory of the tree root.  This
3882  * is kind of like crossing a mount point.
3883  */
3884 static int fixup_tree_root_location(struct btrfs_root *root,
3885                                     struct inode *dir,
3886                                     struct dentry *dentry,
3887                                     struct btrfs_key *location,
3888                                     struct btrfs_root **sub_root)
3889 {
3890         struct btrfs_path *path;
3891         struct btrfs_root *new_root;
3892         struct btrfs_root_ref *ref;
3893         struct extent_buffer *leaf;
3894         int ret;
3895         int err = 0;
3896
3897         path = btrfs_alloc_path();
3898         if (!path) {
3899                 err = -ENOMEM;
3900                 goto out;
3901         }
3902
3903         err = -ENOENT;
3904         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3905                                   BTRFS_I(dir)->root->root_key.objectid,
3906                                   location->objectid);
3907         if (ret) {
3908                 if (ret < 0)
3909                         err = ret;
3910                 goto out;
3911         }
3912
3913         leaf = path->nodes[0];
3914         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3915         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3916             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3917                 goto out;
3918
3919         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3920                                    (unsigned long)(ref + 1),
3921                                    dentry->d_name.len);
3922         if (ret)
3923                 goto out;
3924
3925         btrfs_release_path(path);
3926
3927         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3928         if (IS_ERR(new_root)) {
3929                 err = PTR_ERR(new_root);
3930                 goto out;
3931         }
3932
3933         if (btrfs_root_refs(&new_root->root_item) == 0) {
3934                 err = -ENOENT;
3935                 goto out;
3936         }
3937
3938         *sub_root = new_root;
3939         location->objectid = btrfs_root_dirid(&new_root->root_item);
3940         location->type = BTRFS_INODE_ITEM_KEY;
3941         location->offset = 0;
3942         err = 0;
3943 out:
3944         btrfs_free_path(path);
3945         return err;
3946 }
3947
3948 static void inode_tree_add(struct inode *inode)
3949 {
3950         struct btrfs_root *root = BTRFS_I(inode)->root;
3951         struct btrfs_inode *entry;
3952         struct rb_node **p;
3953         struct rb_node *parent;
3954         u64 ino = btrfs_ino(inode);
3955 again:
3956         p = &root->inode_tree.rb_node;
3957         parent = NULL;
3958
3959         if (inode_unhashed(inode))
3960                 return;
3961
3962         spin_lock(&root->inode_lock);
3963         while (*p) {
3964                 parent = *p;
3965                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3966
3967                 if (ino < btrfs_ino(&entry->vfs_inode))
3968                         p = &parent->rb_left;
3969                 else if (ino > btrfs_ino(&entry->vfs_inode))
3970                         p = &parent->rb_right;
3971                 else {
3972                         WARN_ON(!(entry->vfs_inode.i_state &
3973                                   (I_WILL_FREE | I_FREEING)));
3974                         rb_erase(parent, &root->inode_tree);
3975                         RB_CLEAR_NODE(parent);
3976                         spin_unlock(&root->inode_lock);
3977                         goto again;
3978                 }
3979         }
3980         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3981         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3982         spin_unlock(&root->inode_lock);
3983 }
3984
3985 static void inode_tree_del(struct inode *inode)
3986 {
3987         struct btrfs_root *root = BTRFS_I(inode)->root;
3988         int empty = 0;
3989
3990         spin_lock(&root->inode_lock);
3991         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3992                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3993                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3994                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3995         }
3996         spin_unlock(&root->inode_lock);
3997
3998         /*
3999          * Free space cache has inodes in the tree root, but the tree root has a
4000          * root_refs of 0, so this could end up dropping the tree root as a
4001          * snapshot, so we need the extra !root->fs_info->tree_root check to
4002          * make sure we don't drop it.
4003          */
4004         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
4005             root != root->fs_info->tree_root) {
4006                 synchronize_srcu(&root->fs_info->subvol_srcu);
4007                 spin_lock(&root->inode_lock);
4008                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4009                 spin_unlock(&root->inode_lock);
4010                 if (empty)
4011                         btrfs_add_dead_root(root);
4012         }
4013 }
4014
4015 void btrfs_invalidate_inodes(struct btrfs_root *root)
4016 {
4017         struct rb_node *node;
4018         struct rb_node *prev;
4019         struct btrfs_inode *entry;
4020         struct inode *inode;
4021         u64 objectid = 0;
4022
4023         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4024
4025         spin_lock(&root->inode_lock);
4026 again:
4027         node = root->inode_tree.rb_node;
4028         prev = NULL;
4029         while (node) {
4030                 prev = node;
4031                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4032
4033                 if (objectid < btrfs_ino(&entry->vfs_inode))
4034                         node = node->rb_left;
4035                 else if (objectid > btrfs_ino(&entry->vfs_inode))
4036                         node = node->rb_right;
4037                 else
4038                         break;
4039         }
4040         if (!node) {
4041                 while (prev) {
4042                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
4043                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4044                                 node = prev;
4045                                 break;
4046                         }
4047                         prev = rb_next(prev);
4048                 }
4049         }
4050         while (node) {
4051                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4052                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4053                 inode = igrab(&entry->vfs_inode);
4054                 if (inode) {
4055                         spin_unlock(&root->inode_lock);
4056                         if (atomic_read(&inode->i_count) > 1)
4057                                 d_prune_aliases(inode);
4058                         /*
4059                          * btrfs_drop_inode will have it removed from
4060                          * the inode cache when its usage count
4061                          * hits zero.
4062                          */
4063                         iput(inode);
4064                         cond_resched();
4065                         spin_lock(&root->inode_lock);
4066                         goto again;
4067                 }
4068
4069                 if (cond_resched_lock(&root->inode_lock))
4070                         goto again;
4071
4072                 node = rb_next(node);
4073         }
4074         spin_unlock(&root->inode_lock);
4075 }
4076
4077 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4078 {
4079         struct btrfs_iget_args *args = p;
4080         inode->i_ino = args->ino;
4081         BTRFS_I(inode)->root = args->root;
4082         btrfs_set_inode_space_info(args->root, inode);
4083         return 0;
4084 }
4085
4086 static int btrfs_find_actor(struct inode *inode, void *opaque)
4087 {
4088         struct btrfs_iget_args *args = opaque;
4089         return args->ino == btrfs_ino(inode) &&
4090                 args->root == BTRFS_I(inode)->root;
4091 }
4092
4093 static struct inode *btrfs_iget_locked(struct super_block *s,
4094                                        u64 objectid,
4095                                        struct btrfs_root *root)
4096 {
4097         struct inode *inode;
4098         struct btrfs_iget_args args;
4099         args.ino = objectid;
4100         args.root = root;
4101
4102         inode = iget5_locked(s, objectid, btrfs_find_actor,
4103                              btrfs_init_locked_inode,
4104                              (void *)&args);
4105         return inode;
4106 }
4107
4108 /* Get an inode object given its location and corresponding root.
4109  * Returns in *is_new if the inode was read from disk
4110  */
4111 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4112                          struct btrfs_root *root, int *new)
4113 {
4114         struct inode *inode;
4115
4116         inode = btrfs_iget_locked(s, location->objectid, root);
4117         if (!inode)
4118                 return ERR_PTR(-ENOMEM);
4119
4120         if (inode->i_state & I_NEW) {
4121                 BTRFS_I(inode)->root = root;
4122                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4123                 btrfs_read_locked_inode(inode);
4124                 if (!is_bad_inode(inode)) {
4125                         inode_tree_add(inode);
4126                         unlock_new_inode(inode);
4127                         if (new)
4128                                 *new = 1;
4129                 } else {
4130                         unlock_new_inode(inode);
4131                         iput(inode);
4132                         inode = ERR_PTR(-ESTALE);
4133                 }
4134         }
4135
4136         return inode;
4137 }
4138
4139 static struct inode *new_simple_dir(struct super_block *s,
4140                                     struct btrfs_key *key,
4141                                     struct btrfs_root *root)
4142 {
4143         struct inode *inode = new_inode(s);
4144
4145         if (!inode)
4146                 return ERR_PTR(-ENOMEM);
4147
4148         BTRFS_I(inode)->root = root;
4149         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4150         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4151
4152         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4153         inode->i_op = &btrfs_dir_ro_inode_operations;
4154         inode->i_fop = &simple_dir_operations;
4155         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4156         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4157
4158         return inode;
4159 }
4160
4161 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4162 {
4163         struct inode *inode;
4164         struct btrfs_root *root = BTRFS_I(dir)->root;
4165         struct btrfs_root *sub_root = root;
4166         struct btrfs_key location;
4167         int index;
4168         int ret = 0;
4169
4170         if (dentry->d_name.len > BTRFS_NAME_LEN)
4171                 return ERR_PTR(-ENAMETOOLONG);
4172
4173         if (unlikely(d_need_lookup(dentry))) {
4174                 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
4175                 kfree(dentry->d_fsdata);
4176                 dentry->d_fsdata = NULL;
4177                 /* This thing is hashed, drop it for now */
4178                 d_drop(dentry);
4179         } else {
4180                 ret = btrfs_inode_by_name(dir, dentry, &location);
4181         }
4182
4183         if (ret < 0)
4184                 return ERR_PTR(ret);
4185
4186         if (location.objectid == 0)
4187                 return NULL;
4188
4189         if (location.type == BTRFS_INODE_ITEM_KEY) {
4190                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4191                 return inode;
4192         }
4193
4194         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4195
4196         index = srcu_read_lock(&root->fs_info->subvol_srcu);
4197         ret = fixup_tree_root_location(root, dir, dentry,
4198                                        &location, &sub_root);
4199         if (ret < 0) {
4200                 if (ret != -ENOENT)
4201                         inode = ERR_PTR(ret);
4202                 else
4203                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
4204         } else {
4205                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4206         }
4207         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4208
4209         if (!IS_ERR(inode) && root != sub_root) {
4210                 down_read(&root->fs_info->cleanup_work_sem);
4211                 if (!(inode->i_sb->s_flags & MS_RDONLY))
4212                         ret = btrfs_orphan_cleanup(sub_root);
4213                 up_read(&root->fs_info->cleanup_work_sem);
4214                 if (ret)
4215                         inode = ERR_PTR(ret);
4216         }
4217
4218         return inode;
4219 }
4220
4221 static int btrfs_dentry_delete(const struct dentry *dentry)
4222 {
4223         struct btrfs_root *root;
4224         struct inode *inode = dentry->d_inode;
4225
4226         if (!inode && !IS_ROOT(dentry))
4227                 inode = dentry->d_parent->d_inode;
4228
4229         if (inode) {
4230                 root = BTRFS_I(inode)->root;
4231                 if (btrfs_root_refs(&root->root_item) == 0)
4232                         return 1;
4233
4234                 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
4235                         return 1;
4236         }
4237         return 0;
4238 }
4239
4240 static void btrfs_dentry_release(struct dentry *dentry)
4241 {
4242         if (dentry->d_fsdata)
4243                 kfree(dentry->d_fsdata);
4244 }
4245
4246 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4247                                    struct nameidata *nd)
4248 {
4249         struct dentry *ret;
4250
4251         ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4252         if (unlikely(d_need_lookup(dentry))) {
4253                 spin_lock(&dentry->d_lock);
4254                 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4255                 spin_unlock(&dentry->d_lock);
4256         }
4257         return ret;
4258 }
4259
4260 unsigned char btrfs_filetype_table[] = {
4261         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4262 };
4263
4264 static int btrfs_real_readdir(struct file *filp, void *dirent,
4265                               filldir_t filldir)
4266 {
4267         struct inode *inode = filp->f_dentry->d_inode;
4268         struct btrfs_root *root = BTRFS_I(inode)->root;
4269         struct btrfs_item *item;
4270         struct btrfs_dir_item *di;
4271         struct btrfs_key key;
4272         struct btrfs_key found_key;
4273         struct btrfs_path *path;
4274         struct list_head ins_list;
4275         struct list_head del_list;
4276         int ret;
4277         struct extent_buffer *leaf;
4278         int slot;
4279         unsigned char d_type;
4280         int over = 0;
4281         u32 di_cur;
4282         u32 di_total;
4283         u32 di_len;
4284         int key_type = BTRFS_DIR_INDEX_KEY;
4285         char tmp_name[32];
4286         char *name_ptr;
4287         int name_len;
4288         int is_curr = 0;        /* filp->f_pos points to the current index? */
4289
4290         /* FIXME, use a real flag for deciding about the key type */
4291         if (root->fs_info->tree_root == root)
4292                 key_type = BTRFS_DIR_ITEM_KEY;
4293
4294         /* special case for "." */
4295         if (filp->f_pos == 0) {
4296                 over = filldir(dirent, ".", 1,
4297                                filp->f_pos, btrfs_ino(inode), DT_DIR);
4298                 if (over)
4299                         return 0;
4300                 filp->f_pos = 1;
4301         }
4302         /* special case for .., just use the back ref */
4303         if (filp->f_pos == 1) {
4304                 u64 pino = parent_ino(filp->f_path.dentry);
4305                 over = filldir(dirent, "..", 2,
4306                                filp->f_pos, pino, DT_DIR);
4307                 if (over)
4308                         return 0;
4309                 filp->f_pos = 2;
4310         }
4311         path = btrfs_alloc_path();
4312         if (!path)
4313                 return -ENOMEM;
4314
4315         path->reada = 1;
4316
4317         if (key_type == BTRFS_DIR_INDEX_KEY) {
4318                 INIT_LIST_HEAD(&ins_list);
4319                 INIT_LIST_HEAD(&del_list);
4320                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4321         }
4322
4323         btrfs_set_key_type(&key, key_type);
4324         key.offset = filp->f_pos;
4325         key.objectid = btrfs_ino(inode);
4326
4327         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4328         if (ret < 0)
4329                 goto err;
4330
4331         while (1) {
4332                 leaf = path->nodes[0];
4333                 slot = path->slots[0];
4334                 if (slot >= btrfs_header_nritems(leaf)) {
4335                         ret = btrfs_next_leaf(root, path);
4336                         if (ret < 0)
4337                                 goto err;
4338                         else if (ret > 0)
4339                                 break;
4340                         continue;
4341                 }
4342
4343                 item = btrfs_item_nr(leaf, slot);
4344                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4345
4346                 if (found_key.objectid != key.objectid)
4347                         break;
4348                 if (btrfs_key_type(&found_key) != key_type)
4349                         break;
4350                 if (found_key.offset < filp->f_pos)
4351                         goto next;
4352                 if (key_type == BTRFS_DIR_INDEX_KEY &&
4353                     btrfs_should_delete_dir_index(&del_list,
4354                                                   found_key.offset))
4355                         goto next;
4356
4357                 filp->f_pos = found_key.offset;
4358                 is_curr = 1;
4359
4360                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4361                 di_cur = 0;
4362                 di_total = btrfs_item_size(leaf, item);
4363
4364                 while (di_cur < di_total) {
4365                         struct btrfs_key location;
4366
4367                         if (verify_dir_item(root, leaf, di))
4368                                 break;
4369
4370                         name_len = btrfs_dir_name_len(leaf, di);
4371                         if (name_len <= sizeof(tmp_name)) {
4372                                 name_ptr = tmp_name;
4373                         } else {
4374                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4375                                 if (!name_ptr) {
4376                                         ret = -ENOMEM;
4377                                         goto err;
4378                                 }
4379                         }
4380                         read_extent_buffer(leaf, name_ptr,
4381                                            (unsigned long)(di + 1), name_len);
4382
4383                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4384                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4385
4386
4387                         /* is this a reference to our own snapshot? If so
4388                          * skip it.
