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/xattr.h>
  38 #include <linux/posix_acl.h>
  39 #include <linux/falloc.h>
  40 #include "compat.h"
  41 #include "ctree.h"
  42 #include "disk-io.h"
  43 #include "transaction.h"
  44 #include "btrfs_inode.h"
  45 #include "ioctl.h"
  46 #include "print-tree.h"
  47 #include "volumes.h"
  48 #include "ordered-data.h"
  49 #include "xattr.h"
  50 #include "tree-log.h"
  51 #include "ref-cache.h"
  52 #include "compression.h"
  53 #include "locking.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 static int btrfs_init_inode_security(struct inode *inode,  struct inode *dir)
  95 {
  96         int err;
  97
  98         err = btrfs_init_acl(inode, dir);
  99         if (!err)
 100                 err = btrfs_xattr_security_init(inode, dir);
 101         return err;
 102 }
 103
 104 /*
 105  * this does all the hard work for inserting an inline extent into
 106  * the btree.  The caller should have done a btrfs_drop_extents so that
 107  * no overlapping inline items exist in the btree
 108  */
 109 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 110                                 struct btrfs_root *root, struct inode *inode,
 111                                 u64 start, size_t size, size_t compressed_size,
 112                                 struct page **compressed_pages)
 113 {
 114         struct btrfs_key key;
 115         struct btrfs_path *path;
 116         struct extent_buffer *leaf;
 117         struct page *page = NULL;
 118         char *kaddr;
 119         unsigned long ptr;
 120         struct btrfs_file_extent_item *ei;
 121         int err = 0;
 122         int ret;
 123         size_t cur_size = size;
 124         size_t datasize;
 125         unsigned long offset;
 126         int use_compress = 0;
 127
 128         if (compressed_size && compressed_pages) {
 129                 use_compress = 1;
 130                 cur_size = compressed_size;
 131         }
 132
 133         path = btrfs_alloc_path();
 134         if (!path)
 135                 return -ENOMEM;
 136
 137         path->leave_spinning = 1;
 138         btrfs_set_trans_block_group(trans, inode);
 139
 140         key.objectid = inode->i_ino;
 141         key.offset = start;
 142         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 143         datasize = btrfs_file_extent_calc_inline_size(cur_size);
 144
 145         inode_add_bytes(inode, size);
 146         ret = btrfs_insert_empty_item(trans, root, path, &key,
 147                                       datasize);
 148         BUG_ON(ret);
 149         if (ret) {
 150                 err = ret;
 151                 goto fail;
 152         }
 153         leaf = path->nodes[0];
 154         ei = btrfs_item_ptr(leaf, path->slots[0],
 155                             struct btrfs_file_extent_item);
 156         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 157         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 158         btrfs_set_file_extent_encryption(leaf, ei, 0);
 159         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 160         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 161         ptr = btrfs_file_extent_inline_start(ei);
 162
 163         if (use_compress) {
 164                 struct page *cpage;
 165                 int i = 0;
 166                 while (compressed_size > 0) {
 167                         cpage = compressed_pages[i];
 168                         cur_size = min_t(unsigned long, compressed_size,
 169                                        PAGE_CACHE_SIZE);
 170
 171                         kaddr = kmap_atomic(cpage, KM_USER0);
 172                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 173                         kunmap_atomic(kaddr, KM_USER0);
 174
 175                         i++;
 176                         ptr += cur_size;
 177                         compressed_size -= cur_size;
 178                 }
 179                 btrfs_set_file_extent_compression(leaf, ei,
 180                                                   BTRFS_COMPRESS_ZLIB);
 181         } else {
 182                 page = find_get_page(inode->i_mapping,
 183                                      start >> PAGE_CACHE_SHIFT);
 184                 btrfs_set_file_extent_compression(leaf, ei, 0);
 185                 kaddr = kmap_atomic(page, KM_USER0);
 186                 offset = start & (PAGE_CACHE_SIZE - 1);
 187                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 188                 kunmap_atomic(kaddr, KM_USER0);
 189                 page_cache_release(page);
 190         }
 191         btrfs_mark_buffer_dirty(leaf);
 192         btrfs_free_path(path);
 193
 194         BTRFS_I(inode)->disk_i_size = inode->i_size;
 195         btrfs_update_inode(trans, root, inode);
 196         return 0;
 197 fail:
 198         btrfs_free_path(path);
 199         return err;
 200 }
 201
 202
 203 /*
 204  * conditionally insert an inline extent into the file.  This
 205  * does the checks required to make sure the data is small enough
 206  * to fit as an inline extent.
