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 "compat.h"
  42 #include "ctree.h"
  43 #include "disk-io.h"
  44 #include "transaction.h"
  45 #include "btrfs_inode.h"
  46 #include "ioctl.h"
  47 #include "print-tree.h"
  48 #include "volumes.h"
  49 #include "ordered-data.h"
  50 #include "xattr.h"
  51 #include "tree-log.h"
  52 #include "compression.h"
  53 #include "locking.h"
  54 #include "free-space-cache.h"
  55 #include "inode-map.h"
  56
  57 struct btrfs_iget_args {
  58         u64 ino;
  59         struct btrfs_root *root;
  60 };
  61
  62 static const struct inode_operations btrfs_dir_inode_operations;
  63 static const struct inode_operations btrfs_symlink_inode_operations;
  64 static const struct inode_operations btrfs_dir_ro_inode_operations;
  65 static const struct inode_operations btrfs_special_inode_operations;
  66 static const struct inode_operations btrfs_file_inode_operations;
  67 static const struct address_space_operations btrfs_aops;
  68 static const struct address_space_operations btrfs_symlink_aops;
  69 static const struct file_operations btrfs_dir_file_operations;
  70 static struct extent_io_ops btrfs_extent_io_ops;
  71
  72 static struct kmem_cache *btrfs_inode_cachep;
  73 struct kmem_cache *btrfs_trans_handle_cachep;
  74 struct kmem_cache *btrfs_transaction_cachep;
  75 struct kmem_cache *btrfs_path_cachep;
  76 struct kmem_cache *btrfs_free_space_cachep;
  77
  78 #define S_SHIFT 12
  79 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  80         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  81         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  82         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  83         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  84         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  85         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  86         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  87 };
  88
  89 static int btrfs_setsize(struct inode *inode, loff_t newsize);
  90 static int btrfs_truncate(struct inode *inode);
  91 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
  92 static noinline int cow_file_range(struct inode *inode,
  93                                    struct page *locked_page,
  94                                    u64 start, u64 end, int *page_started,
  95                                    unsigned long *nr_written, int unlock);
  96
  97 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
  98                                      struct inode *inode,  struct inode *dir,
  99                                      const struct qstr *qstr)
 100 {
 101         int err;
 102
 103         err = btrfs_init_acl(trans, inode, dir);
 104         if (!err)
 105                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 106         return err;
 107 }
 108
 109 /*
 110  * this does all the hard work for inserting an inline extent into
 111  * the btree.  The caller should have done a btrfs_drop_extents so that
 112  * no overlapping inline items exist in the btree
 113  */
 114 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 115                                 struct btrfs_root *root, struct inode *inode,
 116                                 u64 start, size_t size, size_t compressed_size,
 117                                 int compress_type,
 118                                 struct page **compressed_pages)
 119 {
 120         struct btrfs_key key;
 121         struct btrfs_path *path;
 122         struct extent_buffer *leaf;
 123         struct page *page = NULL;
 124         char *kaddr;
 125         unsigned long ptr;
 126         struct btrfs_file_extent_item *ei;
 127         int err = 0;
 128         int ret;
 129         size_t cur_size = size;
 130         size_t datasize;
 131         unsigned long offset;
 132
 133         if (compressed_size && compressed_pages)
 134                 cur_size = compressed_size;
 135
 136         path = btrfs_alloc_path();
 137         if (!path)
 138                 return -ENOMEM;
 139
 140         path->leave_spinning = 1;
 141
 142         key.objectid = btrfs_ino(inode);
 143         key.offset = start;
 144         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 145         datasize = btrfs_file_extent_calc_inline_size(cur_size);
 146
 147         inode_add_bytes(inode, size);
 148         ret = btrfs_insert_empty_item(trans, root, path, &key,
 149                                       datasize);
 150         BUG_ON(ret);
 151         if (ret) {
 152                 err = ret;
 153                 goto fail;
 154         }
 155         leaf = path->nodes[0];
 156         ei = btrfs_item_ptr(leaf, path->slots[0],
 157                             struct btrfs_file_extent_item);
 158         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 159         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 160         btrfs_set_file_extent_encryption(leaf, ei, 0);
 161         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 162         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 163         ptr = btrfs_file_extent_inline_start(ei);
 164
 165         if (compress_type != BTRFS_COMPRESS_NONE) {
 166                 struct page *cpage;
 167                 int i = 0;
 168                 while (compressed_size > 0) {
 169                         cpage = compressed_pages[i];
 170                         cur_size = min_t(unsigned long, compressed_size,
 171                                        PAGE_CACHE_SIZE);
 172
 173                         kaddr = kmap_atomic(cpage, KM_USER0);
 174                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 175                         kunmap_atomic(kaddr, KM_USER0);
 176
 177                         i++;
 178                         ptr += cur_size;
 179                         compressed_size -= cur_size;
 180                 }
 181                 btrfs_set_file_extent_compression(leaf, ei,
 182                                                   compress_type);
 183         } else {
 184                 page = find_get_page(inode->i_mapping,
 185                                      start >> PAGE_CACHE_SHIFT);
 186                 btrfs_set_file_extent_compression(leaf, ei, 0);
 187                 kaddr = kmap_atomic(page, KM_USER0);
 188                 offset = start & (PAGE_CACHE_SIZE - 1);
 189                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 190                 kunmap_atomic(kaddr, KM_USER0);
 191                 page_cache_release(page);
 192         }
 193         btrfs_mark_buffer_dirty(leaf);
 194         btrfs_free_path(path);
 195
 196         /*
 197          * we're an inline extent, so nobody can
 198          * extend the file past i_size without locking
 199          * a page we already have locked.
 200          *
 201          * We must do any isize and inode updates
 202          * before we unlock the pages.  Otherwise we
 203          * could end up racing with unlink.
 204          */
 205         BTRFS_I(inode)->disk_i_size = inode->i_size;
 206         btrfs_update_inode(trans, root, inode);
 207
 208         return 0;
 209 fail:
 210         btrfs_free_path(path);
 211         return err;
 212 }
 213
 214
 215 /*
 216  * conditionally insert an inline extent into the file.  This
 217  * does the checks required to make sure the data is small enough
 218  * to fit as an inline extent.
 219  */
 220 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 221                                  struct btrfs_root *root,
 222                                  struct inode *inode, u64 start, u64 end,
 223                                  size_t compressed_size, int compress_type,
 224                                  struct page **compressed_pages)
 225 {
 226         u64 isize = i_size_read(inode);
 227         u64 actual_end = min(end + 1, isize);
 228         u64 inline_len = actual_end - start;
 229         u64 aligned_end = (end + root->sectorsize - 1) &
 230                         ~((u64)root->sectorsize - 1);
 231         u64 hint_byte;
 232         u64 data_len = inline_len;
 233         int ret;
 234
 235         if (compressed_size)
 236                 data_len = compressed_size;
 237
 238         if (start > 0 ||
 239             actual_end >= PAGE_CACHE_SIZE ||
 240             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 241             (!compressed_size &&
 242             (actual_end & (root->sectorsize - 1)) == 0) ||
 243             end + 1 < isize ||
 244             data_len > root->fs_info->max_inline) {
 245                 return 1;
 246         }
 247
 248         ret = btrfs_drop_extents(trans, inode, start, aligned_end,
 249                                  &hint_byte, 1);
 250         BUG_ON(ret);
 251
 252         if (isize > actual_end)
 253                 inline_len = min_t(u64, isize, actual_end);
 254         ret = insert_inline_extent(trans, root, inode, start,
 255                                    inline_len, compressed_size,
 256                                    compress_type, compressed_pages);
 257         BUG_ON(ret);
 258         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 259         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 260         return 0;
 261 }
 262
 263 struct async_extent {
 264         u64 start;
 265         u64 ram_size;
 266         u64 compressed_size;
 267         struct page **pages;
 268         unsigned long nr_pages;
 269         int compress_type;
 270         struct list_head list;
 271 };
 272
 273 struct async_cow {
 274         struct inode *inode;
 275         struct btrfs_root *root;
 276         struct page *locked_page;
 277         u64 start;
 278         u64 end;
 279         struct list_head extents;
 280         struct btrfs_work work;
 281 };
 282
 283 static noinline int add_async_extent(struct async_cow *cow,
 284                                      u64 start, u64 ram_size,
 285                                      u64 compressed_size,
 286                                      struct page **pages,
 287                                      unsigned long nr_pages,
 288                                      int compress_type)
 289 {
 290         struct async_extent *async_extent;
 291
 292         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 293         BUG_ON(!async_extent);
 294         async_extent->start = start;
 295         async_extent->ram_size = ram_size;
 296         async_extent->compressed_size = compressed_size;
 297         async_extent->pages = pages;
 298         async_extent->nr_pages = nr_pages;
 299         async_extent->compress_type = compress_type;
 300         list_add_tail(&async_extent->list, &cow->extents);
 301         return 0;
 302 }
 303
 304 /*
 305  * we create compressed extents in two phases.  The first
 306  * phase compresses a range of pages that have already been
 307  * locked (both pages and state bits are locked).
 308  *
 309  * This is done inside an ordered work queue, and the compression
 310  * is spread across many cpus.  The actual IO submission is step
 311  * two, and the ordered work queue takes care of making sure that
 312  * happens in the same order things were put onto the queue by
 313  * writepages and friends.
 314  *
 315  * If this code finds it can't get good compression, it puts an
 316  * entry onto the work queue to write the uncompressed bytes.  This
 317  * makes sure that both compressed inodes and uncompressed inodes
 318  * are written in the same order that pdflush sent them down.
 319  */
 320 static noinline int compress_file_range(struct inode *inode,
 321                                         struct page *locked_page,
 322                                         u64 start, u64 end,
 323                                         struct async_cow *async_cow,
 324                                         int *num_added)
 325 {
 326         struct btrfs_root *root = BTRFS_I(inode)->root;
 327         struct btrfs_trans_handle *trans;
 328         u64 num_bytes;
 329         u64 blocksize = root->sectorsize;
 330         u64 actual_end;
 331         u64 isize = i_size_read(inode);
 332         int ret = 0;
 333         struct page **pages = NULL;
 334         unsigned long nr_pages;
 335         unsigned long nr_pages_ret = 0;
 336         unsigned long total_compressed = 0;
 337         unsigned long total_in = 0;
 338         unsigned long max_compressed = 128 * 1024;
 339         unsigned long max_uncompressed = 128 * 1024;
 340         int i;
 341         int will_compress;
 342         int compress_type = root->fs_info->compress_type;
 343
 344         /* if this is a small write inside eof, kick off a defragbot */
 345         if (end <= BTRFS_I(inode)->disk_i_size && (end - start + 1) < 16 * 1024)
 346                 btrfs_add_inode_defrag(NULL, inode);
 347
 348         actual_end = min_t(u64, isize, end + 1);
 349 again:
 350         will_compress = 0;
 351         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 352         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 353
 354         /*
 355          * we don't want to send crud past the end of i_size through
 356          * compression, that's just a waste of CPU time.  So, if the
 357          * end of the file is before the start of our current
 358          * requested range of bytes, we bail out to the uncompressed
 359          * cleanup code that can deal with all of this.
 360          *
 361          * It isn't really the fastest way to fix things, but this is a
 362          * very uncommon corner.
 363          */
 364         if (actual_end <= start)
 365                 goto cleanup_and_bail_uncompressed;
 366
 367         total_compressed = actual_end - start;
 368
 369         /* we want to make sure that amount of ram required to uncompress
 370          * an extent is reasonable, so we limit the total size in ram
 371          * of a compressed extent to 128k.  This is a crucial number
 372          * because it also controls how easily we can spread reads across
 373          * cpus for decompression.
 374          *
 375          * We also want to make sure the amount of IO required to do
 376          * a random read is reasonably small, so we limit the size of
 377          * a compressed extent to 128k.
 378          */
 379         total_compressed = min(total_compressed, max_uncompressed);
 380         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 381         num_bytes = max(blocksize,  num_bytes);
 382         total_in = 0;
 383         ret = 0;
 384
 385         /*
 386          * we do compression for mount -o compress and when the
 387          * inode has not been flagged as nocompress.  This flag can
 388          * change at any time if we discover bad compression ratios.
 389          */
 390         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 391             (btrfs_test_opt(root, COMPRESS) ||
 392              (BTRFS_I(inode)->force_compress) ||
 393              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 394                 WARN_ON(pages);
 395                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 396                 if (!pages) {
 397                         /* just bail out to the uncompressed code */
 398                         goto cont;
 399                 }
 400
 401                 if (BTRFS_I(inode)->force_compress)
 402                         compress_type = BTRFS_I(inode)->force_compress;
 403
 404                 ret = btrfs_compress_pages(compress_type,
 405                                            inode->i_mapping, start,
 406                                            total_compressed, pages,
 407                                            nr_pages, &nr_pages_ret,
 408                                            &total_in,
 409                                            &total_compressed,
 410                                            max_compressed);
 411
 412                 if (!ret) {
 413                         unsigned long offset = total_compressed &
 414                                 (PAGE_CACHE_SIZE - 1);
 415                         struct page *page = pages[nr_pages_ret - 1];
 416                         char *kaddr;
 417
 418                         /* zero the tail end of the last page, we might be
 419                          * sending it down to disk
 420                          */
 421                         if (offset) {
 422                                 kaddr = kmap_atomic(page, KM_USER0);
 423                                 memset(kaddr + offset, 0,
 424                                        PAGE_CACHE_SIZE - offset);
 425                                 kunmap_atomic(kaddr, KM_USER0);
 426                         }
 427                         will_compress = 1;
 428                 }
 429         }
 430 cont:
 431         if (start == 0) {
 432                 trans = btrfs_join_transaction(root);
 433                 BUG_ON(IS_ERR(trans));
 434                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 435
 436                 /* lets try to make an inline extent */
 437                 if (ret || total_in < (actual_end - start)) {
 438                         /* we didn't compress the entire range, try
 439                          * to make an uncompressed inline extent.
 440                          */
 441                         ret = cow_file_range_inline(trans, root, inode,
 442                                                     start, end, 0, 0, NULL);
 443                 } else {
 444                         /* try making a compressed inline extent */
 445                         ret = cow_file_range_inline(trans, root, inode,
 446                                                     start, end,
 447                                                     total_compressed,
 448                                                     compress_type, pages);
 449                 }
 450                 if (ret == 0) {
 451                         /*
 452                          * inline extent creation worked, we don't need
 453                          * to create any more async work items.  Unlock
 454                          * and free up our temp pages.
 455                          */
 456                         extent_clear_unlock_delalloc(inode,
 457                              &BTRFS_I(inode)->io_tree,
 458                              start, end, NULL,
 459                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
 460                              EXTENT_CLEAR_DELALLOC |
 461                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
 462
 463                         btrfs_end_transaction(trans, root);
 464                         goto free_pages_out;
 465                 }
 466                 btrfs_end_transaction(trans, root);
 467         }
 468
 469         if (will_compress) {
 470                 /*
 471                  * we aren't doing an inline extent round the compressed size
 472                  * up to a block size boundary so the allocator does sane
 473                  * things
 474                  */
 475                 total_compressed = (total_compressed + blocksize - 1) &
 476                         ~(blocksize - 1);
 477
 478                 /*
 479                  * one last check to make sure the compression is really a
 480                  * win, compare the page count read with the blocks on disk
 481                  */
 482                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
 483                         ~(PAGE_CACHE_SIZE - 1);
 484                 if (total_compressed >= total_in) {
 485                         will_compress = 0;
 486                 } else {
 487                         num_bytes = total_in;
 488                 }
 489         }
 490         if (!will_compress && pages) {
 491                 /*
 492                  * the compression code ran but failed to make things smaller,
 493                  * free any pages it allocated and our page pointer array
 494                  */
 495                 for (i = 0; i < nr_pages_ret; i++) {
 496                         WARN_ON(pages[i]->mapping);
 497                         page_cache_release(pages[i]);
 498                 }
 499                 kfree(pages);
 500                 pages = NULL;
 501                 total_compressed = 0;
 502                 nr_pages_ret = 0;
 503
 504                 /* flag the file so we don't compress in the future */
 505                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 506                     !(BTRFS_I(inode)->force_compress)) {
 507                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 508                 }
 509         }
 510         if (will_compress) {
 511                 *num_added += 1;
 512
 513                 /* the async work queues will take care of doing actual
 514                  * allocation on disk for these compressed pages,
 515                  * and will submit them to the elevator.
 516                  */
 517                 add_async_extent(async_cow, start, num_bytes,
 518                                  total_compressed, pages, nr_pages_ret,
 519                                  compress_type);
 520
 521                 if (start + num_bytes < end) {
 522                         start += num_bytes;
 523                         pages = NULL;
 524                         cond_resched();
 525                         goto again;
 526                 }
 527         } else {
 528 cleanup_and_bail_uncompressed:
 529                 /*
 530                  * No compression, but we still need to write the pages in
 531                  * the file we've been given so far.  redirty the locked
 532                  * page if it corresponds to our extent and set things up
 533                  * for the async work queue to run cow_file_range to do
 534                  * the normal delalloc dance
 535                  */
 536                 if (page_offset(locked_page) >= start &&
 537                     page_offset(locked_page) <= end) {
 538                         __set_page_dirty_nobuffers(locked_page);
 539                         /* unlocked later on in the async handlers */
 540                 }
 541                 add_async_extent(async_cow, start, end - start + 1,
 542                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 543                 *num_added += 1;
 544         }
 545
 546 out:
 547         return 0;
 548
 549 free_pages_out:
 550         for (i = 0; i < nr_pages_ret; i++) {
 551                 WARN_ON(pages[i]->mapping);
 552                 page_cache_release(pages[i]);
 553         }
 554         kfree(pages);
 555
 556         goto out;
 557 }
 558
 559 /*
 560  * phase two of compressed writeback.  This is the ordered portion
 561  * of the code, which only gets called in the order the work was
 562  * queued.  We walk all the async extents created by compress_file_range
 563  * and send them down to the disk.
 564  */
 565 static noinline int submit_compressed_extents(struct inode *inode,
 566                                               struct async_cow *async_cow)
 567 {
 568         struct async_extent *async_extent;
 569         u64 alloc_hint = 0;
 570         struct btrfs_trans_handle *trans;
 571         struct btrfs_key ins;
 572         struct extent_map *em;
 573         struct btrfs_root *root = BTRFS_I(inode)->root;
 574         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 575         struct extent_io_tree *io_tree;
 576         int ret = 0;
 577
 578         if (list_empty(&async_cow->extents))
 579                 return 0;
 580
 581
 582         while (!list_empty(&async_cow->extents)) {
 583                 async_extent = list_entry(async_cow->extents.next,
 584                                           struct async_extent, list);
 585                 list_del(&async_extent->list);
 586
 587                 io_tree = &BTRFS_I(inode)->io_tree;
 588
 589 retry:
 590                 /* did the compression code fall back to uncompressed IO? */
 591                 if (!async_extent->pages) {
 592                         int page_started = 0;
 593                         unsigned long nr_written = 0;
 594
 595                         lock_extent(io_tree, async_extent->start,
 596                                          async_extent->start +
 597                                          async_extent->ram_size - 1, GFP_NOFS);
 598
 599                         /* allocate blocks */
 600                         ret = cow_file_range(inode, async_cow->locked_page,
 601                                              async_extent->start,
 602                                              async_extent->start +
 603                                              async_extent->ram_size - 1,
 604                                              &page_started, &nr_written, 0);
 605
 606                         /*
 607                          * if page_started, cow_file_range inserted an
 608                          * inline extent and took care of all the unlocking
 609                          * and IO for us.  Otherwise, we need to submit
 610                          * all those pages down to the drive.
 611                          */
 612                         if (!page_started && !ret)
 613                                 extent_write_locked_range(io_tree,
 614                                                   inode, async_extent->start,
 615                                                   async_extent->start +
 616                                                   async_extent->ram_size - 1,
 617                                                   btrfs_get_extent,
 618                                                   WB_SYNC_ALL);
 619                         kfree(async_extent);
 620                         cond_resched();
 621                         continue;
 622                 }
 623
 624                 lock_extent(io_tree, async_extent->start,
 625                             async_extent->start + async_extent->ram_size - 1,
 626                             GFP_NOFS);
 627
 628                 trans = btrfs_join_transaction(root);
 629                 BUG_ON(IS_ERR(trans));
 630                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 631                 ret = btrfs_reserve_extent(trans, root,
 632                                            async_extent->compressed_size,
 633                                            async_extent->compressed_size,
 634                                            0, alloc_hint,
 635                                            (u64)-1, &ins, 1);
 636                 btrfs_end_transaction(trans, root);
 637
 638                 if (ret) {
 639                         int i;
 640                         for (i = 0; i < async_extent->nr_pages; i++) {
 641                                 WARN_ON(async_extent->pages[i]->mapping);
 642                                 page_cache_release(async_extent->pages[i]);
 643                         }
 644                         kfree(async_extent->pages);
 645                         async_extent->nr_pages = 0;
 646                         async_extent->pages = NULL;
 647                         unlock_extent(io_tree, async_extent->start,
 648                                       async_extent->start +
 649                                       async_extent->ram_size - 1, GFP_NOFS);
 650                         goto retry;
 651                 }
 652
 653                 /*
 654                  * here we're doing allocation and writeback of the
 655                  * compressed pages
 656                  */
 657                 btrfs_drop_extent_cache(inode, async_extent->start,
 658                                         async_extent->start +
 659                                         async_extent->ram_size - 1, 0);
 660
 661                 em = alloc_extent_map();
 662                 BUG_ON(!em);
 663                 em->start = async_extent->start;
 664                 em->len = async_extent->ram_size;
 665                 em->orig_start = em->start;
 666
 667                 em->block_start = ins.objectid;
 668                 em->block_len = ins.offset;
 669                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 670                 em->compress_type = async_extent->compress_type;
 671                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 672                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 673
 674                 while (1) {
 675                         write_lock(&em_tree->lock);
 676                         ret = add_extent_mapping(em_tree, em);
 677                         write_unlock(&em_tree->lock);
 678                         if (ret != -EEXIST) {
 679                                 free_extent_map(em);
 680                                 break;
 681                         }
 682                         btrfs_drop_extent_cache(inode, async_extent->start,
 683                                                 async_extent->start +
 684                                                 async_extent->ram_size - 1, 0);
 685                 }
 686
 687                 ret = btrfs_add_ordered_extent_compress(inode,
 688                                                 async_extent->start,
 689                                                 ins.objectid,
 690                                                 async_extent->ram_size,
 691                                                 ins.offset,
 692                                                 BTRFS_ORDERED_COMPRESSED,
 693                                                 async_extent->compress_type);
 694                 BUG_ON(ret);
 695
 696                 /*
 697                  * clear dirty, set writeback and unlock the pages.
 698                  */
 699                 extent_clear_unlock_delalloc(inode,
 700                                 &BTRFS_I(inode)->io_tree,
 701                                 async_extent->start,
 702                                 async_extent->start +
 703                                 async_extent->ram_size - 1,
 704                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 705                                 EXTENT_CLEAR_UNLOCK |
 706                                 EXTENT_CLEAR_DELALLOC |
 707                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
 708
 709                 ret = btrfs_submit_compressed_write(inode,
 710                                     async_extent->start,
 711                                     async_extent->ram_size,
 712                                     ins.objectid,
 713                                     ins.offset, async_extent->pages,
 714                                     async_extent->nr_pages);
 715
 716                 BUG_ON(ret);
 717                 alloc_hint = ins.objectid + ins.offset;
 718                 kfree(async_extent);
 719                 cond_resched();
 720         }
 721
 722         return 0;
 723 }
 724
 725 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 726                                       u64 num_bytes)
 727 {
 728         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 729         struct extent_map *em;
 730         u64 alloc_hint = 0;
 731
 732         read_lock(&em_tree->lock);
 733         em = search_extent_mapping(em_tree, start, num_bytes);
 734         if (em) {
 735                 /*
 736                  * if block start isn't an actual block number then find the
 737                  * first block in this inode and use that as a hint.  If that
 738                  * block is also bogus then just don't worry about it.
 739                  */
 740                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 741                         free_extent_map(em);
 742                         em = search_extent_mapping(em_tree, 0, 0);
 743                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 744                                 alloc_hint = em->block_start;
 745                         if (em)
 746                                 free_extent_map(em);
 747                 } else {
 748                         alloc_hint = em->block_start;
 749                         free_extent_map(em);
 750                 }
 751         }
 752         read_unlock(&em_tree->lock);
 753
 754         return alloc_hint;
 755 }
 756
 757 /*
 758  * when extent_io.c finds a delayed allocation range in the file,
 759  * the call backs end up in this code.  The basic idea is to
 760  * allocate extents on disk for the range, and create ordered data structs
 761  * in ram to track those extents.
 762  *
 763  * locked_page is the page that writepage had locked already.  We use
 764  * it to make sure we don't do extra locks or unlocks.
 765  *
 766  * *page_started is set to one if we unlock locked_page and do everything
 767  * required to start IO on it.  It may be clean and already done with
 768  * IO when we return.
 769  */
 770 static noinline int cow_file_range(struct inode *inode,
 771                                    struct page *locked_page,
 772                                    u64 start, u64 end, int *page_started,
 773                                    unsigned long *nr_written,
 774                                    int unlock)
 775 {
 776         struct btrfs_root *root = BTRFS_I(inode)->root;
 777         struct btrfs_trans_handle *trans;
 778         u64 alloc_hint = 0;
 779         u64 num_bytes;
 780         unsigned long ram_size;
 781         u64 disk_num_bytes;
 782         u64 cur_alloc_size;
 783         u64 blocksize = root->sectorsize;
 784         struct btrfs_key ins;
 785         struct extent_map *em;
 786         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 787         int ret = 0;
 788
 789         BUG_ON(btrfs_is_free_space_inode(root, inode));
 790         trans = btrfs_join_transaction(root);
 791         BUG_ON(IS_ERR(trans));
 792         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 793
 794         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 795         num_bytes = max(blocksize,  num_bytes);
 796         disk_num_bytes = num_bytes;
 797         ret = 0;
 798
 799         /* if this is a small write inside eof, kick off defrag */
 800         if (end <= BTRFS_I(inode)->disk_i_size && num_bytes < 64 * 1024)
 801                 btrfs_add_inode_defrag(trans, inode);
 802
 803         if (start == 0) {
 804                 /* lets try to make an inline extent */
 805                 ret = cow_file_range_inline(trans, root, inode,
 806                                             start, end, 0, 0, NULL);
 807                 if (ret == 0) {
 808                         extent_clear_unlock_delalloc(inode,
 809                                      &BTRFS_I(inode)->io_tree,
 810                                      start, end, NULL,
 811                                      EXTENT_CLEAR_UNLOCK_PAGE |
 812                                      EXTENT_CLEAR_UNLOCK |
 813                                      EXTENT_CLEAR_DELALLOC |
 814                                      EXTENT_CLEAR_DIRTY |
 815                                      EXTENT_SET_WRITEBACK |
 816                                      EXTENT_END_WRITEBACK);
 817
 818                         *nr_written = *nr_written +
 819                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 820                         *page_started = 1;
 821                         ret = 0;
 822                         goto out;
 823                 }
 824         }
 825
 826         BUG_ON(disk_num_bytes >
 827                btrfs_super_total_bytes(&root->fs_info->super_copy));
 828
 829         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 830         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 831
 832         while (disk_num_bytes > 0) {
 833                 unsigned long op;
 834
 835                 cur_alloc_size = disk_num_bytes;
 836                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 837                                            root->sectorsize, 0, alloc_hint,
 838                                            (u64)-1, &ins, 1);
 839                 BUG_ON(ret);
 840
 841                 em = alloc_extent_map();
 842                 BUG_ON(!em);
 843                 em->start = start;
 844                 em->orig_start = em->start;
 845                 ram_size = ins.offset;
 846                 em->len = ins.offset;
 847
 848                 em->block_start = ins.objectid;
 849                 em->block_len = ins.offset;
 850                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 851                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 852
 853                 while (1) {
 854                         write_lock(&em_tree->lock);
 855                         ret = add_extent_mapping(em_tree, em);
 856                         write_unlock(&em_tree->lock);
 857                         if (ret != -EEXIST) {
 858                                 free_extent_map(em);
 859                                 break;
 860                         }
 861                         btrfs_drop_extent_cache(inode, start,
 862                                                 start + ram_size - 1, 0);
 863                 }
 864
 865                 cur_alloc_size = ins.offset;
 866                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 867                                                ram_size, cur_alloc_size, 0);
 868                 BUG_ON(ret);
 869
 870                 if (root->root_key.objectid ==
 871                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 872                         ret = btrfs_reloc_clone_csums(inode, start,
 873                                                       cur_alloc_size);
 874                         BUG_ON(ret);
 875                 }
 876
 877                 if (disk_num_bytes < cur_alloc_size)
 878                         break;
 879
 880                 /* we're not doing compressed IO, don't unlock the first
 881                  * page (which the caller expects to stay locked), don't
 882                  * clear any dirty bits and don't set any writeback bits
 883                  *
 884                  * Do set the Private2 bit so we know this page was properly
 885                  * setup for writepage
 886                  */
 887                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
 888                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
 889                         EXTENT_SET_PRIVATE2;
 890
 891                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 892                                              start, start + ram_size - 1,
 893                                              locked_page, op);
 894                 disk_num_bytes -= cur_alloc_size;
 895                 num_bytes -= cur_alloc_size;
 896                 alloc_hint = ins.objectid + ins.offset;
 897                 start += cur_alloc_size;
 898         }
 899 out:
 900         ret = 0;
 901         btrfs_end_transaction(trans, root);
 902
 903         return ret;
 904 }
 905
 906 /*
 907  * work queue call back to started compression on a file and pages
 908  */
 909 static noinline void async_cow_start(struct btrfs_work *work)
 910 {
 911         struct async_cow *async_cow;
 912         int num_added = 0;
 913         async_cow = container_of(work, struct async_cow, work);
 914
 915         compress_file_range(async_cow->inode, async_cow->locked_page,
 916                             async_cow->start, async_cow->end, async_cow,
 917                             &num_added);
 918         if (num_added == 0)
 919                 async_cow->inode = NULL;
 920 }
 921
 922 /*
 923  * work queue call back to submit previously compressed pages
 924  */
 925 static noinline void async_cow_submit(struct btrfs_work *work)
 926 {
 927         struct async_cow *async_cow;
 928         struct btrfs_root *root;
 929         unsigned long nr_pages;
 930
 931         async_cow = container_of(work, struct async_cow, work);
 932
 933         root = async_cow->root;
 934         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
 935                 PAGE_CACHE_SHIFT;
 936
 937         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
 938
 939         if (atomic_read(&root->fs_info->async_delalloc_pages) <
 940             5 * 1042 * 1024 &&
 941             waitqueue_active(&root->fs_info->async_submit_wait))
 942                 wake_up(&root->fs_info->async_submit_wait);
 943
 944         if (async_cow->inode)
 945                 submit_compressed_extents(async_cow->inode, async_cow);
 946 }
 947
 948 static noinline void async_cow_free(struct btrfs_work *work)
 949 {
 950         struct async_cow *async_cow;
 951         async_cow = container_of(work, struct async_cow, work);
 952         kfree(async_cow);
 953 }
 954
 955 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
 956                                 u64 start, u64 end, int *page_started,
 957                                 unsigned long *nr_written)
 958 {
 959         struct async_cow *async_cow;
 960         struct btrfs_root *root = BTRFS_I(inode)->root;
 961         unsigned long nr_pages;
 962         u64 cur_end;
 963         int limit = 10 * 1024 * 1042;
 964
 965         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
 966                          1, 0, NULL, GFP_NOFS);
 967         while (start < end) {
 968                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
 969                 BUG_ON(!async_cow);
 970                 async_cow->inode = inode;
 971                 async_cow->root = root;
 972                 async_cow->locked_page = locked_page;
 973                 async_cow->start = start;
 974
 975                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
 976                         cur_end = end;
 977                 else
 978                         cur_end = min(end, start + 512 * 1024 - 1);
 979
 980                 async_cow->end = cur_end;
 981                 INIT_LIST_HEAD(&async_cow->extents);
 982
 983                 async_cow->work.func = async_cow_start;
 984                 async_cow->work.ordered_func = async_cow_submit;
 985                 async_cow->work.ordered_free = async_cow_free;
 986                 async_cow->work.flags = 0;
 987
 988                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
 989                         PAGE_CACHE_SHIFT;
 990                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
 991
 992                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
 993                                    &async_cow->work);
 994
 995                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
 996                         wait_event(root->fs_info->async_submit_wait,
 997                            (atomic_read(&root->fs_info->async_delalloc_pages) <
 998                             limit));
 999                 }
1000
1001                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1002                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1003                         wait_event(root->fs_info->async_submit_wait,
1004                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1005                            0));
1006                 }
1007
1008                 *nr_written += nr_pages;
1009                 start = cur_end + 1;
1010         }
1011         *page_started = 1;
1012         return 0;
1013 }
1014
1015 static noinline int csum_exist_in_range(struct btrfs_root *root,
1016                                         u64 bytenr, u64 num_bytes)
1017 {
1018         int ret;
1019         struct btrfs_ordered_sum *sums;
1020         LIST_HEAD(list);
1021
1022         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1023                                        bytenr + num_bytes - 1, &list, 0);
1024         if (ret == 0 && list_empty(&list))
1025                 return 0;
1026
1027         while (!list_empty(&list)) {
1028                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1029                 list_del(&sums->list);
1030                 kfree(sums);
1031         }
1032         return 1;
1033 }
1034
1035 /*
1036  * when nowcow writeback call back.  This checks for snapshots or COW copies
1037  * of the extents that exist in the file, and COWs the file as required.