4389                          *
4390                          * In contrast to old kernels, we insert the snapshot's
4391                          * dir item and dir index after it has been created, so
4392                          * we won't find a reference to our own snapshot. We
4393                          * still keep the following code for backward
4394                          * compatibility.
4395                          */
4396                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4397                             location.objectid == root->root_key.objectid) {
4398                                 over = 0;
4399                                 goto skip;
4400                         }
4401                         over = filldir(dirent, name_ptr, name_len,
4402                                        found_key.offset, location.objectid,
4403                                        d_type);
4404
4405 skip:
4406                         if (name_ptr != tmp_name)
4407                                 kfree(name_ptr);
4408
4409                         if (over)
4410                                 goto nopos;
4411                         di_len = btrfs_dir_name_len(leaf, di) +
4412                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4413                         di_cur += di_len;
4414                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4415                 }
4416 next:
4417                 path->slots[0]++;
4418         }
4419
4420         if (key_type == BTRFS_DIR_INDEX_KEY) {
4421                 if (is_curr)
4422                         filp->f_pos++;
4423                 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4424                                                       &ins_list);
4425                 if (ret)
4426                         goto nopos;
4427         }
4428
4429         /* Reached end of directory/root. Bump pos past the last item. */
4430         if (key_type == BTRFS_DIR_INDEX_KEY)
4431                 /*
4432                  * 32-bit glibc will use getdents64, but then strtol -
4433                  * so the last number we can serve is this.
4434                  */
4435                 filp->f_pos = 0x7fffffff;
4436         else
4437                 filp->f_pos++;
4438 nopos:
4439         ret = 0;
4440 err:
4441         if (key_type == BTRFS_DIR_INDEX_KEY)
4442                 btrfs_put_delayed_items(&ins_list, &del_list);
4443         btrfs_free_path(path);
4444         return ret;
4445 }
4446
4447 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4448 {
4449         struct btrfs_root *root = BTRFS_I(inode)->root;
4450         struct btrfs_trans_handle *trans;
4451         int ret = 0;
4452         bool nolock = false;
4453
4454         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4455                 return 0;
4456
4457         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(root, inode))
4458                 nolock = true;
4459
4460         if (wbc->sync_mode == WB_SYNC_ALL) {
4461                 if (nolock)
4462                         trans = btrfs_join_transaction_nolock(root);
4463                 else
4464                         trans = btrfs_join_transaction(root);
4465                 if (IS_ERR(trans))
4466                         return PTR_ERR(trans);
4467                 if (nolock)
4468                         ret = btrfs_end_transaction_nolock(trans, root);
4469                 else
4470                         ret = btrfs_commit_transaction(trans, root);
4471         }
4472         return ret;
4473 }
4474
4475 /*
4476  * This is somewhat expensive, updating the tree every time the
4477  * inode changes.  But, it is most likely to find the inode in cache.
4478  * FIXME, needs more benchmarking...there are no reasons other than performance
4479  * to keep or drop this code.
4480  */
4481 int btrfs_dirty_inode(struct inode *inode)
4482 {
4483         struct btrfs_root *root = BTRFS_I(inode)->root;
4484         struct btrfs_trans_handle *trans;
4485         int ret;
4486
4487         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4488                 return 0;
4489
4490         trans = btrfs_join_transaction(root);
4491         if (IS_ERR(trans))
4492                 return PTR_ERR(trans);
4493
4494         ret = btrfs_update_inode(trans, root, inode);
4495         if (ret && ret == -ENOSPC) {
4496                 /* whoops, lets try again with the full transaction */
4497                 btrfs_end_transaction(trans, root);
4498                 trans = btrfs_start_transaction(root, 1);
4499                 if (IS_ERR(trans))
4500                         return PTR_ERR(trans);
4501
4502                 ret = btrfs_update_inode(trans, root, inode);
4503         }
4504         btrfs_end_transaction(trans, root);
4505         if (BTRFS_I(inode)->delayed_node)
4506                 btrfs_balance_delayed_items(root);
4507
4508         return ret;
4509 }
4510
4511 /*
4512  * This is a copy of file_update_time.  We need this so we can return error on
4513  * ENOSPC for updating the inode in the case of file write and mmap writes.
4514  */
4515 int btrfs_update_time(struct file *file)
4516 {
4517         struct inode *inode = file->f_path.dentry->d_inode;
4518         struct timespec now;
4519         int ret;
4520         enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
4521
4522         /* First try to exhaust all avenues to not sync */
4523         if (IS_NOCMTIME(inode))
4524                 return 0;
4525
4526         now = current_fs_time(inode->i_sb);
4527         if (!timespec_equal(&inode->i_mtime, &now))
4528                 sync_it = S_MTIME;
4529
4530         if (!timespec_equal(&inode->i_ctime, &now))
4531                 sync_it |= S_CTIME;
4532
4533         if (IS_I_VERSION(inode))
4534                 sync_it |= S_VERSION;
4535
4536         if (!sync_it)
4537                 return 0;
4538
4539         /* Finally allowed to write? Takes lock. */
4540         if (mnt_want_write_file(file))
4541                 return 0;
4542
4543         /* Only change inode inside the lock region */
4544         if (sync_it & S_VERSION)
4545                 inode_inc_iversion(inode);
4546         if (sync_it & S_CTIME)
4547                 inode->i_ctime = now;
4548         if (sync_it & S_MTIME)
4549                 inode->i_mtime = now;
4550         ret = btrfs_dirty_inode(inode);
4551         if (!ret)
4552                 mark_inode_dirty_sync(inode);
4553         mnt_drop_write(file->f_path.mnt);
4554         return ret;
4555 }
4556
4557 /*
4558  * find the highest existing sequence number in a directory
4559  * and then set the in-memory index_cnt variable to reflect
4560  * free sequence numbers
4561  */
4562 static int btrfs_set_inode_index_count(struct inode *inode)
4563 {
4564         struct btrfs_root *root = BTRFS_I(inode)->root;
4565         struct btrfs_key key, found_key;
4566         struct btrfs_path *path;
4567         struct extent_buffer *leaf;
4568         int ret;
4569
4570         key.objectid = btrfs_ino(inode);
4571         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4572         key.offset = (u64)-1;
4573
4574         path = btrfs_alloc_path();
4575         if (!path)
4576                 return -ENOMEM;
4577
4578         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4579         if (ret < 0)
4580                 goto out;
4581         /* FIXME: we should be able to handle this */
4582         if (ret == 0)
4583                 goto out;
4584         ret = 0;
4585
4586         /*
4587          * MAGIC NUMBER EXPLANATION:
4588          * since we search a directory based on f_pos we have to start at 2
4589          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4590          * else has to start at 2
4591          */
4592         if (path->slots[0] == 0) {
4593                 BTRFS_I(inode)->index_cnt = 2;
4594                 goto out;
4595         }
4596
4597         path->slots[0]--;
4598
4599         leaf = path->nodes[0];
4600         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4601
4602         if (found_key.objectid != btrfs_ino(inode) ||
4603             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4604                 BTRFS_I(inode)->index_cnt = 2;
4605                 goto out;
4606         }
4607
4608         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4609 out:
4610         btrfs_free_path(path);
4611         return ret;
4612 }
4613
4614 /*
4615  * helper to find a free sequence number in a given directory.  This current
4616  * code is very simple, later versions will do smarter things in the btree
4617  */
4618 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4619 {
4620         int ret = 0;
4621
4622         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4623                 ret = btrfs_inode_delayed_dir_index_count(dir);
4624                 if (ret) {
4625                         ret = btrfs_set_inode_index_count(dir);
4626                         if (ret)
4627                                 return ret;
4628                 }
4629         }
4630
4631         *index = BTRFS_I(dir)->index_cnt;
4632         BTRFS_I(dir)->index_cnt++;
4633
4634         return ret;
4635 }
4636
4637 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4638                                      struct btrfs_root *root,
4639                                      struct inode *dir,
4640                                      const char *name, int name_len,
4641                                      u64 ref_objectid, u64 objectid,
4642                                      umode_t mode, u64 *index)
4643 {
4644         struct inode *inode;
4645         struct btrfs_inode_item *inode_item;
4646         struct btrfs_key *location;
4647         struct btrfs_path *path;
4648         struct btrfs_inode_ref *ref;
4649         struct btrfs_key key[2];
4650         u32 sizes[2];
4651         unsigned long ptr;
4652         int ret;
4653         int owner;
4654
4655         path = btrfs_alloc_path();
4656         if (!path)
4657                 return ERR_PTR(-ENOMEM);
4658
4659         inode = new_inode(root->fs_info->sb);
4660         if (!inode) {
4661                 btrfs_free_path(path);
4662                 return ERR_PTR(-ENOMEM);
4663         }
4664
4665         /*
4666          * we have to initialize this early, so we can reclaim the inode
4667          * number if we fail afterwards in this function.
4668          */
4669         inode->i_ino = objectid;
4670
4671         if (dir) {
4672                 trace_btrfs_inode_request(dir);
4673
4674                 ret = btrfs_set_inode_index(dir, index);
4675                 if (ret) {
4676                         btrfs_free_path(path);
4677                         iput(inode);
4678                         return ERR_PTR(ret);
4679                 }
4680         }
4681         /*
4682          * index_cnt is ignored for everything but a dir,
4683          * btrfs_get_inode_index_count has an explanation for the magic
4684          * number
4685          */
4686         BTRFS_I(inode)->index_cnt = 2;
4687         BTRFS_I(inode)->root = root;
4688         BTRFS_I(inode)->generation = trans->transid;
4689         inode->i_generation = BTRFS_I(inode)->generation;
4690         btrfs_set_inode_space_info(root, inode);
4691
4692         if (S_ISDIR(mode))
4693                 owner = 0;
4694         else
4695                 owner = 1;
4696
4697         key[0].objectid = objectid;
4698         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4699         key[0].offset = 0;
4700
4701         key[1].objectid = objectid;
4702         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4703         key[1].offset = ref_objectid;
4704
4705         sizes[0] = sizeof(struct btrfs_inode_item);
4706         sizes[1] = name_len + sizeof(*ref);
4707
4708         path->leave_spinning = 1;
4709         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4710         if (ret != 0)
4711                 goto fail;
4712
4713         inode_init_owner(inode, dir, mode);
4714         inode_set_bytes(inode, 0);
4715         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4716         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4717                                   struct btrfs_inode_item);
4718         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4719
4720         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4721                              struct btrfs_inode_ref);
4722         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4723         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4724         ptr = (unsigned long)(ref + 1);
4725         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4726
4727         btrfs_mark_buffer_dirty(path->nodes[0]);
4728         btrfs_free_path(path);
4729
4730         location = &BTRFS_I(inode)->location;
4731         location->objectid = objectid;
4732         location->offset = 0;
4733         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4734
4735         btrfs_inherit_iflags(inode, dir);
4736
4737         if (S_ISREG(mode)) {
4738                 if (btrfs_test_opt(root, NODATASUM))
4739                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4740                 if (btrfs_test_opt(root, NODATACOW) ||
4741                     (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4742                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4743         }
4744
4745         insert_inode_hash(inode);
4746         inode_tree_add(inode);
4747
4748         trace_btrfs_inode_new(inode);
4749         btrfs_set_inode_last_trans(trans, inode);
4750
4751         return inode;
4752 fail:
4753         if (dir)
4754                 BTRFS_I(dir)->index_cnt--;
4755         btrfs_free_path(path);
4756         iput(inode);
4757         return ERR_PTR(ret);
4758 }
4759
4760 static inline u8 btrfs_inode_type(struct inode *inode)
4761 {
4762         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4763 }
4764
4765 /*
4766  * utility function to add 'inode' into 'parent_inode' with
4767  * a give name and a given sequence number.
4768  * if 'add_backref' is true, also insert a backref from the
4769  * inode to the parent directory.
4770  */
4771 int btrfs_add_link(struct btrfs_trans_handle *trans,
4772                    struct inode *parent_inode, struct inode *inode,
4773                    const char *name, int name_len, int add_backref, u64 index)
4774 {
4775         int ret = 0;
4776         struct btrfs_key key;
4777         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4778         u64 ino = btrfs_ino(inode);
4779         u64 parent_ino = btrfs_ino(parent_inode);
4780
4781         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4782                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4783         } else {
4784                 key.objectid = ino;
4785                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4786                 key.offset = 0;
4787         }
4788
4789         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4790                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4791                                          key.objectid, root->root_key.objectid,
4792                                          parent_ino, index, name, name_len);
4793         } else if (add_backref) {
4794                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4795                                              parent_ino, index);
4796         }
4797
4798         /* Nothing to clean up yet */
4799         if (ret)
4800                 return ret;
4801
4802         ret = btrfs_insert_dir_item(trans, root, name, name_len,
4803                                     parent_inode, &key,
4804                                     btrfs_inode_type(inode), index);
4805         if (ret == -EEXIST)
4806                 goto fail_dir_item;
4807         else if (ret) {
4808                 btrfs_abort_transaction(trans, root, ret);
4809                 return ret;
4810         }
4811
4812         btrfs_i_size_write(parent_inode, parent_inode->i_size +
4813                            name_len * 2);
4814         inode_inc_iversion(parent_inode);
4815         parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4816         ret = btrfs_update_inode(trans, root, parent_inode);
4817         if (ret)
4818                 btrfs_abort_transaction(trans, root, ret);
4819         return ret;
4820
4821 fail_dir_item:
4822         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4823                 u64 local_index;
4824                 int err;
4825                 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4826                                  key.objectid, root->root_key.objectid,
4827                                  parent_ino, &local_index, name, name_len);
4828
4829         } else if (add_backref) {
4830                 u64 local_index;
4831                 int err;
4832
4833                 err = btrfs_del_inode_ref(trans, root, name, name_len,
4834                                           ino, parent_ino, &local_index);
4835         }
4836         return ret;
4837 }
4838
4839 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4840                             struct inode *dir, struct dentry *dentry,
4841                             struct inode *inode, int backref, u64 index)
4842 {
4843         int err = btrfs_add_link(trans, dir, inode,
4844                                  dentry->d_name.name, dentry->d_name.len,
4845                                  backref, index);
4846         if (err > 0)
4847                 err = -EEXIST;
4848         return err;
4849 }
4850
4851 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4852                         umode_t mode, dev_t rdev)
4853 {
4854         struct btrfs_trans_handle *trans;
4855         struct btrfs_root *root = BTRFS_I(dir)->root;
4856         struct inode *inode = NULL;
4857         int err;
4858         int drop_inode = 0;
4859         u64 objectid;
4860         unsigned long nr = 0;
4861         u64 index = 0;
4862
4863         if (!new_valid_dev(rdev))
4864                 return -EINVAL;
4865
4866         /*
4867          * 2 for inode item and ref
4868          * 2 for dir items
4869          * 1 for xattr if selinux is on
4870          */
4871         trans = btrfs_start_transaction(root, 5);
4872         if (IS_ERR(trans))
4873                 return PTR_ERR(trans);
4874
4875         err = btrfs_find_free_ino(root, &objectid);
4876         if (err)
4877                 goto out_unlock;
4878
4879         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4880                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4881                                 mode, &index);
4882         if (IS_ERR(inode)) {
4883                 err = PTR_ERR(inode);
4884                 goto out_unlock;
4885         }
4886
4887         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4888         if (err) {
4889                 drop_inode = 1;
4890                 goto out_unlock;
4891         }
4892
4893         /*
4894         * If the active LSM wants to access the inode during
4895         * d_instantiate it needs these. Smack checks to see
4896         * if the filesystem supports xattrs by looking at the
4897         * ops vector.