 207  */
 208 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 209                                  struct btrfs_root *root,
 210                                  struct inode *inode, u64 start, u64 end,
 211                                  size_t compressed_size,
 212                                  struct page **compressed_pages)
 213 {
 214         u64 isize = i_size_read(inode);
 215         u64 actual_end = min(end + 1, isize);
 216         u64 inline_len = actual_end - start;
 217         u64 aligned_end = (end + root->sectorsize - 1) &
 218                         ~((u64)root->sectorsize - 1);
 219         u64 hint_byte;
 220         u64 data_len = inline_len;
 221         int ret;
 222
 223         if (compressed_size)
 224                 data_len = compressed_size;
 225
 226         if (start > 0 ||
 227             actual_end >= PAGE_CACHE_SIZE ||
 228             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 229             (!compressed_size &&
 230             (actual_end & (root->sectorsize - 1)) == 0) ||
 231             end + 1 < isize ||
 232             data_len > root->fs_info->max_inline) {
 233                 return 1;
 234         }
 235
 236         ret = btrfs_drop_extents(trans, root, inode, start,
 237                                  aligned_end, start, &hint_byte);
 238         BUG_ON(ret);
 239
 240         if (isize > actual_end)
 241                 inline_len = min_t(u64, isize, actual_end);
 242         ret = insert_inline_extent(trans, root, inode, start,
 243                                    inline_len, compressed_size,
 244                                    compressed_pages);
 245         BUG_ON(ret);
 246         btrfs_drop_extent_cache(inode, start, aligned_end, 0);
 247         return 0;
 248 }
 249
 250 struct async_extent {
 251         u64 start;
 252         u64 ram_size;
 253         u64 compressed_size;
 254         struct page **pages;
 255         unsigned long nr_pages;
 256         struct list_head list;
 257 };
 258
 259 struct async_cow {
 260         struct inode *inode;
 261         struct btrfs_root *root;
 262         struct page *locked_page;
 263         u64 start;
 264         u64 end;
 265         struct list_head extents;
 266         struct btrfs_work work;
 267 };
 268
 269 static noinline int add_async_extent(struct async_cow *cow,
 270                                      u64 start, u64 ram_size,
 271                                      u64 compressed_size,
 272                                      struct page **pages,
 273                                      unsigned long nr_pages)
 274 {
 275         struct async_extent *async_extent;
 276
 277         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 278         async_extent->start = start;
 279         async_extent->ram_size = ram_size;
 280         async_extent->compressed_size = compressed_size;
 281         async_extent->pages = pages;
 282         async_extent->nr_pages = nr_pages;
 283         list_add_tail(&async_extent->list, &cow->extents);
 284         return 0;
 285 }
 286
 287 /*
 288  * we create compressed extents in two phases.  The first
 289  * phase compresses a range of pages that have already been
 290  * locked (both pages and state bits are locked).
 291  *
 292  * This is done inside an ordered work queue, and the compression
 293  * is spread across many cpus.  The actual IO submission is step
 294  * two, and the ordered work queue takes care of making sure that
 295  * happens in the same order things were put onto the queue by
 296  * writepages and friends.
 297  *
 298  * If this code finds it can't get good compression, it puts an
 299  * entry onto the work queue to write the uncompressed bytes.  This
 300  * makes sure that both compressed inodes and uncompressed inodes
 301  * are written in the same order that pdflush sent them down.
