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 "compat.h"
  41 #include "ctree.h"
  42 #include "disk-io.h"
  43 #include "transaction.h"
  44 #include "btrfs_inode.h"
  45 #include "ioctl.h"
  46 #include "print-tree.h"
  47 #include "volumes.h"
  48 #include "ordered-data.h"
  49 #include "xattr.h"
  50 #include "tree-log.h"
  51 #include "compression.h"
  52 #include "locking.h"
  53 #include "free-space-cache.h"
  54
  55 struct btrfs_iget_args {
  56         u64 ino;
  57         struct btrfs_root *root;
  58 };
  59
  60 static const struct inode_operations btrfs_dir_inode_operations;
  61 static const struct inode_operations btrfs_symlink_inode_operations;
  62 static const struct inode_operations btrfs_dir_ro_inode_operations;
  63 static const struct inode_operations btrfs_special_inode_operations;
  64 static const struct inode_operations btrfs_file_inode_operations;
  65 static const struct address_space_operations btrfs_aops;
  66 static const struct address_space_operations btrfs_symlink_aops;
  67 static const struct file_operations btrfs_dir_file_operations;
  68 static struct extent_io_ops btrfs_extent_io_ops;
  69
  70 static struct kmem_cache *btrfs_inode_cachep;
  71 struct kmem_cache *btrfs_trans_handle_cachep;
  72 struct kmem_cache *btrfs_transaction_cachep;
  73 struct kmem_cache *btrfs_path_cachep;
  74 struct kmem_cache *btrfs_free_space_cachep;
  75
  76 #define S_SHIFT 12
  77 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  78         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  79         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  80         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  81         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  82         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  83         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  84         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  85 };
  86
  87 static int btrfs_setsize(struct inode *inode, loff_t newsize);
  88 static int btrfs_truncate(struct inode *inode);
  89 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
  90 static noinline int cow_file_range(struct inode *inode,
  91                                    struct page *locked_page,
  92                                    u64 start, u64 end, int *page_started,
  93                                    unsigned long *nr_written, int unlock);
  94
  95 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
  96                                      struct inode *inode,  struct inode *dir,
  97                                      const struct qstr *qstr)
  98 {
  99         int err;
 100
 101         err = btrfs_init_acl(trans, inode, dir);
 102         if (!err)
 103                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 104         return err;
 105 }
 106
 107 /*
 108  * this does all the hard work for inserting an inline extent into
 109  * the btree.  The caller should have done a btrfs_drop_extents so that
 110  * no overlapping inline items exist in the btree
 111  */
 112 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 113                                 struct btrfs_root *root, struct inode *inode,
 114                                 u64 start, size_t size, size_t compressed_size,
 115                                 int compress_type,
 116                                 struct page **compressed_pages)
 117 {
 118         struct btrfs_key key;
 119         struct btrfs_path *path;
 120         struct extent_buffer *leaf;
 121         struct page *page = NULL;
 122         char *kaddr;
 123         unsigned long ptr;
 124         struct btrfs_file_extent_item *ei;
 125         int err = 0;
 126         int ret;
 127         size_t cur_size = size;
 128         size_t datasize;
 129         unsigned long offset;
 130
 131         if (compressed_size && compressed_pages)
 132                 cur_size = compressed_size;
 133
 134         path = btrfs_alloc_path();
 135         if (!path)
 136                 return -ENOMEM;
 137
 138         path->leave_spinning = 1;
 139         btrfs_set_trans_block_group(trans, inode);
 140
 141         key.objectid = inode->i_ino;
 142         key.offset = start;
 143         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 144         datasize = btrfs_file_extent_calc_inline_size(cur_size);
 145
 146         inode_add_bytes(inode, size);
 147         ret = btrfs_insert_empty_item(trans, root, path, &key,
 148                                       datasize);
 149         BUG_ON(ret);
 150         if (ret) {
 151                 err = ret;
 152                 goto fail;
 153         }
 154         leaf = path->nodes[0];
 155         ei = btrfs_item_ptr(leaf, path->slots[0],
 156                             struct btrfs_file_extent_item);
 157         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 158         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 159         btrfs_set_file_extent_encryption(leaf, ei, 0);
 160         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 161         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 162         ptr = btrfs_file_extent_inline_start(ei);
 163
 164         if (compress_type != BTRFS_COMPRESS_NONE) {
 165                 struct page *cpage;
 166                 int i = 0;
 167                 while (compressed_size > 0) {
 168                         cpage = compressed_pages[i];
 169                         cur_size = min_t(unsigned long, compressed_size,
 170                                        PAGE_CACHE_SIZE);
 171
 172                         kaddr = kmap_atomic(cpage, KM_USER0);
 173                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 174                         kunmap_atomic(kaddr, KM_USER0);
 175
 176                         i++;
 177                         ptr += cur_size;
 178                         compressed_size -= cur_size;
 179                 }
 180                 btrfs_set_file_extent_compression(leaf, ei,
 181                                                   compress_type);
 182         } else {
 183                 page = find_get_page(inode->i_mapping,
 184                                      start >> PAGE_CACHE_SHIFT);
 185                 btrfs_set_file_extent_compression(leaf, ei, 0);
 186                 kaddr = kmap_atomic(page, KM_USER0);
 187                 offset = start & (PAGE_CACHE_SIZE - 1);
 188                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 189                 kunmap_atomic(kaddr, KM_USER0);
 190                 page_cache_release(page);
 191         }
 192         btrfs_mark_buffer_dirty(leaf);
 193         btrfs_free_path(path);
 194
 195         /*
 196          * we're an inline extent, so nobody can
 197          * extend the file past i_size without locking
 198          * a page we already have locked.
 199          *
 200          * We must do any isize and inode updates
 201          * before we unlock the pages.  Otherwise we
 202          * could end up racing with unlink.
 203          */
 204         BTRFS_I(inode)->disk_i_size = inode->i_size;
 205         btrfs_update_inode(trans, root, inode);
 206
 207         return 0;
 208 fail:
 209         btrfs_free_path(path);
 210         return err;
 211 }
 212
 213
 214 /*
 215  * conditionally insert an inline extent into the file.  This
 216  * does the checks required to make sure the data is small enough
 217  * to fit as an inline extent.
 218  */
 219 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 220                                  struct btrfs_root *root,
 221                                  struct inode *inode, u64 start, u64 end,
 222                                  size_t compressed_size, int compress_type,
 223                                  struct page **compressed_pages)
 224 {
 225         u64 isize = i_size_read(inode);
 226         u64 actual_end = min(end + 1, isize);
 227         u64 inline_len = actual_end - start;
 228         u64 aligned_end = (end + root->sectorsize - 1) &
 229                         ~((u64)root->sectorsize - 1);
 230         u64 hint_byte;
 231         u64 data_len = inline_len;
 232         int ret;
 233
 234         if (compressed_size)
 235                 data_len = compressed_size;
 236
 237         if (start > 0 ||
 238             actual_end >= PAGE_CACHE_SIZE ||
 239             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 240             (!compressed_size &&
 241             (actual_end & (root->sectorsize - 1)) == 0) ||
 242             end + 1 < isize ||
 243             data_len > root->fs_info->max_inline) {
 244                 return 1;
 245         }
 246
 247         ret = btrfs_drop_extents(trans, inode, start, aligned_end,
 248                                  &hint_byte, 1);
 249         BUG_ON(ret);
 250
 251         if (isize > actual_end)
 252                 inline_len = min_t(u64, isize, actual_end);
 253         ret = insert_inline_extent(trans, root, inode, start,
 254                                    inline_len, compressed_size,
 255                                    compress_type, compressed_pages);
 256         BUG_ON(ret);
 257         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 258         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 259         return 0;
 260 }
 261
 262 struct async_extent {
 263         u64 start;
 264         u64 ram_size;
 265         u64 compressed_size;
 266         struct page **pages;
 267         unsigned long nr_pages;
 268         int compress_type;
 269         struct list_head list;
 270 };
 271
 272 struct async_cow {
 273         struct inode *inode;
 274         struct btrfs_root *root;
 275         struct page *locked_page;
 276         u64 start;
 277         u64 end;
 278         struct list_head extents;
 279         struct btrfs_work work;
 280 };
 281
 282 static noinline int add_async_extent(struct async_cow *cow,
 283                                      u64 start, u64 ram_size,
 284                                      u64 compressed_size,
 285                                      struct page **pages,
 286                                      unsigned long nr_pages,
 287                                      int compress_type)
 288 {
 289         struct async_extent *async_extent;
 290
 291         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 292         BUG_ON(!async_extent);
 293         async_extent->start = start;
 294         async_extent->ram_size = ram_size;
 295         async_extent->compressed_size = compressed_size;
 296         async_extent->pages = pages;
 297         async_extent->nr_pages = nr_pages;
 298         async_extent->compress_type = compress_type;
 299         list_add_tail(&async_extent->list, &cow->extents);
 300         return 0;
 301 }
 302
 303 /*
 304  * we create compressed extents in two phases.  The first
 305  * phase compresses a range of pages that have already been
 306  * locked (both pages and state bits are locked).
 307  *
 308  * This is done inside an ordered work queue, and the compression
 309  * is spread across many cpus.  The actual IO submission is step
 310  * two, and the ordered work queue takes care of making sure that
 311  * happens in the same order things were put onto the queue by
 312  * writepages and friends.
 313  *
 314  * If this code finds it can't get good compression, it puts an
 315  * entry onto the work queue to write the uncompressed bytes.  This
 316  * makes sure that both compressed inodes and uncompressed inodes
 317  * are written in the same order that pdflush sent them down.
 318  */
 319 static noinline int compress_file_range(struct inode *inode,
 320                                         struct page *locked_page,
 321                                         u64 start, u64 end,
 322                                         struct async_cow *async_cow,
 323                                         int *num_added)
 324 {
 325         struct btrfs_root *root = BTRFS_I(inode)->root;
 326         struct btrfs_trans_handle *trans;
 327         u64 num_bytes;
 328         u64 blocksize = root->sectorsize;
 329         u64 actual_end;
 330         u64 isize = i_size_read(inode);
 331         int ret = 0;
 332         struct page **pages = NULL;
 333         unsigned long nr_pages;
 334         unsigned long nr_pages_ret = 0;
 335         unsigned long total_compressed = 0;
 336         unsigned long total_in = 0;
 337         unsigned long max_compressed = 128 * 1024;
 338         unsigned long max_uncompressed = 128 * 1024;
 339         int i;
 340         int will_compress;
 341         int compress_type = root->fs_info->compress_type;
 342
 343         actual_end = min_t(u64, isize, end + 1);
 344 again:
 345         will_compress = 0;
 346         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 347         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 348
 349         /*
 350          * we don't want to send crud past the end of i_size through
 351          * compression, that's just a waste of CPU time.  So, if the
 352          * end of the file is before the start of our current
 353          * requested range of bytes, we bail out to the uncompressed
 354          * cleanup code that can deal with all of this.
 355          *
 356          * It isn't really the fastest way to fix things, but this is a
 357          * very uncommon corner.
 358          */
 359         if (actual_end <= start)
 360                 goto cleanup_and_bail_uncompressed;
 361
 362         total_compressed = actual_end - start;
 363
 364         /* we want to make sure that amount of ram required to uncompress
 365          * an extent is reasonable, so we limit the total size in ram
 366          * of a compressed extent to 128k.  This is a crucial number
 367          * because it also controls how easily we can spread reads across
 368          * cpus for decompression.
 369          *
 370          * We also want to make sure the amount of IO required to do
 371          * a random read is reasonably small, so we limit the size of
 372          * a compressed extent to 128k.
 373          */
 374         total_compressed = min(total_compressed, max_uncompressed);
 375         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 376         num_bytes = max(blocksize,  num_bytes);
 377         total_in = 0;
 378         ret = 0;
 379
 380         /*
 381          * we do compression for mount -o compress and when the
 382          * inode has not been flagged as nocompress.  This flag can
 383          * change at any time if we discover bad compression ratios.
 384          */
 385         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 386             (btrfs_test_opt(root, COMPRESS) ||
 387              (BTRFS_I(inode)->force_compress) ||
 388              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 389                 WARN_ON(pages);
 390                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 391                 BUG_ON(!pages);
 392
 393                 if (BTRFS_I(inode)->force_compress)
 394                         compress_type = BTRFS_I(inode)->force_compress;
 395
 396                 ret = btrfs_compress_pages(compress_type,
 397                                            inode->i_mapping, start,
 398                                            total_compressed, pages,
 399                                            nr_pages, &nr_pages_ret,
 400                                            &total_in,
 401                                            &total_compressed,
 402                                            max_compressed);
 403
 404                 if (!ret) {
 405                         unsigned long offset = total_compressed &
 406                                 (PAGE_CACHE_SIZE - 1);
 407                         struct page *page = pages[nr_pages_ret - 1];
 408                         char *kaddr;
 409
 410                         /* zero the tail end of the last page, we might be
 411                          * sending it down to disk
 412                          */
 413                         if (offset) {
 414                                 kaddr = kmap_atomic(page, KM_USER0);
 415                                 memset(kaddr + offset, 0,
 416                                        PAGE_CACHE_SIZE - offset);
 417                                 kunmap_atomic(kaddr, KM_USER0);
 418                         }
 419                         will_compress = 1;
 420                 }
 421         }
 422         if (start == 0) {
 423                 trans = btrfs_join_transaction(root);
 424                 BUG_ON(IS_ERR(trans));
 425                 btrfs_set_trans_block_group(trans, inode);
 426                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 427
 428                 /* lets try to make an inline extent */
 429                 if (ret || total_in < (actual_end - start)) {
 430                         /* we didn't compress the entire range, try
 431                          * to make an uncompressed inline extent.
 432                          */
 433                         ret = cow_file_range_inline(trans, root, inode,
 434                                                     start, end, 0, 0, NULL);
 435                 } else {
 436                         /* try making a compressed inline extent */
 437                         ret = cow_file_range_inline(trans, root, inode,
 438                                                     start, end,
 439                                                     total_compressed,
 440                                                     compress_type, pages);
 441                 }
 442                 if (ret == 0) {
 443                         /*
 444                          * inline extent creation worked, we don't need
 445                          * to create any more async work items.  Unlock
 446                          * and free up our temp pages.
 447                          */
 448                         extent_clear_unlock_delalloc(inode,
 449                              &BTRFS_I(inode)->io_tree,
 450                              start, end, NULL,
 451                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
 452                              EXTENT_CLEAR_DELALLOC |
 453                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
 454
 455                         btrfs_end_transaction(trans, root);
 456                         goto free_pages_out;
 457                 }
 458                 btrfs_end_transaction(trans, root);
 459         }
 460
 461         if (will_compress) {
 462                 /*
 463                  * we aren't doing an inline extent round the compressed size
 464                  * up to a block size boundary so the allocator does sane
 465                  * things
 466                  */
 467                 total_compressed = (total_compressed + blocksize - 1) &
 468                         ~(blocksize - 1);
 469
 470                 /*
 471                  * one last check to make sure the compression is really a
 472                  * win, compare the page count read with the blocks on disk
 473                  */
 474                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
 475                         ~(PAGE_CACHE_SIZE - 1);
 476                 if (total_compressed >= total_in) {
 477                         will_compress = 0;
 478                 } else {
 479                         num_bytes = total_in;
 480                 }
 481         }
 482         if (!will_compress && pages) {
 483                 /*
 484                  * the compression code ran but failed to make things smaller,
 485                  * free any pages it allocated and our page pointer array
 486                  */
 487                 for (i = 0; i < nr_pages_ret; i++) {
 488                         WARN_ON(pages[i]->mapping);
 489                         page_cache_release(pages[i]);
 490                 }
 491                 kfree(pages);
 492                 pages = NULL;
 493                 total_compressed = 0;
 494                 nr_pages_ret = 0;
 495
 496                 /* flag the file so we don't compress in the future */
 497                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 498                     !(BTRFS_I(inode)->force_compress)) {
 499                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 500                 }
 501         }
 502         if (will_compress) {
 503                 *num_added += 1;
 504
 505                 /* the async work queues will take care of doing actual
 506                  * allocation on disk for these compressed pages,
 507                  * and will submit them to the elevator.
 508                  */
 509                 add_async_extent(async_cow, start, num_bytes,
 510                                  total_compressed, pages, nr_pages_ret,
 511                                  compress_type);
 512
 513                 if (start + num_bytes < end) {
 514                         start += num_bytes;
 515                         pages = NULL;
 516                         cond_resched();
 517                         goto again;
 518                 }
 519         } else {
 520 cleanup_and_bail_uncompressed:
 521                 /*
 522                  * No compression, but we still need to write the pages in
 523                  * the file we've been given so far.  redirty the locked
 524                  * page if it corresponds to our extent and set things up
 525                  * for the async work queue to run cow_file_range to do
 526                  * the normal delalloc dance
 527                  */
 528                 if (page_offset(locked_page) >= start &&
 529                     page_offset(locked_page) <= end) {
 530                         __set_page_dirty_nobuffers(locked_page);
 531                         /* unlocked later on in the async handlers */
 532                 }
 533                 add_async_extent(async_cow, start, end - start + 1,
 534                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 535                 *num_added += 1;
 536         }
 537
 538 out:
 539         return 0;
 540
 541 free_pages_out:
 542         for (i = 0; i < nr_pages_ret; i++) {
 543                 WARN_ON(pages[i]->mapping);
 544                 page_cache_release(pages[i]);
 545         }
 546         kfree(pages);
 547
 548         goto out;
 549 }
 550
 551 /*
 552  * phase two of compressed writeback.  This is the ordered portion
 553  * of the code, which only gets called in the order the work was
 554  * queued.  We walk all the async extents created by compress_file_range
 555  * and send them down to the disk.
 556  */
 557 static noinline int submit_compressed_extents(struct inode *inode,
 558                                               struct async_cow *async_cow)
 559 {
 560         struct async_extent *async_extent;
 561         u64 alloc_hint = 0;
 562         struct btrfs_trans_handle *trans;
 563         struct btrfs_key ins;
 564         struct extent_map *em;
 565         struct btrfs_root *root = BTRFS_I(inode)->root;
 566         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 567         struct extent_io_tree *io_tree;
 568         int ret = 0;
 569
 570         if (list_empty(&async_cow->extents))
 571                 return 0;
 572
 573
 574         while (!list_empty(&async_cow->extents)) {
 575                 async_extent = list_entry(async_cow->extents.next,
 576                                           struct async_extent, list);
 577                 list_del(&async_extent->list);
 578
 579                 io_tree = &BTRFS_I(inode)->io_tree;
 580
 581 retry:
 582                 /* did the compression code fall back to uncompressed IO? */
 583                 if (!async_extent->pages) {
 584                         int page_started = 0;
 585                         unsigned long nr_written = 0;
 586
 587                         lock_extent(io_tree, async_extent->start,
 588                                          async_extent->start +
 589                                          async_extent->ram_size - 1, GFP_NOFS);
 590
 591                         /* allocate blocks */
 592                         ret = cow_file_range(inode, async_cow->locked_page,
 593                                              async_extent->start,
 594                                              async_extent->start +
 595                                              async_extent->ram_size - 1,
 596                                              &page_started, &nr_written, 0);
 597
 598                         /*
 599                          * if page_started, cow_file_range inserted an
 600                          * inline extent and took care of all the unlocking
 601                          * and IO for us.  Otherwise, we need to submit
 602                          * all those pages down to the drive.
 603                          */
 604                         if (!page_started && !ret)
 605                                 extent_write_locked_range(io_tree,
 606                                                   inode, async_extent->start,
 607                                                   async_extent->start +
 608                                                   async_extent->ram_size - 1,
 609                                                   btrfs_get_extent,
 610                                                   WB_SYNC_ALL);
 611                         kfree(async_extent);
 612                         cond_resched();
 613                         continue;
 614                 }
 615
 616                 lock_extent(io_tree, async_extent->start,
 617                             async_extent->start + async_extent->ram_size - 1,
 618                             GFP_NOFS);
 619
 620                 trans = btrfs_join_transaction(root);
 621                 BUG_ON(IS_ERR(trans));
 622                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 623                 ret = btrfs_reserve_extent(trans, root,
 624                                            async_extent->compressed_size,
 625                                            async_extent->compressed_size,
 626                                            0, alloc_hint,
 627                                            (u64)-1, &ins, 1);
 628                 btrfs_end_transaction(trans, root);
 629
 630                 if (ret) {
 631                         int i;
 632                         for (i = 0; i < async_extent->nr_pages; i++) {
 633                                 WARN_ON(async_extent->pages[i]->mapping);
 634                                 page_cache_release(async_extent->pages[i]);
 635                         }
 636                         kfree(async_extent->pages);
 637                         async_extent->nr_pages = 0;
 638                         async_extent->pages = NULL;
 639                         unlock_extent(io_tree, async_extent->start,
 640                                       async_extent->start +
 641                                       async_extent->ram_size - 1, GFP_NOFS);
 642                         goto retry;
 643                 }
 644
 645                 /*
 646                  * here we're doing allocation and writeback of the
 647                  * compressed pages
 648                  */
 649                 btrfs_drop_extent_cache(inode, async_extent->start,
 650                                         async_extent->start +
 651                                         async_extent->ram_size - 1, 0);
 652
 653                 em = alloc_extent_map(GFP_NOFS);
 654                 BUG_ON(!em);
 655                 em->start = async_extent->start;
 656                 em->len = async_extent->ram_size;
 657                 em->orig_start = em->start;
 658
 659                 em->block_start = ins.objectid;
 660                 em->block_len = ins.offset;
 661                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 662                 em->compress_type = async_extent->compress_type;
 663                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 664                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 665
 666                 while (1) {
 667                         write_lock(&em_tree->lock);
 668                         ret = add_extent_mapping(em_tree, em);
 669                         write_unlock(&em_tree->lock);
 670                         if (ret != -EEXIST) {
 671                                 free_extent_map(em);
 672                                 break;
 673                         }
 674                         btrfs_drop_extent_cache(inode, async_extent->start,
 675                                                 async_extent->start +
 676                                                 async_extent->ram_size - 1, 0);
 677                 }
 678
 679                 ret = btrfs_add_ordered_extent_compress(inode,
 680                                                 async_extent->start,
 681                                                 ins.objectid,
 682                                                 async_extent->ram_size,
 683                                                 ins.offset,
 684                                                 BTRFS_ORDERED_COMPRESSED,
 685                                                 async_extent->compress_type);
 686                 BUG_ON(ret);
 687
 688                 /*
 689                  * clear dirty, set writeback and unlock the pages.
 690                  */
 691                 extent_clear_unlock_delalloc(inode,
 692                                 &BTRFS_I(inode)->io_tree,
 693                                 async_extent->start,
 694                                 async_extent->start +
 695                                 async_extent->ram_size - 1,
 696                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 697                                 EXTENT_CLEAR_UNLOCK |
 698                                 EXTENT_CLEAR_DELALLOC |
 699                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
 700
 701                 ret = btrfs_submit_compressed_write(inode,
 702                                     async_extent->start,
 703                                     async_extent->ram_size,
 704                                     ins.objectid,
 705                                     ins.offset, async_extent->pages,
 706                                     async_extent->nr_pages);
 707
 708                 BUG_ON(ret);
 709                 alloc_hint = ins.objectid + ins.offset;
 710                 kfree(async_extent);
 711                 cond_resched();
 712         }
 713
 714         return 0;
 715 }
 716
 717 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 718                                       u64 num_bytes)
 719 {
 720         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 721         struct extent_map *em;
 722         u64 alloc_hint = 0;
 723
 724         read_lock(&em_tree->lock);
 725         em = search_extent_mapping(em_tree, start, num_bytes);
 726         if (em) {
 727                 /*
 728                  * if block start isn't an actual block number then find the
 729                  * first block in this inode and use that as a hint.  If that
 730                  * block is also bogus then just don't worry about it.
 731                  */
 732                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 733                         free_extent_map(em);
 734                         em = search_extent_mapping(em_tree, 0, 0);
 735                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 736                                 alloc_hint = em->block_start;
 737                         if (em)
 738                                 free_extent_map(em);
 739                 } else {
 740                         alloc_hint = em->block_start;
 741                         free_extent_map(em);
 742                 }
 743         }
 744         read_unlock(&em_tree->lock);
 745
 746         return alloc_hint;
 747 }
 748
 749 /*
 750  * when extent_io.c finds a delayed allocation range in the file,
 751  * the call backs end up in this code.  The basic idea is to
 752  * allocate extents on disk for the range, and create ordered data structs
 753  * in ram to track those extents.
 754  *
 755  * locked_page is the page that writepage had locked already.  We use
 756  * it to make sure we don't do extra locks or unlocks.
 757  *
 758  * *page_started is set to one if we unlock locked_page and do everything
 759  * required to start IO on it.  It may be clean and already done with
 760  * IO when we return.
 761  */
 762 static noinline int cow_file_range(struct inode *inode,
 763                                    struct page *locked_page,
 764                                    u64 start, u64 end, int *page_started,
 765                                    unsigned long *nr_written,
 766                                    int unlock)
 767 {
 768         struct btrfs_root *root = BTRFS_I(inode)->root;
 769         struct btrfs_trans_handle *trans;
 770         u64 alloc_hint = 0;
 771         u64 num_bytes;
 772         unsigned long ram_size;
 773         u64 disk_num_bytes;
 774         u64 cur_alloc_size;
 775         u64 blocksize = root->sectorsize;
 776         struct btrfs_key ins;
 777         struct extent_map *em;
 778         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 779         int ret = 0;
 780
 781         BUG_ON(root == root->fs_info->tree_root);
 782         trans = btrfs_join_transaction(root);
 783         BUG_ON(IS_ERR(trans));
 784         btrfs_set_trans_block_group(trans, inode);
 785         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 786
 787         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 788         num_bytes = max(blocksize,  num_bytes);
 789         disk_num_bytes = num_bytes;
 790         ret = 0;
 791
 792         if (start == 0) {
 793                 /* lets try to make an inline extent */
 794                 ret = cow_file_range_inline(trans, root, inode,
 795                                             start, end, 0, 0, NULL);
 796                 if (ret == 0) {
 797                         extent_clear_unlock_delalloc(inode,
 798                                      &BTRFS_I(inode)->io_tree,
 799                                      start, end, NULL,
 800                                      EXTENT_CLEAR_UNLOCK_PAGE |
 801                                      EXTENT_CLEAR_UNLOCK |
 802                                      EXTENT_CLEAR_DELALLOC |
 803                                      EXTENT_CLEAR_DIRTY |
 804                                      EXTENT_SET_WRITEBACK |
 805                                      EXTENT_END_WRITEBACK);
 806
 807                         *nr_written = *nr_written +
 808                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 809                         *page_started = 1;
 810                         ret = 0;
 811                         goto out;
 812                 }
 813         }
 814
 815         BUG_ON(disk_num_bytes >
 816                btrfs_super_total_bytes(&root->fs_info->super_copy));
 817
 818         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 819         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 820
 821         while (disk_num_bytes > 0) {
 822                 unsigned long op;
 823
 824                 cur_alloc_size = disk_num_bytes;
 825                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 826                                            root->sectorsize, 0, alloc_hint,
 827                                            (u64)-1, &ins, 1);
 828                 BUG_ON(ret);
 829
 830                 em = alloc_extent_map(GFP_NOFS);
 831                 BUG_ON(!em);
 832                 em->start = start;
 833                 em->orig_start = em->start;
 834                 ram_size = ins.offset;
 835                 em->len = ins.offset;
 836
 837                 em->block_start = ins.objectid;
 838                 em->block_len = ins.offset;
 839                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 840                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 841
 842                 while (1) {
 843                         write_lock(&em_tree->lock);
 844                         ret = add_extent_mapping(em_tree, em);
 845                         write_unlock(&em_tree->lock);
 846                         if (ret != -EEXIST) {
 847                                 free_extent_map(em);
 848                                 break;
 849                         }
 850                         btrfs_drop_extent_cache(inode, start,
 851                                                 start + ram_size - 1, 0);
 852                 }
 853
 854                 cur_alloc_size = ins.offset;
 855                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 856                                                ram_size, cur_alloc_size, 0);
 857                 BUG_ON(ret);
 858
 859                 if (root->root_key.objectid ==
 860                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 861                         ret = btrfs_reloc_clone_csums(inode, start,
 862                                                       cur_alloc_size);
 863                         BUG_ON(ret);
 864                 }
 865
 866                 if (disk_num_bytes < cur_alloc_size)
 867                         break;
 868
 869                 /* we're not doing compressed IO, don't unlock the first
 870                  * page (which the caller expects to stay locked), don't
 871                  * clear any dirty bits and don't set any writeback bits
 872                  *
 873                  * Do set the Private2 bit so we know this page was properly
 874                  * setup for writepage
 875                  */
 876                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
 877                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
 878                         EXTENT_SET_PRIVATE2;
 879
 880                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 881                                              start, start + ram_size - 1,
 882                                              locked_page, op);
 883                 disk_num_bytes -= cur_alloc_size;
 884                 num_bytes -= cur_alloc_size;
 885                 alloc_hint = ins.objectid + ins.offset;
 886                 start += cur_alloc_size;
 887         }
 888 out:
 889         ret = 0;
 890         btrfs_end_transaction(trans, root);
 891
 892         return ret;
 893 }
 894
 895 /*
 896  * work queue call back to started compression on a file and pages
 897  */
 898 static noinline void async_cow_start(struct btrfs_work *work)
 899 {
 900         struct async_cow *async_cow;
 901         int num_added = 0;
 902         async_cow = container_of(work, struct async_cow, work);
 903
 904         compress_file_range(async_cow->inode, async_cow->locked_page,
 905                             async_cow->start, async_cow->end, async_cow,
 906                             &num_added);
 907         if (num_added == 0)
 908                 async_cow->inode = NULL;
 909 }
 910
 911 /*
 912  * work queue call back to submit previously compressed pages
 913  */
 914 static noinline void async_cow_submit(struct btrfs_work *work)
 915 {
 916         struct async_cow *async_cow;
 917         struct btrfs_root *root;
 918         unsigned long nr_pages;
 919
 920         async_cow = container_of(work, struct async_cow, work);
 921
 922         root = async_cow->root;
 923         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
 924                 PAGE_CACHE_SHIFT;
 925
 926         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
 927
 928         if (atomic_read(&root->fs_info->async_delalloc_pages) <
 929             5 * 1042 * 1024 &&
 930             waitqueue_active(&root->fs_info->async_submit_wait))
 931                 wake_up(&root->fs_info->async_submit_wait);
 932
 933         if (async_cow->inode)
 934                 submit_compressed_extents(async_cow->inode, async_cow);
 935 }
 936
 937 static noinline void async_cow_free(struct btrfs_work *work)
 938 {
 939         struct async_cow *async_cow;
 940         async_cow = container_of(work, struct async_cow, work);
 941         kfree(async_cow);
 942 }
 943
 944 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
 945                                 u64 start, u64 end, int *page_started,
 946                                 unsigned long *nr_written)
 947 {
 948         struct async_cow *async_cow;
 949         struct btrfs_root *root = BTRFS_I(inode)->root;
 950         unsigned long nr_pages;
 951         u64 cur_end;
 952         int limit = 10 * 1024 * 1042;
 953
 954         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
 955                          1, 0, NULL, GFP_NOFS);
 956         while (start < end) {
 957                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
 958                 BUG_ON(!async_cow);
 959                 async_cow->inode = inode;
 960                 async_cow->root = root;
 961                 async_cow->locked_page = locked_page;
 962                 async_cow->start = start;
 963
 964                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
 965                         cur_end = end;
 966                 else
 967                         cur_end = min(end, start + 512 * 1024 - 1);
 968
 969                 async_cow->end = cur_end;
 970                 INIT_LIST_HEAD(&async_cow->extents);
 971
 972                 async_cow->work.func = async_cow_start;
 973                 async_cow->work.ordered_func = async_cow_submit;
 974                 async_cow->work.ordered_free = async_cow_free;
 975                 async_cow->work.flags = 0;
 976
 977                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
 978                         PAGE_CACHE_SHIFT;
 979                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
 980
 981                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
 982                                    &async_cow->work);
 983
 984                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
 985                         wait_event(root->fs_info->async_submit_wait,
 986                            (atomic_read(&root->fs_info->async_delalloc_pages) <
 987                             limit));
 988                 }
 989
 990                 while (atomic_read(&root->fs_info->async_submit_draining) &&
 991                       atomic_read(&root->fs_info->async_delalloc_pages)) {
 992                         wait_event(root->fs_info->async_submit_wait,
 993                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
 994                            0));
 995                 }
 996
 997                 *nr_written += nr_pages;
 998                 start = cur_end + 1;
 999         }
1000         *page_started = 1;
1001         return 0;
1002 }
1003
1004 static noinline int csum_exist_in_range(struct btrfs_root *root,
1005                                         u64 bytenr, u64 num_bytes)
1006 {
1007         int ret;
1008         struct btrfs_ordered_sum *sums;
1009         LIST_HEAD(list);
1010
1011         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1012                                        bytenr + num_bytes - 1, &list);
1013         if (ret == 0 && list_empty(&list))
1014                 return 0;
1015
1016         while (!list_empty(&list)) {
1017                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1018                 list_del(&sums->list);
1019                 kfree(sums);
1020         }
1021         return 1;
1022 }
1023
1024 /*
1025  * when nowcow writeback call back.  This checks for snapshots or COW copies
1026  * of the extents that exist in the file, and COWs the file as required.