4898         */
4899
4900         inode->i_op = &btrfs_special_inode_operations;
4901         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4902         if (err)
4903                 drop_inode = 1;
4904         else {
4905                 init_special_inode(inode, inode->i_mode, rdev);
4906                 btrfs_update_inode(trans, root, inode);
4907                 d_instantiate(dentry, inode);
4908         }
4909 out_unlock:
4910         nr = trans->blocks_used;
4911         btrfs_end_transaction(trans, root);
4912         btrfs_btree_balance_dirty(root, nr);
4913         if (drop_inode) {
4914                 inode_dec_link_count(inode);
4915                 iput(inode);
4916         }
4917         return err;
4918 }
4919
4920 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4921                         umode_t mode, struct nameidata *nd)
4922 {
4923         struct btrfs_trans_handle *trans;
4924         struct btrfs_root *root = BTRFS_I(dir)->root;
4925         struct inode *inode = NULL;
4926         int drop_inode = 0;
4927         int err;
4928         unsigned long nr = 0;
4929         u64 objectid;
4930         u64 index = 0;
4931
4932         /*
4933          * 2 for inode item and ref
4934          * 2 for dir items
4935          * 1 for xattr if selinux is on
4936          */
4937         trans = btrfs_start_transaction(root, 5);
4938         if (IS_ERR(trans))
4939                 return PTR_ERR(trans);
4940
4941         err = btrfs_find_free_ino(root, &objectid);
4942         if (err)
4943                 goto out_unlock;
4944
4945         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4946                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4947                                 mode, &index);
4948         if (IS_ERR(inode)) {
4949                 err = PTR_ERR(inode);
4950                 goto out_unlock;
4951         }
4952
4953         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4954         if (err) {
4955                 drop_inode = 1;
4956                 goto out_unlock;
4957         }
4958
4959         /*
4960         * If the active LSM wants to access the inode during
4961         * d_instantiate it needs these. Smack checks to see
4962         * if the filesystem supports xattrs by looking at the
4963         * ops vector.
4964         */
4965         inode->i_fop = &btrfs_file_operations;
4966         inode->i_op = &btrfs_file_inode_operations;
4967
4968         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4969         if (err)
4970                 drop_inode = 1;
4971         else {
4972                 inode->i_mapping->a_ops = &btrfs_aops;
4973                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4974                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4975                 d_instantiate(dentry, inode);
4976         }
4977 out_unlock:
4978         nr = trans->blocks_used;
4979         btrfs_end_transaction(trans, root);
4980         if (drop_inode) {
4981                 inode_dec_link_count(inode);
4982                 iput(inode);
4983         }
4984         btrfs_btree_balance_dirty(root, nr);
4985         return err;
4986 }
4987
4988 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4989                       struct dentry *dentry)
4990 {
4991         struct btrfs_trans_handle *trans;
4992         struct btrfs_root *root = BTRFS_I(dir)->root;
4993         struct inode *inode = old_dentry->d_inode;
4994         u64 index;
4995         unsigned long nr = 0;
4996         int err;
4997         int drop_inode = 0;
4998
4999         /* do not allow sys_link's with other subvols of the same device */
5000         if (root->objectid != BTRFS_I(inode)->root->objectid)
5001                 return -EXDEV;
5002
5003         if (inode->i_nlink == ~0U)
5004                 return -EMLINK;
5005
5006         err = btrfs_set_inode_index(dir, &index);
5007         if (err)
5008                 goto fail;
5009
5010         /*
5011          * 2 items for inode and inode ref
5012          * 2 items for dir items
5013          * 1 item for parent inode
5014          */
5015         trans = btrfs_start_transaction(root, 5);
5016         if (IS_ERR(trans)) {
5017                 err = PTR_ERR(trans);
5018                 goto fail;
5019         }
5020
5021         btrfs_inc_nlink(inode);
5022         inode_inc_iversion(inode);
5023         inode->i_ctime = CURRENT_TIME;
5024         ihold(inode);
5025
5026         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5027
5028         if (err) {
5029                 drop_inode = 1;
5030         } else {
5031                 struct dentry *parent = dentry->d_parent;
5032                 err = btrfs_update_inode(trans, root, inode);
5033                 if (err)
5034                         goto fail;
5035                 d_instantiate(dentry, inode);
5036                 btrfs_log_new_name(trans, inode, NULL, parent);
5037         }
5038
5039         nr = trans->blocks_used;
5040         btrfs_end_transaction(trans, root);
5041 fail:
5042         if (drop_inode) {
5043                 inode_dec_link_count(inode);
5044                 iput(inode);
5045         }
5046         btrfs_btree_balance_dirty(root, nr);
5047         return err;
5048 }
5049
5050 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5051 {
5052         struct inode *inode = NULL;
5053         struct btrfs_trans_handle *trans;
5054         struct btrfs_root *root = BTRFS_I(dir)->root;
5055         int err = 0;
5056         int drop_on_err = 0;
5057         u64 objectid = 0;
5058         u64 index = 0;
5059         unsigned long nr = 1;
5060
5061         /*
5062          * 2 items for inode and ref
5063          * 2 items for dir items
5064          * 1 for xattr if selinux is on
5065          */
5066         trans = btrfs_start_transaction(root, 5);
5067         if (IS_ERR(trans))
5068                 return PTR_ERR(trans);
5069
5070         err = btrfs_find_free_ino(root, &objectid);
5071         if (err)
5072                 goto out_fail;
5073
5074         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5075                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5076                                 S_IFDIR | mode, &index);
5077         if (IS_ERR(inode)) {
5078                 err = PTR_ERR(inode);
5079                 goto out_fail;
5080         }
5081
5082         drop_on_err = 1;
5083
5084         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5085         if (err)
5086                 goto out_fail;
5087
5088         inode->i_op = &btrfs_dir_inode_operations;
5089         inode->i_fop = &btrfs_dir_file_operations;
5090
5091         btrfs_i_size_write(inode, 0);
5092         err = btrfs_update_inode(trans, root, inode);
5093         if (err)
5094                 goto out_fail;
5095
5096         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5097                              dentry->d_name.len, 0, index);
5098         if (err)
5099                 goto out_fail;
5100
5101         d_instantiate(dentry, inode);
5102         drop_on_err = 0;
5103
5104 out_fail:
5105         nr = trans->blocks_used;
5106         btrfs_end_transaction(trans, root);
5107         if (drop_on_err)
5108                 iput(inode);
5109         btrfs_btree_balance_dirty(root, nr);
5110         return err;
5111 }
5112
5113 /* helper for btfs_get_extent.  Given an existing extent in the tree,
5114  * and an extent that you want to insert, deal with overlap and insert
5115  * the new extent into the tree.
5116  */
5117 static int merge_extent_mapping(struct extent_map_tree *em_tree,
5118                                 struct extent_map *existing,
5119                                 struct extent_map *em,
5120                                 u64 map_start, u64 map_len)
5121 {
5122         u64 start_diff;
5123
5124         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5125         start_diff = map_start - em->start;
5126         em->start = map_start;
5127         em->len = map_len;
5128         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5129             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
5130                 em->block_start += start_diff;
5131                 em->block_len -= start_diff;
5132         }
5133         return add_extent_mapping(em_tree, em);
5134 }
5135
5136 static noinline int uncompress_inline(struct btrfs_path *path,
5137                                       struct inode *inode, struct page *page,
5138                                       size_t pg_offset, u64 extent_offset,
5139                                       struct btrfs_file_extent_item *item)
5140 {
5141         int ret;
5142         struct extent_buffer *leaf = path->nodes[0];
5143         char *tmp;
5144         size_t max_size;
5145         unsigned long inline_size;
5146         unsigned long ptr;
5147         int compress_type;
5148
5149         WARN_ON(pg_offset != 0);
5150         compress_type = btrfs_file_extent_compression(leaf, item);
5151         max_size = btrfs_file_extent_ram_bytes(leaf, item);
5152         inline_size = btrfs_file_extent_inline_item_len(leaf,
5153                                         btrfs_item_nr(leaf, path->slots[0]));
5154         tmp = kmalloc(inline_size, GFP_NOFS);
5155         if (!tmp)
5156                 return -ENOMEM;
5157         ptr = btrfs_file_extent_inline_start(item);
5158
5159         read_extent_buffer(leaf, tmp, ptr, inline_size);
5160
5161         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5162         ret = btrfs_decompress(compress_type, tmp, page,
5163                                extent_offset, inline_size, max_size);
5164         if (ret) {
5165                 char *kaddr = kmap_atomic(page);
5166                 unsigned long copy_size = min_t(u64,
5167                                   PAGE_CACHE_SIZE - pg_offset,
5168                                   max_size - extent_offset);
5169                 memset(kaddr + pg_offset, 0, copy_size);
5170                 kunmap_atomic(kaddr);
5171         }
5172         kfree(tmp);
5173         return 0;
5174 }
5175
5176 /*
5177  * a bit scary, this does extent mapping from logical file offset to the disk.
5178  * the ugly parts come from merging extents from the disk with the in-ram
5179  * representation.  This gets more complex because of the data=ordered code,
5180  * where the in-ram extents might be locked pending data=ordered completion.
5181  *
5182  * This also copies inline extents directly into the page.
5183  */
5184
5185 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5186                                     size_t pg_offset, u64 start, u64 len,
5187                                     int create)
5188 {
5189         int ret;
5190         int err = 0;
5191         u64 bytenr;
5192         u64 extent_start = 0;
5193         u64 extent_end = 0;
5194         u64 objectid = btrfs_ino(inode);
5195         u32 found_type;
5196         struct btrfs_path *path = NULL;
5197         struct btrfs_root *root = BTRFS_I(inode)->root;
5198         struct btrfs_file_extent_item *item;
5199         struct extent_buffer *leaf;
5200         struct btrfs_key found_key;
5201         struct extent_map *em = NULL;
5202         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5203         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5204         struct btrfs_trans_handle *trans = NULL;
5205         int compress_type;
5206
5207 again:
5208         read_lock(&em_tree->lock);
5209         em = lookup_extent_mapping(em_tree, start, len);
5210         if (em)
5211                 em->bdev = root->fs_info->fs_devices->latest_bdev;
5212         read_unlock(&em_tree->lock);
5213
5214         if (em) {
5215                 if (em->start > start || em->start + em->len <= start)
5216                         free_extent_map(em);
5217                 else if (em->block_start == EXTENT_MAP_INLINE && page)
5218                         free_extent_map(em);
5219                 else
5220                         goto out;
5221         }
5222         em = alloc_extent_map();
5223         if (!em) {
5224                 err = -ENOMEM;
5225                 goto out;
5226         }
5227         em->bdev = root->fs_info->fs_devices->latest_bdev;
5228         em->start = EXTENT_MAP_HOLE;
5229         em->orig_start = EXTENT_MAP_HOLE;
5230         em->len = (u64)-1;
5231         em->block_len = (u64)-1;
5232
5233         if (!path) {
5234                 path = btrfs_alloc_path();
5235                 if (!path) {
5236                         err = -ENOMEM;
5237                         goto out;
5238                 }
5239                 /*
5240                  * Chances are we'll be called again, so go ahead and do
5241                  * readahead
5242                  */
5243                 path->reada = 1;
5244         }
5245
5246         ret = btrfs_lookup_file_extent(trans, root, path,
5247                                        objectid, start, trans != NULL);
5248         if (ret < 0) {
5249                 err = ret;
5250                 goto out;
5251         }
5252
5253         if (ret != 0) {
5254                 if (path->slots[0] == 0)
5255                         goto not_found;
5256                 path->slots[0]--;
5257         }
5258
5259         leaf = path->nodes[0];
5260         item = btrfs_item_ptr(leaf, path->slots[0],
5261                               struct btrfs_file_extent_item);
5262         /* are we inside the extent that was found? */
5263         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5264         found_type = btrfs_key_type(&found_key);
5265         if (found_key.objectid != objectid ||
5266             found_type != BTRFS_EXTENT_DATA_KEY) {
5267                 goto not_found;
5268         }
5269
5270         found_type = btrfs_file_extent_type(leaf, item);
5271         extent_start = found_key.offset;
5272         compress_type = btrfs_file_extent_compression(leaf, item);
5273         if (found_type == BTRFS_FILE_EXTENT_REG ||
5274             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5275                 extent_end = extent_start +
5276                        btrfs_file_extent_num_bytes(leaf, item);
5277         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5278                 size_t size;
5279                 size = btrfs_file_extent_inline_len(leaf, item);
5280                 extent_end = (extent_start + size + root->sectorsize - 1) &
5281                         ~((u64)root->sectorsize - 1);
5282         }
5283
5284         if (start >= extent_end) {
5285                 path->slots[0]++;
5286                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5287                         ret = btrfs_next_leaf(root, path);
5288                         if (ret < 0) {
5289                                 err = ret;
5290                                 goto out;
5291                         }
5292                         if (ret > 0)
5293                                 goto not_found;
5294                         leaf = path->nodes[0];
5295                 }
5296                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5297                 if (found_key.objectid != objectid ||
5298                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5299                         goto not_found;
5300                 if (start + len <= found_key.offset)
5301                         goto not_found;
5302                 em->start = start;
5303                 em->len = found_key.offset - start;
5304                 goto not_found_em;
5305         }
5306
5307         if (found_type == BTRFS_FILE_EXTENT_REG ||
5308             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5309                 em->start = extent_start;
5310                 em->len = extent_end - extent_start;
5311                 em->orig_start = extent_start -
5312                                  btrfs_file_extent_offset(leaf, item);
5313                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5314                 if (bytenr == 0) {
5315                         em->block_start = EXTENT_MAP_HOLE;
5316                         goto insert;
5317                 }
5318                 if (compress_type != BTRFS_COMPRESS_NONE) {
5319                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5320                         em->compress_type = compress_type;
5321                         em->block_start = bytenr;
5322                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5323                                                                          item);
5324                 } else {
5325                         bytenr += btrfs_file_extent_offset(leaf, item);
5326                         em->block_start = bytenr;
5327                         em->block_len = em->len;
5328                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5329                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5330                 }
5331                 goto insert;
5332         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5333                 unsigned long ptr;
5334                 char *map;
5335                 size_t size;
5336                 size_t extent_offset;
5337                 size_t copy_size;
5338
5339                 em->block_start = EXTENT_MAP_INLINE;
5340                 if (!page || create) {
5341                         em->start = extent_start;
5342                         em->len = extent_end - extent_start;
5343                         goto out;
5344                 }
5345
5346                 size = btrfs_file_extent_inline_len(leaf, item);
5347                 extent_offset = page_offset(page) + pg_offset - extent_start;
5348                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5349                                 size - extent_offset);
5350                 em->start = extent_start + extent_offset;
5351                 em->len = (copy_size + root->sectorsize - 1) &
5352                         ~((u64)root->sectorsize - 1);
5353                 em->orig_start = EXTENT_MAP_INLINE;
5354                 if (compress_type) {
5355                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5356                         em->compress_type = compress_type;
5357                 }
5358                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5359                 if (create == 0 && !PageUptodate(page)) {
5360                         if (btrfs_file_extent_compression(leaf, item) !=
5361                             BTRFS_COMPRESS_NONE) {
5362                                 ret = uncompress_inline(path, inode, page,
5363                                                         pg_offset,
5364                                                         extent_offset, item);
5365                                 BUG_ON(ret); /* -ENOMEM */
5366                         } else {
5367                                 map = kmap(page);
5368                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5369                                                    copy_size);
