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/smp_lock.h>
  30 #include <linux/backing-dev.h>
  31 #include <linux/mpage.h>
  32 #include <linux/swap.h>
  33 #include <linux/writeback.h>
  34 #include <linux/statfs.h>
  35 #include <linux/compat.h>
  36 #include <linux/bit_spinlock.h>
  37 #include <linux/version.h>
  38 #include <linux/xattr.h>
  39 #include <linux/posix_acl.h>
  40 #include <linux/falloc.h>
  41 #include "compat.h"
  42 #include "ctree.h"
  43 #include "disk-io.h"
  44 #include "transaction.h"
  45 #include "btrfs_inode.h"
  46 #include "ioctl.h"
  47 #include "print-tree.h"
  48 #include "volumes.h"
  49 #include "ordered-data.h"
  50 #include "xattr.h"
  51 #include "tree-log.h"
  52 #include "ref-cache.h"
  53 #include "compression.h"
  54
  55 struct btrfs_iget_args {
  56         u64 ino;
  57         struct btrfs_root *root;
  58 };
  59
  60 static struct inode_operations btrfs_dir_inode_operations;
  61 static struct inode_operations btrfs_symlink_inode_operations;
  62 static struct inode_operations btrfs_dir_ro_inode_operations;
  63 static struct inode_operations btrfs_special_inode_operations;
  64 static struct inode_operations btrfs_file_inode_operations;
  65 static struct address_space_operations btrfs_aops;
  66 static struct address_space_operations btrfs_symlink_aops;
  67 static struct file_operations btrfs_dir_file_operations;
  68 static struct extent_io_ops btrfs_extent_io_ops;
  69
  70 static struct kmem_cache *btrfs_inode_cachep;
  71 struct kmem_cache *btrfs_trans_handle_cachep;
  72 struct kmem_cache *btrfs_transaction_cachep;
  73 struct kmem_cache *btrfs_bit_radix_cachep;
  74 struct kmem_cache *btrfs_path_cachep;
  75
  76 #define S_SHIFT 12
  77 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  78         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  79         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  80         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  81         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  82         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  83         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  84         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  85 };
  86
  87 static void btrfs_truncate(struct inode *inode);
  88 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
  89 static noinline int cow_file_range(struct inode *inode,
  90                                    struct page *locked_page,
  91                                    u64 start, u64 end, int *page_started,
  92                                    unsigned long *nr_written, int unlock);
  93
  94 /*
  95  * a very lame attempt at stopping writes when the FS is 85% full.  There
  96  * are countless ways this is incorrect, but it is better than nothing.
  97  */
  98 int btrfs_check_free_space(struct btrfs_root *root, u64 num_required,
  99                            int for_del)
 100 {
 101         u64 total;
 102         u64 used;
 103         u64 thresh;
 104         int ret = 0;
 105
 106         spin_lock(&root->fs_info->delalloc_lock);
 107         total = btrfs_super_total_bytes(&root->fs_info->super_copy);
 108         used = btrfs_super_bytes_used(&root->fs_info->super_copy);
 109         if (for_del)
 110                 thresh = total * 90;
 111         else
 112                 thresh = total * 85;
 113
 114         do_div(thresh, 100);
 115
 116         if (used + root->fs_info->delalloc_bytes + num_required > thresh)
 117                 ret = -ENOSPC;
 118         spin_unlock(&root->fs_info->delalloc_lock);
 119         return ret;
 120 }
 121
 122 /*
 123  * this does all the hard work for inserting an inline extent into
 124  * the btree.  The caller should have done a btrfs_drop_extents so that
 125  * no overlapping inline items exist in the btree
 126  */
 127 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 128                                 struct btrfs_root *root, struct inode *inode,
 129                                 u64 start, size_t size, size_t compressed_size,
 130                                 struct page **compressed_pages)
 131 {
 132         struct btrfs_key key;
 133         struct btrfs_path *path;
 134         struct extent_buffer *leaf;
 135         struct page *page = NULL;
 136         char *kaddr;
 137         unsigned long ptr;
 138         struct btrfs_file_extent_item *ei;
 139         int err = 0;
 140         int ret;
 141         size_t cur_size = size;
 142         size_t datasize;
 143         unsigned long offset;
 144         int use_compress = 0;
 145
 146         if (compressed_size && compressed_pages) {
 147                 use_compress = 1;
 148                 cur_size = compressed_size;
 149         }
 150
 151         path = btrfs_alloc_path();
 152         if (!path)
 153                 return -ENOMEM;
 154
 155         btrfs_set_trans_block_group(trans, inode);
 156
 157         key.objectid = inode->i_ino;
 158         key.offset = start;
 159         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 160         datasize = btrfs_file_extent_calc_inline_size(cur_size);
 161
 162         inode_add_bytes(inode, size);
 163         ret = btrfs_insert_empty_item(trans, root, path, &key,
 164                                       datasize);
 165         BUG_ON(ret);
 166         if (ret) {
 167                 err = ret;
 168                 goto fail;
 169         }
 170         leaf = path->nodes[0];
 171         ei = btrfs_item_ptr(leaf, path->slots[0],
 172                             struct btrfs_file_extent_item);
 173         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 174         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 175         btrfs_set_file_extent_encryption(leaf, ei, 0);
 176         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 177         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 178         ptr = btrfs_file_extent_inline_start(ei);
 179
 180         if (use_compress) {
 181                 struct page *cpage;
 182                 int i = 0;
 183                 while (compressed_size > 0) {
 184                         cpage = compressed_pages[i];
 185                         cur_size = min_t(unsigned long, compressed_size,
 186                                        PAGE_CACHE_SIZE);
 187
 188                         kaddr = kmap(cpage);
 189                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 190                         kunmap(cpage);
 191
 192                         i++;
 193                         ptr += cur_size;
 194                         compressed_size -= cur_size;
 195                 }
 196                 btrfs_set_file_extent_compression(leaf, ei,
 197                                                   BTRFS_COMPRESS_ZLIB);
 198         } else {
 199                 page = find_get_page(inode->i_mapping,
 200                                      start >> PAGE_CACHE_SHIFT);
 201                 btrfs_set_file_extent_compression(leaf, ei, 0);
 202                 kaddr = kmap_atomic(page, KM_USER0);
 203                 offset = start & (PAGE_CACHE_SIZE - 1);
 204                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 205                 kunmap_atomic(kaddr, KM_USER0);
 206                 page_cache_release(page);
 207         }
 208         btrfs_mark_buffer_dirty(leaf);
 209         btrfs_free_path(path);
 210
 211         BTRFS_I(inode)->disk_i_size = inode->i_size;
 212         btrfs_update_inode(trans, root, inode);
 213         return 0;
 214 fail:
 215         btrfs_free_path(path);
 216         return err;
 217 }
 218
 219
 220 /*
 221  * conditionally insert an inline extent into the file.  This
 222  * does the checks required to make sure the data is small enough
 223  * to fit as an inline extent.
 224  */
 225 static int cow_file_range_inline(struct btrfs_trans_handle *trans,
 226                                  struct btrfs_root *root,
 227                                  struct inode *inode, u64 start, u64 end,
 228                                  size_t compressed_size,
 229                                  struct page **compressed_pages)
 230 {
 231         u64 isize = i_size_read(inode);
 232         u64 actual_end = min(end + 1, isize);
 233         u64 inline_len = actual_end - start;
 234         u64 aligned_end = (end + root->sectorsize - 1) &
 235                         ~((u64)root->sectorsize - 1);
 236         u64 hint_byte;
 237         u64 data_len = inline_len;
 238         int ret;
 239
 240         if (compressed_size)
 241                 data_len = compressed_size;
 242
 243         if (start > 0 ||
 244             actual_end >= PAGE_CACHE_SIZE ||
 245             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 246             (!compressed_size &&
 247             (actual_end & (root->sectorsize - 1)) == 0) ||
 248             end + 1 < isize ||
 249             data_len > root->fs_info->max_inline) {
 250                 return 1;
 251         }
 252
 253         ret = btrfs_drop_extents(trans, root, inode, start,
 254                                  aligned_end, start, &hint_byte);
 255         BUG_ON(ret);
 256
 257         if (isize > actual_end)
 258                 inline_len = min_t(u64, isize, actual_end);
 259         ret = insert_inline_extent(trans, root, inode, start,
 260                                    inline_len, compressed_size,
 261                                    compressed_pages);
 262         BUG_ON(ret);
 263         btrfs_drop_extent_cache(inode, start, aligned_end, 0);
 264         return 0;
 265 }
 266
 267 struct async_extent {
 268         u64 start;
 269         u64 ram_size;
 270         u64 compressed_size;
 271         struct page **pages;
 272         unsigned long nr_pages;
 273         struct list_head list;
 274 };
 275
 276 struct async_cow {
 277         struct inode *inode;
 278         struct btrfs_root *root;
 279         struct page *locked_page;
 280         u64 start;
 281         u64 end;
 282         struct list_head extents;
 283         struct btrfs_work work;
 284 };
 285
 286 static noinline int add_async_extent(struct async_cow *cow,
 287                                      u64 start, u64 ram_size,
 288                                      u64 compressed_size,
 289                                      struct page **pages,
 290                                      unsigned long nr_pages)
 291 {
 292         struct async_extent *async_extent;
 293
 294         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 295         async_extent->start = start;
 296         async_extent->ram_size = ram_size;
 297         async_extent->compressed_size = compressed_size;
 298         async_extent->pages = pages;
 299         async_extent->nr_pages = nr_pages;
 300         list_add_tail(&async_extent->list, &cow->extents);
 301         return 0;
 302 }
 303
 304 /*
 305  * we create compressed extents in two phases.  The first
 306  * phase compresses a range of pages that have already been
 307  * locked (both pages and state bits are locked).
 308  *
 309  * This is done inside an ordered work queue, and the compression
 310  * is spread across many cpus.  The actual IO submission is step
 311  * two, and the ordered work queue takes care of making sure that
 312  * happens in the same order things were put onto the queue by
 313  * writepages and friends.
 314  *
 315  * If this code finds it can't get good compression, it puts an
 316  * entry onto the work queue to write the uncompressed bytes.  This
 317  * makes sure that both compressed inodes and uncompressed inodes
 318  * are written in the same order that pdflush sent them down.
 319  */
 320 static noinline int compress_file_range(struct inode *inode,
 321                                         struct page *locked_page,
 322                                         u64 start, u64 end,
 323                                         struct async_cow *async_cow,
 324                                         int *num_added)
 325 {
 326         struct btrfs_root *root = BTRFS_I(inode)->root;
 327         struct btrfs_trans_handle *trans;
 328         u64 num_bytes;
 329         u64 orig_start;
 330         u64 disk_num_bytes;
 331         u64 blocksize = root->sectorsize;
 332         u64 actual_end;
 333         u64 isize = i_size_read(inode);
 334         int ret = 0;
 335         struct page **pages = NULL;
 336         unsigned long nr_pages;
 337         unsigned long nr_pages_ret = 0;
 338         unsigned long total_compressed = 0;
 339         unsigned long total_in = 0;
 340         unsigned long max_compressed = 128 * 1024;
 341         unsigned long max_uncompressed = 128 * 1024;
 342         int i;
 343         int will_compress;
 344
 345         orig_start = start;
 346
 347         actual_end = min_t(u64, isize, end + 1);
 348 again:
 349         will_compress = 0;
 350         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 351         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 352
 353         total_compressed = actual_end - start;
 354
 355         /* we want to make sure that amount of ram required to uncompress
 356          * an extent is reasonable, so we limit the total size in ram
 357          * of a compressed extent to 128k.  This is a crucial number
 358          * because it also controls how easily we can spread reads across
 359          * cpus for decompression.
 360          *
 361          * We also want to make sure the amount of IO required to do
 362          * a random read is reasonably small, so we limit the size of
 363          * a compressed extent to 128k.
 364          */
 365         total_compressed = min(total_compressed, max_uncompressed);
 366         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 367         num_bytes = max(blocksize,  num_bytes);
 368         disk_num_bytes = num_bytes;
 369         total_in = 0;
 370         ret = 0;
 371
 372         /*
 373          * we do compression for mount -o compress and when the
 374          * inode has not been flagged as nocompress.  This flag can
 375          * change at any time if we discover bad compression ratios.
 376          */
 377         if (!btrfs_test_flag(inode, NOCOMPRESS) &&
 378             btrfs_test_opt(root, COMPRESS)) {
 379                 WARN_ON(pages);
 380                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 381
 382                 ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
 383                                                 total_compressed, pages,
 384                                                 nr_pages, &nr_pages_ret,
 385                                                 &total_in,
 386                                                 &total_compressed,
 387                                                 max_compressed);
 388
 389                 if (!ret) {
 390                         unsigned long offset = total_compressed &
 391                                 (PAGE_CACHE_SIZE - 1);
 392                         struct page *page = pages[nr_pages_ret - 1];
 393                         char *kaddr;
 394
 395                         /* zero the tail end of the last page, we might be
 396                          * sending it down to disk
 397                          */
 398                         if (offset) {
 399                                 kaddr = kmap_atomic(page, KM_USER0);
 400                                 memset(kaddr + offset, 0,
 401                                        PAGE_CACHE_SIZE - offset);
 402                                 kunmap_atomic(kaddr, KM_USER0);
 403                         }
 404                         will_compress = 1;
 405                 }
 406         }
 407         if (start == 0) {
 408                 trans = btrfs_join_transaction(root, 1);
 409                 BUG_ON(!trans);
 410                 btrfs_set_trans_block_group(trans, inode);
 411
 412                 /* lets try to make an inline extent */
 413                 if (ret || total_in < (actual_end - start)) {
 414                         /* we didn't compress the entire range, try
 415                          * to make an uncompressed inline extent.
 416                          */
 417                         ret = cow_file_range_inline(trans, root, inode,
 418                                                     start, end, 0, NULL);
 419                 } else {
 420                         /* try making a compressed inline extent */
 421                         ret = cow_file_range_inline(trans, root, inode,
 422                                                     start, end,
 423                                                     total_compressed, pages);
 424                 }
 425                 btrfs_end_transaction(trans, root);
 426                 if (ret == 0) {
 427                         /*
 428                          * inline extent creation worked, we don't need
 429                          * to create any more async work items.  Unlock
 430                          * and free up our temp pages.
 431                          */
 432                         extent_clear_unlock_delalloc(inode,
 433                                                      &BTRFS_I(inode)->io_tree,
 434                                                      start, end, NULL, 1, 0,
 435                                                      0, 1, 1, 1);
 436                         ret = 0;
 437                         goto free_pages_out;
 438                 }
 439         }
 440
 441         if (will_compress) {
 442                 /*
 443                  * we aren't doing an inline extent round the compressed size
 444                  * up to a block size boundary so the allocator does sane
 445                  * things
 446                  */
 447                 total_compressed = (total_compressed + blocksize - 1) &
 448                         ~(blocksize - 1);
 449
 450                 /*
 451                  * one last check to make sure the compression is really a
 452                  * win, compare the page count read with the blocks on disk
 453                  */
 454                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
 455                         ~(PAGE_CACHE_SIZE - 1);
 456                 if (total_compressed >= total_in) {
 457                         will_compress = 0;
 458                 } else {
 459                         disk_num_bytes = total_compressed;
 460                         num_bytes = total_in;
 461                 }
 462         }
 463         if (!will_compress && pages) {
 464                 /*
 465                  * the compression code ran but failed to make things smaller,
 466                  * free any pages it allocated and our page pointer array
 467                  */
 468                 for (i = 0; i < nr_pages_ret; i++) {
 469                         WARN_ON(pages[i]->mapping);
 470                         page_cache_release(pages[i]);
 471                 }
 472                 kfree(pages);
 473                 pages = NULL;
 474                 total_compressed = 0;
 475                 nr_pages_ret = 0;
 476
 477                 /* flag the file so we don't compress in the future */
 478                 btrfs_set_flag(inode, NOCOMPRESS);
 479         }
 480         if (will_compress) {
 481                 *num_added += 1;
 482
 483                 /* the async work queues will take care of doing actual
 484                  * allocation on disk for these compressed pages,
 485                  * and will submit them to the elevator.
 486                  */
 487                 add_async_extent(async_cow, start, num_bytes,
 488                                  total_compressed, pages, nr_pages_ret);
 489
 490                 if (start + num_bytes < end && start + num_bytes < actual_end) {
 491                         start += num_bytes;
 492                         pages = NULL;
 493                         cond_resched();
 494                         goto again;
 495                 }
 496         } else {
 497                 /*
 498                  * No compression, but we still need to write the pages in
 499                  * the file we've been given so far.  redirty the locked
 500                  * page if it corresponds to our extent and set things up
 501                  * for the async work queue to run cow_file_range to do
 502                  * the normal delalloc dance
 503                  */
 504                 if (page_offset(locked_page) >= start &&
 505                     page_offset(locked_page) <= end) {
 506                         __set_page_dirty_nobuffers(locked_page);
 507                         /* unlocked later on in the async handlers */
 508                 }
 509                 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0);
 510                 *num_added += 1;
 511         }
 512
 513 out:
 514         return 0;
 515
 516 free_pages_out:
 517         for (i = 0; i < nr_pages_ret; i++) {
 518                 WARN_ON(pages[i]->mapping);
 519                 page_cache_release(pages[i]);
 520         }
 521         kfree(pages);
 522
 523         goto out;
 524 }
 525
 526 /*
 527  * phase two of compressed writeback.  This is the ordered portion
 528  * of the code, which only gets called in the order the work was
 529  * queued.  We walk all the async extents created by compress_file_range
 530  * and send them down to the disk.
 531  */
 532 static noinline int submit_compressed_extents(struct inode *inode,
 533                                               struct async_cow *async_cow)
 534 {
 535         struct async_extent *async_extent;
 536         u64 alloc_hint = 0;
 537         struct btrfs_trans_handle *trans;
 538         struct btrfs_key ins;
 539         struct extent_map *em;
 540         struct btrfs_root *root = BTRFS_I(inode)->root;
 541         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 542         struct extent_io_tree *io_tree;
 543         int ret;
 544
 545         if (list_empty(&async_cow->extents))
 546                 return 0;
 547
 548         trans = btrfs_join_transaction(root, 1);
 549
 550         while (!list_empty(&async_cow->extents)) {
 551                 async_extent = list_entry(async_cow->extents.next,
 552                                           struct async_extent, list);
 553                 list_del(&async_extent->list);
 554
 555                 io_tree = &BTRFS_I(inode)->io_tree;
 556
 557                 /* did the compression code fall back to uncompressed IO? */
 558                 if (!async_extent->pages) {
 559                         int page_started = 0;
 560                         unsigned long nr_written = 0;
 561
 562                         lock_extent(io_tree, async_extent->start,
 563                                     async_extent->start +
 564                                     async_extent->ram_size - 1, GFP_NOFS);
 565
 566                         /* allocate blocks */
 567                         cow_file_range(inode, async_cow->locked_page,
 568                                        async_extent->start,
 569                                        async_extent->start +
 570                                        async_extent->ram_size - 1,
 571                                        &page_started, &nr_written, 0);
 572
 573                         /*
 574                          * if page_started, cow_file_range inserted an
 575                          * inline extent and took care of all the unlocking
 576                          * and IO for us.  Otherwise, we need to submit
 577                          * all those pages down to the drive.
 578                          */
 579                         if (!page_started)
 580                                 extent_write_locked_range(io_tree,
 581                                                   inode, async_extent->start,
 582                                                   async_extent->start +
 583                                                   async_extent->ram_size - 1,
 584                                                   btrfs_get_extent,
 585                                                   WB_SYNC_ALL);
 586                         kfree(async_extent);
 587                         cond_resched();
 588                         continue;
 589                 }
 590
 591                 lock_extent(io_tree, async_extent->start,
 592                             async_extent->start + async_extent->ram_size - 1,
 593                             GFP_NOFS);
 594                 /*
 595                  * here we're doing allocation and writeback of the
 596                  * compressed pages
 597                  */
 598                 btrfs_drop_extent_cache(inode, async_extent->start,
 599                                         async_extent->start +
 600                                         async_extent->ram_size - 1, 0);
 601
 602                 ret = btrfs_reserve_extent(trans, root,
 603                                            async_extent->compressed_size,
 604                                            async_extent->compressed_size,
 605                                            0, alloc_hint,
 606                                            (u64)-1, &ins, 1);
 607                 BUG_ON(ret);
 608                 em = alloc_extent_map(GFP_NOFS);
 609                 em->start = async_extent->start;
 610                 em->len = async_extent->ram_size;
 611                 em->orig_start = em->start;
 612
 613                 em->block_start = ins.objectid;
 614                 em->block_len = ins.offset;
 615                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 616                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 617                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 618
 619                 while (1) {
 620                         spin_lock(&em_tree->lock);
 621                         ret = add_extent_mapping(em_tree, em);
 622                         spin_unlock(&em_tree->lock);
 623                         if (ret != -EEXIST) {
 624                                 free_extent_map(em);
 625                                 break;
 626                         }
 627                         btrfs_drop_extent_cache(inode, async_extent->start,
 628                                                 async_extent->start +
 629                                                 async_extent->ram_size - 1, 0);
 630                 }
 631
 632                 ret = btrfs_add_ordered_extent(inode, async_extent->start,
 633                                                ins.objectid,
 634                                                async_extent->ram_size,
 635                                                ins.offset,
 636                                                BTRFS_ORDERED_COMPRESSED);
 637                 BUG_ON(ret);
 638
 639                 btrfs_end_transaction(trans, root);
 640
 641                 /*
 642                  * clear dirty, set writeback and unlock the pages.