1027  *
1028  * If no cow copies or snapshots exist, we write directly to the existing
1029  * blocks on disk
1030  */
1031 static noinline int run_delalloc_nocow(struct inode *inode,
1032                                        struct page *locked_page,
1033                               u64 start, u64 end, int *page_started, int force,
1034                               unsigned long *nr_written)
1035 {
1036         struct btrfs_root *root = BTRFS_I(inode)->root;
1037         struct btrfs_trans_handle *trans;
1038         struct extent_buffer *leaf;
1039         struct btrfs_path *path;
1040         struct btrfs_file_extent_item *fi;
1041         struct btrfs_key found_key;
1042         u64 cow_start;
1043         u64 cur_offset;
1044         u64 extent_end;
1045         u64 extent_offset;
1046         u64 disk_bytenr;
1047         u64 num_bytes;
1048         int extent_type;
1049         int ret;
1050         int type;
1051         int nocow;
1052         int check_prev = 1;
1053         bool nolock = false;
1054
1055         path = btrfs_alloc_path();
1056         BUG_ON(!path);
1057         if (root == root->fs_info->tree_root) {
1058                 nolock = true;
1059                 trans = btrfs_join_transaction_nolock(root);
1060         } else {
1061                 trans = btrfs_join_transaction(root);
1062         }
1063         BUG_ON(IS_ERR(trans));
1064         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1065
1066         cow_start = (u64)-1;
1067         cur_offset = start;
1068         while (1) {
1069                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
1070                                                cur_offset, 0);
1071                 BUG_ON(ret < 0);
1072                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1073                         leaf = path->nodes[0];
1074                         btrfs_item_key_to_cpu(leaf, &found_key,
1075                                               path->slots[0] - 1);
1076                         if (found_key.objectid == inode->i_ino &&
1077                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1078                                 path->slots[0]--;
1079                 }
1080                 check_prev = 0;
1081 next_slot:
1082                 leaf = path->nodes[0];
1083                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1084                         ret = btrfs_next_leaf(root, path);
1085                         if (ret < 0)
1086                                 BUG_ON(1);
1087                         if (ret > 0)
1088                                 break;
1089                         leaf = path->nodes[0];
1090                 }
1091
1092                 nocow = 0;
1093                 disk_bytenr = 0;
1094                 num_bytes = 0;
1095                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1096
1097                 if (found_key.objectid > inode->i_ino ||
1098                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1099                     found_key.offset > end)
1100                         break;
1101
1102                 if (found_key.offset > cur_offset) {
1103                         extent_end = found_key.offset;
1104                         extent_type = 0;
1105                         goto out_check;
1106                 }
1107
1108                 fi = btrfs_item_ptr(leaf, path->slots[0],
1109                                     struct btrfs_file_extent_item);
1110                 extent_type = btrfs_file_extent_type(leaf, fi);
1111
1112                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1113                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1114                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1115                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1116                         extent_end = found_key.offset +
1117                                 btrfs_file_extent_num_bytes(leaf, fi);
1118                         if (extent_end <= start) {
1119                                 path->slots[0]++;
1120                                 goto next_slot;
1121                         }
1122                         if (disk_bytenr == 0)
1123                                 goto out_check;
1124                         if (btrfs_file_extent_compression(leaf, fi) ||
1125                             btrfs_file_extent_encryption(leaf, fi) ||
1126                             btrfs_file_extent_other_encoding(leaf, fi))
1127                                 goto out_check;
1128                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1129                                 goto out_check;
1130                         if (btrfs_extent_readonly(root, disk_bytenr))
1131                                 goto out_check;
1132                         if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1133                                                   found_key.offset -
1134                                                   extent_offset, disk_bytenr))
1135                                 goto out_check;
1136                         disk_bytenr += extent_offset;
1137                         disk_bytenr += cur_offset - found_key.offset;
1138                         num_bytes = min(end + 1, extent_end) - cur_offset;
1139                         /*
1140                          * force cow if csum exists in the range.
1141                          * this ensure that csum for a given extent are
1142                          * either valid or do not exist.
1143                          */
1144                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1145                                 goto out_check;
1146                         nocow = 1;
1147                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1148                         extent_end = found_key.offset +
1149                                 btrfs_file_extent_inline_len(leaf, fi);
1150                         extent_end = ALIGN(extent_end, root->sectorsize);
1151                 } else {
1152                         BUG_ON(1);
1153                 }
1154 out_check:
1155                 if (extent_end <= start) {
1156                         path->slots[0]++;
1157                         goto next_slot;
1158                 }
1159                 if (!nocow) {
1160                         if (cow_start == (u64)-1)
1161                                 cow_start = cur_offset;
1162                         cur_offset = extent_end;
1163                         if (cur_offset > end)
1164                                 break;
1165                         path->slots[0]++;
1166                         goto next_slot;
1167                 }
1168
1169                 btrfs_release_path(root, path);
1170                 if (cow_start != (u64)-1) {
1171                         ret = cow_file_range(inode, locked_page, cow_start,
1172                                         found_key.offset - 1, page_started,
1173                                         nr_written, 1);
1174                         BUG_ON(ret);
1175                         cow_start = (u64)-1;
1176                 }
1177
1178                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1179                         struct extent_map *em;
1180                         struct extent_map_tree *em_tree;
1181                         em_tree = &BTRFS_I(inode)->extent_tree;
1182                         em = alloc_extent_map(GFP_NOFS);
1183                         BUG_ON(!em);
1184                         em->start = cur_offset;
1185                         em->orig_start = em->start;
1186                         em->len = num_bytes;
1187                         em->block_len = num_bytes;
1188                         em->block_start = disk_bytenr;
1189                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1190                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1191                         while (1) {
1192                                 write_lock(&em_tree->lock);
1193                                 ret = add_extent_mapping(em_tree, em);
1194                                 write_unlock(&em_tree->lock);
1195                                 if (ret != -EEXIST) {
1196                                         free_extent_map(em);
1197                                         break;
1198                                 }
1199                                 btrfs_drop_extent_cache(inode, em->start,
1200                                                 em->start + em->len - 1, 0);
1201                         }
1202                         type = BTRFS_ORDERED_PREALLOC;
1203                 } else {
1204                         type = BTRFS_ORDERED_NOCOW;
1205                 }
1206
1207                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1208                                                num_bytes, num_bytes, type);
1209                 BUG_ON(ret);
1210
1211                 if (root->root_key.objectid ==
1212                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1213                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1214                                                       num_bytes);
1215                         BUG_ON(ret);
1216                 }
1217
1218                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1219                                 cur_offset, cur_offset + num_bytes - 1,
1220                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1221                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1222                                 EXTENT_SET_PRIVATE2);
1223                 cur_offset = extent_end;
1224                 if (cur_offset > end)
1225                         break;
1226         }
1227         btrfs_release_path(root, path);
1228
1229         if (cur_offset <= end && cow_start == (u64)-1)
1230                 cow_start = cur_offset;
1231         if (cow_start != (u64)-1) {
1232                 ret = cow_file_range(inode, locked_page, cow_start, end,
1233                                      page_started, nr_written, 1);
1234                 BUG_ON(ret);
1235         }
1236
1237         if (nolock) {
1238                 ret = btrfs_end_transaction_nolock(trans, root);
1239                 BUG_ON(ret);
1240         } else {
1241                 ret = btrfs_end_transaction(trans, root);
1242                 BUG_ON(ret);
1243         }
1244         btrfs_free_path(path);
1245         return 0;
1246 }
1247
1248 /*
1249  * extent_io.c call back to do delayed allocation processing
1250  */
1251 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1252                               u64 start, u64 end, int *page_started,
1253                               unsigned long *nr_written)
1254 {
1255         int ret;
1256         struct btrfs_root *root = BTRFS_I(inode)->root;
1257
1258         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1259                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1260                                          page_started, 1, nr_written);
1261         else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1262                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1263                                          page_started, 0, nr_written);
1264         else if (!btrfs_test_opt(root, COMPRESS) &&
1265                  !(BTRFS_I(inode)->force_compress) &&
1266                  !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1267                 ret = cow_file_range(inode, locked_page, start, end,
1268                                       page_started, nr_written, 1);
1269         else
1270                 ret = cow_file_range_async(inode, locked_page, start, end,
1271                                            page_started, nr_written);
1272         return ret;
1273 }
1274
1275 static int btrfs_split_extent_hook(struct inode *inode,
1276                                    struct extent_state *orig, u64 split)
1277 {
1278         /* not delalloc, ignore it */
1279         if (!(orig->state & EXTENT_DELALLOC))
1280                 return 0;
1281
1282         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1283         return 0;
1284 }
1285
1286 /*
1287  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1288  * extents so we can keep track of new extents that are just merged onto old
1289  * extents, such as when we are doing sequential writes, so we can properly
1290  * account for the metadata space we'll need.
1291  */
1292 static int btrfs_merge_extent_hook(struct inode *inode,
1293                                    struct extent_state *new,
1294                                    struct extent_state *other)
1295 {
1296         /* not delalloc, ignore it */
1297         if (!(other->state & EXTENT_DELALLOC))
1298                 return 0;
1299
1300         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1301         return 0;
1302 }
1303
1304 /*
1305  * extent_io.c set_bit_hook, used to track delayed allocation
1306  * bytes in this file, and to maintain the list of inodes that
1307  * have pending delalloc work to be done.
1308  */
1309 static int btrfs_set_bit_hook(struct inode *inode,
1310                               struct extent_state *state, int *bits)
1311 {
1312
1313         /*
1314          * set_bit and clear bit hooks normally require _irqsave/restore
1315          * but in this case, we are only testeing for the DELALLOC
1316          * bit, which is only set or cleared with irqs on
1317          */
1318         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1319                 struct btrfs_root *root = BTRFS_I(inode)->root;
1320                 u64 len = state->end + 1 - state->start;
1321                 int do_list = (root->root_key.objectid !=
1322                                BTRFS_ROOT_TREE_OBJECTID);
1323
1324                 if (*bits & EXTENT_FIRST_DELALLOC)
1325                         *bits &= ~EXTENT_FIRST_DELALLOC;
1326                 else
1327                         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1328
1329                 spin_lock(&root->fs_info->delalloc_lock);
1330                 BTRFS_I(inode)->delalloc_bytes += len;
1331                 root->fs_info->delalloc_bytes += len;
1332                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1333                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1334                                       &root->fs_info->delalloc_inodes);
1335                 }
1336                 spin_unlock(&root->fs_info->delalloc_lock);
1337         }
1338         return 0;
1339 }
1340
1341 /*
1342  * extent_io.c clear_bit_hook, see set_bit_hook for why
1343  */
1344 static int btrfs_clear_bit_hook(struct inode *inode,
1345                                 struct extent_state *state, int *bits)
1346 {
1347         /*
1348          * set_bit and clear bit hooks normally require _irqsave/restore
1349          * but in this case, we are only testeing for the DELALLOC
1350          * bit, which is only set or cleared with irqs on
1351          */
1352         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1353                 struct btrfs_root *root = BTRFS_I(inode)->root;
1354                 u64 len = state->end + 1 - state->start;
1355                 int do_list = (root->root_key.objectid !=
1356                                BTRFS_ROOT_TREE_OBJECTID);
1357
1358                 if (*bits & EXTENT_FIRST_DELALLOC)
1359                         *bits &= ~EXTENT_FIRST_DELALLOC;
1360                 else if (!(*bits & EXTENT_DO_ACCOUNTING))
1361                         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1362
1363                 if (*bits & EXTENT_DO_ACCOUNTING)
1364                         btrfs_delalloc_release_metadata(inode, len);
1365
1366                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1367                     && do_list)
1368                         btrfs_free_reserved_data_space(inode, len);
1369
1370                 spin_lock(&root->fs_info->delalloc_lock);
1371                 root->fs_info->delalloc_bytes -= len;
1372                 BTRFS_I(inode)->delalloc_bytes -= len;
1373
1374                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1375                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1376                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1377                 }
1378                 spin_unlock(&root->fs_info->delalloc_lock);
1379         }
1380         return 0;
1381 }
1382
1383 /*
1384  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1385  * we don't create bios that span stripes or chunks
1386  */
1387 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1388                          size_t size, struct bio *bio,
1389                          unsigned long bio_flags)
1390 {
1391         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1392         struct btrfs_mapping_tree *map_tree;
1393         u64 logical = (u64)bio->bi_sector << 9;
1394         u64 length = 0;
1395         u64 map_length;
1396         int ret;
1397
1398         if (bio_flags & EXTENT_BIO_COMPRESSED)
1399                 return 0;
1400
1401         length = bio->bi_size;
1402         map_tree = &root->fs_info->mapping_tree;
1403         map_length = length;
1404         ret = btrfs_map_block(map_tree, READ, logical,
1405                               &map_length, NULL, 0);
1406
1407         if (map_length < length + size)
1408                 return 1;
1409         return ret;
1410 }
1411
1412 /*
1413  * in order to insert checksums into the metadata in large chunks,
1414  * we wait until bio submission time.   All the pages in the bio are
1415  * checksummed and sums are attached onto the ordered extent record.
1416  *
1417  * At IO completion time the cums attached on the ordered extent record
1418  * are inserted into the btree
1419  */
1420 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1421                                     struct bio *bio, int mirror_num,
1422                                     unsigned long bio_flags,
1423                                     u64 bio_offset)
1424 {
1425         struct btrfs_root *root = BTRFS_I(inode)->root;
1426         int ret = 0;
1427
1428         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1429         BUG_ON(ret);
1430         return 0;
1431 }
1432
1433 /*
1434  * in order to insert checksums into the metadata in large chunks,
1435  * we wait until bio submission time.   All the pages in the bio are
1436  * checksummed and sums are attached onto the ordered extent record.
1437  *
1438  * At IO completion time the cums attached on the ordered extent record
1439  * are inserted into the btree
1440  */
1441 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1442                           int mirror_num, unsigned long bio_flags,
1443                           u64 bio_offset)
1444 {
1445         struct btrfs_root *root = BTRFS_I(inode)->root;
1446         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1447 }
1448
1449 /*
1450  * extent_io.c submission hook. This does the right thing for csum calculation
1451  * on write, or reading the csums from the tree before a read
1452  */
1453 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1454                           int mirror_num, unsigned long bio_flags,
1455                           u64 bio_offset)
1456 {
1457         struct btrfs_root *root = BTRFS_I(inode)->root;
1458         int ret = 0;
1459         int skip_sum;
1460
1461         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1462
1463         if (root == root->fs_info->tree_root)
1464                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1465         else
1466                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1467         BUG_ON(ret);
1468
1469         if (!(rw & REQ_WRITE)) {
1470                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1471                         return btrfs_submit_compressed_read(inode, bio,
1472                                                     mirror_num, bio_flags);
1473                 } else if (!skip_sum) {
1474                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1475                         if (ret)
1476                                 return ret;
1477                 }
1478                 goto mapit;
1479         } else if (!skip_sum) {
1480                 /* csum items have already been cloned */
1481                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1482                         goto mapit;
1483                 /* we're doing a write, do the async checksumming */
1484                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1485                                    inode, rw, bio, mirror_num,
1486                                    bio_flags, bio_offset,
1487                                    __btrfs_submit_bio_start,
1488                                    __btrfs_submit_bio_done);
1489         }
1490
1491 mapit:
1492         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1493 }
1494
1495 /*
1496  * given a list of ordered sums record them in the inode.  This happens
1497  * at IO completion time based on sums calculated at bio submission time.
1498  */
1499 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1500                              struct inode *inode, u64 file_offset,
1501                              struct list_head *list)
1502 {
1503         struct btrfs_ordered_sum *sum;
1504
1505         btrfs_set_trans_block_group(trans, inode);
1506
1507         list_for_each_entry(sum, list, list) {
1508                 btrfs_csum_file_blocks(trans,
1509                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1510         }
1511         return 0;
1512 }
1513
1514 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1515                               struct extent_state **cached_state)
1516 {
1517         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1518                 WARN_ON(1);
1519         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1520                                    cached_state, GFP_NOFS);
1521 }
1522
1523 /* see btrfs_writepage_start_hook for details on why this is required */
1524 struct btrfs_writepage_fixup {
1525         struct page *page;
1526         struct btrfs_work work;
1527 };
1528
1529 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1530 {
1531         struct btrfs_writepage_fixup *fixup;
1532         struct btrfs_ordered_extent *ordered;
1533         struct extent_state *cached_state = NULL;
1534         struct page *page;
1535         struct inode *inode;
1536         u64 page_start;
1537         u64 page_end;
1538
1539         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1540         page = fixup->page;
1541 again:
1542         lock_page(page);
1543         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1544                 ClearPageChecked(page);
1545                 goto out_page;
1546         }
1547
1548         inode = page->mapping->host;
1549         page_start = page_offset(page);
1550         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1551
1552         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1553                          &cached_state, GFP_NOFS);
1554
1555         /* already ordered? We're done */
1556         if (PagePrivate2(page))
1557                 goto out;
1558
1559         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1560         if (ordered) {
1561                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1562                                      page_end, &cached_state, GFP_NOFS);
1563                 unlock_page(page);
1564                 btrfs_start_ordered_extent(inode, ordered, 1);
1565                 goto again;
1566         }
1567
1568         BUG();
1569         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1570         ClearPageChecked(page);
1571 out:
1572         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1573                              &cached_state, GFP_NOFS);
1574 out_page:
1575         unlock_page(page);
1576         page_cache_release(page);
1577         kfree(fixup);
1578 }
1579
1580 /*
1581  * There are a few paths in the higher layers of the kernel that directly
1582  * set the page dirty bit without asking the filesystem if it is a
1583  * good idea.  This causes problems because we want to make sure COW
1584  * properly happens and the data=ordered rules are followed.
1585  *
1586  * In our case any range that doesn't have the ORDERED bit set
1587  * hasn't been properly setup for IO.  We kick off an async process
1588  * to fix it up.  The async helper will wait for ordered extents, set
1589  * the delalloc bit and make it safe to write the page.
1590  */
1591 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1592 {
1593         struct inode *inode = page->mapping->host;
1594         struct btrfs_writepage_fixup *fixup;
1595         struct btrfs_root *root = BTRFS_I(inode)->root;
1596
1597         /* this page is properly in the ordered list */
1598         if (TestClearPagePrivate2(page))
1599                 return 0;
1600
1601         if (PageChecked(page))
1602                 return -EAGAIN;
1603
1604         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1605         if (!fixup)
1606                 return -EAGAIN;
1607
1608         SetPageChecked(page);
1609         page_cache_get(page);
1610         fixup->work.func = btrfs_writepage_fixup_worker;
1611         fixup->page = page;
1612         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1613         return -EAGAIN;
1614 }
1615
1616 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1617                                        struct inode *inode, u64 file_pos,
1618                                        u64 disk_bytenr, u64 disk_num_bytes,
1619                                        u64 num_bytes, u64 ram_bytes,
1620                                        u8 compression, u8 encryption,
1621                                        u16 other_encoding, int extent_type)
1622 {
1623         struct btrfs_root *root = BTRFS_I(inode)->root;
1624         struct btrfs_file_extent_item *fi;
1625         struct btrfs_path *path;
1626         struct extent_buffer *leaf;
1627         struct btrfs_key ins;
1628         u64 hint;
1629         int ret;
1630
1631         path = btrfs_alloc_path();
1632         BUG_ON(!path);
1633
1634         path->leave_spinning = 1;
1635
1636         /*
1637          * we may be replacing one extent in the tree with another.
1638          * The new extent is pinned in the extent map, and we don't want
1639          * to drop it from the cache until it is completely in the btree.
1640          *
1641          * So, tell btrfs_drop_extents to leave this extent in the cache.
1642          * the caller is expected to unpin it and allow it to be merged
1643          * with the others.
1644          */
1645         ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1646                                  &hint, 0);
1647         BUG_ON(ret);
1648
1649         ins.objectid = inode->i_ino;
1650         ins.offset = file_pos;
1651         ins.type = BTRFS_EXTENT_DATA_KEY;
1652         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1653         BUG_ON(ret);
1654         leaf = path->nodes[0];
1655         fi = btrfs_item_ptr(leaf, path->slots[0],
1656                             struct btrfs_file_extent_item);
1657         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1658         btrfs_set_file_extent_type(leaf, fi, extent_type);
1659         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1660         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1661         btrfs_set_file_extent_offset(leaf, fi, 0);
1662         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1663         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1664         btrfs_set_file_extent_compression(leaf, fi, compression);
1665         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1666         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1667
1668         btrfs_unlock_up_safe(path, 1);
1669         btrfs_set_lock_blocking(leaf);
1670
1671         btrfs_mark_buffer_dirty(leaf);
1672
1673         inode_add_bytes(inode, num_bytes);
1674
1675         ins.objectid = disk_bytenr;
1676         ins.offset = disk_num_bytes;
1677         ins.type = BTRFS_EXTENT_ITEM_KEY;
1678         ret = btrfs_alloc_reserved_file_extent(trans, root,
1679                                         root->root_key.objectid,
1680                                         inode->i_ino, file_pos, &ins);
1681         BUG_ON(ret);
1682         btrfs_free_path(path);
1683
1684         return 0;
1685 }
1686
1687 /*
1688  * helper function for btrfs_finish_ordered_io, this
1689  * just reads in some of the csum leaves to prime them into ram
1690  * before we start the transaction.  It limits the amount of btree
1691  * reads required while inside the transaction.
1692  */
1693 /* as ordered data IO finishes, this gets called so we can finish
1694  * an ordered extent if the range of bytes in the file it covers are
1695  * fully written.
1696  */
1697 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1698 {
1699         struct btrfs_root *root = BTRFS_I(inode)->root;
1700         struct btrfs_trans_handle *trans = NULL;
1701         struct btrfs_ordered_extent *ordered_extent = NULL;
1702         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1703         struct extent_state *cached_state = NULL;
1704         int compress_type = 0;
1705         int ret;
1706         bool nolock = false;
1707
1708         ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1709                                              end - start + 1);
1710         if (!ret)
1711                 return 0;
1712         BUG_ON(!ordered_extent);
1713
1714         nolock = (root == root->fs_info->tree_root);
1715
1716         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1717                 BUG_ON(!list_empty(&ordered_extent->list));
1718                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1719                 if (!ret) {
1720                         if (nolock)
1721                                 trans = btrfs_join_transaction_nolock(root);
1722                         else
1723                                 trans = btrfs_join_transaction(root);
1724                         BUG_ON(IS_ERR(trans));
1725                         btrfs_set_trans_block_group(trans, inode);
1726                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1727                         ret = btrfs_update_inode(trans, root, inode);
1728                         BUG_ON(ret);
1729                 }
1730                 goto out;
1731         }
1732
1733         lock_extent_bits(io_tree, ordered_extent->file_offset,
1734                          ordered_extent->file_offset + ordered_extent->len - 1,
1735                          0, &cached_state, GFP_NOFS);
1736
1737         if (nolock)
1738                 trans = btrfs_join_transaction_nolock(root);
1739         else
1740                 trans = btrfs_join_transaction(root);
1741         BUG_ON(IS_ERR(trans));
1742         btrfs_set_trans_block_group(trans, inode);
1743         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1744
1745         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1746                 compress_type = ordered_extent->compress_type;
1747         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1748                 BUG_ON(compress_type);
1749                 ret = btrfs_mark_extent_written(trans, inode,
1750                                                 ordered_extent->file_offset,
1751                                                 ordered_extent->file_offset +
1752                                                 ordered_extent->len);
1753                 BUG_ON(ret);
1754         } else {
1755                 BUG_ON(root == root->fs_info->tree_root);
1756                 ret = insert_reserved_file_extent(trans, inode,
1757                                                 ordered_extent->file_offset,
1758                                                 ordered_extent->start,
1759                                                 ordered_extent->disk_len,
1760                                                 ordered_extent->len,
1761                                                 ordered_extent->len,
1762                                                 compress_type, 0, 0,
1763                                                 BTRFS_FILE_EXTENT_REG);
1764                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1765                                    ordered_extent->file_offset,
1766                                    ordered_extent->len);
1767                 BUG_ON(ret);
1768         }
1769         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1770                              ordered_extent->file_offset +
1771                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1772
1773         add_pending_csums(trans, inode, ordered_extent->file_offset,
1774                           &ordered_extent->list);
1775
1776         ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1777         if (!ret) {
1778                 ret = btrfs_update_inode(trans, root, inode);
1779                 BUG_ON(ret);
1780         }
1781         ret = 0;
1782 out:
1783         if (nolock) {
1784                 if (trans)
1785                         btrfs_end_transaction_nolock(trans, root);
1786         } else {
1787                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1788                 if (trans)
1789                         btrfs_end_transaction(trans, root);
1790         }
1791
1792         /* once for us */
1793         btrfs_put_ordered_extent(ordered_extent);
1794         /* once for the tree */
1795         btrfs_put_ordered_extent(ordered_extent);
1796
1797         return 0;
1798 }
1799
1800 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1801                                 struct extent_state *state, int uptodate)
1802 {
1803         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1804
1805         ClearPagePrivate2(page);
1806         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1807 }
1808
1809 /*
1810  * When IO fails, either with EIO or csum verification fails, we
1811  * try other mirrors that might have a good copy of the data.  This
1812  * io_failure_record is used to record state as we go through all the
1813  * mirrors.  If another mirror has good data, the page is set up to date
1814  * and things continue.  If a good mirror can't be found, the original
1815  * bio end_io callback is called to indicate things have failed.