 302  */
 303 static noinline int compress_file_range(struct inode *inode,
 304                                         struct page *locked_page,
 305                                         u64 start, u64 end,
 306                                         struct async_cow *async_cow,
 307                                         int *num_added)
 308 {
 309         struct btrfs_root *root = BTRFS_I(inode)->root;
 310         struct btrfs_trans_handle *trans;
 311         u64 num_bytes;
 312         u64 orig_start;
 313         u64 disk_num_bytes;
 314         u64 blocksize = root->sectorsize;
 315         u64 actual_end;
 316         u64 isize = i_size_read(inode);
 317         int ret = 0;
 318         struct page **pages = NULL;
 319         unsigned long nr_pages;
 320         unsigned long nr_pages_ret = 0;
 321         unsigned long total_compressed = 0;
 322         unsigned long total_in = 0;
 323         unsigned long max_compressed = 128 * 1024;
 324         unsigned long max_uncompressed = 128 * 1024;
 325         int i;
 326         int will_compress;
 327
 328         orig_start = start;
 329
 330         actual_end = min_t(u64, isize, end + 1);
 331 again:
 332         will_compress = 0;
 333         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 334         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 335
 336         /*
 337          * we don't want to send crud past the end of i_size through
 338          * compression, that's just a waste of CPU time.  So, if the
 339          * end of the file is before the start of our current
 340          * requested range of bytes, we bail out to the uncompressed
 341          * cleanup code that can deal with all of this.
 342          *
 343          * It isn't really the fastest way to fix things, but this is a
 344          * very uncommon corner.
 345          */
 346         if (actual_end <= start)
 347                 goto cleanup_and_bail_uncompressed;
 348
 349         total_compressed = actual_end - start;
 350
 351         /* we want to make sure that amount of ram required to uncompress
 352          * an extent is reasonable, so we limit the total size in ram
 353          * of a compressed extent to 128k.  This is a crucial number
 354          * because it also controls how easily we can spread reads across
 355          * cpus for decompression.
 356          *
 357          * We also want to make sure the amount of IO required to do
 358          * a random read is reasonably small, so we limit the size of
 359          * a compressed extent to 128k.
 360          */
 361         total_compressed = min(total_compressed, max_uncompressed);
 362         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 363         num_bytes = max(blocksize,  num_bytes);
 364         disk_num_bytes = num_bytes;
 365         total_in = 0;
 366         ret = 0;
 367
 368         /*
 369          * we do compression for mount -o compress and when the
 370          * inode has not been flagged as nocompress.  This flag can
 371          * change at any time if we discover bad compression ratios.
 372          */
 373         if (!btrfs_test_flag(inode, NOCOMPRESS) &&
 374             btrfs_test_opt(root, COMPRESS)) {
 375                 WARN_ON(pages);
 376                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 377
 378                 ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
 379                                                 total_compressed, pages,
 380                                                 nr_pages, &nr_pages_ret,
 381                                                 &total_in,
 382                                                 &total_compressed,
 383                                                 max_compressed);
 384
 385                 if (!ret) {
 386                         unsigned long offset = total_compressed &
 387                                 (PAGE_CACHE_SIZE - 1);
 388                         struct page *page = pages[nr_pages_ret - 1];
 389                         char *kaddr;
 390
 391                         /* zero the tail end of the last page, we might be
 392                          * sending it down to disk
 393                          */
 394                         if (offset) {
 395                                 kaddr = kmap_atomic(page, KM_USER0);
 396                                 memset(kaddr + offset, 0,
 397                                        PAGE_CACHE_SIZE - offset);
 398                                 kunmap_atomic(kaddr, KM_USER0);
 399                         }
 400                         will_compress = 1;
 401                 }
 402         }
 403         if (start == 0) {
 404                 trans = btrfs_join_transaction(root, 1);
 405                 BUG_ON(!trans);
 406                 btrfs_set_trans_block_group(trans, inode);
 407
 408                 /* lets try to make an inline extent */
 409                 if (ret || total_in < (actual_end - start)) {
 410                         /* we didn't compress the entire range, try
 411                          * to make an uncompressed inline extent.
