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 "compat.h"
  40 #include "ctree.h"
  41 #include "disk-io.h"
  42 #include "transaction.h"
  43 #include "btrfs_inode.h"
  44 #include "ioctl.h"
  45 #include "print-tree.h"
  46 #include "volumes.h"
  47 #include "ordered-data.h"
  48 #include "xattr.h"
  49 #include "tree-log.h"
  50 #include "compression.h"
  51 #include "locking.h"
  52
  53 struct btrfs_iget_args {
  54         u64 ino;
  55         struct btrfs_root *root;
  56 };
  57
  58 static struct inode_operations btrfs_dir_inode_operations;
  59 static struct inode_operations btrfs_symlink_inode_operations;
  60 static struct inode_operations btrfs_dir_ro_inode_operations;
  61 static struct inode_operations btrfs_special_inode_operations;
  62 static struct inode_operations btrfs_file_inode_operations;
  63 static struct address_space_operations btrfs_aops;
  64 static struct address_space_operations btrfs_symlink_aops;
  65 static struct file_operations btrfs_dir_file_operations;
  66 static struct extent_io_ops btrfs_extent_io_ops;
  67
  68 static struct kmem_cache *btrfs_inode_cachep;
  69 struct kmem_cache *btrfs_trans_handle_cachep;
  70 struct kmem_cache *btrfs_transaction_cachep;
  71 struct kmem_cache *btrfs_path_cachep;
  72
  73 #define S_SHIFT 12
  74 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  75         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  76         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  77         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  78         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  79         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  80         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  81         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  82 };
  83
  84 static void btrfs_truncate(struct inode *inode);
  85 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
  86 static noinline int cow_file_range(struct inode *inode,
  87                                    struct page *locked_page,
  88                                    u64 start, u64 end, int *page_started,
  89                                    unsigned long *nr_written, int unlock);
  90
  91 static int btrfs_init_inode_security(struct inode *inode,  struct inode *dir)
  92 {
  93         int err;
  94
  95         err = btrfs_init_acl(inode, dir);
  96         if (!err)
  97                 err = btrfs_xattr_security_init(inode, dir);
  98         return err;
  99 }
 100
 101 /*
 102  * this does all the hard work for inserting an inline extent into
 103  * the btree.  The caller should have done a btrfs_drop_extents so that
 104  * no overlapping inline items exist in the btree
 105  */
 106 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 107                                 struct btrfs_root *root, struct inode *inode,
 108                                 u64 start, size_t size, size_t compressed_size,
 109                                 struct page **compressed_pages)
 110 {
 111         struct btrfs_key key;
 112         struct btrfs_path *path;
 113         struct extent_buffer *leaf;
 114         struct page *page = NULL;
 115         char *kaddr;
 116         unsigned long ptr;
 117         struct btrfs_file_extent_item *ei;
 118         int err = 0;
 119         int ret;
 120         size_t cur_size = size;
 121         size_t datasize;
 122         unsigned long offset;
 123         int use_compress = 0;
 124
 125         if (compressed_size && compressed_pages) {
 126                 use_compress = 1;
 127                 cur_size = compressed_size;
 128         }
 129
 130         path = btrfs_alloc_path();
 131         if (!path)
 132                 return -ENOMEM;
 133
 134         path->leave_spinning = 1;
 135         btrfs_set_trans_block_group(trans, inode);
 136
 137         key.objectid = inode->i_ino;
 138         key.offset = start;
 139         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 140         datasize = btrfs_file_extent_calc_inline_size(cur_size);
 141
 142         inode_add_bytes(inode, size);
 143         ret = btrfs_insert_empty_item(trans, root, path, &key,
 144                                       datasize);
 145         BUG_ON(ret);
 146         if (ret) {
 147                 err = ret;
 148                 goto fail;
 149         }
 150         leaf = path->nodes[0];
 151         ei = btrfs_item_ptr(leaf, path->slots[0],
 152                             struct btrfs_file_extent_item);
 153         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 154         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 155         btrfs_set_file_extent_encryption(leaf, ei, 0);
 156         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 157         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 158         ptr = btrfs_file_extent_inline_start(ei);
 159
 160         if (use_compress) {
 161                 struct page *cpage;
 162                 int i = 0;
 163                 while (compressed_size > 0) {
 164                         cpage = compressed_pages[i];
 165                         cur_size = min_t(unsigned long, compressed_size,
 166                                        PAGE_CACHE_SIZE);
 167
 168                         kaddr = kmap_atomic(cpage, KM_USER0);
 169                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 170                         kunmap_atomic(kaddr, KM_USER0);
 171
 172                         i++;
 173                         ptr += cur_size;
 174                         compressed_size -= cur_size;
 175                 }
 176                 btrfs_set_file_extent_compression(leaf, ei,
 177                                                   BTRFS_COMPRESS_ZLIB);
 178         } else {
 179                 page = find_get_page(inode->i_mapping,
 180                                      start >> PAGE_CACHE_SHIFT);
 181                 btrfs_set_file_extent_compression(leaf, ei, 0);
 182                 kaddr = kmap_atomic(page, KM_USER0);
 183                 offset = start & (PAGE_CACHE_SIZE - 1);
 184                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 185                 kunmap_atomic(kaddr, KM_USER0);
 186                 page_cache_release(page);
 187         }
 188         btrfs_mark_buffer_dirty(leaf);
 189         btrfs_free_path(path);
 190
 191         BTRFS_I(inode)->disk_i_size = inode->i_size;
 192         btrfs_update_inode(trans, root, inode);
 193         return 0;
 194 fail:
 195         btrfs_free_path(path);
 196         return err;
 197 }
 198
 199
 200 /*
 201  * conditionally insert an inline extent into the file.  This
 202  * does the checks required to make sure the data is small enough
 203  * to fit as an inline extent.
 204  */
 205 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 206                                  struct btrfs_root *root,
 207                                  struct inode *inode, u64 start, u64 end,
 208                                  size_t compressed_size,
 209                                  struct page **compressed_pages)
 210 {
 211         u64 isize = i_size_read(inode);
 212         u64 actual_end = min(end + 1, isize);
 213         u64 inline_len = actual_end - start;
 214         u64 aligned_end = (end + root->sectorsize - 1) &
 215                         ~((u64)root->sectorsize - 1);
 216         u64 hint_byte;
 217         u64 data_len = inline_len;
 218         int ret;
 219
 220         if (compressed_size)
 221                 data_len = compressed_size;
 222
 223         if (start > 0 ||
 224             actual_end >= PAGE_CACHE_SIZE ||
 225             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 226             (!compressed_size &&
 227             (actual_end & (root->sectorsize - 1)) == 0) ||
 228             end + 1 < isize ||
 229             data_len > root->fs_info->max_inline) {
 230                 return 1;
 231         }
 232
 233         ret = btrfs_drop_extents(trans, root, inode, start,
 234                                  aligned_end, aligned_end, start,
 235                                  &hint_byte, 1);
 236         BUG_ON(ret);
 237
 238         if (isize > actual_end)
 239                 inline_len = min_t(u64, isize, actual_end);
 240         ret = insert_inline_extent(trans, root, inode, start,
 241                                    inline_len, compressed_size,
 242                                    compressed_pages);
 243         BUG_ON(ret);
 244         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 245         return 0;
 246 }
 247
 248 struct async_extent {
 249         u64 start;
 250         u64 ram_size;
 251         u64 compressed_size;
 252         struct page **pages;
 253         unsigned long nr_pages;
 254         struct list_head list;
 255 };
 256
 257 struct async_cow {
 258         struct inode *inode;
 259         struct btrfs_root *root;
 260         struct page *locked_page;
 261         u64 start;
 262         u64 end;
 263         struct list_head extents;
 264         struct btrfs_work work;
 265 };
 266
 267 static noinline int add_async_extent(struct async_cow *cow,
 268                                      u64 start, u64 ram_size,
 269                                      u64 compressed_size,
 270                                      struct page **pages,
 271                                      unsigned long nr_pages)
 272 {
 273         struct async_extent *async_extent;
 274
 275         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 276         async_extent->start = start;
 277         async_extent->ram_size = ram_size;
 278         async_extent->compressed_size = compressed_size;
 279         async_extent->pages = pages;
 280         async_extent->nr_pages = nr_pages;
 281         list_add_tail(&async_extent->list, &cow->extents);
 282         return 0;
 283 }
 284
 285 /*
 286  * we create compressed extents in two phases.  The first
 287  * phase compresses a range of pages that have already been
 288  * locked (both pages and state bits are locked).
 289  *
 290  * This is done inside an ordered work queue, and the compression
 291  * is spread across many cpus.  The actual IO submission is step
 292  * two, and the ordered work queue takes care of making sure that
 293  * happens in the same order things were put onto the queue by
 294  * writepages and friends.
 295  *
 296  * If this code finds it can't get good compression, it puts an
 297  * entry onto the work queue to write the uncompressed bytes.  This
 298  * makes sure that both compressed inodes and uncompressed inodes
 299  * are written in the same order that pdflush sent them down.
 300  */
 301 static noinline int compress_file_range(struct inode *inode,
 302                                         struct page *locked_page,
 303                                         u64 start, u64 end,
 304                                         struct async_cow *async_cow,
 305                                         int *num_added)
 306 {
 307         struct btrfs_root *root = BTRFS_I(inode)->root;
 308         struct btrfs_trans_handle *trans;
 309         u64 num_bytes;
 310         u64 orig_start;
 311         u64 disk_num_bytes;
 312         u64 blocksize = root->sectorsize;
 313         u64 actual_end;
 314         u64 isize = i_size_read(inode);
 315         int ret = 0;
 316         struct page **pages = NULL;
 317         unsigned long nr_pages;
 318         unsigned long nr_pages_ret = 0;
 319         unsigned long total_compressed = 0;
 320         unsigned long total_in = 0;
 321         unsigned long max_compressed = 128 * 1024;
 322         unsigned long max_uncompressed = 128 * 1024;
 323         int i;
 324         int will_compress;
 325
 326         orig_start = start;
 327
 328         actual_end = min_t(u64, isize, end + 1);
 329 again:
 330         will_compress = 0;
 331         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 332         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 333
 334         /*
 335          * we don't want to send crud past the end of i_size through
 336          * compression, that's just a waste of CPU time.  So, if the
 337          * end of the file is before the start of our current
 338          * requested range of bytes, we bail out to the uncompressed
 339          * cleanup code that can deal with all of this.
 340          *
 341          * It isn't really the fastest way to fix things, but this is a
 342          * very uncommon corner.
 343          */
 344         if (actual_end <= start)
 345                 goto cleanup_and_bail_uncompressed;
 346
 347         total_compressed = actual_end - start;
 348
 349         /* we want to make sure that amount of ram required to uncompress
 350          * an extent is reasonable, so we limit the total size in ram
 351          * of a compressed extent to 128k.  This is a crucial number
 352          * because it also controls how easily we can spread reads across
 353          * cpus for decompression.
 354          *
 355          * We also want to make sure the amount of IO required to do
 356          * a random read is reasonably small, so we limit the size of
 357          * a compressed extent to 128k.
 358          */
 359         total_compressed = min(total_compressed, max_uncompressed);
 360         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 361         num_bytes = max(blocksize,  num_bytes);
 362         disk_num_bytes = num_bytes;
 363         total_in = 0;
 364         ret = 0;
 365
 366         /*
 367          * we do compression for mount -o compress and when the
 368          * inode has not been flagged as nocompress.  This flag can
 369          * change at any time if we discover bad compression ratios.
 370          */
 371         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 372             btrfs_test_opt(root, COMPRESS)) {
 373                 WARN_ON(pages);
 374                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 375
 376                 ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
 377                                                 total_compressed, pages,
 378                                                 nr_pages, &nr_pages_ret,
 379                                                 &total_in,
 380                                                 &total_compressed,
 381                                                 max_compressed);
 382
 383                 if (!ret) {
 384                         unsigned long offset = total_compressed &
 385                                 (PAGE_CACHE_SIZE - 1);
 386                         struct page *page = pages[nr_pages_ret - 1];
 387                         char *kaddr;
 388
 389                         /* zero the tail end of the last page, we might be
 390                          * sending it down to disk
 391                          */
 392                         if (offset) {
 393                                 kaddr = kmap_atomic(page, KM_USER0);
 394                                 memset(kaddr + offset, 0,
 395                                        PAGE_CACHE_SIZE - offset);
 396                                 kunmap_atomic(kaddr, KM_USER0);
 397                         }
 398                         will_compress = 1;
 399                 }
 400         }
 401         if (start == 0) {
 402                 trans = btrfs_join_transaction(root, 1);
 403                 BUG_ON(!trans);
 404                 btrfs_set_trans_block_group(trans, inode);
 405
 406                 /* lets try to make an inline extent */
 407                 if (ret || total_in < (actual_end - start)) {
 408                         /* we didn't compress the entire range, try
 409                          * to make an uncompressed inline extent.
 410                          */
 411                         ret = cow_file_range_inline(trans, root, inode,
 412                                                     start, end, 0, NULL);
 413                 } else {
 414                         /* try making a compressed inline extent */
 415                         ret = cow_file_range_inline(trans, root, inode,
 416                                                     start, end,
 417                                                     total_compressed, pages);
 418                 }
 419                 btrfs_end_transaction(trans, root);
 420                 if (ret == 0) {
 421                         /*
 422                          * inline extent creation worked, we don't need
 423                          * to create any more async work items.  Unlock
 424                          * and free up our temp pages.
 425                          */
 426                         extent_clear_unlock_delalloc(inode,
 427                              &BTRFS_I(inode)->io_tree,
 428                              start, end, NULL,
 429                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
 430                              EXTENT_CLEAR_DELALLOC |
 431                              EXTENT_CLEAR_ACCOUNTING |
 432                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
 433                         ret = 0;
 434                         goto free_pages_out;
 435                 }
 436         }
 437
 438         if (will_compress) {
 439                 /*
 440                  * we aren't doing an inline extent round the compressed size
 441                  * up to a block size boundary so the allocator does sane
 442                  * things
 443                  */
 444                 total_compressed = (total_compressed + blocksize - 1) &
 445                         ~(blocksize - 1);
 446
 447                 /*
 448                  * one last check to make sure the compression is really a
 449                  * win, compare the page count read with the blocks on disk
 450                  */
 451                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
 452                         ~(PAGE_CACHE_SIZE - 1);
 453                 if (total_compressed >= total_in) {
 454                         will_compress = 0;
 455                 } else {
 456                         disk_num_bytes = total_compressed;
 457                         num_bytes = total_in;
 458                 }
 459         }
 460         if (!will_compress && pages) {
 461                 /*
 462                  * the compression code ran but failed to make things smaller,
 463                  * free any pages it allocated and our page pointer array
 464                  */
 465                 for (i = 0; i < nr_pages_ret; i++) {
 466                         WARN_ON(pages[i]->mapping);
 467                         page_cache_release(pages[i]);
 468                 }
 469                 kfree(pages);
 470                 pages = NULL;
 471                 total_compressed = 0;
 472                 nr_pages_ret = 0;
 473
 474                 /* flag the file so we don't compress in the future */
 475                 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 476         }
 477         if (will_compress) {
 478                 *num_added += 1;
 479
 480                 /* the async work queues will take care of doing actual
 481                  * allocation on disk for these compressed pages,
 482                  * and will submit them to the elevator.
 483                  */
 484                 add_async_extent(async_cow, start, num_bytes,
 485                                  total_compressed, pages, nr_pages_ret);
 486
 487                 if (start + num_bytes < end && start + num_bytes < actual_end) {
 488                         start += num_bytes;
 489                         pages = NULL;
 490                         cond_resched();
 491                         goto again;
 492                 }
 493         } else {
 494 cleanup_and_bail_uncompressed:
 495                 /*
 496                  * No compression, but we still need to write the pages in
 497                  * the file we've been given so far.  redirty the locked
 498                  * page if it corresponds to our extent and set things up
 499                  * for the async work queue to run cow_file_range to do
 500                  * the normal delalloc dance
 501                  */
 502                 if (page_offset(locked_page) >= start &&
 503                     page_offset(locked_page) <= end) {
 504                         __set_page_dirty_nobuffers(locked_page);
 505                         /* unlocked later on in the async handlers */
 506                 }
 507                 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0);
 508                 *num_added += 1;
 509         }
 510
 511 out:
 512         return 0;
 513
 514 free_pages_out:
 515         for (i = 0; i < nr_pages_ret; i++) {
 516                 WARN_ON(pages[i]->mapping);
 517                 page_cache_release(pages[i]);
 518         }
 519         kfree(pages);
 520
 521         goto out;
 522 }
 523
 524 /*
 525  * phase two of compressed writeback.  This is the ordered portion
 526  * of the code, which only gets called in the order the work was
 527  * queued.  We walk all the async extents created by compress_file_range
 528  * and send them down to the disk.
 529  */
 530 static noinline int submit_compressed_extents(struct inode *inode,
 531                                               struct async_cow *async_cow)
 532 {
 533         struct async_extent *async_extent;
 534         u64 alloc_hint = 0;
 535         struct btrfs_trans_handle *trans;
 536         struct btrfs_key ins;
 537         struct extent_map *em;
 538         struct btrfs_root *root = BTRFS_I(inode)->root;
 539         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 540         struct extent_io_tree *io_tree;
 541         int ret;
 542
 543         if (list_empty(&async_cow->extents))
 544                 return 0;
 545
 546         trans = btrfs_join_transaction(root, 1);
 547
 548         while (!list_empty(&async_cow->extents)) {
 549                 async_extent = list_entry(async_cow->extents.next,
 550                                           struct async_extent, list);
 551                 list_del(&async_extent->list);
 552
 553                 io_tree = &BTRFS_I(inode)->io_tree;
 554
 555                 /* did the compression code fall back to uncompressed IO? */
 556                 if (!async_extent->pages) {
 557                         int page_started = 0;
 558                         unsigned long nr_written = 0;
 559
 560                         lock_extent(io_tree, async_extent->start,
 561                                     async_extent->start +
 562                                     async_extent->ram_size - 1, GFP_NOFS);
 563
 564                         /* allocate blocks */
 565                         cow_file_range(inode, async_cow->locked_page,
 566                                        async_extent->start,
 567                                        async_extent->start +
 568                                        async_extent->ram_size - 1,
 569                                        &page_started, &nr_written, 0);
 570
 571                         /*
 572                          * if page_started, cow_file_range inserted an
 573                          * inline extent and took care of all the unlocking
 574                          * and IO for us.  Otherwise, we need to submit
 575                          * all those pages down to the drive.
 576                          */
 577                         if (!page_started)
 578                                 extent_write_locked_range(io_tree,
 579                                                   inode, async_extent->start,
 580                                                   async_extent->start +
 581                                                   async_extent->ram_size - 1,
 582                                                   btrfs_get_extent,
 583                                                   WB_SYNC_ALL);
 584                         kfree(async_extent);
 585                         cond_resched();
 586                         continue;
 587                 }
 588
 589                 lock_extent(io_tree, async_extent->start,
 590                             async_extent->start + async_extent->ram_size - 1,
 591                             GFP_NOFS);
 592                 /*
 593                  * here we're doing allocation and writeback of the
 594                  * compressed pages
 595                  */
 596                 btrfs_drop_extent_cache(inode, async_extent->start,
 597                                         async_extent->start +
 598                                         async_extent->ram_size - 1, 0);
 599
 600                 ret = btrfs_reserve_extent(trans, root,
 601                                            async_extent->compressed_size,
 602                                            async_extent->compressed_size,
 603                                            0, alloc_hint,
 604                                            (u64)-1, &ins, 1);
 605                 BUG_ON(ret);
 606                 em = alloc_extent_map(GFP_NOFS);
 607                 em->start = async_extent->start;
 608                 em->len = async_extent->ram_size;
 609                 em->orig_start = em->start;
 610
 611                 em->block_start = ins.objectid;
 612                 em->block_len = ins.offset;
 613                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 614                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 615                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 616
 617                 while (1) {
 618                         write_lock(&em_tree->lock);
 619                         ret = add_extent_mapping(em_tree, em);
 620                         write_unlock(&em_tree->lock);
 621                         if (ret != -EEXIST) {
 622                                 free_extent_map(em);
 623                                 break;
 624                         }
 625                         btrfs_drop_extent_cache(inode, async_extent->start,
 626                                                 async_extent->start +
 627                                                 async_extent->ram_size - 1, 0);
 628                 }
 629
 630                 ret = btrfs_add_ordered_extent(inode, async_extent->start,
 631                                                ins.objectid,
 632                                                async_extent->ram_size,
 633                                                ins.offset,
 634                                                BTRFS_ORDERED_COMPRESSED);
 635                 BUG_ON(ret);
 636
 637                 btrfs_end_transaction(trans, root);
 638
 639                 /*
 640                  * clear dirty, set writeback and unlock the pages.
 641                  */
 642                 extent_clear_unlock_delalloc(inode,
 643                                 &BTRFS_I(inode)->io_tree,
 644                                 async_extent->start,
 645                                 async_extent->start +
 646                                 async_extent->ram_size - 1,
 647                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 648                                 EXTENT_CLEAR_UNLOCK |
 649                                 EXTENT_CLEAR_DELALLOC |
 650                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
 651
 652                 ret = btrfs_submit_compressed_write(inode,
 653                                     async_extent->start,
 654                                     async_extent->ram_size,
 655                                     ins.objectid,
 656                                     ins.offset, async_extent->pages,
 657                                     async_extent->nr_pages);
 658
 659                 BUG_ON(ret);
 660                 trans = btrfs_join_transaction(root, 1);
 661                 alloc_hint = ins.objectid + ins.offset;
 662                 kfree(async_extent);
 663                 cond_resched();
 664         }
 665
 666         btrfs_end_transaction(trans, root);
 667         return 0;
 668 }
 669
 670 /*
 671  * when extent_io.c finds a delayed allocation range in the file,
 672  * the call backs end up in this code.  The basic idea is to
 673  * allocate extents on disk for the range, and create ordered data structs
 674  * in ram to track those extents.
 675  *
 676  * locked_page is the page that writepage had locked already.  We use
 677  * it to make sure we don't do extra locks or unlocks.
 678  *
 679  * *page_started is set to one if we unlock locked_page and do everything
 680  * required to start IO on it.  It may be clean and already done with
 681  * IO when we return.