1038  *
1039  * If no cow copies or snapshots exist, we write directly to the existing
1040  * blocks on disk
1041  */
1042 static noinline int run_delalloc_nocow(struct inode *inode,
1043                                        struct page *locked_page,
1044                               u64 start, u64 end, int *page_started, int force,
1045                               unsigned long *nr_written)
1046 {
1047         struct btrfs_root *root = BTRFS_I(inode)->root;
1048         struct btrfs_trans_handle *trans;
1049         struct extent_buffer *leaf;
1050         struct btrfs_path *path;
1051         struct btrfs_file_extent_item *fi;
1052         struct btrfs_key found_key;
1053         u64 cow_start;
1054         u64 cur_offset;
1055         u64 extent_end;
1056         u64 extent_offset;
1057         u64 disk_bytenr;
1058         u64 num_bytes;
1059         int extent_type;
1060         int ret;
1061         int type;
1062         int nocow;
1063         int check_prev = 1;
1064         bool nolock;
1065         u64 ino = btrfs_ino(inode);
1066
1067         path = btrfs_alloc_path();
1068         if (!path)
1069                 return -ENOMEM;
1070
1071         nolock = btrfs_is_free_space_inode(root, inode);
1072
1073         if (nolock)
1074                 trans = btrfs_join_transaction_nolock(root);
1075         else
1076                 trans = btrfs_join_transaction(root);
1077
1078         BUG_ON(IS_ERR(trans));
1079         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1080
1081         cow_start = (u64)-1;
1082         cur_offset = start;
1083         while (1) {
1084                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1085                                                cur_offset, 0);
1086                 BUG_ON(ret < 0);
1087                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1088                         leaf = path->nodes[0];
1089                         btrfs_item_key_to_cpu(leaf, &found_key,
1090                                               path->slots[0] - 1);
1091                         if (found_key.objectid == ino &&
1092                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1093                                 path->slots[0]--;
1094                 }
1095                 check_prev = 0;
1096 next_slot:
1097                 leaf = path->nodes[0];
1098                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1099                         ret = btrfs_next_leaf(root, path);
1100                         if (ret < 0)
1101                                 BUG_ON(1);
1102                         if (ret > 0)
1103                                 break;
1104                         leaf = path->nodes[0];
1105                 }
1106
1107                 nocow = 0;
1108                 disk_bytenr = 0;
1109                 num_bytes = 0;
1110                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1111
1112                 if (found_key.objectid > ino ||
1113                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1114                     found_key.offset > end)
1115                         break;
1116
1117                 if (found_key.offset > cur_offset) {
1118                         extent_end = found_key.offset;
1119                         extent_type = 0;
1120                         goto out_check;
1121                 }
1122
1123                 fi = btrfs_item_ptr(leaf, path->slots[0],
1124                                     struct btrfs_file_extent_item);
1125                 extent_type = btrfs_file_extent_type(leaf, fi);
1126
1127                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1128                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1129                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1130                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1131                         extent_end = found_key.offset +
1132                                 btrfs_file_extent_num_bytes(leaf, fi);
1133                         if (extent_end <= start) {
1134                                 path->slots[0]++;
1135                                 goto next_slot;
1136                         }
1137                         if (disk_bytenr == 0)
1138                                 goto out_check;
1139                         if (btrfs_file_extent_compression(leaf, fi) ||
1140                             btrfs_file_extent_encryption(leaf, fi) ||
1141                             btrfs_file_extent_other_encoding(leaf, fi))
1142                                 goto out_check;
1143                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1144                                 goto out_check;
1145                         if (btrfs_extent_readonly(root, disk_bytenr))
1146                                 goto out_check;
1147                         if (btrfs_cross_ref_exist(trans, root, ino,
1148                                                   found_key.offset -
1149                                                   extent_offset, disk_bytenr))
1150                                 goto out_check;
1151                         disk_bytenr += extent_offset;
1152                         disk_bytenr += cur_offset - found_key.offset;
1153                         num_bytes = min(end + 1, extent_end) - cur_offset;
1154                         /*
1155                          * force cow if csum exists in the range.
1156                          * this ensure that csum for a given extent are
1157                          * either valid or do not exist.
1158                          */
1159                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1160                                 goto out_check;
1161                         nocow = 1;
1162                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1163                         extent_end = found_key.offset +
1164                                 btrfs_file_extent_inline_len(leaf, fi);
1165                         extent_end = ALIGN(extent_end, root->sectorsize);
1166                 } else {
1167                         BUG_ON(1);
1168                 }
1169 out_check:
1170                 if (extent_end <= start) {
1171                         path->slots[0]++;
1172                         goto next_slot;
1173                 }
1174                 if (!nocow) {
1175                         if (cow_start == (u64)-1)
1176                                 cow_start = cur_offset;
1177                         cur_offset = extent_end;
1178                         if (cur_offset > end)
1179                                 break;
1180                         path->slots[0]++;
1181                         goto next_slot;
1182                 }
1183
1184                 btrfs_release_path(path);
1185                 if (cow_start != (u64)-1) {
1186                         ret = cow_file_range(inode, locked_page, cow_start,
1187                                         found_key.offset - 1, page_started,
1188                                         nr_written, 1);
1189                         BUG_ON(ret);
1190                         cow_start = (u64)-1;
1191                 }
1192
1193                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1194                         struct extent_map *em;
1195                         struct extent_map_tree *em_tree;
1196                         em_tree = &BTRFS_I(inode)->extent_tree;
1197                         em = alloc_extent_map();
1198                         BUG_ON(!em);
1199                         em->start = cur_offset;
1200                         em->orig_start = em->start;
1201                         em->len = num_bytes;
1202                         em->block_len = num_bytes;
1203                         em->block_start = disk_bytenr;
1204                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1205                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1206                         while (1) {
1207                                 write_lock(&em_tree->lock);
1208                                 ret = add_extent_mapping(em_tree, em);
1209                                 write_unlock(&em_tree->lock);
1210                                 if (ret != -EEXIST) {
1211                                         free_extent_map(em);
1212                                         break;
1213                                 }
1214                                 btrfs_drop_extent_cache(inode, em->start,
1215                                                 em->start + em->len - 1, 0);
1216                         }
1217                         type = BTRFS_ORDERED_PREALLOC;
1218                 } else {
1219                         type = BTRFS_ORDERED_NOCOW;
1220                 }
1221
1222                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1223                                                num_bytes, num_bytes, type);
1224                 BUG_ON(ret);
1225
1226                 if (root->root_key.objectid ==
1227                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1228                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1229                                                       num_bytes);
1230                         BUG_ON(ret);
1231                 }
1232
1233                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1234                                 cur_offset, cur_offset + num_bytes - 1,
1235                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1236                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1237                                 EXTENT_SET_PRIVATE2);
1238                 cur_offset = extent_end;
1239                 if (cur_offset > end)
1240                         break;
1241         }
1242         btrfs_release_path(path);
1243
1244         if (cur_offset <= end && cow_start == (u64)-1)
1245                 cow_start = cur_offset;
1246         if (cow_start != (u64)-1) {
1247                 ret = cow_file_range(inode, locked_page, cow_start, end,
1248                                      page_started, nr_written, 1);
1249                 BUG_ON(ret);
1250         }
1251
1252         if (nolock) {
1253                 ret = btrfs_end_transaction_nolock(trans, root);
1254                 BUG_ON(ret);
1255         } else {
1256                 ret = btrfs_end_transaction(trans, root);
1257                 BUG_ON(ret);
1258         }
1259         btrfs_free_path(path);
1260         return 0;
1261 }
1262
1263 /*
1264  * extent_io.c call back to do delayed allocation processing
1265  */
1266 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1267                               u64 start, u64 end, int *page_started,
1268                               unsigned long *nr_written)
1269 {
1270         int ret;
1271         struct btrfs_root *root = BTRFS_I(inode)->root;
1272
1273         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1274                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1275                                          page_started, 1, nr_written);
1276         else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1277                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1278                                          page_started, 0, nr_written);
1279         else if (!btrfs_test_opt(root, COMPRESS) &&
1280                  !(BTRFS_I(inode)->force_compress) &&
1281                  !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1282                 ret = cow_file_range(inode, locked_page, start, end,
1283                                       page_started, nr_written, 1);
1284         else
1285                 ret = cow_file_range_async(inode, locked_page, start, end,
1286                                            page_started, nr_written);
1287         return ret;
1288 }
1289
1290 static void btrfs_split_extent_hook(struct inode *inode,
1291                                     struct extent_state *orig, u64 split)
1292 {
1293         /* not delalloc, ignore it */
1294         if (!(orig->state & EXTENT_DELALLOC))
1295                 return;
1296
1297         spin_lock(&BTRFS_I(inode)->lock);
1298         BTRFS_I(inode)->outstanding_extents++;
1299         spin_unlock(&BTRFS_I(inode)->lock);
1300 }
1301
1302 /*
1303  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1304  * extents so we can keep track of new extents that are just merged onto old
1305  * extents, such as when we are doing sequential writes, so we can properly
1306  * account for the metadata space we'll need.
1307  */
1308 static void btrfs_merge_extent_hook(struct inode *inode,
1309                                     struct extent_state *new,
1310                                     struct extent_state *other)
1311 {
1312         /* not delalloc, ignore it */
1313         if (!(other->state & EXTENT_DELALLOC))
1314                 return;
1315
1316         spin_lock(&BTRFS_I(inode)->lock);
1317         BTRFS_I(inode)->outstanding_extents--;
1318         spin_unlock(&BTRFS_I(inode)->lock);
1319 }
1320
1321 /*
1322  * extent_io.c set_bit_hook, used to track delayed allocation
1323  * bytes in this file, and to maintain the list of inodes that
1324  * have pending delalloc work to be done.
1325  */
1326 static void btrfs_set_bit_hook(struct inode *inode,
1327                                struct extent_state *state, int *bits)
1328 {
1329
1330         /*
1331          * set_bit and clear bit hooks normally require _irqsave/restore
1332          * but in this case, we are only testing for the DELALLOC
1333          * bit, which is only set or cleared with irqs on
1334          */
1335         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1336                 struct btrfs_root *root = BTRFS_I(inode)->root;
1337                 u64 len = state->end + 1 - state->start;
1338                 bool do_list = !btrfs_is_free_space_inode(root, inode);
1339
1340                 if (*bits & EXTENT_FIRST_DELALLOC) {
1341                         *bits &= ~EXTENT_FIRST_DELALLOC;
1342                 } else {
1343                         spin_lock(&BTRFS_I(inode)->lock);
1344                         BTRFS_I(inode)->outstanding_extents++;
1345                         spin_unlock(&BTRFS_I(inode)->lock);
1346                 }
1347
1348                 spin_lock(&root->fs_info->delalloc_lock);
1349                 BTRFS_I(inode)->delalloc_bytes += len;
1350                 root->fs_info->delalloc_bytes += len;
1351                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1352                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1353                                       &root->fs_info->delalloc_inodes);
1354                 }
1355                 spin_unlock(&root->fs_info->delalloc_lock);
1356         }
1357 }
1358
1359 /*
1360  * extent_io.c clear_bit_hook, see set_bit_hook for why
1361  */
1362 static void btrfs_clear_bit_hook(struct inode *inode,
1363                                  struct extent_state *state, int *bits)
1364 {
1365         /*
1366          * set_bit and clear bit hooks normally require _irqsave/restore
1367          * but in this case, we are only testing for the DELALLOC
1368          * bit, which is only set or cleared with irqs on
1369          */
1370         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1371                 struct btrfs_root *root = BTRFS_I(inode)->root;
1372                 u64 len = state->end + 1 - state->start;
1373                 bool do_list = !btrfs_is_free_space_inode(root, inode);
1374
1375                 if (*bits & EXTENT_FIRST_DELALLOC) {
1376                         *bits &= ~EXTENT_FIRST_DELALLOC;
1377                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1378                         spin_lock(&BTRFS_I(inode)->lock);
1379                         BTRFS_I(inode)->outstanding_extents--;
1380                         spin_unlock(&BTRFS_I(inode)->lock);
1381                 }
1382
1383                 if (*bits & EXTENT_DO_ACCOUNTING)
1384                         btrfs_delalloc_release_metadata(inode, len);
1385
1386                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1387                     && do_list)
1388                         btrfs_free_reserved_data_space(inode, len);
1389
1390                 spin_lock(&root->fs_info->delalloc_lock);
1391                 root->fs_info->delalloc_bytes -= len;
1392                 BTRFS_I(inode)->delalloc_bytes -= len;
1393
1394                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1395                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1396                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1397                 }
1398                 spin_unlock(&root->fs_info->delalloc_lock);
1399         }
1400 }
1401
1402 /*
1403  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1404  * we don't create bios that span stripes or chunks
1405  */
1406 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1407                          size_t size, struct bio *bio,
1408                          unsigned long bio_flags)
1409 {
1410         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1411         struct btrfs_mapping_tree *map_tree;
1412         u64 logical = (u64)bio->bi_sector << 9;
1413         u64 length = 0;
1414         u64 map_length;
1415         int ret;
1416
1417         if (bio_flags & EXTENT_BIO_COMPRESSED)
1418                 return 0;
1419
1420         length = bio->bi_size;
1421         map_tree = &root->fs_info->mapping_tree;
1422         map_length = length;
1423         ret = btrfs_map_block(map_tree, READ, logical,
1424                               &map_length, NULL, 0);
1425
1426         if (map_length < length + size)
1427                 return 1;
1428         return ret;
1429 }
1430
1431 /*
1432  * in order to insert checksums into the metadata in large chunks,
1433  * we wait until bio submission time.   All the pages in the bio are
1434  * checksummed and sums are attached onto the ordered extent record.
1435  *
1436  * At IO completion time the cums attached on the ordered extent record
1437  * are inserted into the btree
1438  */
1439 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1440                                     struct bio *bio, int mirror_num,
1441                                     unsigned long bio_flags,
1442                                     u64 bio_offset)
1443 {
1444         struct btrfs_root *root = BTRFS_I(inode)->root;
1445         int ret = 0;
1446
1447         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1448         BUG_ON(ret);
1449         return 0;
1450 }
1451
1452 /*
1453  * in order to insert checksums into the metadata in large chunks,
1454  * we wait until bio submission time.   All the pages in the bio are
1455  * checksummed and sums are attached onto the ordered extent record.
1456  *
1457  * At IO completion time the cums attached on the ordered extent record
1458  * are inserted into the btree
1459  */
1460 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1461                           int mirror_num, unsigned long bio_flags,
1462                           u64 bio_offset)
1463 {
1464         struct btrfs_root *root = BTRFS_I(inode)->root;
1465         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1466 }
1467
1468 /*
1469  * extent_io.c submission hook. This does the right thing for csum calculation
1470  * on write, or reading the csums from the tree before a read
1471  */
1472 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1473                           int mirror_num, unsigned long bio_flags,
1474                           u64 bio_offset)
1475 {
1476         struct btrfs_root *root = BTRFS_I(inode)->root;
1477         int ret = 0;
1478         int skip_sum;
1479
1480         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1481
1482         if (btrfs_is_free_space_inode(root, inode))
1483                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1484         else
1485                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1486         BUG_ON(ret);
1487
1488         if (!(rw & REQ_WRITE)) {
1489                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1490                         return btrfs_submit_compressed_read(inode, bio,
1491                                                     mirror_num, bio_flags);
1492                 } else if (!skip_sum) {
1493                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1494                         if (ret)
1495                                 return ret;
1496                 }
1497                 goto mapit;
1498         } else if (!skip_sum) {
1499                 /* csum items have already been cloned */
1500                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1501                         goto mapit;
1502                 /* we're doing a write, do the async checksumming */
1503                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1504                                    inode, rw, bio, mirror_num,
1505                                    bio_flags, bio_offset,
1506                                    __btrfs_submit_bio_start,
1507                                    __btrfs_submit_bio_done);
1508         }
1509
1510 mapit:
1511         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1512 }
1513
1514 /*
1515  * given a list of ordered sums record them in the inode.  This happens
1516  * at IO completion time based on sums calculated at bio submission time.
1517  */
1518 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1519                              struct inode *inode, u64 file_offset,
1520                              struct list_head *list)
1521 {
1522         struct btrfs_ordered_sum *sum;
1523
1524         list_for_each_entry(sum, list, list) {
1525                 btrfs_csum_file_blocks(trans,
1526                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1527         }
1528         return 0;
1529 }
1530
1531 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1532                               struct extent_state **cached_state)
1533 {
1534         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1535                 WARN_ON(1);
1536         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1537                                    cached_state, GFP_NOFS);
1538 }
1539
1540 /* see btrfs_writepage_start_hook for details on why this is required */
1541 struct btrfs_writepage_fixup {
1542         struct page *page;
1543         struct btrfs_work work;
1544 };
1545
1546 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1547 {
1548         struct btrfs_writepage_fixup *fixup;
1549         struct btrfs_ordered_extent *ordered;
1550         struct extent_state *cached_state = NULL;
1551         struct page *page;
1552         struct inode *inode;
1553         u64 page_start;
1554         u64 page_end;
1555
1556         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1557         page = fixup->page;
1558 again:
1559         lock_page(page);
1560         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1561                 ClearPageChecked(page);
1562                 goto out_page;
1563         }
1564
1565         inode = page->mapping->host;
1566         page_start = page_offset(page);
1567         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1568
1569         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1570                          &cached_state, GFP_NOFS);
1571
1572         /* already ordered? We're done */
1573         if (PagePrivate2(page))
1574                 goto out;
1575
1576         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1577         if (ordered) {
1578                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1579                                      page_end, &cached_state, GFP_NOFS);
1580                 unlock_page(page);
1581                 btrfs_start_ordered_extent(inode, ordered, 1);
1582                 goto again;
1583         }
1584
1585         BUG();
1586         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1587         ClearPageChecked(page);
1588 out:
1589         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1590                              &cached_state, GFP_NOFS);
1591 out_page:
1592         unlock_page(page);
1593         page_cache_release(page);
1594         kfree(fixup);
1595 }
1596
1597 /*
1598  * There are a few paths in the higher layers of the kernel that directly
1599  * set the page dirty bit without asking the filesystem if it is a
1600  * good idea.  This causes problems because we want to make sure COW
1601  * properly happens and the data=ordered rules are followed.
1602  *
1603  * In our case any range that doesn't have the ORDERED bit set
1604  * hasn't been properly setup for IO.  We kick off an async process
1605  * to fix it up.  The async helper will wait for ordered extents, set
1606  * the delalloc bit and make it safe to write the page.
1607  */
1608 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1609 {
1610         struct inode *inode = page->mapping->host;
1611         struct btrfs_writepage_fixup *fixup;
1612         struct btrfs_root *root = BTRFS_I(inode)->root;
1613
1614         /* this page is properly in the ordered list */
1615         if (TestClearPagePrivate2(page))
1616                 return 0;
1617
1618         if (PageChecked(page))
1619                 return -EAGAIN;
1620
1621         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1622         if (!fixup)
1623                 return -EAGAIN;
1624
1625         SetPageChecked(page);
1626         page_cache_get(page);
1627         fixup->work.func = btrfs_writepage_fixup_worker;
1628         fixup->page = page;
1629         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1630         return -EAGAIN;
1631 }
1632
1633 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1634                                        struct inode *inode, u64 file_pos,
1635                                        u64 disk_bytenr, u64 disk_num_bytes,
1636                                        u64 num_bytes, u64 ram_bytes,
1637                                        u8 compression, u8 encryption,
1638                                        u16 other_encoding, int extent_type)
1639 {
1640         struct btrfs_root *root = BTRFS_I(inode)->root;
1641         struct btrfs_file_extent_item *fi;
1642         struct btrfs_path *path;
1643         struct extent_buffer *leaf;
1644         struct btrfs_key ins;
1645         u64 hint;
1646         int ret;
1647
1648         path = btrfs_alloc_path();
1649         if (!path)
1650                 return -ENOMEM;
1651
1652         path->leave_spinning = 1;
1653
1654         /*
1655          * we may be replacing one extent in the tree with another.
1656          * The new extent is pinned in the extent map, and we don't want
1657          * to drop it from the cache until it is completely in the btree.
1658          *
1659          * So, tell btrfs_drop_extents to leave this extent in the cache.
1660          * the caller is expected to unpin it and allow it to be merged
1661          * with the others.
1662          */
1663         ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1664                                  &hint, 0);
1665         BUG_ON(ret);
1666
1667         ins.objectid = btrfs_ino(inode);
1668         ins.offset = file_pos;
1669         ins.type = BTRFS_EXTENT_DATA_KEY;
1670         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1671         BUG_ON(ret);
1672         leaf = path->nodes[0];
1673         fi = btrfs_item_ptr(leaf, path->slots[0],
1674                             struct btrfs_file_extent_item);
1675         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1676         btrfs_set_file_extent_type(leaf, fi, extent_type);
1677         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1678         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1679         btrfs_set_file_extent_offset(leaf, fi, 0);
1680         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1681         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1682         btrfs_set_file_extent_compression(leaf, fi, compression);
1683         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1684         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1685
1686         btrfs_unlock_up_safe(path, 1);
1687         btrfs_set_lock_blocking(leaf);
1688
1689         btrfs_mark_buffer_dirty(leaf);
1690
1691         inode_add_bytes(inode, num_bytes);
1692
1693         ins.objectid = disk_bytenr;
1694         ins.offset = disk_num_bytes;
1695         ins.type = BTRFS_EXTENT_ITEM_KEY;
1696         ret = btrfs_alloc_reserved_file_extent(trans, root,
1697                                         root->root_key.objectid,
1698                                         btrfs_ino(inode), file_pos, &ins);
1699         BUG_ON(ret);
1700         btrfs_free_path(path);
1701
1702         return 0;
1703 }
1704
1705 /*
1706  * helper function for btrfs_finish_ordered_io, this
1707  * just reads in some of the csum leaves to prime them into ram
1708  * before we start the transaction.  It limits the amount of btree
1709  * reads required while inside the transaction.
1710  */
1711 /* as ordered data IO finishes, this gets called so we can finish
1712  * an ordered extent if the range of bytes in the file it covers are
1713  * fully written.
1714  */
1715 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1716 {
1717         struct btrfs_root *root = BTRFS_I(inode)->root;
1718         struct btrfs_trans_handle *trans = NULL;
1719         struct btrfs_ordered_extent *ordered_extent = NULL;
1720         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1721         struct extent_state *cached_state = NULL;
1722         int compress_type = 0;
1723         int ret;
1724         bool nolock;
1725
1726         ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1727                                              end - start + 1);
1728         if (!ret)
1729                 return 0;
1730         BUG_ON(!ordered_extent);
1731
1732         nolock = btrfs_is_free_space_inode(root, inode);
1733
1734         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1735                 BUG_ON(!list_empty(&ordered_extent->list));
1736                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1737                 if (!ret) {
1738                         if (nolock)
1739                                 trans = btrfs_join_transaction_nolock(root);
1740                         else
1741                                 trans = btrfs_join_transaction(root);
1742                         BUG_ON(IS_ERR(trans));
1743                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1744                         ret = btrfs_update_inode(trans, root, inode);
1745                         BUG_ON(ret);
1746                 }
1747                 goto out;
1748         }
1749
1750         lock_extent_bits(io_tree, ordered_extent->file_offset,
1751                          ordered_extent->file_offset + ordered_extent->len - 1,
1752                          0, &cached_state, GFP_NOFS);
1753
1754         if (nolock)
1755                 trans = btrfs_join_transaction_nolock(root);
1756         else
1757                 trans = btrfs_join_transaction(root);
1758         BUG_ON(IS_ERR(trans));
1759         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1760
1761         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1762                 compress_type = ordered_extent->compress_type;
1763         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1764                 BUG_ON(compress_type);
1765                 ret = btrfs_mark_extent_written(trans, inode,
1766                                                 ordered_extent->file_offset,
1767                                                 ordered_extent->file_offset +
1768                                                 ordered_extent->len);
1769                 BUG_ON(ret);
1770         } else {
1771                 BUG_ON(root == root->fs_info->tree_root);
1772                 ret = insert_reserved_file_extent(trans, inode,
1773                                                 ordered_extent->file_offset,
1774                                                 ordered_extent->start,
1775                                                 ordered_extent->disk_len,
1776                                                 ordered_extent->len,
1777                                                 ordered_extent->len,
1778                                                 compress_type, 0, 0,
1779                                                 BTRFS_FILE_EXTENT_REG);
1780                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1781                                    ordered_extent->file_offset,
1782                                    ordered_extent->len);
1783                 BUG_ON(ret);
1784         }
1785         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1786                              ordered_extent->file_offset +
1787                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1788
1789         add_pending_csums(trans, inode, ordered_extent->file_offset,
1790                           &ordered_extent->list);
1791
1792         ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1793         if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1794                 ret = btrfs_update_inode(trans, root, inode);
1795                 BUG_ON(ret);
1796         }
1797         ret = 0;
1798 out:
1799         if (root != root->fs_info->tree_root)
1800                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1801         if (trans) {
1802                 if (nolock)
1803                         btrfs_end_transaction_nolock(trans, root);
1804                 else
1805                         btrfs_end_transaction(trans, root);
1806         }
1807
1808         /* once for us */
1809         btrfs_put_ordered_extent(ordered_extent);
1810         /* once for the tree */
1811         btrfs_put_ordered_extent(ordered_extent);
1812
1813         return 0;
1814 }
1815
1816 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1817                                 struct extent_state *state, int uptodate)
1818 {
1819         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1820
1821         ClearPagePrivate2(page);
1822         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1823 }
1824
1825 /*
1826  * When IO fails, either with EIO or csum verification fails, we
1827  * try other mirrors that might have a good copy of the data.  This
1828  * io_failure_record is used to record state as we go through all the
1829  * mirrors.  If another mirror has good data, the page is set up to date
1830  * and things continue.  If a good mirror can't be found, the original
1831  * bio end_io callback is called to indicate things have failed.