1816  */
1817 struct io_failure_record {
1818         struct page *page;
1819         u64 start;
1820         u64 len;
1821         u64 logical;
1822         unsigned long bio_flags;
1823         int last_mirror;
1824 };
1825
1826 static int btrfs_io_failed_hook(struct bio *failed_bio,
1827                          struct page *page, u64 start, u64 end,
1828                          struct extent_state *state)
1829 {
1830         struct io_failure_record *failrec = NULL;
1831         u64 private;
1832         struct extent_map *em;
1833         struct inode *inode = page->mapping->host;
1834         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1835         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1836         struct bio *bio;
1837         int num_copies;
1838         int ret;
1839         int rw;
1840         u64 logical;
1841
1842         ret = get_state_private(failure_tree, start, &private);
1843         if (ret) {
1844                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1845                 if (!failrec)
1846                         return -ENOMEM;
1847                 failrec->start = start;
1848                 failrec->len = end - start + 1;
1849                 failrec->last_mirror = 0;
1850                 failrec->bio_flags = 0;
1851
1852                 read_lock(&em_tree->lock);
1853                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1854                 if (em->start > start || em->start + em->len < start) {
1855                         free_extent_map(em);
1856                         em = NULL;
1857                 }
1858                 read_unlock(&em_tree->lock);
1859
1860                 if (!em || IS_ERR(em)) {
1861                         kfree(failrec);
1862                         return -EIO;
1863                 }
1864                 logical = start - em->start;
1865                 logical = em->block_start + logical;
1866                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1867                         logical = em->block_start;
1868                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1869                         extent_set_compress_type(&failrec->bio_flags,
1870                                                  em->compress_type);
1871                 }
1872                 failrec->logical = logical;
1873                 free_extent_map(em);
1874                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1875                                 EXTENT_DIRTY, GFP_NOFS);
1876                 set_state_private(failure_tree, start,
1877                                  (u64)(unsigned long)failrec);
1878         } else {
1879                 failrec = (struct io_failure_record *)(unsigned long)private;
1880         }
1881         num_copies = btrfs_num_copies(
1882                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1883                               failrec->logical, failrec->len);
1884         failrec->last_mirror++;
1885         if (!state) {
1886                 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1887                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1888                                                     failrec->start,
1889                                                     EXTENT_LOCKED);
1890                 if (state && state->start != failrec->start)
1891                         state = NULL;
1892                 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1893         }
1894         if (!state || failrec->last_mirror > num_copies) {
1895                 set_state_private(failure_tree, failrec->start, 0);
1896                 clear_extent_bits(failure_tree, failrec->start,
1897                                   failrec->start + failrec->len - 1,
1898                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1899                 kfree(failrec);
1900                 return -EIO;
1901         }
1902         bio = bio_alloc(GFP_NOFS, 1);
1903         bio->bi_private = state;
1904         bio->bi_end_io = failed_bio->bi_end_io;
1905         bio->bi_sector = failrec->logical >> 9;
1906         bio->bi_bdev = failed_bio->bi_bdev;
1907         bio->bi_size = 0;
1908
1909         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1910         if (failed_bio->bi_rw & REQ_WRITE)
1911                 rw = WRITE;
1912         else
1913                 rw = READ;
1914
1915         ret = BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1916                                                       failrec->last_mirror,
1917                                                       failrec->bio_flags, 0);
1918         return ret;
1919 }
1920
1921 /*
1922  * each time an IO finishes, we do a fast check in the IO failure tree
1923  * to see if we need to process or clean up an io_failure_record
1924  */
1925 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1926 {
1927         u64 private;
1928         u64 private_failure;
1929         struct io_failure_record *failure;
1930         int ret;
1931
1932         private = 0;
1933         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1934                              (u64)-1, 1, EXTENT_DIRTY, 0)) {
1935                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1936                                         start, &private_failure);
1937                 if (ret == 0) {
1938                         failure = (struct io_failure_record *)(unsigned long)
1939                                    private_failure;
1940                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1941                                           failure->start, 0);
1942                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1943                                           failure->start,
1944                                           failure->start + failure->len - 1,
1945                                           EXTENT_DIRTY | EXTENT_LOCKED,
1946                                           GFP_NOFS);
1947                         kfree(failure);
1948                 }
1949         }
1950         return 0;
1951 }
1952
1953 /*
1954  * when reads are done, we need to check csums to verify the data is correct
1955  * if there's a match, we allow the bio to finish.  If not, we go through
1956  * the io_failure_record routines to find good copies
1957  */
1958 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1959                                struct extent_state *state)
1960 {
1961         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1962         struct inode *inode = page->mapping->host;
1963         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1964         char *kaddr;
1965         u64 private = ~(u32)0;
1966         int ret;
1967         struct btrfs_root *root = BTRFS_I(inode)->root;
1968         u32 csum = ~(u32)0;
1969
1970         if (PageChecked(page)) {
1971                 ClearPageChecked(page);
1972                 goto good;
1973         }
1974
1975         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1976                 return 0;
1977
1978         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1979             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1980                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1981                                   GFP_NOFS);
1982                 return 0;
1983         }
1984
1985         if (state && state->start == start) {
1986                 private = state->private;
1987                 ret = 0;
1988         } else {
1989                 ret = get_state_private(io_tree, start, &private);
1990         }
1991         kaddr = kmap_atomic(page, KM_USER0);
1992         if (ret)
1993                 goto zeroit;
1994
1995         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
1996         btrfs_csum_final(csum, (char *)&csum);
1997         if (csum != private)
1998                 goto zeroit;
1999
2000         kunmap_atomic(kaddr, KM_USER0);
2001 good:
2002         /* if the io failure tree for this inode is non-empty,
2003          * check to see if we've recovered from a failed IO
2004          */
2005         btrfs_clean_io_failures(inode, start);
2006         return 0;
2007
2008 zeroit:
2009         if (printk_ratelimit()) {
2010                 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
2011                        "private %llu\n", page->mapping->host->i_ino,
2012                        (unsigned long long)start, csum,
2013                        (unsigned long long)private);
2014         }
2015         memset(kaddr + offset, 1, end - start + 1);
2016         flush_dcache_page(page);
2017         kunmap_atomic(kaddr, KM_USER0);
2018         if (private == 0)
2019                 return 0;
2020         return -EIO;
2021 }
2022
2023 struct delayed_iput {
2024         struct list_head list;
2025         struct inode *inode;
2026 };
2027
2028 void btrfs_add_delayed_iput(struct inode *inode)
2029 {
2030         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2031         struct delayed_iput *delayed;
2032
2033         if (atomic_add_unless(&inode->i_count, -1, 1))
2034                 return;
2035
2036         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2037         delayed->inode = inode;
2038
2039         spin_lock(&fs_info->delayed_iput_lock);
2040         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2041         spin_unlock(&fs_info->delayed_iput_lock);
2042 }
2043
2044 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2045 {
2046         LIST_HEAD(list);
2047         struct btrfs_fs_info *fs_info = root->fs_info;
2048         struct delayed_iput *delayed;
2049         int empty;
2050
2051         spin_lock(&fs_info->delayed_iput_lock);
2052         empty = list_empty(&fs_info->delayed_iputs);
2053         spin_unlock(&fs_info->delayed_iput_lock);
2054         if (empty)
2055                 return;
2056
2057         down_read(&root->fs_info->cleanup_work_sem);
2058         spin_lock(&fs_info->delayed_iput_lock);
2059         list_splice_init(&fs_info->delayed_iputs, &list);
2060         spin_unlock(&fs_info->delayed_iput_lock);
2061
2062         while (!list_empty(&list)) {
2063                 delayed = list_entry(list.next, struct delayed_iput, list);
2064                 list_del(&delayed->list);
2065                 iput(delayed->inode);
2066                 kfree(delayed);
2067         }
2068         up_read(&root->fs_info->cleanup_work_sem);
2069 }
2070
2071 /*
2072  * calculate extra metadata reservation when snapshotting a subvolume
2073  * contains orphan files.
2074  */
2075 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans,
2076                                 struct btrfs_pending_snapshot *pending,
2077                                 u64 *bytes_to_reserve)
2078 {
2079         struct btrfs_root *root;
2080         struct btrfs_block_rsv *block_rsv;
2081         u64 num_bytes;
2082         int index;
2083
2084         root = pending->root;
2085         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2086                 return;
2087
2088         block_rsv = root->orphan_block_rsv;
2089
2090         /* orphan block reservation for the snapshot */
2091         num_bytes = block_rsv->size;
2092
2093         /*
2094          * after the snapshot is created, COWing tree blocks may use more
2095          * space than it frees. So we should make sure there is enough
2096          * reserved space.
2097          */
2098         index = trans->transid & 0x1;
2099         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2100                 num_bytes += block_rsv->size -
2101                              (block_rsv->reserved + block_rsv->freed[index]);
2102         }
2103
2104         *bytes_to_reserve += num_bytes;
2105 }
2106
2107 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans,
2108                                 struct btrfs_pending_snapshot *pending)
2109 {
2110         struct btrfs_root *root = pending->root;
2111         struct btrfs_root *snap = pending->snap;
2112         struct btrfs_block_rsv *block_rsv;
2113         u64 num_bytes;
2114         int index;
2115         int ret;
2116
2117         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2118                 return;
2119
2120         /* refill source subvolume's orphan block reservation */
2121         block_rsv = root->orphan_block_rsv;
2122         index = trans->transid & 0x1;
2123         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2124                 num_bytes = block_rsv->size -
2125                             (block_rsv->reserved + block_rsv->freed[index]);
2126                 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2127                                               root->orphan_block_rsv,
2128                                               num_bytes);
2129                 BUG_ON(ret);
2130         }
2131
2132         /* setup orphan block reservation for the snapshot */
2133         block_rsv = btrfs_alloc_block_rsv(snap);
2134         BUG_ON(!block_rsv);
2135
2136         btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2137         snap->orphan_block_rsv = block_rsv;
2138
2139         num_bytes = root->orphan_block_rsv->size;
2140         ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2141                                       block_rsv, num_bytes);
2142         BUG_ON(ret);
2143
2144 #if 0
2145         /* insert orphan item for the snapshot */
2146         WARN_ON(!root->orphan_item_inserted);
2147         ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2148                                        snap->root_key.objectid);
2149         BUG_ON(ret);
2150         snap->orphan_item_inserted = 1;
2151 #endif
2152 }
2153
2154 enum btrfs_orphan_cleanup_state {
2155         ORPHAN_CLEANUP_STARTED  = 1,
2156         ORPHAN_CLEANUP_DONE     = 2,
2157 };
2158
2159 /*
2160  * This is called in transaction commmit time. If there are no orphan
2161  * files in the subvolume, it removes orphan item and frees block_rsv
2162  * structure.
2163  */
2164 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2165                               struct btrfs_root *root)
2166 {
2167         int ret;
2168
2169         if (!list_empty(&root->orphan_list) ||
2170             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2171                 return;
2172
2173         if (root->orphan_item_inserted &&
2174             btrfs_root_refs(&root->root_item) > 0) {
2175                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2176                                             root->root_key.objectid);
2177                 BUG_ON(ret);
2178                 root->orphan_item_inserted = 0;
2179         }
2180
2181         if (root->orphan_block_rsv) {
2182                 WARN_ON(root->orphan_block_rsv->size > 0);
2183                 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2184                 root->orphan_block_rsv = NULL;
2185         }
2186 }
2187
2188 /*
2189  * This creates an orphan entry for the given inode in case something goes
2190  * wrong in the middle of an unlink/truncate.
2191  *
2192  * NOTE: caller of this function should reserve 5 units of metadata for
2193  *       this function.
2194  */
2195 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2196 {
2197         struct btrfs_root *root = BTRFS_I(inode)->root;
2198         struct btrfs_block_rsv *block_rsv = NULL;
2199         int reserve = 0;
2200         int insert = 0;
2201         int ret;
2202
2203         if (!root->orphan_block_rsv) {
2204                 block_rsv = btrfs_alloc_block_rsv(root);
2205                 BUG_ON(!block_rsv);
2206         }
2207
2208         spin_lock(&root->orphan_lock);
2209         if (!root->orphan_block_rsv) {
2210                 root->orphan_block_rsv = block_rsv;
2211         } else if (block_rsv) {
2212                 btrfs_free_block_rsv(root, block_rsv);
2213                 block_rsv = NULL;
2214         }
2215
2216         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2217                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2218 #if 0
2219                 /*
2220                  * For proper ENOSPC handling, we should do orphan
2221                  * cleanup when mounting. But this introduces backward
2222                  * compatibility issue.
2223                  */
2224                 if (!xchg(&root->orphan_item_inserted, 1))
2225                         insert = 2;
2226                 else
2227                         insert = 1;
2228 #endif
2229                 insert = 1;
2230         }
2231
2232         if (!BTRFS_I(inode)->orphan_meta_reserved) {
2233                 BTRFS_I(inode)->orphan_meta_reserved = 1;
2234                 reserve = 1;
2235         }
2236         spin_unlock(&root->orphan_lock);
2237
2238         if (block_rsv)
2239                 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2240
2241         /* grab metadata reservation from transaction handle */
2242         if (reserve) {
2243                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2244                 BUG_ON(ret);
2245         }
2246
2247         /* insert an orphan item to track this unlinked/truncated file */
2248         if (insert >= 1) {
2249                 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
2250                 BUG_ON(ret);
2251         }
2252
2253         /* insert an orphan item to track subvolume contains orphan files */
2254         if (insert >= 2) {
2255                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2256                                                root->root_key.objectid);
2257                 BUG_ON(ret);
2258         }
2259         return 0;
2260 }
2261
2262 /*
2263  * We have done the truncate/delete so we can go ahead and remove the orphan
2264  * item for this particular inode.
2265  */
2266 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2267 {
2268         struct btrfs_root *root = BTRFS_I(inode)->root;
2269         int delete_item = 0;
2270         int release_rsv = 0;
2271         int ret = 0;
2272
2273         spin_lock(&root->orphan_lock);
2274         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2275                 list_del_init(&BTRFS_I(inode)->i_orphan);
2276                 delete_item = 1;
2277         }
2278
2279         if (BTRFS_I(inode)->orphan_meta_reserved) {
2280                 BTRFS_I(inode)->orphan_meta_reserved = 0;
2281                 release_rsv = 1;
2282         }
2283         spin_unlock(&root->orphan_lock);
2284
2285         if (trans && delete_item) {
2286                 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
2287                 BUG_ON(ret);
2288         }
2289
2290         if (release_rsv)
2291                 btrfs_orphan_release_metadata(inode);
2292
2293         return 0;
2294 }
2295
2296 /*
2297  * this cleans up any orphans that may be left on the list from the last use
2298  * of this root.
2299  */
2300 int btrfs_orphan_cleanup(struct btrfs_root *root)
2301 {
2302         struct btrfs_path *path;
2303         struct extent_buffer *leaf;
2304         struct btrfs_key key, found_key;
2305         struct btrfs_trans_handle *trans;
2306         struct inode *inode;
2307         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2308
2309         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2310                 return 0;
2311
2312         path = btrfs_alloc_path();
2313         if (!path) {
2314                 ret = -ENOMEM;
2315                 goto out;
2316         }
2317         path->reada = -1;
2318
2319         key.objectid = BTRFS_ORPHAN_OBJECTID;
2320         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2321         key.offset = (u64)-1;
2322
2323         while (1) {
2324                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2325                 if (ret < 0)
2326                         goto out;
2327
2328                 /*
2329                  * if ret == 0 means we found what we were searching for, which
2330                  * is weird, but possible, so only screw with path if we didn't
2331                  * find the key and see if we have stuff that matches
2332                  */
2333                 if (ret > 0) {
2334                         ret = 0;
2335                         if (path->slots[0] == 0)
2336                                 break;
2337                         path->slots[0]--;
2338                 }
2339
2340                 /* pull out the item */
2341                 leaf = path->nodes[0];
2342                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2343
2344                 /* make sure the item matches what we want */
2345                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2346                         break;
2347                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2348                         break;
2349
2350                 /* release the path since we're done with it */
2351                 btrfs_release_path(root, path);
2352
2353                 /*
2354                  * this is where we are basically btrfs_lookup, without the
2355                  * crossing root thing.  we store the inode number in the
2356                  * offset of the orphan item.
2357                  */
2358                 found_key.objectid = found_key.offset;
2359                 found_key.type = BTRFS_INODE_ITEM_KEY;
2360                 found_key.offset = 0;
2361                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2362                 if (IS_ERR(inode)) {
2363                         ret = PTR_ERR(inode);
2364                         goto out;
2365                 }
2366
2367                 /*
2368                  * add this inode to the orphan list so btrfs_orphan_del does
2369                  * the proper thing when we hit it
2370                  */
2371                 spin_lock(&root->orphan_lock);
2372                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2373                 spin_unlock(&root->orphan_lock);
2374
2375                 /*
2376                  * if this is a bad inode, means we actually succeeded in
2377                  * removing the inode, but not the orphan record, which means
2378                  * we need to manually delete the orphan since iput will just
2379                  * do a destroy_inode
2380                  */
2381                 if (is_bad_inode(inode)) {
2382                         trans = btrfs_start_transaction(root, 0);
2383                         if (IS_ERR(trans)) {
2384                                 ret = PTR_ERR(trans);
2385                                 goto out;
2386                         }
2387                         btrfs_orphan_del(trans, inode);
2388                         btrfs_end_transaction(trans, root);
2389                         iput(inode);
2390                         continue;
2391                 }
2392
2393                 /* if we have links, this was a truncate, lets do that */
2394                 if (inode->i_nlink) {
2395                         if (!S_ISREG(inode->i_mode)) {
2396                                 WARN_ON(1);
2397                                 iput(inode);
2398                                 continue;
2399                         }
2400                         nr_truncate++;
2401                         ret = btrfs_truncate(inode);
2402                 } else {
2403                         nr_unlink++;
2404                 }
2405
2406                 /* this will do delete_inode and everything for us */
2407                 iput(inode);
2408                 if (ret)
2409                         goto out;
2410         }
2411         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2412
2413         if (root->orphan_block_rsv)
2414                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2415                                         (u64)-1);
2416
2417         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2418                 trans = btrfs_join_transaction(root);
2419                 if (!IS_ERR(trans))
2420                         btrfs_end_transaction(trans, root);
2421         }
2422
2423         if (nr_unlink)
2424                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2425         if (nr_truncate)
2426                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2427
2428 out:
2429         if (ret)
2430                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2431         btrfs_free_path(path);
2432         return ret;
2433 }
2434
2435 /*
2436  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2437  * don't find any xattrs, we know there can't be any acls.
2438  *
2439  * slot is the slot the inode is in, objectid is the objectid of the inode
2440  */
2441 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2442                                           int slot, u64 objectid)
2443 {
2444         u32 nritems = btrfs_header_nritems(leaf);
2445         struct btrfs_key found_key;
2446         int scanned = 0;
2447
2448         slot++;
2449         while (slot < nritems) {
2450                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2451
2452                 /* we found a different objectid, there must not be acls */
2453                 if (found_key.objectid != objectid)
2454                         return 0;
2455
2456                 /* we found an xattr, assume we've got an acl */
2457                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2458                         return 1;
2459
2460                 /*
2461                  * we found a key greater than an xattr key, there can't
2462                  * be any acls later on
2463                  */
2464                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2465                         return 0;
2466
2467                 slot++;
2468                 scanned++;
2469
2470                 /*
2471                  * it goes inode, inode backrefs, xattrs, extents,
2472                  * so if there are a ton of hard links to an inode there can
2473                  * be a lot of backrefs.  Don't waste time searching too hard,
2474                  * this is just an optimization
2475                  */
2476                 if (scanned >= 8)
2477                         break;
2478         }
2479         /* we hit the end of the leaf before we found an xattr or
2480          * something larger than an xattr.  We have to assume the inode
2481          * has acls
2482          */
2483         return 1;
2484 }
2485
2486 /*
2487  * read an inode from the btree into the in-memory inode
2488  */
2489 static void btrfs_read_locked_inode(struct inode *inode)
2490 {
2491         struct btrfs_path *path;
2492         struct extent_buffer *leaf;
2493         struct btrfs_inode_item *inode_item;
2494         struct btrfs_timespec *tspec;
2495         struct btrfs_root *root = BTRFS_I(inode)->root;
2496         struct btrfs_key location;
2497         int maybe_acls;
2498         u64 alloc_group_block;
2499         u32 rdev;
2500         int ret;
2501
2502         path = btrfs_alloc_path();
2503         BUG_ON(!path);
2504         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2505
2506         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2507         if (ret)
2508                 goto make_bad;
2509
2510         leaf = path->nodes[0];
2511         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2512                                     struct btrfs_inode_item);
2513
2514         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2515         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2516         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2517         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2518         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2519
2520         tspec = btrfs_inode_atime(inode_item);
2521         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2522         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2523
2524         tspec = btrfs_inode_mtime(inode_item);
2525         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2526         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2527
2528         tspec = btrfs_inode_ctime(inode_item);
2529         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2530         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2531
2532         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2533         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2534         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2535         inode->i_generation = BTRFS_I(inode)->generation;
2536         inode->i_rdev = 0;
2537         rdev = btrfs_inode_rdev(leaf, inode_item);
2538
2539         BTRFS_I(inode)->index_cnt = (u64)-1;
2540         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2541
2542         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2543
2544         /*
2545          * try to precache a NULL acl entry for files that don't have
2546          * any xattrs or acls
2547          */
2548         maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2549         if (!maybe_acls)
2550                 cache_no_acl(inode);
2551
2552         BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2553                                                 alloc_group_block, 0);
2554         btrfs_free_path(path);
2555         inode_item = NULL;
2556
2557         switch (inode->i_mode & S_IFMT) {
2558         case S_IFREG:
2559                 inode->i_mapping->a_ops = &btrfs_aops;
2560                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2561                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2562                 inode->i_fop = &btrfs_file_operations;
2563                 inode->i_op = &btrfs_file_inode_operations;
2564                 break;
2565         case S_IFDIR:
2566                 inode->i_fop = &btrfs_dir_file_operations;
2567                 if (root == root->fs_info->tree_root)
2568                         inode->i_op = &btrfs_dir_ro_inode_operations;
2569                 else
2570                         inode->i_op = &btrfs_dir_inode_operations;
2571                 break;
2572         case S_IFLNK:
2573                 inode->i_op = &btrfs_symlink_inode_operations;
2574                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2575                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2576                 break;
2577         default:
2578                 inode->i_op = &btrfs_special_inode_operations;
2579                 init_special_inode(inode, inode->i_mode, rdev);
2580                 break;
2581         }
2582
2583         btrfs_update_iflags(inode);
2584         return;
2585
2586 make_bad:
2587         btrfs_free_path(path);
2588         make_bad_inode(inode);
2589 }
2590
2591 /*
2592  * given a leaf and an inode, copy the inode fields into the leaf
2593  */
2594 static void fill_inode_item(struct btrfs_trans_handle *trans,
2595                             struct extent_buffer *leaf,
2596                             struct btrfs_inode_item *item,
2597                             struct inode *inode)
2598 {
2599         if (!leaf->map_token)
2600                 map_private_extent_buffer(leaf, (unsigned long)item,
2601                                           sizeof(struct btrfs_inode_item),
2602                                           &leaf->map_token, &leaf->kaddr,
2603                                           &leaf->map_start, &leaf->map_len,
2604                                           KM_USER1);
2605
2606         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2607         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2608         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2609         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2610         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2611
2612         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2613                                inode->i_atime.tv_sec);
2614         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2615                                 inode->i_atime.tv_nsec);
2616
2617         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2618                                inode->i_mtime.tv_sec);
2619         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2620                                 inode->i_mtime.tv_nsec);
2621
2622         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2623                                inode->i_ctime.tv_sec);
2624         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2625                                 inode->i_ctime.tv_nsec);
2626
2627         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2628         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2629         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2630         btrfs_set_inode_transid(leaf, item, trans->transid);
2631         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2632         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2633         btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2634
2635         if (leaf->map_token) {
2636                 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2637                 leaf->map_token = NULL;
2638         }
2639 }
2640
2641 /*
2642  * copy everything in the in-memory inode into the btree.
2643  */
2644 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2645                                 struct btrfs_root *root, struct inode *inode)
2646 {
2647         struct btrfs_inode_item *inode_item;
2648         struct btrfs_path *path;
2649         struct extent_buffer *leaf;
2650         int ret;
2651
2652         path = btrfs_alloc_path();
2653         BUG_ON(!path);
2654         path->leave_spinning = 1;
2655         ret = btrfs_lookup_inode(trans, root, path,
2656                                  &BTRFS_I(inode)->location, 1);
2657         if (ret) {
2658                 if (ret > 0)
2659                         ret = -ENOENT;
2660                 goto failed;
2661         }
2662
2663         btrfs_unlock_up_safe(path, 1);
2664         leaf = path->nodes[0];
2665         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2666                                   struct btrfs_inode_item);
2667
2668         fill_inode_item(trans, leaf, inode_item, inode);
2669         btrfs_mark_buffer_dirty(leaf);
2670         btrfs_set_inode_last_trans(trans, inode);
2671         ret = 0;
2672 failed:
2673         btrfs_free_path(path);
2674         return ret;
2675 }
2676
2677
2678 /*
2679  * unlink helper that gets used here in inode.c and in the tree logging
2680  * recovery code.  It remove a link in a directory with a given name, and
2681  * also drops the back refs in the inode to the directory
2682  */
2683 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2684                                 struct btrfs_root *root,
2685                                 struct inode *dir, struct inode *inode,
2686                                 const char *name, int name_len)
2687 {
2688         struct btrfs_path *path;
2689         int ret = 0;
2690         struct extent_buffer *leaf;
2691         struct btrfs_dir_item *di;
2692         struct btrfs_key key;
2693         u64 index;
2694
2695         path = btrfs_alloc_path();
2696         if (!path) {
2697                 ret = -ENOMEM;
2698                 goto out;
2699         }
2700
2701         path->leave_spinning = 1;
2702         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2703                                     name, name_len, -1);
2704         if (IS_ERR(di)) {
2705                 ret = PTR_ERR(di);
2706                 goto err;
2707         }
2708         if (!di) {
2709                 ret = -ENOENT;
2710                 goto err;
2711         }
2712         leaf = path->nodes[0];
2713         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2714         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2715         if (ret)
2716                 goto err;
2717         btrfs_release_path(root, path);
2718
2719         ret = btrfs_del_inode_ref(trans, root, name, name_len,
2720                                   inode->i_ino,
2721                                   dir->i_ino, &index);
2722         if (ret) {
2723                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2724                        "inode %lu parent %lu\n", name_len, name,
2725                        inode->i_ino, dir->i_ino);
2726                 goto err;
2727         }
2728
2729         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2730                                          index, name, name_len, -1);
2731         if (IS_ERR(di)) {
2732                 ret = PTR_ERR(di);
2733                 goto err;
2734         }
2735         if (!di) {
2736                 ret = -ENOENT;
2737                 goto err;
2738         }
2739         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2740         btrfs_release_path(root, path);
2741
2742         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2743                                          inode, dir->i_ino);
2744         BUG_ON(ret != 0 && ret != -ENOENT);
2745
2746         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2747                                            dir, index);
2748         if (ret == -ENOENT)
2749                 ret = 0;
2750 err:
2751         btrfs_free_path(path);
2752         if (ret)
2753                 goto out;
2754
2755         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2756         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2757         btrfs_update_inode(trans, root, dir);
2758 out:
2759         return ret;
2760 }
2761
2762 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2763                        struct btrfs_root *root,
2764                        struct inode *dir, struct inode *inode,
2765                        const char *name, int name_len)
2766 {
2767         int ret;
2768         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2769         if (!ret) {
2770                 btrfs_drop_nlink(inode);
2771                 ret = btrfs_update_inode(trans, root, inode);
2772         }
2773         return ret;
2774 }
2775
2776
2777 /* helper to check if there is any shared block in the path */
2778 static int check_path_shared(struct btrfs_root *root,
2779                              struct btrfs_path *path)
2780 {
2781         struct extent_buffer *eb;
2782         int level;
2783         u64 refs = 1;
2784
2785         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2786                 int ret;
2787
2788                 if (!path->nodes[level])
2789                         break;
2790                 eb = path->nodes[level];
2791                 if (!btrfs_block_can_be_shared(root, eb))
2792                         continue;
2793                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2794                                                &refs, NULL);
2795                 if (refs > 1)
2796                         return 1;
2797         }
2798         return 0;
2799 }
2800
2801 /*
2802  * helper to start transaction for unlink and rmdir.
2803  *
2804  * unlink and rmdir are special in btrfs, they do not always free space.
2805  * so in enospc case, we should make sure they will free space before
2806  * allowing them to use the global metadata reservation.