 682  */
 683 static noinline int cow_file_range(struct inode *inode,
 684                                    struct page *locked_page,
 685                                    u64 start, u64 end, int *page_started,
 686                                    unsigned long *nr_written,
 687                                    int unlock)
 688 {
 689         struct btrfs_root *root = BTRFS_I(inode)->root;
 690         struct btrfs_trans_handle *trans;
 691         u64 alloc_hint = 0;
 692         u64 num_bytes;
 693         unsigned long ram_size;
 694         u64 disk_num_bytes;
 695         u64 cur_alloc_size;
 696         u64 blocksize = root->sectorsize;
 697         u64 actual_end;
 698         u64 isize = i_size_read(inode);
 699         struct btrfs_key ins;
 700         struct extent_map *em;
 701         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 702         int ret = 0;
 703
 704         trans = btrfs_join_transaction(root, 1);
 705         BUG_ON(!trans);
 706         btrfs_set_trans_block_group(trans, inode);
 707
 708         actual_end = min_t(u64, isize, end + 1);
 709
 710         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 711         num_bytes = max(blocksize,  num_bytes);
 712         disk_num_bytes = num_bytes;
 713         ret = 0;
 714
 715         if (start == 0) {
 716                 /* lets try to make an inline extent */
 717                 ret = cow_file_range_inline(trans, root, inode,
 718                                             start, end, 0, NULL);
 719                 if (ret == 0) {
 720                         extent_clear_unlock_delalloc(inode,
 721                                      &BTRFS_I(inode)->io_tree,
 722                                      start, end, NULL,
 723                                      EXTENT_CLEAR_UNLOCK_PAGE |
 724                                      EXTENT_CLEAR_UNLOCK |
 725                                      EXTENT_CLEAR_DELALLOC |
 726                                      EXTENT_CLEAR_ACCOUNTING |
 727                                      EXTENT_CLEAR_DIRTY |
 728                                      EXTENT_SET_WRITEBACK |
 729                                      EXTENT_END_WRITEBACK);
 730                         *nr_written = *nr_written +
 731                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 732                         *page_started = 1;
 733                         ret = 0;
 734                         goto out;
 735                 }
 736         }
 737
 738         BUG_ON(disk_num_bytes >
 739                btrfs_super_total_bytes(&root->fs_info->super_copy));
 740
 741
 742         read_lock(&BTRFS_I(inode)->extent_tree.lock);
 743         em = search_extent_mapping(&BTRFS_I(inode)->extent_tree,
 744                                    start, num_bytes);
 745         if (em) {
 746                 alloc_hint = em->block_start;
 747                 free_extent_map(em);
 748         }
 749         read_unlock(&BTRFS_I(inode)->extent_tree.lock);
 750         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 751
 752         while (disk_num_bytes > 0) {
 753                 unsigned long op;
 754
 755                 cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent);
 756                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 757                                            root->sectorsize, 0, alloc_hint,
 758                                            (u64)-1, &ins, 1);
 759                 BUG_ON(ret);
 760
 761                 em = alloc_extent_map(GFP_NOFS);
 762                 em->start = start;
 763                 em->orig_start = em->start;
 764                 ram_size = ins.offset;
 765                 em->len = ins.offset;
 766
 767                 em->block_start = ins.objectid;
 768                 em->block_len = ins.offset;
 769                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 770                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 771
 772                 while (1) {
 773                         write_lock(&em_tree->lock);
 774                         ret = add_extent_mapping(em_tree, em);
 775                         write_unlock(&em_tree->lock);
 776                         if (ret != -EEXIST) {
 777                                 free_extent_map(em);
 778                                 break;
 779                         }
 780                         btrfs_drop_extent_cache(inode, start,
 781                                                 start + ram_size - 1, 0);
 782                 }
 783
 784                 cur_alloc_size = ins.offset;
 785                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 786                                                ram_size, cur_alloc_size, 0);
 787                 BUG_ON(ret);
 788
 789                 if (root->root_key.objectid ==
 790                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 791                         ret = btrfs_reloc_clone_csums(inode, start,
 792                                                       cur_alloc_size);
 793                         BUG_ON(ret);
 794                 }
 795
 796                 if (disk_num_bytes < cur_alloc_size)
 797                         break;
 798
 799                 /* we're not doing compressed IO, don't unlock the first
 800                  * page (which the caller expects to stay locked), don't
 801                  * clear any dirty bits and don't set any writeback bits
 802                  *
 803                  * Do set the Private2 bit so we know this page was properly
 804                  * setup for writepage
 805                  */
 806                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
 807                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
 808                         EXTENT_SET_PRIVATE2;
 809
 810                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 811                                              start, start + ram_size - 1,
 812                                              locked_page, op);
 813                 disk_num_bytes -= cur_alloc_size;
 814                 num_bytes -= cur_alloc_size;
 815                 alloc_hint = ins.objectid + ins.offset;
 816                 start += cur_alloc_size;
 817         }
 818 out:
 819         ret = 0;
 820         btrfs_end_transaction(trans, root);
 821
 822         return ret;
 823 }
 824
 825 /*
 826  * work queue call back to started compression on a file and pages
 827  */
 828 static noinline void async_cow_start(struct btrfs_work *work)
 829 {
 830         struct async_cow *async_cow;
 831         int num_added = 0;
 832         async_cow = container_of(work, struct async_cow, work);
 833
 834         compress_file_range(async_cow->inode, async_cow->locked_page,
 835                             async_cow->start, async_cow->end, async_cow,
 836                             &num_added);
 837         if (num_added == 0)
 838                 async_cow->inode = NULL;
 839 }
 840
 841 /*
 842  * work queue call back to submit previously compressed pages
 843  */
 844 static noinline void async_cow_submit(struct btrfs_work *work)
 845 {
 846         struct async_cow *async_cow;
 847         struct btrfs_root *root;
 848         unsigned long nr_pages;
 849
 850         async_cow = container_of(work, struct async_cow, work);
 851
 852         root = async_cow->root;
 853         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
 854                 PAGE_CACHE_SHIFT;
 855
 856         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
 857
 858         if (atomic_read(&root->fs_info->async_delalloc_pages) <
 859             5 * 1042 * 1024 &&
 860             waitqueue_active(&root->fs_info->async_submit_wait))
 861                 wake_up(&root->fs_info->async_submit_wait);
 862
 863         if (async_cow->inode)
 864                 submit_compressed_extents(async_cow->inode, async_cow);
 865 }
 866
 867 static noinline void async_cow_free(struct btrfs_work *work)
 868 {
 869         struct async_cow *async_cow;
 870         async_cow = container_of(work, struct async_cow, work);
 871         kfree(async_cow);
 872 }
 873
 874 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
 875                                 u64 start, u64 end, int *page_started,
 876                                 unsigned long *nr_written)
 877 {
 878         struct async_cow *async_cow;
 879         struct btrfs_root *root = BTRFS_I(inode)->root;
 880         unsigned long nr_pages;
 881         u64 cur_end;
 882         int limit = 10 * 1024 * 1042;
 883
 884         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
 885                          1, 0, NULL, GFP_NOFS);
 886         while (start < end) {
 887                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
 888                 async_cow->inode = inode;
 889                 async_cow->root = root;
 890                 async_cow->locked_page = locked_page;
 891                 async_cow->start = start;
 892
 893                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
 894                         cur_end = end;
 895                 else
 896                         cur_end = min(end, start + 512 * 1024 - 1);
 897
 898                 async_cow->end = cur_end;
 899                 INIT_LIST_HEAD(&async_cow->extents);
 900
 901                 async_cow->work.func = async_cow_start;
 902                 async_cow->work.ordered_func = async_cow_submit;
 903                 async_cow->work.ordered_free = async_cow_free;
 904                 async_cow->work.flags = 0;
 905
 906                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
 907                         PAGE_CACHE_SHIFT;
 908                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
 909
 910                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
 911                                    &async_cow->work);
 912
 913                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
 914                         wait_event(root->fs_info->async_submit_wait,
 915                            (atomic_read(&root->fs_info->async_delalloc_pages) <
 916                             limit));
 917                 }
 918
 919                 while (atomic_read(&root->fs_info->async_submit_draining) &&
 920                       atomic_read(&root->fs_info->async_delalloc_pages)) {
 921                         wait_event(root->fs_info->async_submit_wait,
 922                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
 923                            0));
 924                 }
 925
 926                 *nr_written += nr_pages;
 927                 start = cur_end + 1;
 928         }
 929         *page_started = 1;
 930         return 0;
 931 }
 932
 933 static noinline int csum_exist_in_range(struct btrfs_root *root,
 934                                         u64 bytenr, u64 num_bytes)
 935 {
 936         int ret;
 937         struct btrfs_ordered_sum *sums;
 938         LIST_HEAD(list);
 939
 940         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
 941                                        bytenr + num_bytes - 1, &list);
 942         if (ret == 0 && list_empty(&list))
 943                 return 0;
 944
 945         while (!list_empty(&list)) {
 946                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
 947                 list_del(&sums->list);
 948                 kfree(sums);
 949         }
 950         return 1;
 951 }
 952
 953 /*
 954  * when nowcow writeback call back.  This checks for snapshots or COW copies
 955  * of the extents that exist in the file, and COWs the file as required.
 956  *
 957  * If no cow copies or snapshots exist, we write directly to the existing
 958  * blocks on disk
 959  */
 960 static noinline int run_delalloc_nocow(struct inode *inode,
 961                                        struct page *locked_page,
 962                               u64 start, u64 end, int *page_started, int force,
 963                               unsigned long *nr_written)
 964 {
 965         struct btrfs_root *root = BTRFS_I(inode)->root;
 966         struct btrfs_trans_handle *trans;
 967         struct extent_buffer *leaf;
 968         struct btrfs_path *path;
 969         struct btrfs_file_extent_item *fi;
 970         struct btrfs_key found_key;
 971         u64 cow_start;
 972         u64 cur_offset;
 973         u64 extent_end;
 974         u64 extent_offset;
 975         u64 disk_bytenr;
 976         u64 num_bytes;
 977         int extent_type;
 978         int ret;
 979         int type;
 980         int nocow;
 981         int check_prev = 1;
 982
 983         path = btrfs_alloc_path();
 984         BUG_ON(!path);
 985         trans = btrfs_join_transaction(root, 1);
 986         BUG_ON(!trans);
 987
 988         cow_start = (u64)-1;
 989         cur_offset = start;
 990         while (1) {
 991                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
 992                                                cur_offset, 0);
 993                 BUG_ON(ret < 0);
 994                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
 995                         leaf = path->nodes[0];
 996                         btrfs_item_key_to_cpu(leaf, &found_key,
 997                                               path->slots[0] - 1);
 998                         if (found_key.objectid == inode->i_ino &&
 999                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1000                                 path->slots[0]--;
1001                 }
1002                 check_prev = 0;
1003 next_slot:
1004                 leaf = path->nodes[0];
1005                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1006                         ret = btrfs_next_leaf(root, path);
1007                         if (ret < 0)
1008                                 BUG_ON(1);
1009                         if (ret > 0)
1010                                 break;
1011                         leaf = path->nodes[0];
1012                 }
1013
1014                 nocow = 0;
1015                 disk_bytenr = 0;
1016                 num_bytes = 0;
1017                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1018
1019                 if (found_key.objectid > inode->i_ino ||
1020                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1021                     found_key.offset > end)
1022                         break;
1023
1024                 if (found_key.offset > cur_offset) {
1025                         extent_end = found_key.offset;
1026                         extent_type = 0;
1027                         goto out_check;
1028                 }
1029
1030                 fi = btrfs_item_ptr(leaf, path->slots[0],
1031                                     struct btrfs_file_extent_item);
1032                 extent_type = btrfs_file_extent_type(leaf, fi);
1033
1034                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1035                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1036                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1037                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1038                         extent_end = found_key.offset +
1039                                 btrfs_file_extent_num_bytes(leaf, fi);
1040                         if (extent_end <= start) {
1041                                 path->slots[0]++;
1042                                 goto next_slot;
1043                         }
1044                         if (disk_bytenr == 0)
1045                                 goto out_check;
1046                         if (btrfs_file_extent_compression(leaf, fi) ||
1047                             btrfs_file_extent_encryption(leaf, fi) ||
1048                             btrfs_file_extent_other_encoding(leaf, fi))
1049                                 goto out_check;
1050                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1051                                 goto out_check;
1052                         if (btrfs_extent_readonly(root, disk_bytenr))
1053                                 goto out_check;
1054                         if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1055                                                   found_key.offset -
1056                                                   extent_offset, disk_bytenr))
1057                                 goto out_check;
1058                         disk_bytenr += extent_offset;
1059                         disk_bytenr += cur_offset - found_key.offset;
1060                         num_bytes = min(end + 1, extent_end) - cur_offset;
1061                         /*
1062                          * force cow if csum exists in the range.
1063                          * this ensure that csum for a given extent are
1064                          * either valid or do not exist.
1065                          */
1066                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1067                                 goto out_check;
1068                         nocow = 1;
1069                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1070                         extent_end = found_key.offset +
1071                                 btrfs_file_extent_inline_len(leaf, fi);
1072                         extent_end = ALIGN(extent_end, root->sectorsize);
1073                 } else {
1074                         BUG_ON(1);
1075                 }
1076 out_check:
1077                 if (extent_end <= start) {
1078                         path->slots[0]++;
1079                         goto next_slot;
1080                 }
1081                 if (!nocow) {
1082                         if (cow_start == (u64)-1)
1083                                 cow_start = cur_offset;
1084                         cur_offset = extent_end;
1085                         if (cur_offset > end)
1086                                 break;
1087                         path->slots[0]++;
1088                         goto next_slot;
1089                 }
1090
1091                 btrfs_release_path(root, path);
1092                 if (cow_start != (u64)-1) {
1093                         ret = cow_file_range(inode, locked_page, cow_start,
1094                                         found_key.offset - 1, page_started,
1095                                         nr_written, 1);
1096                         BUG_ON(ret);
1097                         cow_start = (u64)-1;
1098                 }
1099
1100                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1101                         struct extent_map *em;
1102                         struct extent_map_tree *em_tree;
1103                         em_tree = &BTRFS_I(inode)->extent_tree;
1104                         em = alloc_extent_map(GFP_NOFS);
1105                         em->start = cur_offset;
1106                         em->orig_start = em->start;
1107                         em->len = num_bytes;
1108                         em->block_len = num_bytes;
1109                         em->block_start = disk_bytenr;
1110                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1111                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1112                         while (1) {
1113                                 write_lock(&em_tree->lock);
1114                                 ret = add_extent_mapping(em_tree, em);
1115                                 write_unlock(&em_tree->lock);
1116                                 if (ret != -EEXIST) {
1117                                         free_extent_map(em);
1118                                         break;
1119                                 }
1120                                 btrfs_drop_extent_cache(inode, em->start,
1121                                                 em->start + em->len - 1, 0);
1122                         }
1123                         type = BTRFS_ORDERED_PREALLOC;
1124                 } else {
1125                         type = BTRFS_ORDERED_NOCOW;
1126                 }
1127
1128                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1129                                                num_bytes, num_bytes, type);
1130                 BUG_ON(ret);
1131
1132                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1133                                 cur_offset, cur_offset + num_bytes - 1,
1134                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1135                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1136                                 EXTENT_SET_PRIVATE2);
1137                 cur_offset = extent_end;
1138                 if (cur_offset > end)
1139                         break;
1140         }
1141         btrfs_release_path(root, path);
1142
1143         if (cur_offset <= end && cow_start == (u64)-1)
1144                 cow_start = cur_offset;
1145         if (cow_start != (u64)-1) {
1146                 ret = cow_file_range(inode, locked_page, cow_start, end,
1147                                      page_started, nr_written, 1);
1148                 BUG_ON(ret);
1149         }
1150
1151         ret = btrfs_end_transaction(trans, root);
1152         BUG_ON(ret);
1153         btrfs_free_path(path);
1154         return 0;
1155 }
1156
1157 /*
1158  * extent_io.c call back to do delayed allocation processing
1159  */
1160 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1161                               u64 start, u64 end, int *page_started,
1162                               unsigned long *nr_written)
1163 {
1164         int ret;
1165         struct btrfs_root *root = BTRFS_I(inode)->root;
1166
1167         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1168                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1169                                          page_started, 1, nr_written);
1170         else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1171                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1172                                          page_started, 0, nr_written);
1173         else if (!btrfs_test_opt(root, COMPRESS))
1174                 ret = cow_file_range(inode, locked_page, start, end,
1175                                       page_started, nr_written, 1);
1176         else
1177                 ret = cow_file_range_async(inode, locked_page, start, end,
1178                                            page_started, nr_written);
1179         return ret;
1180 }
1181
1182 static int btrfs_split_extent_hook(struct inode *inode,
1183                                     struct extent_state *orig, u64 split)
1184 {
1185         struct btrfs_root *root = BTRFS_I(inode)->root;
1186         u64 size;
1187
1188         if (!(orig->state & EXTENT_DELALLOC))
1189                 return 0;
1190
1191         size = orig->end - orig->start + 1;
1192         if (size > root->fs_info->max_extent) {
1193                 u64 num_extents;
1194                 u64 new_size;
1195
1196                 new_size = orig->end - split + 1;
1197                 num_extents = div64_u64(size + root->fs_info->max_extent - 1,
1198                                         root->fs_info->max_extent);
1199
1200                 /*
1201                  * if we break a large extent up then leave oustanding_extents
1202                  * be, since we've already accounted for the large extent.
1203                  */
1204                 if (div64_u64(new_size + root->fs_info->max_extent - 1,
1205                               root->fs_info->max_extent) < num_extents)
1206                         return 0;
1207         }
1208
1209         spin_lock(&BTRFS_I(inode)->accounting_lock);
1210         BTRFS_I(inode)->outstanding_extents++;
1211         spin_unlock(&BTRFS_I(inode)->accounting_lock);
1212
1213         return 0;
1214 }
1215
1216 /*
1217  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1218  * extents so we can keep track of new extents that are just merged onto old
1219  * extents, such as when we are doing sequential writes, so we can properly
1220  * account for the metadata space we'll need.
1221  */
1222 static int btrfs_merge_extent_hook(struct inode *inode,
1223                                    struct extent_state *new,
1224                                    struct extent_state *other)
1225 {
1226         struct btrfs_root *root = BTRFS_I(inode)->root;
1227         u64 new_size, old_size;
1228         u64 num_extents;
1229
1230         /* not delalloc, ignore it */
1231         if (!(other->state & EXTENT_DELALLOC))
1232                 return 0;
1233
1234         old_size = other->end - other->start + 1;
1235         if (new->start < other->start)
1236                 new_size = other->end - new->start + 1;
1237         else
1238                 new_size = new->end - other->start + 1;
1239
1240         /* we're not bigger than the max, unreserve the space and go */
1241         if (new_size <= root->fs_info->max_extent) {
1242                 spin_lock(&BTRFS_I(inode)->accounting_lock);
1243                 BTRFS_I(inode)->outstanding_extents--;
1244                 spin_unlock(&BTRFS_I(inode)->accounting_lock);
1245                 return 0;
1246         }
1247
1248         /*
1249          * If we grew by another max_extent, just return, we want to keep that
1250          * reserved amount.
1251          */
1252         num_extents = div64_u64(old_size + root->fs_info->max_extent - 1,
1253                                 root->fs_info->max_extent);
1254         if (div64_u64(new_size + root->fs_info->max_extent - 1,
1255                       root->fs_info->max_extent) > num_extents)
1256                 return 0;
1257
1258         spin_lock(&BTRFS_I(inode)->accounting_lock);
1259         BTRFS_I(inode)->outstanding_extents--;
1260         spin_unlock(&BTRFS_I(inode)->accounting_lock);
1261
1262         return 0;
1263 }
1264
1265 /*
1266  * extent_io.c set_bit_hook, used to track delayed allocation
1267  * bytes in this file, and to maintain the list of inodes that
1268  * have pending delalloc work to be done.
1269  */
1270 static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
1271                        unsigned long old, unsigned long bits)
1272 {
1273
1274         /*
1275          * set_bit and clear bit hooks normally require _irqsave/restore
1276          * but in this case, we are only testeing for the DELALLOC
1277          * bit, which is only set or cleared with irqs on
1278          */
1279         if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1280                 struct btrfs_root *root = BTRFS_I(inode)->root;
1281
1282                 spin_lock(&BTRFS_I(inode)->accounting_lock);
1283                 BTRFS_I(inode)->outstanding_extents++;
1284                 spin_unlock(&BTRFS_I(inode)->accounting_lock);
1285                 btrfs_delalloc_reserve_space(root, inode, end - start + 1);
1286                 spin_lock(&root->fs_info->delalloc_lock);
1287                 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
1288                 root->fs_info->delalloc_bytes += end - start + 1;
1289                 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1290                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1291                                       &root->fs_info->delalloc_inodes);
1292                 }
1293                 spin_unlock(&root->fs_info->delalloc_lock);
1294         }
1295         return 0;
1296 }
1297
1298 /*
1299  * extent_io.c clear_bit_hook, see set_bit_hook for why
1300  */
1301 static int btrfs_clear_bit_hook(struct inode *inode,
1302                                 struct extent_state *state, unsigned long bits)
1303 {
1304         /*
1305          * set_bit and clear bit hooks normally require _irqsave/restore
1306          * but in this case, we are only testeing for the DELALLOC
1307          * bit, which is only set or cleared with irqs on
1308          */
1309         if ((state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1310                 struct btrfs_root *root = BTRFS_I(inode)->root;
1311
1312                 if (bits & EXTENT_DO_ACCOUNTING) {
1313                         spin_lock(&BTRFS_I(inode)->accounting_lock);
1314                         BTRFS_I(inode)->outstanding_extents--;
1315                         spin_unlock(&BTRFS_I(inode)->accounting_lock);
1316                         btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
1317                 }
1318
1319                 spin_lock(&root->fs_info->delalloc_lock);
1320                 if (state->end - state->start + 1 >
1321                     root->fs_info->delalloc_bytes) {
1322                         printk(KERN_INFO "btrfs warning: delalloc account "
1323                                "%llu %llu\n",
1324                                (unsigned long long)
1325                                state->end - state->start + 1,
1326                                (unsigned long long)
1327                                root->fs_info->delalloc_bytes);
1328                         btrfs_delalloc_free_space(root, inode, (u64)-1);
1329                         root->fs_info->delalloc_bytes = 0;
1330                         BTRFS_I(inode)->delalloc_bytes = 0;
1331                 } else {
1332                         btrfs_delalloc_free_space(root, inode,
1333                                                   state->end -
1334                                                   state->start + 1);
1335                         root->fs_info->delalloc_bytes -= state->end -
1336                                 state->start + 1;
1337                         BTRFS_I(inode)->delalloc_bytes -= state->end -
1338                                 state->start + 1;
1339                 }
1340                 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
1341                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1342                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1343                 }
1344                 spin_unlock(&root->fs_info->delalloc_lock);
1345         }
1346         return 0;
1347 }
1348
1349 /*
1350  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1351  * we don't create bios that span stripes or chunks
1352  */
1353 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1354                          size_t size, struct bio *bio,
1355                          unsigned long bio_flags)
1356 {
1357         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1358         struct btrfs_mapping_tree *map_tree;
1359         u64 logical = (u64)bio->bi_sector << 9;
1360         u64 length = 0;
1361         u64 map_length;
1362         int ret;
1363
1364         if (bio_flags & EXTENT_BIO_COMPRESSED)
1365                 return 0;
1366
1367         length = bio->bi_size;
1368         map_tree = &root->fs_info->mapping_tree;
1369         map_length = length;
1370         ret = btrfs_map_block(map_tree, READ, logical,
1371                               &map_length, NULL, 0);
1372
1373         if (map_length < length + size)
1374                 return 1;
1375         return 0;
1376 }
1377
1378 /*
1379  * in order to insert checksums into the metadata in large chunks,
1380  * we wait until bio submission time.   All the pages in the bio are
1381  * checksummed and sums are attached onto the ordered extent record.
1382  *
1383  * At IO completion time the cums attached on the ordered extent record
1384  * are inserted into the btree
1385  */
1386 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1387                                     struct bio *bio, int mirror_num,
1388                                     unsigned long bio_flags)
1389 {
1390         struct btrfs_root *root = BTRFS_I(inode)->root;
1391         int ret = 0;
1392
1393         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1394         BUG_ON(ret);
1395         return 0;
1396 }
1397
1398 /*
1399  * in order to insert checksums into the metadata in large chunks,
1400  * we wait until bio submission time.   All the pages in the bio are
1401  * checksummed and sums are attached onto the ordered extent record.