1832  */
1833 struct io_failure_record {
1834         struct page *page;
1835         u64 start;
1836         u64 len;
1837         u64 logical;
1838         unsigned long bio_flags;
1839         int last_mirror;
1840 };
1841
1842 static int btrfs_io_failed_hook(struct bio *failed_bio,
1843                          struct page *page, u64 start, u64 end,
1844                          struct extent_state *state)
1845 {
1846         struct io_failure_record *failrec = NULL;
1847         u64 private;
1848         struct extent_map *em;
1849         struct inode *inode = page->mapping->host;
1850         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1851         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1852         struct bio *bio;
1853         int num_copies;
1854         int ret;
1855         int rw;
1856         u64 logical;
1857
1858         ret = get_state_private(failure_tree, start, &private);
1859         if (ret) {
1860                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1861                 if (!failrec)
1862                         return -ENOMEM;
1863                 failrec->start = start;
1864                 failrec->len = end - start + 1;
1865                 failrec->last_mirror = 0;
1866                 failrec->bio_flags = 0;
1867
1868                 read_lock(&em_tree->lock);
1869                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1870                 if (em->start > start || em->start + em->len < start) {
1871                         free_extent_map(em);
1872                         em = NULL;
1873                 }
1874                 read_unlock(&em_tree->lock);
1875
1876                 if (IS_ERR_OR_NULL(em)) {
1877                         kfree(failrec);
1878                         return -EIO;
1879                 }
1880                 logical = start - em->start;
1881                 logical = em->block_start + logical;
1882                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1883                         logical = em->block_start;
1884                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1885                         extent_set_compress_type(&failrec->bio_flags,
1886                                                  em->compress_type);
1887                 }
1888                 failrec->logical = logical;
1889                 free_extent_map(em);
1890                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1891                                 EXTENT_DIRTY, GFP_NOFS);
1892                 set_state_private(failure_tree, start,
1893                                  (u64)(unsigned long)failrec);
1894         } else {
1895                 failrec = (struct io_failure_record *)(unsigned long)private;
1896         }
1897         num_copies = btrfs_num_copies(
1898                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1899                               failrec->logical, failrec->len);
1900         failrec->last_mirror++;
1901         if (!state) {
1902                 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1903                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1904                                                     failrec->start,
1905                                                     EXTENT_LOCKED);
1906                 if (state && state->start != failrec->start)
1907                         state = NULL;
1908                 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1909         }
1910         if (!state || failrec->last_mirror > num_copies) {
1911                 set_state_private(failure_tree, failrec->start, 0);
1912                 clear_extent_bits(failure_tree, failrec->start,
1913                                   failrec->start + failrec->len - 1,
1914                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1915                 kfree(failrec);
1916                 return -EIO;
1917         }
1918         bio = bio_alloc(GFP_NOFS, 1);
1919         bio->bi_private = state;
1920         bio->bi_end_io = failed_bio->bi_end_io;
1921         bio->bi_sector = failrec->logical >> 9;
1922         bio->bi_bdev = failed_bio->bi_bdev;
1923         bio->bi_size = 0;
1924
1925         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1926         if (failed_bio->bi_rw & REQ_WRITE)
1927                 rw = WRITE;
1928         else
1929                 rw = READ;
1930
1931         ret = BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1932                                                       failrec->last_mirror,
1933                                                       failrec->bio_flags, 0);
1934         return ret;
1935 }
1936
1937 /*
1938  * each time an IO finishes, we do a fast check in the IO failure tree
1939  * to see if we need to process or clean up an io_failure_record
1940  */
1941 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1942 {
1943         u64 private;
1944         u64 private_failure;
1945         struct io_failure_record *failure;
1946         int ret;
1947
1948         private = 0;
1949         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1950                              (u64)-1, 1, EXTENT_DIRTY, 0)) {
1951                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1952                                         start, &private_failure);
1953                 if (ret == 0) {
1954                         failure = (struct io_failure_record *)(unsigned long)
1955                                    private_failure;
1956                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1957                                           failure->start, 0);
1958                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1959                                           failure->start,
1960                                           failure->start + failure->len - 1,
1961                                           EXTENT_DIRTY | EXTENT_LOCKED,
1962                                           GFP_NOFS);
1963                         kfree(failure);
1964                 }
1965         }
1966         return 0;
1967 }
1968
1969 /*
1970  * when reads are done, we need to check csums to verify the data is correct
1971  * if there's a match, we allow the bio to finish.  If not, we go through
1972  * the io_failure_record routines to find good copies
1973  */
1974 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1975                                struct extent_state *state)
1976 {
1977         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1978         struct inode *inode = page->mapping->host;
1979         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1980         char *kaddr;
1981         u64 private = ~(u32)0;
1982         int ret;
1983         struct btrfs_root *root = BTRFS_I(inode)->root;
1984         u32 csum = ~(u32)0;
1985
1986         if (PageChecked(page)) {
1987                 ClearPageChecked(page);
1988                 goto good;
1989         }
1990
1991         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1992                 goto good;
1993
1994         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1995             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1996                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1997                                   GFP_NOFS);
1998                 return 0;
1999         }
2000
2001         if (state && state->start == start) {
2002                 private = state->private;
2003                 ret = 0;
2004         } else {
2005                 ret = get_state_private(io_tree, start, &private);
2006         }
2007         kaddr = kmap_atomic(page, KM_USER0);
2008         if (ret)
2009                 goto zeroit;
2010
2011         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
2012         btrfs_csum_final(csum, (char *)&csum);
2013         if (csum != private)
2014                 goto zeroit;
2015
2016         kunmap_atomic(kaddr, KM_USER0);
2017 good:
2018         /* if the io failure tree for this inode is non-empty,
2019          * check to see if we've recovered from a failed IO
2020          */
2021         btrfs_clean_io_failures(inode, start);
2022         return 0;
2023
2024 zeroit:
2025         printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
2026                        "private %llu\n",
2027                        (unsigned long long)btrfs_ino(page->mapping->host),
2028                        (unsigned long long)start, csum,
2029                        (unsigned long long)private);
2030         memset(kaddr + offset, 1, end - start + 1);
2031         flush_dcache_page(page);
2032         kunmap_atomic(kaddr, KM_USER0);
2033         if (private == 0)
2034                 return 0;
2035         return -EIO;
2036 }
2037
2038 struct delayed_iput {
2039         struct list_head list;
2040         struct inode *inode;
2041 };
2042
2043 void btrfs_add_delayed_iput(struct inode *inode)
2044 {
2045         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2046         struct delayed_iput *delayed;
2047
2048         if (atomic_add_unless(&inode->i_count, -1, 1))
2049                 return;
2050
2051         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2052         delayed->inode = inode;
2053
2054         spin_lock(&fs_info->delayed_iput_lock);
2055         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2056         spin_unlock(&fs_info->delayed_iput_lock);
2057 }
2058
2059 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2060 {
2061         LIST_HEAD(list);
2062         struct btrfs_fs_info *fs_info = root->fs_info;
2063         struct delayed_iput *delayed;
2064         int empty;
2065
2066         spin_lock(&fs_info->delayed_iput_lock);
2067         empty = list_empty(&fs_info->delayed_iputs);
2068         spin_unlock(&fs_info->delayed_iput_lock);
2069         if (empty)
2070                 return;
2071
2072         down_read(&root->fs_info->cleanup_work_sem);
2073         spin_lock(&fs_info->delayed_iput_lock);
2074         list_splice_init(&fs_info->delayed_iputs, &list);
2075         spin_unlock(&fs_info->delayed_iput_lock);
2076
2077         while (!list_empty(&list)) {
2078                 delayed = list_entry(list.next, struct delayed_iput, list);
2079                 list_del(&delayed->list);
2080                 iput(delayed->inode);
2081                 kfree(delayed);
2082         }
2083         up_read(&root->fs_info->cleanup_work_sem);
2084 }
2085
2086 enum btrfs_orphan_cleanup_state {
2087         ORPHAN_CLEANUP_STARTED  = 1,
2088         ORPHAN_CLEANUP_DONE     = 2,
2089 };
2090
2091 /*
2092  * This is called in transaction commmit time. If there are no orphan
2093  * files in the subvolume, it removes orphan item and frees block_rsv
2094  * structure.
2095  */
2096 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2097                               struct btrfs_root *root)
2098 {
2099         int ret;
2100
2101         if (!list_empty(&root->orphan_list) ||
2102             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2103                 return;
2104
2105         if (root->orphan_item_inserted &&
2106             btrfs_root_refs(&root->root_item) > 0) {
2107                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2108                                             root->root_key.objectid);
2109                 BUG_ON(ret);
2110                 root->orphan_item_inserted = 0;
2111         }
2112
2113         if (root->orphan_block_rsv) {
2114                 WARN_ON(root->orphan_block_rsv->size > 0);
2115                 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2116                 root->orphan_block_rsv = NULL;
2117         }
2118 }
2119
2120 /*
2121  * This creates an orphan entry for the given inode in case something goes
2122  * wrong in the middle of an unlink/truncate.
2123  *
2124  * NOTE: caller of this function should reserve 5 units of metadata for
2125  *       this function.
2126  */
2127 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2128 {
2129         struct btrfs_root *root = BTRFS_I(inode)->root;
2130         struct btrfs_block_rsv *block_rsv = NULL;
2131         int reserve = 0;
2132         int insert = 0;
2133         int ret;
2134
2135         if (!root->orphan_block_rsv) {
2136                 block_rsv = btrfs_alloc_block_rsv(root);
2137                 if (!block_rsv)
2138                         return -ENOMEM;
2139         }
2140
2141         spin_lock(&root->orphan_lock);
2142         if (!root->orphan_block_rsv) {
2143                 root->orphan_block_rsv = block_rsv;
2144         } else if (block_rsv) {
2145                 btrfs_free_block_rsv(root, block_rsv);
2146                 block_rsv = NULL;
2147         }
2148
2149         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2150                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2151 #if 0
2152                 /*
2153                  * For proper ENOSPC handling, we should do orphan
2154                  * cleanup when mounting. But this introduces backward
2155                  * compatibility issue.
2156                  */
2157                 if (!xchg(&root->orphan_item_inserted, 1))
2158                         insert = 2;
2159                 else
2160                         insert = 1;
2161 #endif
2162                 insert = 1;
2163         }
2164
2165         if (!BTRFS_I(inode)->orphan_meta_reserved) {
2166                 BTRFS_I(inode)->orphan_meta_reserved = 1;
2167                 reserve = 1;
2168         }
2169         spin_unlock(&root->orphan_lock);
2170
2171         /* grab metadata reservation from transaction handle */
2172         if (reserve) {
2173                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2174                 BUG_ON(ret);
2175         }
2176
2177         /* insert an orphan item to track this unlinked/truncated file */
2178         if (insert >= 1) {
2179                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2180                 BUG_ON(ret);
2181         }
2182
2183         /* insert an orphan item to track subvolume contains orphan files */
2184         if (insert >= 2) {
2185                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2186                                                root->root_key.objectid);
2187                 BUG_ON(ret);
2188         }
2189         return 0;
2190 }
2191
2192 /*
2193  * We have done the truncate/delete so we can go ahead and remove the orphan
2194  * item for this particular inode.
2195  */
2196 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2197 {
2198         struct btrfs_root *root = BTRFS_I(inode)->root;
2199         int delete_item = 0;
2200         int release_rsv = 0;
2201         int ret = 0;
2202
2203         spin_lock(&root->orphan_lock);
2204         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2205                 list_del_init(&BTRFS_I(inode)->i_orphan);
2206                 delete_item = 1;
2207         }
2208
2209         if (BTRFS_I(inode)->orphan_meta_reserved) {
2210                 BTRFS_I(inode)->orphan_meta_reserved = 0;
2211                 release_rsv = 1;
2212         }
2213         spin_unlock(&root->orphan_lock);
2214
2215         if (trans && delete_item) {
2216                 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2217                 BUG_ON(ret);
2218         }
2219
2220         if (release_rsv)
2221                 btrfs_orphan_release_metadata(inode);
2222
2223         return 0;
2224 }
2225
2226 /*
2227  * this cleans up any orphans that may be left on the list from the last use
2228  * of this root.
2229  */
2230 int btrfs_orphan_cleanup(struct btrfs_root *root)
2231 {
2232         struct btrfs_path *path;
2233         struct extent_buffer *leaf;
2234         struct btrfs_key key, found_key;
2235         struct btrfs_trans_handle *trans;
2236         struct inode *inode;
2237         u64 last_objectid = 0;
2238         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2239
2240         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2241                 return 0;
2242
2243         path = btrfs_alloc_path();
2244         if (!path) {
2245                 ret = -ENOMEM;
2246                 goto out;
2247         }
2248         path->reada = -1;
2249
2250         key.objectid = BTRFS_ORPHAN_OBJECTID;
2251         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2252         key.offset = (u64)-1;
2253
2254         while (1) {
2255                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2256                 if (ret < 0)
2257                         goto out;
2258
2259                 /*
2260                  * if ret == 0 means we found what we were searching for, which
2261                  * is weird, but possible, so only screw with path if we didn't
2262                  * find the key and see if we have stuff that matches
2263                  */
2264                 if (ret > 0) {
2265                         ret = 0;
2266                         if (path->slots[0] == 0)
2267                                 break;
2268                         path->slots[0]--;
2269                 }
2270
2271                 /* pull out the item */
2272                 leaf = path->nodes[0];
2273                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2274
2275                 /* make sure the item matches what we want */
2276                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2277                         break;
2278                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2279                         break;
2280
2281                 /* release the path since we're done with it */
2282                 btrfs_release_path(path);
2283
2284                 /*
2285                  * this is where we are basically btrfs_lookup, without the
2286                  * crossing root thing.  we store the inode number in the
2287                  * offset of the orphan item.
2288                  */
2289
2290                 if (found_key.offset == last_objectid) {
2291                         printk(KERN_ERR "btrfs: Error removing orphan entry, "
2292                                "stopping orphan cleanup\n");
2293                         ret = -EINVAL;
2294                         goto out;
2295                 }
2296
2297                 last_objectid = found_key.offset;
2298
2299                 found_key.objectid = found_key.offset;
2300                 found_key.type = BTRFS_INODE_ITEM_KEY;
2301                 found_key.offset = 0;
2302                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2303                 ret = PTR_RET(inode);
2304                 if (ret && ret != -ESTALE)
2305                         goto out;
2306
2307                 /*
2308                  * Inode is already gone but the orphan item is still there,
2309                  * kill the orphan item.
2310                  */
2311                 if (ret == -ESTALE) {
2312                         trans = btrfs_start_transaction(root, 1);
2313                         if (IS_ERR(trans)) {
2314                                 ret = PTR_ERR(trans);
2315                                 goto out;
2316                         }
2317                         ret = btrfs_del_orphan_item(trans, root,
2318                                                     found_key.objectid);
2319                         BUG_ON(ret);
2320                         btrfs_end_transaction(trans, root);
2321                         continue;
2322                 }
2323
2324                 /*
2325                  * add this inode to the orphan list so btrfs_orphan_del does
2326                  * the proper thing when we hit it
2327                  */
2328                 spin_lock(&root->orphan_lock);
2329                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2330                 spin_unlock(&root->orphan_lock);
2331
2332                 /* if we have links, this was a truncate, lets do that */
2333                 if (inode->i_nlink) {
2334                         if (!S_ISREG(inode->i_mode)) {
2335                                 WARN_ON(1);
2336                                 iput(inode);
2337                                 continue;
2338                         }
2339                         nr_truncate++;
2340                         ret = btrfs_truncate(inode);
2341                 } else {
2342                         nr_unlink++;
2343                 }
2344
2345                 /* this will do delete_inode and everything for us */
2346                 iput(inode);
2347                 if (ret)
2348                         goto out;
2349         }
2350         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2351
2352         if (root->orphan_block_rsv)
2353                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2354                                         (u64)-1);
2355
2356         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2357                 trans = btrfs_join_transaction(root);
2358                 if (!IS_ERR(trans))
2359                         btrfs_end_transaction(trans, root);
2360         }
2361
2362         if (nr_unlink)
2363                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2364         if (nr_truncate)
2365                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2366
2367 out:
2368         if (ret)
2369                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2370         btrfs_free_path(path);
2371         return ret;
2372 }
2373
2374 /*
2375  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2376  * don't find any xattrs, we know there can't be any acls.
2377  *
2378  * slot is the slot the inode is in, objectid is the objectid of the inode
2379  */
2380 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2381                                           int slot, u64 objectid)
2382 {
2383         u32 nritems = btrfs_header_nritems(leaf);
2384         struct btrfs_key found_key;
2385         int scanned = 0;
2386
2387         slot++;
2388         while (slot < nritems) {
2389                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2390
2391                 /* we found a different objectid, there must not be acls */
2392                 if (found_key.objectid != objectid)
2393                         return 0;
2394
2395                 /* we found an xattr, assume we've got an acl */
2396                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2397                         return 1;
2398
2399                 /*
2400                  * we found a key greater than an xattr key, there can't
2401                  * be any acls later on
2402                  */
2403                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2404                         return 0;
2405
2406                 slot++;
2407                 scanned++;
2408
2409                 /*
2410                  * it goes inode, inode backrefs, xattrs, extents,
2411                  * so if there are a ton of hard links to an inode there can
2412                  * be a lot of backrefs.  Don't waste time searching too hard,
2413                  * this is just an optimization
2414                  */
2415                 if (scanned >= 8)
2416                         break;
2417         }
2418         /* we hit the end of the leaf before we found an xattr or
2419          * something larger than an xattr.  We have to assume the inode
2420          * has acls
2421          */
2422         return 1;
2423 }
2424
2425 /*
2426  * read an inode from the btree into the in-memory inode
2427  */
2428 static void btrfs_read_locked_inode(struct inode *inode)
2429 {
2430         struct btrfs_path *path;
2431         struct extent_buffer *leaf;
2432         struct btrfs_inode_item *inode_item;
2433         struct btrfs_timespec *tspec;
2434         struct btrfs_root *root = BTRFS_I(inode)->root;
2435         struct btrfs_key location;
2436         int maybe_acls;
2437         u32 rdev;
2438         int ret;
2439         bool filled = false;
2440
2441         ret = btrfs_fill_inode(inode, &rdev);
2442         if (!ret)
2443                 filled = true;
2444
2445         path = btrfs_alloc_path();
2446         if (!path)
2447                 goto make_bad;
2448
2449         path->leave_spinning = 1;
2450         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2451
2452         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2453         if (ret)
2454                 goto make_bad;
2455
2456         leaf = path->nodes[0];
2457
2458         if (filled)
2459                 goto cache_acl;
2460
2461         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2462                                     struct btrfs_inode_item);
2463         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2464         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2465         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2466         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2467         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2468
2469         tspec = btrfs_inode_atime(inode_item);
2470         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2471         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2472
2473         tspec = btrfs_inode_mtime(inode_item);
2474         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2475         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2476
2477         tspec = btrfs_inode_ctime(inode_item);
2478         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2479         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2480
2481         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2482         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2483         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2484         inode->i_generation = BTRFS_I(inode)->generation;
2485         inode->i_rdev = 0;
2486         rdev = btrfs_inode_rdev(leaf, inode_item);
2487
2488         BTRFS_I(inode)->index_cnt = (u64)-1;
2489         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2490 cache_acl:
2491         /*
2492          * try to precache a NULL acl entry for files that don't have
2493          * any xattrs or acls
2494          */
2495         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2496                                            btrfs_ino(inode));
2497         if (!maybe_acls)
2498                 cache_no_acl(inode);
2499
2500         btrfs_free_path(path);
2501
2502         switch (inode->i_mode & S_IFMT) {
2503         case S_IFREG:
2504                 inode->i_mapping->a_ops = &btrfs_aops;
2505                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2506                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2507                 inode->i_fop = &btrfs_file_operations;
2508                 inode->i_op = &btrfs_file_inode_operations;
2509                 break;
2510         case S_IFDIR:
2511                 inode->i_fop = &btrfs_dir_file_operations;
2512                 if (root == root->fs_info->tree_root)
2513                         inode->i_op = &btrfs_dir_ro_inode_operations;
2514                 else
2515                         inode->i_op = &btrfs_dir_inode_operations;
2516                 break;
2517         case S_IFLNK:
2518                 inode->i_op = &btrfs_symlink_inode_operations;
2519                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2520                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2521                 break;
2522         default:
2523                 inode->i_op = &btrfs_special_inode_operations;
2524                 init_special_inode(inode, inode->i_mode, rdev);
2525                 break;
2526         }
2527
2528         btrfs_update_iflags(inode);
2529         return;
2530
2531 make_bad:
2532         btrfs_free_path(path);
2533         make_bad_inode(inode);
2534 }
2535
2536 /*
2537  * given a leaf and an inode, copy the inode fields into the leaf
2538  */
2539 static void fill_inode_item(struct btrfs_trans_handle *trans,
2540                             struct extent_buffer *leaf,
2541                             struct btrfs_inode_item *item,
2542                             struct inode *inode)
2543 {
2544         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2545         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2546         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2547         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2548         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2549
2550         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2551                                inode->i_atime.tv_sec);
2552         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2553                                 inode->i_atime.tv_nsec);
2554
2555         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2556                                inode->i_mtime.tv_sec);
2557         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2558                                 inode->i_mtime.tv_nsec);
2559
2560         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2561                                inode->i_ctime.tv_sec);
2562         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2563                                 inode->i_ctime.tv_nsec);
2564
2565         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2566         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2567         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2568         btrfs_set_inode_transid(leaf, item, trans->transid);
2569         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2570         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2571         btrfs_set_inode_block_group(leaf, item, 0);
2572 }
2573
2574 /*
2575  * copy everything in the in-memory inode into the btree.
2576  */
2577 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2578                                 struct btrfs_root *root, struct inode *inode)
2579 {
2580         struct btrfs_inode_item *inode_item;
2581         struct btrfs_path *path;
2582         struct extent_buffer *leaf;
2583         int ret;
2584
2585         /*
2586          * If the inode is a free space inode, we can deadlock during commit
2587          * if we put it into the delayed code.
2588          *
2589          * The data relocation inode should also be directly updated
2590          * without delay
2591          */
2592         if (!btrfs_is_free_space_inode(root, inode)
2593             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2594                 ret = btrfs_delayed_update_inode(trans, root, inode);
2595                 if (!ret)
2596                         btrfs_set_inode_last_trans(trans, inode);
2597                 return ret;
2598         }
2599
2600         path = btrfs_alloc_path();
2601         if (!path)
2602                 return -ENOMEM;
2603
2604         path->leave_spinning = 1;
2605         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2606                                  1);
2607         if (ret) {
2608                 if (ret > 0)
2609                         ret = -ENOENT;
2610                 goto failed;
2611         }
2612
2613         btrfs_unlock_up_safe(path, 1);
2614         leaf = path->nodes[0];
2615         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2616                                     struct btrfs_inode_item);
2617
2618         fill_inode_item(trans, leaf, inode_item, inode);
2619         btrfs_mark_buffer_dirty(leaf);
2620         btrfs_set_inode_last_trans(trans, inode);
2621         ret = 0;
2622 failed:
2623         btrfs_free_path(path);
2624         return ret;
2625 }
2626
2627 /*
2628  * unlink helper that gets used here in inode.c and in the tree logging
2629  * recovery code.  It remove a link in a directory with a given name, and
2630  * also drops the back refs in the inode to the directory
2631  */
2632 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2633                                 struct btrfs_root *root,
2634                                 struct inode *dir, struct inode *inode,
2635                                 const char *name, int name_len)
2636 {
2637         struct btrfs_path *path;
2638         int ret = 0;
2639         struct extent_buffer *leaf;
2640         struct btrfs_dir_item *di;
2641         struct btrfs_key key;
2642         u64 index;
2643         u64 ino = btrfs_ino(inode);
2644         u64 dir_ino = btrfs_ino(dir);
2645
2646         path = btrfs_alloc_path();
2647         if (!path) {
2648                 ret = -ENOMEM;
2649                 goto out;
2650         }
2651
2652         path->leave_spinning = 1;
2653         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2654                                     name, name_len, -1);
2655         if (IS_ERR(di)) {
2656                 ret = PTR_ERR(di);
2657                 goto err;
2658         }
2659         if (!di) {
2660                 ret = -ENOENT;
2661                 goto err;
2662         }
2663         leaf = path->nodes[0];
2664         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2665         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2666         if (ret)
2667                 goto err;
2668         btrfs_release_path(path);
2669
2670         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2671                                   dir_ino, &index);
2672         if (ret) {
2673                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2674                        "inode %llu parent %llu\n", name_len, name,
2675                        (unsigned long long)ino, (unsigned long long)dir_ino);
2676                 goto err;
2677         }
2678
2679         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2680         if (ret)
2681                 goto err;
2682
2683         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2684                                          inode, dir_ino);
2685         BUG_ON(ret != 0 && ret != -ENOENT);
2686
2687         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2688                                            dir, index);
2689         if (ret == -ENOENT)
2690                 ret = 0;
2691 err:
2692         btrfs_free_path(path);
2693         if (ret)
2694                 goto out;
2695
2696         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2697         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2698         btrfs_update_inode(trans, root, dir);
2699 out:
2700         return ret;
2701 }
2702
2703 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2704                        struct btrfs_root *root,
2705                        struct inode *dir, struct inode *inode,
2706                        const char *name, int name_len)
2707 {
2708         int ret;
2709         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2710         if (!ret) {
2711                 btrfs_drop_nlink(inode);
2712                 ret = btrfs_update_inode(trans, root, inode);
2713         }
2714         return ret;
2715 }
2716
2717
2718 /* helper to check if there is any shared block in the path */
2719 static int check_path_shared(struct btrfs_root *root,
2720                              struct btrfs_path *path)
2721 {
2722         struct extent_buffer *eb;
2723         int level;
2724         u64 refs = 1;
2725
2726         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2727                 int ret;
2728
2729                 if (!path->nodes[level])
2730                         break;
2731                 eb = path->nodes[level];
2732                 if (!btrfs_block_can_be_shared(root, eb))
2733                         continue;
2734                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2735                                                &refs, NULL);
2736                 if (refs > 1)
2737                         return 1;
2738         }
2739         return 0;
2740 }
2741
2742 /*
2743  * helper to start transaction for unlink and rmdir.
2744  *
2745  * unlink and rmdir are special in btrfs, they do not always free space.
2746  * so in enospc case, we should make sure they will free space before
2747  * allowing them to use the global metadata reservation.
2748  */
2749 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2750                                                        struct dentry *dentry)
2751 {
2752         struct btrfs_trans_handle *trans;
2753         struct btrfs_root *root = BTRFS_I(dir)->root;
2754         struct btrfs_path *path;
2755         struct btrfs_inode_ref *ref;
2756         struct btrfs_dir_item *di;
2757         struct inode *inode = dentry->d_inode;
2758         u64 index;
2759         int check_link = 1;
2760         int err = -ENOSPC;
2761         int ret;
2762         u64 ino = btrfs_ino(inode);
2763         u64 dir_ino = btrfs_ino(dir);
2764
2765         /*
2766          * 1 for the possible orphan item
2767          * 1 for the dir item
2768          * 1 for the dir index
2769          * 1 for the inode ref
2770          * 1 for the inode ref in the tree log
2771          * 2 for the dir entries in the log
2772          * 1 for the inode
2773          */
2774         trans = btrfs_start_transaction(root, 8);
2775         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2776                 return trans;
2777
2778         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2779                 return ERR_PTR(-ENOSPC);
2780
2781         /* check if there is someone else holds reference */
2782         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2783                 return ERR_PTR(-ENOSPC);
2784
2785         if (atomic_read(&inode->i_count) > 2)
2786                 return ERR_PTR(-ENOSPC);
2787
2788         if (xchg(&root->fs_info->enospc_unlink, 1))
2789                 return ERR_PTR(-ENOSPC);
2790
2791         path = btrfs_alloc_path();
2792         if (!path) {
2793                 root->fs_info->enospc_unlink = 0;
2794                 return ERR_PTR(-ENOMEM);
2795         }
2796
2797         /* 1 for the orphan item */
2798         trans = btrfs_start_transaction(root, 1);
2799         if (IS_ERR(trans)) {
2800                 btrfs_free_path(path);
2801                 root->fs_info->enospc_unlink = 0;
2802                 return trans;
2803         }
2804
2805         path->skip_locking = 1;
2806         path->search_commit_root = 1;
2807
2808         ret = btrfs_lookup_inode(trans, root, path,
2809                                 &BTRFS_I(dir)->location, 0);
2810         if (ret < 0) {
2811                 err = ret;
2812                 goto out;
2813         }
2814         if (ret == 0) {
2815                 if (check_path_shared(root, path))
2816                         goto out;
2817         } else {
2818                 check_link = 0;
2819         }
2820         btrfs_release_path(path);
2821
2822         ret = btrfs_lookup_inode(trans, root, path,
2823                                 &BTRFS_I(inode)->location, 0);
2824         if (ret < 0) {
2825                 err = ret;
2826                 goto out;
2827         }
2828         if (ret == 0) {
2829                 if (check_path_shared(root, path))
2830                         goto out;
2831         } else {
2832                 check_link = 0;
2833         }
2834         btrfs_release_path(path);
2835
2836         if (ret == 0 && S_ISREG(inode->i_mode)) {
2837                 ret = btrfs_lookup_file_extent(trans, root, path,
2838                                                ino, (u64)-1, 0);
2839                 if (ret < 0) {
2840                         err = ret;
2841                         goto out;
2842                 }
2843                 BUG_ON(ret == 0);
2844                 if (check_path_shared(root, path))
2845                         goto out;
2846                 btrfs_release_path(path);
2847         }
2848
2849         if (!check_link) {
2850                 err = 0;
2851                 goto out;
2852         }
2853
2854         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2855                                 dentry->d_name.name, dentry->d_name.len, 0);
2856         if (IS_ERR(di)) {
2857                 err = PTR_ERR(di);
2858                 goto out;
2859         }
2860         if (di) {
2861                 if (check_path_shared(root, path))
2862                         goto out;
2863         } else {
2864                 err = 0;
2865                 goto out;
2866         }
2867         btrfs_release_path(path);
2868
2869         ref = btrfs_lookup_inode_ref(trans, root, path,
2870                                 dentry->d_name.name, dentry->d_name.len,
2871                                 ino, dir_ino, 0);
2872         if (IS_ERR(ref)) {
2873                 err = PTR_ERR(ref);
2874                 goto out;
2875         }
2876         BUG_ON(!ref);
2877         if (check_path_shared(root, path))
2878                 goto out;
2879         index = btrfs_inode_ref_index(path->nodes[0], ref);
2880         btrfs_release_path(path);
2881
2882         /*
2883          * This is a commit root search, if we can lookup inode item and other
2884          * relative items in the commit root, it means the transaction of
2885          * dir/file creation has been committed, and the dir index item that we
2886          * delay to insert has also been inserted into the commit root. So
2887          * we needn't worry about the delayed insertion of the dir index item
2888          * here.