2807  */
2808 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2809                                                        struct dentry *dentry)
2810 {
2811         struct btrfs_trans_handle *trans;
2812         struct btrfs_root *root = BTRFS_I(dir)->root;
2813         struct btrfs_path *path;
2814         struct btrfs_inode_ref *ref;
2815         struct btrfs_dir_item *di;
2816         struct inode *inode = dentry->d_inode;
2817         u64 index;
2818         int check_link = 1;
2819         int err = -ENOSPC;
2820         int ret;
2821
2822         trans = btrfs_start_transaction(root, 10);
2823         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2824                 return trans;
2825
2826         if (inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2827                 return ERR_PTR(-ENOSPC);
2828
2829         /* check if there is someone else holds reference */
2830         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2831                 return ERR_PTR(-ENOSPC);
2832
2833         if (atomic_read(&inode->i_count) > 2)
2834                 return ERR_PTR(-ENOSPC);
2835
2836         if (xchg(&root->fs_info->enospc_unlink, 1))
2837                 return ERR_PTR(-ENOSPC);
2838
2839         path = btrfs_alloc_path();
2840         if (!path) {
2841                 root->fs_info->enospc_unlink = 0;
2842                 return ERR_PTR(-ENOMEM);
2843         }
2844
2845         trans = btrfs_start_transaction(root, 0);
2846         if (IS_ERR(trans)) {
2847                 btrfs_free_path(path);
2848                 root->fs_info->enospc_unlink = 0;
2849                 return trans;
2850         }
2851
2852         path->skip_locking = 1;
2853         path->search_commit_root = 1;
2854
2855         ret = btrfs_lookup_inode(trans, root, path,
2856                                 &BTRFS_I(dir)->location, 0);
2857         if (ret < 0) {
2858                 err = ret;
2859                 goto out;
2860         }
2861         if (ret == 0) {
2862                 if (check_path_shared(root, path))
2863                         goto out;
2864         } else {
2865                 check_link = 0;
2866         }
2867         btrfs_release_path(root, path);
2868
2869         ret = btrfs_lookup_inode(trans, root, path,
2870                                 &BTRFS_I(inode)->location, 0);
2871         if (ret < 0) {
2872                 err = ret;
2873                 goto out;
2874         }
2875         if (ret == 0) {
2876                 if (check_path_shared(root, path))
2877                         goto out;
2878         } else {
2879                 check_link = 0;
2880         }
2881         btrfs_release_path(root, path);
2882
2883         if (ret == 0 && S_ISREG(inode->i_mode)) {
2884                 ret = btrfs_lookup_file_extent(trans, root, path,
2885                                                inode->i_ino, (u64)-1, 0);
2886                 if (ret < 0) {
2887                         err = ret;
2888                         goto out;
2889                 }
2890                 BUG_ON(ret == 0);
2891                 if (check_path_shared(root, path))
2892                         goto out;
2893                 btrfs_release_path(root, path);
2894         }
2895
2896         if (!check_link) {
2897                 err = 0;
2898                 goto out;
2899         }
2900
2901         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2902                                 dentry->d_name.name, dentry->d_name.len, 0);
2903         if (IS_ERR(di)) {
2904                 err = PTR_ERR(di);
2905                 goto out;
2906         }
2907         if (di) {
2908                 if (check_path_shared(root, path))
2909                         goto out;
2910         } else {
2911                 err = 0;
2912                 goto out;
2913         }
2914         btrfs_release_path(root, path);
2915
2916         ref = btrfs_lookup_inode_ref(trans, root, path,
2917                                 dentry->d_name.name, dentry->d_name.len,
2918                                 inode->i_ino, dir->i_ino, 0);
2919         if (IS_ERR(ref)) {
2920                 err = PTR_ERR(ref);
2921                 goto out;
2922         }
2923         BUG_ON(!ref);
2924         if (check_path_shared(root, path))
2925                 goto out;
2926         index = btrfs_inode_ref_index(path->nodes[0], ref);
2927         btrfs_release_path(root, path);
2928
2929         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, index,
2930                                 dentry->d_name.name, dentry->d_name.len, 0);
2931         if (IS_ERR(di)) {
2932                 err = PTR_ERR(di);
2933                 goto out;
2934         }
2935         BUG_ON(ret == -ENOENT);
2936         if (check_path_shared(root, path))
2937                 goto out;
2938
2939         err = 0;
2940 out:
2941         btrfs_free_path(path);
2942         if (err) {
2943                 btrfs_end_transaction(trans, root);
2944                 root->fs_info->enospc_unlink = 0;
2945                 return ERR_PTR(err);
2946         }
2947
2948         trans->block_rsv = &root->fs_info->global_block_rsv;
2949         return trans;
2950 }
2951
2952 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2953                                struct btrfs_root *root)
2954 {
2955         if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2956                 BUG_ON(!root->fs_info->enospc_unlink);
2957                 root->fs_info->enospc_unlink = 0;
2958         }
2959         btrfs_end_transaction_throttle(trans, root);
2960 }
2961
2962 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2963 {
2964         struct btrfs_root *root = BTRFS_I(dir)->root;
2965         struct btrfs_trans_handle *trans;
2966         struct inode *inode = dentry->d_inode;
2967         int ret;
2968         unsigned long nr = 0;
2969
2970         trans = __unlink_start_trans(dir, dentry);
2971         if (IS_ERR(trans))
2972                 return PTR_ERR(trans);
2973
2974         btrfs_set_trans_block_group(trans, dir);
2975
2976         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2977
2978         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2979                                  dentry->d_name.name, dentry->d_name.len);
2980         BUG_ON(ret);
2981
2982         if (inode->i_nlink == 0) {
2983                 ret = btrfs_orphan_add(trans, inode);
2984                 BUG_ON(ret);
2985         }
2986
2987         nr = trans->blocks_used;
2988         __unlink_end_trans(trans, root);
2989         btrfs_btree_balance_dirty(root, nr);
2990         return ret;
2991 }
2992
2993 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2994                         struct btrfs_root *root,
2995                         struct inode *dir, u64 objectid,
2996                         const char *name, int name_len)
2997 {
2998         struct btrfs_path *path;
2999         struct extent_buffer *leaf;
3000         struct btrfs_dir_item *di;
3001         struct btrfs_key key;
3002         u64 index;
3003         int ret;
3004
3005         path = btrfs_alloc_path();
3006         if (!path)
3007                 return -ENOMEM;
3008
3009         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
3010                                    name, name_len, -1);
3011         BUG_ON(!di || IS_ERR(di));
3012
3013         leaf = path->nodes[0];
3014         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3015         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3016         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3017         BUG_ON(ret);
3018         btrfs_release_path(root, path);
3019
3020         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3021                                  objectid, root->root_key.objectid,
3022                                  dir->i_ino, &index, name, name_len);
3023         if (ret < 0) {
3024                 BUG_ON(ret != -ENOENT);
3025                 di = btrfs_search_dir_index_item(root, path, dir->i_ino,
3026                                                  name, name_len);
3027                 BUG_ON(!di || IS_ERR(di));
3028
3029                 leaf = path->nodes[0];
3030                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3031                 btrfs_release_path(root, path);
3032                 index = key.offset;
3033         }
3034
3035         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
3036                                          index, name, name_len, -1);
3037         BUG_ON(!di || IS_ERR(di));
3038
3039         leaf = path->nodes[0];
3040         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3041         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3042         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3043         BUG_ON(ret);
3044         btrfs_release_path(root, path);
3045
3046         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3047         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3048         ret = btrfs_update_inode(trans, root, dir);
3049         BUG_ON(ret);
3050
3051         btrfs_free_path(path);
3052         return 0;
3053 }
3054
3055 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3056 {
3057         struct inode *inode = dentry->d_inode;
3058         int err = 0;
3059         struct btrfs_root *root = BTRFS_I(dir)->root;
3060         struct btrfs_trans_handle *trans;
3061         unsigned long nr = 0;
3062
3063         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3064             inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
3065                 return -ENOTEMPTY;
3066
3067         trans = __unlink_start_trans(dir, dentry);
3068         if (IS_ERR(trans))
3069                 return PTR_ERR(trans);
3070
3071         btrfs_set_trans_block_group(trans, dir);
3072
3073         if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3074                 err = btrfs_unlink_subvol(trans, root, dir,
3075                                           BTRFS_I(inode)->location.objectid,
3076                                           dentry->d_name.name,
3077                                           dentry->d_name.len);
3078                 goto out;
3079         }
3080
3081         err = btrfs_orphan_add(trans, inode);
3082         if (err)
3083                 goto out;
3084
3085         /* now the directory is empty */
3086         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3087                                  dentry->d_name.name, dentry->d_name.len);
3088         if (!err)
3089                 btrfs_i_size_write(inode, 0);
3090 out:
3091         nr = trans->blocks_used;
3092         __unlink_end_trans(trans, root);
3093         btrfs_btree_balance_dirty(root, nr);
3094
3095         return err;
3096 }
3097
3098 #if 0
3099 /*
3100  * when truncating bytes in a file, it is possible to avoid reading
3101  * the leaves that contain only checksum items.  This can be the
3102  * majority of the IO required to delete a large file, but it must
3103  * be done carefully.
3104  *
3105  * The keys in the level just above the leaves are checked to make sure
3106  * the lowest key in a given leaf is a csum key, and starts at an offset
3107  * after the new  size.
3108  *
3109  * Then the key for the next leaf is checked to make sure it also has
3110  * a checksum item for the same file.  If it does, we know our target leaf
3111  * contains only checksum items, and it can be safely freed without reading
3112  * it.
3113  *
3114  * This is just an optimization targeted at large files.  It may do
3115  * nothing.  It will return 0 unless things went badly.
3116  */
3117 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
3118                                      struct btrfs_root *root,
3119                                      struct btrfs_path *path,
3120                                      struct inode *inode, u64 new_size)
3121 {
3122         struct btrfs_key key;
3123         int ret;
3124         int nritems;
3125         struct btrfs_key found_key;
3126         struct btrfs_key other_key;
3127         struct btrfs_leaf_ref *ref;
3128         u64 leaf_gen;
3129         u64 leaf_start;
3130
3131         path->lowest_level = 1;
3132         key.objectid = inode->i_ino;
3133         key.type = BTRFS_CSUM_ITEM_KEY;
3134         key.offset = new_size;
3135 again:
3136         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3137         if (ret < 0)
3138                 goto out;
3139
3140         if (path->nodes[1] == NULL) {
3141                 ret = 0;
3142                 goto out;
3143         }
3144         ret = 0;
3145         btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
3146         nritems = btrfs_header_nritems(path->nodes[1]);
3147
3148         if (!nritems)
3149                 goto out;
3150
3151         if (path->slots[1] >= nritems)
3152                 goto next_node;
3153
3154         /* did we find a key greater than anything we want to delete? */
3155         if (found_key.objectid > inode->i_ino ||
3156            (found_key.objectid == inode->i_ino && found_key.type > key.type))
3157                 goto out;
3158
3159         /* we check the next key in the node to make sure the leave contains
3160          * only checksum items.  This comparison doesn't work if our
3161          * leaf is the last one in the node
3162          */
3163         if (path->slots[1] + 1 >= nritems) {
3164 next_node:
3165                 /* search forward from the last key in the node, this
3166                  * will bring us into the next node in the tree
3167                  */
3168                 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
3169
3170                 /* unlikely, but we inc below, so check to be safe */
3171                 if (found_key.offset == (u64)-1)
3172                         goto out;
3173
3174                 /* search_forward needs a path with locks held, do the
3175                  * search again for the original key.  It is possible
3176                  * this will race with a balance and return a path that
3177                  * we could modify, but this drop is just an optimization
3178                  * and is allowed to miss some leaves.
3179                  */
3180                 btrfs_release_path(root, path);
3181                 found_key.offset++;
3182
3183                 /* setup a max key for search_forward */
3184                 other_key.offset = (u64)-1;
3185                 other_key.type = key.type;
3186                 other_key.objectid = key.objectid;
3187
3188                 path->keep_locks = 1;
3189                 ret = btrfs_search_forward(root, &found_key, &other_key,
3190                                            path, 0, 0);
3191                 path->keep_locks = 0;
3192                 if (ret || found_key.objectid != key.objectid ||
3193                     found_key.type != key.type) {
3194                         ret = 0;
3195                         goto out;
3196                 }
3197
3198                 key.offset = found_key.offset;
3199                 btrfs_release_path(root, path);
3200                 cond_resched();
3201                 goto again;
3202         }
3203
3204         /* we know there's one more slot after us in the tree,
3205          * read that key so we can verify it is also a checksum item
3206          */
3207         btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
3208
3209         if (found_key.objectid < inode->i_ino)
3210                 goto next_key;
3211
3212         if (found_key.type != key.type || found_key.offset < new_size)
3213                 goto next_key;
3214
3215         /*
3216          * if the key for the next leaf isn't a csum key from this objectid,
3217          * we can't be sure there aren't good items inside this leaf.
3218          * Bail out
3219          */
3220         if (other_key.objectid != inode->i_ino || other_key.type != key.type)
3221                 goto out;
3222
3223         leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
3224         leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
3225         /*
3226          * it is safe to delete this leaf, it contains only
3227          * csum items from this inode at an offset >= new_size
3228          */
3229         ret = btrfs_del_leaf(trans, root, path, leaf_start);
3230         BUG_ON(ret);
3231
3232         if (root->ref_cows && leaf_gen < trans->transid) {
3233                 ref = btrfs_alloc_leaf_ref(root, 0);
3234                 if (ref) {
3235                         ref->root_gen = root->root_key.offset;
3236                         ref->bytenr = leaf_start;
3237                         ref->owner = 0;
3238                         ref->generation = leaf_gen;
3239                         ref->nritems = 0;
3240
3241                         btrfs_sort_leaf_ref(ref);
3242
3243                         ret = btrfs_add_leaf_ref(root, ref, 0);
3244                         WARN_ON(ret);
3245                         btrfs_free_leaf_ref(root, ref);
3246                 } else {
3247                         WARN_ON(1);
3248                 }
3249         }
3250 next_key:
3251         btrfs_release_path(root, path);
3252
3253         if (other_key.objectid == inode->i_ino &&
3254             other_key.type == key.type && other_key.offset > key.offset) {
3255                 key.offset = other_key.offset;
3256                 cond_resched();
3257                 goto again;
3258         }
3259         ret = 0;
3260 out:
3261         /* fixup any changes we've made to the path */
3262         path->lowest_level = 0;
3263         path->keep_locks = 0;
3264         btrfs_release_path(root, path);
3265         return ret;
3266 }
3267
3268 #endif
3269
3270 /*
3271  * this can truncate away extent items, csum items and directory items.
3272  * It starts at a high offset and removes keys until it can't find
3273  * any higher than new_size
3274  *
3275  * csum items that cross the new i_size are truncated to the new size
3276  * as well.
3277  *
3278  * min_type is the minimum key type to truncate down to.  If set to 0, this
3279  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3280  */
3281 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3282                                struct btrfs_root *root,
3283                                struct inode *inode,
3284                                u64 new_size, u32 min_type)
3285 {
3286         struct btrfs_path *path;
3287         struct extent_buffer *leaf;
3288         struct btrfs_file_extent_item *fi;
3289         struct btrfs_key key;
3290         struct btrfs_key found_key;
3291         u64 extent_start = 0;
3292         u64 extent_num_bytes = 0;
3293         u64 extent_offset = 0;
3294         u64 item_end = 0;
3295         u64 mask = root->sectorsize - 1;
3296         u32 found_type = (u8)-1;
3297         int found_extent;
3298         int del_item;
3299         int pending_del_nr = 0;
3300         int pending_del_slot = 0;
3301         int extent_type = -1;
3302         int encoding;
3303         int ret;
3304         int err = 0;
3305
3306         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3307
3308         if (root->ref_cows || root == root->fs_info->tree_root)
3309                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3310
3311         path = btrfs_alloc_path();
3312         BUG_ON(!path);
3313         path->reada = -1;
3314
3315         key.objectid = inode->i_ino;
3316         key.offset = (u64)-1;
3317         key.type = (u8)-1;
3318
3319 search_again:
3320         path->leave_spinning = 1;
3321         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3322         if (ret < 0) {
3323                 err = ret;
3324                 goto out;
3325         }
3326
3327         if (ret > 0) {
3328                 /* there are no items in the tree for us to truncate, we're
3329                  * done
3330                  */
3331                 if (path->slots[0] == 0)
3332                         goto out;
3333                 path->slots[0]--;
3334         }
3335
3336         while (1) {
3337                 fi = NULL;
3338                 leaf = path->nodes[0];
3339                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3340                 found_type = btrfs_key_type(&found_key);
3341                 encoding = 0;
3342
3343                 if (found_key.objectid != inode->i_ino)
3344                         break;
3345
3346                 if (found_type < min_type)
3347                         break;
3348
3349                 item_end = found_key.offset;
3350                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3351                         fi = btrfs_item_ptr(leaf, path->slots[0],
3352                                             struct btrfs_file_extent_item);
3353                         extent_type = btrfs_file_extent_type(leaf, fi);
3354                         encoding = btrfs_file_extent_compression(leaf, fi);
3355                         encoding |= btrfs_file_extent_encryption(leaf, fi);
3356                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3357
3358                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3359                                 item_end +=
3360                                     btrfs_file_extent_num_bytes(leaf, fi);
3361                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3362                                 item_end += btrfs_file_extent_inline_len(leaf,
3363                                                                          fi);
3364                         }
3365                         item_end--;
3366                 }
3367                 if (found_type > min_type) {
3368                         del_item = 1;
3369                 } else {
3370                         if (item_end < new_size)
3371                                 break;
3372                         if (found_key.offset >= new_size)
3373                                 del_item = 1;
3374                         else
3375                                 del_item = 0;
3376                 }
3377                 found_extent = 0;
3378                 /* FIXME, shrink the extent if the ref count is only 1 */
3379                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3380                         goto delete;
3381
3382                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3383                         u64 num_dec;
3384                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3385                         if (!del_item && !encoding) {
3386                                 u64 orig_num_bytes =
3387                                         btrfs_file_extent_num_bytes(leaf, fi);
3388                                 extent_num_bytes = new_size -
3389                                         found_key.offset + root->sectorsize - 1;
3390                                 extent_num_bytes = extent_num_bytes &
3391                                         ~((u64)root->sectorsize - 1);
3392                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3393                                                          extent_num_bytes);
3394                                 num_dec = (orig_num_bytes -
3395                                            extent_num_bytes);
3396                                 if (root->ref_cows && extent_start != 0)
3397                                         inode_sub_bytes(inode, num_dec);
3398                                 btrfs_mark_buffer_dirty(leaf);
3399                         } else {
3400                                 extent_num_bytes =
3401                                         btrfs_file_extent_disk_num_bytes(leaf,
3402                                                                          fi);
3403                                 extent_offset = found_key.offset -
3404                                         btrfs_file_extent_offset(leaf, fi);
3405
3406                                 /* FIXME blocksize != 4096 */
3407                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3408                                 if (extent_start != 0) {
3409                                         found_extent = 1;
3410                                         if (root->ref_cows)
3411                                                 inode_sub_bytes(inode, num_dec);
3412                                 }
3413                         }
3414                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3415                         /*
3416                          * we can't truncate inline items that have had
3417                          * special encodings
3418                          */
3419                         if (!del_item &&
3420                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3421                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3422                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3423                                 u32 size = new_size - found_key.offset;
3424
3425                                 if (root->ref_cows) {
3426                                         inode_sub_bytes(inode, item_end + 1 -
3427                                                         new_size);
3428                                 }
3429                                 size =
3430                                     btrfs_file_extent_calc_inline_size(size);
3431                                 ret = btrfs_truncate_item(trans, root, path,
3432                                                           size, 1);
3433                                 BUG_ON(ret);
3434                         } else if (root->ref_cows) {
3435                                 inode_sub_bytes(inode, item_end + 1 -
3436                                                 found_key.offset);
3437                         }
3438                 }
3439 delete:
3440                 if (del_item) {
3441                         if (!pending_del_nr) {
3442                                 /* no pending yet, add ourselves */
3443                                 pending_del_slot = path->slots[0];
3444                                 pending_del_nr = 1;
3445                         } else if (pending_del_nr &&
3446                                    path->slots[0] + 1 == pending_del_slot) {
3447                                 /* hop on the pending chunk */
3448                                 pending_del_nr++;
3449                                 pending_del_slot = path->slots[0];
3450                         } else {
3451                                 BUG();
3452                         }
3453                 } else {
3454                         break;
3455                 }
3456                 if (found_extent && (root->ref_cows ||
3457                                      root == root->fs_info->tree_root)) {
3458                         btrfs_set_path_blocking(path);
3459                         ret = btrfs_free_extent(trans, root, extent_start,
3460                                                 extent_num_bytes, 0,
3461                                                 btrfs_header_owner(leaf),
3462                                                 inode->i_ino, extent_offset);
3463                         BUG_ON(ret);
3464                 }
3465
3466                 if (found_type == BTRFS_INODE_ITEM_KEY)
3467                         break;
3468
3469                 if (path->slots[0] == 0 ||
3470                     path->slots[0] != pending_del_slot) {
3471                         if (root->ref_cows) {
3472                                 err = -EAGAIN;
3473                                 goto out;
3474                         }
3475                         if (pending_del_nr) {
3476                                 ret = btrfs_del_items(trans, root, path,
3477                                                 pending_del_slot,
3478                                                 pending_del_nr);
3479                                 BUG_ON(ret);
3480                                 pending_del_nr = 0;
3481                         }
3482                         btrfs_release_path(root, path);
3483                         goto search_again;
3484                 } else {
3485                         path->slots[0]--;
3486                 }
3487         }
3488 out:
3489         if (pending_del_nr) {
3490                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3491                                       pending_del_nr);
3492                 BUG_ON(ret);
3493         }
3494         btrfs_free_path(path);
3495         return err;
3496 }
3497
3498 /*
3499  * taken from block_truncate_page, but does cow as it zeros out
3500  * any bytes left in the last page in the file.
3501  */
3502 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3503 {
3504         struct inode *inode = mapping->host;
3505         struct btrfs_root *root = BTRFS_I(inode)->root;
3506         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3507         struct btrfs_ordered_extent *ordered;
3508         struct extent_state *cached_state = NULL;
3509         char *kaddr;
3510         u32 blocksize = root->sectorsize;
3511         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3512         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3513         struct page *page;
3514         int ret = 0;
3515         u64 page_start;
3516         u64 page_end;
3517
3518         if ((offset & (blocksize - 1)) == 0)
3519                 goto out;
3520         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3521         if (ret)
3522                 goto out;
3523
3524         ret = -ENOMEM;
3525 again:
3526         page = grab_cache_page(mapping, index);
3527         if (!page) {
3528                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3529                 goto out;
3530         }
3531
3532         page_start = page_offset(page);
3533         page_end = page_start + PAGE_CACHE_SIZE - 1;
3534
3535         if (!PageUptodate(page)) {
3536                 ret = btrfs_readpage(NULL, page);
3537                 lock_page(page);
3538                 if (page->mapping != mapping) {
3539                         unlock_page(page);
3540                         page_cache_release(page);
3541                         goto again;
3542                 }
3543                 if (!PageUptodate(page)) {
3544                         ret = -EIO;
3545                         goto out_unlock;
3546                 }
3547         }
3548         wait_on_page_writeback(page);
3549
3550         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3551                          GFP_NOFS);
3552         set_page_extent_mapped(page);
3553
3554         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3555         if (ordered) {
3556                 unlock_extent_cached(io_tree, page_start, page_end,
3557                                      &cached_state, GFP_NOFS);
3558                 unlock_page(page);
3559                 page_cache_release(page);
3560                 btrfs_start_ordered_extent(inode, ordered, 1);
3561                 btrfs_put_ordered_extent(ordered);
3562                 goto again;
3563         }
3564
3565         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3566                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3567                           0, 0, &cached_state, GFP_NOFS);
3568
3569         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3570                                         &cached_state);
3571         if (ret) {
3572                 unlock_extent_cached(io_tree, page_start, page_end,
3573                                      &cached_state, GFP_NOFS);
3574                 goto out_unlock;
3575         }
3576
3577         ret = 0;
3578         if (offset != PAGE_CACHE_SIZE) {
3579                 kaddr = kmap(page);
3580                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3581                 flush_dcache_page(page);
3582                 kunmap(page);
3583         }
3584         ClearPageChecked(page);
3585         set_page_dirty(page);
3586         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3587                              GFP_NOFS);
3588
3589 out_unlock:
3590         if (ret)
3591                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3592         unlock_page(page);
3593         page_cache_release(page);
3594 out:
3595         return ret;
3596 }
3597
3598 /*
3599  * This function puts in dummy file extents for the area we're creating a hole
3600  * for.  So if we are truncating this file to a larger size we need to insert
3601  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3602  * the range between oldsize and size
3603  */
3604 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3605 {
3606         struct btrfs_trans_handle *trans;
3607         struct btrfs_root *root = BTRFS_I(inode)->root;
3608         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3609         struct extent_map *em = NULL;
3610         struct extent_state *cached_state = NULL;
3611         u64 mask = root->sectorsize - 1;
3612         u64 hole_start = (oldsize + mask) & ~mask;
3613         u64 block_end = (size + mask) & ~mask;
3614         u64 last_byte;
3615         u64 cur_offset;
3616         u64 hole_size;
3617         int err = 0;
3618
3619         if (size <= hole_start)
3620                 return 0;
3621
3622         while (1) {
3623                 struct btrfs_ordered_extent *ordered;
3624                 btrfs_wait_ordered_range(inode, hole_start,
3625                                          block_end - hole_start);
3626                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3627                                  &cached_state, GFP_NOFS);
3628                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3629                 if (!ordered)
3630                         break;
3631                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3632                                      &cached_state, GFP_NOFS);
3633                 btrfs_put_ordered_extent(ordered);
3634         }
3635
3636         cur_offset = hole_start;
3637         while (1) {
3638                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3639                                 block_end - cur_offset, 0);
3640                 BUG_ON(IS_ERR(em) || !em);
3641                 last_byte = min(extent_map_end(em), block_end);
3642                 last_byte = (last_byte + mask) & ~mask;
3643                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3644                         u64 hint_byte = 0;
3645                         hole_size = last_byte - cur_offset;
3646
3647                         trans = btrfs_start_transaction(root, 2);
3648                         if (IS_ERR(trans)) {
3649                                 err = PTR_ERR(trans);
3650                                 break;
3651                         }
3652                         btrfs_set_trans_block_group(trans, inode);
3653
3654                         err = btrfs_drop_extents(trans, inode, cur_offset,
3655                                                  cur_offset + hole_size,
3656                                                  &hint_byte, 1);
3657                         if (err)
3658                                 break;
3659
3660                         err = btrfs_insert_file_extent(trans, root,
3661                                         inode->i_ino, cur_offset, 0,
3662                                         0, hole_size, 0, hole_size,
3663                                         0, 0, 0);
3664                         if (err)
3665                                 break;
3666
3667                         btrfs_drop_extent_cache(inode, hole_start,
3668                                         last_byte - 1, 0);
3669
3670                         btrfs_end_transaction(trans, root);
3671                 }
3672                 free_extent_map(em);
3673                 em = NULL;
3674                 cur_offset = last_byte;
3675                 if (cur_offset >= block_end)
3676                         break;
3677         }
3678
3679         free_extent_map(em);
3680         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3681                              GFP_NOFS);
3682         return err;
3683 }
3684
3685 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3686 {
3687         loff_t oldsize = i_size_read(inode);
3688         int ret;
3689
3690         if (newsize == oldsize)
3691                 return 0;
3692
3693         if (newsize > oldsize) {
3694                 i_size_write(inode, newsize);
3695                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3696                 truncate_pagecache(inode, oldsize, newsize);
3697                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3698                 if (ret) {
3699                         btrfs_setsize(inode, oldsize);
3700                         return ret;
3701                 }
3702
3703                 mark_inode_dirty(inode);
3704         } else {
3705
3706                 /*
3707                  * We're truncating a file that used to have good data down to
3708                  * zero. Make sure it gets into the ordered flush list so that
3709                  * any new writes get down to disk quickly.
3710                  */
3711                 if (newsize == 0)
3712                         BTRFS_I(inode)->ordered_data_close = 1;
3713
3714                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3715                 truncate_setsize(inode, newsize);
3716                 ret = btrfs_truncate(inode);
3717         }
3718
3719         return ret;
3720 }
3721
3722 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3723 {
3724         struct inode *inode = dentry->d_inode;
3725         struct btrfs_root *root = BTRFS_I(inode)->root;
3726         int err;
3727
3728         if (btrfs_root_readonly(root))
3729                 return -EROFS;
3730
3731         err = inode_change_ok(inode, attr);
3732         if (err)
3733                 return err;
3734
3735         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3736                 err = btrfs_setsize(inode, attr->ia_size);
3737                 if (err)
3738                         return err;
3739         }
3740
3741         if (attr->ia_valid) {
3742                 setattr_copy(inode, attr);
3743                 mark_inode_dirty(inode);
3744
3745                 if (attr->ia_valid & ATTR_MODE)
3746                         err = btrfs_acl_chmod(inode);
3747         }
3748
3749         return err;
3750 }
3751
3752 void btrfs_evict_inode(struct inode *inode)
3753 {
3754         struct btrfs_trans_handle *trans;
3755         struct btrfs_root *root = BTRFS_I(inode)->root;
3756         unsigned long nr;
3757         int ret;
3758
3759         trace_btrfs_inode_evict(inode);
3760
3761         truncate_inode_pages(&inode->i_data, 0);
3762         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3763                                root == root->fs_info->tree_root))
3764                 goto no_delete;
3765
3766         if (is_bad_inode(inode)) {
3767                 btrfs_orphan_del(NULL, inode);
3768                 goto no_delete;
3769         }
3770         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3771         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3772
3773         if (root->fs_info->log_root_recovering) {
3774                 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3775                 goto no_delete;
3776         }
3777
3778         if (inode->i_nlink > 0) {
3779                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3780                 goto no_delete;
3781         }
3782
3783         btrfs_i_size_write(inode, 0);
3784
3785         while (1) {
3786                 trans = btrfs_start_transaction(root, 0);
3787                 BUG_ON(IS_ERR(trans));
3788                 btrfs_set_trans_block_group(trans, inode);
3789                 trans->block_rsv = root->orphan_block_rsv;
3790
3791                 ret = btrfs_block_rsv_check(trans, root,
3792                                             root->orphan_block_rsv, 0, 5);
3793                 if (ret) {
3794                         BUG_ON(ret != -EAGAIN);
3795                         ret = btrfs_commit_transaction(trans, root);
3796                         BUG_ON(ret);
3797                         continue;
3798                 }
3799
3800                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3801                 if (ret != -EAGAIN)
3802                         break;
3803
3804                 nr = trans->blocks_used;
3805                 btrfs_end_transaction(trans, root);
3806                 trans = NULL;
3807                 btrfs_btree_balance_dirty(root, nr);
3808
3809         }
3810
3811         if (ret == 0) {
3812                 ret = btrfs_orphan_del(trans, inode);
3813                 BUG_ON(ret);
3814         }
3815
3816         nr = trans->blocks_used;
3817         btrfs_end_transaction(trans, root);
3818         btrfs_btree_balance_dirty(root, nr);
3819 no_delete:
3820         end_writeback(inode);
3821         return;
3822 }
3823
3824 /*
3825  * this returns the key found in the dir entry in the location pointer.
3826  * If no dir entries were found, location->objectid is 0.
3827  */
3828 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3829                                struct btrfs_key *location)
3830 {
3831         const char *name = dentry->d_name.name;
3832         int namelen = dentry->d_name.len;
3833         struct btrfs_dir_item *di;
3834         struct btrfs_path *path;
3835         struct btrfs_root *root = BTRFS_I(dir)->root;
3836         int ret = 0;
3837
3838         path = btrfs_alloc_path();
3839         BUG_ON(!path);
3840
3841         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3842                                     namelen, 0);
3843         if (IS_ERR(di))
3844                 ret = PTR_ERR(di);
3845
3846         if (!di || IS_ERR(di))
3847                 goto out_err;
3848
3849         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3850 out:
3851         btrfs_free_path(path);
3852         return ret;
3853 out_err:
3854         location->objectid = 0;
3855         goto out;
3856 }
3857
3858 /*
3859  * when we hit a tree root in a directory, the btrfs part of the inode
3860  * needs to be changed to reflect the root directory of the tree root.  This
3861  * is kind of like crossing a mount point.