1402  *
1403  * At IO completion time the cums attached on the ordered extent record
1404  * are inserted into the btree
1405  */
1406 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1407                           int mirror_num, unsigned long bio_flags)
1408 {
1409         struct btrfs_root *root = BTRFS_I(inode)->root;
1410         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1411 }
1412
1413 /*
1414  * extent_io.c submission hook. This does the right thing for csum calculation
1415  * on write, or reading the csums from the tree before a read
1416  */
1417 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1418                           int mirror_num, unsigned long bio_flags)
1419 {
1420         struct btrfs_root *root = BTRFS_I(inode)->root;
1421         int ret = 0;
1422         int skip_sum;
1423
1424         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1425
1426         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1427         BUG_ON(ret);
1428
1429         if (!(rw & (1 << BIO_RW))) {
1430                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1431                         return btrfs_submit_compressed_read(inode, bio,
1432                                                     mirror_num, bio_flags);
1433                 } else if (!skip_sum)
1434                         btrfs_lookup_bio_sums(root, inode, bio, NULL);
1435                 goto mapit;
1436         } else if (!skip_sum) {
1437                 /* csum items have already been cloned */
1438                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1439                         goto mapit;
1440                 /* we're doing a write, do the async checksumming */
1441                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1442                                    inode, rw, bio, mirror_num,
1443                                    bio_flags, __btrfs_submit_bio_start,
1444                                    __btrfs_submit_bio_done);
1445         }
1446
1447 mapit:
1448         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1449 }
1450
1451 /*
1452  * given a list of ordered sums record them in the inode.  This happens
1453  * at IO completion time based on sums calculated at bio submission time.
1454  */
1455 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1456                              struct inode *inode, u64 file_offset,
1457                              struct list_head *list)
1458 {
1459         struct btrfs_ordered_sum *sum;
1460
1461         btrfs_set_trans_block_group(trans, inode);
1462
1463         list_for_each_entry(sum, list, list) {
1464                 btrfs_csum_file_blocks(trans,
1465                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1466         }
1467         return 0;
1468 }
1469
1470 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end)
1471 {
1472         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1473                 WARN_ON(1);
1474         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1475                                    GFP_NOFS);
1476 }
1477
1478 /* see btrfs_writepage_start_hook for details on why this is required */
1479 struct btrfs_writepage_fixup {
1480         struct page *page;
1481         struct btrfs_work work;
1482 };
1483
1484 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1485 {
1486         struct btrfs_writepage_fixup *fixup;
1487         struct btrfs_ordered_extent *ordered;
1488         struct page *page;
1489         struct inode *inode;
1490         u64 page_start;
1491         u64 page_end;
1492
1493         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1494         page = fixup->page;
1495 again:
1496         lock_page(page);
1497         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1498                 ClearPageChecked(page);
1499                 goto out_page;
1500         }
1501
1502         inode = page->mapping->host;
1503         page_start = page_offset(page);
1504         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1505
1506         lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1507
1508         /* already ordered? We're done */
1509         if (PagePrivate2(page))
1510                 goto out;
1511
1512         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1513         if (ordered) {
1514                 unlock_extent(&BTRFS_I(inode)->io_tree, page_start,
1515                               page_end, GFP_NOFS);
1516                 unlock_page(page);
1517                 btrfs_start_ordered_extent(inode, ordered, 1);
1518                 goto again;
1519         }
1520
1521         btrfs_set_extent_delalloc(inode, page_start, page_end);
1522         ClearPageChecked(page);
1523 out:
1524         unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1525 out_page:
1526         unlock_page(page);
1527         page_cache_release(page);
1528 }
1529
1530 /*
1531  * There are a few paths in the higher layers of the kernel that directly
1532  * set the page dirty bit without asking the filesystem if it is a
1533  * good idea.  This causes problems because we want to make sure COW
1534  * properly happens and the data=ordered rules are followed.
1535  *
1536  * In our case any range that doesn't have the ORDERED bit set
1537  * hasn't been properly setup for IO.  We kick off an async process
1538  * to fix it up.  The async helper will wait for ordered extents, set
1539  * the delalloc bit and make it safe to write the page.
1540  */
1541 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1542 {
1543         struct inode *inode = page->mapping->host;
1544         struct btrfs_writepage_fixup *fixup;
1545         struct btrfs_root *root = BTRFS_I(inode)->root;
1546
1547         /* this page is properly in the ordered list */
1548         if (TestClearPagePrivate2(page))
1549                 return 0;
1550
1551         if (PageChecked(page))
1552                 return -EAGAIN;
1553
1554         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1555         if (!fixup)
1556                 return -EAGAIN;
1557
1558         SetPageChecked(page);
1559         page_cache_get(page);
1560         fixup->work.func = btrfs_writepage_fixup_worker;
1561         fixup->page = page;
1562         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1563         return -EAGAIN;
1564 }
1565
1566 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1567                                        struct inode *inode, u64 file_pos,
1568                                        u64 disk_bytenr, u64 disk_num_bytes,
1569                                        u64 num_bytes, u64 ram_bytes,
1570                                        u64 locked_end,
1571                                        u8 compression, u8 encryption,
1572                                        u16 other_encoding, int extent_type)
1573 {
1574         struct btrfs_root *root = BTRFS_I(inode)->root;
1575         struct btrfs_file_extent_item *fi;
1576         struct btrfs_path *path;
1577         struct extent_buffer *leaf;
1578         struct btrfs_key ins;
1579         u64 hint;
1580         int ret;
1581
1582         path = btrfs_alloc_path();
1583         BUG_ON(!path);
1584
1585         path->leave_spinning = 1;
1586
1587         /*
1588          * we may be replacing one extent in the tree with another.
1589          * The new extent is pinned in the extent map, and we don't want
1590          * to drop it from the cache until it is completely in the btree.
1591          *
1592          * So, tell btrfs_drop_extents to leave this extent in the cache.
1593          * the caller is expected to unpin it and allow it to be merged
1594          * with the others.
1595          */
1596         ret = btrfs_drop_extents(trans, root, inode, file_pos,
1597                                  file_pos + num_bytes, locked_end,
1598                                  file_pos, &hint, 0);
1599         BUG_ON(ret);
1600
1601         ins.objectid = inode->i_ino;
1602         ins.offset = file_pos;
1603         ins.type = BTRFS_EXTENT_DATA_KEY;
1604         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1605         BUG_ON(ret);
1606         leaf = path->nodes[0];
1607         fi = btrfs_item_ptr(leaf, path->slots[0],
1608                             struct btrfs_file_extent_item);
1609         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1610         btrfs_set_file_extent_type(leaf, fi, extent_type);
1611         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1612         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1613         btrfs_set_file_extent_offset(leaf, fi, 0);
1614         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1615         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1616         btrfs_set_file_extent_compression(leaf, fi, compression);
1617         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1618         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1619
1620         btrfs_unlock_up_safe(path, 1);
1621         btrfs_set_lock_blocking(leaf);
1622
1623         btrfs_mark_buffer_dirty(leaf);
1624
1625         inode_add_bytes(inode, num_bytes);
1626
1627         ins.objectid = disk_bytenr;
1628         ins.offset = disk_num_bytes;
1629         ins.type = BTRFS_EXTENT_ITEM_KEY;
1630         ret = btrfs_alloc_reserved_file_extent(trans, root,
1631                                         root->root_key.objectid,
1632                                         inode->i_ino, file_pos, &ins);
1633         BUG_ON(ret);
1634         btrfs_free_path(path);
1635
1636         return 0;
1637 }
1638
1639 /*
1640  * helper function for btrfs_finish_ordered_io, this
1641  * just reads in some of the csum leaves to prime them into ram
1642  * before we start the transaction.  It limits the amount of btree
1643  * reads required while inside the transaction.
1644  */
1645 static noinline void reada_csum(struct btrfs_root *root,
1646                                 struct btrfs_path *path,
1647                                 struct btrfs_ordered_extent *ordered_extent)
1648 {
1649         struct btrfs_ordered_sum *sum;
1650         u64 bytenr;
1651
1652         sum = list_entry(ordered_extent->list.next, struct btrfs_ordered_sum,
1653                          list);
1654         bytenr = sum->sums[0].bytenr;
1655
1656         /*
1657          * we don't care about the results, the point of this search is
1658          * just to get the btree leaves into ram
1659          */
1660         btrfs_lookup_csum(NULL, root->fs_info->csum_root, path, bytenr, 0);
1661 }
1662
1663 /* as ordered data IO finishes, this gets called so we can finish
1664  * an ordered extent if the range of bytes in the file it covers are
1665  * fully written.
1666  */
1667 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1668 {
1669         struct btrfs_root *root = BTRFS_I(inode)->root;
1670         struct btrfs_trans_handle *trans;
1671         struct btrfs_ordered_extent *ordered_extent = NULL;
1672         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1673         struct btrfs_path *path;
1674         int compressed = 0;
1675         int ret;
1676
1677         ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
1678         if (!ret)
1679                 return 0;
1680
1681         /*
1682          * before we join the transaction, try to do some of our IO.
1683          * This will limit the amount of IO that we have to do with
1684          * the transaction running.  We're unlikely to need to do any
1685          * IO if the file extents are new, the disk_i_size checks
1686          * covers the most common case.
1687          */
1688         if (start < BTRFS_I(inode)->disk_i_size) {
1689                 path = btrfs_alloc_path();
1690                 if (path) {
1691                         ret = btrfs_lookup_file_extent(NULL, root, path,
1692                                                        inode->i_ino,
1693                                                        start, 0);
1694                         ordered_extent = btrfs_lookup_ordered_extent(inode,
1695                                                                      start);
1696                         if (!list_empty(&ordered_extent->list)) {
1697                                 btrfs_release_path(root, path);
1698                                 reada_csum(root, path, ordered_extent);
1699                         }
1700                         btrfs_free_path(path);
1701                 }
1702         }
1703
1704         trans = btrfs_join_transaction(root, 1);
1705
1706         if (!ordered_extent)
1707                 ordered_extent = btrfs_lookup_ordered_extent(inode, start);
1708         BUG_ON(!ordered_extent);
1709         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
1710                 goto nocow;
1711
1712         lock_extent(io_tree, ordered_extent->file_offset,
1713                     ordered_extent->file_offset + ordered_extent->len - 1,
1714                     GFP_NOFS);
1715
1716         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1717                 compressed = 1;
1718         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1719                 BUG_ON(compressed);
1720                 ret = btrfs_mark_extent_written(trans, root, inode,
1721                                                 ordered_extent->file_offset,
1722                                                 ordered_extent->file_offset +
1723                                                 ordered_extent->len);
1724                 BUG_ON(ret);
1725         } else {
1726                 ret = insert_reserved_file_extent(trans, inode,
1727                                                 ordered_extent->file_offset,
1728                                                 ordered_extent->start,
1729                                                 ordered_extent->disk_len,
1730                                                 ordered_extent->len,
1731                                                 ordered_extent->len,
1732                                                 ordered_extent->file_offset +
1733                                                 ordered_extent->len,
1734                                                 compressed, 0, 0,
1735                                                 BTRFS_FILE_EXTENT_REG);
1736                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1737                                    ordered_extent->file_offset,
1738                                    ordered_extent->len);
1739                 BUG_ON(ret);
1740         }
1741         unlock_extent(io_tree, ordered_extent->file_offset,
1742                     ordered_extent->file_offset + ordered_extent->len - 1,
1743                     GFP_NOFS);
1744 nocow:
1745         add_pending_csums(trans, inode, ordered_extent->file_offset,
1746                           &ordered_extent->list);
1747
1748         mutex_lock(&BTRFS_I(inode)->extent_mutex);
1749         btrfs_ordered_update_i_size(inode, ordered_extent);
1750         btrfs_update_inode(trans, root, inode);
1751         btrfs_remove_ordered_extent(inode, ordered_extent);
1752         mutex_unlock(&BTRFS_I(inode)->extent_mutex);
1753
1754         /* once for us */
1755         btrfs_put_ordered_extent(ordered_extent);
1756         /* once for the tree */
1757         btrfs_put_ordered_extent(ordered_extent);
1758
1759         btrfs_end_transaction(trans, root);
1760         return 0;
1761 }
1762
1763 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1764                                 struct extent_state *state, int uptodate)
1765 {
1766         ClearPagePrivate2(page);
1767         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1768 }
1769
1770 /*
1771  * When IO fails, either with EIO or csum verification fails, we
1772  * try other mirrors that might have a good copy of the data.  This
1773  * io_failure_record is used to record state as we go through all the
1774  * mirrors.  If another mirror has good data, the page is set up to date
1775  * and things continue.  If a good mirror can't be found, the original
1776  * bio end_io callback is called to indicate things have failed.
1777  */
1778 struct io_failure_record {
1779         struct page *page;
1780         u64 start;
1781         u64 len;
1782         u64 logical;
1783         unsigned long bio_flags;
1784         int last_mirror;
1785 };
1786
1787 static int btrfs_io_failed_hook(struct bio *failed_bio,
1788                          struct page *page, u64 start, u64 end,
1789                          struct extent_state *state)
1790 {
1791         struct io_failure_record *failrec = NULL;
1792         u64 private;
1793         struct extent_map *em;
1794         struct inode *inode = page->mapping->host;
1795         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1796         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1797         struct bio *bio;
1798         int num_copies;
1799         int ret;
1800         int rw;
1801         u64 logical;
1802
1803         ret = get_state_private(failure_tree, start, &private);
1804         if (ret) {
1805                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1806                 if (!failrec)
1807                         return -ENOMEM;
1808                 failrec->start = start;
1809                 failrec->len = end - start + 1;
1810                 failrec->last_mirror = 0;
1811                 failrec->bio_flags = 0;
1812
1813                 read_lock(&em_tree->lock);
1814                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1815                 if (em->start > start || em->start + em->len < start) {
1816                         free_extent_map(em);
1817                         em = NULL;
1818                 }
1819                 read_unlock(&em_tree->lock);
1820
1821                 if (!em || IS_ERR(em)) {
1822                         kfree(failrec);
1823                         return -EIO;
1824                 }
1825                 logical = start - em->start;
1826                 logical = em->block_start + logical;
1827                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1828                         logical = em->block_start;
1829                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1830                 }
1831                 failrec->logical = logical;
1832                 free_extent_map(em);
1833                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1834                                 EXTENT_DIRTY, GFP_NOFS);
1835                 set_state_private(failure_tree, start,
1836                                  (u64)(unsigned long)failrec);
1837         } else {
1838                 failrec = (struct io_failure_record *)(unsigned long)private;
1839         }
1840         num_copies = btrfs_num_copies(
1841                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1842                               failrec->logical, failrec->len);
1843         failrec->last_mirror++;
1844         if (!state) {
1845                 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1846                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1847                                                     failrec->start,
1848                                                     EXTENT_LOCKED);
1849                 if (state && state->start != failrec->start)
1850                         state = NULL;
1851                 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1852         }
1853         if (!state || failrec->last_mirror > num_copies) {
1854                 set_state_private(failure_tree, failrec->start, 0);
1855                 clear_extent_bits(failure_tree, failrec->start,
1856                                   failrec->start + failrec->len - 1,
1857                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1858                 kfree(failrec);
1859                 return -EIO;
1860         }
1861         bio = bio_alloc(GFP_NOFS, 1);
1862         bio->bi_private = state;
1863         bio->bi_end_io = failed_bio->bi_end_io;
1864         bio->bi_sector = failrec->logical >> 9;
1865         bio->bi_bdev = failed_bio->bi_bdev;
1866         bio->bi_size = 0;
1867
1868         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1869         if (failed_bio->bi_rw & (1 << BIO_RW))
1870                 rw = WRITE;
1871         else
1872                 rw = READ;
1873
1874         BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1875                                                       failrec->last_mirror,
1876                                                       failrec->bio_flags);
1877         return 0;
1878 }
1879
1880 /*
1881  * each time an IO finishes, we do a fast check in the IO failure tree
1882  * to see if we need to process or clean up an io_failure_record
1883  */
1884 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1885 {
1886         u64 private;
1887         u64 private_failure;
1888         struct io_failure_record *failure;
1889         int ret;
1890
1891         private = 0;
1892         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1893                              (u64)-1, 1, EXTENT_DIRTY)) {
1894                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1895                                         start, &private_failure);
1896                 if (ret == 0) {
1897                         failure = (struct io_failure_record *)(unsigned long)
1898                                    private_failure;
1899                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1900                                           failure->start, 0);
1901                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1902                                           failure->start,
1903                                           failure->start + failure->len - 1,
1904                                           EXTENT_DIRTY | EXTENT_LOCKED,
1905                                           GFP_NOFS);
1906                         kfree(failure);
1907                 }
1908         }
1909         return 0;
1910 }
1911
1912 /*
1913  * when reads are done, we need to check csums to verify the data is correct
1914  * if there's a match, we allow the bio to finish.  If not, we go through
1915  * the io_failure_record routines to find good copies
1916  */
1917 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1918                                struct extent_state *state)
1919 {
1920         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1921         struct inode *inode = page->mapping->host;
1922         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1923         char *kaddr;
1924         u64 private = ~(u32)0;
1925         int ret;
1926         struct btrfs_root *root = BTRFS_I(inode)->root;
1927         u32 csum = ~(u32)0;
1928
1929         if (PageChecked(page)) {
1930                 ClearPageChecked(page);
1931                 goto good;
1932         }
1933
1934         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1935                 return 0;
1936
1937         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1938             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1939                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1940                                   GFP_NOFS);
1941                 return 0;
1942         }
1943
1944         if (state && state->start == start) {
1945                 private = state->private;
1946                 ret = 0;
1947         } else {
1948                 ret = get_state_private(io_tree, start, &private);
1949         }
1950         kaddr = kmap_atomic(page, KM_USER0);
1951         if (ret)
1952                 goto zeroit;
1953
1954         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
1955         btrfs_csum_final(csum, (char *)&csum);
1956         if (csum != private)
1957                 goto zeroit;
1958
1959         kunmap_atomic(kaddr, KM_USER0);
1960 good:
1961         /* if the io failure tree for this inode is non-empty,
1962          * check to see if we've recovered from a failed IO
1963          */
1964         btrfs_clean_io_failures(inode, start);
1965         return 0;
1966
1967 zeroit:
1968         if (printk_ratelimit()) {
1969                 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
1970                        "private %llu\n", page->mapping->host->i_ino,
1971                        (unsigned long long)start, csum,
1972                        (unsigned long long)private);
1973         }
1974         memset(kaddr + offset, 1, end - start + 1);
1975         flush_dcache_page(page);
1976         kunmap_atomic(kaddr, KM_USER0);
1977         if (private == 0)
1978                 return 0;
1979         return -EIO;
1980 }
1981
1982 /*
1983  * This creates an orphan entry for the given inode in case something goes
1984  * wrong in the middle of an unlink/truncate.
1985  */
1986 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
1987 {
1988         struct btrfs_root *root = BTRFS_I(inode)->root;
1989         int ret = 0;
1990
1991         spin_lock(&root->list_lock);
1992
1993         /* already on the orphan list, we're good */
1994         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
1995                 spin_unlock(&root->list_lock);
1996                 return 0;
1997         }
1998
1999         list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2000
2001         spin_unlock(&root->list_lock);
2002
2003         /*
2004          * insert an orphan item to track this unlinked/truncated file
2005          */
2006         ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
2007
2008         return ret;
2009 }
2010
2011 /*
2012  * We have done the truncate/delete so we can go ahead and remove the orphan
2013  * item for this particular inode.
2014  */
2015 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2016 {
2017         struct btrfs_root *root = BTRFS_I(inode)->root;
2018         int ret = 0;
2019
2020         spin_lock(&root->list_lock);
2021
2022         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2023                 spin_unlock(&root->list_lock);
2024                 return 0;
2025         }
2026
2027         list_del_init(&BTRFS_I(inode)->i_orphan);
2028         if (!trans) {
2029                 spin_unlock(&root->list_lock);
2030                 return 0;
2031         }
2032
2033         spin_unlock(&root->list_lock);
2034
2035         ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
2036
2037         return ret;
2038 }
2039
2040 /*
2041  * this cleans up any orphans that may be left on the list from the last use
2042  * of this root.
2043  */
2044 void btrfs_orphan_cleanup(struct btrfs_root *root)
2045 {
2046         struct btrfs_path *path;
2047         struct extent_buffer *leaf;
2048         struct btrfs_item *item;
2049         struct btrfs_key key, found_key;
2050         struct btrfs_trans_handle *trans;
2051         struct inode *inode;
2052         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2053
2054         path = btrfs_alloc_path();
2055         if (!path)
2056                 return;
2057         path->reada = -1;
2058
2059         key.objectid = BTRFS_ORPHAN_OBJECTID;
2060         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2061         key.offset = (u64)-1;
2062
2063
2064         while (1) {
2065                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2066                 if (ret < 0) {
2067                         printk(KERN_ERR "Error searching slot for orphan: %d"
2068                                "\n", ret);
2069                         break;
2070                 }
2071
2072                 /*
2073                  * if ret == 0 means we found what we were searching for, which
2074                  * is weird, but possible, so only screw with path if we didnt
2075                  * find the key and see if we have stuff that matches
2076                  */
2077                 if (ret > 0) {
2078                         if (path->slots[0] == 0)
2079                                 break;
2080                         path->slots[0]--;
2081                 }
2082
2083                 /* pull out the item */
2084                 leaf = path->nodes[0];
2085                 item = btrfs_item_nr(leaf, path->slots[0]);
2086                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2087
2088                 /* make sure the item matches what we want */
2089                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2090                         break;
2091                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2092                         break;
2093
2094                 /* release the path since we're done with it */
2095                 btrfs_release_path(root, path);
2096
2097                 /*
2098                  * this is where we are basically btrfs_lookup, without the
2099                  * crossing root thing.  we store the inode number in the
2100                  * offset of the orphan item.
2101                  */
2102                 found_key.objectid = found_key.offset;
2103                 found_key.type = BTRFS_INODE_ITEM_KEY;
2104                 found_key.offset = 0;
2105                 inode = btrfs_iget(root->fs_info->sb, &found_key, root);
2106                 if (IS_ERR(inode))
2107                         break;
2108
2109                 /*
2110                  * add this inode to the orphan list so btrfs_orphan_del does
2111                  * the proper thing when we hit it
2112                  */
2113                 spin_lock(&root->list_lock);
2114                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2115                 spin_unlock(&root->list_lock);
2116
2117                 /*
2118                  * if this is a bad inode, means we actually succeeded in
2119                  * removing the inode, but not the orphan record, which means
2120                  * we need to manually delete the orphan since iput will just
2121                  * do a destroy_inode
2122                  */
2123                 if (is_bad_inode(inode)) {
2124                         trans = btrfs_start_transaction(root, 1);
2125                         btrfs_orphan_del(trans, inode);
2126                         btrfs_end_transaction(trans, root);
2127                         iput(inode);
2128                         continue;
2129                 }
2130
2131                 /* if we have links, this was a truncate, lets do that */
2132                 if (inode->i_nlink) {
2133                         nr_truncate++;
2134                         btrfs_truncate(inode);
2135                 } else {
2136                         nr_unlink++;
2137                 }
2138
2139                 /* this will do delete_inode and everything for us */
2140                 iput(inode);
2141         }
2142
2143         if (nr_unlink)
2144                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2145         if (nr_truncate)
2146                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2147
2148         btrfs_free_path(path);
2149 }
2150
2151 /*
2152  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2153  * don't find any xattrs, we know there can't be any acls.