2889          */
2890         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
2891                                 dentry->d_name.name, dentry->d_name.len, 0);
2892         if (IS_ERR(di)) {
2893                 err = PTR_ERR(di);
2894                 goto out;
2895         }
2896         BUG_ON(ret == -ENOENT);
2897         if (check_path_shared(root, path))
2898                 goto out;
2899
2900         err = 0;
2901 out:
2902         btrfs_free_path(path);
2903         /* Migrate the orphan reservation over */
2904         if (!err)
2905                 err = btrfs_block_rsv_migrate(trans->block_rsv,
2906                                 &root->fs_info->global_block_rsv,
2907                                 btrfs_calc_trans_metadata_size(root, 1));
2908
2909         if (err) {
2910                 btrfs_end_transaction(trans, root);
2911                 root->fs_info->enospc_unlink = 0;
2912                 return ERR_PTR(err);
2913         }
2914
2915         trans->block_rsv = &root->fs_info->global_block_rsv;
2916         return trans;
2917 }
2918
2919 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2920                                struct btrfs_root *root)
2921 {
2922         if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2923                 BUG_ON(!root->fs_info->enospc_unlink);
2924                 root->fs_info->enospc_unlink = 0;
2925         }
2926         btrfs_end_transaction_throttle(trans, root);
2927 }
2928
2929 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2930 {
2931         struct btrfs_root *root = BTRFS_I(dir)->root;
2932         struct btrfs_trans_handle *trans;
2933         struct inode *inode = dentry->d_inode;
2934         int ret;
2935         unsigned long nr = 0;
2936
2937         trans = __unlink_start_trans(dir, dentry);
2938         if (IS_ERR(trans))
2939                 return PTR_ERR(trans);
2940
2941         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2942
2943         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2944                                  dentry->d_name.name, dentry->d_name.len);
2945         if (ret)
2946                 goto out;
2947
2948         if (inode->i_nlink == 0) {
2949                 ret = btrfs_orphan_add(trans, inode);
2950                 if (ret)
2951                         goto out;
2952         }
2953
2954 out:
2955         nr = trans->blocks_used;
2956         __unlink_end_trans(trans, root);
2957         btrfs_btree_balance_dirty(root, nr);
2958         return ret;
2959 }
2960
2961 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2962                         struct btrfs_root *root,
2963                         struct inode *dir, u64 objectid,
2964                         const char *name, int name_len)
2965 {
2966         struct btrfs_path *path;
2967         struct extent_buffer *leaf;
2968         struct btrfs_dir_item *di;
2969         struct btrfs_key key;
2970         u64 index;
2971         int ret;
2972         u64 dir_ino = btrfs_ino(dir);
2973
2974         path = btrfs_alloc_path();
2975         if (!path)
2976                 return -ENOMEM;
2977
2978         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2979                                    name, name_len, -1);
2980         BUG_ON(IS_ERR_OR_NULL(di));
2981
2982         leaf = path->nodes[0];
2983         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2984         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2985         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2986         BUG_ON(ret);
2987         btrfs_release_path(path);
2988
2989         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2990                                  objectid, root->root_key.objectid,
2991                                  dir_ino, &index, name, name_len);
2992         if (ret < 0) {
2993                 BUG_ON(ret != -ENOENT);
2994                 di = btrfs_search_dir_index_item(root, path, dir_ino,
2995                                                  name, name_len);
2996                 BUG_ON(IS_ERR_OR_NULL(di));
2997
2998                 leaf = path->nodes[0];
2999                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3000                 btrfs_release_path(path);
3001                 index = key.offset;
3002         }
3003         btrfs_release_path(path);
3004
3005         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3006         BUG_ON(ret);
3007
3008         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3009         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3010         ret = btrfs_update_inode(trans, root, dir);
3011         BUG_ON(ret);
3012
3013         btrfs_free_path(path);
3014         return 0;
3015 }
3016
3017 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3018 {
3019         struct inode *inode = dentry->d_inode;
3020         int err = 0;
3021         struct btrfs_root *root = BTRFS_I(dir)->root;
3022         struct btrfs_trans_handle *trans;
3023         unsigned long nr = 0;
3024
3025         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3026             btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3027                 return -ENOTEMPTY;
3028
3029         trans = __unlink_start_trans(dir, dentry);
3030         if (IS_ERR(trans))
3031                 return PTR_ERR(trans);
3032
3033         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3034                 err = btrfs_unlink_subvol(trans, root, dir,
3035                                           BTRFS_I(inode)->location.objectid,
3036                                           dentry->d_name.name,
3037                                           dentry->d_name.len);
3038                 goto out;
3039         }
3040
3041         err = btrfs_orphan_add(trans, inode);
3042         if (err)
3043                 goto out;
3044
3045         /* now the directory is empty */
3046         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3047                                  dentry->d_name.name, dentry->d_name.len);
3048         if (!err)
3049                 btrfs_i_size_write(inode, 0);
3050 out:
3051         nr = trans->blocks_used;
3052         __unlink_end_trans(trans, root);
3053         btrfs_btree_balance_dirty(root, nr);
3054
3055         return err;
3056 }
3057
3058 /*
3059  * this can truncate away extent items, csum items and directory items.
3060  * It starts at a high offset and removes keys until it can't find
3061  * any higher than new_size
3062  *
3063  * csum items that cross the new i_size are truncated to the new size
3064  * as well.
3065  *
3066  * min_type is the minimum key type to truncate down to.  If set to 0, this
3067  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3068  */
3069 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3070                                struct btrfs_root *root,
3071                                struct inode *inode,
3072                                u64 new_size, u32 min_type)
3073 {
3074         struct btrfs_path *path;
3075         struct extent_buffer *leaf;
3076         struct btrfs_file_extent_item *fi;
3077         struct btrfs_key key;
3078         struct btrfs_key found_key;
3079         u64 extent_start = 0;
3080         u64 extent_num_bytes = 0;
3081         u64 extent_offset = 0;
3082         u64 item_end = 0;
3083         u64 mask = root->sectorsize - 1;
3084         u32 found_type = (u8)-1;
3085         int found_extent;
3086         int del_item;
3087         int pending_del_nr = 0;
3088         int pending_del_slot = 0;
3089         int extent_type = -1;
3090         int encoding;
3091         int ret;
3092         int err = 0;
3093         u64 ino = btrfs_ino(inode);
3094
3095         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3096
3097         path = btrfs_alloc_path();
3098         if (!path)
3099                 return -ENOMEM;
3100         path->reada = -1;
3101
3102         if (root->ref_cows || root == root->fs_info->tree_root)
3103                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3104
3105         /*
3106          * This function is also used to drop the items in the log tree before
3107          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3108          * it is used to drop the loged items. So we shouldn't kill the delayed
3109          * items.
3110          */
3111         if (min_type == 0 && root == BTRFS_I(inode)->root)
3112                 btrfs_kill_delayed_inode_items(inode);
3113
3114         key.objectid = ino;
3115         key.offset = (u64)-1;
3116         key.type = (u8)-1;
3117
3118 search_again:
3119         path->leave_spinning = 1;
3120         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3121         if (ret < 0) {
3122                 err = ret;
3123                 goto out;
3124         }
3125
3126         if (ret > 0) {
3127                 /* there are no items in the tree for us to truncate, we're
3128                  * done
3129                  */
3130                 if (path->slots[0] == 0)
3131                         goto out;
3132                 path->slots[0]--;
3133         }
3134
3135         while (1) {
3136                 fi = NULL;
3137                 leaf = path->nodes[0];
3138                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3139                 found_type = btrfs_key_type(&found_key);
3140                 encoding = 0;
3141
3142                 if (found_key.objectid != ino)
3143                         break;
3144
3145                 if (found_type < min_type)
3146                         break;
3147
3148                 item_end = found_key.offset;
3149                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3150                         fi = btrfs_item_ptr(leaf, path->slots[0],
3151                                             struct btrfs_file_extent_item);
3152                         extent_type = btrfs_file_extent_type(leaf, fi);
3153                         encoding = btrfs_file_extent_compression(leaf, fi);
3154                         encoding |= btrfs_file_extent_encryption(leaf, fi);
3155                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3156
3157                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3158                                 item_end +=
3159                                     btrfs_file_extent_num_bytes(leaf, fi);
3160                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3161                                 item_end += btrfs_file_extent_inline_len(leaf,
3162                                                                          fi);
3163                         }
3164                         item_end--;
3165                 }
3166                 if (found_type > min_type) {
3167                         del_item = 1;
3168                 } else {
3169                         if (item_end < new_size)
3170                                 break;
3171                         if (found_key.offset >= new_size)
3172                                 del_item = 1;
3173                         else
3174                                 del_item = 0;
3175                 }
3176                 found_extent = 0;
3177                 /* FIXME, shrink the extent if the ref count is only 1 */
3178                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3179                         goto delete;
3180
3181                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3182                         u64 num_dec;
3183                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3184                         if (!del_item && !encoding) {
3185                                 u64 orig_num_bytes =
3186                                         btrfs_file_extent_num_bytes(leaf, fi);
3187                                 extent_num_bytes = new_size -
3188                                         found_key.offset + root->sectorsize - 1;
3189                                 extent_num_bytes = extent_num_bytes &
3190                                         ~((u64)root->sectorsize - 1);
3191                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3192                                                          extent_num_bytes);
3193                                 num_dec = (orig_num_bytes -
3194                                            extent_num_bytes);
3195                                 if (root->ref_cows && extent_start != 0)
3196                                         inode_sub_bytes(inode, num_dec);
3197                                 btrfs_mark_buffer_dirty(leaf);
3198                         } else {
3199                                 extent_num_bytes =
3200                                         btrfs_file_extent_disk_num_bytes(leaf,
3201                                                                          fi);
3202                                 extent_offset = found_key.offset -
3203                                         btrfs_file_extent_offset(leaf, fi);
3204
3205                                 /* FIXME blocksize != 4096 */
3206                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3207                                 if (extent_start != 0) {
3208                                         found_extent = 1;
3209                                         if (root->ref_cows)
3210                                                 inode_sub_bytes(inode, num_dec);
3211                                 }
3212                         }
3213                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3214                         /*
3215                          * we can't truncate inline items that have had
3216                          * special encodings
3217                          */
3218                         if (!del_item &&
3219                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3220                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3221                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3222                                 u32 size = new_size - found_key.offset;
3223
3224                                 if (root->ref_cows) {
3225                                         inode_sub_bytes(inode, item_end + 1 -
3226                                                         new_size);
3227                                 }
3228                                 size =
3229                                     btrfs_file_extent_calc_inline_size(size);
3230                                 ret = btrfs_truncate_item(trans, root, path,
3231                                                           size, 1);
3232                         } else if (root->ref_cows) {
3233                                 inode_sub_bytes(inode, item_end + 1 -
3234                                                 found_key.offset);
3235                         }
3236                 }
3237 delete:
3238                 if (del_item) {
3239                         if (!pending_del_nr) {
3240                                 /* no pending yet, add ourselves */
3241                                 pending_del_slot = path->slots[0];
3242                                 pending_del_nr = 1;
3243                         } else if (pending_del_nr &&
3244                                    path->slots[0] + 1 == pending_del_slot) {
3245                                 /* hop on the pending chunk */
3246                                 pending_del_nr++;
3247                                 pending_del_slot = path->slots[0];
3248                         } else {
3249                                 BUG();
3250                         }
3251                 } else {
3252                         break;
3253                 }
3254                 if (found_extent && (root->ref_cows ||
3255                                      root == root->fs_info->tree_root)) {
3256                         btrfs_set_path_blocking(path);
3257                         ret = btrfs_free_extent(trans, root, extent_start,
3258                                                 extent_num_bytes, 0,
3259                                                 btrfs_header_owner(leaf),
3260                                                 ino, extent_offset);
3261                         BUG_ON(ret);
3262                 }
3263
3264                 if (found_type == BTRFS_INODE_ITEM_KEY)
3265                         break;
3266
3267                 if (path->slots[0] == 0 ||
3268                     path->slots[0] != pending_del_slot) {
3269                         if (root->ref_cows &&
3270                             BTRFS_I(inode)->location.objectid !=
3271                                                 BTRFS_FREE_INO_OBJECTID) {
3272                                 err = -EAGAIN;
3273                                 goto out;
3274                         }
3275                         if (pending_del_nr) {
3276                                 ret = btrfs_del_items(trans, root, path,
3277                                                 pending_del_slot,
3278                                                 pending_del_nr);
3279                                 BUG_ON(ret);
3280                                 pending_del_nr = 0;
3281                         }
3282                         btrfs_release_path(path);
3283                         goto search_again;
3284                 } else {
3285                         path->slots[0]--;
3286                 }
3287         }
3288 out:
3289         if (pending_del_nr) {
3290                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3291                                       pending_del_nr);
3292                 BUG_ON(ret);
3293         }
3294         btrfs_free_path(path);
3295         return err;
3296 }
3297
3298 /*
3299  * taken from block_truncate_page, but does cow as it zeros out
3300  * any bytes left in the last page in the file.
3301  */
3302 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3303 {
3304         struct inode *inode = mapping->host;
3305         struct btrfs_root *root = BTRFS_I(inode)->root;
3306         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3307         struct btrfs_ordered_extent *ordered;
3308         struct extent_state *cached_state = NULL;
3309         char *kaddr;
3310         u32 blocksize = root->sectorsize;
3311         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3312         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3313         struct page *page;
3314         gfp_t mask = btrfs_alloc_write_mask(mapping);
3315         int ret = 0;
3316         u64 page_start;
3317         u64 page_end;
3318
3319         if ((offset & (blocksize - 1)) == 0)
3320                 goto out;
3321         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3322         if (ret)
3323                 goto out;
3324
3325         ret = -ENOMEM;
3326 again:
3327         page = find_or_create_page(mapping, index, mask);
3328         if (!page) {
3329                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3330                 goto out;
3331         }
3332
3333         page_start = page_offset(page);
3334         page_end = page_start + PAGE_CACHE_SIZE - 1;
3335
3336         if (!PageUptodate(page)) {
3337                 ret = btrfs_readpage(NULL, page);
3338                 lock_page(page);
3339                 if (page->mapping != mapping) {
3340                         unlock_page(page);
3341                         page_cache_release(page);
3342                         goto again;
3343                 }
3344                 if (!PageUptodate(page)) {
3345                         ret = -EIO;
3346                         goto out_unlock;
3347                 }
3348         }
3349         wait_on_page_writeback(page);
3350
3351         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3352                          GFP_NOFS);
3353         set_page_extent_mapped(page);
3354
3355         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3356         if (ordered) {
3357                 unlock_extent_cached(io_tree, page_start, page_end,
3358                                      &cached_state, GFP_NOFS);
3359                 unlock_page(page);
3360                 page_cache_release(page);
3361                 btrfs_start_ordered_extent(inode, ordered, 1);
3362                 btrfs_put_ordered_extent(ordered);
3363                 goto again;
3364         }
3365
3366         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3367                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3368                           0, 0, &cached_state, GFP_NOFS);
3369
3370         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3371                                         &cached_state);
3372         if (ret) {
3373                 unlock_extent_cached(io_tree, page_start, page_end,
3374                                      &cached_state, GFP_NOFS);
3375                 goto out_unlock;
3376         }
3377
3378         ret = 0;
3379         if (offset != PAGE_CACHE_SIZE) {
3380                 kaddr = kmap(page);
3381                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3382                 flush_dcache_page(page);
3383                 kunmap(page);
3384         }
3385         ClearPageChecked(page);
3386         set_page_dirty(page);
3387         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3388                              GFP_NOFS);
3389
3390 out_unlock:
3391         if (ret)
3392                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3393         unlock_page(page);
3394         page_cache_release(page);
3395 out:
3396         return ret;
3397 }
3398
3399 /*
3400  * This function puts in dummy file extents for the area we're creating a hole
3401  * for.  So if we are truncating this file to a larger size we need to insert
3402  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3403  * the range between oldsize and size
3404  */
3405 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3406 {
3407         struct btrfs_trans_handle *trans;
3408         struct btrfs_root *root = BTRFS_I(inode)->root;
3409         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3410         struct extent_map *em = NULL;
3411         struct extent_state *cached_state = NULL;
3412         u64 mask = root->sectorsize - 1;
3413         u64 hole_start = (oldsize + mask) & ~mask;
3414         u64 block_end = (size + mask) & ~mask;
3415         u64 last_byte;
3416         u64 cur_offset;
3417         u64 hole_size;
3418         int err = 0;
3419
3420         if (size <= hole_start)
3421                 return 0;
3422
3423         while (1) {
3424                 struct btrfs_ordered_extent *ordered;
3425                 btrfs_wait_ordered_range(inode, hole_start,
3426                                          block_end - hole_start);
3427                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3428                                  &cached_state, GFP_NOFS);
3429                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3430                 if (!ordered)
3431                         break;
3432                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3433                                      &cached_state, GFP_NOFS);
3434                 btrfs_put_ordered_extent(ordered);
3435         }
3436
3437         cur_offset = hole_start;
3438         while (1) {
3439                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3440                                 block_end - cur_offset, 0);
3441                 BUG_ON(IS_ERR_OR_NULL(em));
3442                 last_byte = min(extent_map_end(em), block_end);
3443                 last_byte = (last_byte + mask) & ~mask;
3444                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3445                         u64 hint_byte = 0;
3446                         hole_size = last_byte - cur_offset;
3447
3448                         trans = btrfs_start_transaction(root, 2);
3449                         if (IS_ERR(trans)) {
3450                                 err = PTR_ERR(trans);
3451                                 break;
3452                         }
3453
3454                         err = btrfs_drop_extents(trans, inode, cur_offset,
3455                                                  cur_offset + hole_size,
3456                                                  &hint_byte, 1);
3457                         if (err) {
3458                                 btrfs_end_transaction(trans, root);
3459                                 break;
3460                         }
3461
3462                         err = btrfs_insert_file_extent(trans, root,
3463                                         btrfs_ino(inode), cur_offset, 0,
3464                                         0, hole_size, 0, hole_size,
3465                                         0, 0, 0);
3466                         if (err) {
3467                                 btrfs_end_transaction(trans, root);
3468                                 break;
3469                         }
3470
3471                         btrfs_drop_extent_cache(inode, hole_start,
3472                                         last_byte - 1, 0);
3473
3474                         btrfs_end_transaction(trans, root);
3475                 }
3476                 free_extent_map(em);
3477                 em = NULL;
3478                 cur_offset = last_byte;
3479                 if (cur_offset >= block_end)
3480                         break;
3481         }
3482
3483         free_extent_map(em);
3484         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3485                              GFP_NOFS);
3486         return err;
3487 }
3488
3489 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3490 {
3491         loff_t oldsize = i_size_read(inode);
3492         int ret;
3493
3494         if (newsize == oldsize)
3495                 return 0;
3496
3497         if (newsize > oldsize) {
3498                 i_size_write(inode, newsize);
3499                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3500                 truncate_pagecache(inode, oldsize, newsize);
3501                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3502                 if (ret) {
3503                         btrfs_setsize(inode, oldsize);
3504                         return ret;
3505                 }
3506
3507                 mark_inode_dirty(inode);
3508         } else {
3509
3510                 /*
3511                  * We're truncating a file that used to have good data down to
3512                  * zero. Make sure it gets into the ordered flush list so that
3513                  * any new writes get down to disk quickly.
3514                  */
3515                 if (newsize == 0)
3516                         BTRFS_I(inode)->ordered_data_close = 1;
3517
3518                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3519                 truncate_setsize(inode, newsize);
3520                 ret = btrfs_truncate(inode);
3521         }
3522
3523         return ret;
3524 }
3525
3526 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3527 {
3528         struct inode *inode = dentry->d_inode;
3529         struct btrfs_root *root = BTRFS_I(inode)->root;
3530         int err;
3531
3532         if (btrfs_root_readonly(root))
3533                 return -EROFS;
3534
3535         err = inode_change_ok(inode, attr);
3536         if (err)
3537                 return err;
3538
3539         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3540                 err = btrfs_setsize(inode, attr->ia_size);
3541                 if (err)
3542                         return err;
3543         }
3544
3545         if (attr->ia_valid) {
3546                 setattr_copy(inode, attr);
3547                 mark_inode_dirty(inode);
3548
3549                 if (attr->ia_valid & ATTR_MODE)
3550                         err = btrfs_acl_chmod(inode);
3551         }
3552
3553         return err;
3554 }
3555
3556 void btrfs_evict_inode(struct inode *inode)
3557 {
3558         struct btrfs_trans_handle *trans;
3559         struct btrfs_root *root = BTRFS_I(inode)->root;
3560         struct btrfs_block_rsv *rsv, *global_rsv;
3561         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
3562         unsigned long nr;
3563         int ret;
3564
3565         trace_btrfs_inode_evict(inode);
3566
3567         truncate_inode_pages(&inode->i_data, 0);
3568         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3569                                btrfs_is_free_space_inode(root, inode)))
3570                 goto no_delete;
3571
3572         if (is_bad_inode(inode)) {
3573                 btrfs_orphan_del(NULL, inode);
3574                 goto no_delete;
3575         }
3576         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3577         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3578
3579         if (root->fs_info->log_root_recovering) {
3580                 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3581                 goto no_delete;
3582         }
3583
3584         if (inode->i_nlink > 0) {
3585                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3586                 goto no_delete;
3587         }
3588
3589         rsv = btrfs_alloc_block_rsv(root);
3590         if (!rsv) {
3591                 btrfs_orphan_del(NULL, inode);
3592                 goto no_delete;
3593         }
3594         rsv->size = min_size;
3595         global_rsv = &root->fs_info->global_block_rsv;
3596
3597         btrfs_i_size_write(inode, 0);
3598
3599         /*
3600          * This is a bit simpler than btrfs_truncate since
3601          *
3602          * 1) We've already reserved our space for our orphan item in the
3603          *    unlink.
3604          * 2) We're going to delete the inode item, so we don't need to update
3605          *    it at all.
3606          *
3607          * So we just need to reserve some slack space in case we add bytes when
3608          * doing the truncate.
3609          */
3610         while (1) {
3611                 ret = btrfs_block_rsv_refill(root, rsv, min_size);
3612
3613                 /*
3614                  * Try and steal from the global reserve since we will
3615                  * likely not use this space anyway, we want to try as
3616                  * hard as possible to get this to work.
3617                  */
3618                 if (ret)
3619                         ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
3620
3621                 if (ret) {
3622                         printk(KERN_WARNING "Could not get space for a "
3623                                "delete, will truncate on mount %d\n", ret);
3624                         btrfs_orphan_del(NULL, inode);
3625                         btrfs_free_block_rsv(root, rsv);
3626                         goto no_delete;
3627                 }
3628
3629                 trans = btrfs_start_transaction(root, 0);
3630                 if (IS_ERR(trans)) {
3631                         btrfs_orphan_del(NULL, inode);
3632                         btrfs_free_block_rsv(root, rsv);
3633                         goto no_delete;
3634                 }
3635
3636                 trans->block_rsv = rsv;
3637
3638                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3639                 if (ret != -EAGAIN)
3640                         break;
3641
3642                 nr = trans->blocks_used;
3643                 btrfs_end_transaction(trans, root);
3644                 trans = NULL;
3645                 btrfs_btree_balance_dirty(root, nr);
3646         }
3647
3648         btrfs_free_block_rsv(root, rsv);
3649
3650         if (ret == 0) {
3651                 trans->block_rsv = root->orphan_block_rsv;
3652                 ret = btrfs_orphan_del(trans, inode);
3653                 BUG_ON(ret);
3654         }
3655
3656         trans->block_rsv = &root->fs_info->trans_block_rsv;
3657         if (!(root == root->fs_info->tree_root ||
3658               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3659                 btrfs_return_ino(root, btrfs_ino(inode));
3660
3661         nr = trans->blocks_used;
3662         btrfs_end_transaction(trans, root);
3663         btrfs_btree_balance_dirty(root, nr);
3664 no_delete:
3665         end_writeback(inode);
3666         return;
3667 }
3668
3669 /*
3670  * this returns the key found in the dir entry in the location pointer.
3671  * If no dir entries were found, location->objectid is 0.
3672  */
3673 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3674                                struct btrfs_key *location)
3675 {
3676         const char *name = dentry->d_name.name;
3677         int namelen = dentry->d_name.len;
3678         struct btrfs_dir_item *di;
3679         struct btrfs_path *path;
3680         struct btrfs_root *root = BTRFS_I(dir)->root;
3681         int ret = 0;
3682
3683         path = btrfs_alloc_path();
3684         if (!path)
3685                 return -ENOMEM;
3686
3687         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3688                                     namelen, 0);
3689         if (IS_ERR(di))
3690                 ret = PTR_ERR(di);
3691
3692         if (IS_ERR_OR_NULL(di))
3693                 goto out_err;
3694
3695         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3696 out:
3697         btrfs_free_path(path);
3698         return ret;
3699 out_err:
3700         location->objectid = 0;
3701         goto out;
3702 }
3703
3704 /*
3705  * when we hit a tree root in a directory, the btrfs part of the inode
3706  * needs to be changed to reflect the root directory of the tree root.  This
3707  * is kind of like crossing a mount point.
3708  */
3709 static int fixup_tree_root_location(struct btrfs_root *root,
3710                                     struct inode *dir,
3711                                     struct dentry *dentry,
3712                                     struct btrfs_key *location,
3713                                     struct btrfs_root **sub_root)
3714 {
3715         struct btrfs_path *path;
3716         struct btrfs_root *new_root;
3717         struct btrfs_root_ref *ref;
3718         struct extent_buffer *leaf;
3719         int ret;
3720         int err = 0;
3721
3722         path = btrfs_alloc_path();
3723         if (!path) {
3724                 err = -ENOMEM;
3725                 goto out;
3726         }
3727
3728         err = -ENOENT;
3729         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3730                                   BTRFS_I(dir)->root->root_key.objectid,
3731                                   location->objectid);
3732         if (ret) {
3733                 if (ret < 0)
3734                         err = ret;
3735                 goto out;
3736         }
3737
3738         leaf = path->nodes[0];
3739         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3740         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3741             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3742                 goto out;
3743
3744         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3745                                    (unsigned long)(ref + 1),
3746                                    dentry->d_name.len);
3747         if (ret)
3748                 goto out;
3749
3750         btrfs_release_path(path);
3751
3752         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3753         if (IS_ERR(new_root)) {
3754                 err = PTR_ERR(new_root);
3755                 goto out;
3756         }
3757
3758         if (btrfs_root_refs(&new_root->root_item) == 0) {
3759                 err = -ENOENT;
3760                 goto out;
3761         }
3762
3763         *sub_root = new_root;
3764         location->objectid = btrfs_root_dirid(&new_root->root_item);
3765         location->type = BTRFS_INODE_ITEM_KEY;
3766         location->offset = 0;
3767         err = 0;
3768 out:
3769         btrfs_free_path(path);
3770         return err;
3771 }
3772
3773 static void inode_tree_add(struct inode *inode)
3774 {
3775         struct btrfs_root *root = BTRFS_I(inode)->root;
3776         struct btrfs_inode *entry;
3777         struct rb_node **p;
3778         struct rb_node *parent;
3779         u64 ino = btrfs_ino(inode);
3780 again:
3781         p = &root->inode_tree.rb_node;
3782         parent = NULL;
3783
3784         if (inode_unhashed(inode))
3785                 return;
3786
3787         spin_lock(&root->inode_lock);
3788         while (*p) {
3789                 parent = *p;
3790                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3791
3792                 if (ino < btrfs_ino(&entry->vfs_inode))
3793                         p = &parent->rb_left;
3794                 else if (ino > btrfs_ino(&entry->vfs_inode))
3795                         p = &parent->rb_right;
3796                 else {
3797                         WARN_ON(!(entry->vfs_inode.i_state &
3798                                   (I_WILL_FREE | I_FREEING)));
3799                         rb_erase(parent, &root->inode_tree);
3800                         RB_CLEAR_NODE(parent);
3801                         spin_unlock(&root->inode_lock);
3802                         goto again;
3803                 }
3804         }
3805         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3806         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3807         spin_unlock(&root->inode_lock);
3808 }
3809
3810 static void inode_tree_del(struct inode *inode)
3811 {
3812         struct btrfs_root *root = BTRFS_I(inode)->root;
3813         int empty = 0;
3814
3815         spin_lock(&root->inode_lock);
3816         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3817                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3818                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3819                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3820         }
3821         spin_unlock(&root->inode_lock);
3822
3823         /*
3824          * Free space cache has inodes in the tree root, but the tree root has a
3825          * root_refs of 0, so this could end up dropping the tree root as a
3826          * snapshot, so we need the extra !root->fs_info->tree_root check to
3827          * make sure we don't drop it.