 412                          */
 413                         ret = cow_file_range_inline(trans, root, inode,
 414                                                     start, end, 0, NULL);
 415                 } else {
 416                         /* try making a compressed inline extent */
 417                         ret = cow_file_range_inline(trans, root, inode,
 418                                                     start, end,
 419                                                     total_compressed, pages);
 420                 }
 421                 btrfs_end_transaction(trans, root);
 422                 if (ret == 0) {
 423                         /*
 424                          * inline extent creation worked, we don't need
 425                          * to create any more async work items.  Unlock
 426                          * and free up our temp pages.
 427                          */
 428                         extent_clear_unlock_delalloc(inode,
 429                                                      &BTRFS_I(inode)->io_tree,
 430                                                      start, end, NULL, 1, 0,
 431                                                      0, 1, 1, 1);
 432                         ret = 0;
 433                         goto free_pages_out;
 434                 }
 435         }
 436
 437         if (will_compress) {
 438                 /*
 439                  * we aren't doing an inline extent round the compressed size
 440                  * up to a block size boundary so the allocator does sane
 441                  * things
 442                  */
 443                 total_compressed = (total_compressed + blocksize - 1) &
 444                         ~(blocksize - 1);
 445
 446                 /*
 447                  * one last check to make sure the compression is really a
 448                  * win, compare the page count read with the blocks on disk
 449                  */
 450                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
 451                         ~(PAGE_CACHE_SIZE - 1);
 452                 if (total_compressed >= total_in) {
 453                         will_compress = 0;
 454                 } else {
 455                         disk_num_bytes = total_compressed;
 456                         num_bytes = total_in;
 457                 }
 458         }
 459         if (!will_compress && pages) {
 460                 /*
 461                  * the compression code ran but failed to make things smaller,
 462                  * free any pages it allocated and our page pointer array
 463                  */
 464                 for (i = 0; i < nr_pages_ret; i++) {
 465                         WARN_ON(pages[i]->mapping);
 466                         page_cache_release(pages[i]);
 467                 }
 468                 kfree(pages);
 469                 pages = NULL;
 470                 total_compressed = 0;
 471                 nr_pages_ret = 0;
 472
 473                 /* flag the file so we don't compress in the future */
 474                 btrfs_set_flag(inode, NOCOMPRESS);
 475         }
 476         if (will_compress) {
 477                 *num_added += 1;
 478
 479                 /* the async work queues will take care of doing actual
 480                  * allocation on disk for these compressed pages,
 481                  * and will submit them to the elevator.
 482                  */
 483                 add_async_extent(async_cow, start, num_bytes,
 484                                  total_compressed, pages, nr_pages_ret);
 485
 486                 if (start + num_bytes < end && start + num_bytes < actual_end) {
 487                         start += num_bytes;
 488                         pages = NULL;
 489                         cond_resched();
 490                         goto again;
 491                 }
 492         } else {
 493 cleanup_and_bail_uncompressed:
 494                 /*
 495                  * No compression, but we still need to write the pages in
 496                  * the file we've been given so far.  redirty the locked
 497                  * page if it corresponds to our extent and set things up
 498                  * for the async work queue to run cow_file_range to do
 499                  * the normal delalloc dance
 500                  */
 501                 if (page_offset(locked_page) >= start &&
 502                     page_offset(locked_page) <= end) {
 503                         __set_page_dirty_nobuffers(locked_page);
 504                         /* unlocked later on in the async handlers */
 505                 }
 506                 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0);
 507                 *num_added += 1;
 508         }
 509
 510 out:
 511         return 0;
 512
 513 free_pages_out:
 514         for (i = 0; i < nr_pages_ret; i++) {
 515                 WARN_ON(pages[i]->mapping);
 516                 page_cache_release(pages[i]);
 517         }
 518         kfree(pages);
 519
 520         goto out;
 521 }
 522
 523 /*
 524  * phase two of compressed writeback.  This is the ordered portion
 525  * of the code, which only gets called in the order the work was
 526  * queued.  We walk all the async extents created by compress_file_range
 527  * and send them down to the disk.