 643                  */
 644                 extent_clear_unlock_delalloc(inode,
 645                                              &BTRFS_I(inode)->io_tree,
 646                                              async_extent->start,
 647                                              async_extent->start +
 648                                              async_extent->ram_size - 1,
 649                                              NULL, 1, 1, 0, 1, 1, 0);
 650
 651                 ret = btrfs_submit_compressed_write(inode,
 652                                     async_extent->start,
 653                                     async_extent->ram_size,
 654                                     ins.objectid,
 655                                     ins.offset, async_extent->pages,
 656                                     async_extent->nr_pages);
 657
 658                 BUG_ON(ret);
 659                 trans = btrfs_join_transaction(root, 1);
 660                 alloc_hint = ins.objectid + ins.offset;
 661                 kfree(async_extent);
 662                 cond_resched();
 663         }
 664
 665         btrfs_end_transaction(trans, root);
 666         return 0;
 667 }
 668
 669 /*
 670  * when extent_io.c finds a delayed allocation range in the file,
 671  * the call backs end up in this code.  The basic idea is to
 672  * allocate extents on disk for the range, and create ordered data structs
 673  * in ram to track those extents.
 674  *
 675  * locked_page is the page that writepage had locked already.  We use
 676  * it to make sure we don't do extra locks or unlocks.
 677  *
 678  * *page_started is set to one if we unlock locked_page and do everything
 679  * required to start IO on it.  It may be clean and already done with
 680  * IO when we return.
 681  */
 682 static noinline int cow_file_range(struct inode *inode,
 683                                    struct page *locked_page,
 684                                    u64 start, u64 end, int *page_started,
 685                                    unsigned long *nr_written,
 686                                    int unlock)
 687 {
 688         struct btrfs_root *root = BTRFS_I(inode)->root;
 689         struct btrfs_trans_handle *trans;
 690         u64 alloc_hint = 0;
 691         u64 num_bytes;
 692         unsigned long ram_size;
 693         u64 disk_num_bytes;
 694         u64 cur_alloc_size;
 695         u64 blocksize = root->sectorsize;
 696         u64 actual_end;
 697         u64 isize = i_size_read(inode);
 698         struct btrfs_key ins;
 699         struct extent_map *em;
 700         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 701         int ret = 0;
 702
 703         trans = btrfs_join_transaction(root, 1);
 704         BUG_ON(!trans);
 705         btrfs_set_trans_block_group(trans, inode);
 706
 707         actual_end = min_t(u64, isize, end + 1);
 708
 709         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 710         num_bytes = max(blocksize,  num_bytes);
 711         disk_num_bytes = num_bytes;
 712         ret = 0;
 713
 714         if (start == 0) {
 715                 /* lets try to make an inline extent */
 716                 ret = cow_file_range_inline(trans, root, inode,
 717                                             start, end, 0, NULL);
 718                 if (ret == 0) {
 719                         extent_clear_unlock_delalloc(inode,
 720                                                      &BTRFS_I(inode)->io_tree,
 721                                                      start, end, NULL, 1, 1,
 722                                                      1, 1, 1, 1);
 723                         *nr_written = *nr_written +
 724                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 725                         *page_started = 1;
 726                         ret = 0;
 727                         goto out;
 728                 }
 729         }
 730
 731         BUG_ON(disk_num_bytes >
 732                btrfs_super_total_bytes(&root->fs_info->super_copy));
 733
 734         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 735
 736         while (disk_num_bytes > 0) {
 737                 cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent);
 738                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 739                                            root->sectorsize, 0, alloc_hint,
 740                                            (u64)-1, &ins, 1);
 741                 BUG_ON(ret);
 742
 743                 em = alloc_extent_map(GFP_NOFS);
 744                 em->start = start;
 745                 em->orig_start = em->start;
 746
 747                 ram_size = ins.offset;
 748                 em->len = ins.offset;
 749
 750                 em->block_start = ins.objectid;
 751                 em->block_len = ins.offset;
 752                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 753                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 754
 755                 while (1) {
 756                         spin_lock(&em_tree->lock);
 757                         ret = add_extent_mapping(em_tree, em);
 758                         spin_unlock(&em_tree->lock);
 759                         if (ret != -EEXIST) {
 760                                 free_extent_map(em);
 761                                 break;
 762                         }
 763                         btrfs_drop_extent_cache(inode, start,
 764                                                 start + ram_size - 1, 0);
 765                 }
 766
 767                 cur_alloc_size = ins.offset;
 768                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 769                                                ram_size, cur_alloc_size, 0);
 770                 BUG_ON(ret);
 771
 772                 if (root->root_key.objectid ==
 773                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 774                         ret = btrfs_reloc_clone_csums(inode, start,
 775                                                       cur_alloc_size);
 776                         BUG_ON(ret);
 777                 }
 778
 779                 if (disk_num_bytes < cur_alloc_size)
 780                         break;
 781
 782                 /* we're not doing compressed IO, don't unlock the first
 783                  * page (which the caller expects to stay locked), don't
 784                  * clear any dirty bits and don't set any writeback bits
 785                  */
 786                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 787                                              start, start + ram_size - 1,
 788                                              locked_page, unlock, 1,
 789                                              1, 0, 0, 0);
 790                 disk_num_bytes -= cur_alloc_size;
 791                 num_bytes -= cur_alloc_size;
 792                 alloc_hint = ins.objectid + ins.offset;
 793                 start += cur_alloc_size;
 794         }
 795 out:
 796         ret = 0;
 797         btrfs_end_transaction(trans, root);
 798
 799         return ret;
 800 }
 801
 802 /*
 803  * work queue call back to started compression on a file and pages
 804  */
 805 static noinline void async_cow_start(struct btrfs_work *work)
 806 {
 807         struct async_cow *async_cow;
 808         int num_added = 0;
 809         async_cow = container_of(work, struct async_cow, work);
 810
 811         compress_file_range(async_cow->inode, async_cow->locked_page,
 812                             async_cow->start, async_cow->end, async_cow,
 813                             &num_added);
 814         if (num_added == 0)
 815                 async_cow->inode = NULL;
 816 }
 817
 818 /*
 819  * work queue call back to submit previously compressed pages
 820  */
 821 static noinline void async_cow_submit(struct btrfs_work *work)
 822 {
 823         struct async_cow *async_cow;
 824         struct btrfs_root *root;
 825         unsigned long nr_pages;
 826
 827         async_cow = container_of(work, struct async_cow, work);
 828
 829         root = async_cow->root;
 830         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
 831                 PAGE_CACHE_SHIFT;
 832
 833         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
 834
 835         if (atomic_read(&root->fs_info->async_delalloc_pages) <
 836             5 * 1042 * 1024 &&
 837             waitqueue_active(&root->fs_info->async_submit_wait))
 838                 wake_up(&root->fs_info->async_submit_wait);
 839
 840         if (async_cow->inode)
 841                 submit_compressed_extents(async_cow->inode, async_cow);
 842 }
 843
 844 static noinline void async_cow_free(struct btrfs_work *work)
 845 {
 846         struct async_cow *async_cow;
 847         async_cow = container_of(work, struct async_cow, work);
 848         kfree(async_cow);
 849 }
 850
 851 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
 852                                 u64 start, u64 end, int *page_started,
 853                                 unsigned long *nr_written)
 854 {
 855         struct async_cow *async_cow;
 856         struct btrfs_root *root = BTRFS_I(inode)->root;
 857         unsigned long nr_pages;
 858         u64 cur_end;
 859         int limit = 10 * 1024 * 1042;
 860
 861         if (!btrfs_test_opt(root, COMPRESS)) {
 862                 return cow_file_range(inode, locked_page, start, end,
 863                                       page_started, nr_written, 1);
 864         }
 865
 866         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED |
 867                          EXTENT_DELALLOC, 1, 0, GFP_NOFS);
 868         while (start < end) {
 869                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
 870                 async_cow->inode = inode;
 871                 async_cow->root = root;
 872                 async_cow->locked_page = locked_page;
 873                 async_cow->start = start;
 874
 875                 if (btrfs_test_flag(inode, NOCOMPRESS))
 876                         cur_end = end;
 877                 else
 878                         cur_end = min(end, start + 512 * 1024 - 1);
 879
 880                 async_cow->end = cur_end;
 881                 INIT_LIST_HEAD(&async_cow->extents);
 882
 883                 async_cow->work.func = async_cow_start;
 884                 async_cow->work.ordered_func = async_cow_submit;
 885                 async_cow->work.ordered_free = async_cow_free;
 886                 async_cow->work.flags = 0;
 887
 888                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
 889                         PAGE_CACHE_SHIFT;
 890                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
 891
 892                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
 893                                    &async_cow->work);
 894
 895                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
 896                         wait_event(root->fs_info->async_submit_wait,
 897                            (atomic_read(&root->fs_info->async_delalloc_pages) <
 898                             limit));
 899                 }
 900
 901                 while (atomic_read(&root->fs_info->async_submit_draining) &&
 902                       atomic_read(&root->fs_info->async_delalloc_pages)) {
 903                         wait_event(root->fs_info->async_submit_wait,
 904                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
 905                            0));
 906                 }
 907
 908                 *nr_written += nr_pages;
 909                 start = cur_end + 1;
 910         }
 911         *page_started = 1;
 912         return 0;
 913 }
 914
 915 static noinline int csum_exist_in_range(struct btrfs_root *root,
 916                                         u64 bytenr, u64 num_bytes)
 917 {
 918         int ret;
 919         struct btrfs_ordered_sum *sums;
 920         LIST_HEAD(list);
 921
 922         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
 923                                        bytenr + num_bytes - 1, &list);
 924         if (ret == 0 && list_empty(&list))
 925                 return 0;
 926
 927         while (!list_empty(&list)) {
 928                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
 929                 list_del(&sums->list);
 930                 kfree(sums);
 931         }
 932         return 1;
 933 }
 934
 935 /*
 936  * when nowcow writeback call back.  This checks for snapshots or COW copies
 937  * of the extents that exist in the file, and COWs the file as required.
 938  *
 939  * If no cow copies or snapshots exist, we write directly to the existing
 940  * blocks on disk
 941  */
 942 static int run_delalloc_nocow(struct inode *inode, struct page *locked_page,
 943                               u64 start, u64 end, int *page_started, int force,
 944                               unsigned long *nr_written)
 945 {
 946         struct btrfs_root *root = BTRFS_I(inode)->root;
 947         struct btrfs_trans_handle *trans;
 948         struct extent_buffer *leaf;
 949         struct btrfs_path *path;
 950         struct btrfs_file_extent_item *fi;
 951         struct btrfs_key found_key;
 952         u64 cow_start;
 953         u64 cur_offset;
 954         u64 extent_end;
 955         u64 disk_bytenr;
 956         u64 num_bytes;
 957         int extent_type;
 958         int ret;
 959         int type;
 960         int nocow;
 961         int check_prev = 1;
 962
 963         path = btrfs_alloc_path();
 964         BUG_ON(!path);
 965         trans = btrfs_join_transaction(root, 1);
 966         BUG_ON(!trans);
 967
 968         cow_start = (u64)-1;
 969         cur_offset = start;
 970         while (1) {
 971                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
 972                                                cur_offset, 0);
 973                 BUG_ON(ret < 0);
 974                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
 975                         leaf = path->nodes[0];
 976                         btrfs_item_key_to_cpu(leaf, &found_key,
 977                                               path->slots[0] - 1);
 978                         if (found_key.objectid == inode->i_ino &&
 979                             found_key.type == BTRFS_EXTENT_DATA_KEY)
 980                                 path->slots[0]--;
 981                 }
 982                 check_prev = 0;
 983 next_slot:
 984                 leaf = path->nodes[0];
 985                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 986                         ret = btrfs_next_leaf(root, path);
 987                         if (ret < 0)
 988                                 BUG_ON(1);
 989                         if (ret > 0)
 990                                 break;
 991                         leaf = path->nodes[0];
 992                 }
 993
 994                 nocow = 0;
 995                 disk_bytenr = 0;
 996                 num_bytes = 0;
 997                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 998
 999                 if (found_key.objectid > inode->i_ino ||
1000                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1001                     found_key.offset > end)
1002                         break;
1003
1004                 if (found_key.offset > cur_offset) {
1005                         extent_end = found_key.offset;
1006                         goto out_check;
1007                 }
1008
1009                 fi = btrfs_item_ptr(leaf, path->slots[0],
1010                                     struct btrfs_file_extent_item);
1011                 extent_type = btrfs_file_extent_type(leaf, fi);
1012
1013                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1014                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1015                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1016                         extent_end = found_key.offset +
1017                                 btrfs_file_extent_num_bytes(leaf, fi);
1018                         if (extent_end <= start) {
1019                                 path->slots[0]++;
1020                                 goto next_slot;
1021                         }
1022                         if (disk_bytenr == 0)
1023                                 goto out_check;
1024                         if (btrfs_file_extent_compression(leaf, fi) ||
1025                             btrfs_file_extent_encryption(leaf, fi) ||
1026                             btrfs_file_extent_other_encoding(leaf, fi))
1027                                 goto out_check;
1028                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1029                                 goto out_check;
1030                         if (btrfs_extent_readonly(root, disk_bytenr))
1031                                 goto out_check;
1032                         if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1033                                                   disk_bytenr))
1034                                 goto out_check;
1035                         disk_bytenr += btrfs_file_extent_offset(leaf, fi);
1036                         disk_bytenr += cur_offset - found_key.offset;
1037                         num_bytes = min(end + 1, extent_end) - cur_offset;
1038                         /*
1039                          * force cow if csum exists in the range.
1040                          * this ensure that csum for a given extent are
1041                          * either valid or do not exist.
1042                          */
1043                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1044                                 goto out_check;
1045                         nocow = 1;
1046                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1047                         extent_end = found_key.offset +
1048                                 btrfs_file_extent_inline_len(leaf, fi);
1049                         extent_end = ALIGN(extent_end, root->sectorsize);
1050                 } else {
1051                         BUG_ON(1);
1052                 }
1053 out_check:
1054                 if (extent_end <= start) {
1055                         path->slots[0]++;
1056                         goto next_slot;
1057                 }
1058                 if (!nocow) {
1059                         if (cow_start == (u64)-1)
1060                                 cow_start = cur_offset;
1061                         cur_offset = extent_end;
1062                         if (cur_offset > end)
1063                                 break;
1064                         path->slots[0]++;
1065                         goto next_slot;
1066                 }
1067
1068                 btrfs_release_path(root, path);
1069                 if (cow_start != (u64)-1) {
1070                         ret = cow_file_range(inode, locked_page, cow_start,
1071                                         found_key.offset - 1, page_started,
1072                                         nr_written, 1);
1073                         BUG_ON(ret);
1074                         cow_start = (u64)-1;
1075                 }
1076
1077                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1078                         struct extent_map *em;
1079                         struct extent_map_tree *em_tree;
1080                         em_tree = &BTRFS_I(inode)->extent_tree;
1081                         em = alloc_extent_map(GFP_NOFS);
1082                         em->start = cur_offset;
1083                         em->orig_start = em->start;
1084                         em->len = num_bytes;
1085                         em->block_len = num_bytes;
1086                         em->block_start = disk_bytenr;
1087                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1088                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1089                         while (1) {
1090                                 spin_lock(&em_tree->lock);
1091                                 ret = add_extent_mapping(em_tree, em);
1092                                 spin_unlock(&em_tree->lock);
1093                                 if (ret != -EEXIST) {
1094                                         free_extent_map(em);
1095                                         break;
1096                                 }
1097                                 btrfs_drop_extent_cache(inode, em->start,
1098                                                 em->start + em->len - 1, 0);
1099                         }
1100                         type = BTRFS_ORDERED_PREALLOC;
1101                 } else {
1102                         type = BTRFS_ORDERED_NOCOW;
1103                 }
1104
1105                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1106                                                num_bytes, num_bytes, type);
1107                 BUG_ON(ret);
1108
1109                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1110                                         cur_offset, cur_offset + num_bytes - 1,
1111                                         locked_page, 1, 1, 1, 0, 0, 0);
1112                 cur_offset = extent_end;
1113                 if (cur_offset > end)
1114                         break;
1115         }
1116         btrfs_release_path(root, path);
1117
1118         if (cur_offset <= end && cow_start == (u64)-1)
1119                 cow_start = cur_offset;
1120         if (cow_start != (u64)-1) {
1121                 ret = cow_file_range(inode, locked_page, cow_start, end,
1122                                      page_started, nr_written, 1);
1123                 BUG_ON(ret);
1124         }
1125
1126         ret = btrfs_end_transaction(trans, root);
1127         BUG_ON(ret);
1128         btrfs_free_path(path);
1129         return 0;
1130 }
1131
1132 /*
1133  * extent_io.c call back to do delayed allocation processing
1134  */
1135 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1136                               u64 start, u64 end, int *page_started,
1137                               unsigned long *nr_written)
1138 {
1139         int ret;
1140
1141         if (btrfs_test_flag(inode, NODATACOW))
1142                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1143                                          page_started, 1, nr_written);
1144         else if (btrfs_test_flag(inode, PREALLOC))
1145                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1146                                          page_started, 0, nr_written);
1147         else
1148                 ret = cow_file_range_async(inode, locked_page, start, end,
1149                                            page_started, nr_written);
1150
1151         return ret;
1152 }
1153
1154 /*
1155  * extent_io.c set_bit_hook, used to track delayed allocation
1156  * bytes in this file, and to maintain the list of inodes that
1157  * have pending delalloc work to be done.
1158  */
1159 static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
1160                        unsigned long old, unsigned long bits)
1161 {
1162         /*
1163          * set_bit and clear bit hooks normally require _irqsave/restore
1164          * but in this case, we are only testeing for the DELALLOC
1165          * bit, which is only set or cleared with irqs on
1166          */
1167         if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1168                 struct btrfs_root *root = BTRFS_I(inode)->root;
1169                 spin_lock(&root->fs_info->delalloc_lock);
1170                 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
1171                 root->fs_info->delalloc_bytes += end - start + 1;
1172                 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1173                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1174                                       &root->fs_info->delalloc_inodes);
1175                 }
1176                 spin_unlock(&root->fs_info->delalloc_lock);
1177         }
1178         return 0;
1179 }
1180
1181 /*
1182  * extent_io.c clear_bit_hook, see set_bit_hook for why
1183  */
1184 static int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
1185                          unsigned long old, unsigned long bits)
1186 {
1187         /*
1188          * set_bit and clear bit hooks normally require _irqsave/restore
1189          * but in this case, we are only testeing for the DELALLOC
1190          * bit, which is only set or cleared with irqs on
1191          */
1192         if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1193                 struct btrfs_root *root = BTRFS_I(inode)->root;
1194
1195                 spin_lock(&root->fs_info->delalloc_lock);
1196                 if (end - start + 1 > root->fs_info->delalloc_bytes) {
1197                         printk(KERN_INFO "btrfs warning: delalloc account "
1198                                "%llu %llu\n",
1199                                (unsigned long long)end - start + 1,
1200                                (unsigned long long)
1201                                root->fs_info->delalloc_bytes);
1202                         root->fs_info->delalloc_bytes = 0;
1203                         BTRFS_I(inode)->delalloc_bytes = 0;
1204                 } else {
1205                         root->fs_info->delalloc_bytes -= end - start + 1;
1206                         BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
1207                 }
1208                 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
1209                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1210                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1211                 }
1212                 spin_unlock(&root->fs_info->delalloc_lock);
1213         }
1214         return 0;
1215 }
1216
1217 /*
1218  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1219  * we don't create bios that span stripes or chunks
1220  */
1221 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1222                          size_t size, struct bio *bio,
1223                          unsigned long bio_flags)
1224 {
1225         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1226         struct btrfs_mapping_tree *map_tree;
1227         u64 logical = (u64)bio->bi_sector << 9;
1228         u64 length = 0;
1229         u64 map_length;
1230         int ret;
1231
1232         if (bio_flags & EXTENT_BIO_COMPRESSED)
1233                 return 0;
1234
1235         length = bio->bi_size;
1236         map_tree = &root->fs_info->mapping_tree;
1237         map_length = length;
1238         ret = btrfs_map_block(map_tree, READ, logical,
1239                               &map_length, NULL, 0);
1240
1241         if (map_length < length + size)
1242                 return 1;
1243         return 0;
1244 }
1245
1246 /*
1247  * in order to insert checksums into the metadata in large chunks,
1248  * we wait until bio submission time.   All the pages in the bio are
1249  * checksummed and sums are attached onto the ordered extent record.