3828          */
3829         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3830             root != root->fs_info->tree_root) {
3831                 synchronize_srcu(&root->fs_info->subvol_srcu);
3832                 spin_lock(&root->inode_lock);
3833                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3834                 spin_unlock(&root->inode_lock);
3835                 if (empty)
3836                         btrfs_add_dead_root(root);
3837         }
3838 }
3839
3840 int btrfs_invalidate_inodes(struct btrfs_root *root)
3841 {
3842         struct rb_node *node;
3843         struct rb_node *prev;
3844         struct btrfs_inode *entry;
3845         struct inode *inode;
3846         u64 objectid = 0;
3847
3848         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3849
3850         spin_lock(&root->inode_lock);
3851 again:
3852         node = root->inode_tree.rb_node;
3853         prev = NULL;
3854         while (node) {
3855                 prev = node;
3856                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3857
3858                 if (objectid < btrfs_ino(&entry->vfs_inode))
3859                         node = node->rb_left;
3860                 else if (objectid > btrfs_ino(&entry->vfs_inode))
3861                         node = node->rb_right;
3862                 else
3863                         break;
3864         }
3865         if (!node) {
3866                 while (prev) {
3867                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
3868                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
3869                                 node = prev;
3870                                 break;
3871                         }
3872                         prev = rb_next(prev);
3873                 }
3874         }
3875         while (node) {
3876                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3877                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
3878                 inode = igrab(&entry->vfs_inode);
3879                 if (inode) {
3880                         spin_unlock(&root->inode_lock);
3881                         if (atomic_read(&inode->i_count) > 1)
3882                                 d_prune_aliases(inode);
3883                         /*
3884                          * btrfs_drop_inode will have it removed from
3885                          * the inode cache when its usage count
3886                          * hits zero.
3887                          */
3888                         iput(inode);
3889                         cond_resched();
3890                         spin_lock(&root->inode_lock);
3891                         goto again;
3892                 }
3893
3894                 if (cond_resched_lock(&root->inode_lock))
3895                         goto again;
3896
3897                 node = rb_next(node);
3898         }
3899         spin_unlock(&root->inode_lock);
3900         return 0;
3901 }
3902
3903 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3904 {
3905         struct btrfs_iget_args *args = p;
3906         inode->i_ino = args->ino;
3907         BTRFS_I(inode)->root = args->root;
3908         btrfs_set_inode_space_info(args->root, inode);
3909         return 0;
3910 }
3911
3912 static int btrfs_find_actor(struct inode *inode, void *opaque)
3913 {
3914         struct btrfs_iget_args *args = opaque;
3915         return args->ino == btrfs_ino(inode) &&
3916                 args->root == BTRFS_I(inode)->root;
3917 }
3918
3919 static struct inode *btrfs_iget_locked(struct super_block *s,
3920                                        u64 objectid,
3921                                        struct btrfs_root *root)
3922 {
3923         struct inode *inode;
3924         struct btrfs_iget_args args;
3925         args.ino = objectid;
3926         args.root = root;
3927
3928         inode = iget5_locked(s, objectid, btrfs_find_actor,
3929                              btrfs_init_locked_inode,
3930                              (void *)&args);
3931         return inode;
3932 }
3933
3934 /* Get an inode object given its location and corresponding root.
3935  * Returns in *is_new if the inode was read from disk
3936  */
3937 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3938                          struct btrfs_root *root, int *new)
3939 {
3940         struct inode *inode;
3941
3942         inode = btrfs_iget_locked(s, location->objectid, root);
3943         if (!inode)
3944                 return ERR_PTR(-ENOMEM);
3945
3946         if (inode->i_state & I_NEW) {
3947                 BTRFS_I(inode)->root = root;
3948                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3949                 btrfs_read_locked_inode(inode);
3950                 if (!is_bad_inode(inode)) {
3951                         inode_tree_add(inode);
3952                         unlock_new_inode(inode);
3953                         if (new)
3954                                 *new = 1;
3955                 } else {
3956                         unlock_new_inode(inode);
3957                         iput(inode);
3958                         inode = ERR_PTR(-ESTALE);
3959                 }
3960         }
3961
3962         return inode;
3963 }
3964
3965 static struct inode *new_simple_dir(struct super_block *s,
3966                                     struct btrfs_key *key,
3967                                     struct btrfs_root *root)
3968 {
3969         struct inode *inode = new_inode(s);
3970
3971         if (!inode)
3972                 return ERR_PTR(-ENOMEM);
3973
3974         BTRFS_I(inode)->root = root;
3975         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
3976         BTRFS_I(inode)->dummy_inode = 1;
3977
3978         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
3979         inode->i_op = &simple_dir_inode_operations;
3980         inode->i_fop = &simple_dir_operations;
3981         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
3982         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3983
3984         return inode;
3985 }
3986
3987 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3988 {
3989         struct inode *inode;
3990         struct btrfs_root *root = BTRFS_I(dir)->root;
3991         struct btrfs_root *sub_root = root;
3992         struct btrfs_key location;
3993         int index;
3994         int ret = 0;
3995
3996         if (dentry->d_name.len > BTRFS_NAME_LEN)
3997                 return ERR_PTR(-ENAMETOOLONG);
3998
3999         if (unlikely(d_need_lookup(dentry))) {
4000                 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
4001                 kfree(dentry->d_fsdata);
4002                 dentry->d_fsdata = NULL;
4003                 /* This thing is hashed, drop it for now */
4004                 d_drop(dentry);
4005         } else {
4006                 ret = btrfs_inode_by_name(dir, dentry, &location);
4007         }
4008
4009         if (ret < 0)
4010                 return ERR_PTR(ret);
4011
4012         if (location.objectid == 0)
4013                 return NULL;
4014
4015         if (location.type == BTRFS_INODE_ITEM_KEY) {
4016                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4017                 return inode;
4018         }
4019
4020         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4021
4022         index = srcu_read_lock(&root->fs_info->subvol_srcu);
4023         ret = fixup_tree_root_location(root, dir, dentry,
4024                                        &location, &sub_root);
4025         if (ret < 0) {
4026                 if (ret != -ENOENT)
4027                         inode = ERR_PTR(ret);
4028                 else
4029                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
4030         } else {
4031                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4032         }
4033         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4034
4035         if (!IS_ERR(inode) && root != sub_root) {
4036                 down_read(&root->fs_info->cleanup_work_sem);
4037                 if (!(inode->i_sb->s_flags & MS_RDONLY))
4038                         ret = btrfs_orphan_cleanup(sub_root);
4039                 up_read(&root->fs_info->cleanup_work_sem);
4040                 if (ret)
4041                         inode = ERR_PTR(ret);
4042         }
4043
4044         return inode;
4045 }
4046
4047 static int btrfs_dentry_delete(const struct dentry *dentry)
4048 {
4049         struct btrfs_root *root;
4050
4051         if (!dentry->d_inode && !IS_ROOT(dentry))
4052                 dentry = dentry->d_parent;
4053
4054         if (dentry->d_inode) {
4055                 root = BTRFS_I(dentry->d_inode)->root;
4056                 if (btrfs_root_refs(&root->root_item) == 0)
4057                         return 1;
4058         }
4059         return 0;
4060 }
4061
4062 static void btrfs_dentry_release(struct dentry *dentry)
4063 {
4064         if (dentry->d_fsdata)
4065                 kfree(dentry->d_fsdata);
4066 }
4067
4068 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4069                                    struct nameidata *nd)
4070 {
4071         struct dentry *ret;
4072
4073         ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4074         if (unlikely(d_need_lookup(dentry))) {
4075                 spin_lock(&dentry->d_lock);
4076                 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4077                 spin_unlock(&dentry->d_lock);
4078         }
4079         return ret;
4080 }
4081
4082 unsigned char btrfs_filetype_table[] = {
4083         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4084 };
4085
4086 static int btrfs_real_readdir(struct file *filp, void *dirent,
4087                               filldir_t filldir)
4088 {
4089         struct inode *inode = filp->f_dentry->d_inode;
4090         struct btrfs_root *root = BTRFS_I(inode)->root;
4091         struct btrfs_item *item;
4092         struct btrfs_dir_item *di;
4093         struct btrfs_key key;
4094         struct btrfs_key found_key;
4095         struct btrfs_path *path;
4096         struct list_head ins_list;
4097         struct list_head del_list;
4098         struct qstr q;
4099         int ret;
4100         struct extent_buffer *leaf;
4101         int slot;
4102         unsigned char d_type;
4103         int over = 0;
4104         u32 di_cur;
4105         u32 di_total;
4106         u32 di_len;
4107         int key_type = BTRFS_DIR_INDEX_KEY;
4108         char tmp_name[32];
4109         char *name_ptr;
4110         int name_len;
4111         int is_curr = 0;        /* filp->f_pos points to the current index? */
4112
4113         /* FIXME, use a real flag for deciding about the key type */
4114         if (root->fs_info->tree_root == root)
4115                 key_type = BTRFS_DIR_ITEM_KEY;
4116
4117         /* special case for "." */
4118         if (filp->f_pos == 0) {
4119                 over = filldir(dirent, ".", 1,
4120                                filp->f_pos, btrfs_ino(inode), DT_DIR);
4121                 if (over)
4122                         return 0;
4123                 filp->f_pos = 1;
4124         }
4125         /* special case for .., just use the back ref */
4126         if (filp->f_pos == 1) {
4127                 u64 pino = parent_ino(filp->f_path.dentry);
4128                 over = filldir(dirent, "..", 2,
4129                                filp->f_pos, pino, DT_DIR);
4130                 if (over)
4131                         return 0;
4132                 filp->f_pos = 2;
4133         }
4134         path = btrfs_alloc_path();
4135         if (!path)
4136                 return -ENOMEM;
4137
4138         path->reada = 1;
4139
4140         if (key_type == BTRFS_DIR_INDEX_KEY) {
4141                 INIT_LIST_HEAD(&ins_list);
4142                 INIT_LIST_HEAD(&del_list);
4143                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4144         }
4145
4146         btrfs_set_key_type(&key, key_type);
4147         key.offset = filp->f_pos;
4148         key.objectid = btrfs_ino(inode);
4149
4150         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4151         if (ret < 0)
4152                 goto err;
4153
4154         while (1) {
4155                 leaf = path->nodes[0];
4156                 slot = path->slots[0];
4157                 if (slot >= btrfs_header_nritems(leaf)) {
4158                         ret = btrfs_next_leaf(root, path);
4159                         if (ret < 0)
4160                                 goto err;
4161                         else if (ret > 0)
4162                                 break;
4163                         continue;
4164                 }
4165
4166                 item = btrfs_item_nr(leaf, slot);
4167                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4168
4169                 if (found_key.objectid != key.objectid)
4170                         break;
4171                 if (btrfs_key_type(&found_key) != key_type)
4172                         break;
4173                 if (found_key.offset < filp->f_pos)
4174                         goto next;
4175                 if (key_type == BTRFS_DIR_INDEX_KEY &&
4176                     btrfs_should_delete_dir_index(&del_list,
4177                                                   found_key.offset))
4178                         goto next;
4179
4180                 filp->f_pos = found_key.offset;
4181                 is_curr = 1;
4182
4183                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4184                 di_cur = 0;
4185                 di_total = btrfs_item_size(leaf, item);
4186
4187                 while (di_cur < di_total) {
4188                         struct btrfs_key location;
4189                         struct dentry *tmp;
4190
4191                         if (verify_dir_item(root, leaf, di))
4192                                 break;
4193
4194                         name_len = btrfs_dir_name_len(leaf, di);
4195                         if (name_len <= sizeof(tmp_name)) {
4196                                 name_ptr = tmp_name;
4197                         } else {
4198                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4199                                 if (!name_ptr) {
4200                                         ret = -ENOMEM;
4201                                         goto err;
4202                                 }
4203                         }
4204                         read_extent_buffer(leaf, name_ptr,
4205                                            (unsigned long)(di + 1), name_len);
4206
4207                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4208                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4209
4210                         q.name = name_ptr;
4211                         q.len = name_len;
4212                         q.hash = full_name_hash(q.name, q.len);
4213                         tmp = d_lookup(filp->f_dentry, &q);
4214                         if (!tmp) {
4215                                 struct btrfs_key *newkey;
4216
4217                                 newkey = kzalloc(sizeof(struct btrfs_key),
4218                                                  GFP_NOFS);
4219                                 if (!newkey)
4220                                         goto no_dentry;
4221                                 tmp = d_alloc(filp->f_dentry, &q);
4222                                 if (!tmp) {
4223                                         kfree(newkey);
4224                                         dput(tmp);
4225                                         goto no_dentry;
4226                                 }
4227                                 memcpy(newkey, &location,
4228                                        sizeof(struct btrfs_key));
4229                                 tmp->d_fsdata = newkey;
4230                                 tmp->d_flags |= DCACHE_NEED_LOOKUP;
4231                                 d_rehash(tmp);
4232                                 dput(tmp);
4233                         } else {
4234                                 dput(tmp);
4235                         }
4236 no_dentry:
4237                         /* is this a reference to our own snapshot? If so
4238                          * skip it
4239                          */
4240                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4241                             location.objectid == root->root_key.objectid) {
4242                                 over = 0;
4243                                 goto skip;
4244                         }
4245                         over = filldir(dirent, name_ptr, name_len,
4246                                        found_key.offset, location.objectid,
4247                                        d_type);
4248
4249 skip:
4250                         if (name_ptr != tmp_name)
4251                                 kfree(name_ptr);
4252
4253                         if (over)
4254                                 goto nopos;
4255                         di_len = btrfs_dir_name_len(leaf, di) +
4256                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4257                         di_cur += di_len;
4258                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4259                 }
4260 next:
4261                 path->slots[0]++;
4262         }
4263
4264         if (key_type == BTRFS_DIR_INDEX_KEY) {
4265                 if (is_curr)
4266                         filp->f_pos++;
4267                 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4268                                                       &ins_list);
4269                 if (ret)
4270                         goto nopos;
4271         }
4272
4273         /* Reached end of directory/root. Bump pos past the last item. */
4274         if (key_type == BTRFS_DIR_INDEX_KEY)
4275                 /*
4276                  * 32-bit glibc will use getdents64, but then strtol -
4277                  * so the last number we can serve is this.
4278                  */
4279                 filp->f_pos = 0x7fffffff;
4280         else
4281                 filp->f_pos++;
4282 nopos:
4283         ret = 0;
4284 err:
4285         if (key_type == BTRFS_DIR_INDEX_KEY)
4286                 btrfs_put_delayed_items(&ins_list, &del_list);
4287         btrfs_free_path(path);
4288         return ret;
4289 }
4290
4291 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4292 {
4293         struct btrfs_root *root = BTRFS_I(inode)->root;
4294         struct btrfs_trans_handle *trans;
4295         int ret = 0;
4296         bool nolock = false;
4297
4298         if (BTRFS_I(inode)->dummy_inode)
4299                 return 0;
4300
4301         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(root, inode))
4302                 nolock = true;
4303
4304         if (wbc->sync_mode == WB_SYNC_ALL) {
4305                 if (nolock)
4306                         trans = btrfs_join_transaction_nolock(root);
4307                 else
4308                         trans = btrfs_join_transaction(root);
4309                 if (IS_ERR(trans))
4310                         return PTR_ERR(trans);
4311                 if (nolock)
4312                         ret = btrfs_end_transaction_nolock(trans, root);
4313                 else
4314                         ret = btrfs_commit_transaction(trans, root);
4315         }
4316         return ret;
4317 }
4318
4319 /*
4320  * This is somewhat expensive, updating the tree every time the
4321  * inode changes.  But, it is most likely to find the inode in cache.
4322  * FIXME, needs more benchmarking...there are no reasons other than performance
4323  * to keep or drop this code.
4324  */
4325 void btrfs_dirty_inode(struct inode *inode, int flags)
4326 {
4327         struct btrfs_root *root = BTRFS_I(inode)->root;
4328         struct btrfs_trans_handle *trans;
4329         int ret;
4330
4331         if (BTRFS_I(inode)->dummy_inode)
4332                 return;
4333
4334         trans = btrfs_join_transaction(root);
4335         BUG_ON(IS_ERR(trans));
4336
4337         ret = btrfs_update_inode(trans, root, inode);
4338         if (ret && ret == -ENOSPC) {
4339                 /* whoops, lets try again with the full transaction */
4340                 btrfs_end_transaction(trans, root);
4341                 trans = btrfs_start_transaction(root, 1);
4342                 if (IS_ERR(trans)) {
4343                         printk_ratelimited(KERN_ERR "btrfs: fail to "
4344                                        "dirty  inode %llu error %ld\n",
4345                                        (unsigned long long)btrfs_ino(inode),
4346                                        PTR_ERR(trans));
4347                         return;
4348                 }
4349
4350                 ret = btrfs_update_inode(trans, root, inode);
4351                 if (ret) {
4352                         printk_ratelimited(KERN_ERR "btrfs: fail to "
4353                                        "dirty  inode %llu error %d\n",
4354                                        (unsigned long long)btrfs_ino(inode),
4355                                        ret);
4356                 }
4357         }
4358         btrfs_end_transaction(trans, root);
4359         if (BTRFS_I(inode)->delayed_node)
4360                 btrfs_balance_delayed_items(root);
4361 }
4362
4363 /*
4364  * find the highest existing sequence number in a directory
4365  * and then set the in-memory index_cnt variable to reflect
4366  * free sequence numbers
4367  */
4368 static int btrfs_set_inode_index_count(struct inode *inode)
4369 {
4370         struct btrfs_root *root = BTRFS_I(inode)->root;
4371         struct btrfs_key key, found_key;
4372         struct btrfs_path *path;
4373         struct extent_buffer *leaf;
4374         int ret;
4375
4376         key.objectid = btrfs_ino(inode);
4377         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4378         key.offset = (u64)-1;
4379
4380         path = btrfs_alloc_path();
4381         if (!path)
4382                 return -ENOMEM;
4383
4384         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4385         if (ret < 0)
4386                 goto out;
4387         /* FIXME: we should be able to handle this */
4388         if (ret == 0)
4389                 goto out;
4390         ret = 0;
4391
4392         /*
4393          * MAGIC NUMBER EXPLANATION:
4394          * since we search a directory based on f_pos we have to start at 2
4395          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4396          * else has to start at 2
4397          */
4398         if (path->slots[0] == 0) {
4399                 BTRFS_I(inode)->index_cnt = 2;
4400                 goto out;
4401         }
4402
4403         path->slots[0]--;
4404
4405         leaf = path->nodes[0];
4406         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4407
4408         if (found_key.objectid != btrfs_ino(inode) ||
4409             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4410                 BTRFS_I(inode)->index_cnt = 2;
4411                 goto out;
4412         }
4413
4414         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4415 out:
4416         btrfs_free_path(path);
4417         return ret;
4418 }
4419
4420 /*
4421  * helper to find a free sequence number in a given directory.  This current
4422  * code is very simple, later versions will do smarter things in the btree
4423  */
4424 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4425 {
4426         int ret = 0;
4427
4428         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4429                 ret = btrfs_inode_delayed_dir_index_count(dir);
4430                 if (ret) {
4431                         ret = btrfs_set_inode_index_count(dir);
4432                         if (ret)
4433                                 return ret;
4434                 }
4435         }
4436
4437         *index = BTRFS_I(dir)->index_cnt;
4438         BTRFS_I(dir)->index_cnt++;
4439
4440         return ret;
4441 }
4442
4443 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4444                                      struct btrfs_root *root,
4445                                      struct inode *dir,
4446                                      const char *name, int name_len,
4447                                      u64 ref_objectid, u64 objectid, int mode,
4448                                      u64 *index)
4449 {
4450         struct inode *inode;
4451         struct btrfs_inode_item *inode_item;
4452         struct btrfs_key *location;
4453         struct btrfs_path *path;
4454         struct btrfs_inode_ref *ref;
4455         struct btrfs_key key[2];
4456         u32 sizes[2];
4457         unsigned long ptr;
4458         int ret;
4459         int owner;
4460
4461         path = btrfs_alloc_path();
4462         if (!path)
4463                 return ERR_PTR(-ENOMEM);
4464
4465         inode = new_inode(root->fs_info->sb);
4466         if (!inode) {
4467                 btrfs_free_path(path);
4468                 return ERR_PTR(-ENOMEM);
4469         }
4470
4471         /*
4472          * we have to initialize this early, so we can reclaim the inode
4473          * number if we fail afterwards in this function.
4474          */
4475         inode->i_ino = objectid;
4476
4477         if (dir) {
4478                 trace_btrfs_inode_request(dir);
4479
4480                 ret = btrfs_set_inode_index(dir, index);
4481                 if (ret) {
4482                         btrfs_free_path(path);
4483                         iput(inode);
4484                         return ERR_PTR(ret);
4485                 }
4486         }
4487         /*
4488          * index_cnt is ignored for everything but a dir,
4489          * btrfs_get_inode_index_count has an explanation for the magic
4490          * number
4491          */
4492         BTRFS_I(inode)->index_cnt = 2;
4493         BTRFS_I(inode)->root = root;
4494         BTRFS_I(inode)->generation = trans->transid;
4495         inode->i_generation = BTRFS_I(inode)->generation;
4496         btrfs_set_inode_space_info(root, inode);
4497
4498         if (S_ISDIR(mode))
4499                 owner = 0;
4500         else
4501                 owner = 1;
4502
4503         key[0].objectid = objectid;
4504         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4505         key[0].offset = 0;
4506
4507         key[1].objectid = objectid;
4508         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4509         key[1].offset = ref_objectid;
4510
4511         sizes[0] = sizeof(struct btrfs_inode_item);
4512         sizes[1] = name_len + sizeof(*ref);
4513
4514         path->leave_spinning = 1;
4515         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4516         if (ret != 0)
4517                 goto fail;
4518
4519         inode_init_owner(inode, dir, mode);
4520         inode_set_bytes(inode, 0);
4521         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4522         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4523                                   struct btrfs_inode_item);
4524         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4525
4526         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4527                              struct btrfs_inode_ref);
4528         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4529         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4530         ptr = (unsigned long)(ref + 1);
4531         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4532
4533         btrfs_mark_buffer_dirty(path->nodes[0]);
4534         btrfs_free_path(path);
4535
4536         location = &BTRFS_I(inode)->location;
4537         location->objectid = objectid;
4538         location->offset = 0;
4539         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4540
4541         btrfs_inherit_iflags(inode, dir);
4542
4543         if (S_ISREG(mode)) {
4544                 if (btrfs_test_opt(root, NODATASUM))
4545                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4546                 if (btrfs_test_opt(root, NODATACOW) ||
4547                     (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4548                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4549         }
4550
4551         insert_inode_hash(inode);
4552         inode_tree_add(inode);
4553
4554         trace_btrfs_inode_new(inode);
4555         btrfs_set_inode_last_trans(trans, inode);
4556
4557         return inode;
4558 fail:
4559         if (dir)
4560                 BTRFS_I(dir)->index_cnt--;
4561         btrfs_free_path(path);
4562         iput(inode);
4563         return ERR_PTR(ret);
4564 }
4565
4566 static inline u8 btrfs_inode_type(struct inode *inode)
4567 {
4568         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4569 }
4570
4571 /*
4572  * utility function to add 'inode' into 'parent_inode' with
4573  * a give name and a given sequence number.
4574  * if 'add_backref' is true, also insert a backref from the
4575  * inode to the parent directory.
4576  */
4577 int btrfs_add_link(struct btrfs_trans_handle *trans,
4578                    struct inode *parent_inode, struct inode *inode,
4579                    const char *name, int name_len, int add_backref, u64 index)
4580 {
4581         int ret = 0;
4582         struct btrfs_key key;
4583         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4584         u64 ino = btrfs_ino(inode);
4585         u64 parent_ino = btrfs_ino(parent_inode);
4586
4587         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4588                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4589         } else {
4590                 key.objectid = ino;
4591                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4592                 key.offset = 0;
4593         }
4594
4595         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4596                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4597                                          key.objectid, root->root_key.objectid,
4598                                          parent_ino, index, name, name_len);
4599         } else if (add_backref) {
4600                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4601                                              parent_ino, index);
4602         }
4603
4604         if (ret == 0) {
4605                 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4606                                             parent_inode, &key,
4607                                             btrfs_inode_type(inode), index);
4608                 BUG_ON(ret);
4609
4610                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4611                                    name_len * 2);
4612                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4613                 ret = btrfs_update_inode(trans, root, parent_inode);
4614         }
4615         return ret;
4616 }
4617
4618 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4619                             struct inode *dir, struct dentry *dentry,
4620                             struct inode *inode, int backref, u64 index)
4621 {
4622         int err = btrfs_add_link(trans, dir, inode,
4623                                  dentry->d_name.name, dentry->d_name.len,
4624                                  backref, index);
4625         if (!err) {
4626                 d_instantiate(dentry, inode);
4627                 return 0;
4628         }
4629         if (err > 0)
4630                 err = -EEXIST;
4631         return err;
4632 }
4633
4634 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4635                         int mode, dev_t rdev)
4636 {
4637         struct btrfs_trans_handle *trans;
4638         struct btrfs_root *root = BTRFS_I(dir)->root;
4639         struct inode *inode = NULL;
4640         int err;
4641         int drop_inode = 0;
4642         u64 objectid;
4643         unsigned long nr = 0;
4644         u64 index = 0;
4645
4646         if (!new_valid_dev(rdev))
4647                 return -EINVAL;
4648
4649         /*
4650          * 2 for inode item and ref
4651          * 2 for dir items
4652          * 1 for xattr if selinux is on
4653          */
4654         trans = btrfs_start_transaction(root, 5);
4655         if (IS_ERR(trans))
4656                 return PTR_ERR(trans);
4657
4658         err = btrfs_find_free_ino(root, &objectid);
4659         if (err)
4660                 goto out_unlock;
4661
4662         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4663                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4664                                 mode, &index);
4665         if (IS_ERR(inode)) {
4666                 err = PTR_ERR(inode);
4667                 goto out_unlock;
4668         }
4669
4670         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4671         if (err) {
4672                 drop_inode = 1;
4673                 goto out_unlock;
4674         }
4675
4676         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4677         if (err)
4678                 drop_inode = 1;
4679         else {
4680                 inode->i_op = &btrfs_special_inode_operations;
4681                 init_special_inode(inode, inode->i_mode, rdev);
4682                 btrfs_update_inode(trans, root, inode);
4683         }
4684 out_unlock:
4685         nr = trans->blocks_used;
4686         btrfs_end_transaction_throttle(trans, root);
4687         btrfs_btree_balance_dirty(root, nr);
4688         if (drop_inode) {
4689                 inode_dec_link_count(inode);
4690                 iput(inode);
4691         }
4692         return err;
4693 }
4694
4695 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4696                         int mode, struct nameidata *nd)
4697 {
4698         struct btrfs_trans_handle *trans;
4699         struct btrfs_root *root = BTRFS_I(dir)->root;
4700         struct inode *inode = NULL;
4701         int drop_inode = 0;
4702         int err;
4703         unsigned long nr = 0;
4704         u64 objectid;
4705         u64 index = 0;
4706
4707         /*
4708          * 2 for inode item and ref
4709          * 2 for dir items
4710          * 1 for xattr if selinux is on
4711          */
4712         trans = btrfs_start_transaction(root, 5);
4713         if (IS_ERR(trans))
4714                 return PTR_ERR(trans);
4715
4716         err = btrfs_find_free_ino(root, &objectid);
4717         if (err)
4718                 goto out_unlock;
4719
4720         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4721                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4722                                 mode, &index);
4723         if (IS_ERR(inode)) {
4724                 err = PTR_ERR(inode);
4725                 goto out_unlock;
4726         }
4727
4728         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4729         if (err) {
4730                 drop_inode = 1;
4731                 goto out_unlock;
4732         }
4733
4734         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4735         if (err)
4736                 drop_inode = 1;
4737         else {
4738                 inode->i_mapping->a_ops = &btrfs_aops;
4739                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4740                 inode->i_fop = &btrfs_file_operations;
4741                 inode->i_op = &btrfs_file_inode_operations;
4742                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4743         }
4744 out_unlock:
4745         nr = trans->blocks_used;
4746         btrfs_end_transaction_throttle(trans, root);
4747         if (drop_inode) {
4748                 inode_dec_link_count(inode);
4749                 iput(inode);
4750         }
4751         btrfs_btree_balance_dirty(root, nr);
4752         return err;
4753 }
4754
4755 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4756                       struct dentry *dentry)
4757 {
4758         struct btrfs_trans_handle *trans;
4759         struct btrfs_root *root = BTRFS_I(dir)->root;
4760         struct inode *inode = old_dentry->d_inode;
4761         u64 index;
4762         unsigned long nr = 0;
4763         int err;
4764         int drop_inode = 0;
4765
4766         /* do not allow sys_link's with other subvols of the same device */
4767         if (root->objectid != BTRFS_I(inode)->root->objectid)
4768                 return -EXDEV;
4769
4770         if (inode->i_nlink == ~0U)
4771                 return -EMLINK;
4772
4773         err = btrfs_set_inode_index(dir, &index);
4774         if (err)
4775                 goto fail;
4776
4777         /*
4778          * 2 items for inode and inode ref
4779          * 2 items for dir items
4780          * 1 item for parent inode
4781          */
4782         trans = btrfs_start_transaction(root, 5);
4783         if (IS_ERR(trans)) {
4784                 err = PTR_ERR(trans);
4785                 goto fail;
4786         }
4787
4788         btrfs_inc_nlink(inode);
4789         inode->i_ctime = CURRENT_TIME;
4790         ihold(inode);
4791
4792         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4793
4794         if (err) {
4795                 drop_inode = 1;
4796         } else {
4797                 struct dentry *parent = dentry->d_parent;
4798                 err = btrfs_update_inode(trans, root, inode);
4799                 BUG_ON(err);
4800                 btrfs_log_new_name(trans, inode, NULL, parent);
4801         }
4802
4803         nr = trans->blocks_used;
4804         btrfs_end_transaction_throttle(trans, root);
4805 fail:
4806         if (drop_inode) {
4807                 inode_dec_link_count(inode);
4808                 iput(inode);
4809         }
4810         btrfs_btree_balance_dirty(root, nr);
4811         return err;
4812 }
4813
4814 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4815 {
4816         struct inode *inode = NULL;
4817         struct btrfs_trans_handle *trans;
4818         struct btrfs_root *root = BTRFS_I(dir)->root;
4819         int err = 0;
4820         int drop_on_err = 0;
4821         u64 objectid = 0;
4822         u64 index = 0;
4823         unsigned long nr = 1;
4824
4825         /*
4826          * 2 items for inode and ref
4827          * 2 items for dir items
4828          * 1 for xattr if selinux is on
4829          */
4830         trans = btrfs_start_transaction(root, 5);
4831         if (IS_ERR(trans))
4832                 return PTR_ERR(trans);
4833
4834         err = btrfs_find_free_ino(root, &objectid);
4835         if (err)
4836                 goto out_fail;
4837
4838         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4839                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4840                                 S_IFDIR | mode, &index);
4841         if (IS_ERR(inode)) {
4842                 err = PTR_ERR(inode);
4843                 goto out_fail;
4844         }
4845
4846         drop_on_err = 1;
4847
4848         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4849         if (err)
4850                 goto out_fail;
4851
4852         inode->i_op = &btrfs_dir_inode_operations;
4853         inode->i_fop = &btrfs_dir_file_operations;
4854
4855         btrfs_i_size_write(inode, 0);
4856         err = btrfs_update_inode(trans, root, inode);
4857         if (err)
4858                 goto out_fail;
4859
4860         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4861                              dentry->d_name.len, 0, index);
4862         if (err)
4863                 goto out_fail;
4864
4865         d_instantiate(dentry, inode);
4866         drop_on_err = 0;
4867
4868 out_fail:
4869         nr = trans->blocks_used;
4870         btrfs_end_transaction_throttle(trans, root);
4871         if (drop_on_err)
4872                 iput(inode);
4873         btrfs_btree_balance_dirty(root, nr);
4874         return err;
4875 }
4876
4877 /* helper for btfs_get_extent.  Given an existing extent in the tree,
4878  * and an extent that you want to insert, deal with overlap and insert
4879  * the new extent into the tree.