5370                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5371                                         memset(map + pg_offset + copy_size, 0,
5372                                                PAGE_CACHE_SIZE - pg_offset -
5373                                                copy_size);
5374                                 }
5375                                 kunmap(page);
5376                         }
5377                         flush_dcache_page(page);
5378                 } else if (create && PageUptodate(page)) {
5379                         BUG();
5380                         if (!trans) {
5381                                 kunmap(page);
5382                                 free_extent_map(em);
5383                                 em = NULL;
5384
5385                                 btrfs_release_path(path);
5386                                 trans = btrfs_join_transaction(root);
5387
5388                                 if (IS_ERR(trans))
5389                                         return ERR_CAST(trans);
5390                                 goto again;
5391                         }
5392                         map = kmap(page);
5393                         write_extent_buffer(leaf, map + pg_offset, ptr,
5394                                             copy_size);
5395                         kunmap(page);
5396                         btrfs_mark_buffer_dirty(leaf);
5397                 }
5398                 set_extent_uptodate(io_tree, em->start,
5399                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
5400                 goto insert;
5401         } else {
5402                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5403                 WARN_ON(1);
5404         }
5405 not_found:
5406         em->start = start;
5407         em->len = len;
5408 not_found_em:
5409         em->block_start = EXTENT_MAP_HOLE;
5410         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5411 insert:
5412         btrfs_release_path(path);
5413         if (em->start > start || extent_map_end(em) <= start) {
5414                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5415                        "[%llu %llu]\n", (unsigned long long)em->start,
5416                        (unsigned long long)em->len,
5417                        (unsigned long long)start,
5418                        (unsigned long long)len);
5419                 err = -EIO;
5420                 goto out;
5421         }
5422
5423         err = 0;
5424         write_lock(&em_tree->lock);
5425         ret = add_extent_mapping(em_tree, em);
5426         /* it is possible that someone inserted the extent into the tree
5427          * while we had the lock dropped.  It is also possible that
5428          * an overlapping map exists in the tree
5429          */
5430         if (ret == -EEXIST) {
5431                 struct extent_map *existing;
5432
5433                 ret = 0;
5434
5435                 existing = lookup_extent_mapping(em_tree, start, len);
5436                 if (existing && (existing->start > start ||
5437                     existing->start + existing->len <= start)) {
5438                         free_extent_map(existing);
5439                         existing = NULL;
5440                 }
5441                 if (!existing) {
5442                         existing = lookup_extent_mapping(em_tree, em->start,
5443                                                          em->len);
5444                         if (existing) {
5445                                 err = merge_extent_mapping(em_tree, existing,
5446                                                            em, start,
5447                                                            root->sectorsize);
5448                                 free_extent_map(existing);
5449                                 if (err) {
5450                                         free_extent_map(em);
5451                                         em = NULL;
5452                                 }
5453                         } else {
5454                                 err = -EIO;
5455                                 free_extent_map(em);
5456                                 em = NULL;
5457                         }
5458                 } else {
5459                         free_extent_map(em);
5460                         em = existing;
5461                         err = 0;
5462                 }
5463         }
5464         write_unlock(&em_tree->lock);
5465 out:
5466
5467         trace_btrfs_get_extent(root, em);
5468
5469         if (path)
5470                 btrfs_free_path(path);
5471         if (trans) {
5472                 ret = btrfs_end_transaction(trans, root);
5473                 if (!err)
5474                         err = ret;
5475         }
5476         if (err) {
5477                 free_extent_map(em);
5478                 return ERR_PTR(err);
5479         }
5480         BUG_ON(!em); /* Error is always set */
5481         return em;
5482 }
5483
5484 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5485                                            size_t pg_offset, u64 start, u64 len,
5486                                            int create)
5487 {
5488         struct extent_map *em;
5489         struct extent_map *hole_em = NULL;
5490         u64 range_start = start;
5491         u64 end;
5492         u64 found;
5493         u64 found_end;
5494         int err = 0;
5495
5496         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5497         if (IS_ERR(em))
5498                 return em;
5499         if (em) {
5500                 /*
5501                  * if our em maps to a hole, there might
5502                  * actually be delalloc bytes behind it
5503                  */
5504                 if (em->block_start != EXTENT_MAP_HOLE)
5505                         return em;
5506                 else
5507                         hole_em = em;
5508         }
5509
5510         /* check to see if we've wrapped (len == -1 or similar) */
5511         end = start + len;
5512         if (end < start)
5513                 end = (u64)-1;
5514         else
5515                 end -= 1;
5516
5517         em = NULL;
5518
5519         /* ok, we didn't find anything, lets look for delalloc */
5520         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5521                                  end, len, EXTENT_DELALLOC, 1);
5522         found_end = range_start + found;
5523         if (found_end < range_start)
5524                 found_end = (u64)-1;
5525
5526         /*
5527          * we didn't find anything useful, return
5528          * the original results from get_extent()
5529          */
5530         if (range_start > end || found_end <= start) {
5531                 em = hole_em;
5532                 hole_em = NULL;
5533                 goto out;
5534         }
5535
5536         /* adjust the range_start to make sure it doesn't
5537          * go backwards from the start they passed in
5538          */
5539         range_start = max(start,range_start);
5540         found = found_end - range_start;
5541
5542         if (found > 0) {
5543                 u64 hole_start = start;
5544                 u64 hole_len = len;
5545
5546                 em = alloc_extent_map();
5547                 if (!em) {
5548                         err = -ENOMEM;
5549                         goto out;
5550                 }
5551                 /*
5552                  * when btrfs_get_extent can't find anything it
5553                  * returns one huge hole
5554                  *
5555                  * make sure what it found really fits our range, and
5556                  * adjust to make sure it is based on the start from
5557                  * the caller
5558                  */
5559                 if (hole_em) {
5560                         u64 calc_end = extent_map_end(hole_em);
5561
5562                         if (calc_end <= start || (hole_em->start > end)) {
5563                                 free_extent_map(hole_em);
5564                                 hole_em = NULL;
5565                         } else {
5566                                 hole_start = max(hole_em->start, start);
5567                                 hole_len = calc_end - hole_start;
5568                         }
5569                 }
5570                 em->bdev = NULL;
5571                 if (hole_em && range_start > hole_start) {
5572                         /* our hole starts before our delalloc, so we
5573                          * have to return just the parts of the hole
5574                          * that go until  the delalloc starts
5575                          */
5576                         em->len = min(hole_len,
5577                                       range_start - hole_start);
5578                         em->start = hole_start;
5579                         em->orig_start = hole_start;
5580                         /*
5581                          * don't adjust block start at all,
5582                          * it is fixed at EXTENT_MAP_HOLE
5583                          */
5584                         em->block_start = hole_em->block_start;
5585                         em->block_len = hole_len;
5586                 } else {
5587                         em->start = range_start;
5588                         em->len = found;
5589                         em->orig_start = range_start;
5590                         em->block_start = EXTENT_MAP_DELALLOC;
5591                         em->block_len = found;
5592                 }
5593         } else if (hole_em) {
5594                 return hole_em;
5595         }
5596 out:
5597
5598         free_extent_map(hole_em);
5599         if (err) {
5600                 free_extent_map(em);
5601                 return ERR_PTR(err);
5602         }
5603         return em;
5604 }
5605
5606 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5607                                                   struct extent_map *em,
5608                                                   u64 start, u64 len)
5609 {
5610         struct btrfs_root *root = BTRFS_I(inode)->root;
5611         struct btrfs_trans_handle *trans;
5612         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5613         struct btrfs_key ins;
5614         u64 alloc_hint;
5615         int ret;
5616         bool insert = false;
5617
5618         /*
5619          * Ok if the extent map we looked up is a hole and is for the exact
5620          * range we want, there is no reason to allocate a new one, however if
5621          * it is not right then we need to free this one and drop the cache for
5622          * our range.
5623          */
5624         if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5625             em->len != len) {
5626                 free_extent_map(em);
5627                 em = NULL;
5628                 insert = true;
5629                 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5630         }
5631
5632         trans = btrfs_join_transaction(root);
5633         if (IS_ERR(trans))
5634                 return ERR_CAST(trans);
5635
5636         if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5637                 btrfs_add_inode_defrag(trans, inode);
5638
5639         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5640
5641         alloc_hint = get_extent_allocation_hint(inode, start, len);
5642         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5643                                    alloc_hint, &ins, 1);
5644         if (ret) {
5645                 em = ERR_PTR(ret);
5646                 goto out;
5647         }
5648
5649         if (!em) {
5650                 em = alloc_extent_map();
5651                 if (!em) {
5652                         em = ERR_PTR(-ENOMEM);
5653                         goto out;
5654                 }
5655         }
5656
5657         em->start = start;
5658         em->orig_start = em->start;
5659         em->len = ins.offset;
5660
5661         em->block_start = ins.objectid;
5662         em->block_len = ins.offset;
5663         em->bdev = root->fs_info->fs_devices->latest_bdev;
5664
5665         /*
5666          * We need to do this because if we're using the original em we searched
5667          * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5668          */
5669         em->flags = 0;
5670         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5671
5672         while (insert) {
5673                 write_lock(&em_tree->lock);
5674                 ret = add_extent_mapping(em_tree, em);
5675                 write_unlock(&em_tree->lock);
5676                 if (ret != -EEXIST)
5677                         break;
5678                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5679         }
5680
5681         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5682                                            ins.offset, ins.offset, 0);
5683         if (ret) {
5684                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5685                 em = ERR_PTR(ret);
5686         }
5687 out:
5688         btrfs_end_transaction(trans, root);
5689         return em;
5690 }
5691
5692 /*
5693  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5694  * block must be cow'd
5695  */
5696 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5697                                       struct inode *inode, u64 offset, u64 len)
5698 {
5699         struct btrfs_path *path;
5700         int ret;
5701         struct extent_buffer *leaf;
5702         struct btrfs_root *root = BTRFS_I(inode)->root;
5703         struct btrfs_file_extent_item *fi;
5704         struct btrfs_key key;
5705         u64 disk_bytenr;
5706         u64 backref_offset;
5707         u64 extent_end;
5708         u64 num_bytes;
5709         int slot;
5710         int found_type;
5711
5712         path = btrfs_alloc_path();
5713         if (!path)
5714                 return -ENOMEM;
5715
5716         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5717                                        offset, 0);
5718         if (ret < 0)
5719                 goto out;
5720
5721         slot = path->slots[0];
5722         if (ret == 1) {
5723                 if (slot == 0) {
5724                         /* can't find the item, must cow */
5725                         ret = 0;
5726                         goto out;
5727                 }
5728                 slot--;
5729         }
5730         ret = 0;
5731         leaf = path->nodes[0];
5732         btrfs_item_key_to_cpu(leaf, &key, slot);
5733         if (key.objectid != btrfs_ino(inode) ||
5734             key.type != BTRFS_EXTENT_DATA_KEY) {
5735                 /* not our file or wrong item type, must cow */
5736                 goto out;
5737         }
5738
5739         if (key.offset > offset) {
5740                 /* Wrong offset, must cow */
5741                 goto out;
5742         }
5743
5744         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5745         found_type = btrfs_file_extent_type(leaf, fi);
5746         if (found_type != BTRFS_FILE_EXTENT_REG &&
5747             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5748                 /* not a regular extent, must cow */
5749                 goto out;
5750         }
5751         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5752         backref_offset = btrfs_file_extent_offset(leaf, fi);
5753
5754         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5755         if (extent_end < offset + len) {
5756                 /* extent doesn't include our full range, must cow */
5757                 goto out;
5758         }
5759
5760         if (btrfs_extent_readonly(root, disk_bytenr))
5761                 goto out;
5762
5763         /*
5764          * look for other files referencing this extent, if we
5765          * find any we must cow
5766          */
5767         if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5768                                   key.offset - backref_offset, disk_bytenr))
5769                 goto out;
5770
5771         /*
5772          * adjust disk_bytenr and num_bytes to cover just the bytes
5773          * in this extent we are about to write.  If there
5774          * are any csums in that range we have to cow in order
5775          * to keep the csums correct
5776          */
5777         disk_bytenr += backref_offset;
5778         disk_bytenr += offset - key.offset;
5779         num_bytes = min(offset + len, extent_end) - offset;
5780         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5781                                 goto out;
5782         /*
5783          * all of the above have passed, it is safe to overwrite this extent
5784          * without cow
5785          */
5786         ret = 1;
5787 out:
5788         btrfs_free_path(path);
5789         return ret;
5790 }
5791
5792 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5793                                    struct buffer_head *bh_result, int create)
5794 {
5795         struct extent_map *em;
5796         struct btrfs_root *root = BTRFS_I(inode)->root;
5797         u64 start = iblock << inode->i_blkbits;
5798         u64 len = bh_result->b_size;
5799         struct btrfs_trans_handle *trans;
5800
5801         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5802         if (IS_ERR(em))
5803                 return PTR_ERR(em);
5804
5805         /*
5806          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5807          * io.  INLINE is special, and we could probably kludge it in here, but
5808          * it's still buffered so for safety lets just fall back to the generic
5809          * buffered path.
5810          *
5811          * For COMPRESSED we _have_ to read the entire extent in so we can
5812          * decompress it, so there will be buffering required no matter what we
5813          * do, so go ahead and fallback to buffered.
5814          *
5815          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5816          * to buffered IO.  Don't blame me, this is the price we pay for using
5817          * the generic code.
5818          */
5819         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5820             em->block_start == EXTENT_MAP_INLINE) {
5821                 free_extent_map(em);
5822                 return -ENOTBLK;
5823         }
5824
5825         /* Just a good old fashioned hole, return */
5826         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5827                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5828                 free_extent_map(em);
5829                 /* DIO will do one hole at a time, so just unlock a sector */
5830                 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5831                               start + root->sectorsize - 1);
5832                 return 0;
5833         }
5834
5835         /*
5836          * We don't allocate a new extent in the following cases
5837          *
5838          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
5839          * existing extent.
5840          * 2) The extent is marked as PREALLOC.  We're good to go here and can
5841          * just use the extent.