2154  *
2155  * slot is the slot the inode is in, objectid is the objectid of the inode
2156  */
2157 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2158                                           int slot, u64 objectid)
2159 {
2160         u32 nritems = btrfs_header_nritems(leaf);
2161         struct btrfs_key found_key;
2162         int scanned = 0;
2163
2164         slot++;
2165         while (slot < nritems) {
2166                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2167
2168                 /* we found a different objectid, there must not be acls */
2169                 if (found_key.objectid != objectid)
2170                         return 0;
2171
2172                 /* we found an xattr, assume we've got an acl */
2173                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2174                         return 1;
2175
2176                 /*
2177                  * we found a key greater than an xattr key, there can't
2178                  * be any acls later on
2179                  */
2180                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2181                         return 0;
2182
2183                 slot++;
2184                 scanned++;
2185
2186                 /*
2187                  * it goes inode, inode backrefs, xattrs, extents,
2188                  * so if there are a ton of hard links to an inode there can
2189                  * be a lot of backrefs.  Don't waste time searching too hard,
2190                  * this is just an optimization
2191                  */
2192                 if (scanned >= 8)
2193                         break;
2194         }
2195         /* we hit the end of the leaf before we found an xattr or
2196          * something larger than an xattr.  We have to assume the inode
2197          * has acls
2198          */
2199         return 1;
2200 }
2201
2202 /*
2203  * read an inode from the btree into the in-memory inode
2204  */
2205 static void btrfs_read_locked_inode(struct inode *inode)
2206 {
2207         struct btrfs_path *path;
2208         struct extent_buffer *leaf;
2209         struct btrfs_inode_item *inode_item;
2210         struct btrfs_timespec *tspec;
2211         struct btrfs_root *root = BTRFS_I(inode)->root;
2212         struct btrfs_key location;
2213         int maybe_acls;
2214         u64 alloc_group_block;
2215         u32 rdev;
2216         int ret;
2217
2218         path = btrfs_alloc_path();
2219         BUG_ON(!path);
2220         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2221
2222         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2223         if (ret)
2224                 goto make_bad;
2225
2226         leaf = path->nodes[0];
2227         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2228                                     struct btrfs_inode_item);
2229
2230         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2231         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2232         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2233         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2234         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2235
2236         tspec = btrfs_inode_atime(inode_item);
2237         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2238         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2239
2240         tspec = btrfs_inode_mtime(inode_item);
2241         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2242         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2243
2244         tspec = btrfs_inode_ctime(inode_item);
2245         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2246         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2247
2248         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2249         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2250         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2251         inode->i_generation = BTRFS_I(inode)->generation;
2252         inode->i_rdev = 0;
2253         rdev = btrfs_inode_rdev(leaf, inode_item);
2254
2255         BTRFS_I(inode)->index_cnt = (u64)-1;
2256         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2257
2258         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2259
2260         /*
2261          * try to precache a NULL acl entry for files that don't have
2262          * any xattrs or acls
2263          */
2264         maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2265         if (!maybe_acls)
2266                 cache_no_acl(inode);
2267
2268         BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2269                                                 alloc_group_block, 0);
2270         btrfs_free_path(path);
2271         inode_item = NULL;
2272
2273         switch (inode->i_mode & S_IFMT) {
2274         case S_IFREG:
2275                 inode->i_mapping->a_ops = &btrfs_aops;
2276                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2277                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2278                 inode->i_fop = &btrfs_file_operations;
2279                 inode->i_op = &btrfs_file_inode_operations;
2280                 break;
2281         case S_IFDIR:
2282                 inode->i_fop = &btrfs_dir_file_operations;
2283                 if (root == root->fs_info->tree_root)
2284                         inode->i_op = &btrfs_dir_ro_inode_operations;
2285                 else
2286                         inode->i_op = &btrfs_dir_inode_operations;
2287                 break;
2288         case S_IFLNK:
2289                 inode->i_op = &btrfs_symlink_inode_operations;
2290                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2291                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2292                 break;
2293         default:
2294                 inode->i_op = &btrfs_special_inode_operations;
2295                 init_special_inode(inode, inode->i_mode, rdev);
2296                 break;
2297         }
2298
2299         btrfs_update_iflags(inode);
2300         return;
2301
2302 make_bad:
2303         btrfs_free_path(path);
2304         make_bad_inode(inode);
2305 }
2306
2307 /*
2308  * given a leaf and an inode, copy the inode fields into the leaf
2309  */
2310 static void fill_inode_item(struct btrfs_trans_handle *trans,
2311                             struct extent_buffer *leaf,
2312                             struct btrfs_inode_item *item,
2313                             struct inode *inode)
2314 {
2315         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2316         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2317         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2318         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2319         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2320
2321         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2322                                inode->i_atime.tv_sec);
2323         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2324                                 inode->i_atime.tv_nsec);
2325
2326         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2327                                inode->i_mtime.tv_sec);
2328         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2329                                 inode->i_mtime.tv_nsec);
2330
2331         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2332                                inode->i_ctime.tv_sec);
2333         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2334                                 inode->i_ctime.tv_nsec);
2335
2336         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2337         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2338         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2339         btrfs_set_inode_transid(leaf, item, trans->transid);
2340         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2341         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2342         btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2343 }
2344
2345 /*
2346  * copy everything in the in-memory inode into the btree.
2347  */
2348 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2349                                 struct btrfs_root *root, struct inode *inode)
2350 {
2351         struct btrfs_inode_item *inode_item;
2352         struct btrfs_path *path;
2353         struct extent_buffer *leaf;
2354         int ret;
2355
2356         path = btrfs_alloc_path();
2357         BUG_ON(!path);
2358         path->leave_spinning = 1;
2359         ret = btrfs_lookup_inode(trans, root, path,
2360                                  &BTRFS_I(inode)->location, 1);
2361         if (ret) {
2362                 if (ret > 0)
2363                         ret = -ENOENT;
2364                 goto failed;
2365         }
2366
2367         btrfs_unlock_up_safe(path, 1);
2368         leaf = path->nodes[0];
2369         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2370                                   struct btrfs_inode_item);
2371
2372         fill_inode_item(trans, leaf, inode_item, inode);
2373         btrfs_mark_buffer_dirty(leaf);
2374         btrfs_set_inode_last_trans(trans, inode);
2375         ret = 0;
2376 failed:
2377         btrfs_free_path(path);
2378         return ret;
2379 }
2380
2381
2382 /*
2383  * unlink helper that gets used here in inode.c and in the tree logging
2384  * recovery code.  It remove a link in a directory with a given name, and
2385  * also drops the back refs in the inode to the directory
2386  */
2387 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2388                        struct btrfs_root *root,
2389                        struct inode *dir, struct inode *inode,
2390                        const char *name, int name_len)
2391 {
2392         struct btrfs_path *path;
2393         int ret = 0;
2394         struct extent_buffer *leaf;
2395         struct btrfs_dir_item *di;
2396         struct btrfs_key key;
2397         u64 index;
2398
2399         path = btrfs_alloc_path();
2400         if (!path) {
2401                 ret = -ENOMEM;
2402                 goto err;
2403         }
2404
2405         path->leave_spinning = 1;
2406         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2407                                     name, name_len, -1);
2408         if (IS_ERR(di)) {
2409                 ret = PTR_ERR(di);
2410                 goto err;
2411         }
2412         if (!di) {
2413                 ret = -ENOENT;
2414                 goto err;
2415         }
2416         leaf = path->nodes[0];
2417         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2418         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2419         if (ret)
2420                 goto err;
2421         btrfs_release_path(root, path);
2422
2423         ret = btrfs_del_inode_ref(trans, root, name, name_len,
2424                                   inode->i_ino,
2425                                   dir->i_ino, &index);
2426         if (ret) {
2427                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2428                        "inode %lu parent %lu\n", name_len, name,
2429                        inode->i_ino, dir->i_ino);
2430                 goto err;
2431         }
2432
2433         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2434                                          index, name, name_len, -1);
2435         if (IS_ERR(di)) {
2436                 ret = PTR_ERR(di);
2437                 goto err;
2438         }
2439         if (!di) {
2440                 ret = -ENOENT;
2441                 goto err;
2442         }
2443         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2444         btrfs_release_path(root, path);
2445
2446         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2447                                          inode, dir->i_ino);
2448         BUG_ON(ret != 0 && ret != -ENOENT);
2449
2450         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2451                                            dir, index);
2452         BUG_ON(ret);
2453 err:
2454         btrfs_free_path(path);
2455         if (ret)
2456                 goto out;
2457
2458         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2459         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2460         btrfs_update_inode(trans, root, dir);
2461         btrfs_drop_nlink(inode);
2462         ret = btrfs_update_inode(trans, root, inode);
2463 out:
2464         return ret;
2465 }
2466
2467 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2468 {
2469         struct btrfs_root *root;
2470         struct btrfs_trans_handle *trans;
2471         struct inode *inode = dentry->d_inode;
2472         int ret;
2473         unsigned long nr = 0;
2474
2475         root = BTRFS_I(dir)->root;
2476
2477         trans = btrfs_start_transaction(root, 1);
2478
2479         btrfs_set_trans_block_group(trans, dir);
2480
2481         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2482
2483         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2484                                  dentry->d_name.name, dentry->d_name.len);
2485
2486         if (inode->i_nlink == 0)
2487                 ret = btrfs_orphan_add(trans, inode);
2488
2489         nr = trans->blocks_used;
2490
2491         btrfs_end_transaction_throttle(trans, root);
2492         btrfs_btree_balance_dirty(root, nr);
2493         return ret;
2494 }
2495
2496 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2497                         struct btrfs_root *root,
2498                         struct inode *dir, u64 objectid,
2499                         const char *name, int name_len)
2500 {
2501         struct btrfs_path *path;
2502         struct extent_buffer *leaf;
2503         struct btrfs_dir_item *di;
2504         struct btrfs_key key;
2505         u64 index;
2506         int ret;
2507
2508         path = btrfs_alloc_path();
2509         if (!path)
2510                 return -ENOMEM;
2511
2512         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2513                                    name, name_len, -1);
2514         BUG_ON(!di || IS_ERR(di));
2515
2516         leaf = path->nodes[0];
2517         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2518         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2519         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2520         BUG_ON(ret);
2521         btrfs_release_path(root, path);
2522
2523         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2524                                  objectid, root->root_key.objectid,
2525                                  dir->i_ino, &index, name, name_len);
2526         if (ret < 0) {
2527                 BUG_ON(ret != -ENOENT);
2528                 di = btrfs_search_dir_index_item(root, path, dir->i_ino,
2529                                                  name, name_len);
2530                 BUG_ON(!di || IS_ERR(di));
2531
2532                 leaf = path->nodes[0];
2533                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2534                 btrfs_release_path(root, path);
2535                 index = key.offset;
2536         }
2537
2538         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2539                                          index, name, name_len, -1);
2540         BUG_ON(!di || IS_ERR(di));
2541
2542         leaf = path->nodes[0];
2543         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2544         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2545         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2546         BUG_ON(ret);
2547         btrfs_release_path(root, path);
2548
2549         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2550         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2551         ret = btrfs_update_inode(trans, root, dir);
2552         BUG_ON(ret);
2553         dir->i_sb->s_dirt = 1;
2554
2555         btrfs_free_path(path);
2556         return 0;
2557 }
2558
2559 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2560 {
2561         struct inode *inode = dentry->d_inode;
2562         int err = 0;
2563         int ret;
2564         struct btrfs_root *root = BTRFS_I(dir)->root;
2565         struct btrfs_trans_handle *trans;
2566         unsigned long nr = 0;
2567
2568         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2569             inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
2570                 return -ENOTEMPTY;
2571
2572         trans = btrfs_start_transaction(root, 1);
2573         btrfs_set_trans_block_group(trans, dir);
2574
2575         if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
2576                 err = btrfs_unlink_subvol(trans, root, dir,
2577                                           BTRFS_I(inode)->location.objectid,
2578                                           dentry->d_name.name,
2579                                           dentry->d_name.len);
2580                 goto out;
2581         }
2582
2583         err = btrfs_orphan_add(trans, inode);
2584         if (err)
2585                 goto out;
2586
2587         /* now the directory is empty */
2588         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2589                                  dentry->d_name.name, dentry->d_name.len);
2590         if (!err)
2591                 btrfs_i_size_write(inode, 0);
2592 out:
2593         nr = trans->blocks_used;
2594         ret = btrfs_end_transaction_throttle(trans, root);
2595         btrfs_btree_balance_dirty(root, nr);
2596
2597         if (ret && !err)
2598                 err = ret;
2599         return err;
2600 }
2601
2602 #if 0
2603 /*
2604  * when truncating bytes in a file, it is possible to avoid reading
2605  * the leaves that contain only checksum items.  This can be the
2606  * majority of the IO required to delete a large file, but it must
2607  * be done carefully.
2608  *
2609  * The keys in the level just above the leaves are checked to make sure
2610  * the lowest key in a given leaf is a csum key, and starts at an offset
2611  * after the new  size.
2612  *
2613  * Then the key for the next leaf is checked to make sure it also has
2614  * a checksum item for the same file.  If it does, we know our target leaf
2615  * contains only checksum items, and it can be safely freed without reading
2616  * it.
2617  *
2618  * This is just an optimization targeted at large files.  It may do
2619  * nothing.  It will return 0 unless things went badly.
2620  */
2621 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
2622                                      struct btrfs_root *root,
2623                                      struct btrfs_path *path,
2624                                      struct inode *inode, u64 new_size)
2625 {
2626         struct btrfs_key key;
2627         int ret;
2628         int nritems;
2629         struct btrfs_key found_key;
2630         struct btrfs_key other_key;
2631         struct btrfs_leaf_ref *ref;
2632         u64 leaf_gen;
2633         u64 leaf_start;
2634
2635         path->lowest_level = 1;
2636         key.objectid = inode->i_ino;
2637         key.type = BTRFS_CSUM_ITEM_KEY;
2638         key.offset = new_size;
2639 again:
2640         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2641         if (ret < 0)
2642                 goto out;
2643
2644         if (path->nodes[1] == NULL) {
2645                 ret = 0;
2646                 goto out;
2647         }
2648         ret = 0;
2649         btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
2650         nritems = btrfs_header_nritems(path->nodes[1]);
2651
2652         if (!nritems)
2653                 goto out;
2654
2655         if (path->slots[1] >= nritems)
2656                 goto next_node;
2657
2658         /* did we find a key greater than anything we want to delete? */
2659         if (found_key.objectid > inode->i_ino ||
2660            (found_key.objectid == inode->i_ino && found_key.type > key.type))
2661                 goto out;
2662
2663         /* we check the next key in the node to make sure the leave contains
2664          * only checksum items.  This comparison doesn't work if our
2665          * leaf is the last one in the node
2666          */
2667         if (path->slots[1] + 1 >= nritems) {
2668 next_node:
2669                 /* search forward from the last key in the node, this
2670                  * will bring us into the next node in the tree
2671                  */
2672                 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
2673
2674                 /* unlikely, but we inc below, so check to be safe */
2675                 if (found_key.offset == (u64)-1)
2676                         goto out;
2677
2678                 /* search_forward needs a path with locks held, do the
2679                  * search again for the original key.  It is possible
2680                  * this will race with a balance and return a path that
2681                  * we could modify, but this drop is just an optimization
2682                  * and is allowed to miss some leaves.
2683                  */
2684                 btrfs_release_path(root, path);
2685                 found_key.offset++;
2686
2687                 /* setup a max key for search_forward */
2688                 other_key.offset = (u64)-1;
2689                 other_key.type = key.type;
2690                 other_key.objectid = key.objectid;
2691
2692                 path->keep_locks = 1;
2693                 ret = btrfs_search_forward(root, &found_key, &other_key,
2694                                            path, 0, 0);
2695                 path->keep_locks = 0;
2696                 if (ret || found_key.objectid != key.objectid ||
2697                     found_key.type != key.type) {
2698                         ret = 0;
2699                         goto out;
2700                 }
2701
2702                 key.offset = found_key.offset;
2703                 btrfs_release_path(root, path);
2704                 cond_resched();
2705                 goto again;
2706         }
2707
2708         /* we know there's one more slot after us in the tree,
2709          * read that key so we can verify it is also a checksum item
2710          */
2711         btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
2712
2713         if (found_key.objectid < inode->i_ino)
2714                 goto next_key;
2715
2716         if (found_key.type != key.type || found_key.offset < new_size)
2717                 goto next_key;
2718
2719         /*
2720          * if the key for the next leaf isn't a csum key from this objectid,
2721          * we can't be sure there aren't good items inside this leaf.
2722          * Bail out
2723          */
2724         if (other_key.objectid != inode->i_ino || other_key.type != key.type)
2725                 goto out;
2726
2727         leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
2728         leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
2729         /*
2730          * it is safe to delete this leaf, it contains only
2731          * csum items from this inode at an offset >= new_size
2732          */
2733         ret = btrfs_del_leaf(trans, root, path, leaf_start);
2734         BUG_ON(ret);
2735
2736         if (root->ref_cows && leaf_gen < trans->transid) {
2737                 ref = btrfs_alloc_leaf_ref(root, 0);
2738                 if (ref) {
2739                         ref->root_gen = root->root_key.offset;
2740                         ref->bytenr = leaf_start;
2741                         ref->owner = 0;
2742                         ref->generation = leaf_gen;
2743                         ref->nritems = 0;
2744
2745                         btrfs_sort_leaf_ref(ref);
2746
2747                         ret = btrfs_add_leaf_ref(root, ref, 0);
2748                         WARN_ON(ret);
2749                         btrfs_free_leaf_ref(root, ref);
2750                 } else {
2751                         WARN_ON(1);
2752                 }
2753         }
2754 next_key:
2755         btrfs_release_path(root, path);
2756
2757         if (other_key.objectid == inode->i_ino &&
2758             other_key.type == key.type && other_key.offset > key.offset) {
2759                 key.offset = other_key.offset;
2760                 cond_resched();
2761                 goto again;
2762         }
2763         ret = 0;
2764 out:
2765         /* fixup any changes we've made to the path */
2766         path->lowest_level = 0;
2767         path->keep_locks = 0;
2768         btrfs_release_path(root, path);
2769         return ret;
2770 }
2771
2772 #endif
2773
2774 /*
2775  * this can truncate away extent items, csum items and directory items.
2776  * It starts at a high offset and removes keys until it can't find
2777  * any higher than new_size
2778  *
2779  * csum items that cross the new i_size are truncated to the new size
2780  * as well.
2781  *
2782  * min_type is the minimum key type to truncate down to.  If set to 0, this
2783  * will kill all the items on this inode, including the INODE_ITEM_KEY.
2784  */
2785 noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2786                                         struct btrfs_root *root,
2787                                         struct inode *inode,
2788                                         u64 new_size, u32 min_type)
2789 {
2790         int ret;
2791         struct btrfs_path *path;
2792         struct btrfs_key key;
2793         struct btrfs_key found_key;
2794         u32 found_type = (u8)-1;
2795         struct extent_buffer *leaf;
2796         struct btrfs_file_extent_item *fi;
2797         u64 extent_start = 0;
2798         u64 extent_num_bytes = 0;
2799         u64 extent_offset = 0;
2800         u64 item_end = 0;
2801         int found_extent;
2802         int del_item;
2803         int pending_del_nr = 0;
2804         int pending_del_slot = 0;
2805         int extent_type = -1;
2806         int encoding;
2807         u64 mask = root->sectorsize - 1;
2808
2809         if (root->ref_cows)
2810                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
2811         path = btrfs_alloc_path();
2812         BUG_ON(!path);
2813         path->reada = -1;
2814
2815         /* FIXME, add redo link to tree so we don't leak on crash */
2816         key.objectid = inode->i_ino;
2817         key.offset = (u64)-1;
2818         key.type = (u8)-1;
2819
2820 search_again:
2821         path->leave_spinning = 1;
2822         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2823         if (ret < 0)
2824                 goto error;
2825
2826         if (ret > 0) {
2827                 /* there are no items in the tree for us to truncate, we're
2828                  * done
2829                  */
2830                 if (path->slots[0] == 0) {
2831                         ret = 0;
2832                         goto error;
2833                 }
2834                 path->slots[0]--;
2835         }
2836
2837         while (1) {
2838                 fi = NULL;
2839                 leaf = path->nodes[0];
2840                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2841                 found_type = btrfs_key_type(&found_key);
2842                 encoding = 0;
2843
2844                 if (found_key.objectid != inode->i_ino)
2845                         break;
2846
2847                 if (found_type < min_type)
2848                         break;
2849
2850                 item_end = found_key.offset;
2851                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
2852                         fi = btrfs_item_ptr(leaf, path->slots[0],
2853                                             struct btrfs_file_extent_item);
2854                         extent_type = btrfs_file_extent_type(leaf, fi);
2855                         encoding = btrfs_file_extent_compression(leaf, fi);
2856                         encoding |= btrfs_file_extent_encryption(leaf, fi);
2857                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
2858
2859                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2860                                 item_end +=
2861                                     btrfs_file_extent_num_bytes(leaf, fi);
2862                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2863                                 item_end += btrfs_file_extent_inline_len(leaf,
2864                                                                          fi);
2865                         }
2866                         item_end--;
2867                 }
2868                 if (item_end < new_size) {
2869                         if (found_type == BTRFS_DIR_ITEM_KEY)
2870                                 found_type = BTRFS_INODE_ITEM_KEY;
2871                         else if (found_type == BTRFS_EXTENT_ITEM_KEY)
2872                                 found_type = BTRFS_EXTENT_DATA_KEY;
2873                         else if (found_type == BTRFS_EXTENT_DATA_KEY)
2874                                 found_type = BTRFS_XATTR_ITEM_KEY;
2875                         else if (found_type == BTRFS_XATTR_ITEM_KEY)
2876                                 found_type = BTRFS_INODE_REF_KEY;
2877                         else if (found_type)
2878                                 found_type--;
2879                         else
2880                                 break;
2881                         btrfs_set_key_type(&key, found_type);
2882                         goto next;
2883                 }
2884                 if (found_key.offset >= new_size)
2885                         del_item = 1;
2886                 else
2887                         del_item = 0;
2888                 found_extent = 0;
2889
2890                 /* FIXME, shrink the extent if the ref count is only 1 */
2891                 if (found_type != BTRFS_EXTENT_DATA_KEY)
2892                         goto delete;
2893
2894                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2895                         u64 num_dec;
2896                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
2897                         if (!del_item && !encoding) {
2898                                 u64 orig_num_bytes =
2899                                         btrfs_file_extent_num_bytes(leaf, fi);
2900                                 extent_num_bytes = new_size -
2901                                         found_key.offset + root->sectorsize - 1;
2902                                 extent_num_bytes = extent_num_bytes &
2903                                         ~((u64)root->sectorsize - 1);
2904                                 btrfs_set_file_extent_num_bytes(leaf, fi,
2905                                                          extent_num_bytes);
2906                                 num_dec = (orig_num_bytes -
2907                                            extent_num_bytes);
2908                                 if (root->ref_cows && extent_start != 0)
2909                                         inode_sub_bytes(inode, num_dec);
2910                                 btrfs_mark_buffer_dirty(leaf);
2911                         } else {
2912                                 extent_num_bytes =
2913                                         btrfs_file_extent_disk_num_bytes(leaf,
2914                                                                          fi);
2915                                 extent_offset = found_key.offset -
2916                                         btrfs_file_extent_offset(leaf, fi);
2917
2918                                 /* FIXME blocksize != 4096 */
2919                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
2920                                 if (extent_start != 0) {
2921                                         found_extent = 1;
2922                                         if (root->ref_cows)
2923                                                 inode_sub_bytes(inode, num_dec);
2924                                 }
2925                         }
2926                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2927                         /*
2928                          * we can't truncate inline items that have had
2929                          * special encodings
2930                          */
2931                         if (!del_item &&
2932                             btrfs_file_extent_compression(leaf, fi) == 0 &&
2933                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
2934                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
2935                                 u32 size = new_size - found_key.offset;
2936
2937                                 if (root->ref_cows) {
2938                                         inode_sub_bytes(inode, item_end + 1 -
2939                                                         new_size);
2940                                 }
2941                                 size =
2942                                     btrfs_file_extent_calc_inline_size(size);
2943                                 ret = btrfs_truncate_item(trans, root, path,
2944                                                           size, 1);
2945                                 BUG_ON(ret);
2946                         } else if (root->ref_cows) {
2947                                 inode_sub_bytes(inode, item_end + 1 -
2948                                                 found_key.offset);
2949                         }
2950                 }
2951 delete:
2952                 if (del_item) {
2953                         if (!pending_del_nr) {
2954                                 /* no pending yet, add ourselves */
2955                                 pending_del_slot = path->slots[0];
2956                                 pending_del_nr = 1;
2957                         } else if (pending_del_nr &&
2958                                    path->slots[0] + 1 == pending_del_slot) {
2959                                 /* hop on the pending chunk */
2960                                 pending_del_nr++;
2961                                 pending_del_slot = path->slots[0];
2962                         } else {
2963                                 BUG();
2964                         }
2965                 } else {
2966                         break;
2967                 }
2968                 if (found_extent && root->ref_cows) {
2969                         btrfs_set_path_blocking(path);
2970                         ret = btrfs_free_extent(trans, root, extent_start,
2971                                                 extent_num_bytes, 0,
2972                                                 btrfs_header_owner(leaf),
2973                                                 inode->i_ino, extent_offset);
2974                         BUG_ON(ret);
2975                 }
2976 next:
2977                 if (path->slots[0] == 0) {
2978                         if (pending_del_nr)
2979                                 goto del_pending;
2980                         btrfs_release_path(root, path);
2981                         if (found_type == BTRFS_INODE_ITEM_KEY)
2982                                 break;
2983                         goto search_again;
2984                 }
2985
2986                 path->slots[0]--;
2987                 if (pending_del_nr &&
2988                     path->slots[0] + 1 != pending_del_slot) {
2989                         struct btrfs_key debug;
2990 del_pending:
2991                         btrfs_item_key_to_cpu(path->nodes[0], &debug,
2992                                               pending_del_slot);
2993                         ret = btrfs_del_items(trans, root, path,
2994                                               pending_del_slot,
2995                                               pending_del_nr);
2996                         BUG_ON(ret);
2997                         pending_del_nr = 0;
2998                         btrfs_release_path(root, path);
2999                         if (found_type == BTRFS_INODE_ITEM_KEY)
3000                                 break;
3001                         goto search_again;
3002                 }
3003         }
3004         ret = 0;
3005 error:
3006         if (pending_del_nr) {
3007                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3008                                       pending_del_nr);
3009         }
3010         btrfs_free_path(path);
3011         return ret;
3012 }
3013
3014 /*
3015  * taken from block_truncate_page, but does cow as it zeros out
3016  * any bytes left in the last page in the file.