1250  *
1251  * At IO completion time the cums attached on the ordered extent record
1252  * are inserted into the btree
1253  */
1254 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1255                                     struct bio *bio, int mirror_num,
1256                                     unsigned long bio_flags)
1257 {
1258         struct btrfs_root *root = BTRFS_I(inode)->root;
1259         int ret = 0;
1260
1261         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1262         BUG_ON(ret);
1263         return 0;
1264 }
1265
1266 /*
1267  * in order to insert checksums into the metadata in large chunks,
1268  * we wait until bio submission time.   All the pages in the bio are
1269  * checksummed and sums are attached onto the ordered extent record.
1270  *
1271  * At IO completion time the cums attached on the ordered extent record
1272  * are inserted into the btree
1273  */
1274 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1275                           int mirror_num, unsigned long bio_flags)
1276 {
1277         struct btrfs_root *root = BTRFS_I(inode)->root;
1278         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1279 }
1280
1281 /*
1282  * extent_io.c submission hook. This does the right thing for csum calculation
1283  * on write, or reading the csums from the tree before a read
1284  */
1285 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1286                           int mirror_num, unsigned long bio_flags)
1287 {
1288         struct btrfs_root *root = BTRFS_I(inode)->root;
1289         int ret = 0;
1290         int skip_sum;
1291
1292         skip_sum = btrfs_test_flag(inode, NODATASUM);
1293
1294         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1295         BUG_ON(ret);
1296
1297         if (!(rw & (1 << BIO_RW))) {
1298                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1299                         return btrfs_submit_compressed_read(inode, bio,
1300                                                     mirror_num, bio_flags);
1301                 } else if (!skip_sum)
1302                         btrfs_lookup_bio_sums(root, inode, bio, NULL);
1303                 goto mapit;
1304         } else if (!skip_sum) {
1305                 /* csum items have already been cloned */
1306                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1307                         goto mapit;
1308                 /* we're doing a write, do the async checksumming */
1309                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1310                                    inode, rw, bio, mirror_num,
1311                                    bio_flags, __btrfs_submit_bio_start,
1312                                    __btrfs_submit_bio_done);
1313         }
1314
1315 mapit:
1316         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1317 }
1318
1319 /*
1320  * given a list of ordered sums record them in the inode.  This happens
1321  * at IO completion time based on sums calculated at bio submission time.
1322  */
1323 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1324                              struct inode *inode, u64 file_offset,
1325                              struct list_head *list)
1326 {
1327         struct list_head *cur;
1328         struct btrfs_ordered_sum *sum;
1329
1330         btrfs_set_trans_block_group(trans, inode);
1331         list_for_each(cur, list) {
1332                 sum = list_entry(cur, struct btrfs_ordered_sum, list);
1333                 btrfs_csum_file_blocks(trans,
1334                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1335         }
1336         return 0;
1337 }
1338
1339 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end)
1340 {
1341         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1342                 WARN_ON(1);
1343         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1344                                    GFP_NOFS);
1345 }
1346
1347 /* see btrfs_writepage_start_hook for details on why this is required */
1348 struct btrfs_writepage_fixup {
1349         struct page *page;
1350         struct btrfs_work work;
1351 };
1352
1353 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1354 {
1355         struct btrfs_writepage_fixup *fixup;
1356         struct btrfs_ordered_extent *ordered;
1357         struct page *page;
1358         struct inode *inode;
1359         u64 page_start;
1360         u64 page_end;
1361
1362         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1363         page = fixup->page;
1364 again:
1365         lock_page(page);
1366         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1367                 ClearPageChecked(page);
1368                 goto out_page;
1369         }
1370
1371         inode = page->mapping->host;
1372         page_start = page_offset(page);
1373         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1374
1375         lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1376
1377         /* already ordered? We're done */
1378         if (test_range_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
1379                              EXTENT_ORDERED, 0)) {
1380                 goto out;
1381         }
1382
1383         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1384         if (ordered) {
1385                 unlock_extent(&BTRFS_I(inode)->io_tree, page_start,
1386                               page_end, GFP_NOFS);
1387                 unlock_page(page);
1388                 btrfs_start_ordered_extent(inode, ordered, 1);
1389                 goto again;
1390         }
1391
1392         btrfs_set_extent_delalloc(inode, page_start, page_end);
1393         ClearPageChecked(page);
1394 out:
1395         unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1396 out_page:
1397         unlock_page(page);
1398         page_cache_release(page);
1399 }
1400
1401 /*
1402  * There are a few paths in the higher layers of the kernel that directly
1403  * set the page dirty bit without asking the filesystem if it is a
1404  * good idea.  This causes problems because we want to make sure COW
1405  * properly happens and the data=ordered rules are followed.
1406  *
1407  * In our case any range that doesn't have the ORDERED bit set
1408  * hasn't been properly setup for IO.  We kick off an async process
1409  * to fix it up.  The async helper will wait for ordered extents, set
1410  * the delalloc bit and make it safe to write the page.
1411  */
1412 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1413 {
1414         struct inode *inode = page->mapping->host;
1415         struct btrfs_writepage_fixup *fixup;
1416         struct btrfs_root *root = BTRFS_I(inode)->root;
1417         int ret;
1418
1419         ret = test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1420                              EXTENT_ORDERED, 0);
1421         if (ret)
1422                 return 0;
1423
1424         if (PageChecked(page))
1425                 return -EAGAIN;
1426
1427         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1428         if (!fixup)
1429                 return -EAGAIN;
1430
1431         SetPageChecked(page);
1432         page_cache_get(page);
1433         fixup->work.func = btrfs_writepage_fixup_worker;
1434         fixup->page = page;
1435         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1436         return -EAGAIN;
1437 }
1438
1439 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1440                                        struct inode *inode, u64 file_pos,
1441                                        u64 disk_bytenr, u64 disk_num_bytes,
1442                                        u64 num_bytes, u64 ram_bytes,
1443                                        u8 compression, u8 encryption,
1444                                        u16 other_encoding, int extent_type)
1445 {
1446         struct btrfs_root *root = BTRFS_I(inode)->root;
1447         struct btrfs_file_extent_item *fi;
1448         struct btrfs_path *path;
1449         struct extent_buffer *leaf;
1450         struct btrfs_key ins;
1451         u64 hint;
1452         int ret;
1453
1454         path = btrfs_alloc_path();
1455         BUG_ON(!path);
1456
1457         ret = btrfs_drop_extents(trans, root, inode, file_pos,
1458                                  file_pos + num_bytes, file_pos, &hint);
1459         BUG_ON(ret);
1460
1461         ins.objectid = inode->i_ino;
1462         ins.offset = file_pos;
1463         ins.type = BTRFS_EXTENT_DATA_KEY;
1464         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1465         BUG_ON(ret);
1466         leaf = path->nodes[0];
1467         fi = btrfs_item_ptr(leaf, path->slots[0],
1468                             struct btrfs_file_extent_item);
1469         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1470         btrfs_set_file_extent_type(leaf, fi, extent_type);
1471         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1472         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1473         btrfs_set_file_extent_offset(leaf, fi, 0);
1474         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1475         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1476         btrfs_set_file_extent_compression(leaf, fi, compression);
1477         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1478         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1479         btrfs_mark_buffer_dirty(leaf);
1480
1481         inode_add_bytes(inode, num_bytes);
1482         btrfs_drop_extent_cache(inode, file_pos, file_pos + num_bytes - 1, 0);
1483
1484         ins.objectid = disk_bytenr;
1485         ins.offset = disk_num_bytes;
1486         ins.type = BTRFS_EXTENT_ITEM_KEY;
1487         ret = btrfs_alloc_reserved_extent(trans, root, leaf->start,
1488                                           root->root_key.objectid,
1489                                           trans->transid, inode->i_ino, &ins);
1490         BUG_ON(ret);
1491
1492         btrfs_free_path(path);
1493         return 0;
1494 }
1495
1496 /* as ordered data IO finishes, this gets called so we can finish
1497  * an ordered extent if the range of bytes in the file it covers are
1498  * fully written.
1499  */
1500 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1501 {
1502         struct btrfs_root *root = BTRFS_I(inode)->root;
1503         struct btrfs_trans_handle *trans;
1504         struct btrfs_ordered_extent *ordered_extent;
1505         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1506         int compressed = 0;
1507         int ret;
1508
1509         ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
1510         if (!ret)
1511                 return 0;
1512
1513         trans = btrfs_join_transaction(root, 1);
1514
1515         ordered_extent = btrfs_lookup_ordered_extent(inode, start);
1516         BUG_ON(!ordered_extent);
1517         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
1518                 goto nocow;
1519
1520         lock_extent(io_tree, ordered_extent->file_offset,
1521                     ordered_extent->file_offset + ordered_extent->len - 1,
1522                     GFP_NOFS);
1523
1524         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1525                 compressed = 1;
1526         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1527                 BUG_ON(compressed);
1528                 ret = btrfs_mark_extent_written(trans, root, inode,
1529                                                 ordered_extent->file_offset,
1530                                                 ordered_extent->file_offset +
1531                                                 ordered_extent->len);
1532                 BUG_ON(ret);
1533         } else {
1534                 ret = insert_reserved_file_extent(trans, inode,
1535                                                 ordered_extent->file_offset,
1536                                                 ordered_extent->start,
1537                                                 ordered_extent->disk_len,
1538                                                 ordered_extent->len,
1539                                                 ordered_extent->len,
1540                                                 compressed, 0, 0,
1541                                                 BTRFS_FILE_EXTENT_REG);
1542                 BUG_ON(ret);
1543         }
1544         unlock_extent(io_tree, ordered_extent->file_offset,
1545                     ordered_extent->file_offset + ordered_extent->len - 1,
1546                     GFP_NOFS);
1547 nocow:
1548         add_pending_csums(trans, inode, ordered_extent->file_offset,
1549                           &ordered_extent->list);
1550
1551         mutex_lock(&BTRFS_I(inode)->extent_mutex);
1552         btrfs_ordered_update_i_size(inode, ordered_extent);
1553         btrfs_update_inode(trans, root, inode);
1554         btrfs_remove_ordered_extent(inode, ordered_extent);
1555         mutex_unlock(&BTRFS_I(inode)->extent_mutex);
1556
1557         /* once for us */
1558         btrfs_put_ordered_extent(ordered_extent);
1559         /* once for the tree */
1560         btrfs_put_ordered_extent(ordered_extent);
1561
1562         btrfs_end_transaction(trans, root);
1563         return 0;
1564 }
1565
1566 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1567                                 struct extent_state *state, int uptodate)
1568 {
1569         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1570 }
1571
1572 /*
1573  * When IO fails, either with EIO or csum verification fails, we
1574  * try other mirrors that might have a good copy of the data.  This
1575  * io_failure_record is used to record state as we go through all the
1576  * mirrors.  If another mirror has good data, the page is set up to date
1577  * and things continue.  If a good mirror can't be found, the original
1578  * bio end_io callback is called to indicate things have failed.
1579  */
1580 struct io_failure_record {
1581         struct page *page;
1582         u64 start;
1583         u64 len;
1584         u64 logical;
1585         unsigned long bio_flags;
1586         int last_mirror;
1587 };
1588
1589 static int btrfs_io_failed_hook(struct bio *failed_bio,
1590                          struct page *page, u64 start, u64 end,
1591                          struct extent_state *state)
1592 {
1593         struct io_failure_record *failrec = NULL;
1594         u64 private;
1595         struct extent_map *em;
1596         struct inode *inode = page->mapping->host;
1597         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1598         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1599         struct bio *bio;
1600         int num_copies;
1601         int ret;
1602         int rw;
1603         u64 logical;
1604
1605         ret = get_state_private(failure_tree, start, &private);
1606         if (ret) {
1607                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1608                 if (!failrec)
1609                         return -ENOMEM;
1610                 failrec->start = start;
1611                 failrec->len = end - start + 1;
1612                 failrec->last_mirror = 0;
1613                 failrec->bio_flags = 0;
1614
1615                 spin_lock(&em_tree->lock);
1616                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1617                 if (em->start > start || em->start + em->len < start) {
1618                         free_extent_map(em);
1619                         em = NULL;
1620                 }
1621                 spin_unlock(&em_tree->lock);
1622
1623                 if (!em || IS_ERR(em)) {
1624                         kfree(failrec);
1625                         return -EIO;
1626                 }
1627                 logical = start - em->start;
1628                 logical = em->block_start + logical;
1629                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1630                         logical = em->block_start;
1631                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1632                 }
1633                 failrec->logical = logical;
1634                 free_extent_map(em);
1635                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1636                                 EXTENT_DIRTY, GFP_NOFS);
1637                 set_state_private(failure_tree, start,
1638                                  (u64)(unsigned long)failrec);
1639         } else {
1640                 failrec = (struct io_failure_record *)(unsigned long)private;
1641         }
1642         num_copies = btrfs_num_copies(
1643                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1644                               failrec->logical, failrec->len);
1645         failrec->last_mirror++;
1646         if (!state) {
1647                 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1648                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1649                                                     failrec->start,
1650                                                     EXTENT_LOCKED);
1651                 if (state && state->start != failrec->start)
1652                         state = NULL;
1653                 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1654         }
1655         if (!state || failrec->last_mirror > num_copies) {
1656                 set_state_private(failure_tree, failrec->start, 0);
1657                 clear_extent_bits(failure_tree, failrec->start,
1658                                   failrec->start + failrec->len - 1,
1659                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1660                 kfree(failrec);
1661                 return -EIO;
1662         }
1663         bio = bio_alloc(GFP_NOFS, 1);
1664         bio->bi_private = state;
1665         bio->bi_end_io = failed_bio->bi_end_io;
1666         bio->bi_sector = failrec->logical >> 9;
1667         bio->bi_bdev = failed_bio->bi_bdev;
1668         bio->bi_size = 0;
1669
1670         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1671         if (failed_bio->bi_rw & (1 << BIO_RW))
1672                 rw = WRITE;
1673         else
1674                 rw = READ;
1675
1676         BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1677                                                       failrec->last_mirror,
1678                                                       failrec->bio_flags);
1679         return 0;
1680 }
1681
1682 /*
1683  * each time an IO finishes, we do a fast check in the IO failure tree
1684  * to see if we need to process or clean up an io_failure_record
1685  */
1686 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1687 {
1688         u64 private;
1689         u64 private_failure;
1690         struct io_failure_record *failure;
1691         int ret;
1692
1693         private = 0;
1694         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1695                              (u64)-1, 1, EXTENT_DIRTY)) {
1696                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1697                                         start, &private_failure);
1698                 if (ret == 0) {
1699                         failure = (struct io_failure_record *)(unsigned long)
1700                                    private_failure;
1701                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1702                                           failure->start, 0);
1703                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1704                                           failure->start,
1705                                           failure->start + failure->len - 1,
1706                                           EXTENT_DIRTY | EXTENT_LOCKED,
1707                                           GFP_NOFS);
1708                         kfree(failure);
1709                 }
1710         }
1711         return 0;
1712 }
1713
1714 /*
1715  * when reads are done, we need to check csums to verify the data is correct
1716  * if there's a match, we allow the bio to finish.  If not, we go through
1717  * the io_failure_record routines to find good copies
1718  */
1719 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1720                                struct extent_state *state)
1721 {
1722         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1723         struct inode *inode = page->mapping->host;
1724         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1725         char *kaddr;
1726         u64 private = ~(u32)0;
1727         int ret;
1728         struct btrfs_root *root = BTRFS_I(inode)->root;
1729         u32 csum = ~(u32)0;
1730         unsigned long flags;
1731
1732         if (PageChecked(page)) {
1733                 ClearPageChecked(page);
1734                 goto good;
1735         }
1736         if (btrfs_test_flag(inode, NODATASUM))
1737                 return 0;
1738
1739         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1740             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1)) {
1741                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1742                                   GFP_NOFS);
1743                 return 0;
1744         }
1745
1746         if (state && state->start == start) {
1747                 private = state->private;
1748                 ret = 0;
1749         } else {
1750                 ret = get_state_private(io_tree, start, &private);
1751         }
1752         local_irq_save(flags);
1753         kaddr = kmap_atomic(page, KM_IRQ0);
1754         if (ret)
1755                 goto zeroit;
1756
1757         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
1758         btrfs_csum_final(csum, (char *)&csum);
1759         if (csum != private)
1760                 goto zeroit;
1761
1762         kunmap_atomic(kaddr, KM_IRQ0);
1763         local_irq_restore(flags);
1764 good:
1765         /* if the io failure tree for this inode is non-empty,
1766          * check to see if we've recovered from a failed IO
1767          */
1768         btrfs_clean_io_failures(inode, start);
1769         return 0;
1770
1771 zeroit:
1772         printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
1773                "private %llu\n", page->mapping->host->i_ino,
1774                (unsigned long long)start, csum,
1775                (unsigned long long)private);
1776         memset(kaddr + offset, 1, end - start + 1);
1777         flush_dcache_page(page);
1778         kunmap_atomic(kaddr, KM_IRQ0);
1779         local_irq_restore(flags);
1780         if (private == 0)
1781                 return 0;
1782         return -EIO;
1783 }
1784
1785 /*
1786  * This creates an orphan entry for the given inode in case something goes
1787  * wrong in the middle of an unlink/truncate.
1788  */
1789 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
1790 {
1791         struct btrfs_root *root = BTRFS_I(inode)->root;
1792         int ret = 0;
1793
1794         spin_lock(&root->list_lock);
1795
1796         /* already on the orphan list, we're good */
1797         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
1798                 spin_unlock(&root->list_lock);
1799                 return 0;
1800         }
1801
1802         list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1803
1804         spin_unlock(&root->list_lock);
1805
1806         /*
1807          * insert an orphan item to track this unlinked/truncated file
1808          */
1809         ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
1810
1811         return ret;
1812 }
1813
1814 /*
1815  * We have done the truncate/delete so we can go ahead and remove the orphan
1816  * item for this particular inode.
1817  */
1818 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
1819 {
1820         struct btrfs_root *root = BTRFS_I(inode)->root;
1821         int ret = 0;
1822
1823         spin_lock(&root->list_lock);
1824
1825         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
1826                 spin_unlock(&root->list_lock);
1827                 return 0;
1828         }
1829
1830         list_del_init(&BTRFS_I(inode)->i_orphan);
1831         if (!trans) {
1832                 spin_unlock(&root->list_lock);
1833                 return 0;
1834         }
1835
1836         spin_unlock(&root->list_lock);
1837
1838         ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
1839
1840         return ret;
1841 }
1842
1843 /*
1844  * this cleans up any orphans that may be left on the list from the last use
1845  * of this root.
1846  */
1847 void btrfs_orphan_cleanup(struct btrfs_root *root)
1848 {
1849         struct btrfs_path *path;
1850         struct extent_buffer *leaf;
1851         struct btrfs_item *item;
1852         struct btrfs_key key, found_key;
1853         struct btrfs_trans_handle *trans;
1854         struct inode *inode;
1855         int ret = 0, nr_unlink = 0, nr_truncate = 0;
1856
1857         path = btrfs_alloc_path();
1858         if (!path)
1859                 return;
1860         path->reada = -1;
1861
1862         key.objectid = BTRFS_ORPHAN_OBJECTID;
1863         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1864         key.offset = (u64)-1;
1865
1866
1867         while (1) {
1868                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1869                 if (ret < 0) {
1870                         printk(KERN_ERR "Error searching slot for orphan: %d"
1871                                "\n", ret);
1872                         break;
1873                 }
1874
1875                 /*
1876                  * if ret == 0 means we found what we were searching for, which
1877                  * is weird, but possible, so only screw with path if we didnt
1878                  * find the key and see if we have stuff that matches
1879                  */
1880                 if (ret > 0) {
1881                         if (path->slots[0] == 0)
1882                                 break;
1883                         path->slots[0]--;
1884                 }
1885
1886                 /* pull out the item */
1887                 leaf = path->nodes[0];
1888                 item = btrfs_item_nr(leaf, path->slots[0]);
1889                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1890
1891                 /* make sure the item matches what we want */
1892                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
1893                         break;
1894                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
1895                         break;
1896
1897                 /* release the path since we're done with it */
1898                 btrfs_release_path(root, path);
1899
1900                 /*
1901                  * this is where we are basically btrfs_lookup, without the
1902                  * crossing root thing.  we store the inode number in the
1903                  * offset of the orphan item.