5842          *
5843          */
5844         if (!create) {
5845                 len = em->len - (start - em->start);
5846                 goto map;
5847         }
5848
5849         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5850             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5851              em->block_start != EXTENT_MAP_HOLE)) {
5852                 int type;
5853                 int ret;
5854                 u64 block_start;
5855
5856                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5857                         type = BTRFS_ORDERED_PREALLOC;
5858                 else
5859                         type = BTRFS_ORDERED_NOCOW;
5860                 len = min(len, em->len - (start - em->start));
5861                 block_start = em->block_start + (start - em->start);
5862
5863                 /*
5864                  * we're not going to log anything, but we do need
5865                  * to make sure the current transaction stays open
5866                  * while we look for nocow cross refs
5867                  */
5868                 trans = btrfs_join_transaction(root);
5869                 if (IS_ERR(trans))
5870                         goto must_cow;
5871
5872                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5873                         ret = btrfs_add_ordered_extent_dio(inode, start,
5874                                            block_start, len, len, type);
5875                         btrfs_end_transaction(trans, root);
5876                         if (ret) {
5877                                 free_extent_map(em);
5878                                 return ret;
5879                         }
5880                         goto unlock;
5881                 }
5882                 btrfs_end_transaction(trans, root);
5883         }
5884 must_cow:
5885         /*
5886          * this will cow the extent, reset the len in case we changed
5887          * it above
5888          */
5889         len = bh_result->b_size;
5890         em = btrfs_new_extent_direct(inode, em, start, len);
5891         if (IS_ERR(em))
5892                 return PTR_ERR(em);
5893         len = min(len, em->len - (start - em->start));
5894 unlock:
5895         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5896                           EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5897                           0, NULL, GFP_NOFS);
5898 map:
5899         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5900                 inode->i_blkbits;
5901         bh_result->b_size = len;
5902         bh_result->b_bdev = em->bdev;
5903         set_buffer_mapped(bh_result);
5904         if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5905                 set_buffer_new(bh_result);
5906
5907         free_extent_map(em);
5908
5909         return 0;
5910 }
5911
5912 struct btrfs_dio_private {
5913         struct inode *inode;
5914         u64 logical_offset;
5915         u64 disk_bytenr;
5916         u64 bytes;
5917         u32 *csums;
5918         void *private;
5919
5920         /* number of bios pending for this dio */
5921         atomic_t pending_bios;
5922
5923         /* IO errors */
5924         int errors;
5925
5926         struct bio *orig_bio;
5927 };
5928
5929 static void btrfs_endio_direct_read(struct bio *bio, int err)
5930 {
5931         struct btrfs_dio_private *dip = bio->bi_private;
5932         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5933         struct bio_vec *bvec = bio->bi_io_vec;
5934         struct inode *inode = dip->inode;
5935         struct btrfs_root *root = BTRFS_I(inode)->root;
5936         u64 start;
5937         u32 *private = dip->csums;
5938
5939         start = dip->logical_offset;
5940         do {
5941                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5942                         struct page *page = bvec->bv_page;
5943                         char *kaddr;
5944                         u32 csum = ~(u32)0;
5945                         unsigned long flags;
5946
5947                         local_irq_save(flags);
5948                         kaddr = kmap_atomic(page);
5949                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5950                                                csum, bvec->bv_len);
5951                         btrfs_csum_final(csum, (char *)&csum);
5952                         kunmap_atomic(kaddr);
5953                         local_irq_restore(flags);
5954
5955                         flush_dcache_page(bvec->bv_page);
5956                         if (csum != *private) {
5957                                 printk(KERN_ERR "btrfs csum failed ino %llu off"
5958                                       " %llu csum %u private %u\n",
5959                                       (unsigned long long)btrfs_ino(inode),
5960                                       (unsigned long long)start,
5961                                       csum, *private);
5962                                 err = -EIO;
5963                         }
5964                 }
5965
5966                 start += bvec->bv_len;
5967                 private++;
5968                 bvec++;
5969         } while (bvec <= bvec_end);
5970
5971         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5972                       dip->logical_offset + dip->bytes - 1);
5973         bio->bi_private = dip->private;
5974
5975         kfree(dip->csums);
5976         kfree(dip);
5977
5978         /* If we had a csum failure make sure to clear the uptodate flag */
5979         if (err)
5980                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5981         dio_end_io(bio, err);
5982 }
5983
5984 static void btrfs_endio_direct_write(struct bio *bio, int err)
5985 {
5986         struct btrfs_dio_private *dip = bio->bi_private;
5987         struct inode *inode = dip->inode;
5988         struct btrfs_root *root = BTRFS_I(inode)->root;
5989         struct btrfs_ordered_extent *ordered = NULL;
5990         u64 ordered_offset = dip->logical_offset;
5991         u64 ordered_bytes = dip->bytes;
5992         int ret;
5993
5994         if (err)
5995                 goto out_done;
5996 again:
5997         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5998                                                    &ordered_offset,
5999                                                    ordered_bytes, !err);
6000         if (!ret)
6001                 goto out_test;
6002
6003         ordered->work.func = finish_ordered_fn;
6004         ordered->work.flags = 0;
6005         btrfs_queue_worker(&root->fs_info->endio_write_workers,
6006                            &ordered->work);
6007 out_test:
6008         /*
6009          * our bio might span multiple ordered extents.  If we haven't
6010          * completed the accounting for the whole dio, go back and try again
6011          */
6012         if (ordered_offset < dip->logical_offset + dip->bytes) {
6013                 ordered_bytes = dip->logical_offset + dip->bytes -
6014                         ordered_offset;
6015                 ordered = NULL;
6016                 goto again;
6017         }
6018 out_done:
6019         bio->bi_private = dip->private;
6020
6021         kfree(dip);
6022
6023         /* If we had an error make sure to clear the uptodate flag */
6024         if (err)
6025                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
6026         dio_end_io(bio, err);
6027 }
6028
6029 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
6030                                     struct bio *bio, int mirror_num,
6031                                     unsigned long bio_flags, u64 offset)
6032 {
6033         int ret;
6034         struct btrfs_root *root = BTRFS_I(inode)->root;
6035         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
6036         BUG_ON(ret); /* -ENOMEM */
6037         return 0;
6038 }
6039
6040 static void btrfs_end_dio_bio(struct bio *bio, int err)
6041 {
6042         struct btrfs_dio_private *dip = bio->bi_private;
6043
6044         if (err) {
6045                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
6046                       "sector %#Lx len %u err no %d\n",
6047                       (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
6048                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
6049                 dip->errors = 1;
6050
6051                 /*
6052                  * before atomic variable goto zero, we must make sure
6053                  * dip->errors is perceived to be set.
6054                  */
6055                 smp_mb__before_atomic_dec();
6056         }
6057
6058         /* if there are more bios still pending for this dio, just exit */
6059         if (!atomic_dec_and_test(&dip->pending_bios))
6060                 goto out;
6061
6062         if (dip->errors)
6063                 bio_io_error(dip->orig_bio);
6064         else {
6065                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
6066                 bio_endio(dip->orig_bio, 0);
6067         }
6068 out:
6069         bio_put(bio);
6070 }
6071
6072 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
6073                                        u64 first_sector, gfp_t gfp_flags)
6074 {
6075         int nr_vecs = bio_get_nr_vecs(bdev);
6076         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
6077 }
6078
6079 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
6080                                          int rw, u64 file_offset, int skip_sum,
6081                                          u32 *csums, int async_submit)
6082 {
6083         int write = rw & REQ_WRITE;
6084         struct btrfs_root *root = BTRFS_I(inode)->root;
6085         int ret;
6086
6087         bio_get(bio);
6088
6089         if (!write) {
6090                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
6091                 if (ret)
6092                         goto err;
6093         }
6094
6095         if (skip_sum)
6096                 goto map;
6097
6098         if (write && async_submit) {
6099                 ret = btrfs_wq_submit_bio(root->fs_info,
6100                                    inode, rw, bio, 0, 0,
6101                                    file_offset,
6102                                    __btrfs_submit_bio_start_direct_io,
6103                                    __btrfs_submit_bio_done);
6104                 goto err;
6105         } else if (write) {
6106                 /*
6107                  * If we aren't doing async submit, calculate the csum of the
6108                  * bio now.
6109                  */
6110                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
6111                 if (ret)
6112                         goto err;
6113         } else if (!skip_sum) {
6114                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
6115                                           file_offset, csums);
6116                 if (ret)
6117                         goto err;
6118         }
6119
6120 map:
6121         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
6122 err:
6123         bio_put(bio);
6124         return ret;
6125 }
6126
6127 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
6128                                     int skip_sum)
6129 {
6130         struct inode *inode = dip->inode;
6131         struct btrfs_root *root = BTRFS_I(inode)->root;
6132         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6133         struct bio *bio;
6134         struct bio *orig_bio = dip->orig_bio;
6135         struct bio_vec *bvec = orig_bio->bi_io_vec;
6136         u64 start_sector = orig_bio->bi_sector;
6137         u64 file_offset = dip->logical_offset;
6138         u64 submit_len = 0;
6139         u64 map_length;
6140         int nr_pages = 0;
6141         u32 *csums = dip->csums;
6142         int ret = 0;
6143         int async_submit = 0;
6144         int write = rw & REQ_WRITE;
6145
6146         map_length = orig_bio->bi_size;
6147         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6148                               &map_length, NULL, 0);
6149         if (ret) {
6150                 bio_put(orig_bio);
6151                 return -EIO;
6152         }
6153
6154         if (map_length >= orig_bio->bi_size) {
6155                 bio = orig_bio;
6156                 goto submit;
6157         }
6158
6159         async_submit = 1;
6160         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
6161         if (!bio)
6162                 return -ENOMEM;
6163         bio->bi_private = dip;
6164         bio->bi_end_io = btrfs_end_dio_bio;
6165         atomic_inc(&dip->pending_bios);
6166
6167         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6168                 if (unlikely(map_length < submit_len + bvec->bv_len ||
6169                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6170                                  bvec->bv_offset) < bvec->bv_len)) {
6171                         /*
6172                          * inc the count before we submit the bio so
6173                          * we know the end IO handler won't happen before
6174                          * we inc the count. Otherwise, the dip might get freed
6175                          * before we're done setting it up
6176                          */
6177                         atomic_inc(&dip->pending_bios);
6178                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
6179                                                      file_offset, skip_sum,
6180                                                      csums, async_submit);
6181                         if (ret) {
6182                                 bio_put(bio);
6183                                 atomic_dec(&dip->pending_bios);
6184                                 goto out_err;
6185                         }
6186
6187                         /* Write's use the ordered csums */
6188                         if (!write && !skip_sum)
6189                                 csums = csums + nr_pages;
6190                         start_sector += submit_len >> 9;
6191                         file_offset += submit_len;
6192
6193                         submit_len = 0;
6194                         nr_pages = 0;
6195
6196                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6197                                                   start_sector, GFP_NOFS);
6198                         if (!bio)
6199                                 goto out_err;
6200                         bio->bi_private = dip;
6201                         bio->bi_end_io = btrfs_end_dio_bio;
6202
6203                         map_length = orig_bio->bi_size;
6204                         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6205                                               &map_length, NULL, 0);
6206                         if (ret) {
6207                                 bio_put(bio);
6208                                 goto out_err;
6209                         }
6210                 } else {
6211                         submit_len += bvec->bv_len;
6212                         nr_pages ++;
6213                         bvec++;
6214                 }
6215         }
6216
6217 submit:
6218         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6219                                      csums, async_submit);
6220         if (!ret)
6221                 return 0;
6222
6223         bio_put(bio);
6224 out_err:
6225         dip->errors = 1;
6226         /*
6227          * before atomic variable goto zero, we must
6228          * make sure dip->errors is perceived to be set.
6229          */
6230         smp_mb__before_atomic_dec();
6231         if (atomic_dec_and_test(&dip->pending_bios))
6232                 bio_io_error(dip->orig_bio);
6233
6234         /* bio_end_io() will handle error, so we needn't return it */
6235         return 0;
6236 }
6237
6238 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6239                                 loff_t file_offset)
6240 {
6241         struct btrfs_root *root = BTRFS_I(inode)->root;
6242         struct btrfs_dio_private *dip;
6243         struct bio_vec *bvec = bio->bi_io_vec;
6244         int skip_sum;
6245         int write = rw & REQ_WRITE;
6246         int ret = 0;
6247
6248         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6249
6250         dip = kmalloc(sizeof(*dip), GFP_NOFS);
6251         if (!dip) {
6252                 ret = -ENOMEM;
6253                 goto free_ordered;
6254         }
6255         dip->csums = NULL;
6256
6257         /* Write's use the ordered csum stuff, so we don't need dip->csums */
6258         if (!write && !skip_sum) {
6259                 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6260                 if (!dip->csums) {
6261                         kfree(dip);
6262                         ret = -ENOMEM;
6263                         goto free_ordered;
6264                 }
6265         }
6266
6267         dip->private = bio->bi_private;
6268         dip->inode = inode;
6269         dip->logical_offset = file_offset;
6270
6271         dip->bytes = 0;
6272         do {
6273                 dip->bytes += bvec->bv_len;
6274                 bvec++;
6275         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6276
6277         dip->disk_bytenr = (u64)bio->bi_sector << 9;
6278         bio->bi_private = dip;
6279         dip->errors = 0;
6280         dip->orig_bio = bio;
6281         atomic_set(&dip->pending_bios, 0);
6282
6283         if (write)
6284                 bio->bi_end_io = btrfs_endio_direct_write;
6285         else
6286                 bio->bi_end_io = btrfs_endio_direct_read;
6287
6288         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6289         if (!ret)
6290                 return;
6291 free_ordered:
6292         /*
6293          * If this is a write, we need to clean up the reserved space and kill
6294          * the ordered extent.
6295          */
6296         if (write) {
6297                 struct btrfs_ordered_extent *ordered;
6298                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6299                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6300                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6301                         btrfs_free_reserved_extent(root, ordered->start,
6302                                                    ordered->disk_len);
6303                 btrfs_put_ordered_extent(ordered);
6304                 btrfs_put_ordered_extent(ordered);
6305         }
6306         bio_endio(bio, ret);
6307 }
6308
6309 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6310                         const struct iovec *iov, loff_t offset,
6311                         unsigned long nr_segs)
6312 {
6313         int seg;
6314         int i;
6315         size_t size;
6316         unsigned long addr;
6317         unsigned blocksize_mask = root->sectorsize - 1;
6318         ssize_t retval = -EINVAL;
6319         loff_t end = offset;
6320
6321         if (offset & blocksize_mask)
6322                 goto out;
6323
6324         /* Check the memory alignment.  Blocks cannot straddle pages */
6325         for (seg = 0; seg < nr_segs; seg++) {
6326                 addr = (unsigned long)iov[seg].iov_base;
6327                 size = iov[seg].iov_len;
6328                 end += size;
6329                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6330                         goto out;
6331
6332                 /* If this is a write we don't need to check anymore */
6333                 if (rw & WRITE)
6334                         continue;
6335
6336                 /*
6337                  * Check to make sure we don't have duplicate iov_base's in this
6338                  * iovec, if so return EINVAL, otherwise we'll get csum errors
6339                  * when reading back.
6340                  */
6341                 for (i = seg + 1; i < nr_segs; i++) {
6342                         if (iov[seg].iov_base == iov[i].iov_base)
6343                                 goto out;
6344                 }
6345         }
6346         retval = 0;
6347 out:
6348         return retval;
6349 }
6350 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6351                         const struct iovec *iov, loff_t offset,
6352                         unsigned long nr_segs)
6353 {
6354         struct file *file = iocb->ki_filp;
6355         struct inode *inode = file->f_mapping->host;
6356         struct btrfs_ordered_extent *ordered;
6357         struct extent_state *cached_state = NULL;
6358         u64 lockstart, lockend;
6359         ssize_t ret;
6360         int writing = rw & WRITE;
6361         int write_bits = 0;
6362         size_t count = iov_length(iov, nr_segs);
6363
6364         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6365                             offset, nr_segs)) {
6366                 return 0;
6367         }
6368
6369         lockstart = offset;
6370         lockend = offset + count - 1;
6371
6372         if (writing) {
6373                 ret = btrfs_delalloc_reserve_space(inode, count);
6374                 if (ret)
6375                         goto out;
6376         }
6377
6378         while (1) {
6379                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6380                                  0, &cached_state);
6381                 /*
6382                  * We're concerned with the entire range that we're going to be
6383                  * doing DIO to, so we need to make sure theres no ordered
6384                  * extents in this range.