3862  */
3863 static int fixup_tree_root_location(struct btrfs_root *root,
3864                                     struct inode *dir,
3865                                     struct dentry *dentry,
3866                                     struct btrfs_key *location,
3867                                     struct btrfs_root **sub_root)
3868 {
3869         struct btrfs_path *path;
3870         struct btrfs_root *new_root;
3871         struct btrfs_root_ref *ref;
3872         struct extent_buffer *leaf;
3873         int ret;
3874         int err = 0;
3875
3876         path = btrfs_alloc_path();
3877         if (!path) {
3878                 err = -ENOMEM;
3879                 goto out;
3880         }
3881
3882         err = -ENOENT;
3883         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3884                                   BTRFS_I(dir)->root->root_key.objectid,
3885                                   location->objectid);
3886         if (ret) {
3887                 if (ret < 0)
3888                         err = ret;
3889                 goto out;
3890         }
3891
3892         leaf = path->nodes[0];
3893         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3894         if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino ||
3895             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3896                 goto out;
3897
3898         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3899                                    (unsigned long)(ref + 1),
3900                                    dentry->d_name.len);
3901         if (ret)
3902                 goto out;
3903
3904         btrfs_release_path(root->fs_info->tree_root, path);
3905
3906         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3907         if (IS_ERR(new_root)) {
3908                 err = PTR_ERR(new_root);
3909                 goto out;
3910         }
3911
3912         if (btrfs_root_refs(&new_root->root_item) == 0) {
3913                 err = -ENOENT;
3914                 goto out;
3915         }
3916
3917         *sub_root = new_root;
3918         location->objectid = btrfs_root_dirid(&new_root->root_item);
3919         location->type = BTRFS_INODE_ITEM_KEY;
3920         location->offset = 0;
3921         err = 0;
3922 out:
3923         btrfs_free_path(path);
3924         return err;
3925 }
3926
3927 static void inode_tree_add(struct inode *inode)
3928 {
3929         struct btrfs_root *root = BTRFS_I(inode)->root;
3930         struct btrfs_inode *entry;
3931         struct rb_node **p;
3932         struct rb_node *parent;
3933 again:
3934         p = &root->inode_tree.rb_node;
3935         parent = NULL;
3936
3937         if (inode_unhashed(inode))
3938                 return;
3939
3940         spin_lock(&root->inode_lock);
3941         while (*p) {
3942                 parent = *p;
3943                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3944
3945                 if (inode->i_ino < entry->vfs_inode.i_ino)
3946                         p = &parent->rb_left;
3947                 else if (inode->i_ino > entry->vfs_inode.i_ino)
3948                         p = &parent->rb_right;
3949                 else {
3950                         WARN_ON(!(entry->vfs_inode.i_state &
3951                                   (I_WILL_FREE | I_FREEING)));
3952                         rb_erase(parent, &root->inode_tree);
3953                         RB_CLEAR_NODE(parent);
3954                         spin_unlock(&root->inode_lock);
3955                         goto again;
3956                 }
3957         }
3958         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3959         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3960         spin_unlock(&root->inode_lock);
3961 }
3962
3963 static void inode_tree_del(struct inode *inode)
3964 {
3965         struct btrfs_root *root = BTRFS_I(inode)->root;
3966         int empty = 0;
3967
3968         spin_lock(&root->inode_lock);
3969         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3970                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3971                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3972                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3973         }
3974         spin_unlock(&root->inode_lock);
3975
3976         /*
3977          * Free space cache has inodes in the tree root, but the tree root has a
3978          * root_refs of 0, so this could end up dropping the tree root as a
3979          * snapshot, so we need the extra !root->fs_info->tree_root check to
3980          * make sure we don't drop it.
3981          */
3982         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3983             root != root->fs_info->tree_root) {
3984                 synchronize_srcu(&root->fs_info->subvol_srcu);
3985                 spin_lock(&root->inode_lock);
3986                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3987                 spin_unlock(&root->inode_lock);
3988                 if (empty)
3989                         btrfs_add_dead_root(root);
3990         }
3991 }
3992
3993 int btrfs_invalidate_inodes(struct btrfs_root *root)
3994 {
3995         struct rb_node *node;
3996         struct rb_node *prev;
3997         struct btrfs_inode *entry;
3998         struct inode *inode;
3999         u64 objectid = 0;
4000
4001         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4002
4003         spin_lock(&root->inode_lock);
4004 again:
4005         node = root->inode_tree.rb_node;
4006         prev = NULL;
4007         while (node) {
4008                 prev = node;
4009                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4010
4011                 if (objectid < entry->vfs_inode.i_ino)
4012                         node = node->rb_left;
4013                 else if (objectid > entry->vfs_inode.i_ino)
4014                         node = node->rb_right;
4015                 else
4016                         break;
4017         }
4018         if (!node) {
4019                 while (prev) {
4020                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
4021                         if (objectid <= entry->vfs_inode.i_ino) {
4022                                 node = prev;
4023                                 break;
4024                         }
4025                         prev = rb_next(prev);
4026                 }
4027         }
4028         while (node) {
4029                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4030                 objectid = entry->vfs_inode.i_ino + 1;
4031                 inode = igrab(&entry->vfs_inode);
4032                 if (inode) {
4033                         spin_unlock(&root->inode_lock);
4034                         if (atomic_read(&inode->i_count) > 1)
4035                                 d_prune_aliases(inode);
4036                         /*
4037                          * btrfs_drop_inode will have it removed from
4038                          * the inode cache when its usage count
4039                          * hits zero.
4040                          */
4041                         iput(inode);
4042                         cond_resched();
4043                         spin_lock(&root->inode_lock);
4044                         goto again;
4045                 }
4046
4047                 if (cond_resched_lock(&root->inode_lock))
4048                         goto again;
4049
4050                 node = rb_next(node);
4051         }
4052         spin_unlock(&root->inode_lock);
4053         return 0;
4054 }
4055
4056 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4057 {
4058         struct btrfs_iget_args *args = p;
4059         inode->i_ino = args->ino;
4060         BTRFS_I(inode)->root = args->root;
4061         btrfs_set_inode_space_info(args->root, inode);
4062         return 0;
4063 }
4064
4065 static int btrfs_find_actor(struct inode *inode, void *opaque)
4066 {
4067         struct btrfs_iget_args *args = opaque;
4068         return args->ino == inode->i_ino &&
4069                 args->root == BTRFS_I(inode)->root;
4070 }
4071
4072 static struct inode *btrfs_iget_locked(struct super_block *s,
4073                                        u64 objectid,
4074                                        struct btrfs_root *root)
4075 {
4076         struct inode *inode;
4077         struct btrfs_iget_args args;
4078         args.ino = objectid;
4079         args.root = root;
4080
4081         inode = iget5_locked(s, objectid, btrfs_find_actor,
4082                              btrfs_init_locked_inode,
4083                              (void *)&args);
4084         return inode;
4085 }
4086
4087 /* Get an inode object given its location and corresponding root.
4088  * Returns in *is_new if the inode was read from disk
4089  */
4090 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4091                          struct btrfs_root *root, int *new)
4092 {
4093         struct inode *inode;
4094
4095         inode = btrfs_iget_locked(s, location->objectid, root);
4096         if (!inode)
4097                 return ERR_PTR(-ENOMEM);
4098
4099         if (inode->i_state & I_NEW) {
4100                 BTRFS_I(inode)->root = root;
4101                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4102                 btrfs_read_locked_inode(inode);
4103                 inode_tree_add(inode);
4104                 unlock_new_inode(inode);
4105                 if (new)
4106                         *new = 1;
4107         }
4108
4109         return inode;
4110 }
4111
4112 static struct inode *new_simple_dir(struct super_block *s,
4113                                     struct btrfs_key *key,
4114                                     struct btrfs_root *root)
4115 {
4116         struct inode *inode = new_inode(s);
4117
4118         if (!inode)
4119                 return ERR_PTR(-ENOMEM);
4120
4121         BTRFS_I(inode)->root = root;
4122         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4123         BTRFS_I(inode)->dummy_inode = 1;
4124
4125         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4126         inode->i_op = &simple_dir_inode_operations;
4127         inode->i_fop = &simple_dir_operations;
4128         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4129         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4130
4131         return inode;
4132 }
4133
4134 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4135 {
4136         struct inode *inode;
4137         struct btrfs_root *root = BTRFS_I(dir)->root;
4138         struct btrfs_root *sub_root = root;
4139         struct btrfs_key location;
4140         int index;
4141         int ret;
4142
4143         if (dentry->d_name.len > BTRFS_NAME_LEN)
4144                 return ERR_PTR(-ENAMETOOLONG);
4145
4146         ret = btrfs_inode_by_name(dir, dentry, &location);
4147
4148         if (ret < 0)
4149                 return ERR_PTR(ret);
4150
4151         if (location.objectid == 0)
4152                 return NULL;
4153
4154         if (location.type == BTRFS_INODE_ITEM_KEY) {
4155                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4156                 return inode;
4157         }
4158
4159         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4160
4161         index = srcu_read_lock(&root->fs_info->subvol_srcu);
4162         ret = fixup_tree_root_location(root, dir, dentry,
4163                                        &location, &sub_root);
4164         if (ret < 0) {
4165                 if (ret != -ENOENT)
4166                         inode = ERR_PTR(ret);
4167                 else
4168                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
4169         } else {
4170                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4171         }
4172         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4173
4174         if (!IS_ERR(inode) && root != sub_root) {
4175                 down_read(&root->fs_info->cleanup_work_sem);
4176                 if (!(inode->i_sb->s_flags & MS_RDONLY))
4177                         ret = btrfs_orphan_cleanup(sub_root);
4178                 up_read(&root->fs_info->cleanup_work_sem);
4179                 if (ret)
4180                         inode = ERR_PTR(ret);
4181         }
4182
4183         return inode;
4184 }
4185
4186 static int btrfs_dentry_delete(const struct dentry *dentry)
4187 {
4188         struct btrfs_root *root;
4189
4190         if (!dentry->d_inode && !IS_ROOT(dentry))
4191                 dentry = dentry->d_parent;
4192
4193         if (dentry->d_inode) {
4194                 root = BTRFS_I(dentry->d_inode)->root;
4195                 if (btrfs_root_refs(&root->root_item) == 0)
4196                         return 1;
4197         }
4198         return 0;
4199 }
4200
4201 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4202                                    struct nameidata *nd)
4203 {
4204         struct inode *inode;
4205
4206         inode = btrfs_lookup_dentry(dir, dentry);
4207         if (IS_ERR(inode))
4208                 return ERR_CAST(inode);
4209
4210         return d_splice_alias(inode, dentry);
4211 }
4212
4213 static unsigned char btrfs_filetype_table[] = {
4214         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4215 };
4216
4217 static int btrfs_real_readdir(struct file *filp, void *dirent,
4218                               filldir_t filldir)
4219 {
4220         struct inode *inode = filp->f_dentry->d_inode;
4221         struct btrfs_root *root = BTRFS_I(inode)->root;
4222         struct btrfs_item *item;
4223         struct btrfs_dir_item *di;
4224         struct btrfs_key key;
4225         struct btrfs_key found_key;
4226         struct btrfs_path *path;
4227         int ret;
4228         struct extent_buffer *leaf;
4229         int slot;
4230         unsigned char d_type;
4231         int over = 0;
4232         u32 di_cur;
4233         u32 di_total;
4234         u32 di_len;
4235         int key_type = BTRFS_DIR_INDEX_KEY;
4236         char tmp_name[32];
4237         char *name_ptr;
4238         int name_len;
4239
4240         /* FIXME, use a real flag for deciding about the key type */
4241         if (root->fs_info->tree_root == root)
4242                 key_type = BTRFS_DIR_ITEM_KEY;
4243
4244         /* special case for "." */
4245         if (filp->f_pos == 0) {
4246                 over = filldir(dirent, ".", 1,
4247                                1, inode->i_ino,
4248                                DT_DIR);
4249                 if (over)
4250                         return 0;
4251                 filp->f_pos = 1;
4252         }
4253         /* special case for .., just use the back ref */
4254         if (filp->f_pos == 1) {
4255                 u64 pino = parent_ino(filp->f_path.dentry);
4256                 over = filldir(dirent, "..", 2,
4257                                2, pino, DT_DIR);
4258                 if (over)
4259                         return 0;
4260                 filp->f_pos = 2;
4261         }
4262         path = btrfs_alloc_path();
4263         path->reada = 2;
4264
4265         btrfs_set_key_type(&key, key_type);
4266         key.offset = filp->f_pos;
4267         key.objectid = inode->i_ino;
4268
4269         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4270         if (ret < 0)
4271                 goto err;
4272
4273         while (1) {
4274                 leaf = path->nodes[0];
4275                 slot = path->slots[0];
4276                 if (slot >= btrfs_header_nritems(leaf)) {
4277                         ret = btrfs_next_leaf(root, path);
4278                         if (ret < 0)
4279                                 goto err;
4280                         else if (ret > 0)
4281                                 break;
4282                         continue;
4283                 }
4284
4285                 item = btrfs_item_nr(leaf, slot);
4286                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4287
4288                 if (found_key.objectid != key.objectid)
4289                         break;
4290                 if (btrfs_key_type(&found_key) != key_type)
4291                         break;
4292                 if (found_key.offset < filp->f_pos)
4293                         goto next;
4294
4295                 filp->f_pos = found_key.offset;
4296
4297                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4298                 di_cur = 0;
4299                 di_total = btrfs_item_size(leaf, item);
4300
4301                 while (di_cur < di_total) {
4302                         struct btrfs_key location;
4303
4304                         if (verify_dir_item(root, leaf, di))
4305                                 break;
4306
4307                         name_len = btrfs_dir_name_len(leaf, di);
4308                         if (name_len <= sizeof(tmp_name)) {
4309                                 name_ptr = tmp_name;
4310                         } else {
4311                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4312                                 if (!name_ptr) {
4313                                         ret = -ENOMEM;
4314                                         goto err;
4315                                 }
4316                         }
4317                         read_extent_buffer(leaf, name_ptr,
4318                                            (unsigned long)(di + 1), name_len);
4319
4320                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4321                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4322
4323                         /* is this a reference to our own snapshot? If so
4324                          * skip it
4325                          */
4326                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4327                             location.objectid == root->root_key.objectid) {
4328                                 over = 0;
4329                                 goto skip;
4330                         }
4331                         over = filldir(dirent, name_ptr, name_len,
4332                                        found_key.offset, location.objectid,
4333                                        d_type);
4334
4335 skip:
4336                         if (name_ptr != tmp_name)
4337                                 kfree(name_ptr);
4338
4339                         if (over)
4340                                 goto nopos;
4341                         di_len = btrfs_dir_name_len(leaf, di) +
4342                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4343                         di_cur += di_len;
4344                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4345                 }
4346 next:
4347                 path->slots[0]++;
4348         }
4349
4350         /* Reached end of directory/root. Bump pos past the last item. */
4351         if (key_type == BTRFS_DIR_INDEX_KEY)
4352                 /*
4353                  * 32-bit glibc will use getdents64, but then strtol -
4354                  * so the last number we can serve is this.
4355                  */
4356                 filp->f_pos = 0x7fffffff;
4357         else
4358                 filp->f_pos++;
4359 nopos:
4360         ret = 0;
4361 err:
4362         btrfs_free_path(path);
4363         return ret;
4364 }
4365
4366 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4367 {
4368         struct btrfs_root *root = BTRFS_I(inode)->root;
4369         struct btrfs_trans_handle *trans;
4370         int ret = 0;
4371         bool nolock = false;
4372
4373         if (BTRFS_I(inode)->dummy_inode)
4374                 return 0;
4375
4376         smp_mb();
4377         nolock = (root->fs_info->closing && root == root->fs_info->tree_root);
4378
4379         if (wbc->sync_mode == WB_SYNC_ALL) {
4380                 if (nolock)
4381                         trans = btrfs_join_transaction_nolock(root);
4382                 else
4383                         trans = btrfs_join_transaction(root);
4384                 if (IS_ERR(trans))
4385                         return PTR_ERR(trans);
4386                 btrfs_set_trans_block_group(trans, inode);
4387                 if (nolock)
4388                         ret = btrfs_end_transaction_nolock(trans, root);
4389                 else
4390                         ret = btrfs_commit_transaction(trans, root);
4391         }
4392         return ret;
4393 }
4394
4395 /*
4396  * This is somewhat expensive, updating the tree every time the
4397  * inode changes.  But, it is most likely to find the inode in cache.
4398  * FIXME, needs more benchmarking...there are no reasons other than performance
4399  * to keep or drop this code.
4400  */
4401 void btrfs_dirty_inode(struct inode *inode)
4402 {
4403         struct btrfs_root *root = BTRFS_I(inode)->root;
4404         struct btrfs_trans_handle *trans;
4405         int ret;
4406
4407         if (BTRFS_I(inode)->dummy_inode)
4408                 return;
4409
4410         trans = btrfs_join_transaction(root);
4411         BUG_ON(IS_ERR(trans));
4412         btrfs_set_trans_block_group(trans, inode);
4413
4414         ret = btrfs_update_inode(trans, root, inode);
4415         if (ret && ret == -ENOSPC) {
4416                 /* whoops, lets try again with the full transaction */
4417                 btrfs_end_transaction(trans, root);
4418                 trans = btrfs_start_transaction(root, 1);
4419                 if (IS_ERR(trans)) {
4420                         if (printk_ratelimit()) {
4421                                 printk(KERN_ERR "btrfs: fail to "
4422                                        "dirty  inode %lu error %ld\n",
4423                                        inode->i_ino, PTR_ERR(trans));
4424                         }
4425                         return;
4426                 }
4427                 btrfs_set_trans_block_group(trans, inode);
4428
4429                 ret = btrfs_update_inode(trans, root, inode);
4430                 if (ret) {
4431                         if (printk_ratelimit()) {
4432                                 printk(KERN_ERR "btrfs: fail to "
4433                                        "dirty  inode %lu error %d\n",
4434                                        inode->i_ino, ret);
4435                         }
4436                 }
4437         }
4438         btrfs_end_transaction(trans, root);
4439 }
4440
4441 /*
4442  * find the highest existing sequence number in a directory
4443  * and then set the in-memory index_cnt variable to reflect
4444  * free sequence numbers
4445  */
4446 static int btrfs_set_inode_index_count(struct inode *inode)
4447 {
4448         struct btrfs_root *root = BTRFS_I(inode)->root;
4449         struct btrfs_key key, found_key;
4450         struct btrfs_path *path;
4451         struct extent_buffer *leaf;
4452         int ret;
4453
4454         key.objectid = inode->i_ino;
4455         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4456         key.offset = (u64)-1;
4457
4458         path = btrfs_alloc_path();
4459         if (!path)
4460                 return -ENOMEM;
4461
4462         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4463         if (ret < 0)
4464                 goto out;
4465         /* FIXME: we should be able to handle this */
4466         if (ret == 0)
4467                 goto out;
4468         ret = 0;
4469
4470         /*
4471          * MAGIC NUMBER EXPLANATION:
4472          * since we search a directory based on f_pos we have to start at 2
4473          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4474          * else has to start at 2
4475          */
4476         if (path->slots[0] == 0) {
4477                 BTRFS_I(inode)->index_cnt = 2;
4478                 goto out;
4479         }
4480
4481         path->slots[0]--;
4482
4483         leaf = path->nodes[0];
4484         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4485
4486         if (found_key.objectid != inode->i_ino ||
4487             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4488                 BTRFS_I(inode)->index_cnt = 2;
4489                 goto out;
4490         }
4491
4492         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4493 out:
4494         btrfs_free_path(path);
4495         return ret;
4496 }
4497
4498 /*
4499  * helper to find a free sequence number in a given directory.  This current
4500  * code is very simple, later versions will do smarter things in the btree
4501  */
4502 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4503 {
4504         int ret = 0;
4505
4506         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4507                 ret = btrfs_set_inode_index_count(dir);
4508                 if (ret)
4509                         return ret;
4510         }
4511
4512         *index = BTRFS_I(dir)->index_cnt;
4513         BTRFS_I(dir)->index_cnt++;
4514
4515         return ret;
4516 }
4517
4518 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4519                                      struct btrfs_root *root,
4520                                      struct inode *dir,
4521                                      const char *name, int name_len,
4522                                      u64 ref_objectid, u64 objectid,
4523                                      u64 alloc_hint, int mode, u64 *index)
4524 {
4525         struct inode *inode;
4526         struct btrfs_inode_item *inode_item;
4527         struct btrfs_key *location;
4528         struct btrfs_path *path;
4529         struct btrfs_inode_ref *ref;
4530         struct btrfs_key key[2];
4531         u32 sizes[2];
4532         unsigned long ptr;
4533         int ret;
4534         int owner;
4535
4536         path = btrfs_alloc_path();
4537         BUG_ON(!path);
4538
4539         inode = new_inode(root->fs_info->sb);
4540         if (!inode) {
4541                 btrfs_free_path(path);
4542                 return ERR_PTR(-ENOMEM);
4543         }
4544
4545         if (dir) {
4546                 trace_btrfs_inode_request(dir);
4547
4548                 ret = btrfs_set_inode_index(dir, index);
4549                 if (ret) {
4550                         btrfs_free_path(path);
4551                         iput(inode);
4552                         return ERR_PTR(ret);
4553                 }
4554         }
4555         /*
4556          * index_cnt is ignored for everything but a dir,
4557          * btrfs_get_inode_index_count has an explanation for the magic
4558          * number
4559          */
4560         BTRFS_I(inode)->index_cnt = 2;
4561         BTRFS_I(inode)->root = root;
4562         BTRFS_I(inode)->generation = trans->transid;
4563         inode->i_generation = BTRFS_I(inode)->generation;
4564         btrfs_set_inode_space_info(root, inode);
4565
4566         if (mode & S_IFDIR)
4567                 owner = 0;
4568         else
4569                 owner = 1;
4570         BTRFS_I(inode)->block_group =
4571                         btrfs_find_block_group(root, 0, alloc_hint, owner);
4572
4573         key[0].objectid = objectid;
4574         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4575         key[0].offset = 0;
4576
4577         key[1].objectid = objectid;
4578         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4579         key[1].offset = ref_objectid;
4580
4581         sizes[0] = sizeof(struct btrfs_inode_item);
4582         sizes[1] = name_len + sizeof(*ref);
4583
4584         path->leave_spinning = 1;
4585         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4586         if (ret != 0)
4587                 goto fail;
4588
4589         inode_init_owner(inode, dir, mode);
4590         inode->i_ino = objectid;
4591         inode_set_bytes(inode, 0);
4592         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4593         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4594                                   struct btrfs_inode_item);
4595         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4596
4597         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4598                              struct btrfs_inode_ref);
4599         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4600         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4601         ptr = (unsigned long)(ref + 1);
4602         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4603
4604         btrfs_mark_buffer_dirty(path->nodes[0]);
4605         btrfs_free_path(path);
4606
4607         location = &BTRFS_I(inode)->location;
4608         location->objectid = objectid;
4609         location->offset = 0;
4610         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4611
4612         btrfs_inherit_iflags(inode, dir);
4613
4614         if ((mode & S_IFREG)) {
4615                 if (btrfs_test_opt(root, NODATASUM))
4616                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4617                 if (btrfs_test_opt(root, NODATACOW) ||
4618                     (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4619                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4620         }
4621
4622         insert_inode_hash(inode);
4623         inode_tree_add(inode);
4624
4625         trace_btrfs_inode_new(inode);
4626
4627         return inode;
4628 fail:
4629         if (dir)
4630                 BTRFS_I(dir)->index_cnt--;
4631         btrfs_free_path(path);
4632         iput(inode);
4633         return ERR_PTR(ret);
4634 }
4635
4636 static inline u8 btrfs_inode_type(struct inode *inode)
4637 {
4638         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4639 }
4640
4641 /*
4642  * utility function to add 'inode' into 'parent_inode' with
4643  * a give name and a given sequence number.
4644  * if 'add_backref' is true, also insert a backref from the
4645  * inode to the parent directory.
4646  */
4647 int btrfs_add_link(struct btrfs_trans_handle *trans,
4648                    struct inode *parent_inode, struct inode *inode,
4649                    const char *name, int name_len, int add_backref, u64 index)
4650 {
4651         int ret = 0;
4652         struct btrfs_key key;
4653         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4654
4655         if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4656                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4657         } else {
4658                 key.objectid = inode->i_ino;
4659                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4660                 key.offset = 0;
4661         }
4662
4663         if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4664                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4665                                          key.objectid, root->root_key.objectid,
4666                                          parent_inode->i_ino,
4667                                          index, name, name_len);
4668         } else if (add_backref) {
4669                 ret = btrfs_insert_inode_ref(trans, root,
4670                                              name, name_len, inode->i_ino,
4671                                              parent_inode->i_ino, index);
4672         }
4673
4674         if (ret == 0) {
4675                 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4676                                             parent_inode->i_ino, &key,
4677                                             btrfs_inode_type(inode), index);
4678                 BUG_ON(ret);
4679
4680                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4681                                    name_len * 2);
4682                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4683                 ret = btrfs_update_inode(trans, root, parent_inode);
4684         }
4685         return ret;
4686 }
4687
4688 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4689                             struct inode *dir, struct dentry *dentry,
4690                             struct inode *inode, int backref, u64 index)
4691 {
4692         int err = btrfs_add_link(trans, dir, inode,
4693                                  dentry->d_name.name, dentry->d_name.len,
4694                                  backref, index);
4695         if (!err) {
4696                 d_instantiate(dentry, inode);
4697                 return 0;
4698         }
4699         if (err > 0)
4700                 err = -EEXIST;
4701         return err;
4702 }
4703
4704 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4705                         int mode, dev_t rdev)
4706 {
4707         struct btrfs_trans_handle *trans;
4708         struct btrfs_root *root = BTRFS_I(dir)->root;
4709         struct inode *inode = NULL;
4710         int err;
4711         int drop_inode = 0;
4712         u64 objectid;
4713         unsigned long nr = 0;
4714         u64 index = 0;
4715
4716         if (!new_valid_dev(rdev))
4717                 return -EINVAL;
4718
4719         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4720         if (err)
4721                 return err;
4722
4723         /*
4724          * 2 for inode item and ref
4725          * 2 for dir items
4726          * 1 for xattr if selinux is on
4727          */
4728         trans = btrfs_start_transaction(root, 5);
4729         if (IS_ERR(trans))
4730                 return PTR_ERR(trans);
4731
4732         btrfs_set_trans_block_group(trans, dir);
4733
4734         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4735                                 dentry->d_name.len, dir->i_ino, objectid,
4736                                 BTRFS_I(dir)->block_group, mode, &index);
4737         if (IS_ERR(inode)) {
4738                 err = PTR_ERR(inode);
4739                 goto out_unlock;
4740         }
4741
4742         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4743         if (err) {
4744                 drop_inode = 1;
4745                 goto out_unlock;
4746         }
4747
4748         btrfs_set_trans_block_group(trans, inode);
4749         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4750         if (err)
4751                 drop_inode = 1;
4752         else {
4753                 inode->i_op = &btrfs_special_inode_operations;
4754                 init_special_inode(inode, inode->i_mode, rdev);
4755                 btrfs_update_inode(trans, root, inode);
4756         }
4757         btrfs_update_inode_block_group(trans, inode);
4758         btrfs_update_inode_block_group(trans, dir);
4759 out_unlock:
4760         nr = trans->blocks_used;
4761         btrfs_end_transaction_throttle(trans, root);
4762         btrfs_btree_balance_dirty(root, nr);
4763         if (drop_inode) {
4764                 inode_dec_link_count(inode);
4765                 iput(inode);
4766         }
4767         return err;
4768 }
4769
4770 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4771                         int mode, struct nameidata *nd)
4772 {
4773         struct btrfs_trans_handle *trans;
4774         struct btrfs_root *root = BTRFS_I(dir)->root;
4775         struct inode *inode = NULL;
4776         int drop_inode = 0;
4777         int err;
4778         unsigned long nr = 0;
4779         u64 objectid;
4780         u64 index = 0;
4781
4782         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4783         if (err)
4784                 return err;
4785         /*
4786          * 2 for inode item and ref
4787          * 2 for dir items
4788          * 1 for xattr if selinux is on
4789          */
4790         trans = btrfs_start_transaction(root, 5);
4791         if (IS_ERR(trans))
4792                 return PTR_ERR(trans);
4793
4794         btrfs_set_trans_block_group(trans, dir);
4795
4796         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4797                                 dentry->d_name.len, dir->i_ino, objectid,
4798                                 BTRFS_I(dir)->block_group, mode, &index);
4799         if (IS_ERR(inode)) {
4800                 err = PTR_ERR(inode);
4801                 goto out_unlock;
4802         }
4803
4804         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4805         if (err) {
4806                 drop_inode = 1;
4807                 goto out_unlock;
4808         }
4809
4810         btrfs_set_trans_block_group(trans, inode);
4811         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4812         if (err)
4813                 drop_inode = 1;
4814         else {
4815                 inode->i_mapping->a_ops = &btrfs_aops;
4816                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4817                 inode->i_fop = &btrfs_file_operations;
4818                 inode->i_op = &btrfs_file_inode_operations;
4819                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4820         }
4821         btrfs_update_inode_block_group(trans, inode);
4822         btrfs_update_inode_block_group(trans, dir);
4823 out_unlock:
4824         nr = trans->blocks_used;
4825         btrfs_end_transaction_throttle(trans, root);
4826         if (drop_inode) {
4827                 inode_dec_link_count(inode);
4828                 iput(inode);
4829         }
4830         btrfs_btree_balance_dirty(root, nr);
4831         return err;
4832 }
4833
4834 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4835                       struct dentry *dentry)
4836 {
4837         struct btrfs_trans_handle *trans;
4838         struct btrfs_root *root = BTRFS_I(dir)->root;
4839         struct inode *inode = old_dentry->d_inode;
4840         u64 index;
4841         unsigned long nr = 0;
4842         int err;
4843         int drop_inode = 0;
4844
4845         /* do not allow sys_link's with other subvols of the same device */
4846         if (root->objectid != BTRFS_I(inode)->root->objectid)
4847                 return -EXDEV;
4848
4849         if (inode->i_nlink == ~0U)
4850                 return -EMLINK;
4851
4852         err = btrfs_set_inode_index(dir, &index);
4853         if (err)
4854                 goto fail;
4855
4856         /*
4857          * 2 items for inode and inode ref
4858          * 2 items for dir items
4859          * 1 item for parent inode
4860          */
4861         trans = btrfs_start_transaction(root, 5);
4862         if (IS_ERR(trans)) {
4863                 err = PTR_ERR(trans);
4864                 goto fail;
4865         }
4866
4867         btrfs_inc_nlink(inode);
4868         inode->i_ctime = CURRENT_TIME;
4869
4870         btrfs_set_trans_block_group(trans, dir);
4871         ihold(inode);
4872
4873         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4874
4875         if (err) {
4876                 drop_inode = 1;
4877         } else {
4878                 struct dentry *parent = dget_parent(dentry);
4879                 btrfs_update_inode_block_group(trans, dir);
4880                 err = btrfs_update_inode(trans, root, inode);
4881                 BUG_ON(err);
4882                 btrfs_log_new_name(trans, inode, NULL, parent);
4883                 dput(parent);
4884         }
4885
4886         nr = trans->blocks_used;
4887         btrfs_end_transaction_throttle(trans, root);
4888 fail:
4889         if (drop_inode) {
4890                 inode_dec_link_count(inode);
4891                 iput(inode);
4892         }
4893         btrfs_btree_balance_dirty(root, nr);
4894         return err;
4895 }
4896
4897 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4898 {
4899         struct inode *inode = NULL;
4900         struct btrfs_trans_handle *trans;
4901         struct btrfs_root *root = BTRFS_I(dir)->root;
4902         int err = 0;
4903         int drop_on_err = 0;
4904         u64 objectid = 0;
4905         u64 index = 0;
4906         unsigned long nr = 1;
4907
4908         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4909         if (err)
4910                 return err;
4911
4912         /*
4913          * 2 items for inode and ref
4914          * 2 items for dir items
4915          * 1 for xattr if selinux is on
4916          */
4917         trans = btrfs_start_transaction(root, 5);
4918         if (IS_ERR(trans))
4919                 return PTR_ERR(trans);
4920         btrfs_set_trans_block_group(trans, dir);
4921
4922         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4923                                 dentry->d_name.len, dir->i_ino, objectid,
4924                                 BTRFS_I(dir)->block_group, S_IFDIR | mode,
4925                                 &index);
4926         if (IS_ERR(inode)) {
4927                 err = PTR_ERR(inode);
4928                 goto out_fail;
4929         }
4930
4931         drop_on_err = 1;
4932
4933         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4934         if (err)
4935                 goto out_fail;
4936
4937         inode->i_op = &btrfs_dir_inode_operations;
4938         inode->i_fop = &btrfs_dir_file_operations;
4939         btrfs_set_trans_block_group(trans, inode);
4940
4941         btrfs_i_size_write(inode, 0);
4942         err = btrfs_update_inode(trans, root, inode);
4943         if (err)
4944                 goto out_fail;
4945
4946         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4947                              dentry->d_name.len, 0, index);
4948         if (err)
4949                 goto out_fail;
4950
4951         d_instantiate(dentry, inode);
4952         drop_on_err = 0;
4953         btrfs_update_inode_block_group(trans, inode);
4954         btrfs_update_inode_block_group(trans, dir);
4955
4956 out_fail:
4957         nr = trans->blocks_used;
4958         btrfs_end_transaction_throttle(trans, root);
4959         if (drop_on_err)
4960                 iput(inode);
4961         btrfs_btree_balance_dirty(root, nr);
4962         return err;
4963 }
4964
4965 /* helper for btfs_get_extent.  Given an existing extent in the tree,
4966  * and an extent that you want to insert, deal with overlap and insert
4967  * the new extent into the tree.