1904                  */
1905                 inode = btrfs_iget_locked(root->fs_info->sb,
1906                                           found_key.offset, root);
1907                 if (!inode)
1908                         break;
1909
1910                 if (inode->i_state & I_NEW) {
1911                         BTRFS_I(inode)->root = root;
1912
1913                         /* have to set the location manually */
1914                         BTRFS_I(inode)->location.objectid = inode->i_ino;
1915                         BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
1916                         BTRFS_I(inode)->location.offset = 0;
1917
1918                         btrfs_read_locked_inode(inode);
1919                         unlock_new_inode(inode);
1920                 }
1921
1922                 /*
1923                  * add this inode to the orphan list so btrfs_orphan_del does
1924                  * the proper thing when we hit it
1925                  */
1926                 spin_lock(&root->list_lock);
1927                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1928                 spin_unlock(&root->list_lock);
1929
1930                 /*
1931                  * if this is a bad inode, means we actually succeeded in
1932                  * removing the inode, but not the orphan record, which means
1933                  * we need to manually delete the orphan since iput will just
1934                  * do a destroy_inode
1935                  */
1936                 if (is_bad_inode(inode)) {
1937                         trans = btrfs_start_transaction(root, 1);
1938                         btrfs_orphan_del(trans, inode);
1939                         btrfs_end_transaction(trans, root);
1940                         iput(inode);
1941                         continue;
1942                 }
1943
1944                 /* if we have links, this was a truncate, lets do that */
1945                 if (inode->i_nlink) {
1946                         nr_truncate++;
1947                         btrfs_truncate(inode);
1948                 } else {
1949                         nr_unlink++;
1950                 }
1951
1952                 /* this will do delete_inode and everything for us */
1953                 iput(inode);
1954         }
1955
1956         if (nr_unlink)
1957                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
1958         if (nr_truncate)
1959                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
1960
1961         btrfs_free_path(path);
1962 }
1963
1964 /*
1965  * read an inode from the btree into the in-memory inode
1966  */
1967 void btrfs_read_locked_inode(struct inode *inode)
1968 {
1969         struct btrfs_path *path;
1970         struct extent_buffer *leaf;
1971         struct btrfs_inode_item *inode_item;
1972         struct btrfs_timespec *tspec;
1973         struct btrfs_root *root = BTRFS_I(inode)->root;
1974         struct btrfs_key location;
1975         u64 alloc_group_block;
1976         u32 rdev;
1977         int ret;
1978
1979         path = btrfs_alloc_path();
1980         BUG_ON(!path);
1981         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
1982
1983         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
1984         if (ret)
1985                 goto make_bad;
1986
1987         leaf = path->nodes[0];
1988         inode_item = btrfs_item_ptr(leaf, path->slots[0],
1989                                     struct btrfs_inode_item);
1990
1991         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
1992         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
1993         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
1994         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
1995         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
1996
1997         tspec = btrfs_inode_atime(inode_item);
1998         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1999         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2000
2001         tspec = btrfs_inode_mtime(inode_item);
2002         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2003         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2004
2005         tspec = btrfs_inode_ctime(inode_item);
2006         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2007         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2008
2009         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2010         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2011         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2012         inode->i_generation = BTRFS_I(inode)->generation;
2013         inode->i_rdev = 0;
2014         rdev = btrfs_inode_rdev(leaf, inode_item);
2015
2016         BTRFS_I(inode)->index_cnt = (u64)-1;
2017         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2018
2019         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2020         BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2021                                                 alloc_group_block, 0);
2022         btrfs_free_path(path);
2023         inode_item = NULL;
2024
2025         switch (inode->i_mode & S_IFMT) {
2026         case S_IFREG:
2027                 inode->i_mapping->a_ops = &btrfs_aops;
2028                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2029                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2030                 inode->i_fop = &btrfs_file_operations;
2031                 inode->i_op = &btrfs_file_inode_operations;
2032                 break;
2033         case S_IFDIR:
2034                 inode->i_fop = &btrfs_dir_file_operations;
2035                 if (root == root->fs_info->tree_root)
2036                         inode->i_op = &btrfs_dir_ro_inode_operations;
2037                 else
2038                         inode->i_op = &btrfs_dir_inode_operations;
2039                 break;
2040         case S_IFLNK:
2041                 inode->i_op = &btrfs_symlink_inode_operations;
2042                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2043                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2044                 break;
2045         default:
2046                 init_special_inode(inode, inode->i_mode, rdev);
2047                 break;
2048         }
2049         return;
2050
2051 make_bad:
2052         btrfs_free_path(path);
2053         make_bad_inode(inode);
2054 }
2055
2056 /*
2057  * given a leaf and an inode, copy the inode fields into the leaf
2058  */
2059 static void fill_inode_item(struct btrfs_trans_handle *trans,
2060                             struct extent_buffer *leaf,
2061                             struct btrfs_inode_item *item,
2062                             struct inode *inode)
2063 {
2064         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2065         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2066         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2067         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2068         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2069
2070         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2071                                inode->i_atime.tv_sec);
2072         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2073                                 inode->i_atime.tv_nsec);
2074
2075         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2076                                inode->i_mtime.tv_sec);
2077         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2078                                 inode->i_mtime.tv_nsec);
2079
2080         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2081                                inode->i_ctime.tv_sec);
2082         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2083                                 inode->i_ctime.tv_nsec);
2084
2085         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2086         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2087         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2088         btrfs_set_inode_transid(leaf, item, trans->transid);
2089         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2090         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2091         btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2092 }
2093
2094 /*
2095  * copy everything in the in-memory inode into the btree.
2096  */
2097 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2098                                 struct btrfs_root *root, struct inode *inode)
2099 {
2100         struct btrfs_inode_item *inode_item;
2101         struct btrfs_path *path;
2102         struct extent_buffer *leaf;
2103         int ret;
2104
2105         path = btrfs_alloc_path();
2106         BUG_ON(!path);
2107         ret = btrfs_lookup_inode(trans, root, path,
2108                                  &BTRFS_I(inode)->location, 1);
2109         if (ret) {
2110                 if (ret > 0)
2111                         ret = -ENOENT;
2112                 goto failed;
2113         }
2114
2115         leaf = path->nodes[0];
2116         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2117                                   struct btrfs_inode_item);
2118
2119         fill_inode_item(trans, leaf, inode_item, inode);
2120         btrfs_mark_buffer_dirty(leaf);
2121         btrfs_set_inode_last_trans(trans, inode);
2122         ret = 0;
2123 failed:
2124         btrfs_free_path(path);
2125         return ret;
2126 }
2127
2128
2129 /*
2130  * unlink helper that gets used here in inode.c and in the tree logging
2131  * recovery code.  It remove a link in a directory with a given name, and
2132  * also drops the back refs in the inode to the directory
2133  */
2134 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2135                        struct btrfs_root *root,
2136                        struct inode *dir, struct inode *inode,
2137                        const char *name, int name_len)
2138 {
2139         struct btrfs_path *path;
2140         int ret = 0;
2141         struct extent_buffer *leaf;
2142         struct btrfs_dir_item *di;
2143         struct btrfs_key key;
2144         u64 index;
2145
2146         path = btrfs_alloc_path();
2147         if (!path) {
2148                 ret = -ENOMEM;
2149                 goto err;
2150         }
2151
2152         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2153                                     name, name_len, -1);
2154         if (IS_ERR(di)) {
2155                 ret = PTR_ERR(di);
2156                 goto err;
2157         }
2158         if (!di) {
2159                 ret = -ENOENT;
2160                 goto err;
2161         }
2162         leaf = path->nodes[0];
2163         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2164         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2165         if (ret)
2166                 goto err;
2167         btrfs_release_path(root, path);
2168
2169         ret = btrfs_del_inode_ref(trans, root, name, name_len,
2170                                   inode->i_ino,
2171                                   dir->i_ino, &index);
2172         if (ret) {
2173                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2174                        "inode %lu parent %lu\n", name_len, name,
2175                        inode->i_ino, dir->i_ino);
2176                 goto err;
2177         }
2178
2179         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2180                                          index, name, name_len, -1);
2181         if (IS_ERR(di)) {
2182                 ret = PTR_ERR(di);
2183                 goto err;
2184         }
2185         if (!di) {
2186                 ret = -ENOENT;
2187                 goto err;
2188         }
2189         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2190         btrfs_release_path(root, path);
2191
2192         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2193                                          inode, dir->i_ino);
2194         BUG_ON(ret != 0 && ret != -ENOENT);
2195         if (ret != -ENOENT)
2196                 BTRFS_I(dir)->log_dirty_trans = trans->transid;
2197
2198         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2199                                            dir, index);
2200         BUG_ON(ret);
2201 err:
2202         btrfs_free_path(path);
2203         if (ret)
2204                 goto out;
2205
2206         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2207         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2208         btrfs_update_inode(trans, root, dir);
2209         btrfs_drop_nlink(inode);
2210         ret = btrfs_update_inode(trans, root, inode);
2211         dir->i_sb->s_dirt = 1;
2212 out:
2213         return ret;
2214 }
2215
2216 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2217 {
2218         struct btrfs_root *root;
2219         struct btrfs_trans_handle *trans;
2220         struct inode *inode = dentry->d_inode;
2221         int ret;
2222         unsigned long nr = 0;
2223
2224         root = BTRFS_I(dir)->root;
2225
2226         ret = btrfs_check_free_space(root, 1, 1);
2227         if (ret)
2228                 goto fail;
2229
2230         trans = btrfs_start_transaction(root, 1);
2231
2232         btrfs_set_trans_block_group(trans, dir);
2233         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2234                                  dentry->d_name.name, dentry->d_name.len);
2235
2236         if (inode->i_nlink == 0)
2237                 ret = btrfs_orphan_add(trans, inode);
2238
2239         nr = trans->blocks_used;
2240
2241         btrfs_end_transaction_throttle(trans, root);
2242 fail:
2243         btrfs_btree_balance_dirty(root, nr);
2244         return ret;
2245 }
2246
2247 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2248 {
2249         struct inode *inode = dentry->d_inode;
2250         int err = 0;
2251         int ret;
2252         struct btrfs_root *root = BTRFS_I(dir)->root;
2253         struct btrfs_trans_handle *trans;
2254         unsigned long nr = 0;
2255
2256         /*
2257          * the FIRST_FREE_OBJECTID check makes sure we don't try to rmdir
2258          * the root of a subvolume or snapshot
2259          */
2260         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2261             inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) {
2262                 return -ENOTEMPTY;
2263         }
2264
2265         ret = btrfs_check_free_space(root, 1, 1);
2266         if (ret)
2267                 goto fail;
2268
2269         trans = btrfs_start_transaction(root, 1);
2270         btrfs_set_trans_block_group(trans, dir);
2271
2272         err = btrfs_orphan_add(trans, inode);
2273         if (err)
2274                 goto fail_trans;
2275
2276         /* now the directory is empty */
2277         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2278                                  dentry->d_name.name, dentry->d_name.len);
2279         if (!err)
2280                 btrfs_i_size_write(inode, 0);
2281
2282 fail_trans:
2283         nr = trans->blocks_used;
2284         ret = btrfs_end_transaction_throttle(trans, root);
2285 fail:
2286         btrfs_btree_balance_dirty(root, nr);
2287
2288         if (ret && !err)
2289                 err = ret;
2290         return err;
2291 }
2292
2293 #if 0
2294 /*
2295  * when truncating bytes in a file, it is possible to avoid reading
2296  * the leaves that contain only checksum items.  This can be the
2297  * majority of the IO required to delete a large file, but it must
2298  * be done carefully.
2299  *
2300  * The keys in the level just above the leaves are checked to make sure
2301  * the lowest key in a given leaf is a csum key, and starts at an offset
2302  * after the new  size.
2303  *
2304  * Then the key for the next leaf is checked to make sure it also has
2305  * a checksum item for the same file.  If it does, we know our target leaf
2306  * contains only checksum items, and it can be safely freed without reading
2307  * it.
2308  *
2309  * This is just an optimization targeted at large files.  It may do
2310  * nothing.  It will return 0 unless things went badly.
2311  */
2312 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
2313                                      struct btrfs_root *root,
2314                                      struct btrfs_path *path,
2315                                      struct inode *inode, u64 new_size)
2316 {
2317         struct btrfs_key key;
2318         int ret;
2319         int nritems;
2320         struct btrfs_key found_key;
2321         struct btrfs_key other_key;
2322         struct btrfs_leaf_ref *ref;
2323         u64 leaf_gen;
2324         u64 leaf_start;
2325
2326         path->lowest_level = 1;
2327         key.objectid = inode->i_ino;
2328         key.type = BTRFS_CSUM_ITEM_KEY;
2329         key.offset = new_size;
2330 again:
2331         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2332         if (ret < 0)
2333                 goto out;
2334
2335         if (path->nodes[1] == NULL) {
2336                 ret = 0;
2337                 goto out;
2338         }
2339         ret = 0;
2340         btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
2341         nritems = btrfs_header_nritems(path->nodes[1]);
2342
2343         if (!nritems)
2344                 goto out;
2345
2346         if (path->slots[1] >= nritems)
2347                 goto next_node;
2348
2349         /* did we find a key greater than anything we want to delete? */
2350         if (found_key.objectid > inode->i_ino ||
2351            (found_key.objectid == inode->i_ino && found_key.type > key.type))
2352                 goto out;
2353
2354         /* we check the next key in the node to make sure the leave contains
2355          * only checksum items.  This comparison doesn't work if our
2356          * leaf is the last one in the node
2357          */
2358         if (path->slots[1] + 1 >= nritems) {
2359 next_node:
2360                 /* search forward from the last key in the node, this
2361                  * will bring us into the next node in the tree
2362                  */
2363                 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
2364
2365                 /* unlikely, but we inc below, so check to be safe */
2366                 if (found_key.offset == (u64)-1)
2367                         goto out;
2368
2369                 /* search_forward needs a path with locks held, do the
2370                  * search again for the original key.  It is possible
2371                  * this will race with a balance and return a path that
2372                  * we could modify, but this drop is just an optimization
2373                  * and is allowed to miss some leaves.
2374                  */
2375                 btrfs_release_path(root, path);
2376                 found_key.offset++;
2377
2378                 /* setup a max key for search_forward */
2379                 other_key.offset = (u64)-1;
2380                 other_key.type = key.type;
2381                 other_key.objectid = key.objectid;
2382
2383                 path->keep_locks = 1;
2384                 ret = btrfs_search_forward(root, &found_key, &other_key,
2385                                            path, 0, 0);
2386                 path->keep_locks = 0;
2387                 if (ret || found_key.objectid != key.objectid ||
2388                     found_key.type != key.type) {
2389                         ret = 0;
2390                         goto out;
2391                 }
2392
2393                 key.offset = found_key.offset;
2394                 btrfs_release_path(root, path);
2395                 cond_resched();
2396                 goto again;
2397         }
2398
2399         /* we know there's one more slot after us in the tree,
2400          * read that key so we can verify it is also a checksum item
2401          */
2402         btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
2403
2404         if (found_key.objectid < inode->i_ino)
2405                 goto next_key;
2406
2407         if (found_key.type != key.type || found_key.offset < new_size)
2408                 goto next_key;
2409
2410         /*
2411          * if the key for the next leaf isn't a csum key from this objectid,
2412          * we can't be sure there aren't good items inside this leaf.
2413          * Bail out
2414          */
2415         if (other_key.objectid != inode->i_ino || other_key.type != key.type)
2416                 goto out;
2417
2418         leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
2419         leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
2420         /*
2421          * it is safe to delete this leaf, it contains only
2422          * csum items from this inode at an offset >= new_size
2423          */
2424         ret = btrfs_del_leaf(trans, root, path, leaf_start);
2425         BUG_ON(ret);
2426
2427         if (root->ref_cows && leaf_gen < trans->transid) {
2428                 ref = btrfs_alloc_leaf_ref(root, 0);
2429                 if (ref) {
2430                         ref->root_gen = root->root_key.offset;
2431                         ref->bytenr = leaf_start;
2432                         ref->owner = 0;
2433                         ref->generation = leaf_gen;
2434                         ref->nritems = 0;
2435
2436                         ret = btrfs_add_leaf_ref(root, ref, 0);
2437                         WARN_ON(ret);
2438                         btrfs_free_leaf_ref(root, ref);
2439                 } else {
2440                         WARN_ON(1);
2441                 }
2442         }
2443 next_key:
2444         btrfs_release_path(root, path);
2445
2446         if (other_key.objectid == inode->i_ino &&
2447             other_key.type == key.type && other_key.offset > key.offset) {
2448                 key.offset = other_key.offset;
2449                 cond_resched();
2450                 goto again;
2451         }
2452         ret = 0;
2453 out:
2454         /* fixup any changes we've made to the path */
2455         path->lowest_level = 0;
2456         path->keep_locks = 0;
2457         btrfs_release_path(root, path);
2458         return ret;
2459 }
2460
2461 #endif
2462
2463 /*
2464  * this can truncate away extent items, csum items and directory items.
2465  * It starts at a high offset and removes keys until it can't find
2466  * any higher than new_size
2467  *
2468  * csum items that cross the new i_size are truncated to the new size
2469  * as well.
2470  *
2471  * min_type is the minimum key type to truncate down to.  If set to 0, this
2472  * will kill all the items on this inode, including the INODE_ITEM_KEY.