6385                  */
6386                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6387                                                      lockend - lockstart + 1);
6388                 if (!ordered)
6389                         break;
6390                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6391                                      &cached_state, GFP_NOFS);
6392                 btrfs_start_ordered_extent(inode, ordered, 1);
6393                 btrfs_put_ordered_extent(ordered);
6394                 cond_resched();
6395         }
6396
6397         /*
6398          * we don't use btrfs_set_extent_delalloc because we don't want
6399          * the dirty or uptodate bits
6400          */
6401         if (writing) {
6402                 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6403                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6404                                      EXTENT_DELALLOC, NULL, &cached_state,
6405                                      GFP_NOFS);
6406                 if (ret) {
6407                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6408                                          lockend, EXTENT_LOCKED | write_bits,
6409                                          1, 0, &cached_state, GFP_NOFS);
6410                         goto out;
6411                 }
6412         }
6413
6414         free_extent_state(cached_state);
6415         cached_state = NULL;
6416
6417         ret = __blockdev_direct_IO(rw, iocb, inode,
6418                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6419                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6420                    btrfs_submit_direct, 0);
6421
6422         if (ret < 0 && ret != -EIOCBQUEUED) {
6423                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6424                               offset + iov_length(iov, nr_segs) - 1,
6425                               EXTENT_LOCKED | write_bits, 1, 0,
6426                               &cached_state, GFP_NOFS);
6427         } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6428                 /*
6429                  * We're falling back to buffered, unlock the section we didn't
6430                  * do IO on.
6431                  */
6432                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6433                               offset + iov_length(iov, nr_segs) - 1,
6434                               EXTENT_LOCKED | write_bits, 1, 0,
6435                               &cached_state, GFP_NOFS);
6436         }
6437 out:
6438         free_extent_state(cached_state);
6439         return ret;
6440 }
6441
6442 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6443                 __u64 start, __u64 len)
6444 {
6445         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6446 }
6447
6448 int btrfs_readpage(struct file *file, struct page *page)
6449 {
6450         struct extent_io_tree *tree;
6451         tree = &BTRFS_I(page->mapping->host)->io_tree;
6452         return extent_read_full_page(tree, page, btrfs_get_extent, 0);
6453 }
6454
6455 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6456 {
6457         struct extent_io_tree *tree;
6458
6459
6460         if (current->flags & PF_MEMALLOC) {
6461                 redirty_page_for_writepage(wbc, page);
6462                 unlock_page(page);
6463                 return 0;
6464         }
6465         tree = &BTRFS_I(page->mapping->host)->io_tree;
6466         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6467 }
6468
6469 int btrfs_writepages(struct address_space *mapping,
6470                      struct writeback_control *wbc)
6471 {
6472         struct extent_io_tree *tree;
6473
6474         tree = &BTRFS_I(mapping->host)->io_tree;
6475         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6476 }
6477
6478 static int
6479 btrfs_readpages(struct file *file, struct address_space *mapping,
6480                 struct list_head *pages, unsigned nr_pages)
6481 {
6482         struct extent_io_tree *tree;
6483         tree = &BTRFS_I(mapping->host)->io_tree;
6484         return extent_readpages(tree, mapping, pages, nr_pages,
6485                                 btrfs_get_extent);
6486 }
6487 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6488 {
6489         struct extent_io_tree *tree;
6490         struct extent_map_tree *map;
6491         int ret;
6492
6493         tree = &BTRFS_I(page->mapping->host)->io_tree;
6494         map = &BTRFS_I(page->mapping->host)->extent_tree;
6495         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6496         if (ret == 1) {
6497                 ClearPagePrivate(page);
6498                 set_page_private(page, 0);
6499                 page_cache_release(page);
6500         }
6501         return ret;
6502 }
6503
6504 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6505 {
6506         if (PageWriteback(page) || PageDirty(page))
6507                 return 0;
6508         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6509 }
6510
6511 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6512 {
6513         struct inode *inode = page->mapping->host;
6514         struct extent_io_tree *tree;
6515         struct btrfs_ordered_extent *ordered;
6516         struct extent_state *cached_state = NULL;
6517         u64 page_start = page_offset(page);
6518         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6519
6520         /*
6521          * we have the page locked, so new writeback can't start,
6522          * and the dirty bit won't be cleared while we are here.
6523          *
6524          * Wait for IO on this page so that we can safely clear
6525          * the PagePrivate2 bit and do ordered accounting
6526          */
6527         wait_on_page_writeback(page);
6528
6529         tree = &BTRFS_I(inode)->io_tree;
6530         if (offset) {
6531                 btrfs_releasepage(page, GFP_NOFS);
6532                 return;
6533         }
6534         lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6535         ordered = btrfs_lookup_ordered_extent(inode,
6536                                            page_offset(page));
6537         if (ordered) {
6538                 /*
6539                  * IO on this page will never be started, so we need
6540                  * to account for any ordered extents now
6541                  */
6542                 clear_extent_bit(tree, page_start, page_end,
6543                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6544                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6545                                  &cached_state, GFP_NOFS);
6546                 /*
6547                  * whoever cleared the private bit is responsible
6548                  * for the finish_ordered_io
6549                  */
6550                 if (TestClearPagePrivate2(page) &&
6551                     btrfs_dec_test_ordered_pending(inode, &ordered, page_start,
6552                                                    PAGE_CACHE_SIZE, 1)) {
6553                         btrfs_finish_ordered_io(ordered);
6554                 }
6555                 btrfs_put_ordered_extent(ordered);
6556                 cached_state = NULL;
6557                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6558         }
6559         clear_extent_bit(tree, page_start, page_end,
6560                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6561                  EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6562         __btrfs_releasepage(page, GFP_NOFS);
6563
6564         ClearPageChecked(page);
6565         if (PagePrivate(page)) {
6566                 ClearPagePrivate(page);
6567                 set_page_private(page, 0);
6568                 page_cache_release(page);
6569         }
6570 }
6571
6572 /*
6573  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6574  * called from a page fault handler when a page is first dirtied. Hence we must
6575  * be careful to check for EOF conditions here. We set the page up correctly
6576  * for a written page which means we get ENOSPC checking when writing into
6577  * holes and correct delalloc and unwritten extent mapping on filesystems that
6578  * support these features.
6579  *
6580  * We are not allowed to take the i_mutex here so we have to play games to
6581  * protect against truncate races as the page could now be beyond EOF.  Because
6582  * vmtruncate() writes the inode size before removing pages, once we have the
6583  * page lock we can determine safely if the page is beyond EOF. If it is not
6584  * beyond EOF, then the page is guaranteed safe against truncation until we
6585  * unlock the page.
6586  */
6587 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6588 {
6589         struct page *page = vmf->page;
6590         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6591         struct btrfs_root *root = BTRFS_I(inode)->root;
6592         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6593         struct btrfs_ordered_extent *ordered;
6594         struct extent_state *cached_state = NULL;
6595         char *kaddr;
6596         unsigned long zero_start;
6597         loff_t size;
6598         int ret;
6599         int reserved = 0;
6600         u64 page_start;
6601         u64 page_end;
6602
6603         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6604         if (!ret) {
6605                 ret = btrfs_update_time(vma->vm_file);
6606                 reserved = 1;
6607         }
6608         if (ret) {
6609                 if (ret == -ENOMEM)
6610                         ret = VM_FAULT_OOM;
6611                 else /* -ENOSPC, -EIO, etc */
6612                         ret = VM_FAULT_SIGBUS;
6613                 if (reserved)
6614                         goto out;
6615                 goto out_noreserve;
6616         }
6617
6618         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6619 again:
6620         lock_page(page);
6621         size = i_size_read(inode);
6622         page_start = page_offset(page);
6623         page_end = page_start + PAGE_CACHE_SIZE - 1;
6624
6625         if ((page->mapping != inode->i_mapping) ||
6626             (page_start >= size)) {
6627                 /* page got truncated out from underneath us */
6628                 goto out_unlock;
6629         }
6630         wait_on_page_writeback(page);
6631
6632         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
6633         set_page_extent_mapped(page);
6634
6635         /*
6636          * we can't set the delalloc bits if there are pending ordered
6637          * extents.  Drop our locks and wait for them to finish
6638          */
6639         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6640         if (ordered) {
6641                 unlock_extent_cached(io_tree, page_start, page_end,
6642                                      &cached_state, GFP_NOFS);
6643                 unlock_page(page);
6644                 btrfs_start_ordered_extent(inode, ordered, 1);
6645                 btrfs_put_ordered_extent(ordered);
6646                 goto again;
6647         }
6648
6649         /*
6650          * XXX - page_mkwrite gets called every time the page is dirtied, even
6651          * if it was already dirty, so for space accounting reasons we need to
6652          * clear any delalloc bits for the range we are fixing to save.  There
6653          * is probably a better way to do this, but for now keep consistent with
6654          * prepare_pages in the normal write path.
6655          */
6656         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6657                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6658                           0, 0, &cached_state, GFP_NOFS);
6659
6660         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6661                                         &cached_state);
6662         if (ret) {
6663                 unlock_extent_cached(io_tree, page_start, page_end,
6664                                      &cached_state, GFP_NOFS);
6665                 ret = VM_FAULT_SIGBUS;
6666                 goto out_unlock;
6667         }
6668         ret = 0;
6669
6670         /* page is wholly or partially inside EOF */
6671         if (page_start + PAGE_CACHE_SIZE > size)
6672                 zero_start = size & ~PAGE_CACHE_MASK;
6673         else
6674                 zero_start = PAGE_CACHE_SIZE;
6675
6676         if (zero_start != PAGE_CACHE_SIZE) {
6677                 kaddr = kmap(page);
6678                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6679                 flush_dcache_page(page);
6680                 kunmap(page);
6681         }
6682         ClearPageChecked(page);
6683         set_page_dirty(page);
6684         SetPageUptodate(page);
6685
6686         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6687         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6688
6689         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6690
6691 out_unlock:
6692         if (!ret)
6693                 return VM_FAULT_LOCKED;
6694         unlock_page(page);
6695 out:
6696         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6697 out_noreserve:
6698         return ret;
6699 }
6700
6701 static int btrfs_truncate(struct inode *inode)
6702 {
6703         struct btrfs_root *root = BTRFS_I(inode)->root;
6704         struct btrfs_block_rsv *rsv;
6705         int ret;
6706         int err = 0;
6707         struct btrfs_trans_handle *trans;
6708         unsigned long nr;
6709         u64 mask = root->sectorsize - 1;
6710         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
6711
6712         ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6713         if (ret)
6714                 return ret;
6715
6716         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6717         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6718
6719         /*
6720          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
6721          * 3 things going on here
6722          *
6723          * 1) We need to reserve space for our orphan item and the space to
6724          * delete our orphan item.  Lord knows we don't want to have a dangling
6725          * orphan item because we didn't reserve space to remove it.
6726          *
6727          * 2) We need to reserve space to update our inode.
6728          *
6729          * 3) We need to have something to cache all the space that is going to
6730          * be free'd up by the truncate operation, but also have some slack
6731          * space reserved in case it uses space during the truncate (thank you
6732          * very much snapshotting).
6733          *
6734          * And we need these to all be seperate.  The fact is we can use alot of
6735          * space doing the truncate, and we have no earthly idea how much space
6736          * we will use, so we need the truncate reservation to be seperate so it
6737          * doesn't end up using space reserved for updating the inode or
6738          * removing the orphan item.  We also need to be able to stop the
6739          * transaction and start a new one, which means we need to be able to
6740          * update the inode several times, and we have no idea of knowing how
6741          * many times that will be, so we can't just reserve 1 item for the
6742          * entirety of the opration, so that has to be done seperately as well.
6743          * Then there is the orphan item, which does indeed need to be held on
6744          * to for the whole operation, and we need nobody to touch this reserved
6745          * space except the orphan code.
6746          *
6747          * So that leaves us with
6748          *
6749          * 1) root->orphan_block_rsv - for the orphan deletion.
6750          * 2) rsv - for the truncate reservation, which we will steal from the
6751          * transaction reservation.
6752          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6753          * updating the inode.
6754          */
6755         rsv = btrfs_alloc_block_rsv(root);
6756         if (!rsv)
6757                 return -ENOMEM;
6758         rsv->size = min_size;
6759
6760         /*
6761          * 1 for the truncate slack space
6762          * 1 for the orphan item we're going to add
6763          * 1 for the orphan item deletion
6764          * 1 for updating the inode.
6765          */
6766         trans = btrfs_start_transaction(root, 4);
6767         if (IS_ERR(trans)) {
6768                 err = PTR_ERR(trans);
6769                 goto out;
6770         }
6771
6772         /* Migrate the slack space for the truncate to our reserve */
6773         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6774                                       min_size);
6775         BUG_ON(ret);
6776
6777         ret = btrfs_orphan_add(trans, inode);
6778         if (ret) {
6779                 btrfs_end_transaction(trans, root);
6780                 goto out;
6781         }
6782
6783         /*
6784          * setattr is responsible for setting the ordered_data_close flag,
6785          * but that is only tested during the last file release.  That
6786          * could happen well after the next commit, leaving a great big
6787          * window where new writes may get lost if someone chooses to write
6788          * to this file after truncating to zero
6789          *
6790          * The inode doesn't have any dirty data here, and so if we commit
6791          * this is a noop.  If someone immediately starts writing to the inode
6792          * it is very likely we'll catch some of their writes in this
6793          * transaction, and the commit will find this file on the ordered
6794          * data list with good things to send down.
6795          *
6796          * This is a best effort solution, there is still a window where
6797          * using truncate to replace the contents of the file will
6798          * end up with a zero length file after a crash.
6799          */
6800         if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
6801                                            &BTRFS_I(inode)->runtime_flags))
6802                 btrfs_add_ordered_operation(trans, root, inode);
6803
6804         while (1) {
6805                 ret = btrfs_block_rsv_refill(root, rsv, min_size);
6806                 if (ret) {
6807                         /*
6808                          * This can only happen with the original transaction we
6809                          * started above, every other time we shouldn't have a
6810                          * transaction started yet.