3017  */
3018 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3019 {
3020         struct inode *inode = mapping->host;
3021         struct btrfs_root *root = BTRFS_I(inode)->root;
3022         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3023         struct btrfs_ordered_extent *ordered;
3024         char *kaddr;
3025         u32 blocksize = root->sectorsize;
3026         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3027         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3028         struct page *page;
3029         int ret = 0;
3030         u64 page_start;
3031         u64 page_end;
3032
3033         if ((offset & (blocksize - 1)) == 0)
3034                 goto out;
3035
3036         ret = -ENOMEM;
3037 again:
3038         page = grab_cache_page(mapping, index);
3039         if (!page)
3040                 goto out;
3041
3042         page_start = page_offset(page);
3043         page_end = page_start + PAGE_CACHE_SIZE - 1;
3044
3045         if (!PageUptodate(page)) {
3046                 ret = btrfs_readpage(NULL, page);
3047                 lock_page(page);
3048                 if (page->mapping != mapping) {
3049                         unlock_page(page);
3050                         page_cache_release(page);
3051                         goto again;
3052                 }
3053                 if (!PageUptodate(page)) {
3054                         ret = -EIO;
3055                         goto out_unlock;
3056                 }
3057         }
3058         wait_on_page_writeback(page);
3059
3060         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
3061         set_page_extent_mapped(page);
3062
3063         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3064         if (ordered) {
3065                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3066                 unlock_page(page);
3067                 page_cache_release(page);
3068                 btrfs_start_ordered_extent(inode, ordered, 1);
3069                 btrfs_put_ordered_extent(ordered);
3070                 goto again;
3071         }
3072
3073         ret = btrfs_set_extent_delalloc(inode, page_start, page_end);
3074         if (ret) {
3075                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3076                 goto out_unlock;
3077         }
3078
3079         ret = 0;
3080         if (offset != PAGE_CACHE_SIZE) {
3081                 kaddr = kmap(page);
3082                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3083                 flush_dcache_page(page);
3084                 kunmap(page);
3085         }
3086         ClearPageChecked(page);
3087         set_page_dirty(page);
3088         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3089
3090 out_unlock:
3091         unlock_page(page);
3092         page_cache_release(page);
3093 out:
3094         return ret;
3095 }
3096
3097 int btrfs_cont_expand(struct inode *inode, loff_t size)
3098 {
3099         struct btrfs_trans_handle *trans;
3100         struct btrfs_root *root = BTRFS_I(inode)->root;
3101         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3102         struct extent_map *em;
3103         u64 mask = root->sectorsize - 1;
3104         u64 hole_start = (inode->i_size + mask) & ~mask;
3105         u64 block_end = (size + mask) & ~mask;
3106         u64 last_byte;
3107         u64 cur_offset;
3108         u64 hole_size;
3109         int err = 0;
3110
3111         if (size <= hole_start)
3112                 return 0;
3113
3114         btrfs_truncate_page(inode->i_mapping, inode->i_size);
3115
3116         while (1) {
3117                 struct btrfs_ordered_extent *ordered;
3118                 btrfs_wait_ordered_range(inode, hole_start,
3119                                          block_end - hole_start);
3120                 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
3121                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3122                 if (!ordered)
3123                         break;
3124                 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
3125                 btrfs_put_ordered_extent(ordered);
3126         }
3127
3128         trans = btrfs_start_transaction(root, 1);
3129         btrfs_set_trans_block_group(trans, inode);
3130
3131         cur_offset = hole_start;
3132         while (1) {
3133                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3134                                 block_end - cur_offset, 0);
3135                 BUG_ON(IS_ERR(em) || !em);
3136                 last_byte = min(extent_map_end(em), block_end);
3137                 last_byte = (last_byte + mask) & ~mask;
3138                 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
3139                         u64 hint_byte = 0;
3140                         hole_size = last_byte - cur_offset;
3141                         err = btrfs_drop_extents(trans, root, inode,
3142                                                  cur_offset,
3143                                                  cur_offset + hole_size,
3144                                                  block_end,
3145                                                  cur_offset, &hint_byte, 1);
3146                         if (err)
3147                                 break;
3148
3149                         err = btrfs_reserve_metadata_space(root, 1);
3150                         if (err)
3151                                 break;
3152
3153                         err = btrfs_insert_file_extent(trans, root,
3154                                         inode->i_ino, cur_offset, 0,
3155                                         0, hole_size, 0, hole_size,
3156                                         0, 0, 0);
3157                         btrfs_drop_extent_cache(inode, hole_start,
3158                                         last_byte - 1, 0);
3159                         btrfs_unreserve_metadata_space(root, 1);
3160                 }
3161                 free_extent_map(em);
3162                 cur_offset = last_byte;
3163                 if (err || cur_offset >= block_end)
3164                         break;
3165         }
3166
3167         btrfs_end_transaction(trans, root);
3168         unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
3169         return err;
3170 }
3171
3172 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3173 {
3174         struct inode *inode = dentry->d_inode;
3175         int err;
3176
3177         err = inode_change_ok(inode, attr);
3178         if (err)
3179                 return err;
3180
3181         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3182                 if (attr->ia_size > inode->i_size) {
3183                         err = btrfs_cont_expand(inode, attr->ia_size);
3184                         if (err)
3185                                 return err;
3186                 } else if (inode->i_size > 0 &&
3187                            attr->ia_size == 0) {
3188
3189                         /* we're truncating a file that used to have good
3190                          * data down to zero.  Make sure it gets into
3191                          * the ordered flush list so that any new writes
3192                          * get down to disk quickly.
3193                          */
3194                         BTRFS_I(inode)->ordered_data_close = 1;
3195                 }
3196         }
3197
3198         err = inode_setattr(inode, attr);
3199
3200         if (!err && ((attr->ia_valid & ATTR_MODE)))
3201                 err = btrfs_acl_chmod(inode);
3202         return err;
3203 }
3204
3205 void btrfs_delete_inode(struct inode *inode)
3206 {
3207         struct btrfs_trans_handle *trans;
3208         struct btrfs_root *root = BTRFS_I(inode)->root;
3209         unsigned long nr;
3210         int ret;
3211
3212         truncate_inode_pages(&inode->i_data, 0);
3213         if (is_bad_inode(inode)) {
3214                 btrfs_orphan_del(NULL, inode);
3215                 goto no_delete;
3216         }
3217         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3218
3219         if (inode->i_nlink > 0) {
3220                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3221                 goto no_delete;
3222         }
3223
3224         btrfs_i_size_write(inode, 0);
3225         trans = btrfs_join_transaction(root, 1);
3226
3227         btrfs_set_trans_block_group(trans, inode);
3228         ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0);
3229         if (ret) {
3230                 btrfs_orphan_del(NULL, inode);
3231                 goto no_delete_lock;
3232         }
3233
3234         btrfs_orphan_del(trans, inode);
3235
3236         nr = trans->blocks_used;
3237         clear_inode(inode);
3238
3239         btrfs_end_transaction(trans, root);
3240         btrfs_btree_balance_dirty(root, nr);
3241         return;
3242
3243 no_delete_lock:
3244         nr = trans->blocks_used;
3245         btrfs_end_transaction(trans, root);
3246         btrfs_btree_balance_dirty(root, nr);
3247 no_delete:
3248         clear_inode(inode);
3249 }
3250
3251 /*
3252  * this returns the key found in the dir entry in the location pointer.
3253  * If no dir entries were found, location->objectid is 0.
3254  */
3255 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3256                                struct btrfs_key *location)
3257 {
3258         const char *name = dentry->d_name.name;
3259         int namelen = dentry->d_name.len;
3260         struct btrfs_dir_item *di;
3261         struct btrfs_path *path;
3262         struct btrfs_root *root = BTRFS_I(dir)->root;
3263         int ret = 0;
3264
3265         path = btrfs_alloc_path();
3266         BUG_ON(!path);
3267
3268         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3269                                     namelen, 0);
3270         if (IS_ERR(di))
3271                 ret = PTR_ERR(di);
3272
3273         if (!di || IS_ERR(di))
3274                 goto out_err;
3275
3276         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3277 out:
3278         btrfs_free_path(path);
3279         return ret;
3280 out_err:
3281         location->objectid = 0;
3282         goto out;
3283 }
3284
3285 /*
3286  * when we hit a tree root in a directory, the btrfs part of the inode
3287  * needs to be changed to reflect the root directory of the tree root.  This
3288  * is kind of like crossing a mount point.
3289  */
3290 static int fixup_tree_root_location(struct btrfs_root *root,
3291                                     struct inode *dir,
3292                                     struct dentry *dentry,
3293                                     struct btrfs_key *location,
3294                                     struct btrfs_root **sub_root)
3295 {
3296         struct btrfs_path *path;
3297         struct btrfs_root *new_root;
3298         struct btrfs_root_ref *ref;
3299         struct extent_buffer *leaf;
3300         int ret;
3301         int err = 0;
3302
3303         path = btrfs_alloc_path();
3304         if (!path) {
3305                 err = -ENOMEM;
3306                 goto out;
3307         }
3308
3309         err = -ENOENT;
3310         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3311                                   BTRFS_I(dir)->root->root_key.objectid,
3312                                   location->objectid);
3313         if (ret) {
3314                 if (ret < 0)
3315                         err = ret;
3316                 goto out;
3317         }
3318
3319         leaf = path->nodes[0];
3320         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3321         if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino ||
3322             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3323                 goto out;
3324
3325         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3326                                    (unsigned long)(ref + 1),
3327                                    dentry->d_name.len);
3328         if (ret)
3329                 goto out;
3330
3331         btrfs_release_path(root->fs_info->tree_root, path);
3332
3333         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3334         if (IS_ERR(new_root)) {
3335                 err = PTR_ERR(new_root);
3336                 goto out;
3337         }
3338
3339         if (btrfs_root_refs(&new_root->root_item) == 0) {
3340                 err = -ENOENT;
3341                 goto out;
3342         }
3343
3344         *sub_root = new_root;
3345         location->objectid = btrfs_root_dirid(&new_root->root_item);
3346         location->type = BTRFS_INODE_ITEM_KEY;
3347         location->offset = 0;
3348         err = 0;
3349 out:
3350         btrfs_free_path(path);
3351         return err;
3352 }
3353
3354 static void inode_tree_add(struct inode *inode)
3355 {
3356         struct btrfs_root *root = BTRFS_I(inode)->root;
3357         struct btrfs_inode *entry;
3358         struct rb_node **p;
3359         struct rb_node *parent;
3360 again:
3361         p = &root->inode_tree.rb_node;
3362         parent = NULL;
3363
3364         if (hlist_unhashed(&inode->i_hash))
3365                 return;
3366
3367         spin_lock(&root->inode_lock);
3368         while (*p) {
3369                 parent = *p;
3370                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3371
3372                 if (inode->i_ino < entry->vfs_inode.i_ino)
3373                         p = &parent->rb_left;
3374                 else if (inode->i_ino > entry->vfs_inode.i_ino)
3375                         p = &parent->rb_right;
3376                 else {
3377                         WARN_ON(!(entry->vfs_inode.i_state &
3378                                   (I_WILL_FREE | I_FREEING | I_CLEAR)));
3379                         rb_erase(parent, &root->inode_tree);
3380                         RB_CLEAR_NODE(parent);
3381                         spin_unlock(&root->inode_lock);
3382                         goto again;
3383                 }
3384         }
3385         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3386         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3387         spin_unlock(&root->inode_lock);
3388 }
3389
3390 static void inode_tree_del(struct inode *inode)
3391 {
3392         struct btrfs_root *root = BTRFS_I(inode)->root;
3393         int empty = 0;
3394
3395         spin_lock(&root->inode_lock);
3396         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3397                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3398                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3399                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3400         }
3401         spin_unlock(&root->inode_lock);
3402
3403         if (empty && btrfs_root_refs(&root->root_item) == 0) {
3404                 synchronize_srcu(&root->fs_info->subvol_srcu);
3405                 spin_lock(&root->inode_lock);
3406                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3407                 spin_unlock(&root->inode_lock);
3408                 if (empty)
3409                         btrfs_add_dead_root(root);
3410         }
3411 }
3412
3413 int btrfs_invalidate_inodes(struct btrfs_root *root)
3414 {
3415         struct rb_node *node;
3416         struct rb_node *prev;
3417         struct btrfs_inode *entry;
3418         struct inode *inode;
3419         u64 objectid = 0;
3420
3421         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3422
3423         spin_lock(&root->inode_lock);
3424 again:
3425         node = root->inode_tree.rb_node;
3426         prev = NULL;
3427         while (node) {
3428                 prev = node;
3429                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3430
3431                 if (objectid < entry->vfs_inode.i_ino)
3432                         node = node->rb_left;
3433                 else if (objectid > entry->vfs_inode.i_ino)
3434                         node = node->rb_right;
3435                 else
3436                         break;
3437         }
3438         if (!node) {
3439                 while (prev) {
3440                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
3441                         if (objectid <= entry->vfs_inode.i_ino) {
3442                                 node = prev;
3443                                 break;
3444                         }
3445                         prev = rb_next(prev);
3446                 }
3447         }
3448         while (node) {
3449                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3450                 objectid = entry->vfs_inode.i_ino + 1;
3451                 inode = igrab(&entry->vfs_inode);
3452                 if (inode) {
3453                         spin_unlock(&root->inode_lock);
3454                         if (atomic_read(&inode->i_count) > 1)
3455                                 d_prune_aliases(inode);
3456                         /*
3457                          * btrfs_drop_inode will remove it from
3458                          * the inode cache when its usage count
3459                          * hits zero.
3460                          */
3461                         iput(inode);
3462                         cond_resched();
3463                         spin_lock(&root->inode_lock);
3464                         goto again;
3465                 }
3466
3467                 if (cond_resched_lock(&root->inode_lock))
3468                         goto again;
3469
3470                 node = rb_next(node);
3471         }
3472         spin_unlock(&root->inode_lock);
3473         return 0;
3474 }
3475
3476 static noinline void init_btrfs_i(struct inode *inode)
3477 {
3478         struct btrfs_inode *bi = BTRFS_I(inode);
3479
3480         bi->generation = 0;
3481         bi->sequence = 0;
3482         bi->last_trans = 0;
3483         bi->last_sub_trans = 0;
3484         bi->logged_trans = 0;
3485         bi->delalloc_bytes = 0;
3486         bi->reserved_bytes = 0;
3487         bi->disk_i_size = 0;
3488         bi->flags = 0;
3489         bi->index_cnt = (u64)-1;
3490         bi->last_unlink_trans = 0;
3491         bi->ordered_data_close = 0;
3492         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
3493         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
3494                              inode->i_mapping, GFP_NOFS);
3495         extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
3496                              inode->i_mapping, GFP_NOFS);
3497         INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
3498         INIT_LIST_HEAD(&BTRFS_I(inode)->ordered_operations);
3499         RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3500         btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
3501         mutex_init(&BTRFS_I(inode)->extent_mutex);
3502         mutex_init(&BTRFS_I(inode)->log_mutex);
3503 }
3504
3505 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3506 {
3507         struct btrfs_iget_args *args = p;
3508         inode->i_ino = args->ino;
3509         init_btrfs_i(inode);
3510         BTRFS_I(inode)->root = args->root;
3511         btrfs_set_inode_space_info(args->root, inode);
3512         return 0;
3513 }
3514
3515 static int btrfs_find_actor(struct inode *inode, void *opaque)
3516 {
3517         struct btrfs_iget_args *args = opaque;
3518         return args->ino == inode->i_ino &&
3519                 args->root == BTRFS_I(inode)->root;
3520 }
3521
3522 static struct inode *btrfs_iget_locked(struct super_block *s,
3523                                        u64 objectid,
3524                                        struct btrfs_root *root)
3525 {
3526         struct inode *inode;
3527         struct btrfs_iget_args args;
3528         args.ino = objectid;
3529         args.root = root;
3530
3531         inode = iget5_locked(s, objectid, btrfs_find_actor,
3532                              btrfs_init_locked_inode,
3533                              (void *)&args);
3534         return inode;
3535 }
3536
3537 /* Get an inode object given its location and corresponding root.
3538  * Returns in *is_new if the inode was read from disk
3539  */
3540 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3541                          struct btrfs_root *root)
3542 {
3543         struct inode *inode;
3544
3545         inode = btrfs_iget_locked(s, location->objectid, root);
3546         if (!inode)
3547                 return ERR_PTR(-ENOMEM);
3548
3549         if (inode->i_state & I_NEW) {
3550                 BTRFS_I(inode)->root = root;
3551                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3552                 btrfs_read_locked_inode(inode);
3553
3554                 inode_tree_add(inode);
3555                 unlock_new_inode(inode);
3556         }
3557
3558         return inode;
3559 }
3560
3561 static struct inode *new_simple_dir(struct super_block *s,
3562                                     struct btrfs_key *key,
3563                                     struct btrfs_root *root)
3564 {
3565         struct inode *inode = new_inode(s);
3566
3567         if (!inode)
3568                 return ERR_PTR(-ENOMEM);
3569
3570         init_btrfs_i(inode);
3571
3572         BTRFS_I(inode)->root = root;
3573         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
3574         BTRFS_I(inode)->dummy_inode = 1;
3575
3576         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
3577         inode->i_op = &simple_dir_inode_operations;
3578         inode->i_fop = &simple_dir_operations;
3579         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
3580         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3581
3582         return inode;
3583 }
3584
3585 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3586 {
3587         struct inode *inode;
3588         struct btrfs_root *root = BTRFS_I(dir)->root;
3589         struct btrfs_root *sub_root = root;
3590         struct btrfs_key location;
3591         int index;
3592         int ret;
3593
3594         dentry->d_op = &btrfs_dentry_operations;
3595
3596         if (dentry->d_name.len > BTRFS_NAME_LEN)
3597                 return ERR_PTR(-ENAMETOOLONG);
3598
3599         ret = btrfs_inode_by_name(dir, dentry, &location);
3600
3601         if (ret < 0)
3602                 return ERR_PTR(ret);
3603
3604         if (location.objectid == 0)
3605                 return NULL;
3606
3607         if (location.type == BTRFS_INODE_ITEM_KEY) {
3608                 inode = btrfs_iget(dir->i_sb, &location, root);
3609                 return inode;
3610         }
3611
3612         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
3613
3614         index = srcu_read_lock(&root->fs_info->subvol_srcu);
3615         ret = fixup_tree_root_location(root, dir, dentry,
3616                                        &location, &sub_root);
3617         if (ret < 0) {
3618                 if (ret != -ENOENT)
3619                         inode = ERR_PTR(ret);
3620                 else
3621                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
3622         } else {
3623                 inode = btrfs_iget(dir->i_sb, &location, sub_root);
3624         }
3625         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
3626
3627         return inode;
3628 }
3629
3630 static int btrfs_dentry_delete(struct dentry *dentry)
3631 {
3632         struct btrfs_root *root;
3633
3634         if (!dentry->d_inode && !IS_ROOT(dentry))
3635                 dentry = dentry->d_parent;
3636
3637         if (dentry->d_inode) {
3638                 root = BTRFS_I(dentry->d_inode)->root;
3639                 if (btrfs_root_refs(&root->root_item) == 0)
3640                         return 1;
3641         }
3642         return 0;
3643 }
3644
3645 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3646                                    struct nameidata *nd)
3647 {
3648         struct inode *inode;
3649
3650         inode = btrfs_lookup_dentry(dir, dentry);
3651         if (IS_ERR(inode))
3652                 return ERR_CAST(inode);
3653
3654         return d_splice_alias(inode, dentry);
3655 }
3656
3657 static unsigned char btrfs_filetype_table[] = {
3658         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
3659 };
3660
3661 static int btrfs_real_readdir(struct file *filp, void *dirent,
3662                               filldir_t filldir)
3663 {
3664         struct inode *inode = filp->f_dentry->d_inode;
3665         struct btrfs_root *root = BTRFS_I(inode)->root;
3666         struct btrfs_item *item;
3667         struct btrfs_dir_item *di;
3668         struct btrfs_key key;
3669         struct btrfs_key found_key;
3670         struct btrfs_path *path;
3671         int ret;
3672         u32 nritems;
3673         struct extent_buffer *leaf;
3674         int slot;
3675         int advance;
3676         unsigned char d_type;
3677         int over = 0;
3678         u32 di_cur;
3679         u32 di_total;
3680         u32 di_len;
3681         int key_type = BTRFS_DIR_INDEX_KEY;
3682         char tmp_name[32];
3683         char *name_ptr;
3684         int name_len;
3685
3686         /* FIXME, use a real flag for deciding about the key type */
3687         if (root->fs_info->tree_root == root)
3688                 key_type = BTRFS_DIR_ITEM_KEY;
3689
3690         /* special case for "." */
3691         if (filp->f_pos == 0) {
3692                 over = filldir(dirent, ".", 1,
3693                                1, inode->i_ino,
3694                                DT_DIR);
3695                 if (over)
3696                         return 0;
3697                 filp->f_pos = 1;
3698         }
3699         /* special case for .., just use the back ref */
3700         if (filp->f_pos == 1) {
3701                 u64 pino = parent_ino(filp->f_path.dentry);
3702                 over = filldir(dirent, "..", 2,
3703                                2, pino, DT_DIR);
3704                 if (over)
3705                         return 0;
3706                 filp->f_pos = 2;
3707         }
3708         path = btrfs_alloc_path();
3709         path->reada = 2;
3710
3711         btrfs_set_key_type(&key, key_type);
3712         key.offset = filp->f_pos;
3713         key.objectid = inode->i_ino;
3714
3715         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3716         if (ret < 0)
3717                 goto err;
3718         advance = 0;
3719
3720         while (1) {
3721                 leaf = path->nodes[0];
3722                 nritems = btrfs_header_nritems(leaf);
3723                 slot = path->slots[0];
3724                 if (advance || slot >= nritems) {
3725                         if (slot >= nritems - 1) {
3726                                 ret = btrfs_next_leaf(root, path);
3727                                 if (ret)
3728                                         break;
3729                                 leaf = path->nodes[0];
3730                                 nritems = btrfs_header_nritems(leaf);
3731                                 slot = path->slots[0];
3732                         } else {
3733                                 slot++;
3734                                 path->slots[0]++;
3735                         }
3736                 }
3737
3738                 advance = 1;
3739                 item = btrfs_item_nr(leaf, slot);
3740                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3741
3742                 if (found_key.objectid != key.objectid)
3743                         break;
3744                 if (btrfs_key_type(&found_key) != key_type)
3745                         break;
3746                 if (found_key.offset < filp->f_pos)
3747                         continue;
3748
3749                 filp->f_pos = found_key.offset;
3750
3751                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
3752                 di_cur = 0;
3753                 di_total = btrfs_item_size(leaf, item);
3754
3755                 while (di_cur < di_total) {
3756                         struct btrfs_key location;
3757
3758                         name_len = btrfs_dir_name_len(leaf, di);
3759                         if (name_len <= sizeof(tmp_name)) {
3760                                 name_ptr = tmp_name;
3761                         } else {
3762                                 name_ptr = kmalloc(name_len, GFP_NOFS);
3763                                 if (!name_ptr) {
3764                                         ret = -ENOMEM;
3765                                         goto err;
3766                                 }
3767                         }
3768                         read_extent_buffer(leaf, name_ptr,
3769                                            (unsigned long)(di + 1), name_len);
3770
3771                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
3772                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
3773
3774                         /* is this a reference to our own snapshot? If so
3775                          * skip it
3776                          */
3777                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
3778                             location.objectid == root->root_key.objectid) {
3779                                 over = 0;
3780                                 goto skip;
3781                         }
3782                         over = filldir(dirent, name_ptr, name_len,
3783                                        found_key.offset, location.objectid,
3784                                        d_type);
3785
3786 skip:
3787                         if (name_ptr != tmp_name)
3788                                 kfree(name_ptr);
3789
3790                         if (over)
3791                                 goto nopos;
3792                         di_len = btrfs_dir_name_len(leaf, di) +
3793                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
3794                         di_cur += di_len;
3795                         di = (struct btrfs_dir_item *)((char *)di + di_len);
3796                 }
3797         }
3798
3799         /* Reached end of directory/root. Bump pos past the last item. */
3800         if (key_type == BTRFS_DIR_INDEX_KEY)
3801                 filp->f_pos = INT_LIMIT(off_t);
3802         else
3803                 filp->f_pos++;
3804 nopos:
3805         ret = 0;
3806 err:
3807         btrfs_free_path(path);
3808         return ret;
3809 }
3810
3811 int btrfs_write_inode(struct inode *inode, int wait)
3812 {
3813         struct btrfs_root *root = BTRFS_I(inode)->root;
3814         struct btrfs_trans_handle *trans;
3815         int ret = 0;
3816
3817         if (root->fs_info->btree_inode == inode)
3818                 return 0;
3819
3820         if (wait) {
3821                 trans = btrfs_join_transaction(root, 1);
3822                 btrfs_set_trans_block_group(trans, inode);
3823                 ret = btrfs_commit_transaction(trans, root);
3824         }
3825         return ret;
3826 }
3827
3828 /*
3829  * This is somewhat expensive, updating the tree every time the
3830  * inode changes.  But, it is most likely to find the inode in cache.