 528  */
 529 static noinline int submit_compressed_extents(struct inode *inode,
 530                                               struct async_cow *async_cow)
 531 {
 532         struct async_extent *async_extent;
 533         u64 alloc_hint = 0;
 534         struct btrfs_trans_handle *trans;
 535         struct btrfs_key ins;
 536         struct extent_map *em;
 537         struct btrfs_root *root = BTRFS_I(inode)->root;
 538         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 539         struct extent_io_tree *io_tree;
 540         int ret;
 541
 542         if (list_empty(&async_cow->extents))
 543                 return 0;
 544
 545         trans = btrfs_join_transaction(root, 1);
 546
 547         while (!list_empty(&async_cow->extents)) {
 548                 async_extent = list_entry(async_cow->extents.next,
 549                                           struct async_extent, list);
 550                 list_del(&async_extent->list);
 551
 552                 io_tree = &BTRFS_I(inode)->io_tree;
 553
 554                 /* did the compression code fall back to uncompressed IO? */
 555                 if (!async_extent->pages) {
 556                         int page_started = 0;
 557                         unsigned long nr_written = 0;
 558
 559                         lock_extent(io_tree, async_extent->start,
 560                                     async_extent->start +
 561                                     async_extent->ram_size - 1, GFP_NOFS);
 562
 563                         /* allocate blocks */
 564                         cow_file_range(inode, async_cow->locked_page,
 565                                        async_extent->start,
 566                                        async_extent->start +
 567                                        async_extent->ram_size - 1,
 568                                        &page_started, &nr_written, 0);
 569
 570                         /*
 571                          * if page_started, cow_file_range inserted an
 572                          * inline extent and took care of all the unlocking
 573                          * and IO for us.  Otherwise, we need to submit
 574                          * all those pages down to the drive.
 575                          */
 576                         if (!page_started)
 577                                 extent_write_locked_range(io_tree,
 578                                                   inode, async_extent->start,
 579                                                   async_extent->start +
 580                                                   async_extent->ram_size - 1,
 581                                                   btrfs_get_extent,
 582                                                   WB_SYNC_ALL);
 583                         kfree(async_extent);
 584                         cond_resched();
 585                         continue;
 586                 }
 587
 588                 lock_extent(io_tree, async_extent->start,
 589                             async_extent->start + async_extent->ram_size - 1,
 590                             GFP_NOFS);
 591                 /*
 592                  * here we're doing allocation and writeback of the
 593                  * compressed pages
 594                  */
 595                 btrfs_drop_extent_cache(inode, async_extent->start,
 596                                         async_extent->start +
 597                                         async_extent->ram_size - 1, 0);
 598
 599                 ret = btrfs_reserve_extent(trans, root,
 600                                            async_extent->compressed_size,
 601                                            async_extent->compressed_size,
 602                                            0, alloc_hint,
 603                                            (u64)-1, &ins, 1);
 604                 BUG_ON(ret);
 605                 em = alloc_extent_map(GFP_NOFS);
 606                 em->start = async_extent->start;
 607                 em->len = async_extent->ram_size;
 608                 em->orig_start = em->start;
 609
 610                 em->block_start = ins.objectid;
 611                 em->block_len = ins.offset;
 612                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 613                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 614                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 615
 616                 while (1) {
 617                         spin_lock(&em_tree->lock);
 618                         ret = add_extent_mapping(em_tree, em);
 619                         spin_unlock(&em_tree->lock);
 620                         if (ret != -EEXIST) {
 621                                 free_extent_map(em);
 622                                 break;
 623                         }
 624                         btrfs_drop_extent_cache(inode, async_extent->start,
 625                                                 async_extent->start +
 626                                                 async_extent->ram_size - 1, 0);
 627                 }
 628
 629                 ret = btrfs_add_ordered_extent(inode, async_extent->start,
 630                                                ins.objectid,
 631                                                async_extent->ram_size,
 632                                                ins.offset,
 633                                                BTRFS_ORDERED_COMPRESSED);
 634                 BUG_ON(ret);
 635
 636                 btrfs_end_transaction(trans, root);
 637
 638                 /*
 639                  * clear dirty, set writeback and unlock the pages.
