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