6811                          */
6812                         if (ret == -EAGAIN)
6813                                 goto end_trans;
6814                         err = ret;
6815                         break;
6816                 }
6817
6818                 if (!trans) {
6819                         /* Just need the 1 for updating the inode */
6820                         trans = btrfs_start_transaction(root, 1);
6821                         if (IS_ERR(trans)) {
6822                                 ret = err = PTR_ERR(trans);
6823                                 trans = NULL;
6824                                 break;
6825                         }
6826                 }
6827
6828                 trans->block_rsv = rsv;
6829
6830                 ret = btrfs_truncate_inode_items(trans, root, inode,
6831                                                  inode->i_size,
6832                                                  BTRFS_EXTENT_DATA_KEY);
6833                 if (ret != -EAGAIN) {
6834                         err = ret;
6835                         break;
6836                 }
6837
6838                 trans->block_rsv = &root->fs_info->trans_block_rsv;
6839                 ret = btrfs_update_inode(trans, root, inode);
6840                 if (ret) {
6841                         err = ret;
6842                         break;
6843                 }
6844 end_trans:
6845                 nr = trans->blocks_used;
6846                 btrfs_end_transaction(trans, root);
6847                 trans = NULL;
6848                 btrfs_btree_balance_dirty(root, nr);
6849         }
6850
6851         if (ret == 0 && inode->i_nlink > 0) {
6852                 trans->block_rsv = root->orphan_block_rsv;
6853                 ret = btrfs_orphan_del(trans, inode);
6854                 if (ret)
6855                         err = ret;
6856         } else if (ret && inode->i_nlink > 0) {
6857                 /*
6858                  * Failed to do the truncate, remove us from the in memory
6859                  * orphan list.
6860                  */
6861                 ret = btrfs_orphan_del(NULL, inode);
6862         }
6863
6864         if (trans) {
6865                 trans->block_rsv = &root->fs_info->trans_block_rsv;
6866                 ret = btrfs_update_inode(trans, root, inode);
6867                 if (ret && !err)
6868                         err = ret;
6869
6870                 nr = trans->blocks_used;
6871                 ret = btrfs_end_transaction(trans, root);
6872                 btrfs_btree_balance_dirty(root, nr);
6873         }
6874
6875 out:
6876         btrfs_free_block_rsv(root, rsv);
6877
6878         if (ret && !err)
6879                 err = ret;
6880
6881         return err;
6882 }
6883
6884 /*
6885  * create a new subvolume directory/inode (helper for the ioctl).
6886  */
6887 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6888                              struct btrfs_root *new_root, u64 new_dirid)
6889 {
6890         struct inode *inode;
6891         int err;
6892         u64 index = 0;
6893
6894         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
6895                                 new_dirid, new_dirid,
6896                                 S_IFDIR | (~current_umask() & S_IRWXUGO),
6897                                 &index);
6898         if (IS_ERR(inode))
6899                 return PTR_ERR(inode);
6900         inode->i_op = &btrfs_dir_inode_operations;
6901         inode->i_fop = &btrfs_dir_file_operations;
6902
6903         set_nlink(inode, 1);
6904         btrfs_i_size_write(inode, 0);
6905
6906         err = btrfs_update_inode(trans, new_root, inode);
6907
6908         iput(inode);
6909         return err;
6910 }
6911
6912 struct inode *btrfs_alloc_inode(struct super_block *sb)
6913 {
6914         struct btrfs_inode *ei;
6915         struct inode *inode;
6916
6917         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6918         if (!ei)
6919                 return NULL;
6920
6921         ei->root = NULL;
6922         ei->space_info = NULL;
6923         ei->generation = 0;
6924         ei->last_trans = 0;
6925         ei->last_sub_trans = 0;
6926         ei->logged_trans = 0;
6927         ei->delalloc_bytes = 0;
6928         ei->disk_i_size = 0;
6929         ei->flags = 0;
6930         ei->csum_bytes = 0;
6931         ei->index_cnt = (u64)-1;
6932         ei->last_unlink_trans = 0;
6933
6934         spin_lock_init(&ei->lock);
6935         ei->outstanding_extents = 0;
6936         ei->reserved_extents = 0;
6937
6938         ei->runtime_flags = 0;
6939         ei->force_compress = BTRFS_COMPRESS_NONE;
6940
6941         ei->delayed_node = NULL;
6942
6943         inode = &ei->vfs_inode;
6944         extent_map_tree_init(&ei->extent_tree);
6945         extent_io_tree_init(&ei->io_tree, &inode->i_data);
6946         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
6947         ei->io_tree.track_uptodate = 1;
6948         ei->io_failure_tree.track_uptodate = 1;
6949         mutex_init(&ei->log_mutex);
6950         mutex_init(&ei->delalloc_mutex);
6951         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6952         INIT_LIST_HEAD(&ei->delalloc_inodes);
6953         INIT_LIST_HEAD(&ei->ordered_operations);
6954         RB_CLEAR_NODE(&ei->rb_node);
6955
6956         return inode;
6957 }
6958
6959 static void btrfs_i_callback(struct rcu_head *head)
6960 {
6961         struct inode *inode = container_of(head, struct inode, i_rcu);
6962         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6963 }
6964
6965 void btrfs_destroy_inode(struct inode *inode)
6966 {
6967         struct btrfs_ordered_extent *ordered;
6968         struct btrfs_root *root = BTRFS_I(inode)->root;
6969
6970         WARN_ON(!list_empty(&inode->i_dentry));
6971         WARN_ON(inode->i_data.nrpages);
6972         WARN_ON(BTRFS_I(inode)->outstanding_extents);
6973         WARN_ON(BTRFS_I(inode)->reserved_extents);
6974         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
6975         WARN_ON(BTRFS_I(inode)->csum_bytes);
6976
6977         /*
6978          * This can happen where we create an inode, but somebody else also
6979          * created the same inode and we need to destroy the one we already
6980          * created.
6981          */
6982         if (!root)
6983                 goto free;
6984
6985         /*
6986          * Make sure we're properly removed from the ordered operation
6987          * lists.
6988          */
6989         smp_mb();
6990         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6991                 spin_lock(&root->fs_info->ordered_extent_lock);
6992                 list_del_init(&BTRFS_I(inode)->ordered_operations);
6993                 spin_unlock(&root->fs_info->ordered_extent_lock);
6994         }
6995
6996         if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
6997                      &BTRFS_I(inode)->runtime_flags)) {
6998                 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
6999                        (unsigned long long)btrfs_ino(inode));
7000                 atomic_dec(&root->orphan_inodes);
7001         }
7002
7003         while (1) {
7004                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7005                 if (!ordered)
7006                         break;
7007                 else {
7008                         printk(KERN_ERR "btrfs found ordered "
7009                                "extent %llu %llu on inode cleanup\n",
7010                                (unsigned long long)ordered->file_offset,
7011                                (unsigned long long)ordered->len);
7012                         btrfs_remove_ordered_extent(inode, ordered);
7013                         btrfs_put_ordered_extent(ordered);
7014                         btrfs_put_ordered_extent(ordered);
7015                 }
7016         }
7017         inode_tree_del(inode);
7018         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
7019 free:
7020         btrfs_remove_delayed_node(inode);
7021         call_rcu(&inode->i_rcu, btrfs_i_callback);
7022 }
7023
7024 int btrfs_drop_inode(struct inode *inode)
7025 {
7026         struct btrfs_root *root = BTRFS_I(inode)->root;
7027
7028         if (btrfs_root_refs(&root->root_item) == 0 &&
7029             !btrfs_is_free_space_inode(root, inode))
7030                 return 1;
7031         else
7032                 return generic_drop_inode(inode);
7033 }
7034
7035 static void init_once(void *foo)
7036 {
7037         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
7038
7039         inode_init_once(&ei->vfs_inode);
7040 }
7041
7042 void btrfs_destroy_cachep(void)
7043 {
7044         if (btrfs_inode_cachep)
7045                 kmem_cache_destroy(btrfs_inode_cachep);
7046         if (btrfs_trans_handle_cachep)
7047                 kmem_cache_destroy(btrfs_trans_handle_cachep);
7048         if (btrfs_transaction_cachep)
7049                 kmem_cache_destroy(btrfs_transaction_cachep);
7050         if (btrfs_path_cachep)
7051                 kmem_cache_destroy(btrfs_path_cachep);
7052         if (btrfs_free_space_cachep)
7053                 kmem_cache_destroy(btrfs_free_space_cachep);
7054 }
7055
7056 int btrfs_init_cachep(void)
7057 {
7058         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
7059                         sizeof(struct btrfs_inode), 0,
7060                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
7061         if (!btrfs_inode_cachep)
7062                 goto fail;
7063
7064         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
7065                         sizeof(struct btrfs_trans_handle), 0,
7066                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7067         if (!btrfs_trans_handle_cachep)
7068                 goto fail;
7069
7070         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
7071                         sizeof(struct btrfs_transaction), 0,
7072                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7073         if (!btrfs_transaction_cachep)
7074                 goto fail;
7075
7076         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
7077                         sizeof(struct btrfs_path), 0,
7078                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7079         if (!btrfs_path_cachep)
7080                 goto fail;
7081
7082         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
7083                         sizeof(struct btrfs_free_space), 0,
7084                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7085         if (!btrfs_free_space_cachep)
7086                 goto fail;
7087
7088         return 0;
7089 fail:
7090         btrfs_destroy_cachep();
7091         return -ENOMEM;
7092 }
7093
7094 static int btrfs_getattr(struct vfsmount *mnt,
7095                          struct dentry *dentry, struct kstat *stat)
7096 {
7097         struct inode *inode = dentry->d_inode;
7098         u32 blocksize = inode->i_sb->s_blocksize;
7099
7100         generic_fillattr(inode, stat);
7101         stat->dev = BTRFS_I(inode)->root->anon_dev;
7102         stat->blksize = PAGE_CACHE_SIZE;
7103         stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
7104                 ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9;
7105         return 0;
7106 }
7107
7108 /*
7109  * If a file is moved, it will inherit the cow and compression flags of the new
7110  * directory.
7111  */
7112 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
7113 {
7114         struct btrfs_inode *b_dir = BTRFS_I(dir);
7115         struct btrfs_inode *b_inode = BTRFS_I(inode);
7116
7117         if (b_dir->flags & BTRFS_INODE_NODATACOW)
7118                 b_inode->flags |= BTRFS_INODE_NODATACOW;
7119         else
7120                 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
7121
7122         if (b_dir->flags & BTRFS_INODE_COMPRESS)
7123                 b_inode->flags |= BTRFS_INODE_COMPRESS;
7124         else
7125                 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
7126 }
7127
7128 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
7129                            struct inode *new_dir, struct dentry *new_dentry)
7130 {
7131         struct btrfs_trans_handle *trans;
7132         struct btrfs_root *root = BTRFS_I(old_dir)->root;
7133         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
7134         struct inode *new_inode = new_dentry->d_inode;
7135         struct inode *old_inode = old_dentry->d_inode;
7136         struct timespec ctime = CURRENT_TIME;
7137         u64 index = 0;
7138         u64 root_objectid;
7139         int ret;
7140         u64 old_ino = btrfs_ino(old_inode);
7141
7142         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
7143                 return -EPERM;
7144
7145         /* we only allow rename subvolume link between subvolumes */
7146         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
7147                 return -EXDEV;
7148
7149         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
7150             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
7151                 return -ENOTEMPTY;
7152
7153         if (S_ISDIR(old_inode->i_mode) && new_inode &&
7154             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
7155                 return -ENOTEMPTY;
7156         /*
7157          * we're using rename to replace one file with another.
7158          * and the replacement file is large.  Start IO on it now so
7159          * we don't add too much work to the end of the transaction
7160          */
7161         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
7162             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
7163                 filemap_flush(old_inode->i_mapping);
7164
7165         /* close the racy window with snapshot create/destroy ioctl */
7166         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7167                 down_read(&root->fs_info->subvol_sem);
7168         /*
7169          * We want to reserve the absolute worst case amount of items.  So if
7170          * both inodes are subvols and we need to unlink them then that would
7171          * require 4 item modifications, but if they are both normal inodes it
7172          * would require 5 item modifications, so we'll assume their normal
7173          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7174          * should cover the worst case number of items we'll modify.
7175          */
7176         trans = btrfs_start_transaction(root, 20);
7177         if (IS_ERR(trans)) {
7178                 ret = PTR_ERR(trans);
7179                 goto out_notrans;
7180         }
7181
7182         if (dest != root)
7183                 btrfs_record_root_in_trans(trans, dest);
7184
7185         ret = btrfs_set_inode_index(new_dir, &index);
7186         if (ret)
7187                 goto out_fail;
7188
7189         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7190                 /* force full log commit if subvolume involved. */
7191                 root->fs_info->last_trans_log_full_commit = trans->transid;
7192         } else {
7193                 ret = btrfs_insert_inode_ref(trans, dest,
7194                                              new_dentry->d_name.name,
7195                                              new_dentry->d_name.len,
7196                                              old_ino,
7197                                              btrfs_ino(new_dir), index);
7198                 if (ret)
7199                         goto out_fail;
7200                 /*
7201                  * this is an ugly little race, but the rename is required
7202                  * to make sure that if we crash, the inode is either at the
7203                  * old name or the new one.  pinning the log transaction lets
7204                  * us make sure we don't allow a log commit to come in after
7205                  * we unlink the name but before we add the new name back in.
7206                  */
7207                 btrfs_pin_log_trans(root);
7208         }
7209         /*
7210          * make sure the inode gets flushed if it is replacing
7211          * something.
7212          */
7213         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7214                 btrfs_add_ordered_operation(trans, root, old_inode);
7215
7216         inode_inc_iversion(old_dir);
7217         inode_inc_iversion(new_dir);
7218         inode_inc_iversion(old_inode);
7219         old_dir->i_ctime = old_dir->i_mtime = ctime;
7220         new_dir->i_ctime = new_dir->i_mtime = ctime;
7221         old_inode->i_ctime = ctime;
7222
7223         if (old_dentry->d_parent != new_dentry->d_parent)
7224                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7225
7226         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7227                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7228                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7229                                         old_dentry->d_name.name,
7230                                         old_dentry->d_name.len);
7231         } else {
7232                 ret = __btrfs_unlink_inode(trans, root, old_dir,
7233                                         old_dentry->d_inode,
7234                                         old_dentry->d_name.name,
7235                                         old_dentry->d_name.len);
7236                 if (!ret)
7237                         ret = btrfs_update_inode(trans, root, old_inode);
7238         }
7239         if (ret) {
7240                 btrfs_abort_transaction(trans, root, ret);
7241                 goto out_fail;
7242         }
7243
7244         if (new_inode) {
7245                 inode_inc_iversion(new_inode);
7246                 new_inode->i_ctime = CURRENT_TIME;
7247                 if (unlikely(btrfs_ino(new_inode) ==
7248                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7249                         root_objectid = BTRFS_I(new_inode)->location.objectid;
7250                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
7251                                                 root_objectid,
7252                                                 new_dentry->d_name.name,
7253                                                 new_dentry->d_name.len);
7254                         BUG_ON(new_inode->i_nlink == 0);
7255                 } else {
7256                         ret = btrfs_unlink_inode(trans, dest, new_dir,
7257                                                  new_dentry->d_inode,
7258                                                  new_dentry->d_name.name,
7259                                                  new_dentry->d_name.len);
7260                 }
7261                 if (!ret && new_inode->i_nlink == 0) {
7262                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7263                         BUG_ON(ret);
7264                 }
7265                 if (ret) {
7266                         btrfs_abort_transaction(trans, root, ret);
7267                         goto out_fail;
7268                 }
7269         }
7270
7271         fixup_inode_flags(new_dir, old_inode);
7272
7273         ret = btrfs_add_link(trans, new_dir, old_inode,
7274                              new_dentry->d_name.name,
7275                              new_dentry->d_name.len, 0, index);
7276         if (ret) {
7277                 btrfs_abort_transaction(trans, root, ret);
7278                 goto out_fail;
7279         }
7280
7281         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7282                 struct dentry *parent = new_dentry->d_parent;
7283                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7284                 btrfs_end_log_trans(root);
7285         }
7286 out_fail:
7287         btrfs_end_transaction(trans, root);
7288 out_notrans:
7289         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7290                 up_read(&root->fs_info->subvol_sem);
7291
7292         return ret;
7293 }
7294
7295 /*
7296  * some fairly slow code that needs optimization. This walks the list
7297  * of all the inodes with pending delalloc and forces them to disk.