3831  * FIXME, needs more benchmarking...there are no reasons other than performance
3832  * to keep or drop this code.
3833  */
3834 void btrfs_dirty_inode(struct inode *inode)
3835 {
3836         struct btrfs_root *root = BTRFS_I(inode)->root;
3837         struct btrfs_trans_handle *trans;
3838
3839         trans = btrfs_join_transaction(root, 1);
3840         btrfs_set_trans_block_group(trans, inode);
3841         btrfs_update_inode(trans, root, inode);
3842         btrfs_end_transaction(trans, root);
3843 }
3844
3845 /*
3846  * find the highest existing sequence number in a directory
3847  * and then set the in-memory index_cnt variable to reflect
3848  * free sequence numbers
3849  */
3850 static int btrfs_set_inode_index_count(struct inode *inode)
3851 {
3852         struct btrfs_root *root = BTRFS_I(inode)->root;
3853         struct btrfs_key key, found_key;
3854         struct btrfs_path *path;
3855         struct extent_buffer *leaf;
3856         int ret;
3857
3858         key.objectid = inode->i_ino;
3859         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
3860         key.offset = (u64)-1;
3861
3862         path = btrfs_alloc_path();
3863         if (!path)
3864                 return -ENOMEM;
3865
3866         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3867         if (ret < 0)
3868                 goto out;
3869         /* FIXME: we should be able to handle this */
3870         if (ret == 0)
3871                 goto out;
3872         ret = 0;
3873
3874         /*
3875          * MAGIC NUMBER EXPLANATION:
3876          * since we search a directory based on f_pos we have to start at 2
3877          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
3878          * else has to start at 2
3879          */
3880         if (path->slots[0] == 0) {
3881                 BTRFS_I(inode)->index_cnt = 2;
3882                 goto out;
3883         }
3884
3885         path->slots[0]--;
3886
3887         leaf = path->nodes[0];
3888         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3889
3890         if (found_key.objectid != inode->i_ino ||
3891             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
3892                 BTRFS_I(inode)->index_cnt = 2;
3893                 goto out;
3894         }
3895
3896         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
3897 out:
3898         btrfs_free_path(path);
3899         return ret;
3900 }
3901
3902 /*
3903  * helper to find a free sequence number in a given directory.  This current
3904  * code is very simple, later versions will do smarter things in the btree
3905  */
3906 int btrfs_set_inode_index(struct inode *dir, u64 *index)
3907 {
3908         int ret = 0;
3909
3910         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
3911                 ret = btrfs_set_inode_index_count(dir);
3912                 if (ret)
3913                         return ret;
3914         }
3915
3916         *index = BTRFS_I(dir)->index_cnt;
3917         BTRFS_I(dir)->index_cnt++;
3918
3919         return ret;
3920 }
3921
3922 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
3923                                      struct btrfs_root *root,
3924                                      struct inode *dir,
3925                                      const char *name, int name_len,
3926                                      u64 ref_objectid, u64 objectid,
3927                                      u64 alloc_hint, int mode, u64 *index)
3928 {
3929         struct inode *inode;
3930         struct btrfs_inode_item *inode_item;
3931         struct btrfs_key *location;
3932         struct btrfs_path *path;
3933         struct btrfs_inode_ref *ref;
3934         struct btrfs_key key[2];
3935         u32 sizes[2];
3936         unsigned long ptr;
3937         int ret;
3938         int owner;
3939
3940         path = btrfs_alloc_path();
3941         BUG_ON(!path);
3942
3943         inode = new_inode(root->fs_info->sb);
3944         if (!inode)
3945                 return ERR_PTR(-ENOMEM);
3946
3947         if (dir) {
3948                 ret = btrfs_set_inode_index(dir, index);
3949                 if (ret) {
3950                         iput(inode);
3951                         return ERR_PTR(ret);
3952                 }
3953         }
3954         /*
3955          * index_cnt is ignored for everything but a dir,
3956          * btrfs_get_inode_index_count has an explanation for the magic
3957          * number
3958          */
3959         init_btrfs_i(inode);
3960         BTRFS_I(inode)->index_cnt = 2;
3961         BTRFS_I(inode)->root = root;
3962         BTRFS_I(inode)->generation = trans->transid;
3963         btrfs_set_inode_space_info(root, inode);
3964
3965         if (mode & S_IFDIR)
3966                 owner = 0;
3967         else
3968                 owner = 1;
3969         BTRFS_I(inode)->block_group =
3970                         btrfs_find_block_group(root, 0, alloc_hint, owner);
3971
3972         key[0].objectid = objectid;
3973         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
3974         key[0].offset = 0;
3975
3976         key[1].objectid = objectid;
3977         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
3978         key[1].offset = ref_objectid;
3979
3980         sizes[0] = sizeof(struct btrfs_inode_item);
3981         sizes[1] = name_len + sizeof(*ref);
3982
3983         path->leave_spinning = 1;
3984         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
3985         if (ret != 0)
3986                 goto fail;
3987
3988         inode->i_uid = current_fsuid();
3989
3990         if (dir && (dir->i_mode & S_ISGID)) {
3991                 inode->i_gid = dir->i_gid;
3992                 if (S_ISDIR(mode))
3993                         mode |= S_ISGID;
3994         } else
3995                 inode->i_gid = current_fsgid();
3996
3997         inode->i_mode = mode;
3998         inode->i_ino = objectid;
3999         inode_set_bytes(inode, 0);
4000         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4001         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4002                                   struct btrfs_inode_item);
4003         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4004
4005         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4006                              struct btrfs_inode_ref);
4007         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4008         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4009         ptr = (unsigned long)(ref + 1);
4010         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4011
4012         btrfs_mark_buffer_dirty(path->nodes[0]);
4013         btrfs_free_path(path);
4014
4015         location = &BTRFS_I(inode)->location;
4016         location->objectid = objectid;
4017         location->offset = 0;
4018         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4019
4020         btrfs_inherit_iflags(inode, dir);
4021
4022         if ((mode & S_IFREG)) {
4023                 if (btrfs_test_opt(root, NODATASUM))
4024                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4025                 if (btrfs_test_opt(root, NODATACOW))
4026                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4027         }
4028
4029         insert_inode_hash(inode);
4030         inode_tree_add(inode);
4031         return inode;
4032 fail:
4033         if (dir)
4034                 BTRFS_I(dir)->index_cnt--;
4035         btrfs_free_path(path);
4036         iput(inode);
4037         return ERR_PTR(ret);
4038 }
4039
4040 static inline u8 btrfs_inode_type(struct inode *inode)
4041 {
4042         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4043 }
4044
4045 /*
4046  * utility function to add 'inode' into 'parent_inode' with
4047  * a give name and a given sequence number.
4048  * if 'add_backref' is true, also insert a backref from the
4049  * inode to the parent directory.
4050  */
4051 int btrfs_add_link(struct btrfs_trans_handle *trans,
4052                    struct inode *parent_inode, struct inode *inode,
4053                    const char *name, int name_len, int add_backref, u64 index)
4054 {
4055         int ret = 0;
4056         struct btrfs_key key;
4057         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4058
4059         if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4060                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4061         } else {
4062                 key.objectid = inode->i_ino;
4063                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4064                 key.offset = 0;
4065         }
4066
4067         if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4068                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4069                                          key.objectid, root->root_key.objectid,
4070                                          parent_inode->i_ino,
4071                                          index, name, name_len);
4072         } else if (add_backref) {
4073                 ret = btrfs_insert_inode_ref(trans, root,
4074                                              name, name_len, inode->i_ino,
4075                                              parent_inode->i_ino, index);
4076         }
4077
4078         if (ret == 0) {
4079                 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4080                                             parent_inode->i_ino, &key,
4081                                             btrfs_inode_type(inode), index);
4082                 BUG_ON(ret);
4083
4084                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4085                                    name_len * 2);
4086                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4087                 ret = btrfs_update_inode(trans, root, parent_inode);
4088         }
4089         return ret;
4090 }
4091
4092 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4093                             struct dentry *dentry, struct inode *inode,
4094                             int backref, u64 index)
4095 {
4096         int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
4097                                  inode, dentry->d_name.name,
4098                                  dentry->d_name.len, backref, index);
4099         if (!err) {
4100                 d_instantiate(dentry, inode);
4101                 return 0;
4102         }
4103         if (err > 0)
4104                 err = -EEXIST;
4105         return err;
4106 }
4107
4108 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4109                         int mode, dev_t rdev)
4110 {
4111         struct btrfs_trans_handle *trans;
4112         struct btrfs_root *root = BTRFS_I(dir)->root;
4113         struct inode *inode = NULL;
4114         int err;
4115         int drop_inode = 0;
4116         u64 objectid;
4117         unsigned long nr = 0;
4118         u64 index = 0;
4119
4120         if (!new_valid_dev(rdev))
4121                 return -EINVAL;
4122
4123         /*
4124          * 2 for inode item and ref
4125          * 2 for dir items
4126          * 1 for xattr if selinux is on
4127          */
4128         err = btrfs_reserve_metadata_space(root, 5);
4129         if (err)
4130                 return err;
4131
4132         trans = btrfs_start_transaction(root, 1);
4133         if (!trans)
4134                 goto fail;
4135         btrfs_set_trans_block_group(trans, dir);
4136
4137         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4138         if (err) {
4139                 err = -ENOSPC;
4140                 goto out_unlock;
4141         }
4142
4143         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4144                                 dentry->d_name.len,
4145                                 dentry->d_parent->d_inode->i_ino, objectid,
4146                                 BTRFS_I(dir)->block_group, mode, &index);
4147         err = PTR_ERR(inode);
4148         if (IS_ERR(inode))
4149                 goto out_unlock;
4150
4151         err = btrfs_init_inode_security(inode, dir);
4152         if (err) {
4153                 drop_inode = 1;
4154                 goto out_unlock;
4155         }
4156
4157         btrfs_set_trans_block_group(trans, inode);
4158         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
4159         if (err)
4160                 drop_inode = 1;
4161         else {
4162                 inode->i_op = &btrfs_special_inode_operations;
4163                 init_special_inode(inode, inode->i_mode, rdev);
4164                 btrfs_update_inode(trans, root, inode);
4165         }
4166         btrfs_update_inode_block_group(trans, inode);
4167         btrfs_update_inode_block_group(trans, dir);
4168 out_unlock:
4169         nr = trans->blocks_used;
4170         btrfs_end_transaction_throttle(trans, root);
4171 fail:
4172         btrfs_unreserve_metadata_space(root, 5);
4173         if (drop_inode) {
4174                 inode_dec_link_count(inode);
4175                 iput(inode);
4176         }
4177         btrfs_btree_balance_dirty(root, nr);
4178         return err;
4179 }
4180
4181 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4182                         int mode, struct nameidata *nd)
4183 {
4184         struct btrfs_trans_handle *trans;
4185         struct btrfs_root *root = BTRFS_I(dir)->root;
4186         struct inode *inode = NULL;
4187         int err;
4188         int drop_inode = 0;
4189         unsigned long nr = 0;
4190         u64 objectid;
4191         u64 index = 0;
4192
4193         /*
4194          * 2 for inode item and ref
4195          * 2 for dir items
4196          * 1 for xattr if selinux is on
4197          */
4198         err = btrfs_reserve_metadata_space(root, 5);
4199         if (err)
4200                 return err;
4201
4202         trans = btrfs_start_transaction(root, 1);
4203         if (!trans)
4204                 goto fail;
4205         btrfs_set_trans_block_group(trans, dir);
4206
4207         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4208         if (err) {
4209                 err = -ENOSPC;
4210                 goto out_unlock;
4211         }
4212
4213         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4214                                 dentry->d_name.len,
4215                                 dentry->d_parent->d_inode->i_ino,
4216                                 objectid, BTRFS_I(dir)->block_group, mode,
4217                                 &index);
4218         err = PTR_ERR(inode);
4219         if (IS_ERR(inode))
4220                 goto out_unlock;
4221
4222         err = btrfs_init_inode_security(inode, dir);
4223         if (err) {
4224                 drop_inode = 1;
4225                 goto out_unlock;
4226         }
4227
4228         btrfs_set_trans_block_group(trans, inode);
4229         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
4230         if (err)
4231                 drop_inode = 1;
4232         else {
4233                 inode->i_mapping->a_ops = &btrfs_aops;
4234                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4235                 inode->i_fop = &btrfs_file_operations;
4236                 inode->i_op = &btrfs_file_inode_operations;
4237                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4238         }
4239         btrfs_update_inode_block_group(trans, inode);
4240         btrfs_update_inode_block_group(trans, dir);
4241 out_unlock:
4242         nr = trans->blocks_used;
4243         btrfs_end_transaction_throttle(trans, root);
4244 fail:
4245         btrfs_unreserve_metadata_space(root, 5);
4246         if (drop_inode) {
4247                 inode_dec_link_count(inode);
4248                 iput(inode);
4249         }
4250         btrfs_btree_balance_dirty(root, nr);
4251         return err;
4252 }
4253
4254 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4255                       struct dentry *dentry)
4256 {
4257         struct btrfs_trans_handle *trans;
4258         struct btrfs_root *root = BTRFS_I(dir)->root;
4259         struct inode *inode = old_dentry->d_inode;
4260         u64 index;
4261         unsigned long nr = 0;
4262         int err;
4263         int drop_inode = 0;
4264
4265         if (inode->i_nlink == 0)
4266                 return -ENOENT;
4267
4268         /*
4269          * 1 item for inode ref
4270          * 2 items for dir items
4271          */
4272         err = btrfs_reserve_metadata_space(root, 3);
4273         if (err)
4274                 return err;
4275
4276         btrfs_inc_nlink(inode);
4277
4278         err = btrfs_set_inode_index(dir, &index);
4279         if (err)
4280                 goto fail;
4281
4282         trans = btrfs_start_transaction(root, 1);
4283
4284         btrfs_set_trans_block_group(trans, dir);
4285         atomic_inc(&inode->i_count);
4286
4287         err = btrfs_add_nondir(trans, dentry, inode, 1, index);
4288
4289         if (err) {
4290                 drop_inode = 1;
4291         } else {
4292                 btrfs_update_inode_block_group(trans, dir);
4293                 err = btrfs_update_inode(trans, root, inode);
4294                 BUG_ON(err);
4295                 btrfs_log_new_name(trans, inode, NULL, dentry->d_parent);
4296         }
4297
4298         nr = trans->blocks_used;
4299         btrfs_end_transaction_throttle(trans, root);
4300 fail:
4301         btrfs_unreserve_metadata_space(root, 3);
4302         if (drop_inode) {
4303                 inode_dec_link_count(inode);
4304                 iput(inode);
4305         }
4306         btrfs_btree_balance_dirty(root, nr);
4307         return err;
4308 }
4309
4310 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4311 {
4312         struct inode *inode = NULL;
4313         struct btrfs_trans_handle *trans;
4314         struct btrfs_root *root = BTRFS_I(dir)->root;
4315         int err = 0;
4316         int drop_on_err = 0;
4317         u64 objectid = 0;
4318         u64 index = 0;
4319         unsigned long nr = 1;
4320
4321         /*
4322          * 2 items for inode and ref
4323          * 2 items for dir items
4324          * 1 for xattr if selinux is on
4325          */
4326         err = btrfs_reserve_metadata_space(root, 5);
4327         if (err)
4328                 return err;
4329
4330         trans = btrfs_start_transaction(root, 1);
4331         if (!trans) {
4332                 err = -ENOMEM;
4333                 goto out_unlock;
4334         }
4335         btrfs_set_trans_block_group(trans, dir);
4336
4337         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4338         if (err) {
4339                 err = -ENOSPC;
4340                 goto out_unlock;
4341         }
4342
4343         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4344                                 dentry->d_name.len,
4345                                 dentry->d_parent->d_inode->i_ino, objectid,
4346                                 BTRFS_I(dir)->block_group, S_IFDIR | mode,
4347                                 &index);
4348         if (IS_ERR(inode)) {
4349                 err = PTR_ERR(inode);
4350                 goto out_fail;
4351         }
4352
4353         drop_on_err = 1;
4354
4355         err = btrfs_init_inode_security(inode, dir);
4356         if (err)
4357                 goto out_fail;
4358
4359         inode->i_op = &btrfs_dir_inode_operations;
4360         inode->i_fop = &btrfs_dir_file_operations;
4361         btrfs_set_trans_block_group(trans, inode);
4362
4363         btrfs_i_size_write(inode, 0);
4364         err = btrfs_update_inode(trans, root, inode);
4365         if (err)
4366                 goto out_fail;
4367
4368         err = btrfs_add_link(trans, dentry->d_parent->d_inode,
4369                                  inode, dentry->d_name.name,
4370                                  dentry->d_name.len, 0, index);
4371         if (err)
4372                 goto out_fail;
4373
4374         d_instantiate(dentry, inode);
4375         drop_on_err = 0;
4376         btrfs_update_inode_block_group(trans, inode);
4377         btrfs_update_inode_block_group(trans, dir);
4378
4379 out_fail:
4380         nr = trans->blocks_used;
4381         btrfs_end_transaction_throttle(trans, root);
4382
4383 out_unlock:
4384         btrfs_unreserve_metadata_space(root, 5);
4385         if (drop_on_err)
4386                 iput(inode);
4387         btrfs_btree_balance_dirty(root, nr);
4388         return err;
4389 }
4390
4391 /* helper for btfs_get_extent.  Given an existing extent in the tree,
4392  * and an extent that you want to insert, deal with overlap and insert
4393  * the new extent into the tree.
4394  */
4395 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4396                                 struct extent_map *existing,
4397                                 struct extent_map *em,
4398                                 u64 map_start, u64 map_len)
4399 {
4400         u64 start_diff;
4401
4402         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4403         start_diff = map_start - em->start;
4404         em->start = map_start;
4405         em->len = map_len;
4406         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4407             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4408                 em->block_start += start_diff;
4409                 em->block_len -= start_diff;
4410         }
4411         return add_extent_mapping(em_tree, em);
4412 }
4413
4414 static noinline int uncompress_inline(struct btrfs_path *path,
4415                                       struct inode *inode, struct page *page,
4416                                       size_t pg_offset, u64 extent_offset,
4417                                       struct btrfs_file_extent_item *item)
4418 {
4419         int ret;
4420         struct extent_buffer *leaf = path->nodes[0];
4421         char *tmp;
4422         size_t max_size;
4423         unsigned long inline_size;
4424         unsigned long ptr;
4425
4426         WARN_ON(pg_offset != 0);
4427         max_size = btrfs_file_extent_ram_bytes(leaf, item);
4428         inline_size = btrfs_file_extent_inline_item_len(leaf,
4429                                         btrfs_item_nr(leaf, path->slots[0]));
4430         tmp = kmalloc(inline_size, GFP_NOFS);
4431         ptr = btrfs_file_extent_inline_start(item);
4432
4433         read_extent_buffer(leaf, tmp, ptr, inline_size);
4434
4435         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4436         ret = btrfs_zlib_decompress(tmp, page, extent_offset,
4437                                     inline_size, max_size);
4438         if (ret) {
4439                 char *kaddr = kmap_atomic(page, KM_USER0);
4440                 unsigned long copy_size = min_t(u64,
4441                                   PAGE_CACHE_SIZE - pg_offset,
4442                                   max_size - extent_offset);
4443                 memset(kaddr + pg_offset, 0, copy_size);
4444                 kunmap_atomic(kaddr, KM_USER0);
4445         }
4446         kfree(tmp);
4447         return 0;
4448 }
4449
4450 /*
4451  * a bit scary, this does extent mapping from logical file offset to the disk.
4452  * the ugly parts come from merging extents from the disk with the in-ram
4453  * representation.  This gets more complex because of the data=ordered code,
4454  * where the in-ram extents might be locked pending data=ordered completion.
4455  *
4456  * This also copies inline extents directly into the page.