4968  */
4969 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4970                                 struct extent_map *existing,
4971                                 struct extent_map *em,
4972                                 u64 map_start, u64 map_len)
4973 {
4974         u64 start_diff;
4975
4976         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4977         start_diff = map_start - em->start;
4978         em->start = map_start;
4979         em->len = map_len;
4980         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4981             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4982                 em->block_start += start_diff;
4983                 em->block_len -= start_diff;
4984         }
4985         return add_extent_mapping(em_tree, em);
4986 }
4987
4988 static noinline int uncompress_inline(struct btrfs_path *path,
4989                                       struct inode *inode, struct page *page,
4990                                       size_t pg_offset, u64 extent_offset,
4991                                       struct btrfs_file_extent_item *item)
4992 {
4993         int ret;
4994         struct extent_buffer *leaf = path->nodes[0];
4995         char *tmp;
4996         size_t max_size;
4997         unsigned long inline_size;
4998         unsigned long ptr;
4999         int compress_type;
5000
5001         WARN_ON(pg_offset != 0);
5002         compress_type = btrfs_file_extent_compression(leaf, item);
5003         max_size = btrfs_file_extent_ram_bytes(leaf, item);
5004         inline_size = btrfs_file_extent_inline_item_len(leaf,
5005                                         btrfs_item_nr(leaf, path->slots[0]));
5006         tmp = kmalloc(inline_size, GFP_NOFS);
5007         if (!tmp)
5008                 return -ENOMEM;
5009         ptr = btrfs_file_extent_inline_start(item);
5010
5011         read_extent_buffer(leaf, tmp, ptr, inline_size);
5012
5013         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5014         ret = btrfs_decompress(compress_type, tmp, page,
5015                                extent_offset, inline_size, max_size);
5016         if (ret) {
5017                 char *kaddr = kmap_atomic(page, KM_USER0);
5018                 unsigned long copy_size = min_t(u64,
5019                                   PAGE_CACHE_SIZE - pg_offset,
5020                                   max_size - extent_offset);
5021                 memset(kaddr + pg_offset, 0, copy_size);
5022                 kunmap_atomic(kaddr, KM_USER0);
5023         }
5024         kfree(tmp);
5025         return 0;
5026 }
5027
5028 /*
5029  * a bit scary, this does extent mapping from logical file offset to the disk.
5030  * the ugly parts come from merging extents from the disk with the in-ram
5031  * representation.  This gets more complex because of the data=ordered code,
5032  * where the in-ram extents might be locked pending data=ordered completion.
5033  *
5034  * This also copies inline extents directly into the page.
5035  */
5036
5037 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5038                                     size_t pg_offset, u64 start, u64 len,
5039                                     int create)
5040 {
5041         int ret;
5042         int err = 0;
5043         u64 bytenr;
5044         u64 extent_start = 0;
5045         u64 extent_end = 0;
5046         u64 objectid = inode->i_ino;
5047         u32 found_type;
5048         struct btrfs_path *path = NULL;
5049         struct btrfs_root *root = BTRFS_I(inode)->root;
5050         struct btrfs_file_extent_item *item;
5051         struct extent_buffer *leaf;
5052         struct btrfs_key found_key;
5053         struct extent_map *em = NULL;
5054         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5055         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5056         struct btrfs_trans_handle *trans = NULL;
5057         int compress_type;
5058
5059 again:
5060         read_lock(&em_tree->lock);
5061         em = lookup_extent_mapping(em_tree, start, len);
5062         if (em)
5063                 em->bdev = root->fs_info->fs_devices->latest_bdev;
5064         read_unlock(&em_tree->lock);
5065
5066         if (em) {
5067                 if (em->start > start || em->start + em->len <= start)
5068                         free_extent_map(em);
5069                 else if (em->block_start == EXTENT_MAP_INLINE && page)
5070                         free_extent_map(em);
5071                 else
5072                         goto out;
5073         }
5074         em = alloc_extent_map(GFP_NOFS);
5075         if (!em) {
5076                 err = -ENOMEM;
5077                 goto out;
5078         }
5079         em->bdev = root->fs_info->fs_devices->latest_bdev;
5080         em->start = EXTENT_MAP_HOLE;
5081         em->orig_start = EXTENT_MAP_HOLE;
5082         em->len = (u64)-1;
5083         em->block_len = (u64)-1;
5084
5085         if (!path) {
5086                 path = btrfs_alloc_path();
5087                 BUG_ON(!path);
5088         }
5089
5090         ret = btrfs_lookup_file_extent(trans, root, path,
5091                                        objectid, start, trans != NULL);
5092         if (ret < 0) {
5093                 err = ret;
5094                 goto out;
5095         }
5096
5097         if (ret != 0) {
5098                 if (path->slots[0] == 0)
5099                         goto not_found;
5100                 path->slots[0]--;
5101         }
5102
5103         leaf = path->nodes[0];
5104         item = btrfs_item_ptr(leaf, path->slots[0],
5105                               struct btrfs_file_extent_item);
5106         /* are we inside the extent that was found? */
5107         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5108         found_type = btrfs_key_type(&found_key);
5109         if (found_key.objectid != objectid ||
5110             found_type != BTRFS_EXTENT_DATA_KEY) {
5111                 goto not_found;
5112         }
5113
5114         found_type = btrfs_file_extent_type(leaf, item);
5115         extent_start = found_key.offset;
5116         compress_type = btrfs_file_extent_compression(leaf, item);
5117         if (found_type == BTRFS_FILE_EXTENT_REG ||
5118             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5119                 extent_end = extent_start +
5120                        btrfs_file_extent_num_bytes(leaf, item);
5121         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5122                 size_t size;
5123                 size = btrfs_file_extent_inline_len(leaf, item);
5124                 extent_end = (extent_start + size + root->sectorsize - 1) &
5125                         ~((u64)root->sectorsize - 1);
5126         }
5127
5128         if (start >= extent_end) {
5129                 path->slots[0]++;
5130                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5131                         ret = btrfs_next_leaf(root, path);
5132                         if (ret < 0) {
5133                                 err = ret;
5134                                 goto out;
5135                         }
5136                         if (ret > 0)
5137                                 goto not_found;
5138                         leaf = path->nodes[0];
5139                 }
5140                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5141                 if (found_key.objectid != objectid ||
5142                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5143                         goto not_found;
5144                 if (start + len <= found_key.offset)
5145                         goto not_found;
5146                 em->start = start;
5147                 em->len = found_key.offset - start;
5148                 goto not_found_em;
5149         }
5150
5151         if (found_type == BTRFS_FILE_EXTENT_REG ||
5152             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5153                 em->start = extent_start;
5154                 em->len = extent_end - extent_start;
5155                 em->orig_start = extent_start -
5156                                  btrfs_file_extent_offset(leaf, item);
5157                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5158                 if (bytenr == 0) {
5159                         em->block_start = EXTENT_MAP_HOLE;
5160                         goto insert;
5161                 }
5162                 if (compress_type != BTRFS_COMPRESS_NONE) {
5163                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5164                         em->compress_type = compress_type;
5165                         em->block_start = bytenr;
5166                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5167                                                                          item);
5168                 } else {
5169                         bytenr += btrfs_file_extent_offset(leaf, item);
5170                         em->block_start = bytenr;
5171                         em->block_len = em->len;
5172                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5173                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5174                 }
5175                 goto insert;
5176         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5177                 unsigned long ptr;
5178                 char *map;
5179                 size_t size;
5180                 size_t extent_offset;
5181                 size_t copy_size;
5182
5183                 em->block_start = EXTENT_MAP_INLINE;
5184                 if (!page || create) {
5185                         em->start = extent_start;
5186                         em->len = extent_end - extent_start;
5187                         goto out;
5188                 }
5189
5190                 size = btrfs_file_extent_inline_len(leaf, item);
5191                 extent_offset = page_offset(page) + pg_offset - extent_start;
5192                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5193                                 size - extent_offset);
5194                 em->start = extent_start + extent_offset;
5195                 em->len = (copy_size + root->sectorsize - 1) &
5196                         ~((u64)root->sectorsize - 1);
5197                 em->orig_start = EXTENT_MAP_INLINE;
5198                 if (compress_type) {
5199                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5200                         em->compress_type = compress_type;
5201                 }
5202                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5203                 if (create == 0 && !PageUptodate(page)) {
5204                         if (btrfs_file_extent_compression(leaf, item) !=
5205                             BTRFS_COMPRESS_NONE) {
5206                                 ret = uncompress_inline(path, inode, page,
5207                                                         pg_offset,
5208                                                         extent_offset, item);
5209                                 BUG_ON(ret);
5210                         } else {
5211                                 map = kmap(page);
5212                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5213                                                    copy_size);
5214                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5215                                         memset(map + pg_offset + copy_size, 0,
5216                                                PAGE_CACHE_SIZE - pg_offset -
5217                                                copy_size);
5218                                 }
5219                                 kunmap(page);
5220                         }
5221                         flush_dcache_page(page);
5222                 } else if (create && PageUptodate(page)) {
5223                         WARN_ON(1);
5224                         if (!trans) {
5225                                 kunmap(page);
5226                                 free_extent_map(em);
5227                                 em = NULL;
5228                                 btrfs_release_path(root, path);
5229                                 trans = btrfs_join_transaction(root);
5230                                 if (IS_ERR(trans))
5231                                         return ERR_CAST(trans);
5232                                 goto again;
5233                         }
5234                         map = kmap(page);
5235                         write_extent_buffer(leaf, map + pg_offset, ptr,
5236                                             copy_size);
5237                         kunmap(page);
5238                         btrfs_mark_buffer_dirty(leaf);
5239                 }
5240                 set_extent_uptodate(io_tree, em->start,
5241                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
5242                 goto insert;
5243         } else {
5244                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5245                 WARN_ON(1);
5246         }
5247 not_found:
5248         em->start = start;
5249         em->len = len;
5250 not_found_em:
5251         em->block_start = EXTENT_MAP_HOLE;
5252         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5253 insert:
5254         btrfs_release_path(root, path);
5255         if (em->start > start || extent_map_end(em) <= start) {
5256                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5257                        "[%llu %llu]\n", (unsigned long long)em->start,
5258                        (unsigned long long)em->len,
5259                        (unsigned long long)start,
5260                        (unsigned long long)len);
5261                 err = -EIO;
5262                 goto out;
5263         }
5264
5265         err = 0;
5266         write_lock(&em_tree->lock);
5267         ret = add_extent_mapping(em_tree, em);
5268         /* it is possible that someone inserted the extent into the tree
5269          * while we had the lock dropped.  It is also possible that
5270          * an overlapping map exists in the tree
5271          */
5272         if (ret == -EEXIST) {
5273                 struct extent_map *existing;
5274
5275                 ret = 0;
5276
5277                 existing = lookup_extent_mapping(em_tree, start, len);
5278                 if (existing && (existing->start > start ||
5279                     existing->start + existing->len <= start)) {
5280                         free_extent_map(existing);
5281                         existing = NULL;
5282                 }
5283                 if (!existing) {
5284                         existing = lookup_extent_mapping(em_tree, em->start,
5285                                                          em->len);
5286                         if (existing) {
5287                                 err = merge_extent_mapping(em_tree, existing,
5288                                                            em, start,
5289                                                            root->sectorsize);
5290                                 free_extent_map(existing);
5291                                 if (err) {
5292                                         free_extent_map(em);
5293                                         em = NULL;
5294                                 }
5295                         } else {
5296                                 err = -EIO;
5297                                 free_extent_map(em);
5298                                 em = NULL;
5299                         }
5300                 } else {
5301                         free_extent_map(em);
5302                         em = existing;
5303                         err = 0;
5304                 }
5305         }
5306         write_unlock(&em_tree->lock);
5307 out:
5308
5309         trace_btrfs_get_extent(root, em);
5310
5311         if (path)
5312                 btrfs_free_path(path);
5313         if (trans) {
5314                 ret = btrfs_end_transaction(trans, root);
5315                 if (!err)
5316                         err = ret;
5317         }
5318         if (err) {
5319                 free_extent_map(em);
5320                 return ERR_PTR(err);
5321         }
5322         return em;
5323 }
5324
5325 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5326                                            size_t pg_offset, u64 start, u64 len,
5327                                            int create)
5328 {
5329         struct extent_map *em;
5330         struct extent_map *hole_em = NULL;
5331         u64 range_start = start;
5332         u64 end;
5333         u64 found;
5334         u64 found_end;
5335         int err = 0;
5336
5337         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5338         if (IS_ERR(em))
5339                 return em;
5340         if (em) {
5341                 /*
5342                  * if our em maps to a hole, there might
5343                  * actually be delalloc bytes behind it
5344                  */
5345                 if (em->block_start != EXTENT_MAP_HOLE)
5346                         return em;
5347                 else
5348                         hole_em = em;
5349         }
5350
5351         /* check to see if we've wrapped (len == -1 or similar) */
5352         end = start + len;
5353         if (end < start)
5354                 end = (u64)-1;
5355         else
5356                 end -= 1;
5357
5358         em = NULL;
5359
5360         /* ok, we didn't find anything, lets look for delalloc */
5361         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5362                                  end, len, EXTENT_DELALLOC, 1);
5363         found_end = range_start + found;
5364         if (found_end < range_start)
5365                 found_end = (u64)-1;
5366
5367         /*
5368          * we didn't find anything useful, return
5369          * the original results from get_extent()
5370          */
5371         if (range_start > end || found_end <= start) {
5372                 em = hole_em;
5373                 hole_em = NULL;
5374                 goto out;
5375         }
5376
5377         /* adjust the range_start to make sure it doesn't
5378          * go backwards from the start they passed in
5379          */
5380         range_start = max(start,range_start);
5381         found = found_end - range_start;
5382
5383         if (found > 0) {
5384                 u64 hole_start = start;
5385                 u64 hole_len = len;
5386
5387                 em = alloc_extent_map(GFP_NOFS);
5388                 if (!em) {
5389                         err = -ENOMEM;
5390                         goto out;
5391                 }
5392                 /*
5393                  * when btrfs_get_extent can't find anything it
5394                  * returns one huge hole
5395                  *
5396                  * make sure what it found really fits our range, and
5397                  * adjust to make sure it is based on the start from
5398                  * the caller
5399                  */
5400                 if (hole_em) {
5401                         u64 calc_end = extent_map_end(hole_em);
5402
5403                         if (calc_end <= start || (hole_em->start > end)) {
5404                                 free_extent_map(hole_em);
5405                                 hole_em = NULL;
5406                         } else {
5407                                 hole_start = max(hole_em->start, start);
5408                                 hole_len = calc_end - hole_start;
5409                         }
5410                 }
5411                 em->bdev = NULL;
5412                 if (hole_em && range_start > hole_start) {
5413                         /* our hole starts before our delalloc, so we
5414                          * have to return just the parts of the hole
5415                          * that go until  the delalloc starts
5416                          */
5417                         em->len = min(hole_len,
5418                                       range_start - hole_start);
5419                         em->start = hole_start;
5420                         em->orig_start = hole_start;
5421                         /*
5422                          * don't adjust block start at all,
5423                          * it is fixed at EXTENT_MAP_HOLE
5424                          */
5425                         em->block_start = hole_em->block_start;
5426                         em->block_len = hole_len;
5427                 } else {
5428                         em->start = range_start;
5429                         em->len = found;
5430                         em->orig_start = range_start;
5431                         em->block_start = EXTENT_MAP_DELALLOC;
5432                         em->block_len = found;
5433                 }
5434         } else if (hole_em) {
5435                 return hole_em;
5436         }
5437 out:
5438
5439         free_extent_map(hole_em);
5440         if (err) {
5441                 free_extent_map(em);
5442                 return ERR_PTR(err);
5443         }
5444         return em;
5445 }
5446
5447 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5448                                                   struct extent_map *em,
5449                                                   u64 start, u64 len)
5450 {
5451         struct btrfs_root *root = BTRFS_I(inode)->root;
5452         struct btrfs_trans_handle *trans;
5453         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5454         struct btrfs_key ins;
5455         u64 alloc_hint;
5456         int ret;
5457         bool insert = false;
5458
5459         /*
5460          * Ok if the extent map we looked up is a hole and is for the exact
5461          * range we want, there is no reason to allocate a new one, however if
5462          * it is not right then we need to free this one and drop the cache for
5463          * our range.
5464          */
5465         if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5466             em->len != len) {
5467                 free_extent_map(em);
5468                 em = NULL;
5469                 insert = true;
5470                 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5471         }
5472
5473         trans = btrfs_join_transaction(root);
5474         if (IS_ERR(trans))
5475                 return ERR_CAST(trans);
5476
5477         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5478
5479         alloc_hint = get_extent_allocation_hint(inode, start, len);
5480         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5481                                    alloc_hint, (u64)-1, &ins, 1);
5482         if (ret) {
5483                 em = ERR_PTR(ret);
5484                 goto out;
5485         }
5486
5487         if (!em) {
5488                 em = alloc_extent_map(GFP_NOFS);
5489                 if (!em) {
5490                         em = ERR_PTR(-ENOMEM);
5491                         goto out;
5492                 }
5493         }
5494
5495         em->start = start;
5496         em->orig_start = em->start;
5497         em->len = ins.offset;
5498
5499         em->block_start = ins.objectid;
5500         em->block_len = ins.offset;
5501         em->bdev = root->fs_info->fs_devices->latest_bdev;
5502
5503         /*
5504          * We need to do this because if we're using the original em we searched
5505          * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5506          */
5507         em->flags = 0;
5508         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5509
5510         while (insert) {
5511                 write_lock(&em_tree->lock);
5512                 ret = add_extent_mapping(em_tree, em);
5513                 write_unlock(&em_tree->lock);
5514                 if (ret != -EEXIST)
5515                         break;
5516                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5517         }
5518
5519         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5520                                            ins.offset, ins.offset, 0);
5521         if (ret) {
5522                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5523                 em = ERR_PTR(ret);
5524         }
5525 out:
5526         btrfs_end_transaction(trans, root);
5527         return em;
5528 }
5529
5530 /*
5531  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5532  * block must be cow'd
5533  */
5534 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5535                                       struct inode *inode, u64 offset, u64 len)
5536 {
5537         struct btrfs_path *path;
5538         int ret;
5539         struct extent_buffer *leaf;
5540         struct btrfs_root *root = BTRFS_I(inode)->root;
5541         struct btrfs_file_extent_item *fi;
5542         struct btrfs_key key;
5543         u64 disk_bytenr;
5544         u64 backref_offset;
5545         u64 extent_end;
5546         u64 num_bytes;
5547         int slot;
5548         int found_type;
5549
5550         path = btrfs_alloc_path();
5551         if (!path)
5552                 return -ENOMEM;
5553
5554         ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
5555                                        offset, 0);
5556         if (ret < 0)
5557                 goto out;
5558
5559         slot = path->slots[0];
5560         if (ret == 1) {
5561                 if (slot == 0) {
5562                         /* can't find the item, must cow */
5563                         ret = 0;
5564                         goto out;
5565                 }
5566                 slot--;
5567         }
5568         ret = 0;
5569         leaf = path->nodes[0];
5570         btrfs_item_key_to_cpu(leaf, &key, slot);
5571         if (key.objectid != inode->i_ino ||
5572             key.type != BTRFS_EXTENT_DATA_KEY) {
5573                 /* not our file or wrong item type, must cow */
5574                 goto out;
5575         }
5576
5577         if (key.offset > offset) {
5578                 /* Wrong offset, must cow */
5579                 goto out;
5580         }
5581
5582         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5583         found_type = btrfs_file_extent_type(leaf, fi);
5584         if (found_type != BTRFS_FILE_EXTENT_REG &&
5585             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5586                 /* not a regular extent, must cow */
5587                 goto out;
5588         }
5589         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5590         backref_offset = btrfs_file_extent_offset(leaf, fi);
5591
5592         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5593         if (extent_end < offset + len) {
5594                 /* extent doesn't include our full range, must cow */
5595                 goto out;
5596         }
5597
5598         if (btrfs_extent_readonly(root, disk_bytenr))
5599                 goto out;
5600
5601         /*
5602          * look for other files referencing this extent, if we
5603          * find any we must cow
5604          */
5605         if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
5606                                   key.offset - backref_offset, disk_bytenr))
5607                 goto out;
5608
5609         /*
5610          * adjust disk_bytenr and num_bytes to cover just the bytes
5611          * in this extent we are about to write.  If there
5612          * are any csums in that range we have to cow in order
5613          * to keep the csums correct
5614          */
5615         disk_bytenr += backref_offset;
5616         disk_bytenr += offset - key.offset;
5617         num_bytes = min(offset + len, extent_end) - offset;
5618         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5619                                 goto out;
5620         /*
5621          * all of the above have passed, it is safe to overwrite this extent
5622          * without cow
5623          */
5624         ret = 1;
5625 out:
5626         btrfs_free_path(path);
5627         return ret;
5628 }
5629
5630 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5631                                    struct buffer_head *bh_result, int create)
5632 {
5633         struct extent_map *em;
5634         struct btrfs_root *root = BTRFS_I(inode)->root;
5635         u64 start = iblock << inode->i_blkbits;
5636         u64 len = bh_result->b_size;
5637         struct btrfs_trans_handle *trans;
5638
5639         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5640         if (IS_ERR(em))
5641                 return PTR_ERR(em);
5642
5643         /*
5644          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5645          * io.  INLINE is special, and we could probably kludge it in here, but
5646          * it's still buffered so for safety lets just fall back to the generic
5647          * buffered path.
5648          *
5649          * For COMPRESSED we _have_ to read the entire extent in so we can
5650          * decompress it, so there will be buffering required no matter what we
5651          * do, so go ahead and fallback to buffered.
5652          *
5653          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5654          * to buffered IO.  Don't blame me, this is the price we pay for using
5655          * the generic code.
5656          */
5657         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5658             em->block_start == EXTENT_MAP_INLINE) {
5659                 free_extent_map(em);
5660                 return -ENOTBLK;
5661         }
5662
5663         /* Just a good old fashioned hole, return */
5664         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5665                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5666                 free_extent_map(em);
5667                 /* DIO will do one hole at a time, so just unlock a sector */
5668                 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5669                               start + root->sectorsize - 1, GFP_NOFS);
5670                 return 0;
5671         }
5672
5673         /*
5674          * We don't allocate a new extent in the following cases
5675          *
5676          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
5677          * existing extent.
5678          * 2) The extent is marked as PREALLOC.  We're good to go here and can
5679          * just use the extent.
5680          *
5681          */
5682         if (!create) {
5683                 len = em->len - (start - em->start);
5684                 goto map;
5685         }
5686
5687         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5688             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5689              em->block_start != EXTENT_MAP_HOLE)) {
5690                 int type;
5691                 int ret;
5692                 u64 block_start;
5693
5694                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5695                         type = BTRFS_ORDERED_PREALLOC;
5696                 else
5697                         type = BTRFS_ORDERED_NOCOW;
5698                 len = min(len, em->len - (start - em->start));
5699                 block_start = em->block_start + (start - em->start);
5700
5701                 /*
5702                  * we're not going to log anything, but we do need
5703                  * to make sure the current transaction stays open
5704                  * while we look for nocow cross refs
5705                  */
5706                 trans = btrfs_join_transaction(root);
5707                 if (IS_ERR(trans))
5708                         goto must_cow;
5709
5710                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5711                         ret = btrfs_add_ordered_extent_dio(inode, start,
5712                                            block_start, len, len, type);
5713                         btrfs_end_transaction(trans, root);
5714                         if (ret) {
5715                                 free_extent_map(em);
5716                                 return ret;
5717                         }
5718                         goto unlock;
5719                 }
5720                 btrfs_end_transaction(trans, root);
5721         }
5722 must_cow:
5723         /*
5724          * this will cow the extent, reset the len in case we changed
5725          * it above
5726          */
5727         len = bh_result->b_size;
5728         em = btrfs_new_extent_direct(inode, em, start, len);
5729         if (IS_ERR(em))
5730                 return PTR_ERR(em);
5731         len = min(len, em->len - (start - em->start));
5732 unlock:
5733         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5734                           EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5735                           0, NULL, GFP_NOFS);
5736 map:
5737         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5738                 inode->i_blkbits;
5739         bh_result->b_size = len;
5740         bh_result->b_bdev = em->bdev;
5741         set_buffer_mapped(bh_result);
5742         if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5743                 set_buffer_new(bh_result);
5744
5745         free_extent_map(em);
5746
5747         return 0;
5748 }
5749
5750 struct btrfs_dio_private {
5751         struct inode *inode;
5752         u64 logical_offset;
5753         u64 disk_bytenr;
5754         u64 bytes;
5755         u32 *csums;
5756         void *private;
5757
5758         /* number of bios pending for this dio */
5759         atomic_t pending_bios;
5760
5761         /* IO errors */
5762         int errors;
5763
5764         struct bio *orig_bio;
5765 };
5766
5767 static void btrfs_endio_direct_read(struct bio *bio, int err)
5768 {
5769         struct btrfs_dio_private *dip = bio->bi_private;
5770         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5771         struct bio_vec *bvec = bio->bi_io_vec;
5772         struct inode *inode = dip->inode;
5773         struct btrfs_root *root = BTRFS_I(inode)->root;
5774         u64 start;
5775         u32 *private = dip->csums;
5776
5777         start = dip->logical_offset;
5778         do {
5779                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5780                         struct page *page = bvec->bv_page;
5781                         char *kaddr;
5782                         u32 csum = ~(u32)0;
5783                         unsigned long flags;
5784
5785                         local_irq_save(flags);
5786                         kaddr = kmap_atomic(page, KM_IRQ0);
5787                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5788                                                csum, bvec->bv_len);
5789                         btrfs_csum_final(csum, (char *)&csum);
5790                         kunmap_atomic(kaddr, KM_IRQ0);
5791                         local_irq_restore(flags);
5792
5793                         flush_dcache_page(bvec->bv_page);
5794                         if (csum != *private) {
5795                                 printk(KERN_ERR "btrfs csum failed ino %lu off"
5796                                       " %llu csum %u private %u\n",
5797                                       inode->i_ino, (unsigned long long)start,
5798                                       csum, *private);
5799                                 err = -EIO;
5800                         }
5801                 }
5802
5803                 start += bvec->bv_len;
5804                 private++;
5805                 bvec++;
5806         } while (bvec <= bvec_end);
5807
5808         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5809                       dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5810         bio->bi_private = dip->private;
5811
5812         kfree(dip->csums);
5813         kfree(dip);
5814
5815         /* If we had a csum failure make sure to clear the uptodate flag */
5816         if (err)
5817                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5818         dio_end_io(bio, err);
5819 }
5820
5821 static void btrfs_endio_direct_write(struct bio *bio, int err)
5822 {
5823         struct btrfs_dio_private *dip = bio->bi_private;
5824         struct inode *inode = dip->inode;
5825         struct btrfs_root *root = BTRFS_I(inode)->root;
5826         struct btrfs_trans_handle *trans;
5827         struct btrfs_ordered_extent *ordered = NULL;
5828         struct extent_state *cached_state = NULL;
5829         u64 ordered_offset = dip->logical_offset;
5830         u64 ordered_bytes = dip->bytes;
5831         int ret;
5832
5833         if (err)
5834                 goto out_done;
5835 again:
5836         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5837                                                    &ordered_offset,
5838                                                    ordered_bytes);
5839         if (!ret)
5840                 goto out_test;
5841
5842         BUG_ON(!ordered);
5843
5844         trans = btrfs_join_transaction(root);
5845         if (IS_ERR(trans)) {
5846                 err = -ENOMEM;
5847                 goto out;
5848         }
5849         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5850
5851         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5852                 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5853                 if (!ret)
5854                         ret = btrfs_update_inode(trans, root, inode);
5855                 err = ret;
5856                 goto out;
5857         }
5858
5859         lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5860                          ordered->file_offset + ordered->len - 1, 0,
5861                          &cached_state, GFP_NOFS);
5862
5863         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5864                 ret = btrfs_mark_extent_written(trans, inode,
5865                                                 ordered->file_offset,
5866                                                 ordered->file_offset +
5867                                                 ordered->len);
5868                 if (ret) {
5869                         err = ret;
5870                         goto out_unlock;
5871                 }
5872         } else {
5873                 ret = insert_reserved_file_extent(trans, inode,
5874                                                   ordered->file_offset,
5875                                                   ordered->start,
5876                                                   ordered->disk_len,
5877                                                   ordered->len,
5878                                                   ordered->len,
5879                                                   0, 0, 0,
5880                                                   BTRFS_FILE_EXTENT_REG);
5881                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5882                                    ordered->file_offset, ordered->len);
5883                 if (ret) {
5884                         err = ret;
5885                         WARN_ON(1);
5886                         goto out_unlock;
5887                 }
5888         }
5889
5890         add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5891         ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5892         if (!ret)
5893                 btrfs_update_inode(trans, root, inode);
5894         ret = 0;
5895 out_unlock:
5896         unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5897                              ordered->file_offset + ordered->len - 1,
5898                              &cached_state, GFP_NOFS);
5899 out:
5900         btrfs_delalloc_release_metadata(inode, ordered->len);
5901         btrfs_end_transaction(trans, root);
5902         ordered_offset = ordered->file_offset + ordered->len;
5903         btrfs_put_ordered_extent(ordered);
5904         btrfs_put_ordered_extent(ordered);
5905
5906 out_test:
5907         /*
5908          * our bio might span multiple ordered extents.  If we haven't
5909          * completed the accounting for the whole dio, go back and try again
5910          */
5911         if (ordered_offset < dip->logical_offset + dip->bytes) {
5912                 ordered_bytes = dip->logical_offset + dip->bytes -
5913                         ordered_offset;
5914                 goto again;
5915         }
5916 out_done:
5917         bio->bi_private = dip->private;
5918
5919         kfree(dip->csums);
5920         kfree(dip);
5921
5922         /* If we had an error make sure to clear the uptodate flag */
5923         if (err)
5924                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5925         dio_end_io(bio, err);
5926 }
5927
5928 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5929                                     struct bio *bio, int mirror_num,
5930                                     unsigned long bio_flags, u64 offset)
5931 {
5932         int ret;
5933         struct btrfs_root *root = BTRFS_I(inode)->root;
5934         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5935         BUG_ON(ret);
5936         return 0;
5937 }
5938
5939 static void btrfs_end_dio_bio(struct bio *bio, int err)
5940 {
5941         struct btrfs_dio_private *dip = bio->bi_private;
5942
5943         if (err) {
5944                 printk(KERN_ERR "btrfs direct IO failed ino %lu rw %lu "
5945                       "sector %#Lx len %u err no %d\n",
5946                       dip->inode->i_ino, bio->bi_rw,
5947                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
5948                 dip->errors = 1;
5949
5950                 /*
5951                  * before atomic variable goto zero, we must make sure
5952                  * dip->errors is perceived to be set.