4880  */
4881 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4882                                 struct extent_map *existing,
4883                                 struct extent_map *em,
4884                                 u64 map_start, u64 map_len)
4885 {
4886         u64 start_diff;
4887
4888         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4889         start_diff = map_start - em->start;
4890         em->start = map_start;
4891         em->len = map_len;
4892         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4893             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4894                 em->block_start += start_diff;
4895                 em->block_len -= start_diff;
4896         }
4897         return add_extent_mapping(em_tree, em);
4898 }
4899
4900 static noinline int uncompress_inline(struct btrfs_path *path,
4901                                       struct inode *inode, struct page *page,
4902                                       size_t pg_offset, u64 extent_offset,
4903                                       struct btrfs_file_extent_item *item)
4904 {
4905         int ret;
4906         struct extent_buffer *leaf = path->nodes[0];
4907         char *tmp;
4908         size_t max_size;
4909         unsigned long inline_size;
4910         unsigned long ptr;
4911         int compress_type;
4912
4913         WARN_ON(pg_offset != 0);
4914         compress_type = btrfs_file_extent_compression(leaf, item);
4915         max_size = btrfs_file_extent_ram_bytes(leaf, item);
4916         inline_size = btrfs_file_extent_inline_item_len(leaf,
4917                                         btrfs_item_nr(leaf, path->slots[0]));
4918         tmp = kmalloc(inline_size, GFP_NOFS);
4919         if (!tmp)
4920                 return -ENOMEM;
4921         ptr = btrfs_file_extent_inline_start(item);
4922
4923         read_extent_buffer(leaf, tmp, ptr, inline_size);
4924
4925         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4926         ret = btrfs_decompress(compress_type, tmp, page,
4927                                extent_offset, inline_size, max_size);
4928         if (ret) {
4929                 char *kaddr = kmap_atomic(page, KM_USER0);
4930                 unsigned long copy_size = min_t(u64,
4931                                   PAGE_CACHE_SIZE - pg_offset,
4932                                   max_size - extent_offset);
4933                 memset(kaddr + pg_offset, 0, copy_size);
4934                 kunmap_atomic(kaddr, KM_USER0);
4935         }
4936         kfree(tmp);
4937         return 0;
4938 }
4939
4940 /*
4941  * a bit scary, this does extent mapping from logical file offset to the disk.
4942  * the ugly parts come from merging extents from the disk with the in-ram
4943  * representation.  This gets more complex because of the data=ordered code,
4944  * where the in-ram extents might be locked pending data=ordered completion.
4945  *
4946  * This also copies inline extents directly into the page.
4947  */
4948
4949 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4950                                     size_t pg_offset, u64 start, u64 len,
4951                                     int create)
4952 {
4953         int ret;
4954         int err = 0;
4955         u64 bytenr;
4956         u64 extent_start = 0;
4957         u64 extent_end = 0;
4958         u64 objectid = btrfs_ino(inode);
4959         u32 found_type;
4960         struct btrfs_path *path = NULL;
4961         struct btrfs_root *root = BTRFS_I(inode)->root;
4962         struct btrfs_file_extent_item *item;
4963         struct extent_buffer *leaf;
4964         struct btrfs_key found_key;
4965         struct extent_map *em = NULL;
4966         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4967         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4968         struct btrfs_trans_handle *trans = NULL;
4969         int compress_type;
4970
4971 again:
4972         read_lock(&em_tree->lock);
4973         em = lookup_extent_mapping(em_tree, start, len);
4974         if (em)
4975                 em->bdev = root->fs_info->fs_devices->latest_bdev;
4976         read_unlock(&em_tree->lock);
4977
4978         if (em) {
4979                 if (em->start > start || em->start + em->len <= start)
4980                         free_extent_map(em);
4981                 else if (em->block_start == EXTENT_MAP_INLINE && page)
4982                         free_extent_map(em);
4983                 else
4984                         goto out;
4985         }
4986         em = alloc_extent_map();
4987         if (!em) {
4988                 err = -ENOMEM;
4989                 goto out;
4990         }
4991         em->bdev = root->fs_info->fs_devices->latest_bdev;
4992         em->start = EXTENT_MAP_HOLE;
4993         em->orig_start = EXTENT_MAP_HOLE;
4994         em->len = (u64)-1;
4995         em->block_len = (u64)-1;
4996
4997         if (!path) {
4998                 path = btrfs_alloc_path();
4999                 if (!path) {
5000                         err = -ENOMEM;
5001                         goto out;
5002                 }
5003                 /*
5004                  * Chances are we'll be called again, so go ahead and do
5005                  * readahead
5006                  */
5007                 path->reada = 1;
5008         }
5009
5010         ret = btrfs_lookup_file_extent(trans, root, path,
5011                                        objectid, start, trans != NULL);
5012         if (ret < 0) {
5013                 err = ret;
5014                 goto out;
5015         }
5016
5017         if (ret != 0) {
5018                 if (path->slots[0] == 0)
5019                         goto not_found;
5020                 path->slots[0]--;
5021         }
5022
5023         leaf = path->nodes[0];
5024         item = btrfs_item_ptr(leaf, path->slots[0],
5025                               struct btrfs_file_extent_item);
5026         /* are we inside the extent that was found? */
5027         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5028         found_type = btrfs_key_type(&found_key);
5029         if (found_key.objectid != objectid ||
5030             found_type != BTRFS_EXTENT_DATA_KEY) {
5031                 goto not_found;
5032         }
5033
5034         found_type = btrfs_file_extent_type(leaf, item);
5035         extent_start = found_key.offset;
5036         compress_type = btrfs_file_extent_compression(leaf, item);
5037         if (found_type == BTRFS_FILE_EXTENT_REG ||
5038             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5039                 extent_end = extent_start +
5040                        btrfs_file_extent_num_bytes(leaf, item);
5041         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5042                 size_t size;
5043                 size = btrfs_file_extent_inline_len(leaf, item);
5044                 extent_end = (extent_start + size + root->sectorsize - 1) &
5045                         ~((u64)root->sectorsize - 1);
5046         }
5047
5048         if (start >= extent_end) {
5049                 path->slots[0]++;
5050                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5051                         ret = btrfs_next_leaf(root, path);
5052                         if (ret < 0) {
5053                                 err = ret;
5054                                 goto out;
5055                         }
5056                         if (ret > 0)
5057                                 goto not_found;
5058                         leaf = path->nodes[0];
5059                 }
5060                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5061                 if (found_key.objectid != objectid ||
5062                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5063                         goto not_found;
5064                 if (start + len <= found_key.offset)
5065                         goto not_found;
5066                 em->start = start;
5067                 em->len = found_key.offset - start;
5068                 goto not_found_em;
5069         }
5070
5071         if (found_type == BTRFS_FILE_EXTENT_REG ||
5072             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5073                 em->start = extent_start;
5074                 em->len = extent_end - extent_start;
5075                 em->orig_start = extent_start -
5076                                  btrfs_file_extent_offset(leaf, item);
5077                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5078                 if (bytenr == 0) {
5079                         em->block_start = EXTENT_MAP_HOLE;
5080                         goto insert;
5081                 }
5082                 if (compress_type != BTRFS_COMPRESS_NONE) {
5083                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5084                         em->compress_type = compress_type;
5085                         em->block_start = bytenr;
5086                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5087                                                                          item);
5088                 } else {
5089                         bytenr += btrfs_file_extent_offset(leaf, item);
5090                         em->block_start = bytenr;
5091                         em->block_len = em->len;
5092                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5093                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5094                 }
5095                 goto insert;
5096         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5097                 unsigned long ptr;
5098                 char *map;
5099                 size_t size;
5100                 size_t extent_offset;
5101                 size_t copy_size;
5102
5103                 em->block_start = EXTENT_MAP_INLINE;
5104                 if (!page || create) {
5105                         em->start = extent_start;
5106                         em->len = extent_end - extent_start;
5107                         goto out;
5108                 }
5109
5110                 size = btrfs_file_extent_inline_len(leaf, item);
5111                 extent_offset = page_offset(page) + pg_offset - extent_start;
5112                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5113                                 size - extent_offset);
5114                 em->start = extent_start + extent_offset;
5115                 em->len = (copy_size + root->sectorsize - 1) &
5116                         ~((u64)root->sectorsize - 1);
5117                 em->orig_start = EXTENT_MAP_INLINE;
5118                 if (compress_type) {
5119                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5120                         em->compress_type = compress_type;
5121                 }
5122                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5123                 if (create == 0 && !PageUptodate(page)) {
5124                         if (btrfs_file_extent_compression(leaf, item) !=
5125                             BTRFS_COMPRESS_NONE) {
5126                                 ret = uncompress_inline(path, inode, page,
5127                                                         pg_offset,
5128                                                         extent_offset, item);
5129                                 BUG_ON(ret);
5130                         } else {
5131                                 map = kmap(page);
5132                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5133                                                    copy_size);
5134                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5135                                         memset(map + pg_offset + copy_size, 0,
5136                                                PAGE_CACHE_SIZE - pg_offset -
5137                                                copy_size);
5138                                 }
5139                                 kunmap(page);
5140                         }
5141                         flush_dcache_page(page);
5142                 } else if (create && PageUptodate(page)) {
5143                         WARN_ON(1);
5144                         if (!trans) {
5145                                 kunmap(page);
5146                                 free_extent_map(em);
5147                                 em = NULL;
5148
5149                                 btrfs_release_path(path);
5150                                 trans = btrfs_join_transaction(root);
5151
5152                                 if (IS_ERR(trans))
5153                                         return ERR_CAST(trans);
5154                                 goto again;
5155                         }
5156                         map = kmap(page);
5157                         write_extent_buffer(leaf, map + pg_offset, ptr,
5158                                             copy_size);
5159                         kunmap(page);
5160                         btrfs_mark_buffer_dirty(leaf);
5161                 }
5162                 set_extent_uptodate(io_tree, em->start,
5163                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
5164                 goto insert;
5165         } else {
5166                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5167                 WARN_ON(1);
5168         }
5169 not_found:
5170         em->start = start;
5171         em->len = len;
5172 not_found_em:
5173         em->block_start = EXTENT_MAP_HOLE;
5174         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5175 insert:
5176         btrfs_release_path(path);
5177         if (em->start > start || extent_map_end(em) <= start) {
5178                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5179                        "[%llu %llu]\n", (unsigned long long)em->start,
5180                        (unsigned long long)em->len,
5181                        (unsigned long long)start,
5182                        (unsigned long long)len);
5183                 err = -EIO;
5184                 goto out;
5185         }
5186
5187         err = 0;
5188         write_lock(&em_tree->lock);
5189         ret = add_extent_mapping(em_tree, em);
5190         /* it is possible that someone inserted the extent into the tree
5191          * while we had the lock dropped.  It is also possible that
5192          * an overlapping map exists in the tree
5193          */
5194         if (ret == -EEXIST) {
5195                 struct extent_map *existing;
5196
5197                 ret = 0;
5198
5199                 existing = lookup_extent_mapping(em_tree, start, len);
5200                 if (existing && (existing->start > start ||
5201                     existing->start + existing->len <= start)) {
5202                         free_extent_map(existing);
5203                         existing = NULL;
5204                 }
5205                 if (!existing) {
5206                         existing = lookup_extent_mapping(em_tree, em->start,
5207                                                          em->len);
5208                         if (existing) {
5209                                 err = merge_extent_mapping(em_tree, existing,
5210                                                            em, start,
5211                                                            root->sectorsize);
5212                                 free_extent_map(existing);
5213                                 if (err) {
5214                                         free_extent_map(em);
5215                                         em = NULL;
5216                                 }
5217                         } else {
5218                                 err = -EIO;
5219                                 free_extent_map(em);
5220                                 em = NULL;
5221                         }
5222                 } else {
5223                         free_extent_map(em);
5224                         em = existing;
5225                         err = 0;
5226                 }
5227         }
5228         write_unlock(&em_tree->lock);
5229 out:
5230
5231         trace_btrfs_get_extent(root, em);
5232
5233         if (path)
5234                 btrfs_free_path(path);
5235         if (trans) {
5236                 ret = btrfs_end_transaction(trans, root);
5237                 if (!err)
5238                         err = ret;
5239         }
5240         if (err) {
5241                 free_extent_map(em);
5242                 return ERR_PTR(err);
5243         }
5244         return em;
5245 }
5246
5247 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5248                                            size_t pg_offset, u64 start, u64 len,
5249                                            int create)
5250 {
5251         struct extent_map *em;
5252         struct extent_map *hole_em = NULL;
5253         u64 range_start = start;
5254         u64 end;
5255         u64 found;
5256         u64 found_end;
5257         int err = 0;
5258
5259         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5260         if (IS_ERR(em))
5261                 return em;
5262         if (em) {
5263                 /*
5264                  * if our em maps to a hole, there might
5265                  * actually be delalloc bytes behind it
5266                  */
5267                 if (em->block_start != EXTENT_MAP_HOLE)
5268                         return em;
5269                 else
5270                         hole_em = em;
5271         }
5272
5273         /* check to see if we've wrapped (len == -1 or similar) */
5274         end = start + len;
5275         if (end < start)
5276                 end = (u64)-1;
5277         else
5278                 end -= 1;
5279
5280         em = NULL;
5281
5282         /* ok, we didn't find anything, lets look for delalloc */
5283         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5284                                  end, len, EXTENT_DELALLOC, 1);
5285         found_end = range_start + found;
5286         if (found_end < range_start)
5287                 found_end = (u64)-1;
5288
5289         /*
5290          * we didn't find anything useful, return
5291          * the original results from get_extent()
5292          */
5293         if (range_start > end || found_end <= start) {
5294                 em = hole_em;
5295                 hole_em = NULL;
5296                 goto out;
5297         }
5298
5299         /* adjust the range_start to make sure it doesn't
5300          * go backwards from the start they passed in
5301          */
5302         range_start = max(start,range_start);
5303         found = found_end - range_start;
5304
5305         if (found > 0) {
5306                 u64 hole_start = start;
5307                 u64 hole_len = len;
5308
5309                 em = alloc_extent_map();
5310                 if (!em) {
5311                         err = -ENOMEM;
5312                         goto out;
5313                 }
5314                 /*
5315                  * when btrfs_get_extent can't find anything it
5316                  * returns one huge hole
5317                  *
5318                  * make sure what it found really fits our range, and
5319                  * adjust to make sure it is based on the start from
5320                  * the caller
5321                  */
5322                 if (hole_em) {
5323                         u64 calc_end = extent_map_end(hole_em);
5324
5325                         if (calc_end <= start || (hole_em->start > end)) {
5326                                 free_extent_map(hole_em);
5327                                 hole_em = NULL;
5328                         } else {
5329                                 hole_start = max(hole_em->start, start);
5330                                 hole_len = calc_end - hole_start;
5331                         }
5332                 }
5333                 em->bdev = NULL;
5334                 if (hole_em && range_start > hole_start) {
5335                         /* our hole starts before our delalloc, so we
5336                          * have to return just the parts of the hole
5337                          * that go until  the delalloc starts
5338                          */
5339                         em->len = min(hole_len,
5340                                       range_start - hole_start);
5341                         em->start = hole_start;
5342                         em->orig_start = hole_start;
5343                         /*
5344                          * don't adjust block start at all,
5345                          * it is fixed at EXTENT_MAP_HOLE
5346                          */
5347                         em->block_start = hole_em->block_start;
5348                         em->block_len = hole_len;
5349                 } else {
5350                         em->start = range_start;
5351                         em->len = found;
5352                         em->orig_start = range_start;
5353                         em->block_start = EXTENT_MAP_DELALLOC;
5354                         em->block_len = found;
5355                 }
5356         } else if (hole_em) {
5357                 return hole_em;
5358         }
5359 out:
5360
5361         free_extent_map(hole_em);
5362         if (err) {
5363                 free_extent_map(em);
5364                 return ERR_PTR(err);
5365         }
5366         return em;
5367 }
5368
5369 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5370                                                   struct extent_map *em,
5371                                                   u64 start, u64 len)
5372 {
5373         struct btrfs_root *root = BTRFS_I(inode)->root;
5374         struct btrfs_trans_handle *trans;
5375         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5376         struct btrfs_key ins;
5377         u64 alloc_hint;
5378         int ret;
5379         bool insert = false;
5380
5381         /*
5382          * Ok if the extent map we looked up is a hole and is for the exact
5383          * range we want, there is no reason to allocate a new one, however if
5384          * it is not right then we need to free this one and drop the cache for
5385          * our range.
5386          */
5387         if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5388             em->len != len) {
5389                 free_extent_map(em);
5390                 em = NULL;
5391                 insert = true;
5392                 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5393         }
5394
5395         trans = btrfs_join_transaction(root);
5396         if (IS_ERR(trans))
5397                 return ERR_CAST(trans);
5398
5399         if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5400                 btrfs_add_inode_defrag(trans, inode);
5401
5402         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5403
5404         alloc_hint = get_extent_allocation_hint(inode, start, len);
5405         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5406                                    alloc_hint, (u64)-1, &ins, 1);
5407         if (ret) {
5408                 em = ERR_PTR(ret);
5409                 goto out;
5410         }
5411
5412         if (!em) {
5413                 em = alloc_extent_map();
5414                 if (!em) {
5415                         em = ERR_PTR(-ENOMEM);
5416                         goto out;
5417                 }
5418         }
5419
5420         em->start = start;
5421         em->orig_start = em->start;
5422         em->len = ins.offset;
5423
5424         em->block_start = ins.objectid;
5425         em->block_len = ins.offset;
5426         em->bdev = root->fs_info->fs_devices->latest_bdev;
5427
5428         /*
5429          * We need to do this because if we're using the original em we searched
5430          * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5431          */
5432         em->flags = 0;
5433         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5434
5435         while (insert) {
5436                 write_lock(&em_tree->lock);
5437                 ret = add_extent_mapping(em_tree, em);
5438                 write_unlock(&em_tree->lock);
5439                 if (ret != -EEXIST)
5440                         break;
5441                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5442         }
5443
5444         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5445                                            ins.offset, ins.offset, 0);
5446         if (ret) {
5447                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5448                 em = ERR_PTR(ret);
5449         }
5450 out:
5451         btrfs_end_transaction(trans, root);
5452         return em;
5453 }
5454
5455 /*
5456  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5457  * block must be cow'd
5458  */
5459 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5460                                       struct inode *inode, u64 offset, u64 len)
5461 {
5462         struct btrfs_path *path;
5463         int ret;
5464         struct extent_buffer *leaf;
5465         struct btrfs_root *root = BTRFS_I(inode)->root;
5466         struct btrfs_file_extent_item *fi;
5467         struct btrfs_key key;
5468         u64 disk_bytenr;
5469         u64 backref_offset;
5470         u64 extent_end;
5471         u64 num_bytes;
5472         int slot;
5473         int found_type;
5474
5475         path = btrfs_alloc_path();
5476         if (!path)
5477                 return -ENOMEM;
5478
5479         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5480                                        offset, 0);
5481         if (ret < 0)
5482                 goto out;
5483
5484         slot = path->slots[0];
5485         if (ret == 1) {
5486                 if (slot == 0) {
5487                         /* can't find the item, must cow */
5488                         ret = 0;
5489                         goto out;
5490                 }
5491                 slot--;
5492         }
5493         ret = 0;
5494         leaf = path->nodes[0];
5495         btrfs_item_key_to_cpu(leaf, &key, slot);
5496         if (key.objectid != btrfs_ino(inode) ||
5497             key.type != BTRFS_EXTENT_DATA_KEY) {
5498                 /* not our file or wrong item type, must cow */
5499                 goto out;
5500         }
5501
5502         if (key.offset > offset) {
5503                 /* Wrong offset, must cow */
5504                 goto out;
5505         }
5506
5507         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5508         found_type = btrfs_file_extent_type(leaf, fi);
5509         if (found_type != BTRFS_FILE_EXTENT_REG &&
5510             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5511                 /* not a regular extent, must cow */
5512                 goto out;
5513         }
5514         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5515         backref_offset = btrfs_file_extent_offset(leaf, fi);
5516
5517         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5518         if (extent_end < offset + len) {
5519                 /* extent doesn't include our full range, must cow */
5520                 goto out;
5521         }
5522
5523         if (btrfs_extent_readonly(root, disk_bytenr))
5524                 goto out;
5525
5526         /*
5527          * look for other files referencing this extent, if we
5528          * find any we must cow
5529          */
5530         if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5531                                   key.offset - backref_offset, disk_bytenr))
5532                 goto out;
5533
5534         /*
5535          * adjust disk_bytenr and num_bytes to cover just the bytes
5536          * in this extent we are about to write.  If there
5537          * are any csums in that range we have to cow in order
5538          * to keep the csums correct
5539          */
5540         disk_bytenr += backref_offset;
5541         disk_bytenr += offset - key.offset;
5542         num_bytes = min(offset + len, extent_end) - offset;
5543         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5544                                 goto out;
5545         /*
5546          * all of the above have passed, it is safe to overwrite this extent
5547          * without cow
5548          */
5549         ret = 1;
5550 out:
5551         btrfs_free_path(path);
5552         return ret;
5553 }
5554
5555 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5556                                    struct buffer_head *bh_result, int create)
5557 {
5558         struct extent_map *em;
5559         struct btrfs_root *root = BTRFS_I(inode)->root;
5560         u64 start = iblock << inode->i_blkbits;
5561         u64 len = bh_result->b_size;
5562         struct btrfs_trans_handle *trans;
5563
5564         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5565         if (IS_ERR(em))
5566                 return PTR_ERR(em);
5567
5568         /*
5569          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5570          * io.  INLINE is special, and we could probably kludge it in here, but
5571          * it's still buffered so for safety lets just fall back to the generic
5572          * buffered path.
5573          *
5574          * For COMPRESSED we _have_ to read the entire extent in so we can
5575          * decompress it, so there will be buffering required no matter what we
5576          * do, so go ahead and fallback to buffered.
5577          *
5578          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5579          * to buffered IO.  Don't blame me, this is the price we pay for using
5580          * the generic code.
5581          */
5582         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5583             em->block_start == EXTENT_MAP_INLINE) {
5584                 free_extent_map(em);
5585                 return -ENOTBLK;
5586         }
5587
5588         /* Just a good old fashioned hole, return */
5589         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5590                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5591                 free_extent_map(em);
5592                 /* DIO will do one hole at a time, so just unlock a sector */
5593                 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5594                               start + root->sectorsize - 1, GFP_NOFS);
5595                 return 0;
5596         }
5597
5598         /*
5599          * We don't allocate a new extent in the following cases
5600          *
5601          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
5602          * existing extent.
5603          * 2) The extent is marked as PREALLOC.  We're good to go here and can
5604          * just use the extent.