4457  */
4458
4459 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4460                                     size_t pg_offset, u64 start, u64 len,
4461                                     int create)
4462 {
4463         int ret;
4464         int err = 0;
4465         u64 bytenr;
4466         u64 extent_start = 0;
4467         u64 extent_end = 0;
4468         u64 objectid = inode->i_ino;
4469         u32 found_type;
4470         struct btrfs_path *path = NULL;
4471         struct btrfs_root *root = BTRFS_I(inode)->root;
4472         struct btrfs_file_extent_item *item;
4473         struct extent_buffer *leaf;
4474         struct btrfs_key found_key;
4475         struct extent_map *em = NULL;
4476         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4477         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4478         struct btrfs_trans_handle *trans = NULL;
4479         int compressed;
4480
4481 again:
4482         read_lock(&em_tree->lock);
4483         em = lookup_extent_mapping(em_tree, start, len);
4484         if (em)
4485                 em->bdev = root->fs_info->fs_devices->latest_bdev;
4486         read_unlock(&em_tree->lock);
4487
4488         if (em) {
4489                 if (em->start > start || em->start + em->len <= start)
4490                         free_extent_map(em);
4491                 else if (em->block_start == EXTENT_MAP_INLINE && page)
4492                         free_extent_map(em);
4493                 else
4494                         goto out;
4495         }
4496         em = alloc_extent_map(GFP_NOFS);
4497         if (!em) {
4498                 err = -ENOMEM;
4499                 goto out;
4500         }
4501         em->bdev = root->fs_info->fs_devices->latest_bdev;
4502         em->start = EXTENT_MAP_HOLE;
4503         em->orig_start = EXTENT_MAP_HOLE;
4504         em->len = (u64)-1;
4505         em->block_len = (u64)-1;
4506
4507         if (!path) {
4508                 path = btrfs_alloc_path();
4509                 BUG_ON(!path);
4510         }
4511
4512         ret = btrfs_lookup_file_extent(trans, root, path,
4513                                        objectid, start, trans != NULL);
4514         if (ret < 0) {
4515                 err = ret;
4516                 goto out;
4517         }
4518
4519         if (ret != 0) {
4520                 if (path->slots[0] == 0)
4521                         goto not_found;
4522                 path->slots[0]--;
4523         }
4524
4525         leaf = path->nodes[0];
4526         item = btrfs_item_ptr(leaf, path->slots[0],
4527                               struct btrfs_file_extent_item);
4528         /* are we inside the extent that was found? */
4529         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4530         found_type = btrfs_key_type(&found_key);
4531         if (found_key.objectid != objectid ||
4532             found_type != BTRFS_EXTENT_DATA_KEY) {
4533                 goto not_found;
4534         }
4535
4536         found_type = btrfs_file_extent_type(leaf, item);
4537         extent_start = found_key.offset;
4538         compressed = btrfs_file_extent_compression(leaf, item);
4539         if (found_type == BTRFS_FILE_EXTENT_REG ||
4540             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4541                 extent_end = extent_start +
4542                        btrfs_file_extent_num_bytes(leaf, item);
4543         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4544                 size_t size;
4545                 size = btrfs_file_extent_inline_len(leaf, item);
4546                 extent_end = (extent_start + size + root->sectorsize - 1) &
4547                         ~((u64)root->sectorsize - 1);
4548         }
4549
4550         if (start >= extent_end) {
4551                 path->slots[0]++;
4552                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
4553                         ret = btrfs_next_leaf(root, path);
4554                         if (ret < 0) {
4555                                 err = ret;
4556                                 goto out;
4557                         }
4558                         if (ret > 0)
4559                                 goto not_found;
4560                         leaf = path->nodes[0];
4561                 }
4562                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4563                 if (found_key.objectid != objectid ||
4564                     found_key.type != BTRFS_EXTENT_DATA_KEY)
4565                         goto not_found;
4566                 if (start + len <= found_key.offset)
4567                         goto not_found;
4568                 em->start = start;
4569                 em->len = found_key.offset - start;
4570                 goto not_found_em;
4571         }
4572
4573         if (found_type == BTRFS_FILE_EXTENT_REG ||
4574             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4575                 em->start = extent_start;
4576                 em->len = extent_end - extent_start;
4577                 em->orig_start = extent_start -
4578                                  btrfs_file_extent_offset(leaf, item);
4579                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
4580                 if (bytenr == 0) {
4581                         em->block_start = EXTENT_MAP_HOLE;
4582                         goto insert;
4583                 }
4584                 if (compressed) {
4585                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4586                         em->block_start = bytenr;
4587                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
4588                                                                          item);
4589                 } else {
4590                         bytenr += btrfs_file_extent_offset(leaf, item);
4591                         em->block_start = bytenr;
4592                         em->block_len = em->len;
4593                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
4594                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
4595                 }
4596                 goto insert;
4597         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4598                 unsigned long ptr;
4599                 char *map;
4600                 size_t size;
4601                 size_t extent_offset;
4602                 size_t copy_size;
4603
4604                 em->block_start = EXTENT_MAP_INLINE;
4605                 if (!page || create) {
4606                         em->start = extent_start;
4607                         em->len = extent_end - extent_start;
4608                         goto out;
4609                 }
4610
4611                 size = btrfs_file_extent_inline_len(leaf, item);
4612                 extent_offset = page_offset(page) + pg_offset - extent_start;
4613                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
4614                                 size - extent_offset);
4615                 em->start = extent_start + extent_offset;
4616                 em->len = (copy_size + root->sectorsize - 1) &
4617                         ~((u64)root->sectorsize - 1);
4618                 em->orig_start = EXTENT_MAP_INLINE;
4619                 if (compressed)
4620                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4621                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
4622                 if (create == 0 && !PageUptodate(page)) {
4623                         if (btrfs_file_extent_compression(leaf, item) ==
4624                             BTRFS_COMPRESS_ZLIB) {
4625                                 ret = uncompress_inline(path, inode, page,
4626                                                         pg_offset,
4627                                                         extent_offset, item);
4628                                 BUG_ON(ret);
4629                         } else {
4630                                 map = kmap(page);
4631                                 read_extent_buffer(leaf, map + pg_offset, ptr,
4632                                                    copy_size);
4633                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
4634                                         memset(map + pg_offset + copy_size, 0,
4635                                                PAGE_CACHE_SIZE - pg_offset -
4636                                                copy_size);
4637                                 }
4638                                 kunmap(page);
4639                         }
4640                         flush_dcache_page(page);
4641                 } else if (create && PageUptodate(page)) {
4642                         if (!trans) {
4643                                 kunmap(page);
4644                                 free_extent_map(em);
4645                                 em = NULL;
4646                                 btrfs_release_path(root, path);
4647                                 trans = btrfs_join_transaction(root, 1);
4648                                 goto again;
4649                         }
4650                         map = kmap(page);
4651                         write_extent_buffer(leaf, map + pg_offset, ptr,
4652                                             copy_size);
4653                         kunmap(page);
4654                         btrfs_mark_buffer_dirty(leaf);
4655                 }
4656                 set_extent_uptodate(io_tree, em->start,
4657                                     extent_map_end(em) - 1, GFP_NOFS);
4658                 goto insert;
4659         } else {
4660                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
4661                 WARN_ON(1);
4662         }
4663 not_found:
4664         em->start = start;
4665         em->len = len;
4666 not_found_em:
4667         em->block_start = EXTENT_MAP_HOLE;
4668         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
4669 insert:
4670         btrfs_release_path(root, path);
4671         if (em->start > start || extent_map_end(em) <= start) {
4672                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
4673                        "[%llu %llu]\n", (unsigned long long)em->start,
4674                        (unsigned long long)em->len,
4675                        (unsigned long long)start,
4676                        (unsigned long long)len);
4677                 err = -EIO;
4678                 goto out;
4679         }
4680
4681         err = 0;
4682         write_lock(&em_tree->lock);
4683         ret = add_extent_mapping(em_tree, em);
4684         /* it is possible that someone inserted the extent into the tree
4685          * while we had the lock dropped.  It is also possible that
4686          * an overlapping map exists in the tree
4687          */
4688         if (ret == -EEXIST) {
4689                 struct extent_map *existing;
4690
4691                 ret = 0;
4692
4693                 existing = lookup_extent_mapping(em_tree, start, len);
4694                 if (existing && (existing->start > start ||
4695                     existing->start + existing->len <= start)) {
4696                         free_extent_map(existing);
4697                         existing = NULL;
4698                 }
4699                 if (!existing) {
4700                         existing = lookup_extent_mapping(em_tree, em->start,
4701                                                          em->len);
4702                         if (existing) {
4703                                 err = merge_extent_mapping(em_tree, existing,
4704                                                            em, start,
4705                                                            root->sectorsize);
4706                                 free_extent_map(existing);
4707                                 if (err) {
4708                                         free_extent_map(em);
4709                                         em = NULL;
4710                                 }
4711                         } else {
4712                                 err = -EIO;
4713                                 free_extent_map(em);
4714                                 em = NULL;
4715                         }
4716                 } else {
4717                         free_extent_map(em);
4718                         em = existing;
4719                         err = 0;
4720                 }
4721         }
4722         write_unlock(&em_tree->lock);
4723 out:
4724         if (path)
4725                 btrfs_free_path(path);
4726         if (trans) {
4727                 ret = btrfs_end_transaction(trans, root);
4728                 if (!err)
4729                         err = ret;
4730         }
4731         if (err) {
4732                 free_extent_map(em);
4733                 return ERR_PTR(err);
4734         }
4735         return em;
4736 }
4737
4738 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
4739                         const struct iovec *iov, loff_t offset,
4740                         unsigned long nr_segs)
4741 {
4742         return -EINVAL;
4743 }
4744
4745 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4746                 __u64 start, __u64 len)
4747 {
4748         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent);
4749 }
4750
4751 int btrfs_readpage(struct file *file, struct page *page)
4752 {
4753         struct extent_io_tree *tree;
4754         tree = &BTRFS_I(page->mapping->host)->io_tree;
4755         return extent_read_full_page(tree, page, btrfs_get_extent);
4756 }
4757
4758 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
4759 {
4760         struct extent_io_tree *tree;
4761
4762
4763         if (current->flags & PF_MEMALLOC) {
4764                 redirty_page_for_writepage(wbc, page);
4765                 unlock_page(page);
4766                 return 0;
4767         }
4768         tree = &BTRFS_I(page->mapping->host)->io_tree;
4769         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
4770 }
4771
4772 int btrfs_writepages(struct address_space *mapping,
4773                      struct writeback_control *wbc)
4774 {
4775         struct extent_io_tree *tree;
4776
4777         tree = &BTRFS_I(mapping->host)->io_tree;
4778         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
4779 }
4780
4781 static int
4782 btrfs_readpages(struct file *file, struct address_space *mapping,
4783                 struct list_head *pages, unsigned nr_pages)
4784 {
4785         struct extent_io_tree *tree;
4786         tree = &BTRFS_I(mapping->host)->io_tree;
4787         return extent_readpages(tree, mapping, pages, nr_pages,
4788                                 btrfs_get_extent);
4789 }
4790 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4791 {
4792         struct extent_io_tree *tree;
4793         struct extent_map_tree *map;
4794         int ret;
4795
4796         tree = &BTRFS_I(page->mapping->host)->io_tree;
4797         map = &BTRFS_I(page->mapping->host)->extent_tree;
4798         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
4799         if (ret == 1) {
4800                 ClearPagePrivate(page);
4801                 set_page_private(page, 0);
4802                 page_cache_release(page);
4803         }
4804         return ret;
4805 }
4806
4807 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4808 {
4809         if (PageWriteback(page) || PageDirty(page))
4810                 return 0;
4811         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
4812 }
4813
4814 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
4815 {
4816         struct extent_io_tree *tree;
4817         struct btrfs_ordered_extent *ordered;
4818         u64 page_start = page_offset(page);
4819         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
4820
4821
4822         /*
4823          * we have the page locked, so new writeback can't start,
4824          * and the dirty bit won't be cleared while we are here.
4825          *
4826          * Wait for IO on this page so that we can safely clear
4827          * the PagePrivate2 bit and do ordered accounting
4828          */
4829         wait_on_page_writeback(page);
4830
4831         tree = &BTRFS_I(page->mapping->host)->io_tree;
4832         if (offset) {
4833                 btrfs_releasepage(page, GFP_NOFS);
4834                 return;
4835         }
4836         lock_extent(tree, page_start, page_end, GFP_NOFS);
4837         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
4838                                            page_offset(page));
4839         if (ordered) {
4840                 /*
4841                  * IO on this page will never be started, so we need
4842                  * to account for any ordered extents now
4843                  */
4844                 clear_extent_bit(tree, page_start, page_end,
4845                                  EXTENT_DIRTY | EXTENT_DELALLOC |
4846                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
4847                                  NULL, GFP_NOFS);
4848                 /*
4849                  * whoever cleared the private bit is responsible
4850                  * for the finish_ordered_io
4851                  */
4852                 if (TestClearPagePrivate2(page)) {
4853                         btrfs_finish_ordered_io(page->mapping->host,
4854                                                 page_start, page_end);
4855                 }
4856                 btrfs_put_ordered_extent(ordered);
4857                 lock_extent(tree, page_start, page_end, GFP_NOFS);
4858         }
4859         clear_extent_bit(tree, page_start, page_end,
4860                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4861                  EXTENT_DO_ACCOUNTING, 1, 1, NULL, GFP_NOFS);
4862         __btrfs_releasepage(page, GFP_NOFS);
4863
4864         ClearPageChecked(page);
4865         if (PagePrivate(page)) {
4866                 ClearPagePrivate(page);
4867                 set_page_private(page, 0);
4868                 page_cache_release(page);
4869         }
4870 }
4871
4872 /*
4873  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
4874  * called from a page fault handler when a page is first dirtied. Hence we must
4875  * be careful to check for EOF conditions here. We set the page up correctly
4876  * for a written page which means we get ENOSPC checking when writing into
4877  * holes and correct delalloc and unwritten extent mapping on filesystems that
4878  * support these features.
4879  *
4880  * We are not allowed to take the i_mutex here so we have to play games to
4881  * protect against truncate races as the page could now be beyond EOF.  Because
4882  * vmtruncate() writes the inode size before removing pages, once we have the
4883  * page lock we can determine safely if the page is beyond EOF. If it is not
4884  * beyond EOF, then the page is guaranteed safe against truncation until we
4885  * unlock the page.
4886  */
4887 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
4888 {
4889         struct page *page = vmf->page;
4890         struct inode *inode = fdentry(vma->vm_file)->d_inode;
4891         struct btrfs_root *root = BTRFS_I(inode)->root;
4892         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4893         struct btrfs_ordered_extent *ordered;
4894         char *kaddr;
4895         unsigned long zero_start;
4896         loff_t size;
4897         int ret;
4898         u64 page_start;
4899         u64 page_end;
4900
4901         ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE);
4902         if (ret) {
4903                 if (ret == -ENOMEM)
4904                         ret = VM_FAULT_OOM;
4905                 else /* -ENOSPC, -EIO, etc */
4906                         ret = VM_FAULT_SIGBUS;
4907                 goto out;
4908         }
4909
4910         ret = btrfs_reserve_metadata_for_delalloc(root, inode, 1);
4911         if (ret) {
4912                 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
4913                 ret = VM_FAULT_SIGBUS;
4914                 goto out;
4915         }
4916
4917         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
4918 again:
4919         lock_page(page);
4920         size = i_size_read(inode);
4921         page_start = page_offset(page);
4922         page_end = page_start + PAGE_CACHE_SIZE - 1;
4923
4924         if ((page->mapping != inode->i_mapping) ||
4925             (page_start >= size)) {
4926                 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
4927                 /* page got truncated out from underneath us */
4928                 goto out_unlock;
4929         }
4930         wait_on_page_writeback(page);
4931
4932         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
4933         set_page_extent_mapped(page);
4934
4935         /*
4936          * we can't set the delalloc bits if there are pending ordered
4937          * extents.  Drop our locks and wait for them to finish
4938          */
4939         ordered = btrfs_lookup_ordered_extent(inode, page_start);
4940         if (ordered) {
4941                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4942                 unlock_page(page);
4943                 btrfs_start_ordered_extent(inode, ordered, 1);
4944                 btrfs_put_ordered_extent(ordered);
4945                 goto again;
4946         }
4947
4948         /*
4949          * XXX - page_mkwrite gets called every time the page is dirtied, even
4950          * if it was already dirty, so for space accounting reasons we need to
4951          * clear any delalloc bits for the range we are fixing to save.  There
4952          * is probably a better way to do this, but for now keep consistent with
4953          * prepare_pages in the normal write path.
4954          */
4955         clear_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
4956                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
4957                           GFP_NOFS);
4958
4959         ret = btrfs_set_extent_delalloc(inode, page_start, page_end);
4960         if (ret) {
4961                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4962                 ret = VM_FAULT_SIGBUS;
4963                 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
4964                 goto out_unlock;
4965         }
4966         ret = 0;
4967
4968         /* page is wholly or partially inside EOF */
4969         if (page_start + PAGE_CACHE_SIZE > size)
4970                 zero_start = size & ~PAGE_CACHE_MASK;
4971         else
4972                 zero_start = PAGE_CACHE_SIZE;
4973
4974         if (zero_start != PAGE_CACHE_SIZE) {
4975                 kaddr = kmap(page);
4976                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
4977                 flush_dcache_page(page);
4978                 kunmap(page);
4979         }
4980         ClearPageChecked(page);
4981         set_page_dirty(page);
4982         SetPageUptodate(page);
4983
4984         BTRFS_I(inode)->last_trans = root->fs_info->generation;
4985         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
4986
4987         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4988
4989 out_unlock:
4990         btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
4991         if (!ret)
4992                 return VM_FAULT_LOCKED;
4993         unlock_page(page);
4994 out:
4995         return ret;
4996 }
4997
4998 static void btrfs_truncate(struct inode *inode)
4999 {
5000         struct btrfs_root *root = BTRFS_I(inode)->root;
5001         int ret;
5002         struct btrfs_trans_handle *trans;
5003         unsigned long nr;
5004         u64 mask = root->sectorsize - 1;
5005
5006         if (!S_ISREG(inode->i_mode))
5007                 return;
5008         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
5009                 return;
5010
5011         btrfs_truncate_page(inode->i_mapping, inode->i_size);
5012         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
5013
5014         trans = btrfs_start_transaction(root, 1);
5015
5016         /*
5017          * setattr is responsible for setting the ordered_data_close flag,
5018          * but that is only tested during the last file release.  That
5019          * could happen well after the next commit, leaving a great big
5020          * window where new writes may get lost if someone chooses to write
5021          * to this file after truncating to zero
5022          *
5023          * The inode doesn't have any dirty data here, and so if we commit
5024          * this is a noop.  If someone immediately starts writing to the inode
5025          * it is very likely we'll catch some of their writes in this
5026          * transaction, and the commit will find this file on the ordered
5027          * data list with good things to send down.
5028          *
5029          * This is a best effort solution, there is still a window where
5030          * using truncate to replace the contents of the file will
5031          * end up with a zero length file after a crash.
5032          */
5033         if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
5034                 btrfs_add_ordered_operation(trans, root, inode);
5035
5036         btrfs_set_trans_block_group(trans, inode);
5037         btrfs_i_size_write(inode, inode->i_size);
5038
5039         ret = btrfs_orphan_add(trans, inode);
5040         if (ret)
5041                 goto out;
5042         /* FIXME, add redo link to tree so we don't leak on crash */
5043         ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size,
5044                                       BTRFS_EXTENT_DATA_KEY);
5045         btrfs_update_inode(trans, root, inode);
5046
5047         ret = btrfs_orphan_del(trans, inode);
5048         BUG_ON(ret);
5049
5050 out:
5051         nr = trans->blocks_used;
5052         ret = btrfs_end_transaction_throttle(trans, root);
5053         BUG_ON(ret);
5054         btrfs_btree_balance_dirty(root, nr);
5055 }
5056
5057 /*
5058  * create a new subvolume directory/inode (helper for the ioctl).
5059  */
5060 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
5061                              struct btrfs_root *new_root,
5062                              u64 new_dirid, u64 alloc_hint)
5063 {
5064         struct inode *inode;
5065         int err;
5066         u64 index = 0;
5067
5068         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
5069                                 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
5070         if (IS_ERR(inode))
5071                 return PTR_ERR(inode);
5072         inode->i_op = &btrfs_dir_inode_operations;
5073         inode->i_fop = &btrfs_dir_file_operations;
5074
5075         inode->i_nlink = 1;
5076         btrfs_i_size_write(inode, 0);
5077
5078         err = btrfs_update_inode(trans, new_root, inode);
5079         BUG_ON(err);
5080
5081         iput(inode);
5082         return 0;
5083 }
5084
5085 /* helper function for file defrag and space balancing.  This
5086  * forces readahead on a given range of bytes in an inode
5087  */
5088 unsigned long btrfs_force_ra(struct address_space *mapping,
5089                               struct file_ra_state *ra, struct file *file,
5090                               pgoff_t offset, pgoff_t last_index)
5091 {
5092         pgoff_t req_size = last_index - offset + 1;
5093
5094         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
5095         return offset + req_size;
5096 }
5097
5098 struct inode *btrfs_alloc_inode(struct super_block *sb)
5099 {
5100         struct btrfs_inode *ei;
5101
5102         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
5103         if (!ei)
5104                 return NULL;
5105         ei->last_trans = 0;
5106         ei->last_sub_trans = 0;
5107         ei->logged_trans = 0;
5108         ei->outstanding_extents = 0;
5109         ei->reserved_extents = 0;
5110         spin_lock_init(&ei->accounting_lock);
5111         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
5112         INIT_LIST_HEAD(&ei->i_orphan);
5113         INIT_LIST_HEAD(&ei->ordered_operations);
5114         return &ei->vfs_inode;
5115 }
5116
5117 void btrfs_destroy_inode(struct inode *inode)
5118 {
5119         struct btrfs_ordered_extent *ordered;
5120         struct btrfs_root *root = BTRFS_I(inode)->root;
5121
5122         WARN_ON(!list_empty(&inode->i_dentry));
5123         WARN_ON(inode->i_data.nrpages);
5124
5125         /*
5126          * Make sure we're properly removed from the ordered operation
5127          * lists.