2473  */
2474 noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2475                                         struct btrfs_root *root,
2476                                         struct inode *inode,
2477                                         u64 new_size, u32 min_type)
2478 {
2479         int ret;
2480         struct btrfs_path *path;
2481         struct btrfs_key key;
2482         struct btrfs_key found_key;
2483         u32 found_type;
2484         struct extent_buffer *leaf;
2485         struct btrfs_file_extent_item *fi;
2486         u64 extent_start = 0;
2487         u64 extent_num_bytes = 0;
2488         u64 item_end = 0;
2489         u64 root_gen = 0;
2490         u64 root_owner = 0;
2491         int found_extent;
2492         int del_item;
2493         int pending_del_nr = 0;
2494         int pending_del_slot = 0;
2495         int extent_type = -1;
2496         int encoding;
2497         u64 mask = root->sectorsize - 1;
2498
2499         if (root->ref_cows)
2500                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
2501         path = btrfs_alloc_path();
2502         path->reada = -1;
2503         BUG_ON(!path);
2504
2505         /* FIXME, add redo link to tree so we don't leak on crash */
2506         key.objectid = inode->i_ino;
2507         key.offset = (u64)-1;
2508         key.type = (u8)-1;
2509
2510         btrfs_init_path(path);
2511
2512 search_again:
2513         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2514         if (ret < 0)
2515                 goto error;
2516
2517         if (ret > 0) {
2518                 /* there are no items in the tree for us to truncate, we're
2519                  * done
2520                  */
2521                 if (path->slots[0] == 0) {
2522                         ret = 0;
2523                         goto error;
2524                 }
2525                 path->slots[0]--;
2526         }
2527
2528         while (1) {
2529                 fi = NULL;
2530                 leaf = path->nodes[0];
2531                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2532                 found_type = btrfs_key_type(&found_key);
2533                 encoding = 0;
2534
2535                 if (found_key.objectid != inode->i_ino)
2536                         break;
2537
2538                 if (found_type < min_type)
2539                         break;
2540
2541                 item_end = found_key.offset;
2542                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
2543                         fi = btrfs_item_ptr(leaf, path->slots[0],
2544                                             struct btrfs_file_extent_item);
2545                         extent_type = btrfs_file_extent_type(leaf, fi);
2546                         encoding = btrfs_file_extent_compression(leaf, fi);
2547                         encoding |= btrfs_file_extent_encryption(leaf, fi);
2548                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
2549
2550                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2551                                 item_end +=
2552                                     btrfs_file_extent_num_bytes(leaf, fi);
2553                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2554                                 item_end += btrfs_file_extent_inline_len(leaf,
2555                                                                          fi);
2556                         }
2557                         item_end--;
2558                 }
2559                 if (item_end < new_size) {
2560                         if (found_type == BTRFS_DIR_ITEM_KEY)
2561                                 found_type = BTRFS_INODE_ITEM_KEY;
2562                         else if (found_type == BTRFS_EXTENT_ITEM_KEY)
2563                                 found_type = BTRFS_EXTENT_DATA_KEY;
2564                         else if (found_type == BTRFS_EXTENT_DATA_KEY)
2565                                 found_type = BTRFS_XATTR_ITEM_KEY;
2566                         else if (found_type == BTRFS_XATTR_ITEM_KEY)
2567                                 found_type = BTRFS_INODE_REF_KEY;
2568                         else if (found_type)
2569                                 found_type--;
2570                         else
2571                                 break;
2572                         btrfs_set_key_type(&key, found_type);
2573                         goto next;
2574                 }
2575                 if (found_key.offset >= new_size)
2576                         del_item = 1;
2577                 else
2578                         del_item = 0;
2579                 found_extent = 0;
2580
2581                 /* FIXME, shrink the extent if the ref count is only 1 */
2582                 if (found_type != BTRFS_EXTENT_DATA_KEY)
2583                         goto delete;
2584
2585                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2586                         u64 num_dec;
2587                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
2588                         if (!del_item && !encoding) {
2589                                 u64 orig_num_bytes =
2590                                         btrfs_file_extent_num_bytes(leaf, fi);
2591                                 extent_num_bytes = new_size -
2592                                         found_key.offset + root->sectorsize - 1;
2593                                 extent_num_bytes = extent_num_bytes &
2594                                         ~((u64)root->sectorsize - 1);
2595                                 btrfs_set_file_extent_num_bytes(leaf, fi,
2596                                                          extent_num_bytes);
2597                                 num_dec = (orig_num_bytes -
2598                                            extent_num_bytes);
2599                                 if (root->ref_cows && extent_start != 0)
2600                                         inode_sub_bytes(inode, num_dec);
2601                                 btrfs_mark_buffer_dirty(leaf);
2602                         } else {
2603                                 extent_num_bytes =
2604                                         btrfs_file_extent_disk_num_bytes(leaf,
2605                                                                          fi);
2606                                 /* FIXME blocksize != 4096 */
2607                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
2608                                 if (extent_start != 0) {
2609                                         found_extent = 1;
2610                                         if (root->ref_cows)
2611                                                 inode_sub_bytes(inode, num_dec);
2612                                 }
2613                                 root_gen = btrfs_header_generation(leaf);
2614                                 root_owner = btrfs_header_owner(leaf);
2615                         }
2616                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2617                         /*
2618                          * we can't truncate inline items that have had
2619                          * special encodings
2620                          */
2621                         if (!del_item &&
2622                             btrfs_file_extent_compression(leaf, fi) == 0 &&
2623                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
2624                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
2625                                 u32 size = new_size - found_key.offset;
2626
2627                                 if (root->ref_cows) {
2628                                         inode_sub_bytes(inode, item_end + 1 -
2629                                                         new_size);
2630                                 }
2631                                 size =
2632                                     btrfs_file_extent_calc_inline_size(size);
2633                                 ret = btrfs_truncate_item(trans, root, path,
2634                                                           size, 1);
2635                                 BUG_ON(ret);
2636                         } else if (root->ref_cows) {
2637                                 inode_sub_bytes(inode, item_end + 1 -
2638                                                 found_key.offset);
2639                         }
2640                 }
2641 delete:
2642                 if (del_item) {
2643                         if (!pending_del_nr) {
2644                                 /* no pending yet, add ourselves */
2645                                 pending_del_slot = path->slots[0];
2646                                 pending_del_nr = 1;
2647                         } else if (pending_del_nr &&
2648                                    path->slots[0] + 1 == pending_del_slot) {
2649                                 /* hop on the pending chunk */
2650                                 pending_del_nr++;
2651                                 pending_del_slot = path->slots[0];
2652                         } else {
2653                                 BUG();
2654                         }
2655                 } else {
2656                         break;
2657                 }
2658                 if (found_extent) {
2659                         ret = btrfs_free_extent(trans, root, extent_start,
2660                                                 extent_num_bytes,
2661                                                 leaf->start, root_owner,
2662                                                 root_gen, inode->i_ino, 0);
2663                         BUG_ON(ret);
2664                 }
2665 next:
2666                 if (path->slots[0] == 0) {
2667                         if (pending_del_nr)
2668                                 goto del_pending;
2669                         btrfs_release_path(root, path);
2670                         goto search_again;
2671                 }
2672
2673                 path->slots[0]--;
2674                 if (pending_del_nr &&
2675                     path->slots[0] + 1 != pending_del_slot) {
2676                         struct btrfs_key debug;
2677 del_pending:
2678                         btrfs_item_key_to_cpu(path->nodes[0], &debug,
2679                                               pending_del_slot);
2680                         ret = btrfs_del_items(trans, root, path,
2681                                               pending_del_slot,
2682                                               pending_del_nr);
2683                         BUG_ON(ret);
2684                         pending_del_nr = 0;
2685                         btrfs_release_path(root, path);
2686                         goto search_again;
2687                 }
2688         }
2689         ret = 0;
2690 error:
2691         if (pending_del_nr) {
2692                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
2693                                       pending_del_nr);
2694         }
2695         btrfs_free_path(path);
2696         inode->i_sb->s_dirt = 1;
2697         return ret;
2698 }
2699
2700 /*
2701  * taken from block_truncate_page, but does cow as it zeros out
2702  * any bytes left in the last page in the file.
2703  */
2704 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
2705 {
2706         struct inode *inode = mapping->host;
2707         struct btrfs_root *root = BTRFS_I(inode)->root;
2708         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2709         struct btrfs_ordered_extent *ordered;
2710         char *kaddr;
2711         u32 blocksize = root->sectorsize;
2712         pgoff_t index = from >> PAGE_CACHE_SHIFT;
2713         unsigned offset = from & (PAGE_CACHE_SIZE-1);
2714         struct page *page;
2715         int ret = 0;
2716         u64 page_start;
2717         u64 page_end;
2718
2719         if ((offset & (blocksize - 1)) == 0)
2720                 goto out;
2721
2722         ret = -ENOMEM;
2723 again:
2724         page = grab_cache_page(mapping, index);
2725         if (!page)
2726                 goto out;
2727
2728         page_start = page_offset(page);
2729         page_end = page_start + PAGE_CACHE_SIZE - 1;
2730
2731         if (!PageUptodate(page)) {
2732                 ret = btrfs_readpage(NULL, page);
2733                 lock_page(page);
2734                 if (page->mapping != mapping) {
2735                         unlock_page(page);
2736                         page_cache_release(page);
2737                         goto again;
2738                 }
2739                 if (!PageUptodate(page)) {
2740                         ret = -EIO;
2741                         goto out_unlock;
2742                 }
2743         }
2744         wait_on_page_writeback(page);
2745
2746         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2747         set_page_extent_mapped(page);
2748
2749         ordered = btrfs_lookup_ordered_extent(inode, page_start);
2750         if (ordered) {
2751                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2752                 unlock_page(page);
2753                 page_cache_release(page);
2754                 btrfs_start_ordered_extent(inode, ordered, 1);
2755                 btrfs_put_ordered_extent(ordered);
2756                 goto again;
2757         }
2758
2759         btrfs_set_extent_delalloc(inode, page_start, page_end);
2760         ret = 0;
2761         if (offset != PAGE_CACHE_SIZE) {
2762                 kaddr = kmap(page);
2763                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
2764                 flush_dcache_page(page);
2765                 kunmap(page);
2766         }
2767         ClearPageChecked(page);
2768         set_page_dirty(page);
2769         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2770
2771 out_unlock:
2772         unlock_page(page);
2773         page_cache_release(page);
2774 out:
2775         return ret;
2776 }
2777
2778 int btrfs_cont_expand(struct inode *inode, loff_t size)
2779 {
2780         struct btrfs_trans_handle *trans;
2781         struct btrfs_root *root = BTRFS_I(inode)->root;
2782         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2783         struct extent_map *em;
2784         u64 mask = root->sectorsize - 1;
2785         u64 hole_start = (inode->i_size + mask) & ~mask;
2786         u64 block_end = (size + mask) & ~mask;
2787         u64 last_byte;
2788         u64 cur_offset;
2789         u64 hole_size;
2790         int err;
2791
2792         if (size <= hole_start)
2793                 return 0;
2794
2795         err = btrfs_check_free_space(root, 1, 0);
2796         if (err)
2797                 return err;
2798
2799         btrfs_truncate_page(inode->i_mapping, inode->i_size);
2800
2801         while (1) {
2802                 struct btrfs_ordered_extent *ordered;
2803                 btrfs_wait_ordered_range(inode, hole_start,
2804                                          block_end - hole_start);
2805                 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2806                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
2807                 if (!ordered)
2808                         break;
2809                 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2810                 btrfs_put_ordered_extent(ordered);
2811         }
2812
2813         trans = btrfs_start_transaction(root, 1);
2814         btrfs_set_trans_block_group(trans, inode);
2815
2816         cur_offset = hole_start;
2817         while (1) {
2818                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2819                                 block_end - cur_offset, 0);
2820                 BUG_ON(IS_ERR(em) || !em);
2821                 last_byte = min(extent_map_end(em), block_end);
2822                 last_byte = (last_byte + mask) & ~mask;
2823                 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
2824                         u64 hint_byte = 0;
2825                         hole_size = last_byte - cur_offset;
2826                         err = btrfs_drop_extents(trans, root, inode,
2827                                                  cur_offset,
2828                                                  cur_offset + hole_size,
2829                                                  cur_offset, &hint_byte);
2830                         if (err)
2831                                 break;
2832                         err = btrfs_insert_file_extent(trans, root,
2833                                         inode->i_ino, cur_offset, 0,
2834                                         0, hole_size, 0, hole_size,
2835                                         0, 0, 0);
2836                         btrfs_drop_extent_cache(inode, hole_start,
2837                                         last_byte - 1, 0);
2838                 }
2839                 free_extent_map(em);
2840                 cur_offset = last_byte;
2841                 if (err || cur_offset >= block_end)
2842                         break;
2843         }
2844
2845         btrfs_end_transaction(trans, root);
2846         unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2847         return err;
2848 }
2849
2850 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
2851 {
2852         struct inode *inode = dentry->d_inode;
2853         int err;
2854
2855         err = inode_change_ok(inode, attr);
2856         if (err)
2857                 return err;
2858
2859         if (S_ISREG(inode->i_mode) &&
2860             attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
2861                 err = btrfs_cont_expand(inode, attr->ia_size);
2862                 if (err)
2863                         return err;
2864         }
2865
2866         err = inode_setattr(inode, attr);
2867
2868         if (!err && ((attr->ia_valid & ATTR_MODE)))
2869                 err = btrfs_acl_chmod(inode);
2870         return err;
2871 }
2872
2873 void btrfs_delete_inode(struct inode *inode)
2874 {
2875         struct btrfs_trans_handle *trans;
2876         struct btrfs_root *root = BTRFS_I(inode)->root;
2877         unsigned long nr;
2878         int ret;
2879
2880         truncate_inode_pages(&inode->i_data, 0);
2881         if (is_bad_inode(inode)) {
2882                 btrfs_orphan_del(NULL, inode);
2883                 goto no_delete;
2884         }
2885         btrfs_wait_ordered_range(inode, 0, (u64)-1);
2886
2887         btrfs_i_size_write(inode, 0);
2888         trans = btrfs_join_transaction(root, 1);
2889
2890         btrfs_set_trans_block_group(trans, inode);
2891         ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0);
2892         if (ret) {
2893                 btrfs_orphan_del(NULL, inode);
2894                 goto no_delete_lock;
2895         }
2896
2897         btrfs_orphan_del(trans, inode);
2898
2899         nr = trans->blocks_used;
2900         clear_inode(inode);
2901
2902         btrfs_end_transaction(trans, root);
2903         btrfs_btree_balance_dirty(root, nr);
2904         return;
2905
2906 no_delete_lock:
2907         nr = trans->blocks_used;
2908         btrfs_end_transaction(trans, root);
2909         btrfs_btree_balance_dirty(root, nr);
2910 no_delete:
2911         clear_inode(inode);
2912 }
2913
2914 /*
2915  * this returns the key found in the dir entry in the location pointer.
2916  * If no dir entries were found, location->objectid is 0.
2917  */
2918 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
2919                                struct btrfs_key *location)
2920 {
2921         const char *name = dentry->d_name.name;
2922         int namelen = dentry->d_name.len;
2923         struct btrfs_dir_item *di;
2924         struct btrfs_path *path;
2925         struct btrfs_root *root = BTRFS_I(dir)->root;
2926         int ret = 0;
2927
2928         path = btrfs_alloc_path();
2929         BUG_ON(!path);
2930
2931         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
2932                                     namelen, 0);
2933         if (IS_ERR(di))
2934                 ret = PTR_ERR(di);
2935
2936         if (!di || IS_ERR(di))
2937                 goto out_err;
2938
2939         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
2940 out:
2941         btrfs_free_path(path);
2942         return ret;
2943 out_err:
2944         location->objectid = 0;
2945         goto out;
2946 }
2947
2948 /*
2949  * when we hit a tree root in a directory, the btrfs part of the inode
2950  * needs to be changed to reflect the root directory of the tree root.  This
2951  * is kind of like crossing a mount point.
2952  */
2953 static int fixup_tree_root_location(struct btrfs_root *root,
2954                              struct btrfs_key *location,
2955                              struct btrfs_root **sub_root,
2956                              struct dentry *dentry)
2957 {
2958         struct btrfs_root_item *ri;
2959
2960         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
2961                 return 0;
2962         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
2963                 return 0;
2964
2965         *sub_root = btrfs_read_fs_root(root->fs_info, location,
2966                                         dentry->d_name.name,
2967                                         dentry->d_name.len);
2968         if (IS_ERR(*sub_root))
2969                 return PTR_ERR(*sub_root);
2970
2971         ri = &(*sub_root)->root_item;
2972         location->objectid = btrfs_root_dirid(ri);
2973         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
2974         location->offset = 0;
2975
2976         return 0;
2977 }
2978
2979 static noinline void init_btrfs_i(struct inode *inode)
2980 {
2981         struct btrfs_inode *bi = BTRFS_I(inode);
2982
2983         bi->i_acl = NULL;
2984         bi->i_default_acl = NULL;
2985
2986         bi->generation = 0;
2987         bi->sequence = 0;
2988         bi->last_trans = 0;
2989         bi->logged_trans = 0;
2990         bi->delalloc_bytes = 0;
2991         bi->disk_i_size = 0;
2992         bi->flags = 0;
2993         bi->index_cnt = (u64)-1;
2994         bi->log_dirty_trans = 0;
2995         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
2996         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
2997                              inode->i_mapping, GFP_NOFS);
2998         extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
2999                              inode->i_mapping, GFP_NOFS);
3000         INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
3001         btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
3002         mutex_init(&BTRFS_I(inode)->extent_mutex);
3003         mutex_init(&BTRFS_I(inode)->log_mutex);
3004 }
3005
3006 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3007 {
3008         struct btrfs_iget_args *args = p;
3009         inode->i_ino = args->ino;
3010         init_btrfs_i(inode);
3011         BTRFS_I(inode)->root = args->root;
3012         return 0;
3013 }
3014
3015 static int btrfs_find_actor(struct inode *inode, void *opaque)
3016 {
3017         struct btrfs_iget_args *args = opaque;
3018         return args->ino == inode->i_ino &&
3019                 args->root == BTRFS_I(inode)->root;
3020 }
3021
3022 struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
3023                             struct btrfs_root *root, int wait)
3024 {
3025         struct inode *inode;
3026         struct btrfs_iget_args args;
3027         args.ino = objectid;
3028         args.root = root;
3029
3030         if (wait) {
3031                 inode = ilookup5(s, objectid, btrfs_find_actor,
3032                                  (void *)&args);
3033         } else {
3034                 inode = ilookup5_nowait(s, objectid, btrfs_find_actor,
3035                                         (void *)&args);
3036         }
3037         return inode;
3038 }
3039
3040 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
3041                                 struct btrfs_root *root)
3042 {
3043         struct inode *inode;
3044         struct btrfs_iget_args args;
3045         args.ino = objectid;
3046         args.root = root;
3047
3048         inode = iget5_locked(s, objectid, btrfs_find_actor,
3049                              btrfs_init_locked_inode,
3050                              (void *)&args);
3051         return inode;
3052 }
3053
3054 /* Get an inode object given its location and corresponding root.
3055  * Returns in *is_new if the inode was read from disk
3056  */
3057 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3058                          struct btrfs_root *root, int *is_new)
3059 {
3060         struct inode *inode;
3061
3062         inode = btrfs_iget_locked(s, location->objectid, root);
3063         if (!inode)
3064                 return ERR_PTR(-EACCES);
3065
3066         if (inode->i_state & I_NEW) {
3067                 BTRFS_I(inode)->root = root;
3068                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3069                 btrfs_read_locked_inode(inode);
3070                 unlock_new_inode(inode);
3071                 if (is_new)
3072                         *is_new = 1;
3073         } else {
3074                 if (is_new)
3075                         *is_new = 0;
3076         }
3077
3078         return inode;
3079 }
3080
3081 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3082 {
3083         struct inode *inode;
3084         struct btrfs_inode *bi = BTRFS_I(dir);
3085         struct btrfs_root *root = bi->root;
3086         struct btrfs_root *sub_root = root;
3087         struct btrfs_key location;
3088         int ret, new;
3089
3090         if (dentry->d_name.len > BTRFS_NAME_LEN)
3091                 return ERR_PTR(-ENAMETOOLONG);
3092
3093         ret = btrfs_inode_by_name(dir, dentry, &location);
3094
3095         if (ret < 0)
3096                 return ERR_PTR(ret);
3097
3098         inode = NULL;
3099         if (location.objectid) {
3100                 ret = fixup_tree_root_location(root, &location, &sub_root,
3101                                                 dentry);
3102                 if (ret < 0)
3103                         return ERR_PTR(ret);
3104                 if (ret > 0)
3105                         return ERR_PTR(-ENOENT);
3106                 inode = btrfs_iget(dir->i_sb, &location, sub_root, &new);
3107                 if (IS_ERR(inode))
3108                         return ERR_CAST(inode);
3109         }
3110         return inode;
3111 }
3112
3113 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3114                                    struct nameidata *nd)
3115 {
3116         struct inode *inode;
3117
3118         if (dentry->d_name.len > BTRFS_NAME_LEN)
3119                 return ERR_PTR(-ENAMETOOLONG);
3120
3121         inode = btrfs_lookup_dentry(dir, dentry);
3122         if (IS_ERR(inode))
3123                 return ERR_CAST(inode);
3124
3125         return d_splice_alias(inode, dentry);
3126 }
3127
3128 static unsigned char btrfs_filetype_table[] = {
3129         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
3130 };
3131
3132 static int btrfs_real_readdir(struct file *filp, void *dirent,
3133                               filldir_t filldir)
3134 {
3135         struct inode *inode = filp->f_dentry->d_inode;
3136         struct btrfs_root *root = BTRFS_I(inode)->root;
3137         struct btrfs_item *item;
3138         struct btrfs_dir_item *di;
3139         struct btrfs_key key;
3140         struct btrfs_key found_key;
3141         struct btrfs_path *path;
3142         int ret;
3143         u32 nritems;
3144         struct extent_buffer *leaf;
3145         int slot;
3146         int advance;
3147         unsigned char d_type;
3148         int over = 0;
3149         u32 di_cur;
3150         u32 di_total;
3151         u32 di_len;
3152         int key_type = BTRFS_DIR_INDEX_KEY;
3153         char tmp_name[32];
3154         char *name_ptr;
3155         int name_len;
3156
3157         /* FIXME, use a real flag for deciding about the key type */
3158         if (root->fs_info->tree_root == root)
3159                 key_type = BTRFS_DIR_ITEM_KEY;
3160
3161         /* special case for "." */
3162         if (filp->f_pos == 0) {
3163                 over = filldir(dirent, ".", 1,
3164                                1, inode->i_ino,
3165                                DT_DIR);
3166                 if (over)
3167                         return 0;
3168                 filp->f_pos = 1;
3169         }
3170         /* special case for .., just use the back ref */
3171         if (filp->f_pos == 1) {
3172                 u64 pino = parent_ino(filp->f_path.dentry);
3173                 over = filldir(dirent, "..", 2,
3174                                2, pino, DT_DIR);
3175                 if (over)
3176                         return 0;
3177                 filp->f_pos = 2;
3178         }
3179         path = btrfs_alloc_path();
3180         path->reada = 2;
3181
3182         btrfs_set_key_type(&key, key_type);
3183         key.offset = filp->f_pos;
3184         key.objectid = inode->i_ino;
3185
3186         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3187         if (ret < 0)
3188                 goto err;
3189         advance = 0;
3190
3191         while (1) {
3192                 leaf = path->nodes[0];
3193                 nritems = btrfs_header_nritems(leaf);
3194                 slot = path->slots[0];
3195                 if (advance || slot >= nritems) {
3196                         if (slot >= nritems - 1) {
3197                                 ret = btrfs_next_leaf(root, path);
3198                                 if (ret)
3199                                         break;
3200                                 leaf = path->nodes[0];
3201                                 nritems = btrfs_header_nritems(leaf);
3202                                 slot = path->slots[0];
3203                         } else {
3204                                 slot++;
3205                                 path->slots[0]++;
3206                         }
3207                 }
3208
3209                 advance = 1;
3210                 item = btrfs_item_nr(leaf, slot);
3211                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3212
3213                 if (found_key.objectid != key.objectid)
3214                         break;
3215                 if (btrfs_key_type(&found_key) != key_type)
3216                         break;
3217                 if (found_key.offset < filp->f_pos)
3218                         continue;
3219
3220                 filp->f_pos = found_key.offset;
3221
3222                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
3223                 di_cur = 0;
3224                 di_total = btrfs_item_size(leaf, item);
3225
3226                 while (di_cur < di_total) {
3227                         struct btrfs_key location;
3228
3229                         name_len = btrfs_dir_name_len(leaf, di);
3230                         if (name_len <= sizeof(tmp_name)) {
3231                                 name_ptr = tmp_name;
3232                         } else {
3233                                 name_ptr = kmalloc(name_len, GFP_NOFS);
3234                                 if (!name_ptr) {
3235                                         ret = -ENOMEM;
3236                                         goto err;
3237                                 }
3238                         }
3239                         read_extent_buffer(leaf, name_ptr,
3240                                            (unsigned long)(di + 1), name_len);
3241
3242                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
3243                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
3244
3245                         /* is this a reference to our own snapshot? If so
3246                          * skip it
3247                          */
3248                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
3249                             location.objectid == root->root_key.objectid) {
3250                                 over = 0;
3251                                 goto skip;
3252                         }
3253                         over = filldir(dirent, name_ptr, name_len,
3254                                        found_key.offset, location.objectid,
3255                                        d_type);
3256
3257 skip:
3258                         if (name_ptr != tmp_name)
3259                                 kfree(name_ptr);
3260
3261                         if (over)
3262                                 goto nopos;
3263                         di_len = btrfs_dir_name_len(leaf, di) +
3264                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
3265                         di_cur += di_len;
3266                         di = (struct btrfs_dir_item *)((char *)di + di_len);
3267                 }
3268         }
3269
3270         /* Reached end of directory/root. Bump pos past the last item. */
3271         if (key_type == BTRFS_DIR_INDEX_KEY)
3272                 filp->f_pos = INT_LIMIT(typeof(filp->f_pos));
3273         else
3274                 filp->f_pos++;
3275 nopos:
3276         ret = 0;
3277 err:
3278         btrfs_free_path(path);
3279         return ret;
3280 }
3281
3282 int btrfs_write_inode(struct inode *inode, int wait)
3283 {
3284         struct btrfs_root *root = BTRFS_I(inode)->root;
3285         struct btrfs_trans_handle *trans;
3286         int ret = 0;
3287
3288         if (root->fs_info->btree_inode == inode)
3289                 return 0;
3290
3291         if (wait) {
3292                 trans = btrfs_join_transaction(root, 1);
3293                 btrfs_set_trans_block_group(trans, inode);
3294                 ret = btrfs_commit_transaction(trans, root);
3295         }
3296         return ret;
3297 }
3298
3299 /*
3300  * This is somewhat expensive, updating the tree every time the
3301  * inode changes.  But, it is most likely to find the inode in cache.