7298  */
7299 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7300 {
7301         struct list_head *head = &root->fs_info->delalloc_inodes;
7302         struct btrfs_inode *binode;
7303         struct inode *inode;
7304
7305         if (root->fs_info->sb->s_flags & MS_RDONLY)
7306                 return -EROFS;
7307
7308         spin_lock(&root->fs_info->delalloc_lock);
7309         while (!list_empty(head)) {
7310                 binode = list_entry(head->next, struct btrfs_inode,
7311                                     delalloc_inodes);
7312                 inode = igrab(&binode->vfs_inode);
7313                 if (!inode)
7314                         list_del_init(&binode->delalloc_inodes);
7315                 spin_unlock(&root->fs_info->delalloc_lock);
7316                 if (inode) {
7317                         filemap_flush(inode->i_mapping);
7318                         if (delay_iput)
7319                                 btrfs_add_delayed_iput(inode);
7320                         else
7321                                 iput(inode);
7322                 }
7323                 cond_resched();
7324                 spin_lock(&root->fs_info->delalloc_lock);
7325         }
7326         spin_unlock(&root->fs_info->delalloc_lock);
7327
7328         /* the filemap_flush will queue IO into the worker threads, but
7329          * we have to make sure the IO is actually started and that
7330          * ordered extents get created before we return
7331          */
7332         atomic_inc(&root->fs_info->async_submit_draining);
7333         while (atomic_read(&root->fs_info->nr_async_submits) ||
7334               atomic_read(&root->fs_info->async_delalloc_pages)) {
7335                 wait_event(root->fs_info->async_submit_wait,
7336                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7337                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7338         }
7339         atomic_dec(&root->fs_info->async_submit_draining);
7340         return 0;
7341 }
7342
7343 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7344                          const char *symname)
7345 {
7346         struct btrfs_trans_handle *trans;
7347         struct btrfs_root *root = BTRFS_I(dir)->root;
7348         struct btrfs_path *path;
7349         struct btrfs_key key;
7350         struct inode *inode = NULL;
7351         int err;
7352         int drop_inode = 0;
7353         u64 objectid;
7354         u64 index = 0 ;
7355         int name_len;
7356         int datasize;
7357         unsigned long ptr;
7358         struct btrfs_file_extent_item *ei;
7359         struct extent_buffer *leaf;
7360         unsigned long nr = 0;
7361
7362         name_len = strlen(symname) + 1;
7363         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7364                 return -ENAMETOOLONG;
7365
7366         /*
7367          * 2 items for inode item and ref
7368          * 2 items for dir items
7369          * 1 item for xattr if selinux is on
7370          */
7371         trans = btrfs_start_transaction(root, 5);
7372         if (IS_ERR(trans))
7373                 return PTR_ERR(trans);
7374
7375         err = btrfs_find_free_ino(root, &objectid);
7376         if (err)
7377                 goto out_unlock;
7378
7379         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7380                                 dentry->d_name.len, btrfs_ino(dir), objectid,
7381                                 S_IFLNK|S_IRWXUGO, &index);
7382         if (IS_ERR(inode)) {
7383                 err = PTR_ERR(inode);
7384                 goto out_unlock;
7385         }
7386
7387         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7388         if (err) {
7389                 drop_inode = 1;
7390                 goto out_unlock;
7391         }
7392
7393         /*
7394         * If the active LSM wants to access the inode during
7395         * d_instantiate it needs these. Smack checks to see
7396         * if the filesystem supports xattrs by looking at the
7397         * ops vector.
7398         */
7399         inode->i_fop = &btrfs_file_operations;
7400         inode->i_op = &btrfs_file_inode_operations;
7401
7402         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7403         if (err)
7404                 drop_inode = 1;
7405         else {
7406                 inode->i_mapping->a_ops = &btrfs_aops;
7407                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7408                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7409         }
7410         if (drop_inode)
7411                 goto out_unlock;
7412
7413         path = btrfs_alloc_path();
7414         if (!path) {
7415                 err = -ENOMEM;
7416                 drop_inode = 1;
7417                 goto out_unlock;
7418         }
7419         key.objectid = btrfs_ino(inode);
7420         key.offset = 0;
7421         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7422         datasize = btrfs_file_extent_calc_inline_size(name_len);
7423         err = btrfs_insert_empty_item(trans, root, path, &key,
7424                                       datasize);
7425         if (err) {
7426                 drop_inode = 1;
7427                 btrfs_free_path(path);
7428                 goto out_unlock;
7429         }
7430         leaf = path->nodes[0];
7431         ei = btrfs_item_ptr(leaf, path->slots[0],
7432                             struct btrfs_file_extent_item);
7433         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7434         btrfs_set_file_extent_type(leaf, ei,
7435                                    BTRFS_FILE_EXTENT_INLINE);
7436         btrfs_set_file_extent_encryption(leaf, ei, 0);
7437         btrfs_set_file_extent_compression(leaf, ei, 0);
7438         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7439         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7440
7441         ptr = btrfs_file_extent_inline_start(ei);
7442         write_extent_buffer(leaf, symname, ptr, name_len);
7443         btrfs_mark_buffer_dirty(leaf);
7444         btrfs_free_path(path);
7445
7446         inode->i_op = &btrfs_symlink_inode_operations;
7447         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7448         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7449         inode_set_bytes(inode, name_len);
7450         btrfs_i_size_write(inode, name_len - 1);
7451         err = btrfs_update_inode(trans, root, inode);
7452         if (err)
7453                 drop_inode = 1;
7454
7455 out_unlock:
7456         if (!err)
7457                 d_instantiate(dentry, inode);
7458         nr = trans->blocks_used;
7459         btrfs_end_transaction(trans, root);
7460         if (drop_inode) {
7461                 inode_dec_link_count(inode);
7462                 iput(inode);
7463         }
7464         btrfs_btree_balance_dirty(root, nr);
7465         return err;
7466 }
7467
7468 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7469                                        u64 start, u64 num_bytes, u64 min_size,
7470                                        loff_t actual_len, u64 *alloc_hint,
7471                                        struct btrfs_trans_handle *trans)
7472 {
7473         struct btrfs_root *root = BTRFS_I(inode)->root;
7474         struct btrfs_key ins;
7475         u64 cur_offset = start;
7476         u64 i_size;
7477         int ret = 0;
7478         bool own_trans = true;
7479
7480         if (trans)
7481                 own_trans = false;
7482         while (num_bytes > 0) {
7483                 if (own_trans) {
7484                         trans = btrfs_start_transaction(root, 3);
7485                         if (IS_ERR(trans)) {
7486                                 ret = PTR_ERR(trans);
7487                                 break;
7488                         }
7489                 }
7490
7491                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7492                                            0, *alloc_hint, &ins, 1);
7493                 if (ret) {
7494                         if (own_trans)
7495                                 btrfs_end_transaction(trans, root);
7496                         break;
7497                 }
7498
7499                 ret = insert_reserved_file_extent(trans, inode,
7500                                                   cur_offset, ins.objectid,
7501                                                   ins.offset, ins.offset,
7502                                                   ins.offset, 0, 0, 0,
7503                                                   BTRFS_FILE_EXTENT_PREALLOC);
7504                 if (ret) {
7505                         btrfs_abort_transaction(trans, root, ret);
7506                         if (own_trans)
7507                                 btrfs_end_transaction(trans, root);
7508                         break;
7509                 }
7510                 btrfs_drop_extent_cache(inode, cur_offset,
7511                                         cur_offset + ins.offset -1, 0);
7512
7513                 num_bytes -= ins.offset;
7514                 cur_offset += ins.offset;
7515                 *alloc_hint = ins.objectid + ins.offset;
7516
7517                 inode_inc_iversion(inode);
7518                 inode->i_ctime = CURRENT_TIME;
7519                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7520                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7521                     (actual_len > inode->i_size) &&
7522                     (cur_offset > inode->i_size)) {
7523                         if (cur_offset > actual_len)
7524                                 i_size = actual_len;
7525                         else
7526                                 i_size = cur_offset;
7527                         i_size_write(inode, i_size);
7528                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7529                 }
7530
7531                 ret = btrfs_update_inode(trans, root, inode);
7532
7533                 if (ret) {
7534                         btrfs_abort_transaction(trans, root, ret);
7535                         if (own_trans)
7536                                 btrfs_end_transaction(trans, root);
7537                         break;
7538                 }
7539
7540                 if (own_trans)
7541                         btrfs_end_transaction(trans, root);
7542         }
7543         return ret;
7544 }
7545
7546 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7547                               u64 start, u64 num_bytes, u64 min_size,
7548                               loff_t actual_len, u64 *alloc_hint)
7549 {
7550         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7551                                            min_size, actual_len, alloc_hint,
7552                                            NULL);
7553 }
7554
7555 int btrfs_prealloc_file_range_trans(struct inode *inode,
7556                                     struct btrfs_trans_handle *trans, int mode,
7557                                     u64 start, u64 num_bytes, u64 min_size,
7558                                     loff_t actual_len, u64 *alloc_hint)
7559 {
7560         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7561                                            min_size, actual_len, alloc_hint, trans);
7562 }
7563
7564 static int btrfs_set_page_dirty(struct page *page)
7565 {
7566         return __set_page_dirty_nobuffers(page);
7567 }
7568
7569 static int btrfs_permission(struct inode *inode, int mask)
7570 {
7571         struct btrfs_root *root = BTRFS_I(inode)->root;
7572         umode_t mode = inode->i_mode;
7573
7574         if (mask & MAY_WRITE &&
7575             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7576                 if (btrfs_root_readonly(root))
7577                         return -EROFS;
7578                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7579                         return -EACCES;
7580         }
7581         return generic_permission(inode, mask);
7582 }
7583
7584 static const struct inode_operations btrfs_dir_inode_operations = {
7585         .getattr        = btrfs_getattr,
7586         .lookup         = btrfs_lookup,
7587         .create         = btrfs_create,
7588         .unlink         = btrfs_unlink,
7589         .link           = btrfs_link,
7590         .mkdir          = btrfs_mkdir,
7591         .rmdir          = btrfs_rmdir,
7592         .rename         = btrfs_rename,
7593         .symlink        = btrfs_symlink,
7594         .setattr        = btrfs_setattr,
7595         .mknod          = btrfs_mknod,
7596         .setxattr       = btrfs_setxattr,
7597         .getxattr       = btrfs_getxattr,
7598         .listxattr      = btrfs_listxattr,
7599         .removexattr    = btrfs_removexattr,
7600         .permission     = btrfs_permission,
7601         .get_acl        = btrfs_get_acl,
7602 };
7603 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7604         .lookup         = btrfs_lookup,
7605         .permission     = btrfs_permission,
7606         .get_acl        = btrfs_get_acl,
7607 };
7608
7609 static const struct file_operations btrfs_dir_file_operations = {
7610         .llseek         = generic_file_llseek,
7611         .read           = generic_read_dir,
7612         .readdir        = btrfs_real_readdir,
7613         .unlocked_ioctl = btrfs_ioctl,
7614 #ifdef CONFIG_COMPAT
7615         .compat_ioctl   = btrfs_ioctl,
7616 #endif
7617         .release        = btrfs_release_file,
7618         .fsync          = btrfs_sync_file,
7619 };
7620
7621 static struct extent_io_ops btrfs_extent_io_ops = {
7622         .fill_delalloc = run_delalloc_range,
7623         .submit_bio_hook = btrfs_submit_bio_hook,
7624         .merge_bio_hook = btrfs_merge_bio_hook,
7625         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7626         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7627         .writepage_start_hook = btrfs_writepage_start_hook,
7628         .set_bit_hook = btrfs_set_bit_hook,
7629         .clear_bit_hook = btrfs_clear_bit_hook,
7630         .merge_extent_hook = btrfs_merge_extent_hook,
7631         .split_extent_hook = btrfs_split_extent_hook,
7632 };
7633
7634 /*
7635  * btrfs doesn't support the bmap operation because swapfiles
7636  * use bmap to make a mapping of extents in the file.  They assume
7637  * these extents won't change over the life of the file and they
7638  * use the bmap result to do IO directly to the drive.
7639  *
7640  * the btrfs bmap call would return logical addresses that aren't
7641  * suitable for IO and they also will change frequently as COW
7642  * operations happen.  So, swapfile + btrfs == corruption.
7643  *
7644  * For now we're avoiding this by dropping bmap.
7645  */
7646 static const struct address_space_operations btrfs_aops = {
7647         .readpage       = btrfs_readpage,
7648         .writepage      = btrfs_writepage,
7649         .writepages     = btrfs_writepages,
7650         .readpages      = btrfs_readpages,
7651         .direct_IO      = btrfs_direct_IO,
7652         .invalidatepage = btrfs_invalidatepage,
7653         .releasepage    = btrfs_releasepage,
7654         .set_page_dirty = btrfs_set_page_dirty,
7655         .error_remove_page = generic_error_remove_page,
7656 };
7657
7658 static const struct address_space_operations btrfs_symlink_aops = {
7659         .readpage       = btrfs_readpage,
7660         .writepage      = btrfs_writepage,
7661         .invalidatepage = btrfs_invalidatepage,
7662         .releasepage    = btrfs_releasepage,
7663 };
7664
7665 static const struct inode_operations btrfs_file_inode_operations = {
7666         .getattr        = btrfs_getattr,
7667         .setattr        = btrfs_setattr,
7668         .setxattr       = btrfs_setxattr,
7669         .getxattr       = btrfs_getxattr,
7670         .listxattr      = btrfs_listxattr,
7671         .removexattr    = btrfs_removexattr,
7672         .permission     = btrfs_permission,
7673         .fiemap         = btrfs_fiemap,
7674         .get_acl        = btrfs_get_acl,
7675 };
7676 static const struct inode_operations btrfs_special_inode_operations = {
7677         .getattr        = btrfs_getattr,
7678         .setattr        = btrfs_setattr,
7679         .permission     = btrfs_permission,
7680         .setxattr       = btrfs_setxattr,
7681         .getxattr       = btrfs_getxattr,
7682         .listxattr      = btrfs_listxattr,
7683         .removexattr    = btrfs_removexattr,
7684         .get_acl        = btrfs_get_acl,
7685 };
7686 static const struct inode_operations btrfs_symlink_inode_operations = {
7687         .readlink       = generic_readlink,
7688         .follow_link    = page_follow_link_light,
7689         .put_link       = page_put_link,
7690         .getattr        = btrfs_getattr,
7691         .setattr        = btrfs_setattr,
7692         .permission     = btrfs_permission,
7693         .setxattr       = btrfs_setxattr,
7694         .getxattr       = btrfs_getxattr,
7695         .listxattr      = btrfs_listxattr,
7696         .removexattr    = btrfs_removexattr,
7697         .get_acl        = btrfs_get_acl,
7698 };
7699
7700 const struct dentry_operations btrfs_dentry_operations = {
7701         .d_delete       = btrfs_dentry_delete,
7702         .d_release      = btrfs_dentry_release,
7703 };