 640                  */
 641                 extent_clear_unlock_delalloc(inode,
 642                                              &BTRFS_I(inode)->io_tree,
 643                                              async_extent->start,
 644                                              async_extent->start +
 645                                              async_extent->ram_size - 1,
 646                                              NULL, 1, 1, 0, 1, 1, 0);
 647
 648                 ret = btrfs_submit_compressed_write(inode,
 649                                     async_extent->start,
 650                                     async_extent->ram_size,
 651                                     ins.objectid,
 652                                     ins.offset, async_extent->pages,
 653                                     async_extent->nr_pages);
 654
 655                 BUG_ON(ret);
 656                 trans = btrfs_join_transaction(root, 1);
 657                 alloc_hint = ins.objectid + ins.offset;
 658                 kfree(async_extent);
 659                 cond_resched();
 660         }
 661
 662         btrfs_end_transaction(trans, root);
 663         return 0;
 664 }
 665
 666 /*
 667  * when extent_io.c finds a delayed allocation range in the file,
 668  * the call backs end up in this code.  The basic idea is to
 669  * allocate extents on disk for the range, and create ordered data structs
 670  * in ram to track those extents.
 671  *
 672  * locked_page is the page that writepage had locked already.  We use
 673  * it to make sure we don't do extra locks or unlocks.
 674  *
 675  * *page_started is set to one if we unlock locked_page and do everything
 676  * required to start IO on it.  It may be clean and already done with
 677  * IO when we return.
 678  */
 679 static noinline int cow_file_range(struct inode *inode,
 680                                    struct page *locked_page,
 681                                    u64 start, u64 end, int *page_started,
 682                                    unsigned long *nr_written,
 683                                    int unlock)
 684 {
 685         struct btrfs_root *root = BTRFS_I(inode)->root;
 686         struct btrfs_trans_handle *trans;
 687         u64 alloc_hint = 0;
 688         u64 num_bytes;
 689         unsigned long ram_size;
 690         u64 disk_num_bytes;
 691         u64 cur_alloc_size;
 692         u64 blocksize = root->sectorsize;
 693         u64 actual_end;
 694         u64 isize = i_size_read(inode);
 695         struct btrfs_key ins;
 696         struct extent_map *em;
 697         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 698         int ret = 0;
 699
 700         trans = btrfs_join_transaction(root, 1);
 701         BUG_ON(!trans);
 702         btrfs_set_trans_block_group(trans, inode);
 703
 704         actual_end = min_t(u64, isize, end + 1);
 705
 706         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 707         num_bytes = max(blocksize,  num_bytes);
 708         disk_num_bytes = num_bytes;
 709         ret = 0;
 710
 711         if (start == 0) {
 712                 /* lets try to make an inline extent */
 713                 ret = cow_file_range_inline(trans, root, inode,
 714                                             start, end, 0, NULL);
 715                 if (ret == 0) {
 716                         extent_clear_unlock_delalloc(inode,
 717                                                      &BTRFS_I(inode)->io_tree,
 718                                                      start, end, NULL, 1, 1,
 719                                                      1, 1, 1, 1);
 720                         *nr_written = *nr_written +
 721                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 722                         *page_started = 1;
 723                         ret = 0;
 724                         goto out;
 725                 }
 726         }
 727
 728         BUG_ON(disk_num_bytes >
 729                btrfs_super_total_bytes(&root->fs_info->super_copy));
 730
 731         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 732
 733         while (disk_num_bytes > 0) {
 734                 cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent);
 735                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 736                                            root->sectorsize, 0, alloc_hint,
 737                                            (u64)-1, &ins, 1);
 738                 BUG_ON(ret);
 739
 740                 em = alloc_extent_map(GFP_NOFS);
 741                 em->start = start;
 742                 em->orig_start = em->start;
 743
 744                 ram_size = ins.offset;
 745                 em->len = ins.offset;