3302  * FIXME, needs more benchmarking...there are no reasons other than performance
3303  * to keep or drop this code.
3304  */
3305 void btrfs_dirty_inode(struct inode *inode)
3306 {
3307         struct btrfs_root *root = BTRFS_I(inode)->root;
3308         struct btrfs_trans_handle *trans;
3309
3310         trans = btrfs_join_transaction(root, 1);
3311         btrfs_set_trans_block_group(trans, inode);
3312         btrfs_update_inode(trans, root, inode);
3313         btrfs_end_transaction(trans, root);
3314 }
3315
3316 /*
3317  * find the highest existing sequence number in a directory
3318  * and then set the in-memory index_cnt variable to reflect
3319  * free sequence numbers
3320  */
3321 static int btrfs_set_inode_index_count(struct inode *inode)
3322 {
3323         struct btrfs_root *root = BTRFS_I(inode)->root;
3324         struct btrfs_key key, found_key;
3325         struct btrfs_path *path;
3326         struct extent_buffer *leaf;
3327         int ret;
3328
3329         key.objectid = inode->i_ino;
3330         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
3331         key.offset = (u64)-1;
3332
3333         path = btrfs_alloc_path();
3334         if (!path)
3335                 return -ENOMEM;
3336
3337         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3338         if (ret < 0)
3339                 goto out;
3340         /* FIXME: we should be able to handle this */
3341         if (ret == 0)
3342                 goto out;
3343         ret = 0;
3344
3345         /*
3346          * MAGIC NUMBER EXPLANATION:
3347          * since we search a directory based on f_pos we have to start at 2
3348          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
3349          * else has to start at 2
3350          */
3351         if (path->slots[0] == 0) {
3352                 BTRFS_I(inode)->index_cnt = 2;
3353                 goto out;
3354         }
3355
3356         path->slots[0]--;
3357
3358         leaf = path->nodes[0];
3359         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3360
3361         if (found_key.objectid != inode->i_ino ||
3362             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
3363                 BTRFS_I(inode)->index_cnt = 2;
3364                 goto out;
3365         }
3366
3367         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
3368 out:
3369         btrfs_free_path(path);
3370         return ret;
3371 }
3372
3373 /*
3374  * helper to find a free sequence number in a given directory.  This current
3375  * code is very simple, later versions will do smarter things in the btree
3376  */
3377 int btrfs_set_inode_index(struct inode *dir, u64 *index)
3378 {
3379         int ret = 0;
3380
3381         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
3382                 ret = btrfs_set_inode_index_count(dir);
3383                 if (ret)
3384                         return ret;
3385         }
3386
3387         *index = BTRFS_I(dir)->index_cnt;
3388         BTRFS_I(dir)->index_cnt++;
3389
3390         return ret;
3391 }
3392
3393 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
3394                                      struct btrfs_root *root,
3395                                      struct inode *dir,
3396                                      const char *name, int name_len,
3397                                      u64 ref_objectid, u64 objectid,
3398                                      u64 alloc_hint, int mode, u64 *index)
3399 {
3400         struct inode *inode;
3401         struct btrfs_inode_item *inode_item;
3402         struct btrfs_key *location;
3403         struct btrfs_path *path;
3404         struct btrfs_inode_ref *ref;
3405         struct btrfs_key key[2];
3406         u32 sizes[2];
3407         unsigned long ptr;
3408         int ret;
3409         int owner;
3410
3411         path = btrfs_alloc_path();
3412         BUG_ON(!path);
3413
3414         inode = new_inode(root->fs_info->sb);
3415         if (!inode)
3416                 return ERR_PTR(-ENOMEM);
3417
3418         if (dir) {
3419                 ret = btrfs_set_inode_index(dir, index);
3420                 if (ret)
3421                         return ERR_PTR(ret);
3422         }
3423         /*
3424          * index_cnt is ignored for everything but a dir,
3425          * btrfs_get_inode_index_count has an explanation for the magic
3426          * number
3427          */
3428         init_btrfs_i(inode);
3429         BTRFS_I(inode)->index_cnt = 2;
3430         BTRFS_I(inode)->root = root;
3431         BTRFS_I(inode)->generation = trans->transid;
3432
3433         if (mode & S_IFDIR)
3434                 owner = 0;
3435         else
3436                 owner = 1;
3437         BTRFS_I(inode)->block_group =
3438                         btrfs_find_block_group(root, 0, alloc_hint, owner);
3439         if ((mode & S_IFREG)) {
3440                 if (btrfs_test_opt(root, NODATASUM))
3441                         btrfs_set_flag(inode, NODATASUM);
3442                 if (btrfs_test_opt(root, NODATACOW))
3443                         btrfs_set_flag(inode, NODATACOW);
3444         }
3445
3446         key[0].objectid = objectid;
3447         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
3448         key[0].offset = 0;
3449
3450         key[1].objectid = objectid;
3451         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
3452         key[1].offset = ref_objectid;
3453
3454         sizes[0] = sizeof(struct btrfs_inode_item);
3455         sizes[1] = name_len + sizeof(*ref);
3456
3457         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
3458         if (ret != 0)
3459                 goto fail;
3460
3461         if (objectid > root->highest_inode)
3462                 root->highest_inode = objectid;
3463
3464         inode->i_uid = current_fsuid();
3465         inode->i_gid = current_fsgid();
3466         inode->i_mode = mode;
3467         inode->i_ino = objectid;
3468         inode_set_bytes(inode, 0);
3469         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3470         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3471                                   struct btrfs_inode_item);
3472         fill_inode_item(trans, path->nodes[0], inode_item, inode);
3473
3474         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3475                              struct btrfs_inode_ref);
3476         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
3477         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
3478         ptr = (unsigned long)(ref + 1);
3479         write_extent_buffer(path->nodes[0], name, ptr, name_len);
3480
3481         btrfs_mark_buffer_dirty(path->nodes[0]);
3482         btrfs_free_path(path);
3483
3484         location = &BTRFS_I(inode)->location;
3485         location->objectid = objectid;
3486         location->offset = 0;
3487         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
3488
3489         insert_inode_hash(inode);
3490         return inode;
3491 fail:
3492         if (dir)
3493                 BTRFS_I(dir)->index_cnt--;
3494         btrfs_free_path(path);
3495         return ERR_PTR(ret);
3496 }
3497
3498 static inline u8 btrfs_inode_type(struct inode *inode)
3499 {
3500         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
3501 }
3502
3503 /*
3504  * utility function to add 'inode' into 'parent_inode' with
3505  * a give name and a given sequence number.
3506  * if 'add_backref' is true, also insert a backref from the
3507  * inode to the parent directory.
3508  */
3509 int btrfs_add_link(struct btrfs_trans_handle *trans,
3510                    struct inode *parent_inode, struct inode *inode,
3511                    const char *name, int name_len, int add_backref, u64 index)
3512 {
3513         int ret;
3514         struct btrfs_key key;
3515         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
3516
3517         key.objectid = inode->i_ino;
3518         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
3519         key.offset = 0;
3520
3521         ret = btrfs_insert_dir_item(trans, root, name, name_len,
3522                                     parent_inode->i_ino,
3523                                     &key, btrfs_inode_type(inode),
3524                                     index);
3525         if (ret == 0) {
3526                 if (add_backref) {
3527                         ret = btrfs_insert_inode_ref(trans, root,
3528                                                      name, name_len,
3529                                                      inode->i_ino,
3530                                                      parent_inode->i_ino,
3531                                                      index);
3532                 }
3533                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
3534                                    name_len * 2);
3535                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
3536                 ret = btrfs_update_inode(trans, root, parent_inode);
3537         }
3538         return ret;
3539 }
3540
3541 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
3542                             struct dentry *dentry, struct inode *inode,
3543                             int backref, u64 index)
3544 {
3545         int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3546                                  inode, dentry->d_name.name,
3547                                  dentry->d_name.len, backref, index);
3548         if (!err) {
3549                 d_instantiate(dentry, inode);
3550                 return 0;
3551         }
3552         if (err > 0)
3553                 err = -EEXIST;
3554         return err;
3555 }
3556
3557 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
3558                         int mode, dev_t rdev)
3559 {
3560         struct btrfs_trans_handle *trans;
3561         struct btrfs_root *root = BTRFS_I(dir)->root;
3562         struct inode *inode = NULL;
3563         int err;
3564         int drop_inode = 0;
3565         u64 objectid;
3566         unsigned long nr = 0;
3567         u64 index = 0;
3568
3569         if (!new_valid_dev(rdev))
3570                 return -EINVAL;
3571
3572         err = btrfs_check_free_space(root, 1, 0);
3573         if (err)
3574                 goto fail;
3575
3576         trans = btrfs_start_transaction(root, 1);
3577         btrfs_set_trans_block_group(trans, dir);
3578
3579         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3580         if (err) {
3581                 err = -ENOSPC;
3582                 goto out_unlock;
3583         }
3584
3585         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3586                                 dentry->d_name.len,
3587                                 dentry->d_parent->d_inode->i_ino, objectid,
3588                                 BTRFS_I(dir)->block_group, mode, &index);
3589         err = PTR_ERR(inode);
3590         if (IS_ERR(inode))
3591                 goto out_unlock;
3592
3593         err = btrfs_init_acl(inode, dir);
3594         if (err) {
3595                 drop_inode = 1;
3596                 goto out_unlock;
3597         }
3598
3599         btrfs_set_trans_block_group(trans, inode);
3600         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3601         if (err)
3602                 drop_inode = 1;
3603         else {
3604                 inode->i_op = &btrfs_special_inode_operations;
3605                 init_special_inode(inode, inode->i_mode, rdev);
3606                 btrfs_update_inode(trans, root, inode);
3607         }
3608         dir->i_sb->s_dirt = 1;
3609         btrfs_update_inode_block_group(trans, inode);
3610         btrfs_update_inode_block_group(trans, dir);
3611 out_unlock:
3612         nr = trans->blocks_used;
3613         btrfs_end_transaction_throttle(trans, root);
3614 fail:
3615         if (drop_inode) {
3616                 inode_dec_link_count(inode);
3617                 iput(inode);
3618         }
3619         btrfs_btree_balance_dirty(root, nr);
3620         return err;
3621 }
3622
3623 static int btrfs_create(struct inode *dir, struct dentry *dentry,
3624                         int mode, struct nameidata *nd)
3625 {
3626         struct btrfs_trans_handle *trans;
3627         struct btrfs_root *root = BTRFS_I(dir)->root;
3628         struct inode *inode = NULL;
3629         int err;
3630         int drop_inode = 0;
3631         unsigned long nr = 0;
3632         u64 objectid;
3633         u64 index = 0;
3634
3635         err = btrfs_check_free_space(root, 1, 0);
3636         if (err)
3637                 goto fail;
3638         trans = btrfs_start_transaction(root, 1);
3639         btrfs_set_trans_block_group(trans, dir);
3640
3641         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3642         if (err) {
3643                 err = -ENOSPC;
3644                 goto out_unlock;
3645         }
3646
3647         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3648                                 dentry->d_name.len,
3649                                 dentry->d_parent->d_inode->i_ino,
3650                                 objectid, BTRFS_I(dir)->block_group, mode,
3651                                 &index);
3652         err = PTR_ERR(inode);
3653         if (IS_ERR(inode))
3654                 goto out_unlock;
3655
3656         err = btrfs_init_acl(inode, dir);
3657         if (err) {
3658                 drop_inode = 1;
3659                 goto out_unlock;
3660         }
3661
3662         btrfs_set_trans_block_group(trans, inode);
3663         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3664         if (err)
3665                 drop_inode = 1;
3666         else {
3667                 inode->i_mapping->a_ops = &btrfs_aops;
3668                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3669                 inode->i_fop = &btrfs_file_operations;
3670                 inode->i_op = &btrfs_file_inode_operations;
3671                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3672         }
3673         dir->i_sb->s_dirt = 1;
3674         btrfs_update_inode_block_group(trans, inode);
3675         btrfs_update_inode_block_group(trans, dir);
3676 out_unlock:
3677         nr = trans->blocks_used;
3678         btrfs_end_transaction_throttle(trans, root);
3679 fail:
3680         if (drop_inode) {
3681                 inode_dec_link_count(inode);
3682                 iput(inode);
3683         }
3684         btrfs_btree_balance_dirty(root, nr);
3685         return err;
3686 }
3687
3688 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
3689                       struct dentry *dentry)
3690 {
3691         struct btrfs_trans_handle *trans;
3692         struct btrfs_root *root = BTRFS_I(dir)->root;
3693         struct inode *inode = old_dentry->d_inode;
3694         u64 index;
3695         unsigned long nr = 0;
3696         int err;
3697         int drop_inode = 0;
3698
3699         if (inode->i_nlink == 0)
3700                 return -ENOENT;
3701
3702         btrfs_inc_nlink(inode);
3703         err = btrfs_check_free_space(root, 1, 0);
3704         if (err)
3705                 goto fail;
3706         err = btrfs_set_inode_index(dir, &index);
3707         if (err)
3708                 goto fail;
3709
3710         trans = btrfs_start_transaction(root, 1);
3711
3712         btrfs_set_trans_block_group(trans, dir);
3713         atomic_inc(&inode->i_count);
3714
3715         err = btrfs_add_nondir(trans, dentry, inode, 1, index);
3716
3717         if (err)
3718                 drop_inode = 1;
3719
3720         dir->i_sb->s_dirt = 1;
3721         btrfs_update_inode_block_group(trans, dir);
3722         err = btrfs_update_inode(trans, root, inode);
3723
3724         if (err)
3725                 drop_inode = 1;
3726
3727         nr = trans->blocks_used;
3728         btrfs_end_transaction_throttle(trans, root);
3729 fail:
3730         if (drop_inode) {
3731                 inode_dec_link_count(inode);
3732                 iput(inode);
3733         }
3734         btrfs_btree_balance_dirty(root, nr);
3735         return err;
3736 }
3737
3738 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
3739 {
3740         struct inode *inode = NULL;
3741         struct btrfs_trans_handle *trans;
3742         struct btrfs_root *root = BTRFS_I(dir)->root;
3743         int err = 0;
3744         int drop_on_err = 0;
3745         u64 objectid = 0;
3746         u64 index = 0;
3747         unsigned long nr = 1;
3748
3749         err = btrfs_check_free_space(root, 1, 0);
3750         if (err)
3751                 goto out_unlock;
3752
3753         trans = btrfs_start_transaction(root, 1);
3754         btrfs_set_trans_block_group(trans, dir);
3755
3756         if (IS_ERR(trans)) {
3757                 err = PTR_ERR(trans);
3758                 goto out_unlock;
3759         }
3760
3761         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3762         if (err) {
3763                 err = -ENOSPC;
3764                 goto out_unlock;
3765         }
3766
3767         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3768                                 dentry->d_name.len,
3769                                 dentry->d_parent->d_inode->i_ino, objectid,
3770                                 BTRFS_I(dir)->block_group, S_IFDIR | mode,
3771                                 &index);
3772         if (IS_ERR(inode)) {
3773                 err = PTR_ERR(inode);
3774                 goto out_fail;
3775         }
3776
3777         drop_on_err = 1;
3778
3779         err = btrfs_init_acl(inode, dir);
3780         if (err)
3781                 goto out_fail;
3782
3783         inode->i_op = &btrfs_dir_inode_operations;
3784         inode->i_fop = &btrfs_dir_file_operations;
3785         btrfs_set_trans_block_group(trans, inode);
3786
3787         btrfs_i_size_write(inode, 0);
3788         err = btrfs_update_inode(trans, root, inode);
3789         if (err)
3790                 goto out_fail;
3791
3792         err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3793                                  inode, dentry->d_name.name,
3794                                  dentry->d_name.len, 0, index);
3795         if (err)
3796                 goto out_fail;
3797
3798         d_instantiate(dentry, inode);
3799         drop_on_err = 0;
3800         dir->i_sb->s_dirt = 1;
3801         btrfs_update_inode_block_group(trans, inode);
3802         btrfs_update_inode_block_group(trans, dir);
3803
3804 out_fail:
3805         nr = trans->blocks_used;
3806         btrfs_end_transaction_throttle(trans, root);
3807
3808 out_unlock:
3809         if (drop_on_err)
3810                 iput(inode);
3811         btrfs_btree_balance_dirty(root, nr);
3812         return err;
3813 }
3814
3815 /* helper for btfs_get_extent.  Given an existing extent in the tree,
3816  * and an extent that you want to insert, deal with overlap and insert
3817  * the new extent into the tree.
3818  */
3819 static int merge_extent_mapping(struct extent_map_tree *em_tree,
3820                                 struct extent_map *existing,
3821                                 struct extent_map *em,
3822                                 u64 map_start, u64 map_len)
3823 {
3824         u64 start_diff;
3825
3826         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
3827         start_diff = map_start - em->start;
3828         em->start = map_start;
3829         em->len = map_len;
3830         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
3831             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
3832                 em->block_start += start_diff;
3833                 em->block_len -= start_diff;
3834         }
3835         return add_extent_mapping(em_tree, em);
3836 }
3837
3838 static noinline int uncompress_inline(struct btrfs_path *path,
3839                                       struct inode *inode, struct page *page,
3840                                       size_t pg_offset, u64 extent_offset,
3841                                       struct btrfs_file_extent_item *item)
3842 {
3843         int ret;
3844         struct extent_buffer *leaf = path->nodes[0];
3845         char *tmp;
3846         size_t max_size;
3847         unsigned long inline_size;
3848         unsigned long ptr;
3849
3850         WARN_ON(pg_offset != 0);
3851         max_size = btrfs_file_extent_ram_bytes(leaf, item);
3852         inline_size = btrfs_file_extent_inline_item_len(leaf,
3853                                         btrfs_item_nr(leaf, path->slots[0]));
3854         tmp = kmalloc(inline_size, GFP_NOFS);
3855         ptr = btrfs_file_extent_inline_start(item);
3856
3857         read_extent_buffer(leaf, tmp, ptr, inline_size);
3858
3859         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
3860         ret = btrfs_zlib_decompress(tmp, page, extent_offset,
3861                                     inline_size, max_size);
3862         if (ret) {
3863                 char *kaddr = kmap_atomic(page, KM_USER0);
3864                 unsigned long copy_size = min_t(u64,
3865                                   PAGE_CACHE_SIZE - pg_offset,
3866                                   max_size - extent_offset);
3867                 memset(kaddr + pg_offset, 0, copy_size);
3868                 kunmap_atomic(kaddr, KM_USER0);
3869         }
3870         kfree(tmp);
3871         return 0;
3872 }
3873
3874 /*
3875  * a bit scary, this does extent mapping from logical file offset to the disk.