5128          */
5129         smp_mb();
5130         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
5131                 spin_lock(&root->fs_info->ordered_extent_lock);
5132                 list_del_init(&BTRFS_I(inode)->ordered_operations);
5133                 spin_unlock(&root->fs_info->ordered_extent_lock);
5134         }
5135
5136         spin_lock(&root->list_lock);
5137         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
5138                 printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
5139                        " list\n", inode->i_ino);
5140                 dump_stack();
5141         }
5142         spin_unlock(&root->list_lock);
5143
5144         while (1) {
5145                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
5146                 if (!ordered)
5147                         break;
5148                 else {
5149                         printk(KERN_ERR "btrfs found ordered "
5150                                "extent %llu %llu on inode cleanup\n",
5151                                (unsigned long long)ordered->file_offset,
5152                                (unsigned long long)ordered->len);
5153                         btrfs_remove_ordered_extent(inode, ordered);
5154                         btrfs_put_ordered_extent(ordered);
5155                         btrfs_put_ordered_extent(ordered);
5156                 }
5157         }
5158         inode_tree_del(inode);
5159         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
5160         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
5161 }
5162
5163 void btrfs_drop_inode(struct inode *inode)
5164 {
5165         struct btrfs_root *root = BTRFS_I(inode)->root;
5166
5167         if (inode->i_nlink > 0 && btrfs_root_refs(&root->root_item) == 0)
5168                 generic_delete_inode(inode);
5169         else
5170                 generic_drop_inode(inode);
5171 }
5172
5173 static void init_once(void *foo)
5174 {
5175         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
5176
5177         inode_init_once(&ei->vfs_inode);
5178 }
5179
5180 void btrfs_destroy_cachep(void)
5181 {
5182         if (btrfs_inode_cachep)
5183                 kmem_cache_destroy(btrfs_inode_cachep);
5184         if (btrfs_trans_handle_cachep)
5185                 kmem_cache_destroy(btrfs_trans_handle_cachep);
5186         if (btrfs_transaction_cachep)
5187                 kmem_cache_destroy(btrfs_transaction_cachep);
5188         if (btrfs_path_cachep)
5189                 kmem_cache_destroy(btrfs_path_cachep);
5190 }
5191
5192 int btrfs_init_cachep(void)
5193 {
5194         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
5195                         sizeof(struct btrfs_inode), 0,
5196                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
5197         if (!btrfs_inode_cachep)
5198                 goto fail;
5199
5200         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
5201                         sizeof(struct btrfs_trans_handle), 0,
5202                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
5203         if (!btrfs_trans_handle_cachep)
5204                 goto fail;
5205
5206         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
5207                         sizeof(struct btrfs_transaction), 0,
5208                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
5209         if (!btrfs_transaction_cachep)
5210                 goto fail;
5211
5212         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
5213                         sizeof(struct btrfs_path), 0,
5214                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
5215         if (!btrfs_path_cachep)
5216                 goto fail;
5217
5218         return 0;
5219 fail:
5220         btrfs_destroy_cachep();
5221         return -ENOMEM;
5222 }
5223
5224 static int btrfs_getattr(struct vfsmount *mnt,
5225                          struct dentry *dentry, struct kstat *stat)
5226 {
5227         struct inode *inode = dentry->d_inode;
5228         generic_fillattr(inode, stat);
5229         stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
5230         stat->blksize = PAGE_CACHE_SIZE;
5231         stat->blocks = (inode_get_bytes(inode) +
5232                         BTRFS_I(inode)->delalloc_bytes) >> 9;
5233         return 0;
5234 }
5235
5236 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
5237                            struct inode *new_dir, struct dentry *new_dentry)
5238 {
5239         struct btrfs_trans_handle *trans;
5240         struct btrfs_root *root = BTRFS_I(old_dir)->root;
5241         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
5242         struct inode *new_inode = new_dentry->d_inode;
5243         struct inode *old_inode = old_dentry->d_inode;
5244         struct timespec ctime = CURRENT_TIME;
5245         u64 index = 0;
5246         u64 root_objectid;
5247         int ret;
5248
5249         if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5250                 return -EPERM;
5251
5252         /* we only allow rename subvolume link between subvolumes */
5253         if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
5254                 return -EXDEV;
5255
5256         if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
5257             (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID))
5258                 return -ENOTEMPTY;
5259
5260         if (S_ISDIR(old_inode->i_mode) && new_inode &&
5261             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
5262                 return -ENOTEMPTY;
5263
5264         /*
5265          * 2 items for dir items
5266          * 1 item for orphan entry
5267          * 1 item for ref
5268          */
5269         ret = btrfs_reserve_metadata_space(root, 4);
5270         if (ret)
5271                 return ret;
5272
5273         /*
5274          * we're using rename to replace one file with another.
5275          * and the replacement file is large.  Start IO on it now so
5276          * we don't add too much work to the end of the transaction
5277          */
5278         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
5279             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
5280                 filemap_flush(old_inode->i_mapping);
5281
5282         /* close the racy window with snapshot create/destroy ioctl */
5283         if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
5284                 down_read(&root->fs_info->subvol_sem);
5285
5286         trans = btrfs_start_transaction(root, 1);
5287         btrfs_set_trans_block_group(trans, new_dir);
5288
5289         if (dest != root)
5290                 btrfs_record_root_in_trans(trans, dest);
5291
5292         ret = btrfs_set_inode_index(new_dir, &index);
5293         if (ret)
5294                 goto out_fail;
5295
5296         if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
5297                 /* force full log commit if subvolume involved. */
5298                 root->fs_info->last_trans_log_full_commit = trans->transid;
5299         } else {
5300                 ret = btrfs_insert_inode_ref(trans, dest,
5301                                              new_dentry->d_name.name,
5302                                              new_dentry->d_name.len,
5303                                              old_inode->i_ino,
5304                                              new_dir->i_ino, index);
5305                 if (ret)
5306                         goto out_fail;
5307                 /*
5308                  * this is an ugly little race, but the rename is required
5309                  * to make sure that if we crash, the inode is either at the
5310                  * old name or the new one.  pinning the log transaction lets
5311                  * us make sure we don't allow a log commit to come in after
5312                  * we unlink the name but before we add the new name back in.
5313                  */
5314                 btrfs_pin_log_trans(root);
5315         }
5316         /*
5317          * make sure the inode gets flushed if it is replacing
5318          * something.
5319          */
5320         if (new_inode && new_inode->i_size &&
5321             old_inode && S_ISREG(old_inode->i_mode)) {
5322                 btrfs_add_ordered_operation(trans, root, old_inode);
5323         }
5324
5325         old_dir->i_ctime = old_dir->i_mtime = ctime;
5326         new_dir->i_ctime = new_dir->i_mtime = ctime;
5327         old_inode->i_ctime = ctime;
5328
5329         if (old_dentry->d_parent != new_dentry->d_parent)
5330                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
5331
5332         if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
5333                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
5334                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
5335                                         old_dentry->d_name.name,
5336                                         old_dentry->d_name.len);
5337         } else {
5338                 btrfs_inc_nlink(old_dentry->d_inode);
5339                 ret = btrfs_unlink_inode(trans, root, old_dir,
5340                                          old_dentry->d_inode,
5341                                          old_dentry->d_name.name,
5342                                          old_dentry->d_name.len);
5343         }
5344         BUG_ON(ret);
5345
5346         if (new_inode) {
5347                 new_inode->i_ctime = CURRENT_TIME;
5348                 if (unlikely(new_inode->i_ino ==
5349                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
5350                         root_objectid = BTRFS_I(new_inode)->location.objectid;
5351                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
5352                                                 root_objectid,
5353                                                 new_dentry->d_name.name,
5354                                                 new_dentry->d_name.len);
5355                         BUG_ON(new_inode->i_nlink == 0);
5356                 } else {
5357                         ret = btrfs_unlink_inode(trans, dest, new_dir,
5358                                                  new_dentry->d_inode,
5359                                                  new_dentry->d_name.name,
5360                                                  new_dentry->d_name.len);
5361                 }
5362                 BUG_ON(ret);
5363                 if (new_inode->i_nlink == 0) {
5364                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
5365                         BUG_ON(ret);
5366                 }
5367         }
5368
5369         ret = btrfs_add_link(trans, new_dir, old_inode,
5370                              new_dentry->d_name.name,
5371                              new_dentry->d_name.len, 0, index);
5372         BUG_ON(ret);
5373
5374         if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
5375                 btrfs_log_new_name(trans, old_inode, old_dir,
5376                                    new_dentry->d_parent);
5377                 btrfs_end_log_trans(root);
5378         }
5379 out_fail:
5380         btrfs_end_transaction_throttle(trans, root);
5381
5382         if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
5383                 up_read(&root->fs_info->subvol_sem);
5384
5385         btrfs_unreserve_metadata_space(root, 4);
5386         return ret;
5387 }
5388
5389 /*
5390  * some fairly slow code that needs optimization. This walks the list
5391  * of all the inodes with pending delalloc and forces them to disk.
5392  */
5393 int btrfs_start_delalloc_inodes(struct btrfs_root *root)
5394 {
5395         struct list_head *head = &root->fs_info->delalloc_inodes;
5396         struct btrfs_inode *binode;
5397         struct inode *inode;
5398
5399         if (root->fs_info->sb->s_flags & MS_RDONLY)
5400                 return -EROFS;
5401
5402         spin_lock(&root->fs_info->delalloc_lock);
5403         while (!list_empty(head)) {
5404                 binode = list_entry(head->next, struct btrfs_inode,
5405                                     delalloc_inodes);
5406                 inode = igrab(&binode->vfs_inode);
5407                 if (!inode)
5408                         list_del_init(&binode->delalloc_inodes);
5409                 spin_unlock(&root->fs_info->delalloc_lock);
5410                 if (inode) {
5411                         filemap_flush(inode->i_mapping);
5412                         iput(inode);
5413                 }
5414                 cond_resched();
5415                 spin_lock(&root->fs_info->delalloc_lock);
5416         }
5417         spin_unlock(&root->fs_info->delalloc_lock);
5418
5419         /* the filemap_flush will queue IO into the worker threads, but
5420          * we have to make sure the IO is actually started and that
5421          * ordered extents get created before we return
5422          */
5423         atomic_inc(&root->fs_info->async_submit_draining);
5424         while (atomic_read(&root->fs_info->nr_async_submits) ||
5425               atomic_read(&root->fs_info->async_delalloc_pages)) {
5426                 wait_event(root->fs_info->async_submit_wait,
5427                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
5428                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
5429         }
5430         atomic_dec(&root->fs_info->async_submit_draining);
5431         return 0;
5432 }
5433
5434 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
5435                          const char *symname)
5436 {
5437         struct btrfs_trans_handle *trans;
5438         struct btrfs_root *root = BTRFS_I(dir)->root;
5439         struct btrfs_path *path;
5440         struct btrfs_key key;
5441         struct inode *inode = NULL;
5442         int err;
5443         int drop_inode = 0;
5444         u64 objectid;
5445         u64 index = 0 ;
5446         int name_len;
5447         int datasize;
5448         unsigned long ptr;
5449         struct btrfs_file_extent_item *ei;
5450         struct extent_buffer *leaf;
5451         unsigned long nr = 0;
5452
5453         name_len = strlen(symname) + 1;
5454         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
5455                 return -ENAMETOOLONG;
5456
5457         /*
5458          * 2 items for inode item and ref
5459          * 2 items for dir items
5460          * 1 item for xattr if selinux is on
5461          */
5462         err = btrfs_reserve_metadata_space(root, 5);
5463         if (err)
5464                 return err;
5465
5466         trans = btrfs_start_transaction(root, 1);
5467         if (!trans)
5468                 goto out_fail;
5469         btrfs_set_trans_block_group(trans, dir);
5470
5471         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
5472         if (err) {
5473                 err = -ENOSPC;
5474                 goto out_unlock;
5475         }
5476
5477         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5478                                 dentry->d_name.len,
5479                                 dentry->d_parent->d_inode->i_ino, objectid,
5480                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
5481                                 &index);
5482         err = PTR_ERR(inode);
5483         if (IS_ERR(inode))
5484                 goto out_unlock;
5485
5486         err = btrfs_init_inode_security(inode, dir);
5487         if (err) {
5488                 drop_inode = 1;
5489                 goto out_unlock;
5490         }
5491
5492         btrfs_set_trans_block_group(trans, inode);
5493         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
5494         if (err)
5495                 drop_inode = 1;
5496         else {
5497                 inode->i_mapping->a_ops = &btrfs_aops;
5498                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5499                 inode->i_fop = &btrfs_file_operations;
5500                 inode->i_op = &btrfs_file_inode_operations;
5501                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5502         }
5503         btrfs_update_inode_block_group(trans, inode);
5504         btrfs_update_inode_block_group(trans, dir);
5505         if (drop_inode)
5506                 goto out_unlock;
5507
5508         path = btrfs_alloc_path();
5509         BUG_ON(!path);
5510         key.objectid = inode->i_ino;
5511         key.offset = 0;
5512         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
5513         datasize = btrfs_file_extent_calc_inline_size(name_len);
5514         err = btrfs_insert_empty_item(trans, root, path, &key,
5515                                       datasize);
5516         if (err) {
5517                 drop_inode = 1;
5518                 goto out_unlock;
5519         }
5520         leaf = path->nodes[0];
5521         ei = btrfs_item_ptr(leaf, path->slots[0],
5522                             struct btrfs_file_extent_item);
5523         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
5524         btrfs_set_file_extent_type(leaf, ei,
5525                                    BTRFS_FILE_EXTENT_INLINE);
5526         btrfs_set_file_extent_encryption(leaf, ei, 0);
5527         btrfs_set_file_extent_compression(leaf, ei, 0);
5528         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
5529         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
5530
5531         ptr = btrfs_file_extent_inline_start(ei);
5532         write_extent_buffer(leaf, symname, ptr, name_len);
5533         btrfs_mark_buffer_dirty(leaf);
5534         btrfs_free_path(path);
5535
5536         inode->i_op = &btrfs_symlink_inode_operations;
5537         inode->i_mapping->a_ops = &btrfs_symlink_aops;
5538         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5539         inode_set_bytes(inode, name_len);
5540         btrfs_i_size_write(inode, name_len - 1);
5541         err = btrfs_update_inode(trans, root, inode);
5542         if (err)
5543                 drop_inode = 1;
5544
5545 out_unlock:
5546         nr = trans->blocks_used;
5547         btrfs_end_transaction_throttle(trans, root);
5548 out_fail:
5549         btrfs_unreserve_metadata_space(root, 5);
5550         if (drop_inode) {
5551                 inode_dec_link_count(inode);
5552                 iput(inode);
5553         }
5554         btrfs_btree_balance_dirty(root, nr);
5555         return err;
5556 }
5557
5558 static int prealloc_file_range(struct btrfs_trans_handle *trans,
5559                                struct inode *inode, u64 start, u64 end,
5560                                u64 locked_end, u64 alloc_hint, int mode)
5561 {
5562         struct btrfs_root *root = BTRFS_I(inode)->root;
5563         struct btrfs_key ins;
5564         u64 alloc_size;
5565         u64 cur_offset = start;
5566         u64 num_bytes = end - start;
5567         int ret = 0;
5568
5569         while (num_bytes > 0) {
5570                 alloc_size = min(num_bytes, root->fs_info->max_extent);
5571
5572                 ret = btrfs_reserve_metadata_space(root, 1);
5573                 if (ret)
5574                         goto out;
5575
5576                 ret = btrfs_reserve_extent(trans, root, alloc_size,
5577                                            root->sectorsize, 0, alloc_hint,
5578                                            (u64)-1, &ins, 1);
5579                 if (ret) {
5580                         WARN_ON(1);
5581                         goto out;
5582                 }
5583                 ret = insert_reserved_file_extent(trans, inode,
5584                                                   cur_offset, ins.objectid,
5585                                                   ins.offset, ins.offset,
5586                                                   ins.offset, locked_end,
5587                                                   0, 0, 0,
5588                                                   BTRFS_FILE_EXTENT_PREALLOC);
5589                 BUG_ON(ret);
5590                 btrfs_drop_extent_cache(inode, cur_offset,
5591                                         cur_offset + ins.offset -1, 0);
5592                 num_bytes -= ins.offset;
5593                 cur_offset += ins.offset;
5594                 alloc_hint = ins.objectid + ins.offset;
5595                 btrfs_unreserve_metadata_space(root, 1);
5596         }
5597 out:
5598         if (cur_offset > start) {
5599                 inode->i_ctime = CURRENT_TIME;
5600                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
5601                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
5602                     cur_offset > i_size_read(inode))
5603                         btrfs_i_size_write(inode, cur_offset);
5604                 ret = btrfs_update_inode(trans, root, inode);
5605                 BUG_ON(ret);
5606         }
5607
5608         return ret;
5609 }
5610
5611 static long btrfs_fallocate(struct inode *inode, int mode,
5612                             loff_t offset, loff_t len)
5613 {
5614         u64 cur_offset;
5615         u64 last_byte;
5616         u64 alloc_start;
5617         u64 alloc_end;
5618         u64 alloc_hint = 0;
5619         u64 locked_end;
5620         u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
5621         struct extent_map *em;
5622         struct btrfs_trans_handle *trans;
5623         struct btrfs_root *root;
5624         int ret;
5625
5626         alloc_start = offset & ~mask;
5627         alloc_end =  (offset + len + mask) & ~mask;
5628
5629         /*
5630          * wait for ordered IO before we have any locks.  We'll loop again
5631          * below with the locks held.
5632          */
5633         btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start);
5634
5635         mutex_lock(&inode->i_mutex);
5636         if (alloc_start > inode->i_size) {
5637                 ret = btrfs_cont_expand(inode, alloc_start);
5638                 if (ret)
5639                         goto out;
5640         }
5641
5642         root = BTRFS_I(inode)->root;
5643
5644         ret = btrfs_check_data_free_space(root, inode,
5645                                           alloc_end - alloc_start);
5646         if (ret)
5647                 goto out;
5648
5649         locked_end = alloc_end - 1;
5650         while (1) {
5651                 struct btrfs_ordered_extent *ordered;
5652
5653                 trans = btrfs_start_transaction(BTRFS_I(inode)->root, 1);
5654                 if (!trans) {
5655                         ret = -EIO;
5656                         goto out_free;
5657                 }
5658
5659                 /* the extent lock is ordered inside the running
5660                  * transaction
5661                  */
5662                 lock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
5663                             GFP_NOFS);
5664                 ordered = btrfs_lookup_first_ordered_extent(inode,
5665                                                             alloc_end - 1);
5666                 if (ordered &&
5667                     ordered->file_offset + ordered->len > alloc_start &&
5668                     ordered->file_offset < alloc_end) {
5669                         btrfs_put_ordered_extent(ordered);
5670                         unlock_extent(&BTRFS_I(inode)->io_tree,
5671                                       alloc_start, locked_end, GFP_NOFS);
5672                         btrfs_end_transaction(trans, BTRFS_I(inode)->root);
5673
5674                         /*
5675                          * we can't wait on the range with the transaction
5676                          * running or with the extent lock held
5677                          */
5678                         btrfs_wait_ordered_range(inode, alloc_start,
5679                                                  alloc_end - alloc_start);
5680                 } else {
5681                         if (ordered)
5682                                 btrfs_put_ordered_extent(ordered);
5683                         break;
5684                 }
5685         }
5686
5687         cur_offset = alloc_start;
5688         while (1) {
5689                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
5690                                       alloc_end - cur_offset, 0);
5691                 BUG_ON(IS_ERR(em) || !em);
5692                 last_byte = min(extent_map_end(em), alloc_end);
5693                 last_byte = (last_byte + mask) & ~mask;
5694                 if (em->block_start == EXTENT_MAP_HOLE) {
5695                         ret = prealloc_file_range(trans, inode, cur_offset,
5696                                         last_byte, locked_end + 1,
5697                                         alloc_hint, mode);
5698                         if (ret < 0) {
5699                                 free_extent_map(em);
5700                                 break;
5701                         }
5702                 }
5703                 if (em->block_start <= EXTENT_MAP_LAST_BYTE)
5704                         alloc_hint = em->block_start;
5705                 free_extent_map(em);
5706
5707                 cur_offset = last_byte;
5708                 if (cur_offset >= alloc_end) {
5709                         ret = 0;
5710                         break;
5711                 }
5712         }
5713         unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
5714                       GFP_NOFS);
5715
5716         btrfs_end_transaction(trans, BTRFS_I(inode)->root);
5717 out_free:
5718         btrfs_free_reserved_data_space(root, inode, alloc_end - alloc_start);
5719 out:
5720         mutex_unlock(&inode->i_mutex);
5721         return ret;
5722 }
5723
5724 static int btrfs_set_page_dirty(struct page *page)
5725 {
5726         return __set_page_dirty_nobuffers(page);
5727 }
5728
5729 static int btrfs_permission(struct inode *inode, int mask)
5730 {
5731         if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
5732                 return -EACCES;
5733         return generic_permission(inode, mask, btrfs_check_acl);
5734 }
5735
5736 static struct inode_operations btrfs_dir_inode_operations = {
5737         .getattr        = btrfs_getattr,
5738         .lookup         = btrfs_lookup,
5739         .create         = btrfs_create,
5740         .unlink         = btrfs_unlink,
5741         .link           = btrfs_link,
5742         .mkdir          = btrfs_mkdir,
5743         .rmdir          = btrfs_rmdir,
5744         .rename         = btrfs_rename,
5745         .symlink        = btrfs_symlink,
5746         .setattr        = btrfs_setattr,
5747         .mknod          = btrfs_mknod,
5748         .setxattr       = btrfs_setxattr,
5749         .getxattr       = btrfs_getxattr,
5750         .listxattr      = btrfs_listxattr,
5751         .removexattr    = btrfs_removexattr,
5752         .permission     = btrfs_permission,
5753 };
5754 static struct inode_operations btrfs_dir_ro_inode_operations = {
5755         .lookup         = btrfs_lookup,
5756         .permission     = btrfs_permission,
5757 };
5758
5759 static struct file_operations btrfs_dir_file_operations = {
5760         .llseek         = generic_file_llseek,
5761         .read           = generic_read_dir,
5762         .readdir        = btrfs_real_readdir,
5763         .unlocked_ioctl = btrfs_ioctl,
5764 #ifdef CONFIG_COMPAT
5765         .compat_ioctl   = btrfs_ioctl,
5766 #endif
5767         .release        = btrfs_release_file,
5768         .fsync          = btrfs_sync_file,
5769 };
5770
5771 static struct extent_io_ops btrfs_extent_io_ops = {
5772         .fill_delalloc = run_delalloc_range,
5773         .submit_bio_hook = btrfs_submit_bio_hook,
5774         .merge_bio_hook = btrfs_merge_bio_hook,
5775         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
5776         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
5777         .writepage_start_hook = btrfs_writepage_start_hook,
5778         .readpage_io_failed_hook = btrfs_io_failed_hook,
5779         .set_bit_hook = btrfs_set_bit_hook,
5780         .clear_bit_hook = btrfs_clear_bit_hook,
5781         .merge_extent_hook = btrfs_merge_extent_hook,
5782         .split_extent_hook = btrfs_split_extent_hook,
5783 };
5784
5785 /*
5786  * btrfs doesn't support the bmap operation because swapfiles
5787  * use bmap to make a mapping of extents in the file.  They assume
5788  * these extents won't change over the life of the file and they
5789  * use the bmap result to do IO directly to the drive.
5790  *
5791  * the btrfs bmap call would return logical addresses that aren't
5792  * suitable for IO and they also will change frequently as COW
5793  * operations happen.  So, swapfile + btrfs == corruption.
5794  *
5795  * For now we're avoiding this by dropping bmap.
5796  */
5797 static struct address_space_operations btrfs_aops = {
5798         .readpage       = btrfs_readpage,
5799         .writepage      = btrfs_writepage,
5800         .writepages     = btrfs_writepages,
5801         .readpages      = btrfs_readpages,
5802         .sync_page      = block_sync_page,
5803         .direct_IO      = btrfs_direct_IO,
5804         .invalidatepage = btrfs_invalidatepage,
5805         .releasepage    = btrfs_releasepage,
5806         .set_page_dirty = btrfs_set_page_dirty,
5807 };
5808
5809 static struct address_space_operations btrfs_symlink_aops = {
5810         .readpage       = btrfs_readpage,
5811         .writepage      = btrfs_writepage,
5812         .invalidatepage = btrfs_invalidatepage,
5813         .releasepage    = btrfs_releasepage,
5814 };
5815
5816 static struct inode_operations btrfs_file_inode_operations = {
5817         .truncate       = btrfs_truncate,
5818         .getattr        = btrfs_getattr,
5819         .setattr        = btrfs_setattr,
5820         .setxattr       = btrfs_setxattr,
5821         .getxattr       = btrfs_getxattr,
5822         .listxattr      = btrfs_listxattr,
5823         .removexattr    = btrfs_removexattr,
5824         .permission     = btrfs_permission,
5825         .fallocate      = btrfs_fallocate,
5826         .fiemap         = btrfs_fiemap,
5827 };
5828 static struct inode_operations btrfs_special_inode_operations = {
5829         .getattr        = btrfs_getattr,
5830         .setattr        = btrfs_setattr,
5831         .permission     = btrfs_permission,
5832         .setxattr       = btrfs_setxattr,
5833         .getxattr       = btrfs_getxattr,
5834         .listxattr      = btrfs_listxattr,
5835         .removexattr    = btrfs_removexattr,
5836 };
5837 static struct inode_operations btrfs_symlink_inode_operations = {
5838         .readlink       = generic_readlink,
5839         .follow_link    = page_follow_link_light,
5840         .put_link       = page_put_link,
5841         .permission     = btrfs_permission,
5842         .setxattr       = btrfs_setxattr,
5843         .getxattr       = btrfs_getxattr,
5844         .listxattr      = btrfs_listxattr,
5845         .removexattr    = btrfs_removexattr,
5846 };
5847
5848 const struct dentry_operations btrfs_dentry_operations = {
5849         .d_delete       = btrfs_dentry_delete,
5850 };