5953                  */
5954                 smp_mb__before_atomic_dec();
5955         }
5956
5957         /* if there are more bios still pending for this dio, just exit */
5958         if (!atomic_dec_and_test(&dip->pending_bios))
5959                 goto out;
5960
5961         if (dip->errors)
5962                 bio_io_error(dip->orig_bio);
5963         else {
5964                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5965                 bio_endio(dip->orig_bio, 0);
5966         }
5967 out:
5968         bio_put(bio);
5969 }
5970
5971 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5972                                        u64 first_sector, gfp_t gfp_flags)
5973 {
5974         int nr_vecs = bio_get_nr_vecs(bdev);
5975         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5976 }
5977
5978 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5979                                          int rw, u64 file_offset, int skip_sum,
5980                                          u32 *csums, int async_submit)
5981 {
5982         int write = rw & REQ_WRITE;
5983         struct btrfs_root *root = BTRFS_I(inode)->root;
5984         int ret;
5985
5986         bio_get(bio);
5987         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5988         if (ret)
5989                 goto err;
5990
5991         if (skip_sum)
5992                 goto map;
5993
5994         if (write && async_submit) {
5995                 ret = btrfs_wq_submit_bio(root->fs_info,
5996                                    inode, rw, bio, 0, 0,
5997                                    file_offset,
5998                                    __btrfs_submit_bio_start_direct_io,
5999                                    __btrfs_submit_bio_done);
6000                 goto err;
6001         } else if (write) {
6002                 /*
6003                  * If we aren't doing async submit, calculate the csum of the
6004                  * bio now.
6005                  */
6006                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
6007                 if (ret)
6008                         goto err;
6009         } else if (!skip_sum) {
6010                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
6011                                           file_offset, csums);
6012                 if (ret)
6013                         goto err;
6014         }
6015
6016 map:
6017         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
6018 err:
6019         bio_put(bio);
6020         return ret;
6021 }
6022
6023 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
6024                                     int skip_sum)
6025 {
6026         struct inode *inode = dip->inode;
6027         struct btrfs_root *root = BTRFS_I(inode)->root;
6028         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6029         struct bio *bio;
6030         struct bio *orig_bio = dip->orig_bio;
6031         struct bio_vec *bvec = orig_bio->bi_io_vec;
6032         u64 start_sector = orig_bio->bi_sector;
6033         u64 file_offset = dip->logical_offset;
6034         u64 submit_len = 0;
6035         u64 map_length;
6036         int nr_pages = 0;
6037         u32 *csums = dip->csums;
6038         int ret = 0;
6039         int async_submit = 0;
6040         int write = rw & REQ_WRITE;
6041
6042         map_length = orig_bio->bi_size;
6043         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6044                               &map_length, NULL, 0);
6045         if (ret) {
6046                 bio_put(orig_bio);
6047                 return -EIO;
6048         }
6049
6050         if (map_length >= orig_bio->bi_size) {
6051                 bio = orig_bio;
6052                 goto submit;
6053         }
6054
6055         async_submit = 1;
6056         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
6057         if (!bio)
6058                 return -ENOMEM;
6059         bio->bi_private = dip;
6060         bio->bi_end_io = btrfs_end_dio_bio;
6061         atomic_inc(&dip->pending_bios);
6062
6063         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6064                 if (unlikely(map_length < submit_len + bvec->bv_len ||
6065                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6066                                  bvec->bv_offset) < bvec->bv_len)) {
6067                         /*
6068                          * inc the count before we submit the bio so
6069                          * we know the end IO handler won't happen before
6070                          * we inc the count. Otherwise, the dip might get freed
6071                          * before we're done setting it up
6072                          */
6073                         atomic_inc(&dip->pending_bios);
6074                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
6075                                                      file_offset, skip_sum,
6076                                                      csums, async_submit);
6077                         if (ret) {
6078                                 bio_put(bio);
6079                                 atomic_dec(&dip->pending_bios);
6080                                 goto out_err;
6081                         }
6082
6083                         /* Write's use the ordered csums */
6084                         if (!write && !skip_sum)
6085                                 csums = csums + nr_pages;
6086                         start_sector += submit_len >> 9;
6087                         file_offset += submit_len;
6088
6089                         submit_len = 0;
6090                         nr_pages = 0;
6091
6092                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6093                                                   start_sector, GFP_NOFS);
6094                         if (!bio)
6095                                 goto out_err;
6096                         bio->bi_private = dip;
6097                         bio->bi_end_io = btrfs_end_dio_bio;
6098
6099                         map_length = orig_bio->bi_size;
6100                         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6101                                               &map_length, NULL, 0);
6102                         if (ret) {
6103                                 bio_put(bio);
6104                                 goto out_err;
6105                         }
6106                 } else {
6107                         submit_len += bvec->bv_len;
6108                         nr_pages ++;
6109                         bvec++;
6110                 }
6111         }
6112
6113 submit:
6114         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6115                                      csums, async_submit);
6116         if (!ret)
6117                 return 0;
6118
6119         bio_put(bio);
6120 out_err:
6121         dip->errors = 1;
6122         /*
6123          * before atomic variable goto zero, we must
6124          * make sure dip->errors is perceived to be set.
6125          */
6126         smp_mb__before_atomic_dec();
6127         if (atomic_dec_and_test(&dip->pending_bios))
6128                 bio_io_error(dip->orig_bio);
6129
6130         /* bio_end_io() will handle error, so we needn't return it */
6131         return 0;
6132 }
6133
6134 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6135                                 loff_t file_offset)
6136 {
6137         struct btrfs_root *root = BTRFS_I(inode)->root;
6138         struct btrfs_dio_private *dip;
6139         struct bio_vec *bvec = bio->bi_io_vec;
6140         int skip_sum;
6141         int write = rw & REQ_WRITE;
6142         int ret = 0;
6143
6144         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6145
6146         dip = kmalloc(sizeof(*dip), GFP_NOFS);
6147         if (!dip) {
6148                 ret = -ENOMEM;
6149                 goto free_ordered;
6150         }
6151         dip->csums = NULL;
6152
6153         /* Write's use the ordered csum stuff, so we don't need dip->csums */
6154         if (!write && !skip_sum) {
6155                 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6156                 if (!dip->csums) {
6157                         kfree(dip);
6158                         ret = -ENOMEM;
6159                         goto free_ordered;
6160                 }
6161         }
6162
6163         dip->private = bio->bi_private;
6164         dip->inode = inode;
6165         dip->logical_offset = file_offset;
6166
6167         dip->bytes = 0;
6168         do {
6169                 dip->bytes += bvec->bv_len;
6170                 bvec++;
6171         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6172
6173         dip->disk_bytenr = (u64)bio->bi_sector << 9;
6174         bio->bi_private = dip;
6175         dip->errors = 0;
6176         dip->orig_bio = bio;
6177         atomic_set(&dip->pending_bios, 0);
6178
6179         if (write)
6180                 bio->bi_end_io = btrfs_endio_direct_write;
6181         else
6182                 bio->bi_end_io = btrfs_endio_direct_read;
6183
6184         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6185         if (!ret)
6186                 return;
6187 free_ordered:
6188         /*
6189          * If this is a write, we need to clean up the reserved space and kill
6190          * the ordered extent.
6191          */
6192         if (write) {
6193                 struct btrfs_ordered_extent *ordered;
6194                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6195                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6196                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6197                         btrfs_free_reserved_extent(root, ordered->start,
6198                                                    ordered->disk_len);
6199                 btrfs_put_ordered_extent(ordered);
6200                 btrfs_put_ordered_extent(ordered);
6201         }
6202         bio_endio(bio, ret);
6203 }
6204
6205 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6206                         const struct iovec *iov, loff_t offset,
6207                         unsigned long nr_segs)
6208 {
6209         int seg;
6210         int i;
6211         size_t size;
6212         unsigned long addr;
6213         unsigned blocksize_mask = root->sectorsize - 1;
6214         ssize_t retval = -EINVAL;
6215         loff_t end = offset;
6216
6217         if (offset & blocksize_mask)
6218                 goto out;
6219
6220         /* Check the memory alignment.  Blocks cannot straddle pages */
6221         for (seg = 0; seg < nr_segs; seg++) {
6222                 addr = (unsigned long)iov[seg].iov_base;
6223                 size = iov[seg].iov_len;
6224                 end += size;
6225                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6226                         goto out;
6227
6228                 /* If this is a write we don't need to check anymore */
6229                 if (rw & WRITE)
6230                         continue;
6231
6232                 /*
6233                  * Check to make sure we don't have duplicate iov_base's in this
6234                  * iovec, if so return EINVAL, otherwise we'll get csum errors
6235                  * when reading back.
6236                  */
6237                 for (i = seg + 1; i < nr_segs; i++) {
6238                         if (iov[seg].iov_base == iov[i].iov_base)
6239                                 goto out;
6240                 }
6241         }
6242         retval = 0;
6243 out:
6244         return retval;
6245 }
6246 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6247                         const struct iovec *iov, loff_t offset,
6248                         unsigned long nr_segs)
6249 {
6250         struct file *file = iocb->ki_filp;
6251         struct inode *inode = file->f_mapping->host;
6252         struct btrfs_ordered_extent *ordered;
6253         struct extent_state *cached_state = NULL;
6254         u64 lockstart, lockend;
6255         ssize_t ret;
6256         int writing = rw & WRITE;
6257         int write_bits = 0;
6258         size_t count = iov_length(iov, nr_segs);
6259
6260         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6261                             offset, nr_segs)) {
6262                 return 0;
6263         }
6264
6265         lockstart = offset;
6266         lockend = offset + count - 1;
6267
6268         if (writing) {
6269                 ret = btrfs_delalloc_reserve_space(inode, count);
6270                 if (ret)
6271                         goto out;
6272         }
6273
6274         while (1) {
6275                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6276                                  0, &cached_state, GFP_NOFS);
6277                 /*
6278                  * We're concerned with the entire range that we're going to be
6279                  * doing DIO to, so we need to make sure theres no ordered
6280                  * extents in this range.
6281                  */
6282                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6283                                                      lockend - lockstart + 1);
6284                 if (!ordered)
6285                         break;
6286                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6287                                      &cached_state, GFP_NOFS);
6288                 btrfs_start_ordered_extent(inode, ordered, 1);
6289                 btrfs_put_ordered_extent(ordered);
6290                 cond_resched();
6291         }
6292
6293         /*
6294          * we don't use btrfs_set_extent_delalloc because we don't want
6295          * the dirty or uptodate bits
6296          */
6297         if (writing) {
6298                 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6299                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6300                                      EXTENT_DELALLOC, 0, NULL, &cached_state,
6301                                      GFP_NOFS);
6302                 if (ret) {
6303                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6304                                          lockend, EXTENT_LOCKED | write_bits,
6305                                          1, 0, &cached_state, GFP_NOFS);
6306                         goto out;
6307                 }
6308         }
6309
6310         free_extent_state(cached_state);
6311         cached_state = NULL;
6312
6313         ret = __blockdev_direct_IO(rw, iocb, inode,
6314                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6315                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6316                    btrfs_submit_direct, 0);
6317
6318         if (ret < 0 && ret != -EIOCBQUEUED) {
6319                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6320                               offset + iov_length(iov, nr_segs) - 1,
6321                               EXTENT_LOCKED | write_bits, 1, 0,
6322                               &cached_state, GFP_NOFS);
6323         } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6324                 /*
6325                  * We're falling back to buffered, unlock the section we didn't
6326                  * do IO on.
6327                  */
6328                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6329                               offset + iov_length(iov, nr_segs) - 1,
6330                               EXTENT_LOCKED | write_bits, 1, 0,
6331                               &cached_state, GFP_NOFS);
6332         }
6333 out:
6334         free_extent_state(cached_state);
6335         return ret;
6336 }
6337
6338 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6339                 __u64 start, __u64 len)
6340 {
6341         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6342 }
6343
6344 int btrfs_readpage(struct file *file, struct page *page)
6345 {
6346         struct extent_io_tree *tree;
6347         tree = &BTRFS_I(page->mapping->host)->io_tree;
6348         return extent_read_full_page(tree, page, btrfs_get_extent);
6349 }
6350
6351 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6352 {
6353         struct extent_io_tree *tree;
6354
6355
6356         if (current->flags & PF_MEMALLOC) {
6357                 redirty_page_for_writepage(wbc, page);
6358                 unlock_page(page);
6359                 return 0;
6360         }
6361         tree = &BTRFS_I(page->mapping->host)->io_tree;
6362         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6363 }
6364
6365 int btrfs_writepages(struct address_space *mapping,
6366                      struct writeback_control *wbc)
6367 {
6368         struct extent_io_tree *tree;
6369
6370         tree = &BTRFS_I(mapping->host)->io_tree;
6371         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6372 }
6373
6374 static int
6375 btrfs_readpages(struct file *file, struct address_space *mapping,
6376                 struct list_head *pages, unsigned nr_pages)
6377 {
6378         struct extent_io_tree *tree;
6379         tree = &BTRFS_I(mapping->host)->io_tree;
6380         return extent_readpages(tree, mapping, pages, nr_pages,
6381                                 btrfs_get_extent);
6382 }
6383 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6384 {
6385         struct extent_io_tree *tree;
6386         struct extent_map_tree *map;
6387         int ret;
6388
6389         tree = &BTRFS_I(page->mapping->host)->io_tree;
6390         map = &BTRFS_I(page->mapping->host)->extent_tree;
6391         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6392         if (ret == 1) {
6393                 ClearPagePrivate(page);
6394                 set_page_private(page, 0);
6395                 page_cache_release(page);
6396         }
6397         return ret;
6398 }
6399
6400 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6401 {
6402         if (PageWriteback(page) || PageDirty(page))
6403                 return 0;
6404         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6405 }
6406
6407 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6408 {
6409         struct extent_io_tree *tree;
6410         struct btrfs_ordered_extent *ordered;
6411         struct extent_state *cached_state = NULL;
6412         u64 page_start = page_offset(page);
6413         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6414
6415
6416         /*
6417          * we have the page locked, so new writeback can't start,
6418          * and the dirty bit won't be cleared while we are here.
6419          *
6420          * Wait for IO on this page so that we can safely clear
6421          * the PagePrivate2 bit and do ordered accounting
6422          */
6423         wait_on_page_writeback(page);
6424
6425         tree = &BTRFS_I(page->mapping->host)->io_tree;
6426         if (offset) {
6427                 btrfs_releasepage(page, GFP_NOFS);
6428                 return;
6429         }
6430         lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6431                          GFP_NOFS);
6432         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6433                                            page_offset(page));
6434         if (ordered) {
6435                 /*
6436                  * IO on this page will never be started, so we need
6437                  * to account for any ordered extents now
6438                  */
6439                 clear_extent_bit(tree, page_start, page_end,
6440                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6441                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6442                                  &cached_state, GFP_NOFS);
6443                 /*
6444                  * whoever cleared the private bit is responsible
6445                  * for the finish_ordered_io
6446                  */
6447                 if (TestClearPagePrivate2(page)) {
6448                         btrfs_finish_ordered_io(page->mapping->host,
6449                                                 page_start, page_end);
6450                 }
6451                 btrfs_put_ordered_extent(ordered);
6452                 cached_state = NULL;
6453                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6454                                  GFP_NOFS);
6455         }
6456         clear_extent_bit(tree, page_start, page_end,
6457                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6458                  EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6459         __btrfs_releasepage(page, GFP_NOFS);
6460
6461         ClearPageChecked(page);
6462         if (PagePrivate(page)) {
6463                 ClearPagePrivate(page);
6464                 set_page_private(page, 0);
6465                 page_cache_release(page);
6466         }
6467 }
6468
6469 /*
6470  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6471  * called from a page fault handler when a page is first dirtied. Hence we must
6472  * be careful to check for EOF conditions here. We set the page up correctly
6473  * for a written page which means we get ENOSPC checking when writing into
6474  * holes and correct delalloc and unwritten extent mapping on filesystems that
6475  * support these features.
6476  *
6477  * We are not allowed to take the i_mutex here so we have to play games to
6478  * protect against truncate races as the page could now be beyond EOF.  Because
6479  * vmtruncate() writes the inode size before removing pages, once we have the
6480  * page lock we can determine safely if the page is beyond EOF. If it is not
6481  * beyond EOF, then the page is guaranteed safe against truncation until we
6482  * unlock the page.
6483  */
6484 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6485 {
6486         struct page *page = vmf->page;
6487         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6488         struct btrfs_root *root = BTRFS_I(inode)->root;
6489         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6490         struct btrfs_ordered_extent *ordered;
6491         struct extent_state *cached_state = NULL;
6492         char *kaddr;
6493         unsigned long zero_start;
6494         loff_t size;
6495         int ret;
6496         u64 page_start;
6497         u64 page_end;
6498
6499         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6500         if (ret) {
6501                 if (ret == -ENOMEM)
6502                         ret = VM_FAULT_OOM;
6503                 else /* -ENOSPC, -EIO, etc */
6504                         ret = VM_FAULT_SIGBUS;
6505                 goto out;
6506         }
6507
6508         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6509 again:
6510         lock_page(page);
6511         size = i_size_read(inode);
6512         page_start = page_offset(page);
6513         page_end = page_start + PAGE_CACHE_SIZE - 1;
6514
6515         if ((page->mapping != inode->i_mapping) ||
6516             (page_start >= size)) {
6517                 /* page got truncated out from underneath us */
6518                 goto out_unlock;
6519         }
6520         wait_on_page_writeback(page);
6521
6522         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6523                          GFP_NOFS);
6524         set_page_extent_mapped(page);
6525
6526         /*
6527          * we can't set the delalloc bits if there are pending ordered
6528          * extents.  Drop our locks and wait for them to finish
6529          */
6530         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6531         if (ordered) {
6532                 unlock_extent_cached(io_tree, page_start, page_end,
6533                                      &cached_state, GFP_NOFS);
6534                 unlock_page(page);
6535                 btrfs_start_ordered_extent(inode, ordered, 1);
6536                 btrfs_put_ordered_extent(ordered);
6537                 goto again;
6538         }
6539
6540         /*
6541          * XXX - page_mkwrite gets called every time the page is dirtied, even
6542          * if it was already dirty, so for space accounting reasons we need to
6543          * clear any delalloc bits for the range we are fixing to save.  There
6544          * is probably a better way to do this, but for now keep consistent with
6545          * prepare_pages in the normal write path.
6546          */
6547         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6548                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6549                           0, 0, &cached_state, GFP_NOFS);
6550
6551         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6552                                         &cached_state);
6553         if (ret) {
6554                 unlock_extent_cached(io_tree, page_start, page_end,
6555                                      &cached_state, GFP_NOFS);
6556                 ret = VM_FAULT_SIGBUS;
6557                 goto out_unlock;
6558         }
6559         ret = 0;
6560
6561         /* page is wholly or partially inside EOF */
6562         if (page_start + PAGE_CACHE_SIZE > size)
6563                 zero_start = size & ~PAGE_CACHE_MASK;
6564         else
6565                 zero_start = PAGE_CACHE_SIZE;
6566
6567         if (zero_start != PAGE_CACHE_SIZE) {
6568                 kaddr = kmap(page);
6569                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6570                 flush_dcache_page(page);
6571                 kunmap(page);
6572         }
6573         ClearPageChecked(page);
6574         set_page_dirty(page);
6575         SetPageUptodate(page);
6576
6577         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6578         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6579
6580         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6581
6582 out_unlock:
6583         if (!ret)
6584                 return VM_FAULT_LOCKED;
6585         unlock_page(page);
6586         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6587 out:
6588         return ret;
6589 }
6590
6591 static int btrfs_truncate(struct inode *inode)
6592 {
6593         struct btrfs_root *root = BTRFS_I(inode)->root;
6594         struct btrfs_block_rsv *rsv;
6595         int ret;
6596         int err = 0;
6597         struct btrfs_trans_handle *trans;
6598         unsigned long nr;
6599         u64 mask = root->sectorsize - 1;
6600
6601         ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6602         if (ret)
6603                 return ret;
6604
6605         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6606         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6607
6608         /*
6609          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
6610          * 3 things going on here
6611          *
6612          * 1) We need to reserve space for our orphan item and the space to
6613          * delete our orphan item.  Lord knows we don't want to have a dangling
6614          * orphan item because we didn't reserve space to remove it.
6615          *
6616          * 2) We need to reserve space to update our inode.
6617          *
6618          * 3) We need to have something to cache all the space that is going to
6619          * be free'd up by the truncate operation, but also have some slack
6620          * space reserved in case it uses space during the truncate (thank you
6621          * very much snapshotting).
6622          *
6623          * And we need these to all be seperate.  The fact is we can use alot of
6624          * space doing the truncate, and we have no earthly idea how much space
6625          * we will use, so we need the truncate reservation to be seperate so it
6626          * doesn't end up using space reserved for updating the inode or
6627          * removing the orphan item.  We also need to be able to stop the
6628          * transaction and start a new one, which means we need to be able to
6629          * update the inode several times, and we have no idea of knowing how
6630          * many times that will be, so we can't just reserve 1 item for the
6631          * entirety of the opration, so that has to be done seperately as well.
6632          * Then there is the orphan item, which does indeed need to be held on
6633          * to for the whole operation, and we need nobody to touch this reserved
6634          * space except the orphan code.
6635          *
6636          * So that leaves us with
6637          *
6638          * 1) root->orphan_block_rsv - for the orphan deletion.
6639          * 2) rsv - for the truncate reservation, which we will steal from the
6640          * transaction reservation.
6641          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6642          * updating the inode.
6643          */
6644         rsv = btrfs_alloc_block_rsv(root);
6645         if (!rsv)
6646                 return -ENOMEM;
6647         btrfs_add_durable_block_rsv(root->fs_info, rsv);
6648
6649         trans = btrfs_start_transaction(root, 4);
6650         if (IS_ERR(trans)) {
6651                 err = PTR_ERR(trans);
6652                 goto out;
6653         }
6654
6655         btrfs_set_trans_block_group(trans, inode);
6656
6657         /*
6658          * Reserve space for the truncate process.  Truncate should be adding
6659          * space, but if there are snapshots it may end up using space.
6660          */
6661         ret = btrfs_truncate_reserve_metadata(trans, root, rsv);
6662         BUG_ON(ret);
6663
6664         ret = btrfs_orphan_add(trans, inode);
6665         if (ret) {
6666                 btrfs_end_transaction(trans, root);
6667                 goto out;
6668         }
6669
6670         nr = trans->blocks_used;
6671         btrfs_end_transaction(trans, root);
6672         btrfs_btree_balance_dirty(root, nr);
6673
6674         /*
6675          * Ok so we've already migrated our bytes over for the truncate, so here
6676          * just reserve the one slot we need for updating the inode.
6677          */
6678         trans = btrfs_start_transaction(root, 1);
6679         if (IS_ERR(trans)) {
6680                 err = PTR_ERR(trans);
6681                 goto out;
6682         }
6683         btrfs_set_trans_block_group(trans, inode);
6684         trans->block_rsv = rsv;
6685
6686         /*
6687          * setattr is responsible for setting the ordered_data_close flag,
6688          * but that is only tested during the last file release.  That
6689          * could happen well after the next commit, leaving a great big
6690          * window where new writes may get lost if someone chooses to write
6691          * to this file after truncating to zero
6692          *
6693          * The inode doesn't have any dirty data here, and so if we commit
6694          * this is a noop.  If someone immediately starts writing to the inode
6695          * it is very likely we'll catch some of their writes in this
6696          * transaction, and the commit will find this file on the ordered
6697          * data list with good things to send down.
6698          *
6699          * This is a best effort solution, there is still a window where
6700          * using truncate to replace the contents of the file will
6701          * end up with a zero length file after a crash.
6702          */
6703         if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6704                 btrfs_add_ordered_operation(trans, root, inode);
6705
6706         while (1) {
6707                 if (!trans) {
6708                         trans = btrfs_start_transaction(root, 3);
6709                         if (IS_ERR(trans)) {
6710                                 err = PTR_ERR(trans);
6711                                 goto out;
6712                         }
6713
6714                         ret = btrfs_truncate_reserve_metadata(trans, root,
6715                                                               rsv);
6716                         BUG_ON(ret);
6717
6718                         btrfs_set_trans_block_group(trans, inode);
6719                         trans->block_rsv = rsv;
6720                 }
6721
6722                 ret = btrfs_truncate_inode_items(trans, root, inode,
6723                                                  inode->i_size,
6724                                                  BTRFS_EXTENT_DATA_KEY);
6725                 if (ret != -EAGAIN) {
6726                         err = ret;
6727                         break;
6728                 }
6729
6730                 trans->block_rsv = &root->fs_info->trans_block_rsv;
6731                 ret = btrfs_update_inode(trans, root, inode);
6732                 if (ret) {
6733                         err = ret;
6734                         break;
6735                 }
6736
6737                 nr = trans->blocks_used;
6738                 btrfs_end_transaction(trans, root);
6739                 trans = NULL;
6740                 btrfs_btree_balance_dirty(root, nr);
6741         }
6742
6743         if (ret == 0 && inode->i_nlink > 0) {
6744                 trans->block_rsv = root->orphan_block_rsv;
6745                 ret = btrfs_orphan_del(trans, inode);
6746                 if (ret)
6747                         err = ret;
6748         } else if (ret && inode->i_nlink > 0) {
6749                 /*
6750                  * Failed to do the truncate, remove us from the in memory
6751                  * orphan list.
6752                  */
6753                 ret = btrfs_orphan_del(NULL, inode);
6754         }
6755
6756         trans->block_rsv = &root->fs_info->trans_block_rsv;
6757         ret = btrfs_update_inode(trans, root, inode);
6758         if (ret && !err)
6759                 err = ret;
6760
6761         nr = trans->blocks_used;
6762         ret = btrfs_end_transaction_throttle(trans, root);
6763         btrfs_btree_balance_dirty(root, nr);
6764
6765 out:
6766         btrfs_free_block_rsv(root, rsv);
6767
6768         if (ret && !err)
6769                 err = ret;
6770
6771         return err;
6772 }
6773
6774 /*
6775  * create a new subvolume directory/inode (helper for the ioctl).