3876  * the ugly parts come from merging extents from the disk with the in-ram
3877  * representation.  This gets more complex because of the data=ordered code,
3878  * where the in-ram extents might be locked pending data=ordered completion.
3879  *
3880  * This also copies inline extents directly into the page.
3881  */
3882
3883 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
3884                                     size_t pg_offset, u64 start, u64 len,
3885                                     int create)
3886 {
3887         int ret;
3888         int err = 0;
3889         u64 bytenr;
3890         u64 extent_start = 0;
3891         u64 extent_end = 0;
3892         u64 objectid = inode->i_ino;
3893         u32 found_type;
3894         struct btrfs_path *path = NULL;
3895         struct btrfs_root *root = BTRFS_I(inode)->root;
3896         struct btrfs_file_extent_item *item;
3897         struct extent_buffer *leaf;
3898         struct btrfs_key found_key;
3899         struct extent_map *em = NULL;
3900         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3901         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3902         struct btrfs_trans_handle *trans = NULL;
3903         int compressed;
3904
3905 again:
3906         spin_lock(&em_tree->lock);
3907         em = lookup_extent_mapping(em_tree, start, len);
3908         if (em)
3909                 em->bdev = root->fs_info->fs_devices->latest_bdev;
3910         spin_unlock(&em_tree->lock);
3911
3912         if (em) {
3913                 if (em->start > start || em->start + em->len <= start)
3914                         free_extent_map(em);
3915                 else if (em->block_start == EXTENT_MAP_INLINE && page)
3916                         free_extent_map(em);
3917                 else
3918                         goto out;
3919         }
3920         em = alloc_extent_map(GFP_NOFS);
3921         if (!em) {
3922                 err = -ENOMEM;
3923                 goto out;
3924         }
3925         em->bdev = root->fs_info->fs_devices->latest_bdev;
3926         em->start = EXTENT_MAP_HOLE;
3927         em->orig_start = EXTENT_MAP_HOLE;
3928         em->len = (u64)-1;
3929         em->block_len = (u64)-1;
3930
3931         if (!path) {
3932                 path = btrfs_alloc_path();
3933                 BUG_ON(!path);
3934         }
3935
3936         ret = btrfs_lookup_file_extent(trans, root, path,
3937                                        objectid, start, trans != NULL);
3938         if (ret < 0) {
3939                 err = ret;
3940                 goto out;
3941         }
3942
3943         if (ret != 0) {
3944                 if (path->slots[0] == 0)
3945                         goto not_found;
3946                 path->slots[0]--;
3947         }
3948
3949         leaf = path->nodes[0];
3950         item = btrfs_item_ptr(leaf, path->slots[0],
3951                               struct btrfs_file_extent_item);
3952         /* are we inside the extent that was found? */
3953         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3954         found_type = btrfs_key_type(&found_key);
3955         if (found_key.objectid != objectid ||
3956             found_type != BTRFS_EXTENT_DATA_KEY) {
3957                 goto not_found;
3958         }
3959
3960         found_type = btrfs_file_extent_type(leaf, item);
3961         extent_start = found_key.offset;
3962         compressed = btrfs_file_extent_compression(leaf, item);
3963         if (found_type == BTRFS_FILE_EXTENT_REG ||
3964             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
3965                 extent_end = extent_start +
3966                        btrfs_file_extent_num_bytes(leaf, item);
3967         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
3968                 size_t size;
3969                 size = btrfs_file_extent_inline_len(leaf, item);
3970                 extent_end = (extent_start + size + root->sectorsize - 1) &
3971                         ~((u64)root->sectorsize - 1);
3972         }
3973
3974         if (start >= extent_end) {
3975                 path->slots[0]++;
3976                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3977                         ret = btrfs_next_leaf(root, path);
3978                         if (ret < 0) {
3979                                 err = ret;
3980                                 goto out;
3981                         }
3982                         if (ret > 0)
3983                                 goto not_found;
3984                         leaf = path->nodes[0];
3985                 }
3986                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3987                 if (found_key.objectid != objectid ||
3988                     found_key.type != BTRFS_EXTENT_DATA_KEY)
3989                         goto not_found;
3990                 if (start + len <= found_key.offset)
3991                         goto not_found;
3992                 em->start = start;
3993                 em->len = found_key.offset - start;
3994                 goto not_found_em;
3995         }
3996
3997         if (found_type == BTRFS_FILE_EXTENT_REG ||
3998             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
3999                 em->start = extent_start;
4000                 em->len = extent_end - extent_start;
4001                 em->orig_start = extent_start -
4002                                  btrfs_file_extent_offset(leaf, item);
4003                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
4004                 if (bytenr == 0) {
4005                         em->block_start = EXTENT_MAP_HOLE;
4006                         goto insert;
4007                 }
4008                 if (compressed) {
4009                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4010                         em->block_start = bytenr;
4011                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
4012                                                                          item);
4013                 } else {
4014                         bytenr += btrfs_file_extent_offset(leaf, item);
4015                         em->block_start = bytenr;
4016                         em->block_len = em->len;
4017                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
4018                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
4019                 }
4020                 goto insert;
4021         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4022                 unsigned long ptr;
4023                 char *map;
4024                 size_t size;
4025                 size_t extent_offset;
4026                 size_t copy_size;
4027
4028                 em->block_start = EXTENT_MAP_INLINE;
4029                 if (!page || create) {
4030                         em->start = extent_start;
4031                         em->len = extent_end - extent_start;
4032                         goto out;
4033                 }
4034
4035                 size = btrfs_file_extent_inline_len(leaf, item);
4036                 extent_offset = page_offset(page) + pg_offset - extent_start;
4037                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
4038                                 size - extent_offset);
4039                 em->start = extent_start + extent_offset;
4040                 em->len = (copy_size + root->sectorsize - 1) &
4041                         ~((u64)root->sectorsize - 1);
4042                 em->orig_start = EXTENT_MAP_INLINE;
4043                 if (compressed)
4044                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4045                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
4046                 if (create == 0 && !PageUptodate(page)) {
4047                         if (btrfs_file_extent_compression(leaf, item) ==
4048                             BTRFS_COMPRESS_ZLIB) {
4049                                 ret = uncompress_inline(path, inode, page,
4050                                                         pg_offset,
4051                                                         extent_offset, item);
4052                                 BUG_ON(ret);
4053                         } else {
4054                                 map = kmap(page);
4055                                 read_extent_buffer(leaf, map + pg_offset, ptr,
4056                                                    copy_size);
4057                                 kunmap(page);
4058                         }
4059                         flush_dcache_page(page);
4060                 } else if (create && PageUptodate(page)) {
4061                         if (!trans) {
4062                                 kunmap(page);
4063                                 free_extent_map(em);
4064                                 em = NULL;
4065                                 btrfs_release_path(root, path);
4066                                 trans = btrfs_join_transaction(root, 1);
4067                                 goto again;
4068                         }
4069                         map = kmap(page);
4070                         write_extent_buffer(leaf, map + pg_offset, ptr,
4071                                             copy_size);
4072                         kunmap(page);
4073                         btrfs_mark_buffer_dirty(leaf);
4074                 }
4075                 set_extent_uptodate(io_tree, em->start,
4076                                     extent_map_end(em) - 1, GFP_NOFS);
4077                 goto insert;
4078         } else {
4079                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
4080                 WARN_ON(1);
4081         }
4082 not_found:
4083         em->start = start;
4084         em->len = len;
4085 not_found_em:
4086         em->block_start = EXTENT_MAP_HOLE;
4087         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
4088 insert:
4089         btrfs_release_path(root, path);
4090         if (em->start > start || extent_map_end(em) <= start) {
4091                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
4092                        "[%llu %llu]\n", (unsigned long long)em->start,
4093                        (unsigned long long)em->len,
4094                        (unsigned long long)start,
4095                        (unsigned long long)len);
4096                 err = -EIO;
4097                 goto out;
4098         }
4099
4100         err = 0;
4101         spin_lock(&em_tree->lock);
4102         ret = add_extent_mapping(em_tree, em);
4103         /* it is possible that someone inserted the extent into the tree
4104          * while we had the lock dropped.  It is also possible that
4105          * an overlapping map exists in the tree
4106          */
4107         if (ret == -EEXIST) {
4108                 struct extent_map *existing;
4109
4110                 ret = 0;
4111
4112                 existing = lookup_extent_mapping(em_tree, start, len);
4113                 if (existing && (existing->start > start ||
4114                     existing->start + existing->len <= start)) {
4115                         free_extent_map(existing);
4116                         existing = NULL;
4117                 }
4118                 if (!existing) {
4119                         existing = lookup_extent_mapping(em_tree, em->start,
4120                                                          em->len);
4121                         if (existing) {
4122                                 err = merge_extent_mapping(em_tree, existing,
4123                                                            em, start,
4124                                                            root->sectorsize);
4125                                 free_extent_map(existing);
4126                                 if (err) {
4127                                         free_extent_map(em);
4128                                         em = NULL;
4129                                 }
4130                         } else {
4131                                 err = -EIO;
4132                                 free_extent_map(em);
4133                                 em = NULL;
4134                         }
4135                 } else {
4136                         free_extent_map(em);
4137                         em = existing;
4138                         err = 0;
4139                 }
4140         }
4141         spin_unlock(&em_tree->lock);
4142 out:
4143         if (path)
4144                 btrfs_free_path(path);
4145         if (trans) {
4146                 ret = btrfs_end_transaction(trans, root);
4147                 if (!err)
4148                         err = ret;
4149         }
4150         if (err) {
4151                 free_extent_map(em);
4152                 WARN_ON(1);
4153                 return ERR_PTR(err);
4154         }
4155         return em;
4156 }
4157
4158 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
4159                         const struct iovec *iov, loff_t offset,
4160                         unsigned long nr_segs)
4161 {
4162         return -EINVAL;
4163 }
4164
4165 static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
4166 {
4167         return extent_bmap(mapping, iblock, btrfs_get_extent);
4168 }
4169
4170 int btrfs_readpage(struct file *file, struct page *page)
4171 {
4172         struct extent_io_tree *tree;
4173         tree = &BTRFS_I(page->mapping->host)->io_tree;
4174         return extent_read_full_page(tree, page, btrfs_get_extent);
4175 }
4176
4177 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
4178 {
4179         struct extent_io_tree *tree;
4180
4181
4182         if (current->flags & PF_MEMALLOC) {
4183                 redirty_page_for_writepage(wbc, page);
4184                 unlock_page(page);
4185                 return 0;
4186         }
4187         tree = &BTRFS_I(page->mapping->host)->io_tree;
4188         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
4189 }
4190
4191 int btrfs_writepages(struct address_space *mapping,
4192                      struct writeback_control *wbc)
4193 {
4194         struct extent_io_tree *tree;
4195
4196         tree = &BTRFS_I(mapping->host)->io_tree;
4197         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
4198 }
4199
4200 static int
4201 btrfs_readpages(struct file *file, struct address_space *mapping,
4202                 struct list_head *pages, unsigned nr_pages)
4203 {
4204         struct extent_io_tree *tree;
4205         tree = &BTRFS_I(mapping->host)->io_tree;
4206         return extent_readpages(tree, mapping, pages, nr_pages,
4207                                 btrfs_get_extent);
4208 }
4209 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4210 {
4211         struct extent_io_tree *tree;
4212         struct extent_map_tree *map;
4213         int ret;
4214
4215         tree = &BTRFS_I(page->mapping->host)->io_tree;
4216         map = &BTRFS_I(page->mapping->host)->extent_tree;
4217         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
4218         if (ret == 1) {
4219                 ClearPagePrivate(page);
4220                 set_page_private(page, 0);
4221                 page_cache_release(page);
4222         }
4223         return ret;
4224 }
4225
4226 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4227 {
4228         if (PageWriteback(page) || PageDirty(page))
4229                 return 0;
4230         return __btrfs_releasepage(page, gfp_flags);
4231 }
4232
4233 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
4234 {
4235         struct extent_io_tree *tree;
4236         struct btrfs_ordered_extent *ordered;
4237         u64 page_start = page_offset(page);
4238         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
4239
4240         wait_on_page_writeback(page);
4241         tree = &BTRFS_I(page->mapping->host)->io_tree;
4242         if (offset) {
4243                 btrfs_releasepage(page, GFP_NOFS);
4244                 return;
4245         }
4246
4247         lock_extent(tree, page_start, page_end, GFP_NOFS);
4248         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
4249                                            page_offset(page));
4250         if (ordered) {
4251                 /*
4252                  * IO on this page will never be started, so we need
4253                  * to account for any ordered extents now
4254                  */
4255                 clear_extent_bit(tree, page_start, page_end,
4256                                  EXTENT_DIRTY | EXTENT_DELALLOC |
4257                                  EXTENT_LOCKED, 1, 0, GFP_NOFS);
4258                 btrfs_finish_ordered_io(page->mapping->host,
4259                                         page_start, page_end);
4260                 btrfs_put_ordered_extent(ordered);
4261                 lock_extent(tree, page_start, page_end, GFP_NOFS);
4262         }
4263         clear_extent_bit(tree, page_start, page_end,
4264                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4265                  EXTENT_ORDERED,
4266                  1, 1, GFP_NOFS);
4267         __btrfs_releasepage(page, GFP_NOFS);
4268
4269         ClearPageChecked(page);
4270         if (PagePrivate(page)) {
4271                 ClearPagePrivate(page);
4272                 set_page_private(page, 0);
4273                 page_cache_release(page);
4274         }
4275 }
4276
4277 /*
4278  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
4279  * called from a page fault handler when a page is first dirtied. Hence we must
4280  * be careful to check for EOF conditions here. We set the page up correctly
4281  * for a written page which means we get ENOSPC checking when writing into
4282  * holes and correct delalloc and unwritten extent mapping on filesystems that
4283  * support these features.
4284  *
4285  * We are not allowed to take the i_mutex here so we have to play games to
4286  * protect against truncate races as the page could now be beyond EOF.  Because
4287  * vmtruncate() writes the inode size before removing pages, once we have the
4288  * page lock we can determine safely if the page is beyond EOF. If it is not
4289  * beyond EOF, then the page is guaranteed safe against truncation until we
4290  * unlock the page.
4291  */
4292 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
4293 {
4294         struct inode *inode = fdentry(vma->vm_file)->d_inode;
4295         struct btrfs_root *root = BTRFS_I(inode)->root;
4296         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4297         struct btrfs_ordered_extent *ordered;
4298         char *kaddr;
4299         unsigned long zero_start;
4300         loff_t size;
4301         int ret;
4302         u64 page_start;
4303         u64 page_end;
4304
4305         ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
4306         if (ret)
4307                 goto out;
4308
4309         ret = -EINVAL;
4310 again:
4311         lock_page(page);
4312         size = i_size_read(inode);
4313         page_start = page_offset(page);
4314         page_end = page_start + PAGE_CACHE_SIZE - 1;
4315
4316         if ((page->mapping != inode->i_mapping) ||
4317             (page_start >= size)) {
4318                 /* page got truncated out from underneath us */
4319                 goto out_unlock;
4320         }
4321         wait_on_page_writeback(page);
4322
4323         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
4324         set_page_extent_mapped(page);
4325
4326         /*
4327          * we can't set the delalloc bits if there are pending ordered
4328          * extents.  Drop our locks and wait for them to finish
4329          */
4330         ordered = btrfs_lookup_ordered_extent(inode, page_start);
4331         if (ordered) {
4332                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4333                 unlock_page(page);
4334                 btrfs_start_ordered_extent(inode, ordered, 1);
4335                 btrfs_put_ordered_extent(ordered);
4336                 goto again;
4337         }
4338
4339         btrfs_set_extent_delalloc(inode, page_start, page_end);
4340         ret = 0;
4341
4342         /* page is wholly or partially inside EOF */
4343         if (page_start + PAGE_CACHE_SIZE > size)
4344                 zero_start = size & ~PAGE_CACHE_MASK;
4345         else
4346                 zero_start = PAGE_CACHE_SIZE;
4347
4348         if (zero_start != PAGE_CACHE_SIZE) {
4349                 kaddr = kmap(page);
4350                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
4351                 flush_dcache_page(page);
4352                 kunmap(page);
4353         }
4354         ClearPageChecked(page);
4355         set_page_dirty(page);
4356         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4357
4358 out_unlock:
4359         unlock_page(page);
4360 out:
4361         return ret;
4362 }
4363
4364 static void btrfs_truncate(struct inode *inode)
4365 {
4366         struct btrfs_root *root = BTRFS_I(inode)->root;
4367         int ret;
4368         struct btrfs_trans_handle *trans;
4369         unsigned long nr;
4370         u64 mask = root->sectorsize - 1;
4371
4372         if (!S_ISREG(inode->i_mode))
4373                 return;
4374         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4375                 return;
4376
4377         btrfs_truncate_page(inode->i_mapping, inode->i_size);
4378         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
4379
4380         trans = btrfs_start_transaction(root, 1);
4381         btrfs_set_trans_block_group(trans, inode);
4382         btrfs_i_size_write(inode, inode->i_size);
4383
4384         ret = btrfs_orphan_add(trans, inode);
4385         if (ret)
4386                 goto out;
4387         /* FIXME, add redo link to tree so we don't leak on crash */
4388         ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size,
4389                                       BTRFS_EXTENT_DATA_KEY);
4390         btrfs_update_inode(trans, root, inode);
4391
4392         ret = btrfs_orphan_del(trans, inode);
4393         BUG_ON(ret);
4394
4395 out:
4396         nr = trans->blocks_used;
4397         ret = btrfs_end_transaction_throttle(trans, root);
4398         BUG_ON(ret);
4399         btrfs_btree_balance_dirty(root, nr);
4400 }
4401
4402 /*
4403  * create a new subvolume directory/inode (helper for the ioctl).