5605          *
5606          */
5607         if (!create) {
5608                 len = em->len - (start - em->start);
5609                 goto map;
5610         }
5611
5612         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5613             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5614              em->block_start != EXTENT_MAP_HOLE)) {
5615                 int type;
5616                 int ret;
5617                 u64 block_start;
5618
5619                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5620                         type = BTRFS_ORDERED_PREALLOC;
5621                 else
5622                         type = BTRFS_ORDERED_NOCOW;
5623                 len = min(len, em->len - (start - em->start));
5624                 block_start = em->block_start + (start - em->start);
5625
5626                 /*
5627                  * we're not going to log anything, but we do need
5628                  * to make sure the current transaction stays open
5629                  * while we look for nocow cross refs
5630                  */
5631                 trans = btrfs_join_transaction(root);
5632                 if (IS_ERR(trans))
5633                         goto must_cow;
5634
5635                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5636                         ret = btrfs_add_ordered_extent_dio(inode, start,
5637                                            block_start, len, len, type);
5638                         btrfs_end_transaction(trans, root);
5639                         if (ret) {
5640                                 free_extent_map(em);
5641                                 return ret;
5642                         }
5643                         goto unlock;
5644                 }
5645                 btrfs_end_transaction(trans, root);
5646         }
5647 must_cow:
5648         /*
5649          * this will cow the extent, reset the len in case we changed
5650          * it above
5651          */
5652         len = bh_result->b_size;
5653         em = btrfs_new_extent_direct(inode, em, start, len);
5654         if (IS_ERR(em))
5655                 return PTR_ERR(em);
5656         len = min(len, em->len - (start - em->start));
5657 unlock:
5658         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5659                           EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5660                           0, NULL, GFP_NOFS);
5661 map:
5662         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5663                 inode->i_blkbits;
5664         bh_result->b_size = len;
5665         bh_result->b_bdev = em->bdev;
5666         set_buffer_mapped(bh_result);
5667         if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5668                 set_buffer_new(bh_result);
5669
5670         free_extent_map(em);
5671
5672         return 0;
5673 }
5674
5675 struct btrfs_dio_private {
5676         struct inode *inode;
5677         u64 logical_offset;
5678         u64 disk_bytenr;
5679         u64 bytes;
5680         u32 *csums;
5681         void *private;
5682
5683         /* number of bios pending for this dio */
5684         atomic_t pending_bios;
5685
5686         /* IO errors */
5687         int errors;
5688
5689         struct bio *orig_bio;
5690 };
5691
5692 static void btrfs_endio_direct_read(struct bio *bio, int err)
5693 {
5694         struct btrfs_dio_private *dip = bio->bi_private;
5695         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5696         struct bio_vec *bvec = bio->bi_io_vec;
5697         struct inode *inode = dip->inode;
5698         struct btrfs_root *root = BTRFS_I(inode)->root;
5699         u64 start;
5700         u32 *private = dip->csums;
5701
5702         start = dip->logical_offset;
5703         do {
5704                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5705                         struct page *page = bvec->bv_page;
5706                         char *kaddr;
5707                         u32 csum = ~(u32)0;
5708                         unsigned long flags;
5709
5710                         local_irq_save(flags);
5711                         kaddr = kmap_atomic(page, KM_IRQ0);
5712                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5713                                                csum, bvec->bv_len);
5714                         btrfs_csum_final(csum, (char *)&csum);
5715                         kunmap_atomic(kaddr, KM_IRQ0);
5716                         local_irq_restore(flags);
5717
5718                         flush_dcache_page(bvec->bv_page);
5719                         if (csum != *private) {
5720                                 printk(KERN_ERR "btrfs csum failed ino %llu off"
5721                                       " %llu csum %u private %u\n",
5722                                       (unsigned long long)btrfs_ino(inode),
5723                                       (unsigned long long)start,
5724                                       csum, *private);
5725                                 err = -EIO;
5726                         }
5727                 }
5728
5729                 start += bvec->bv_len;
5730                 private++;
5731                 bvec++;
5732         } while (bvec <= bvec_end);
5733
5734         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5735                       dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5736         bio->bi_private = dip->private;
5737
5738         kfree(dip->csums);
5739         kfree(dip);
5740
5741         /* If we had a csum failure make sure to clear the uptodate flag */
5742         if (err)
5743                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5744         dio_end_io(bio, err);
5745 }
5746
5747 static void btrfs_endio_direct_write(struct bio *bio, int err)
5748 {
5749         struct btrfs_dio_private *dip = bio->bi_private;
5750         struct inode *inode = dip->inode;
5751         struct btrfs_root *root = BTRFS_I(inode)->root;
5752         struct btrfs_trans_handle *trans;
5753         struct btrfs_ordered_extent *ordered = NULL;
5754         struct extent_state *cached_state = NULL;
5755         u64 ordered_offset = dip->logical_offset;
5756         u64 ordered_bytes = dip->bytes;
5757         int ret;
5758
5759         if (err)
5760                 goto out_done;
5761 again:
5762         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5763                                                    &ordered_offset,
5764                                                    ordered_bytes);
5765         if (!ret)
5766                 goto out_test;
5767
5768         BUG_ON(!ordered);
5769
5770         trans = btrfs_join_transaction(root);
5771         if (IS_ERR(trans)) {
5772                 err = -ENOMEM;
5773                 goto out;
5774         }
5775         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5776
5777         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5778                 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5779                 if (!ret)
5780                         err = btrfs_update_inode(trans, root, inode);
5781                 goto out;
5782         }
5783
5784         lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5785                          ordered->file_offset + ordered->len - 1, 0,
5786                          &cached_state, GFP_NOFS);
5787
5788         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5789                 ret = btrfs_mark_extent_written(trans, inode,
5790                                                 ordered->file_offset,
5791                                                 ordered->file_offset +
5792                                                 ordered->len);
5793                 if (ret) {
5794                         err = ret;
5795                         goto out_unlock;
5796                 }
5797         } else {
5798                 ret = insert_reserved_file_extent(trans, inode,
5799                                                   ordered->file_offset,
5800                                                   ordered->start,
5801                                                   ordered->disk_len,
5802                                                   ordered->len,
5803                                                   ordered->len,
5804                                                   0, 0, 0,
5805                                                   BTRFS_FILE_EXTENT_REG);
5806                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5807                                    ordered->file_offset, ordered->len);
5808                 if (ret) {
5809                         err = ret;
5810                         WARN_ON(1);
5811                         goto out_unlock;
5812                 }
5813         }
5814
5815         add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5816         ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5817         if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags))
5818                 btrfs_update_inode(trans, root, inode);
5819         ret = 0;
5820 out_unlock:
5821         unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5822                              ordered->file_offset + ordered->len - 1,
5823                              &cached_state, GFP_NOFS);
5824 out:
5825         btrfs_delalloc_release_metadata(inode, ordered->len);
5826         btrfs_end_transaction(trans, root);
5827         ordered_offset = ordered->file_offset + ordered->len;
5828         btrfs_put_ordered_extent(ordered);
5829         btrfs_put_ordered_extent(ordered);
5830
5831 out_test:
5832         /*
5833          * our bio might span multiple ordered extents.  If we haven't
5834          * completed the accounting for the whole dio, go back and try again
5835          */
5836         if (ordered_offset < dip->logical_offset + dip->bytes) {
5837                 ordered_bytes = dip->logical_offset + dip->bytes -
5838                         ordered_offset;
5839                 goto again;
5840         }
5841 out_done:
5842         bio->bi_private = dip->private;
5843
5844         kfree(dip->csums);
5845         kfree(dip);
5846
5847         /* If we had an error make sure to clear the uptodate flag */
5848         if (err)
5849                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5850         dio_end_io(bio, err);
5851 }
5852
5853 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5854                                     struct bio *bio, int mirror_num,
5855                                     unsigned long bio_flags, u64 offset)
5856 {
5857         int ret;
5858         struct btrfs_root *root = BTRFS_I(inode)->root;
5859         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5860         BUG_ON(ret);
5861         return 0;
5862 }
5863
5864 static void btrfs_end_dio_bio(struct bio *bio, int err)
5865 {
5866         struct btrfs_dio_private *dip = bio->bi_private;
5867
5868         if (err) {
5869                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
5870                       "sector %#Lx len %u err no %d\n",
5871                       (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
5872                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
5873                 dip->errors = 1;
5874
5875                 /*
5876                  * before atomic variable goto zero, we must make sure
5877                  * dip->errors is perceived to be set.
5878                  */
5879                 smp_mb__before_atomic_dec();
5880         }
5881
5882         /* if there are more bios still pending for this dio, just exit */
5883         if (!atomic_dec_and_test(&dip->pending_bios))
5884                 goto out;
5885
5886         if (dip->errors)
5887                 bio_io_error(dip->orig_bio);
5888         else {
5889                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5890                 bio_endio(dip->orig_bio, 0);
5891         }
5892 out:
5893         bio_put(bio);
5894 }
5895
5896 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5897                                        u64 first_sector, gfp_t gfp_flags)
5898 {
5899         int nr_vecs = bio_get_nr_vecs(bdev);
5900         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5901 }
5902
5903 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5904                                          int rw, u64 file_offset, int skip_sum,
5905                                          u32 *csums, int async_submit)
5906 {
5907         int write = rw & REQ_WRITE;
5908         struct btrfs_root *root = BTRFS_I(inode)->root;
5909         int ret;
5910
5911         bio_get(bio);
5912         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5913         if (ret)
5914                 goto err;
5915
5916         if (skip_sum)
5917                 goto map;
5918
5919         if (write && async_submit) {
5920                 ret = btrfs_wq_submit_bio(root->fs_info,
5921                                    inode, rw, bio, 0, 0,
5922                                    file_offset,
5923                                    __btrfs_submit_bio_start_direct_io,
5924                                    __btrfs_submit_bio_done);
5925                 goto err;
5926         } else if (write) {
5927                 /*
5928                  * If we aren't doing async submit, calculate the csum of the
5929                  * bio now.
5930                  */
5931                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
5932                 if (ret)
5933                         goto err;
5934         } else if (!skip_sum) {
5935                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
5936                                           file_offset, csums);
5937                 if (ret)
5938                         goto err;
5939         }
5940
5941 map:
5942         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
5943 err:
5944         bio_put(bio);
5945         return ret;
5946 }
5947
5948 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5949                                     int skip_sum)
5950 {
5951         struct inode *inode = dip->inode;
5952         struct btrfs_root *root = BTRFS_I(inode)->root;
5953         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5954         struct bio *bio;
5955         struct bio *orig_bio = dip->orig_bio;
5956         struct bio_vec *bvec = orig_bio->bi_io_vec;
5957         u64 start_sector = orig_bio->bi_sector;
5958         u64 file_offset = dip->logical_offset;
5959         u64 submit_len = 0;
5960         u64 map_length;
5961         int nr_pages = 0;
5962         u32 *csums = dip->csums;
5963         int ret = 0;
5964         int async_submit = 0;
5965         int write = rw & REQ_WRITE;
5966
5967         map_length = orig_bio->bi_size;
5968         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5969                               &map_length, NULL, 0);
5970         if (ret) {
5971                 bio_put(orig_bio);
5972                 return -EIO;
5973         }
5974
5975         if (map_length >= orig_bio->bi_size) {
5976                 bio = orig_bio;
5977                 goto submit;
5978         }
5979
5980         async_submit = 1;
5981         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5982         if (!bio)
5983                 return -ENOMEM;
5984         bio->bi_private = dip;
5985         bio->bi_end_io = btrfs_end_dio_bio;
5986         atomic_inc(&dip->pending_bios);
5987
5988         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
5989                 if (unlikely(map_length < submit_len + bvec->bv_len ||
5990                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5991                                  bvec->bv_offset) < bvec->bv_len)) {
5992                         /*
5993                          * inc the count before we submit the bio so
5994                          * we know the end IO handler won't happen before
5995                          * we inc the count. Otherwise, the dip might get freed
5996                          * before we're done setting it up
5997                          */
5998                         atomic_inc(&dip->pending_bios);
5999                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
6000                                                      file_offset, skip_sum,
6001                                                      csums, async_submit);
6002                         if (ret) {
6003                                 bio_put(bio);
6004                                 atomic_dec(&dip->pending_bios);
6005                                 goto out_err;
6006                         }
6007
6008                         /* Write's use the ordered csums */
6009                         if (!write && !skip_sum)
6010                                 csums = csums + nr_pages;
6011                         start_sector += submit_len >> 9;
6012                         file_offset += submit_len;
6013
6014                         submit_len = 0;
6015                         nr_pages = 0;
6016
6017                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6018                                                   start_sector, GFP_NOFS);
6019                         if (!bio)
6020                                 goto out_err;
6021                         bio->bi_private = dip;
6022                         bio->bi_end_io = btrfs_end_dio_bio;
6023
6024                         map_length = orig_bio->bi_size;
6025                         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6026                                               &map_length, NULL, 0);
6027                         if (ret) {
6028                                 bio_put(bio);
6029                                 goto out_err;
6030                         }
6031                 } else {
6032                         submit_len += bvec->bv_len;
6033                         nr_pages ++;
6034                         bvec++;
6035                 }
6036         }
6037
6038 submit:
6039         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6040                                      csums, async_submit);
6041         if (!ret)
6042                 return 0;
6043
6044         bio_put(bio);
6045 out_err:
6046         dip->errors = 1;
6047         /*
6048          * before atomic variable goto zero, we must
6049          * make sure dip->errors is perceived to be set.
6050          */
6051         smp_mb__before_atomic_dec();
6052         if (atomic_dec_and_test(&dip->pending_bios))
6053                 bio_io_error(dip->orig_bio);
6054
6055         /* bio_end_io() will handle error, so we needn't return it */
6056         return 0;
6057 }
6058
6059 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6060                                 loff_t file_offset)
6061 {
6062         struct btrfs_root *root = BTRFS_I(inode)->root;
6063         struct btrfs_dio_private *dip;
6064         struct bio_vec *bvec = bio->bi_io_vec;
6065         int skip_sum;
6066         int write = rw & REQ_WRITE;
6067         int ret = 0;
6068
6069         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6070
6071         dip = kmalloc(sizeof(*dip), GFP_NOFS);
6072         if (!dip) {
6073                 ret = -ENOMEM;
6074                 goto free_ordered;
6075         }
6076         dip->csums = NULL;
6077
6078         /* Write's use the ordered csum stuff, so we don't need dip->csums */
6079         if (!write && !skip_sum) {
6080                 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6081                 if (!dip->csums) {
6082                         kfree(dip);
6083                         ret = -ENOMEM;
6084                         goto free_ordered;
6085                 }
6086         }
6087
6088         dip->private = bio->bi_private;
6089         dip->inode = inode;
6090         dip->logical_offset = file_offset;
6091
6092         dip->bytes = 0;
6093         do {
6094                 dip->bytes += bvec->bv_len;
6095                 bvec++;
6096         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6097
6098         dip->disk_bytenr = (u64)bio->bi_sector << 9;
6099         bio->bi_private = dip;
6100         dip->errors = 0;
6101         dip->orig_bio = bio;
6102         atomic_set(&dip->pending_bios, 0);
6103
6104         if (write)
6105                 bio->bi_end_io = btrfs_endio_direct_write;
6106         else
6107                 bio->bi_end_io = btrfs_endio_direct_read;
6108
6109         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6110         if (!ret)
6111                 return;
6112 free_ordered:
6113         /*
6114          * If this is a write, we need to clean up the reserved space and kill
6115          * the ordered extent.
6116          */
6117         if (write) {
6118                 struct btrfs_ordered_extent *ordered;
6119                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6120                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6121                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6122                         btrfs_free_reserved_extent(root, ordered->start,
6123                                                    ordered->disk_len);
6124                 btrfs_put_ordered_extent(ordered);
6125                 btrfs_put_ordered_extent(ordered);
6126         }
6127         bio_endio(bio, ret);
6128 }
6129
6130 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6131                         const struct iovec *iov, loff_t offset,
6132                         unsigned long nr_segs)
6133 {
6134         int seg;
6135         int i;
6136         size_t size;
6137         unsigned long addr;
6138         unsigned blocksize_mask = root->sectorsize - 1;
6139         ssize_t retval = -EINVAL;
6140         loff_t end = offset;
6141
6142         if (offset & blocksize_mask)
6143                 goto out;
6144
6145         /* Check the memory alignment.  Blocks cannot straddle pages */
6146         for (seg = 0; seg < nr_segs; seg++) {
6147                 addr = (unsigned long)iov[seg].iov_base;
6148                 size = iov[seg].iov_len;
6149                 end += size;
6150                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6151                         goto out;
6152
6153                 /* If this is a write we don't need to check anymore */
6154                 if (rw & WRITE)
6155                         continue;
6156
6157                 /*
6158                  * Check to make sure we don't have duplicate iov_base's in this
6159                  * iovec, if so return EINVAL, otherwise we'll get csum errors
6160                  * when reading back.
6161                  */
6162                 for (i = seg + 1; i < nr_segs; i++) {
6163                         if (iov[seg].iov_base == iov[i].iov_base)
6164                                 goto out;
6165                 }
6166         }
6167         retval = 0;
6168 out:
6169         return retval;
6170 }
6171 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6172                         const struct iovec *iov, loff_t offset,
6173                         unsigned long nr_segs)
6174 {
6175         struct file *file = iocb->ki_filp;
6176         struct inode *inode = file->f_mapping->host;
6177         struct btrfs_ordered_extent *ordered;
6178         struct extent_state *cached_state = NULL;
6179         u64 lockstart, lockend;
6180         ssize_t ret;
6181         int writing = rw & WRITE;
6182         int write_bits = 0;
6183         size_t count = iov_length(iov, nr_segs);
6184
6185         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6186                             offset, nr_segs)) {
6187                 return 0;
6188         }
6189
6190         lockstart = offset;
6191         lockend = offset + count - 1;
6192
6193         if (writing) {
6194                 ret = btrfs_delalloc_reserve_space(inode, count);
6195                 if (ret)
6196                         goto out;
6197         }
6198
6199         while (1) {
6200                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6201                                  0, &cached_state, GFP_NOFS);
6202                 /*
6203                  * We're concerned with the entire range that we're going to be
6204                  * doing DIO to, so we need to make sure theres no ordered
6205                  * extents in this range.
6206                  */
6207                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6208                                                      lockend - lockstart + 1);
6209                 if (!ordered)
6210                         break;
6211                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6212                                      &cached_state, GFP_NOFS);
6213                 btrfs_start_ordered_extent(inode, ordered, 1);
6214                 btrfs_put_ordered_extent(ordered);
6215                 cond_resched();
6216         }
6217
6218         /*
6219          * we don't use btrfs_set_extent_delalloc because we don't want
6220          * the dirty or uptodate bits
6221          */
6222         if (writing) {
6223                 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6224                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6225                                      EXTENT_DELALLOC, 0, NULL, &cached_state,
6226                                      GFP_NOFS);
6227                 if (ret) {
6228                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6229                                          lockend, EXTENT_LOCKED | write_bits,
6230                                          1, 0, &cached_state, GFP_NOFS);
6231                         goto out;
6232                 }
6233         }
6234
6235         free_extent_state(cached_state);
6236         cached_state = NULL;
6237
6238         ret = __blockdev_direct_IO(rw, iocb, inode,
6239                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6240                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6241                    btrfs_submit_direct, 0);
6242
6243         if (ret < 0 && ret != -EIOCBQUEUED) {
6244                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6245                               offset + iov_length(iov, nr_segs) - 1,
6246                               EXTENT_LOCKED | write_bits, 1, 0,
6247                               &cached_state, GFP_NOFS);
6248         } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6249                 /*
6250                  * We're falling back to buffered, unlock the section we didn't
6251                  * do IO on.
6252                  */
6253                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6254                               offset + iov_length(iov, nr_segs) - 1,
6255                               EXTENT_LOCKED | write_bits, 1, 0,
6256                               &cached_state, GFP_NOFS);
6257         }
6258 out:
6259         free_extent_state(cached_state);
6260         return ret;
6261 }
6262
6263 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6264                 __u64 start, __u64 len)
6265 {
6266         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6267 }
6268
6269 int btrfs_readpage(struct file *file, struct page *page)
6270 {
6271         struct extent_io_tree *tree;
6272         tree = &BTRFS_I(page->mapping->host)->io_tree;
6273         return extent_read_full_page(tree, page, btrfs_get_extent);
6274 }
6275
6276 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6277 {
6278         struct extent_io_tree *tree;
6279
6280
6281         if (current->flags & PF_MEMALLOC) {
6282                 redirty_page_for_writepage(wbc, page);
6283                 unlock_page(page);
6284                 return 0;
6285         }
6286         tree = &BTRFS_I(page->mapping->host)->io_tree;
6287         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6288 }
6289
6290 int btrfs_writepages(struct address_space *mapping,
6291                      struct writeback_control *wbc)
6292 {
6293         struct extent_io_tree *tree;
6294
6295         tree = &BTRFS_I(mapping->host)->io_tree;
6296         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6297 }
6298
6299 static int
6300 btrfs_readpages(struct file *file, struct address_space *mapping,
6301                 struct list_head *pages, unsigned nr_pages)
6302 {
6303         struct extent_io_tree *tree;
6304         tree = &BTRFS_I(mapping->host)->io_tree;
6305         return extent_readpages(tree, mapping, pages, nr_pages,
6306                                 btrfs_get_extent);
6307 }
6308 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6309 {
6310         struct extent_io_tree *tree;
6311         struct extent_map_tree *map;
6312         int ret;
6313
6314         tree = &BTRFS_I(page->mapping->host)->io_tree;
6315         map = &BTRFS_I(page->mapping->host)->extent_tree;
6316         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6317         if (ret == 1) {
6318                 ClearPagePrivate(page);
6319                 set_page_private(page, 0);
6320                 page_cache_release(page);
6321         }
6322         return ret;
6323 }
6324
6325 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6326 {
6327         if (PageWriteback(page) || PageDirty(page))
6328                 return 0;
6329         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6330 }
6331
6332 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6333 {
6334         struct extent_io_tree *tree;
6335         struct btrfs_ordered_extent *ordered;
6336         struct extent_state *cached_state = NULL;
6337         u64 page_start = page_offset(page);
6338         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6339
6340
6341         /*
6342          * we have the page locked, so new writeback can't start,
6343          * and the dirty bit won't be cleared while we are here.
6344          *
6345          * Wait for IO on this page so that we can safely clear
6346          * the PagePrivate2 bit and do ordered accounting
6347          */
6348         wait_on_page_writeback(page);
6349
6350         tree = &BTRFS_I(page->mapping->host)->io_tree;
6351         if (offset) {
6352                 btrfs_releasepage(page, GFP_NOFS);
6353                 return;
6354         }
6355         lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6356                          GFP_NOFS);
6357         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6358                                            page_offset(page));
6359         if (ordered) {
6360                 /*
6361                  * IO on this page will never be started, so we need
6362                  * to account for any ordered extents now
6363                  */
6364                 clear_extent_bit(tree, page_start, page_end,
6365                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6366                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6367                                  &cached_state, GFP_NOFS);
6368                 /*
6369                  * whoever cleared the private bit is responsible
6370                  * for the finish_ordered_io
6371                  */
6372                 if (TestClearPagePrivate2(page)) {
6373                         btrfs_finish_ordered_io(page->mapping->host,
6374                                                 page_start, page_end);
6375                 }
6376                 btrfs_put_ordered_extent(ordered);
6377                 cached_state = NULL;
6378                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6379                                  GFP_NOFS);
6380         }
6381         clear_extent_bit(tree, page_start, page_end,
6382                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6383                  EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6384         __btrfs_releasepage(page, GFP_NOFS);
6385
6386         ClearPageChecked(page);
6387         if (PagePrivate(page)) {
6388                 ClearPagePrivate(page);
6389                 set_page_private(page, 0);
6390                 page_cache_release(page);
6391         }
6392 }
6393
6394 /*
6395  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6396  * called from a page fault handler when a page is first dirtied. Hence we must
6397  * be careful to check for EOF conditions here. We set the page up correctly
6398  * for a written page which means we get ENOSPC checking when writing into
6399  * holes and correct delalloc and unwritten extent mapping on filesystems that
6400  * support these features.
6401  *
6402  * We are not allowed to take the i_mutex here so we have to play games to
6403  * protect against truncate races as the page could now be beyond EOF.  Because
6404  * vmtruncate() writes the inode size before removing pages, once we have the
6405  * page lock we can determine safely if the page is beyond EOF. If it is not
6406  * beyond EOF, then the page is guaranteed safe against truncation until we
6407  * unlock the page.
6408  */
6409 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6410 {
6411         struct page *page = vmf->page;
6412         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6413         struct btrfs_root *root = BTRFS_I(inode)->root;
6414         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6415         struct btrfs_ordered_extent *ordered;
6416         struct extent_state *cached_state = NULL;
6417         char *kaddr;
6418         unsigned long zero_start;
6419         loff_t size;
6420         int ret;
6421         u64 page_start;
6422         u64 page_end;
6423
6424         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6425         if (ret) {
6426                 if (ret == -ENOMEM)
6427                         ret = VM_FAULT_OOM;
6428                 else /* -ENOSPC, -EIO, etc */
6429                         ret = VM_FAULT_SIGBUS;
6430                 goto out;
6431         }
6432
6433         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6434 again:
6435         lock_page(page);
6436         size = i_size_read(inode);
6437         page_start = page_offset(page);
6438         page_end = page_start + PAGE_CACHE_SIZE - 1;
6439
6440         if ((page->mapping != inode->i_mapping) ||
6441             (page_start >= size)) {
6442                 /* page got truncated out from underneath us */
6443                 goto out_unlock;
6444         }
6445         wait_on_page_writeback(page);
6446
6447         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6448                          GFP_NOFS);
6449         set_page_extent_mapped(page);
6450
6451         /*
6452          * we can't set the delalloc bits if there are pending ordered
6453          * extents.  Drop our locks and wait for them to finish
6454          */
6455         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6456         if (ordered) {
6457                 unlock_extent_cached(io_tree, page_start, page_end,
6458                                      &cached_state, GFP_NOFS);
6459                 unlock_page(page);
6460                 btrfs_start_ordered_extent(inode, ordered, 1);
6461                 btrfs_put_ordered_extent(ordered);
6462                 goto again;
6463         }
6464
6465         /*
6466          * XXX - page_mkwrite gets called every time the page is dirtied, even
6467          * if it was already dirty, so for space accounting reasons we need to
6468          * clear any delalloc bits for the range we are fixing to save.  There
6469          * is probably a better way to do this, but for now keep consistent with
6470          * prepare_pages in the normal write path.
6471          */
6472         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6473                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6474                           0, 0, &cached_state, GFP_NOFS);
6475
6476         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6477                                         &cached_state);
6478         if (ret) {
6479                 unlock_extent_cached(io_tree, page_start, page_end,
6480                                      &cached_state, GFP_NOFS);
6481                 ret = VM_FAULT_SIGBUS;
6482                 goto out_unlock;
6483         }
6484         ret = 0;
6485
6486         /* page is wholly or partially inside EOF */
6487         if (page_start + PAGE_CACHE_SIZE > size)
6488                 zero_start = size & ~PAGE_CACHE_MASK;
6489         else
6490                 zero_start = PAGE_CACHE_SIZE;
6491
6492         if (zero_start != PAGE_CACHE_SIZE) {
6493                 kaddr = kmap(page);
6494                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6495                 flush_dcache_page(page);
6496                 kunmap(page);
6497         }
6498         ClearPageChecked(page);
6499         set_page_dirty(page);
6500         SetPageUptodate(page);
6501
6502         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6503         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6504
6505         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6506
6507 out_unlock:
6508         if (!ret)
6509                 return VM_FAULT_LOCKED;
6510         unlock_page(page);
6511         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6512 out:
6513         return ret;
6514 }
6515
6516 static int btrfs_truncate(struct inode *inode)
6517 {
6518         struct btrfs_root *root = BTRFS_I(inode)->root;
6519         struct btrfs_block_rsv *rsv;
6520         int ret;
6521         int err = 0;
6522         struct btrfs_trans_handle *trans;
6523         unsigned long nr;
6524         u64 mask = root->sectorsize - 1;
6525         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
6526
6527         ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6528         if (ret)
6529                 return ret;
6530
6531         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6532         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6533
6534         /*
6535          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
6536          * 3 things going on here
6537          *
6538          * 1) We need to reserve space for our orphan item and the space to
6539          * delete our orphan item.  Lord knows we don't want to have a dangling
6540          * orphan item because we didn't reserve space to remove it.
6541          *
6542          * 2) We need to reserve space to update our inode.
6543          *
6544          * 3) We need to have something to cache all the space that is going to
6545          * be free'd up by the truncate operation, but also have some slack
6546          * space reserved in case it uses space during the truncate (thank you
6547          * very much snapshotting).
6548          *
6549          * And we need these to all be seperate.  The fact is we can use alot of
6550          * space doing the truncate, and we have no earthly idea how much space
6551          * we will use, so we need the truncate reservation to be seperate so it
6552          * doesn't end up using space reserved for updating the inode or
6553          * removing the orphan item.  We also need to be able to stop the
6554          * transaction and start a new one, which means we need to be able to
6555          * update the inode several times, and we have no idea of knowing how
6556          * many times that will be, so we can't just reserve 1 item for the
6557          * entirety of the opration, so that has to be done seperately as well.
6558          * Then there is the orphan item, which does indeed need to be held on
6559          * to for the whole operation, and we need nobody to touch this reserved
6560          * space except the orphan code.
6561          *
6562          * So that leaves us with
6563          *
6564          * 1) root->orphan_block_rsv - for the orphan deletion.
6565          * 2) rsv - for the truncate reservation, which we will steal from the
6566          * transaction reservation.
6567          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6568          * updating the inode.
6569          */
6570         rsv = btrfs_alloc_block_rsv(root);
6571         if (!rsv)
6572                 return -ENOMEM;
6573         rsv->size = min_size;
6574
6575         /*
6576          * 1 for the truncate slack space
6577          * 1 for the orphan item we're going to add
6578          * 1 for the orphan item deletion
6579          * 1 for updating the inode.
6580          */
6581         trans = btrfs_start_transaction(root, 4);
6582         if (IS_ERR(trans)) {
6583                 err = PTR_ERR(trans);
6584                 goto out;
6585         }
6586
6587         /* Migrate the slack space for the truncate to our reserve */
6588         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6589                                       min_size);
6590         BUG_ON(ret);
6591
6592         ret = btrfs_orphan_add(trans, inode);
6593         if (ret) {
6594                 btrfs_end_transaction(trans, root);
6595                 goto out;
6596         }
6597
6598         /*
6599          * setattr is responsible for setting the ordered_data_close flag,
6600          * but that is only tested during the last file release.  That
6601          * could happen well after the next commit, leaving a great big
6602          * window where new writes may get lost if someone chooses to write
6603          * to this file after truncating to zero
6604          *
6605          * The inode doesn't have any dirty data here, and so if we commit
6606          * this is a noop.  If someone immediately starts writing to the inode
6607          * it is very likely we'll catch some of their writes in this
6608          * transaction, and the commit will find this file on the ordered
6609          * data list with good things to send down.
6610          *
6611          * This is a best effort solution, there is still a window where
6612          * using truncate to replace the contents of the file will
6613          * end up with a zero length file after a crash.
6614          */
6615         if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6616                 btrfs_add_ordered_operation(trans, root, inode);
6617
6618         while (1) {
6619                 ret = btrfs_block_rsv_refill(root, rsv, min_size);
6620                 if (ret) {
6621                         /*
6622                          * This can only happen with the original transaction we
6623                          * started above, every other time we shouldn't have a
6624                          * transaction started yet.