6776  */
6777 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6778                              struct btrfs_root *new_root,
6779                              u64 new_dirid, u64 alloc_hint)
6780 {
6781         struct inode *inode;
6782         int err;
6783         u64 index = 0;
6784
6785         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6786                                 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
6787         if (IS_ERR(inode))
6788                 return PTR_ERR(inode);
6789         inode->i_op = &btrfs_dir_inode_operations;
6790         inode->i_fop = &btrfs_dir_file_operations;
6791
6792         inode->i_nlink = 1;
6793         btrfs_i_size_write(inode, 0);
6794
6795         err = btrfs_update_inode(trans, new_root, inode);
6796         BUG_ON(err);
6797
6798         iput(inode);
6799         return 0;
6800 }
6801
6802 /* helper function for file defrag and space balancing.  This
6803  * forces readahead on a given range of bytes in an inode
6804  */
6805 unsigned long btrfs_force_ra(struct address_space *mapping,
6806                               struct file_ra_state *ra, struct file *file,
6807                               pgoff_t offset, pgoff_t last_index)
6808 {
6809         pgoff_t req_size = last_index - offset + 1;
6810
6811         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
6812         return offset + req_size;
6813 }
6814
6815 struct inode *btrfs_alloc_inode(struct super_block *sb)
6816 {
6817         struct btrfs_inode *ei;
6818         struct inode *inode;
6819
6820         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6821         if (!ei)
6822                 return NULL;
6823
6824         ei->root = NULL;
6825         ei->space_info = NULL;
6826         ei->generation = 0;
6827         ei->sequence = 0;
6828         ei->last_trans = 0;
6829         ei->last_sub_trans = 0;
6830         ei->logged_trans = 0;
6831         ei->delalloc_bytes = 0;
6832         ei->reserved_bytes = 0;
6833         ei->disk_i_size = 0;
6834         ei->flags = 0;
6835         ei->index_cnt = (u64)-1;
6836         ei->last_unlink_trans = 0;
6837
6838         atomic_set(&ei->outstanding_extents, 0);
6839         atomic_set(&ei->reserved_extents, 0);
6840
6841         ei->ordered_data_close = 0;
6842         ei->orphan_meta_reserved = 0;
6843         ei->dummy_inode = 0;
6844         ei->force_compress = BTRFS_COMPRESS_NONE;
6845
6846         inode = &ei->vfs_inode;
6847         extent_map_tree_init(&ei->extent_tree, GFP_NOFS);
6848         extent_io_tree_init(&ei->io_tree, &inode->i_data, GFP_NOFS);
6849         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data, GFP_NOFS);
6850         mutex_init(&ei->log_mutex);
6851         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6852         INIT_LIST_HEAD(&ei->i_orphan);
6853         INIT_LIST_HEAD(&ei->delalloc_inodes);
6854         INIT_LIST_HEAD(&ei->ordered_operations);
6855         RB_CLEAR_NODE(&ei->rb_node);
6856
6857         return inode;
6858 }
6859
6860 static void btrfs_i_callback(struct rcu_head *head)
6861 {
6862         struct inode *inode = container_of(head, struct inode, i_rcu);
6863         INIT_LIST_HEAD(&inode->i_dentry);
6864         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6865 }
6866
6867 void btrfs_destroy_inode(struct inode *inode)
6868 {
6869         struct btrfs_ordered_extent *ordered;
6870         struct btrfs_root *root = BTRFS_I(inode)->root;
6871
6872         WARN_ON(!list_empty(&inode->i_dentry));
6873         WARN_ON(inode->i_data.nrpages);
6874         WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents));
6875         WARN_ON(atomic_read(&BTRFS_I(inode)->reserved_extents));
6876
6877         /*
6878          * This can happen where we create an inode, but somebody else also
6879          * created the same inode and we need to destroy the one we already
6880          * created.
6881          */
6882         if (!root)
6883                 goto free;
6884
6885         /*
6886          * Make sure we're properly removed from the ordered operation
6887          * lists.
6888          */
6889         smp_mb();
6890         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6891                 spin_lock(&root->fs_info->ordered_extent_lock);
6892                 list_del_init(&BTRFS_I(inode)->ordered_operations);
6893                 spin_unlock(&root->fs_info->ordered_extent_lock);
6894         }
6895
6896         if (root == root->fs_info->tree_root) {
6897                 struct btrfs_block_group_cache *block_group;
6898
6899                 block_group = btrfs_lookup_block_group(root->fs_info,
6900                                                 BTRFS_I(inode)->block_group);
6901                 if (block_group && block_group->inode == inode) {
6902                         spin_lock(&block_group->lock);
6903                         block_group->inode = NULL;
6904                         spin_unlock(&block_group->lock);
6905                         btrfs_put_block_group(block_group);
6906                 } else if (block_group) {
6907                         btrfs_put_block_group(block_group);
6908                 }
6909         }
6910
6911         spin_lock(&root->orphan_lock);
6912         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6913                 printk(KERN_INFO "BTRFS: inode %lu still on the orphan list\n",
6914                        inode->i_ino);
6915                 list_del_init(&BTRFS_I(inode)->i_orphan);
6916         }
6917         spin_unlock(&root->orphan_lock);
6918
6919         while (1) {
6920                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6921                 if (!ordered)
6922                         break;
6923                 else {
6924                         printk(KERN_ERR "btrfs found ordered "
6925                                "extent %llu %llu on inode cleanup\n",
6926                                (unsigned long long)ordered->file_offset,
6927                                (unsigned long long)ordered->len);
6928                         btrfs_remove_ordered_extent(inode, ordered);
6929                         btrfs_put_ordered_extent(ordered);
6930                         btrfs_put_ordered_extent(ordered);
6931                 }
6932         }
6933         inode_tree_del(inode);
6934         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6935 free:
6936         call_rcu(&inode->i_rcu, btrfs_i_callback);
6937 }
6938
6939 int btrfs_drop_inode(struct inode *inode)
6940 {
6941         struct btrfs_root *root = BTRFS_I(inode)->root;
6942
6943         if (btrfs_root_refs(&root->root_item) == 0 &&
6944             root != root->fs_info->tree_root)
6945                 return 1;
6946         else
6947                 return generic_drop_inode(inode);
6948 }
6949
6950 static void init_once(void *foo)
6951 {
6952         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6953
6954         inode_init_once(&ei->vfs_inode);
6955 }
6956
6957 void btrfs_destroy_cachep(void)
6958 {
6959         if (btrfs_inode_cachep)
6960                 kmem_cache_destroy(btrfs_inode_cachep);
6961         if (btrfs_trans_handle_cachep)
6962                 kmem_cache_destroy(btrfs_trans_handle_cachep);
6963         if (btrfs_transaction_cachep)
6964                 kmem_cache_destroy(btrfs_transaction_cachep);
6965         if (btrfs_path_cachep)
6966                 kmem_cache_destroy(btrfs_path_cachep);
6967         if (btrfs_free_space_cachep)
6968                 kmem_cache_destroy(btrfs_free_space_cachep);
6969 }
6970
6971 int btrfs_init_cachep(void)
6972 {
6973         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6974                         sizeof(struct btrfs_inode), 0,
6975                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6976         if (!btrfs_inode_cachep)
6977                 goto fail;
6978
6979         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6980                         sizeof(struct btrfs_trans_handle), 0,
6981                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6982         if (!btrfs_trans_handle_cachep)
6983                 goto fail;
6984
6985         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6986                         sizeof(struct btrfs_transaction), 0,
6987                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6988         if (!btrfs_transaction_cachep)
6989                 goto fail;
6990
6991         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6992                         sizeof(struct btrfs_path), 0,
6993                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6994         if (!btrfs_path_cachep)
6995                 goto fail;
6996
6997         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6998                         sizeof(struct btrfs_free_space), 0,
6999                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7000         if (!btrfs_free_space_cachep)
7001                 goto fail;
7002
7003         return 0;
7004 fail:
7005         btrfs_destroy_cachep();
7006         return -ENOMEM;
7007 }
7008
7009 static int btrfs_getattr(struct vfsmount *mnt,
7010                          struct dentry *dentry, struct kstat *stat)
7011 {
7012         struct inode *inode = dentry->d_inode;
7013         generic_fillattr(inode, stat);
7014         stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
7015         stat->blksize = PAGE_CACHE_SIZE;
7016         stat->blocks = (inode_get_bytes(inode) +
7017                         BTRFS_I(inode)->delalloc_bytes) >> 9;
7018         return 0;
7019 }
7020
7021 /*
7022  * If a file is moved, it will inherit the cow and compression flags of the new
7023  * directory.
7024  */
7025 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
7026 {
7027         struct btrfs_inode *b_dir = BTRFS_I(dir);
7028         struct btrfs_inode *b_inode = BTRFS_I(inode);
7029
7030         if (b_dir->flags & BTRFS_INODE_NODATACOW)
7031                 b_inode->flags |= BTRFS_INODE_NODATACOW;
7032         else
7033                 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
7034
7035         if (b_dir->flags & BTRFS_INODE_COMPRESS)
7036                 b_inode->flags |= BTRFS_INODE_COMPRESS;
7037         else
7038                 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
7039 }
7040
7041 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
7042                            struct inode *new_dir, struct dentry *new_dentry)
7043 {
7044         struct btrfs_trans_handle *trans;
7045         struct btrfs_root *root = BTRFS_I(old_dir)->root;
7046         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
7047         struct inode *new_inode = new_dentry->d_inode;
7048         struct inode *old_inode = old_dentry->d_inode;
7049         struct timespec ctime = CURRENT_TIME;
7050         u64 index = 0;
7051         u64 root_objectid;
7052         int ret;
7053
7054         if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
7055                 return -EPERM;
7056
7057         /* we only allow rename subvolume link between subvolumes */
7058         if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
7059                 return -EXDEV;
7060
7061         if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
7062             (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID))
7063                 return -ENOTEMPTY;
7064
7065         if (S_ISDIR(old_inode->i_mode) && new_inode &&
7066             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
7067                 return -ENOTEMPTY;
7068         /*
7069          * we're using rename to replace one file with another.
7070          * and the replacement file is large.  Start IO on it now so
7071          * we don't add too much work to the end of the transaction
7072          */
7073         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
7074             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
7075                 filemap_flush(old_inode->i_mapping);
7076
7077         /* close the racy window with snapshot create/destroy ioctl */
7078         if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
7079                 down_read(&root->fs_info->subvol_sem);
7080         /*
7081          * We want to reserve the absolute worst case amount of items.  So if
7082          * both inodes are subvols and we need to unlink them then that would
7083          * require 4 item modifications, but if they are both normal inodes it
7084          * would require 5 item modifications, so we'll assume their normal
7085          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7086          * should cover the worst case number of items we'll modify.
7087          */
7088         trans = btrfs_start_transaction(root, 20);
7089         if (IS_ERR(trans)) {
7090                 ret = PTR_ERR(trans);
7091                 goto out_notrans;
7092         }
7093
7094         btrfs_set_trans_block_group(trans, new_dir);
7095
7096         if (dest != root)
7097                 btrfs_record_root_in_trans(trans, dest);
7098
7099         ret = btrfs_set_inode_index(new_dir, &index);
7100         if (ret)
7101                 goto out_fail;
7102
7103         if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7104                 /* force full log commit if subvolume involved. */
7105                 root->fs_info->last_trans_log_full_commit = trans->transid;
7106         } else {
7107                 ret = btrfs_insert_inode_ref(trans, dest,
7108                                              new_dentry->d_name.name,
7109                                              new_dentry->d_name.len,
7110                                              old_inode->i_ino,
7111                                              new_dir->i_ino, index);
7112                 if (ret)
7113                         goto out_fail;
7114                 /*
7115                  * this is an ugly little race, but the rename is required
7116                  * to make sure that if we crash, the inode is either at the
7117                  * old name or the new one.  pinning the log transaction lets
7118                  * us make sure we don't allow a log commit to come in after
7119                  * we unlink the name but before we add the new name back in.
7120                  */
7121                 btrfs_pin_log_trans(root);
7122         }
7123         /*
7124          * make sure the inode gets flushed if it is replacing
7125          * something.
7126          */
7127         if (new_inode && new_inode->i_size &&
7128             old_inode && S_ISREG(old_inode->i_mode)) {
7129                 btrfs_add_ordered_operation(trans, root, old_inode);
7130         }
7131
7132         old_dir->i_ctime = old_dir->i_mtime = ctime;
7133         new_dir->i_ctime = new_dir->i_mtime = ctime;
7134         old_inode->i_ctime = ctime;
7135
7136         if (old_dentry->d_parent != new_dentry->d_parent)
7137                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7138
7139         if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7140                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7141                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7142                                         old_dentry->d_name.name,
7143                                         old_dentry->d_name.len);
7144         } else {
7145                 ret = __btrfs_unlink_inode(trans, root, old_dir,
7146                                         old_dentry->d_inode,
7147                                         old_dentry->d_name.name,
7148                                         old_dentry->d_name.len);
7149                 if (!ret)
7150                         ret = btrfs_update_inode(trans, root, old_inode);
7151         }
7152         BUG_ON(ret);
7153
7154         if (new_inode) {
7155                 new_inode->i_ctime = CURRENT_TIME;
7156                 if (unlikely(new_inode->i_ino ==
7157                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7158                         root_objectid = BTRFS_I(new_inode)->location.objectid;
7159                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
7160                                                 root_objectid,
7161                                                 new_dentry->d_name.name,
7162                                                 new_dentry->d_name.len);
7163                         BUG_ON(new_inode->i_nlink == 0);
7164                 } else {
7165                         ret = btrfs_unlink_inode(trans, dest, new_dir,
7166                                                  new_dentry->d_inode,
7167                                                  new_dentry->d_name.name,
7168                                                  new_dentry->d_name.len);
7169                 }
7170                 BUG_ON(ret);
7171                 if (new_inode->i_nlink == 0) {
7172                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7173                         BUG_ON(ret);
7174                 }
7175         }
7176
7177         fixup_inode_flags(new_dir, old_inode);
7178
7179         ret = btrfs_add_link(trans, new_dir, old_inode,
7180                              new_dentry->d_name.name,
7181                              new_dentry->d_name.len, 0, index);
7182         BUG_ON(ret);
7183
7184         if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
7185                 struct dentry *parent = dget_parent(new_dentry);
7186                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7187                 dput(parent);
7188                 btrfs_end_log_trans(root);
7189         }
7190 out_fail:
7191         btrfs_end_transaction_throttle(trans, root);
7192 out_notrans:
7193         if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
7194                 up_read(&root->fs_info->subvol_sem);
7195
7196         return ret;
7197 }
7198
7199 /*
7200  * some fairly slow code that needs optimization. This walks the list
7201  * of all the inodes with pending delalloc and forces them to disk.
7202  */
7203 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7204 {
7205         struct list_head *head = &root->fs_info->delalloc_inodes;
7206         struct btrfs_inode *binode;
7207         struct inode *inode;
7208
7209         if (root->fs_info->sb->s_flags & MS_RDONLY)
7210                 return -EROFS;
7211
7212         spin_lock(&root->fs_info->delalloc_lock);
7213         while (!list_empty(head)) {
7214                 binode = list_entry(head->next, struct btrfs_inode,
7215                                     delalloc_inodes);
7216                 inode = igrab(&binode->vfs_inode);
7217                 if (!inode)
7218                         list_del_init(&binode->delalloc_inodes);
7219                 spin_unlock(&root->fs_info->delalloc_lock);
7220                 if (inode) {
7221                         filemap_flush(inode->i_mapping);
7222                         if (delay_iput)
7223                                 btrfs_add_delayed_iput(inode);
7224                         else
7225                                 iput(inode);
7226                 }
7227                 cond_resched();
7228                 spin_lock(&root->fs_info->delalloc_lock);
7229         }
7230         spin_unlock(&root->fs_info->delalloc_lock);
7231
7232         /* the filemap_flush will queue IO into the worker threads, but
7233          * we have to make sure the IO is actually started and that
7234          * ordered extents get created before we return
7235          */
7236         atomic_inc(&root->fs_info->async_submit_draining);
7237         while (atomic_read(&root->fs_info->nr_async_submits) ||
7238               atomic_read(&root->fs_info->async_delalloc_pages)) {
7239                 wait_event(root->fs_info->async_submit_wait,
7240                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7241                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7242         }
7243         atomic_dec(&root->fs_info->async_submit_draining);
7244         return 0;
7245 }
7246
7247 int btrfs_start_one_delalloc_inode(struct btrfs_root *root, int delay_iput,
7248                                    int sync)
7249 {
7250         struct btrfs_inode *binode;
7251         struct inode *inode = NULL;
7252
7253         spin_lock(&root->fs_info->delalloc_lock);
7254         while (!list_empty(&root->fs_info->delalloc_inodes)) {
7255                 binode = list_entry(root->fs_info->delalloc_inodes.next,
7256                                     struct btrfs_inode, delalloc_inodes);
7257                 inode = igrab(&binode->vfs_inode);
7258                 if (inode) {
7259                         list_move_tail(&binode->delalloc_inodes,
7260                                        &root->fs_info->delalloc_inodes);
7261                         break;
7262                 }
7263
7264                 list_del_init(&binode->delalloc_inodes);
7265                 cond_resched_lock(&root->fs_info->delalloc_lock);
7266         }
7267         spin_unlock(&root->fs_info->delalloc_lock);
7268
7269         if (inode) {
7270                 if (sync) {
7271                         filemap_write_and_wait(inode->i_mapping);
7272                         /*
7273                          * We have to do this because compression doesn't
7274                          * actually set PG_writeback until it submits the pages
7275                          * for IO, which happens in an async thread, so we could
7276                          * race and not actually wait for any writeback pages
7277                          * because they've not been submitted yet.  Technically
7278                          * this could still be the case for the ordered stuff
7279                          * since the async thread may not have started to do its
7280                          * work yet.  If this becomes the case then we need to
7281                          * figure out a way to make sure that in writepage we
7282                          * wait for any async pages to be submitted before
7283                          * returning so that fdatawait does what its supposed to
7284                          * do.
7285                          */
7286                         btrfs_wait_ordered_range(inode, 0, (u64)-1);
7287                 } else {
7288                         filemap_flush(inode->i_mapping);
7289                 }
7290                 if (delay_iput)
7291                         btrfs_add_delayed_iput(inode);
7292                 else
7293                         iput(inode);
7294                 return 1;
7295         }
7296         return 0;
7297 }
7298
7299 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7300                          const char *symname)
7301 {
7302         struct btrfs_trans_handle *trans;
7303         struct btrfs_root *root = BTRFS_I(dir)->root;
7304         struct btrfs_path *path;
7305         struct btrfs_key key;
7306         struct inode *inode = NULL;
7307         int err;
7308         int drop_inode = 0;
7309         u64 objectid;
7310         u64 index = 0 ;
7311         int name_len;
7312         int datasize;
7313         unsigned long ptr;
7314         struct btrfs_file_extent_item *ei;
7315         struct extent_buffer *leaf;
7316         unsigned long nr = 0;
7317
7318         name_len = strlen(symname) + 1;
7319         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7320                 return -ENAMETOOLONG;
7321
7322         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
7323         if (err)
7324                 return err;
7325         /*
7326          * 2 items for inode item and ref
7327          * 2 items for dir items
7328          * 1 item for xattr if selinux is on
7329          */
7330         trans = btrfs_start_transaction(root, 5);
7331         if (IS_ERR(trans))
7332                 return PTR_ERR(trans);
7333
7334         btrfs_set_trans_block_group(trans, dir);
7335
7336         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7337                                 dentry->d_name.len, dir->i_ino, objectid,
7338                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
7339                                 &index);
7340         if (IS_ERR(inode)) {
7341                 err = PTR_ERR(inode);
7342                 goto out_unlock;
7343         }
7344
7345         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7346         if (err) {
7347                 drop_inode = 1;
7348                 goto out_unlock;
7349         }
7350
7351         btrfs_set_trans_block_group(trans, inode);
7352         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7353         if (err)
7354                 drop_inode = 1;
7355         else {
7356                 inode->i_mapping->a_ops = &btrfs_aops;
7357                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7358                 inode->i_fop = &btrfs_file_operations;
7359                 inode->i_op = &btrfs_file_inode_operations;
7360                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7361         }
7362         btrfs_update_inode_block_group(trans, inode);
7363         btrfs_update_inode_block_group(trans, dir);
7364         if (drop_inode)
7365                 goto out_unlock;
7366
7367         path = btrfs_alloc_path();
7368         BUG_ON(!path);
7369         key.objectid = inode->i_ino;
7370         key.offset = 0;
7371         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7372         datasize = btrfs_file_extent_calc_inline_size(name_len);
7373         err = btrfs_insert_empty_item(trans, root, path, &key,
7374                                       datasize);
7375         if (err) {
7376                 drop_inode = 1;
7377                 goto out_unlock;
7378         }
7379         leaf = path->nodes[0];
7380         ei = btrfs_item_ptr(leaf, path->slots[0],
7381                             struct btrfs_file_extent_item);
7382         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7383         btrfs_set_file_extent_type(leaf, ei,
7384                                    BTRFS_FILE_EXTENT_INLINE);
7385         btrfs_set_file_extent_encryption(leaf, ei, 0);
7386         btrfs_set_file_extent_compression(leaf, ei, 0);
7387         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7388         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7389
7390         ptr = btrfs_file_extent_inline_start(ei);
7391         write_extent_buffer(leaf, symname, ptr, name_len);
7392         btrfs_mark_buffer_dirty(leaf);
7393         btrfs_free_path(path);
7394
7395         inode->i_op = &btrfs_symlink_inode_operations;
7396         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7397         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7398         inode_set_bytes(inode, name_len);
7399         btrfs_i_size_write(inode, name_len - 1);
7400         err = btrfs_update_inode(trans, root, inode);
7401         if (err)
7402                 drop_inode = 1;
7403
7404 out_unlock:
7405         nr = trans->blocks_used;
7406         btrfs_end_transaction_throttle(trans, root);
7407         if (drop_inode) {
7408                 inode_dec_link_count(inode);
7409                 iput(inode);
7410         }
7411         btrfs_btree_balance_dirty(root, nr);
7412         return err;
7413 }
7414
7415 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7416                                        u64 start, u64 num_bytes, u64 min_size,
7417                                        loff_t actual_len, u64 *alloc_hint,
7418                                        struct btrfs_trans_handle *trans)
7419 {
7420         struct btrfs_root *root = BTRFS_I(inode)->root;
7421         struct btrfs_key ins;
7422         u64 cur_offset = start;
7423         u64 i_size;
7424         int ret = 0;
7425         bool own_trans = true;
7426
7427         if (trans)
7428                 own_trans = false;
7429         while (num_bytes > 0) {
7430                 if (own_trans) {
7431                         trans = btrfs_start_transaction(root, 3);
7432                         if (IS_ERR(trans)) {
7433                                 ret = PTR_ERR(trans);
7434                                 break;
7435                         }
7436                 }
7437
7438                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7439                                            0, *alloc_hint, (u64)-1, &ins, 1);
7440                 if (ret) {
7441                         if (own_trans)
7442                                 btrfs_end_transaction(trans, root);
7443                         break;
7444                 }
7445
7446                 ret = insert_reserved_file_extent(trans, inode,
7447                                                   cur_offset, ins.objectid,
7448                                                   ins.offset, ins.offset,
7449                                                   ins.offset, 0, 0, 0,
7450                                                   BTRFS_FILE_EXTENT_PREALLOC);
7451                 BUG_ON(ret);
7452                 btrfs_drop_extent_cache(inode, cur_offset,
7453                                         cur_offset + ins.offset -1, 0);
7454
7455                 num_bytes -= ins.offset;
7456                 cur_offset += ins.offset;
7457                 *alloc_hint = ins.objectid + ins.offset;
7458
7459                 inode->i_ctime = CURRENT_TIME;
7460                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7461                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7462                     (actual_len > inode->i_size) &&
7463                     (cur_offset > inode->i_size)) {
7464                         if (cur_offset > actual_len)
7465                                 i_size = actual_len;
7466                         else
7467                                 i_size = cur_offset;
7468                         i_size_write(inode, i_size);
7469                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7470                 }
7471
7472                 ret = btrfs_update_inode(trans, root, inode);
7473                 BUG_ON(ret);
7474
7475                 if (own_trans)
7476                         btrfs_end_transaction(trans, root);
7477         }
7478         return ret;
7479 }
7480
7481 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7482                               u64 start, u64 num_bytes, u64 min_size,
7483                               loff_t actual_len, u64 *alloc_hint)
7484 {
7485         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7486                                            min_size, actual_len, alloc_hint,
7487                                            NULL);
7488 }
7489
7490 int btrfs_prealloc_file_range_trans(struct inode *inode,
7491                                     struct btrfs_trans_handle *trans, int mode,
7492                                     u64 start, u64 num_bytes, u64 min_size,
7493                                     loff_t actual_len, u64 *alloc_hint)
7494 {
7495         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7496                                            min_size, actual_len, alloc_hint, trans);
7497 }
7498
7499 static int btrfs_set_page_dirty(struct page *page)
7500 {
7501         return __set_page_dirty_nobuffers(page);
7502 }
7503
7504 static int btrfs_permission(struct inode *inode, int mask, unsigned int flags)
7505 {
7506         struct btrfs_root *root = BTRFS_I(inode)->root;
7507
7508         if (btrfs_root_readonly(root) && (mask & MAY_WRITE))
7509                 return -EROFS;
7510         if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
7511                 return -EACCES;
7512         return generic_permission(inode, mask, flags, btrfs_check_acl);
7513 }
7514
7515 static const struct inode_operations btrfs_dir_inode_operations = {
7516         .getattr        = btrfs_getattr,
7517         .lookup         = btrfs_lookup,
7518         .create         = btrfs_create,
7519         .unlink         = btrfs_unlink,
7520         .link           = btrfs_link,
7521         .mkdir          = btrfs_mkdir,
7522         .rmdir          = btrfs_rmdir,
7523         .rename         = btrfs_rename,
7524         .symlink        = btrfs_symlink,
7525         .setattr        = btrfs_setattr,
7526         .mknod          = btrfs_mknod,
7527         .setxattr       = btrfs_setxattr,
7528         .getxattr       = btrfs_getxattr,
7529         .listxattr      = btrfs_listxattr,
7530         .removexattr    = btrfs_removexattr,
7531         .permission     = btrfs_permission,
7532 };
7533 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7534         .lookup         = btrfs_lookup,
7535         .permission     = btrfs_permission,
7536 };
7537
7538 static const struct file_operations btrfs_dir_file_operations = {
7539         .llseek         = generic_file_llseek,
7540         .read           = generic_read_dir,
7541         .readdir        = btrfs_real_readdir,
7542         .unlocked_ioctl = btrfs_ioctl,
7543 #ifdef CONFIG_COMPAT
7544         .compat_ioctl   = btrfs_ioctl,
7545 #endif
7546         .release        = btrfs_release_file,
7547         .fsync          = btrfs_sync_file,
7548 };
7549
7550 static struct extent_io_ops btrfs_extent_io_ops = {
7551         .fill_delalloc = run_delalloc_range,
7552         .submit_bio_hook = btrfs_submit_bio_hook,
7553         .merge_bio_hook = btrfs_merge_bio_hook,
7554         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7555         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7556         .writepage_start_hook = btrfs_writepage_start_hook,
7557         .readpage_io_failed_hook = btrfs_io_failed_hook,
7558         .set_bit_hook = btrfs_set_bit_hook,
7559         .clear_bit_hook = btrfs_clear_bit_hook,
7560         .merge_extent_hook = btrfs_merge_extent_hook,
7561         .split_extent_hook = btrfs_split_extent_hook,
7562 };
7563
7564 /*
7565  * btrfs doesn't support the bmap operation because swapfiles
7566  * use bmap to make a mapping of extents in the file.  They assume
7567  * these extents won't change over the life of the file and they
7568  * use the bmap result to do IO directly to the drive.
7569  *
7570  * the btrfs bmap call would return logical addresses that aren't
7571  * suitable for IO and they also will change frequently as COW
7572  * operations happen.  So, swapfile + btrfs == corruption.
7573  *
7574  * For now we're avoiding this by dropping bmap.
7575  */
7576 static const struct address_space_operations btrfs_aops = {
7577         .readpage       = btrfs_readpage,
7578         .writepage      = btrfs_writepage,
7579         .writepages     = btrfs_writepages,
7580         .readpages      = btrfs_readpages,
7581         .direct_IO      = btrfs_direct_IO,
7582         .invalidatepage = btrfs_invalidatepage,
7583         .releasepage    = btrfs_releasepage,
7584         .set_page_dirty = btrfs_set_page_dirty,
7585         .error_remove_page = generic_error_remove_page,
7586 };
7587
7588 static const struct address_space_operations btrfs_symlink_aops = {
7589         .readpage       = btrfs_readpage,
7590         .writepage      = btrfs_writepage,
7591         .invalidatepage = btrfs_invalidatepage,
7592         .releasepage    = btrfs_releasepage,
7593 };
7594
7595 static const struct inode_operations btrfs_file_inode_operations = {
7596         .getattr        = btrfs_getattr,
7597         .setattr        = btrfs_setattr,
7598         .setxattr       = btrfs_setxattr,
7599         .getxattr       = btrfs_getxattr,
7600         .listxattr      = btrfs_listxattr,
7601         .removexattr    = btrfs_removexattr,
7602         .permission     = btrfs_permission,
7603         .fiemap         = btrfs_fiemap,
7604 };
7605 static const struct inode_operations btrfs_special_inode_operations = {
7606         .getattr        = btrfs_getattr,
7607         .setattr        = btrfs_setattr,
7608         .permission     = btrfs_permission,
7609         .setxattr       = btrfs_setxattr,
7610         .getxattr       = btrfs_getxattr,
7611         .listxattr      = btrfs_listxattr,
7612         .removexattr    = btrfs_removexattr,
7613 };
7614 static const struct inode_operations btrfs_symlink_inode_operations = {
7615         .readlink       = generic_readlink,
7616         .follow_link    = page_follow_link_light,
7617         .put_link       = page_put_link,
7618         .getattr        = btrfs_getattr,
7619         .permission     = btrfs_permission,
7620         .setxattr       = btrfs_setxattr,
7621         .getxattr       = btrfs_getxattr,
7622         .listxattr      = btrfs_listxattr,
7623         .removexattr    = btrfs_removexattr,
7624 };
7625
7626 const struct dentry_operations btrfs_dentry_operations = {
7627         .d_delete       = btrfs_dentry_delete,
7628 };