4404  */
4405 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
4406                              struct btrfs_root *new_root, struct dentry *dentry,
4407                              u64 new_dirid, u64 alloc_hint)
4408 {
4409         struct inode *inode;
4410         int error;
4411         u64 index = 0;
4412
4413         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
4414                                 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
4415         if (IS_ERR(inode))
4416                 return PTR_ERR(inode);
4417         inode->i_op = &btrfs_dir_inode_operations;
4418         inode->i_fop = &btrfs_dir_file_operations;
4419
4420         inode->i_nlink = 1;
4421         btrfs_i_size_write(inode, 0);
4422
4423         error = btrfs_update_inode(trans, new_root, inode);
4424         if (error)
4425                 return error;
4426
4427         d_instantiate(dentry, inode);
4428         return 0;
4429 }
4430
4431 /* helper function for file defrag and space balancing.  This
4432  * forces readahead on a given range of bytes in an inode
4433  */
4434 unsigned long btrfs_force_ra(struct address_space *mapping,
4435                               struct file_ra_state *ra, struct file *file,
4436                               pgoff_t offset, pgoff_t last_index)
4437 {
4438         pgoff_t req_size = last_index - offset + 1;
4439
4440         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
4441         return offset + req_size;
4442 }
4443
4444 struct inode *btrfs_alloc_inode(struct super_block *sb)
4445 {
4446         struct btrfs_inode *ei;
4447
4448         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
4449         if (!ei)
4450                 return NULL;
4451         ei->last_trans = 0;
4452         ei->logged_trans = 0;
4453         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
4454         ei->i_acl = BTRFS_ACL_NOT_CACHED;
4455         ei->i_default_acl = BTRFS_ACL_NOT_CACHED;
4456         INIT_LIST_HEAD(&ei->i_orphan);
4457         return &ei->vfs_inode;
4458 }
4459
4460 void btrfs_destroy_inode(struct inode *inode)
4461 {
4462         struct btrfs_ordered_extent *ordered;
4463         WARN_ON(!list_empty(&inode->i_dentry));
4464         WARN_ON(inode->i_data.nrpages);
4465
4466         if (BTRFS_I(inode)->i_acl &&
4467             BTRFS_I(inode)->i_acl != BTRFS_ACL_NOT_CACHED)
4468                 posix_acl_release(BTRFS_I(inode)->i_acl);
4469         if (BTRFS_I(inode)->i_default_acl &&
4470             BTRFS_I(inode)->i_default_acl != BTRFS_ACL_NOT_CACHED)
4471                 posix_acl_release(BTRFS_I(inode)->i_default_acl);
4472
4473         spin_lock(&BTRFS_I(inode)->root->list_lock);
4474         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
4475                 printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
4476                        " list\n", inode->i_ino);
4477                 dump_stack();
4478         }
4479         spin_unlock(&BTRFS_I(inode)->root->list_lock);
4480
4481         while (1) {
4482                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
4483                 if (!ordered)
4484                         break;
4485                 else {
4486                         printk(KERN_ERR "btrfs found ordered "
4487                                "extent %llu %llu on inode cleanup\n",
4488                                (unsigned long long)ordered->file_offset,
4489                                (unsigned long long)ordered->len);
4490                         btrfs_remove_ordered_extent(inode, ordered);
4491                         btrfs_put_ordered_extent(ordered);
4492                         btrfs_put_ordered_extent(ordered);
4493                 }
4494         }
4495         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
4496         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
4497 }
4498
4499 static void init_once(void *foo)
4500 {
4501         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
4502
4503         inode_init_once(&ei->vfs_inode);
4504 }
4505
4506 void btrfs_destroy_cachep(void)
4507 {
4508         if (btrfs_inode_cachep)
4509                 kmem_cache_destroy(btrfs_inode_cachep);
4510         if (btrfs_trans_handle_cachep)
4511                 kmem_cache_destroy(btrfs_trans_handle_cachep);
4512         if (btrfs_transaction_cachep)
4513                 kmem_cache_destroy(btrfs_transaction_cachep);
4514         if (btrfs_bit_radix_cachep)
4515                 kmem_cache_destroy(btrfs_bit_radix_cachep);
4516         if (btrfs_path_cachep)
4517                 kmem_cache_destroy(btrfs_path_cachep);
4518 }
4519
4520 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
4521                                        unsigned long extra_flags,
4522                                        void (*ctor)(void *))
4523 {
4524         return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
4525                                  SLAB_MEM_SPREAD | extra_flags), ctor);
4526 }
4527
4528 int btrfs_init_cachep(void)
4529 {
4530         btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
4531                                           sizeof(struct btrfs_inode),
4532                                           0, init_once);
4533         if (!btrfs_inode_cachep)
4534                 goto fail;
4535         btrfs_trans_handle_cachep =
4536                         btrfs_cache_create("btrfs_trans_handle_cache",
4537                                            sizeof(struct btrfs_trans_handle),
4538                                            0, NULL);
4539         if (!btrfs_trans_handle_cachep)
4540                 goto fail;
4541         btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
4542                                              sizeof(struct btrfs_transaction),
4543                                              0, NULL);
4544         if (!btrfs_transaction_cachep)
4545                 goto fail;
4546         btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
4547                                          sizeof(struct btrfs_path),
4548                                          0, NULL);
4549         if (!btrfs_path_cachep)
4550                 goto fail;
4551         btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
4552                                               SLAB_DESTROY_BY_RCU, NULL);
4553         if (!btrfs_bit_radix_cachep)
4554                 goto fail;
4555         return 0;
4556 fail:
4557         btrfs_destroy_cachep();
4558         return -ENOMEM;
4559 }
4560
4561 static int btrfs_getattr(struct vfsmount *mnt,
4562                          struct dentry *dentry, struct kstat *stat)
4563 {
4564         struct inode *inode = dentry->d_inode;
4565         generic_fillattr(inode, stat);
4566         stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
4567         stat->blksize = PAGE_CACHE_SIZE;
4568         stat->blocks = (inode_get_bytes(inode) +
4569                         BTRFS_I(inode)->delalloc_bytes) >> 9;
4570         return 0;
4571 }
4572
4573 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4574                            struct inode *new_dir, struct dentry *new_dentry)
4575 {
4576         struct btrfs_trans_handle *trans;
4577         struct btrfs_root *root = BTRFS_I(old_dir)->root;
4578         struct inode *new_inode = new_dentry->d_inode;
4579         struct inode *old_inode = old_dentry->d_inode;
4580         struct timespec ctime = CURRENT_TIME;
4581         u64 index = 0;
4582         int ret;
4583
4584         /* we're not allowed to rename between subvolumes */
4585         if (BTRFS_I(old_inode)->root->root_key.objectid !=
4586             BTRFS_I(new_dir)->root->root_key.objectid)
4587                 return -EXDEV;
4588
4589         if (S_ISDIR(old_inode->i_mode) && new_inode &&
4590             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
4591                 return -ENOTEMPTY;
4592         }
4593
4594         /* to rename a snapshot or subvolume, we need to juggle the
4595          * backrefs.  This isn't coded yet
4596          */
4597         if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
4598                 return -EXDEV;
4599
4600         ret = btrfs_check_free_space(root, 1, 0);
4601         if (ret)
4602                 goto out_unlock;
4603
4604         trans = btrfs_start_transaction(root, 1);
4605
4606         btrfs_set_trans_block_group(trans, new_dir);
4607
4608         btrfs_inc_nlink(old_dentry->d_inode);
4609         old_dir->i_ctime = old_dir->i_mtime = ctime;
4610         new_dir->i_ctime = new_dir->i_mtime = ctime;
4611         old_inode->i_ctime = ctime;
4612
4613         ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode,
4614                                  old_dentry->d_name.name,
4615                                  old_dentry->d_name.len);
4616         if (ret)
4617                 goto out_fail;
4618
4619         if (new_inode) {
4620                 new_inode->i_ctime = CURRENT_TIME;
4621                 ret = btrfs_unlink_inode(trans, root, new_dir,
4622                                          new_dentry->d_inode,
4623                                          new_dentry->d_name.name,
4624                                          new_dentry->d_name.len);
4625                 if (ret)
4626                         goto out_fail;
4627                 if (new_inode->i_nlink == 0) {
4628                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
4629                         if (ret)
4630                                 goto out_fail;
4631                 }
4632
4633         }
4634         ret = btrfs_set_inode_index(new_dir, &index);
4635         if (ret)
4636                 goto out_fail;
4637
4638         ret = btrfs_add_link(trans, new_dentry->d_parent->d_inode,
4639                              old_inode, new_dentry->d_name.name,
4640                              new_dentry->d_name.len, 1, index);
4641         if (ret)
4642                 goto out_fail;
4643
4644 out_fail:
4645         btrfs_end_transaction_throttle(trans, root);
4646 out_unlock:
4647         return ret;
4648 }
4649
4650 /*
4651  * some fairly slow code that needs optimization. This walks the list
4652  * of all the inodes with pending delalloc and forces them to disk.
4653  */
4654 int btrfs_start_delalloc_inodes(struct btrfs_root *root)
4655 {
4656         struct list_head *head = &root->fs_info->delalloc_inodes;
4657         struct btrfs_inode *binode;
4658         struct inode *inode;
4659
4660         if (root->fs_info->sb->s_flags & MS_RDONLY)
4661                 return -EROFS;
4662
4663         spin_lock(&root->fs_info->delalloc_lock);
4664         while (!list_empty(head)) {
4665                 binode = list_entry(head->next, struct btrfs_inode,
4666                                     delalloc_inodes);
4667                 inode = igrab(&binode->vfs_inode);
4668                 if (!inode)
4669                         list_del_init(&binode->delalloc_inodes);
4670                 spin_unlock(&root->fs_info->delalloc_lock);
4671                 if (inode) {
4672                         filemap_flush(inode->i_mapping);
4673                         iput(inode);
4674                 }
4675                 cond_resched();
4676                 spin_lock(&root->fs_info->delalloc_lock);
4677         }
4678         spin_unlock(&root->fs_info->delalloc_lock);
4679
4680         /* the filemap_flush will queue IO into the worker threads, but
4681          * we have to make sure the IO is actually started and that
4682          * ordered extents get created before we return
4683          */
4684         atomic_inc(&root->fs_info->async_submit_draining);
4685         while (atomic_read(&root->fs_info->nr_async_submits) ||
4686               atomic_read(&root->fs_info->async_delalloc_pages)) {
4687                 wait_event(root->fs_info->async_submit_wait,
4688                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
4689                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
4690         }
4691         atomic_dec(&root->fs_info->async_submit_draining);
4692         return 0;
4693 }
4694
4695 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
4696                          const char *symname)
4697 {
4698         struct btrfs_trans_handle *trans;
4699         struct btrfs_root *root = BTRFS_I(dir)->root;
4700         struct btrfs_path *path;
4701         struct btrfs_key key;
4702         struct inode *inode = NULL;
4703         int err;
4704         int drop_inode = 0;
4705         u64 objectid;
4706         u64 index = 0 ;
4707         int name_len;
4708         int datasize;
4709         unsigned long ptr;
4710         struct btrfs_file_extent_item *ei;
4711         struct extent_buffer *leaf;
4712         unsigned long nr = 0;
4713
4714         name_len = strlen(symname) + 1;
4715         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
4716                 return -ENAMETOOLONG;
4717
4718         err = btrfs_check_free_space(root, 1, 0);
4719         if (err)
4720                 goto out_fail;
4721
4722         trans = btrfs_start_transaction(root, 1);
4723         btrfs_set_trans_block_group(trans, dir);
4724
4725         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4726         if (err) {
4727                 err = -ENOSPC;
4728                 goto out_unlock;
4729         }
4730
4731         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4732                                 dentry->d_name.len,
4733                                 dentry->d_parent->d_inode->i_ino, objectid,
4734                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
4735                                 &index);
4736         err = PTR_ERR(inode);
4737         if (IS_ERR(inode))
4738                 goto out_unlock;
4739
4740         err = btrfs_init_acl(inode, dir);
4741         if (err) {
4742                 drop_inode = 1;
4743                 goto out_unlock;
4744         }
4745
4746         btrfs_set_trans_block_group(trans, inode);
4747         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
4748         if (err)
4749                 drop_inode = 1;
4750         else {
4751                 inode->i_mapping->a_ops = &btrfs_aops;
4752                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4753                 inode->i_fop = &btrfs_file_operations;
4754                 inode->i_op = &btrfs_file_inode_operations;
4755                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4756         }
4757         dir->i_sb->s_dirt = 1;
4758         btrfs_update_inode_block_group(trans, inode);
4759         btrfs_update_inode_block_group(trans, dir);
4760         if (drop_inode)
4761                 goto out_unlock;
4762
4763         path = btrfs_alloc_path();
4764         BUG_ON(!path);
4765         key.objectid = inode->i_ino;
4766         key.offset = 0;
4767         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
4768         datasize = btrfs_file_extent_calc_inline_size(name_len);
4769         err = btrfs_insert_empty_item(trans, root, path, &key,
4770                                       datasize);
4771         if (err) {
4772                 drop_inode = 1;
4773                 goto out_unlock;
4774         }
4775         leaf = path->nodes[0];
4776         ei = btrfs_item_ptr(leaf, path->slots[0],
4777                             struct btrfs_file_extent_item);
4778         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
4779         btrfs_set_file_extent_type(leaf, ei,
4780                                    BTRFS_FILE_EXTENT_INLINE);
4781         btrfs_set_file_extent_encryption(leaf, ei, 0);
4782         btrfs_set_file_extent_compression(leaf, ei, 0);
4783         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
4784         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
4785
4786         ptr = btrfs_file_extent_inline_start(ei);
4787         write_extent_buffer(leaf, symname, ptr, name_len);
4788         btrfs_mark_buffer_dirty(leaf);
4789         btrfs_free_path(path);
4790
4791         inode->i_op = &btrfs_symlink_inode_operations;
4792         inode->i_mapping->a_ops = &btrfs_symlink_aops;
4793         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4794         inode_set_bytes(inode, name_len);
4795         btrfs_i_size_write(inode, name_len - 1);
4796         err = btrfs_update_inode(trans, root, inode);
4797         if (err)
4798                 drop_inode = 1;
4799
4800 out_unlock:
4801         nr = trans->blocks_used;
4802         btrfs_end_transaction_throttle(trans, root);
4803 out_fail:
4804         if (drop_inode) {
4805                 inode_dec_link_count(inode);
4806                 iput(inode);
4807         }
4808         btrfs_btree_balance_dirty(root, nr);
4809         return err;
4810 }
4811
4812 static int prealloc_file_range(struct inode *inode, u64 start, u64 end,
4813                                u64 alloc_hint, int mode)
4814 {
4815         struct btrfs_trans_handle *trans;
4816         struct btrfs_root *root = BTRFS_I(inode)->root;
4817         struct btrfs_key ins;
4818         u64 alloc_size;
4819         u64 cur_offset = start;
4820         u64 num_bytes = end - start;
4821         int ret = 0;
4822
4823         trans = btrfs_join_transaction(root, 1);
4824         BUG_ON(!trans);
4825         btrfs_set_trans_block_group(trans, inode);
4826
4827         while (num_bytes > 0) {
4828                 alloc_size = min(num_bytes, root->fs_info->max_extent);
4829                 ret = btrfs_reserve_extent(trans, root, alloc_size,
4830                                            root->sectorsize, 0, alloc_hint,
4831                                            (u64)-1, &ins, 1);
4832                 if (ret) {
4833                         WARN_ON(1);
4834                         goto out;
4835                 }
4836                 ret = insert_reserved_file_extent(trans, inode,
4837                                                   cur_offset, ins.objectid,
4838                                                   ins.offset, ins.offset,
4839                                                   ins.offset, 0, 0, 0,
4840                                                   BTRFS_FILE_EXTENT_PREALLOC);
4841                 BUG_ON(ret);
4842                 num_bytes -= ins.offset;
4843                 cur_offset += ins.offset;
4844                 alloc_hint = ins.objectid + ins.offset;
4845         }
4846 out:
4847         if (cur_offset > start) {
4848                 inode->i_ctime = CURRENT_TIME;
4849                 btrfs_set_flag(inode, PREALLOC);
4850                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4851                     cur_offset > i_size_read(inode))
4852                         btrfs_i_size_write(inode, cur_offset);
4853                 ret = btrfs_update_inode(trans, root, inode);
4854                 BUG_ON(ret);
4855         }
4856
4857         btrfs_end_transaction(trans, root);
4858         return ret;
4859 }
4860
4861 static long btrfs_fallocate(struct inode *inode, int mode,
4862                             loff_t offset, loff_t len)
4863 {
4864         u64 cur_offset;
4865         u64 last_byte;
4866         u64 alloc_start;
4867         u64 alloc_end;
4868         u64 alloc_hint = 0;
4869         u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
4870         struct extent_map *em;
4871         int ret;
4872
4873         alloc_start = offset & ~mask;
4874         alloc_end =  (offset + len + mask) & ~mask;
4875
4876         mutex_lock(&inode->i_mutex);
4877         if (alloc_start > inode->i_size) {
4878                 ret = btrfs_cont_expand(inode, alloc_start);
4879                 if (ret)
4880                         goto out;
4881         }
4882
4883         while (1) {
4884                 struct btrfs_ordered_extent *ordered;
4885                 lock_extent(&BTRFS_I(inode)->io_tree, alloc_start,
4886                             alloc_end - 1, GFP_NOFS);
4887                 ordered = btrfs_lookup_first_ordered_extent(inode,
4888                                                             alloc_end - 1);
4889                 if (ordered &&
4890                     ordered->file_offset + ordered->len > alloc_start &&
4891                     ordered->file_offset < alloc_end) {
4892                         btrfs_put_ordered_extent(ordered);
4893                         unlock_extent(&BTRFS_I(inode)->io_tree,
4894                                       alloc_start, alloc_end - 1, GFP_NOFS);
4895                         btrfs_wait_ordered_range(inode, alloc_start,
4896                                                  alloc_end - alloc_start);
4897                 } else {
4898                         if (ordered)
4899                                 btrfs_put_ordered_extent(ordered);
4900                         break;
4901                 }
4902         }
4903
4904         cur_offset = alloc_start;
4905         while (1) {
4906                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4907                                       alloc_end - cur_offset, 0);
4908                 BUG_ON(IS_ERR(em) || !em);
4909                 last_byte = min(extent_map_end(em), alloc_end);
4910                 last_byte = (last_byte + mask) & ~mask;
4911                 if (em->block_start == EXTENT_MAP_HOLE) {
4912                         ret = prealloc_file_range(inode, cur_offset,
4913                                         last_byte, alloc_hint, mode);
4914                         if (ret < 0) {
4915                                 free_extent_map(em);
4916                                 break;
4917                         }
4918                 }
4919                 if (em->block_start <= EXTENT_MAP_LAST_BYTE)
4920                         alloc_hint = em->block_start;
4921                 free_extent_map(em);
4922
4923                 cur_offset = last_byte;
4924                 if (cur_offset >= alloc_end) {
4925                         ret = 0;
4926                         break;
4927                 }
4928         }
4929         unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, alloc_end - 1,
4930                       GFP_NOFS);
4931 out:
4932         mutex_unlock(&inode->i_mutex);
4933         return ret;
4934 }
4935
4936 static int btrfs_set_page_dirty(struct page *page)
4937 {
4938         return __set_page_dirty_nobuffers(page);
4939 }
4940
4941 static int btrfs_permission(struct inode *inode, int mask)
4942 {
4943         if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
4944                 return -EACCES;
4945         return generic_permission(inode, mask, btrfs_check_acl);
4946 }
4947
4948 static struct inode_operations btrfs_dir_inode_operations = {
4949         .getattr        = btrfs_getattr,
4950         .lookup         = btrfs_lookup,
4951         .create         = btrfs_create,
4952         .unlink         = btrfs_unlink,
4953         .link           = btrfs_link,
4954         .mkdir          = btrfs_mkdir,
4955         .rmdir          = btrfs_rmdir,
4956         .rename         = btrfs_rename,
4957         .symlink        = btrfs_symlink,
4958         .setattr        = btrfs_setattr,
4959         .mknod          = btrfs_mknod,
4960         .setxattr       = btrfs_setxattr,
4961         .getxattr       = btrfs_getxattr,
4962         .listxattr      = btrfs_listxattr,
4963         .removexattr    = btrfs_removexattr,
4964         .permission     = btrfs_permission,
4965 };
4966 static struct inode_operations btrfs_dir_ro_inode_operations = {
4967         .lookup         = btrfs_lookup,
4968         .permission     = btrfs_permission,
4969 };
4970 static struct file_operations btrfs_dir_file_operations = {
4971         .llseek         = generic_file_llseek,
4972         .read           = generic_read_dir,
4973         .readdir        = btrfs_real_readdir,
4974         .unlocked_ioctl = btrfs_ioctl,
4975 #ifdef CONFIG_COMPAT
4976         .compat_ioctl   = btrfs_ioctl,
4977 #endif
4978         .release        = btrfs_release_file,
4979         .fsync          = btrfs_sync_file,
4980 };
4981
4982 static struct extent_io_ops btrfs_extent_io_ops = {
4983         .fill_delalloc = run_delalloc_range,
4984         .submit_bio_hook = btrfs_submit_bio_hook,
4985         .merge_bio_hook = btrfs_merge_bio_hook,
4986         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
4987         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
4988         .writepage_start_hook = btrfs_writepage_start_hook,
4989         .readpage_io_failed_hook = btrfs_io_failed_hook,
4990         .set_bit_hook = btrfs_set_bit_hook,
4991         .clear_bit_hook = btrfs_clear_bit_hook,
4992 };
4993
4994 static struct address_space_operations btrfs_aops = {
4995         .readpage       = btrfs_readpage,
4996         .writepage      = btrfs_writepage,
4997         .writepages     = btrfs_writepages,
4998         .readpages      = btrfs_readpages,
4999         .sync_page      = block_sync_page,
5000         .bmap           = btrfs_bmap,
5001         .direct_IO      = btrfs_direct_IO,
5002         .invalidatepage = btrfs_invalidatepage,
5003         .releasepage    = btrfs_releasepage,
5004         .set_page_dirty = btrfs_set_page_dirty,
5005 };
5006
5007 static struct address_space_operations btrfs_symlink_aops = {
5008         .readpage       = btrfs_readpage,
5009         .writepage      = btrfs_writepage,
5010         .invalidatepage = btrfs_invalidatepage,
5011         .releasepage    = btrfs_releasepage,
5012 };
5013
5014 static struct inode_operations btrfs_file_inode_operations = {
5015         .truncate       = btrfs_truncate,
5016         .getattr        = btrfs_getattr,
5017         .setattr        = btrfs_setattr,
5018         .setxattr       = btrfs_setxattr,
5019         .getxattr       = btrfs_getxattr,
5020         .listxattr      = btrfs_listxattr,
5021         .removexattr    = btrfs_removexattr,
5022         .permission     = btrfs_permission,
5023         .fallocate      = btrfs_fallocate,
5024 };
5025 static struct inode_operations btrfs_special_inode_operations = {
5026         .getattr        = btrfs_getattr,
5027         .setattr        = btrfs_setattr,
5028         .permission     = btrfs_permission,
5029         .setxattr       = btrfs_setxattr,
5030         .getxattr       = btrfs_getxattr,
5031         .listxattr      = btrfs_listxattr,
5032         .removexattr    = btrfs_removexattr,
5033 };
5034 static struct inode_operations btrfs_symlink_inode_operations = {
5035         .readlink       = generic_readlink,
5036         .follow_link    = page_follow_link_light,
5037         .put_link       = page_put_link,
5038         .permission     = btrfs_permission,
5039 };