6625                          */
6626                         if (ret == -EAGAIN)
6627                                 goto end_trans;
6628                         err = ret;
6629                         break;
6630                 }
6631
6632                 if (!trans) {
6633                         /* Just need the 1 for updating the inode */
6634                         trans = btrfs_start_transaction(root, 1);
6635                         if (IS_ERR(trans)) {
6636                                 err = PTR_ERR(trans);
6637                                 goto out;
6638                         }
6639                 }
6640
6641                 trans->block_rsv = rsv;
6642
6643                 ret = btrfs_truncate_inode_items(trans, root, inode,
6644                                                  inode->i_size,
6645                                                  BTRFS_EXTENT_DATA_KEY);
6646                 if (ret != -EAGAIN) {
6647                         err = ret;
6648                         break;
6649                 }
6650
6651                 trans->block_rsv = &root->fs_info->trans_block_rsv;
6652                 ret = btrfs_update_inode(trans, root, inode);
6653                 if (ret) {
6654                         err = ret;
6655                         break;
6656                 }
6657 end_trans:
6658                 nr = trans->blocks_used;
6659                 btrfs_end_transaction(trans, root);
6660                 trans = NULL;
6661                 btrfs_btree_balance_dirty(root, nr);
6662         }
6663
6664         if (ret == 0 && inode->i_nlink > 0) {
6665                 trans->block_rsv = root->orphan_block_rsv;
6666                 ret = btrfs_orphan_del(trans, inode);
6667                 if (ret)
6668                         err = ret;
6669         } else if (ret && inode->i_nlink > 0) {
6670                 /*
6671                  * Failed to do the truncate, remove us from the in memory
6672                  * orphan list.
6673                  */
6674                 ret = btrfs_orphan_del(NULL, inode);
6675         }
6676
6677         trans->block_rsv = &root->fs_info->trans_block_rsv;
6678         ret = btrfs_update_inode(trans, root, inode);
6679         if (ret && !err)
6680                 err = ret;
6681
6682         nr = trans->blocks_used;
6683         ret = btrfs_end_transaction_throttle(trans, root);
6684         btrfs_btree_balance_dirty(root, nr);
6685
6686 out:
6687         btrfs_free_block_rsv(root, rsv);
6688
6689         if (ret && !err)
6690                 err = ret;
6691
6692         return err;
6693 }
6694
6695 /*
6696  * create a new subvolume directory/inode (helper for the ioctl).
6697  */
6698 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6699                              struct btrfs_root *new_root, u64 new_dirid)
6700 {
6701         struct inode *inode;
6702         int err;
6703         u64 index = 0;
6704
6705         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6706                                 new_dirid, S_IFDIR | 0700, &index);
6707         if (IS_ERR(inode))
6708                 return PTR_ERR(inode);
6709         inode->i_op = &btrfs_dir_inode_operations;
6710         inode->i_fop = &btrfs_dir_file_operations;
6711
6712         inode->i_nlink = 1;
6713         btrfs_i_size_write(inode, 0);
6714
6715         err = btrfs_update_inode(trans, new_root, inode);
6716         BUG_ON(err);
6717
6718         iput(inode);
6719         return 0;
6720 }
6721
6722 struct inode *btrfs_alloc_inode(struct super_block *sb)
6723 {
6724         struct btrfs_inode *ei;
6725         struct inode *inode;
6726
6727         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6728         if (!ei)
6729                 return NULL;
6730
6731         ei->root = NULL;
6732         ei->space_info = NULL;
6733         ei->generation = 0;
6734         ei->sequence = 0;
6735         ei->last_trans = 0;
6736         ei->last_sub_trans = 0;
6737         ei->logged_trans = 0;
6738         ei->delalloc_bytes = 0;
6739         ei->disk_i_size = 0;
6740         ei->flags = 0;
6741         ei->csum_bytes = 0;
6742         ei->index_cnt = (u64)-1;
6743         ei->last_unlink_trans = 0;
6744
6745         spin_lock_init(&ei->lock);
6746         ei->outstanding_extents = 0;
6747         ei->reserved_extents = 0;
6748
6749         ei->ordered_data_close = 0;
6750         ei->orphan_meta_reserved = 0;
6751         ei->dummy_inode = 0;
6752         ei->in_defrag = 0;
6753         ei->force_compress = BTRFS_COMPRESS_NONE;
6754
6755         ei->delayed_node = NULL;
6756
6757         inode = &ei->vfs_inode;
6758         extent_map_tree_init(&ei->extent_tree);
6759         extent_io_tree_init(&ei->io_tree, &inode->i_data);
6760         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
6761         mutex_init(&ei->log_mutex);
6762         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6763         INIT_LIST_HEAD(&ei->i_orphan);
6764         INIT_LIST_HEAD(&ei->delalloc_inodes);
6765         INIT_LIST_HEAD(&ei->ordered_operations);
6766         RB_CLEAR_NODE(&ei->rb_node);
6767
6768         return inode;
6769 }
6770
6771 static void btrfs_i_callback(struct rcu_head *head)
6772 {
6773         struct inode *inode = container_of(head, struct inode, i_rcu);
6774         INIT_LIST_HEAD(&inode->i_dentry);
6775         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6776 }
6777
6778 void btrfs_destroy_inode(struct inode *inode)
6779 {
6780         struct btrfs_ordered_extent *ordered;
6781         struct btrfs_root *root = BTRFS_I(inode)->root;
6782
6783         WARN_ON(!list_empty(&inode->i_dentry));
6784         WARN_ON(inode->i_data.nrpages);
6785         WARN_ON(BTRFS_I(inode)->outstanding_extents);
6786         WARN_ON(BTRFS_I(inode)->reserved_extents);
6787         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
6788         WARN_ON(BTRFS_I(inode)->csum_bytes);
6789
6790         /*
6791          * This can happen where we create an inode, but somebody else also
6792          * created the same inode and we need to destroy the one we already
6793          * created.
6794          */
6795         if (!root)
6796                 goto free;
6797
6798         /*
6799          * Make sure we're properly removed from the ordered operation
6800          * lists.
6801          */
6802         smp_mb();
6803         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6804                 spin_lock(&root->fs_info->ordered_extent_lock);
6805                 list_del_init(&BTRFS_I(inode)->ordered_operations);
6806                 spin_unlock(&root->fs_info->ordered_extent_lock);
6807         }
6808
6809         spin_lock(&root->orphan_lock);
6810         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6811                 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
6812                        (unsigned long long)btrfs_ino(inode));
6813                 list_del_init(&BTRFS_I(inode)->i_orphan);
6814         }
6815         spin_unlock(&root->orphan_lock);
6816
6817         while (1) {
6818                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6819                 if (!ordered)
6820                         break;
6821                 else {
6822                         printk(KERN_ERR "btrfs found ordered "
6823                                "extent %llu %llu on inode cleanup\n",
6824                                (unsigned long long)ordered->file_offset,
6825                                (unsigned long long)ordered->len);
6826                         btrfs_remove_ordered_extent(inode, ordered);
6827                         btrfs_put_ordered_extent(ordered);
6828                         btrfs_put_ordered_extent(ordered);
6829                 }
6830         }
6831         inode_tree_del(inode);
6832         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6833 free:
6834         btrfs_remove_delayed_node(inode);
6835         call_rcu(&inode->i_rcu, btrfs_i_callback);
6836 }
6837
6838 int btrfs_drop_inode(struct inode *inode)
6839 {
6840         struct btrfs_root *root = BTRFS_I(inode)->root;
6841
6842         if (btrfs_root_refs(&root->root_item) == 0 &&
6843             !btrfs_is_free_space_inode(root, inode))
6844                 return 1;
6845         else
6846                 return generic_drop_inode(inode);
6847 }
6848
6849 static void init_once(void *foo)
6850 {
6851         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6852
6853         inode_init_once(&ei->vfs_inode);
6854 }
6855
6856 void btrfs_destroy_cachep(void)
6857 {
6858         if (btrfs_inode_cachep)
6859                 kmem_cache_destroy(btrfs_inode_cachep);
6860         if (btrfs_trans_handle_cachep)
6861                 kmem_cache_destroy(btrfs_trans_handle_cachep);
6862         if (btrfs_transaction_cachep)
6863                 kmem_cache_destroy(btrfs_transaction_cachep);
6864         if (btrfs_path_cachep)
6865                 kmem_cache_destroy(btrfs_path_cachep);
6866         if (btrfs_free_space_cachep)
6867                 kmem_cache_destroy(btrfs_free_space_cachep);
6868 }
6869
6870 int btrfs_init_cachep(void)
6871 {
6872         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6873                         sizeof(struct btrfs_inode), 0,
6874                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6875         if (!btrfs_inode_cachep)
6876                 goto fail;
6877
6878         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6879                         sizeof(struct btrfs_trans_handle), 0,
6880                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6881         if (!btrfs_trans_handle_cachep)
6882                 goto fail;
6883
6884         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6885                         sizeof(struct btrfs_transaction), 0,
6886                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6887         if (!btrfs_transaction_cachep)
6888                 goto fail;
6889
6890         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6891                         sizeof(struct btrfs_path), 0,
6892                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6893         if (!btrfs_path_cachep)
6894                 goto fail;
6895
6896         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6897                         sizeof(struct btrfs_free_space), 0,
6898                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6899         if (!btrfs_free_space_cachep)
6900                 goto fail;
6901
6902         return 0;
6903 fail:
6904         btrfs_destroy_cachep();
6905         return -ENOMEM;
6906 }
6907
6908 static int btrfs_getattr(struct vfsmount *mnt,
6909                          struct dentry *dentry, struct kstat *stat)
6910 {
6911         struct inode *inode = dentry->d_inode;
6912         generic_fillattr(inode, stat);
6913         stat->dev = BTRFS_I(inode)->root->anon_dev;
6914         stat->blksize = PAGE_CACHE_SIZE;
6915         stat->blocks = (inode_get_bytes(inode) +
6916                         BTRFS_I(inode)->delalloc_bytes) >> 9;
6917         return 0;
6918 }
6919
6920 /*
6921  * If a file is moved, it will inherit the cow and compression flags of the new
6922  * directory.
6923  */
6924 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6925 {
6926         struct btrfs_inode *b_dir = BTRFS_I(dir);
6927         struct btrfs_inode *b_inode = BTRFS_I(inode);
6928
6929         if (b_dir->flags & BTRFS_INODE_NODATACOW)
6930                 b_inode->flags |= BTRFS_INODE_NODATACOW;
6931         else
6932                 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6933
6934         if (b_dir->flags & BTRFS_INODE_COMPRESS)
6935                 b_inode->flags |= BTRFS_INODE_COMPRESS;
6936         else
6937                 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6938 }
6939
6940 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6941                            struct inode *new_dir, struct dentry *new_dentry)
6942 {
6943         struct btrfs_trans_handle *trans;
6944         struct btrfs_root *root = BTRFS_I(old_dir)->root;
6945         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6946         struct inode *new_inode = new_dentry->d_inode;
6947         struct inode *old_inode = old_dentry->d_inode;
6948         struct timespec ctime = CURRENT_TIME;
6949         u64 index = 0;
6950         u64 root_objectid;
6951         int ret;
6952         u64 old_ino = btrfs_ino(old_inode);
6953
6954         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6955                 return -EPERM;
6956
6957         /* we only allow rename subvolume link between subvolumes */
6958         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6959                 return -EXDEV;
6960
6961         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6962             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
6963                 return -ENOTEMPTY;
6964
6965         if (S_ISDIR(old_inode->i_mode) && new_inode &&
6966             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6967                 return -ENOTEMPTY;
6968         /*
6969          * we're using rename to replace one file with another.
6970          * and the replacement file is large.  Start IO on it now so
6971          * we don't add too much work to the end of the transaction
6972          */
6973         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
6974             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6975                 filemap_flush(old_inode->i_mapping);
6976
6977         /* close the racy window with snapshot create/destroy ioctl */
6978         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
6979                 down_read(&root->fs_info->subvol_sem);
6980         /*
6981          * We want to reserve the absolute worst case amount of items.  So if
6982          * both inodes are subvols and we need to unlink them then that would
6983          * require 4 item modifications, but if they are both normal inodes it
6984          * would require 5 item modifications, so we'll assume their normal
6985          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6986          * should cover the worst case number of items we'll modify.
6987          */
6988         trans = btrfs_start_transaction(root, 20);
6989         if (IS_ERR(trans)) {
6990                 ret = PTR_ERR(trans);
6991                 goto out_notrans;
6992         }
6993
6994         if (dest != root)
6995                 btrfs_record_root_in_trans(trans, dest);
6996
6997         ret = btrfs_set_inode_index(new_dir, &index);
6998         if (ret)
6999                 goto out_fail;
7000
7001         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7002                 /* force full log commit if subvolume involved. */
7003                 root->fs_info->last_trans_log_full_commit = trans->transid;
7004         } else {
7005                 ret = btrfs_insert_inode_ref(trans, dest,
7006                                              new_dentry->d_name.name,
7007                                              new_dentry->d_name.len,
7008                                              old_ino,
7009                                              btrfs_ino(new_dir), index);
7010                 if (ret)
7011                         goto out_fail;
7012                 /*
7013                  * this is an ugly little race, but the rename is required
7014                  * to make sure that if we crash, the inode is either at the
7015                  * old name or the new one.  pinning the log transaction lets
7016                  * us make sure we don't allow a log commit to come in after
7017                  * we unlink the name but before we add the new name back in.
7018                  */
7019                 btrfs_pin_log_trans(root);
7020         }
7021         /*
7022          * make sure the inode gets flushed if it is replacing
7023          * something.
7024          */
7025         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7026                 btrfs_add_ordered_operation(trans, root, old_inode);
7027
7028         old_dir->i_ctime = old_dir->i_mtime = ctime;
7029         new_dir->i_ctime = new_dir->i_mtime = ctime;
7030         old_inode->i_ctime = ctime;
7031
7032         if (old_dentry->d_parent != new_dentry->d_parent)
7033                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7034
7035         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7036                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7037                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7038                                         old_dentry->d_name.name,
7039                                         old_dentry->d_name.len);
7040         } else {
7041                 ret = __btrfs_unlink_inode(trans, root, old_dir,
7042                                         old_dentry->d_inode,
7043                                         old_dentry->d_name.name,
7044                                         old_dentry->d_name.len);
7045                 if (!ret)
7046                         ret = btrfs_update_inode(trans, root, old_inode);
7047         }
7048         BUG_ON(ret);
7049
7050         if (new_inode) {
7051                 new_inode->i_ctime = CURRENT_TIME;
7052                 if (unlikely(btrfs_ino(new_inode) ==
7053                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7054                         root_objectid = BTRFS_I(new_inode)->location.objectid;
7055                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
7056                                                 root_objectid,
7057                                                 new_dentry->d_name.name,
7058                                                 new_dentry->d_name.len);
7059                         BUG_ON(new_inode->i_nlink == 0);
7060                 } else {
7061                         ret = btrfs_unlink_inode(trans, dest, new_dir,
7062                                                  new_dentry->d_inode,
7063                                                  new_dentry->d_name.name,
7064                                                  new_dentry->d_name.len);
7065                 }
7066                 BUG_ON(ret);
7067                 if (new_inode->i_nlink == 0) {
7068                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7069                         BUG_ON(ret);
7070                 }
7071         }
7072
7073         fixup_inode_flags(new_dir, old_inode);
7074
7075         ret = btrfs_add_link(trans, new_dir, old_inode,
7076                              new_dentry->d_name.name,
7077                              new_dentry->d_name.len, 0, index);
7078         BUG_ON(ret);
7079
7080         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7081                 struct dentry *parent = new_dentry->d_parent;
7082                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7083                 btrfs_end_log_trans(root);
7084         }
7085 out_fail:
7086         btrfs_end_transaction_throttle(trans, root);
7087 out_notrans:
7088         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7089                 up_read(&root->fs_info->subvol_sem);
7090
7091         return ret;
7092 }
7093
7094 /*
7095  * some fairly slow code that needs optimization. This walks the list
7096  * of all the inodes with pending delalloc and forces them to disk.
7097  */
7098 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7099 {
7100         struct list_head *head = &root->fs_info->delalloc_inodes;
7101         struct btrfs_inode *binode;
7102         struct inode *inode;
7103
7104         if (root->fs_info->sb->s_flags & MS_RDONLY)
7105                 return -EROFS;
7106
7107         spin_lock(&root->fs_info->delalloc_lock);
7108         while (!list_empty(head)) {
7109                 binode = list_entry(head->next, struct btrfs_inode,
7110                                     delalloc_inodes);
7111                 inode = igrab(&binode->vfs_inode);
7112                 if (!inode)
7113                         list_del_init(&binode->delalloc_inodes);
7114                 spin_unlock(&root->fs_info->delalloc_lock);
7115                 if (inode) {
7116                         filemap_flush(inode->i_mapping);
7117                         if (delay_iput)
7118                                 btrfs_add_delayed_iput(inode);
7119                         else
7120                                 iput(inode);
7121                 }
7122                 cond_resched();
7123                 spin_lock(&root->fs_info->delalloc_lock);
7124         }
7125         spin_unlock(&root->fs_info->delalloc_lock);
7126
7127         /* the filemap_flush will queue IO into the worker threads, but
7128          * we have to make sure the IO is actually started and that
7129          * ordered extents get created before we return
7130          */
7131         atomic_inc(&root->fs_info->async_submit_draining);
7132         while (atomic_read(&root->fs_info->nr_async_submits) ||
7133               atomic_read(&root->fs_info->async_delalloc_pages)) {
7134                 wait_event(root->fs_info->async_submit_wait,
7135                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7136                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7137         }
7138         atomic_dec(&root->fs_info->async_submit_draining);
7139         return 0;
7140 }
7141
7142 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7143                          const char *symname)
7144 {
7145         struct btrfs_trans_handle *trans;
7146         struct btrfs_root *root = BTRFS_I(dir)->root;
7147         struct btrfs_path *path;
7148         struct btrfs_key key;
7149         struct inode *inode = NULL;
7150         int err;
7151         int drop_inode = 0;
7152         u64 objectid;
7153         u64 index = 0 ;
7154         int name_len;
7155         int datasize;
7156         unsigned long ptr;
7157         struct btrfs_file_extent_item *ei;
7158         struct extent_buffer *leaf;
7159         unsigned long nr = 0;
7160
7161         name_len = strlen(symname) + 1;
7162         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7163                 return -ENAMETOOLONG;
7164
7165         /*
7166          * 2 items for inode item and ref
7167          * 2 items for dir items
7168          * 1 item for xattr if selinux is on
7169          */
7170         trans = btrfs_start_transaction(root, 5);
7171         if (IS_ERR(trans))
7172                 return PTR_ERR(trans);
7173
7174         err = btrfs_find_free_ino(root, &objectid);
7175         if (err)
7176                 goto out_unlock;
7177
7178         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7179                                 dentry->d_name.len, btrfs_ino(dir), objectid,
7180                                 S_IFLNK|S_IRWXUGO, &index);
7181         if (IS_ERR(inode)) {
7182                 err = PTR_ERR(inode);
7183                 goto out_unlock;
7184         }
7185
7186         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7187         if (err) {
7188                 drop_inode = 1;
7189                 goto out_unlock;
7190         }
7191
7192         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7193         if (err)
7194                 drop_inode = 1;
7195         else {
7196                 inode->i_mapping->a_ops = &btrfs_aops;
7197                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7198                 inode->i_fop = &btrfs_file_operations;
7199                 inode->i_op = &btrfs_file_inode_operations;
7200                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7201         }
7202         if (drop_inode)
7203                 goto out_unlock;
7204
7205         path = btrfs_alloc_path();
7206         if (!path) {
7207                 err = -ENOMEM;
7208                 drop_inode = 1;
7209                 goto out_unlock;
7210         }
7211         key.objectid = btrfs_ino(inode);
7212         key.offset = 0;
7213         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7214         datasize = btrfs_file_extent_calc_inline_size(name_len);
7215         err = btrfs_insert_empty_item(trans, root, path, &key,
7216                                       datasize);
7217         if (err) {
7218                 drop_inode = 1;
7219                 btrfs_free_path(path);
7220                 goto out_unlock;
7221         }
7222         leaf = path->nodes[0];
7223         ei = btrfs_item_ptr(leaf, path->slots[0],
7224                             struct btrfs_file_extent_item);
7225         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7226         btrfs_set_file_extent_type(leaf, ei,
7227                                    BTRFS_FILE_EXTENT_INLINE);
7228         btrfs_set_file_extent_encryption(leaf, ei, 0);
7229         btrfs_set_file_extent_compression(leaf, ei, 0);
7230         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7231         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7232
7233         ptr = btrfs_file_extent_inline_start(ei);
7234         write_extent_buffer(leaf, symname, ptr, name_len);
7235         btrfs_mark_buffer_dirty(leaf);
7236         btrfs_free_path(path);
7237
7238         inode->i_op = &btrfs_symlink_inode_operations;
7239         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7240         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7241         inode_set_bytes(inode, name_len);
7242         btrfs_i_size_write(inode, name_len - 1);
7243         err = btrfs_update_inode(trans, root, inode);
7244         if (err)
7245                 drop_inode = 1;
7246
7247 out_unlock:
7248         nr = trans->blocks_used;
7249         btrfs_end_transaction_throttle(trans, root);
7250         if (drop_inode) {
7251                 inode_dec_link_count(inode);
7252                 iput(inode);
7253         }
7254         btrfs_btree_balance_dirty(root, nr);
7255         return err;
7256 }
7257
7258 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7259                                        u64 start, u64 num_bytes, u64 min_size,
7260                                        loff_t actual_len, u64 *alloc_hint,
7261                                        struct btrfs_trans_handle *trans)
7262 {
7263         struct btrfs_root *root = BTRFS_I(inode)->root;
7264         struct btrfs_key ins;
7265         u64 cur_offset = start;
7266         u64 i_size;
7267         int ret = 0;
7268         bool own_trans = true;
7269
7270         if (trans)
7271                 own_trans = false;
7272         while (num_bytes > 0) {
7273                 if (own_trans) {
7274                         trans = btrfs_start_transaction(root, 3);
7275                         if (IS_ERR(trans)) {
7276                                 ret = PTR_ERR(trans);
7277                                 break;
7278                         }
7279                 }
7280
7281                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7282                                            0, *alloc_hint, (u64)-1, &ins, 1);
7283                 if (ret) {
7284                         if (own_trans)
7285                                 btrfs_end_transaction(trans, root);
7286                         break;
7287                 }
7288
7289                 ret = insert_reserved_file_extent(trans, inode,
7290                                                   cur_offset, ins.objectid,
7291                                                   ins.offset, ins.offset,
7292                                                   ins.offset, 0, 0, 0,
7293                                                   BTRFS_FILE_EXTENT_PREALLOC);
7294                 BUG_ON(ret);
7295                 btrfs_drop_extent_cache(inode, cur_offset,
7296                                         cur_offset + ins.offset -1, 0);
7297
7298                 num_bytes -= ins.offset;
7299                 cur_offset += ins.offset;
7300                 *alloc_hint = ins.objectid + ins.offset;
7301
7302                 inode->i_ctime = CURRENT_TIME;
7303                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7304                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7305                     (actual_len > inode->i_size) &&
7306                     (cur_offset > inode->i_size)) {
7307                         if (cur_offset > actual_len)
7308                                 i_size = actual_len;
7309                         else
7310                                 i_size = cur_offset;
7311                         i_size_write(inode, i_size);
7312                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7313                 }
7314
7315                 ret = btrfs_update_inode(trans, root, inode);
7316                 BUG_ON(ret);
7317
7318                 if (own_trans)
7319                         btrfs_end_transaction(trans, root);
7320         }
7321         return ret;
7322 }
7323
7324 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7325                               u64 start, u64 num_bytes, u64 min_size,
7326                               loff_t actual_len, u64 *alloc_hint)
7327 {
7328         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7329                                            min_size, actual_len, alloc_hint,
7330                                            NULL);
7331 }
7332
7333 int btrfs_prealloc_file_range_trans(struct inode *inode,
7334                                     struct btrfs_trans_handle *trans, int mode,
7335                                     u64 start, u64 num_bytes, u64 min_size,
7336                                     loff_t actual_len, u64 *alloc_hint)
7337 {
7338         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7339                                            min_size, actual_len, alloc_hint, trans);
7340 }
7341
7342 static int btrfs_set_page_dirty(struct page *page)
7343 {
7344         return __set_page_dirty_nobuffers(page);
7345 }
7346
7347 static int btrfs_permission(struct inode *inode, int mask)
7348 {
7349         struct btrfs_root *root = BTRFS_I(inode)->root;
7350         umode_t mode = inode->i_mode;
7351
7352         if (mask & MAY_WRITE &&
7353             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7354                 if (btrfs_root_readonly(root))
7355                         return -EROFS;
7356                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7357                         return -EACCES;
7358         }
7359         return generic_permission(inode, mask);
7360 }
7361
7362 static const struct inode_operations btrfs_dir_inode_operations = {
7363         .getattr        = btrfs_getattr,
7364         .lookup         = btrfs_lookup,
7365         .create         = btrfs_create,
7366         .unlink         = btrfs_unlink,
7367         .link           = btrfs_link,
7368         .mkdir          = btrfs_mkdir,
7369         .rmdir          = btrfs_rmdir,
7370         .rename         = btrfs_rename,
7371         .symlink        = btrfs_symlink,
7372         .setattr        = btrfs_setattr,
7373         .mknod          = btrfs_mknod,
7374         .setxattr       = btrfs_setxattr,
7375         .getxattr       = btrfs_getxattr,
7376         .listxattr      = btrfs_listxattr,
7377         .removexattr    = btrfs_removexattr,
7378         .permission     = btrfs_permission,
7379         .get_acl        = btrfs_get_acl,
7380 };
7381 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7382         .lookup         = btrfs_lookup,
7383         .permission     = btrfs_permission,
7384         .get_acl        = btrfs_get_acl,
7385 };
7386
7387 static const struct file_operations btrfs_dir_file_operations = {
7388         .llseek         = generic_file_llseek,
7389         .read           = generic_read_dir,
7390         .readdir        = btrfs_real_readdir,
7391         .unlocked_ioctl = btrfs_ioctl,
7392 #ifdef CONFIG_COMPAT
7393         .compat_ioctl   = btrfs_ioctl,
7394 #endif
7395         .release        = btrfs_release_file,
7396         .fsync          = btrfs_sync_file,
7397 };
7398
7399 static struct extent_io_ops btrfs_extent_io_ops = {
7400         .fill_delalloc = run_delalloc_range,
7401         .submit_bio_hook = btrfs_submit_bio_hook,
7402         .merge_bio_hook = btrfs_merge_bio_hook,
7403         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7404         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7405         .writepage_start_hook = btrfs_writepage_start_hook,
7406         .readpage_io_failed_hook = btrfs_io_failed_hook,
7407         .set_bit_hook = btrfs_set_bit_hook,
7408         .clear_bit_hook = btrfs_clear_bit_hook,
7409         .merge_extent_hook = btrfs_merge_extent_hook,
7410         .split_extent_hook = btrfs_split_extent_hook,
7411 };
7412
7413 /*
7414  * btrfs doesn't support the bmap operation because swapfiles
7415  * use bmap to make a mapping of extents in the file.  They assume
7416  * these extents won't change over the life of the file and they
7417  * use the bmap result to do IO directly to the drive.
7418  *
7419  * the btrfs bmap call would return logical addresses that aren't
7420  * suitable for IO and they also will change frequently as COW
7421  * operations happen.  So, swapfile + btrfs == corruption.
7422  *
7423  * For now we're avoiding this by dropping bmap.
7424  */
7425 static const struct address_space_operations btrfs_aops = {
7426         .readpage       = btrfs_readpage,
7427         .writepage      = btrfs_writepage,
7428         .writepages     = btrfs_writepages,
7429         .readpages      = btrfs_readpages,
7430         .direct_IO      = btrfs_direct_IO,
7431         .invalidatepage = btrfs_invalidatepage,
7432         .releasepage    = btrfs_releasepage,
7433         .set_page_dirty = btrfs_set_page_dirty,
7434         .error_remove_page = generic_error_remove_page,
7435 };
7436
7437 static const struct address_space_operations btrfs_symlink_aops = {
7438         .readpage       = btrfs_readpage,
7439         .writepage      = btrfs_writepage,
7440         .invalidatepage = btrfs_invalidatepage,
7441         .releasepage    = btrfs_releasepage,
7442 };
7443
7444 static const struct inode_operations btrfs_file_inode_operations = {
7445         .getattr        = btrfs_getattr,
7446         .setattr        = btrfs_setattr,
7447         .setxattr       = btrfs_setxattr,
7448         .getxattr       = btrfs_getxattr,
7449         .listxattr      = btrfs_listxattr,
7450         .removexattr    = btrfs_removexattr,
7451         .permission     = btrfs_permission,
7452         .fiemap         = btrfs_fiemap,
7453         .get_acl        = btrfs_get_acl,
7454 };
7455 static const struct inode_operations btrfs_special_inode_operations = {
7456         .getattr        = btrfs_getattr,
7457         .setattr        = btrfs_setattr,
7458         .permission     = btrfs_permission,
7459         .setxattr       = btrfs_setxattr,
7460         .getxattr       = btrfs_getxattr,
7461         .listxattr      = btrfs_listxattr,
7462         .removexattr    = btrfs_removexattr,
7463         .get_acl        = btrfs_get_acl,
7464 };
7465 static const struct inode_operations btrfs_symlink_inode_operations = {
7466         .readlink       = generic_readlink,
7467         .follow_link    = page_follow_link_light,
7468         .put_link       = page_put_link,
7469         .getattr        = btrfs_getattr,
7470         .permission     = btrfs_permission,
7471         .setxattr       = btrfs_setxattr,
7472         .getxattr       = btrfs_getxattr,
7473         .listxattr      = btrfs_listxattr,
7474         .removexattr    = btrfs_removexattr,
7475         .get_acl        = btrfs_get_acl,
7476 };
7477
7478 const struct dentry_operations btrfs_dentry_operations = {
7479         .d_delete       = btrfs_dentry_delete,
7480         .d_release      = btrfs_dentry_release,
7481 };