 746
 747                 em->block_start = ins.objectid;
 748                 em->block_len = ins.offset;
 749                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 750                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 751
 752                 while (1) {
 753                         spin_lock(&em_tree->lock);
 754                         ret = add_extent_mapping(em_tree, em);
 755                         spin_unlock(&em_tree->lock);
 756                         if (ret != -EEXIST) {
 757                                 free_extent_map(em);
 758                                 break;
 759                         }
 760                         btrfs_drop_extent_cache(inode, start,
 761                                                 start + ram_size - 1, 0);
 762                 }
 763
 764                 cur_alloc_size = ins.offset;
 765                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 766                                                ram_size, cur_alloc_size, 0);
 767                 BUG_ON(ret);
 768
 769                 if (root->root_key.objectid ==
 770                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 771                         ret = btrfs_reloc_clone_csums(inode, start,
 772                                                       cur_alloc_size);
 773                         BUG_ON(ret);
 774                 }
 775
 776                 if (disk_num_bytes < cur_alloc_size)
 777                         break;
 778
 779                 /* we're not doing compressed IO, don't unlock the first
 780                  * page (which the caller expects to stay locked), don't
 781                  * clear any dirty bits and don't set any writeback bits
 782                  */
 783                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 784                                              start, start + ram_size - 1,
 785                                              locked_page, unlock, 1,
 786                                              1, 0, 0, 0);
 787                 disk_num_bytes -= cur_alloc_size;
 788                 num_bytes -= cur_alloc_size;
 789                 alloc_hint = ins.objectid + ins.offset;
 790                 start += cur_alloc_size;
 791         }
 792 out:
 793         ret = 0;
 794         btrfs_end_transaction(trans, root);
 795
 796         return ret;
 797 }
 798
 799 /*
 800  * work queue call back to started compression on a file and pages
 801  */
 802 static noinline void async_cow_start(struct btrfs_work *work)
 803 {
 804         struct async_cow *async_cow;
 805         int num_added = 0;
 806         async_cow = container_of(work, struct async_cow, work);
 807
 808         compress_file_range(async_cow->inode, async_cow->locked_page,
 809                             async_cow->start, async_cow->end, async_cow,
 810                             &num_added);
 811         if (num_added == 0)
 812                 async_cow->inode = NULL;
 813 }
 814
 815 /*
 816  * work queue call back to submit previously compressed pages
 817  */
 818 static noinline void async_cow_submit(struct btrfs_work *work)
 819 {
 820         struct async_cow *async_cow;
 821         struct btrfs_root *root;
 822         unsigned long nr_pages;
 823
 824         async_cow = container_of(work, struct async_cow, work);
 825
 826         root = async_cow->root;
 827         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
 828                 PAGE_CACHE_SHIFT;
 829
 830         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
 831
 832         if (atomic_read(&root->fs_info->async_delalloc_pages) <
 833             5 * 1042 * 1024 &&
 834             waitqueue_active(&root->fs_info->async_submit_wait))
 835                 wake_up(&root->fs_info->async_submit_wait);
 836
 837         if (async_cow->inode)
 838                 submit_compressed_extents(async_cow->inode, async_cow);
 839 }
 840
 841 static noinline void async_cow_free(struct btrfs_work *work)
 842 {
 843         struct async_cow *async_cow;
 844         async_cow = container_of(work, struct async_cow, work);
 845         kfree(async_cow);
 846 }
 847
 848 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
 849                                 u64 start, u64 end, int *page_started,
 850                                 unsigned long *nr_written)
 851 {
 852         struct async_cow *async_cow;
 853         struct btrfs_root *root = BTRFS_I(inode)->root;
 854         unsigned long nr_pages;
 855         u64 cur_end;
 856         int limit = 10 * 1024 * 1042;
 857
 858         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED |
 859                          EXTENT_DELALLOC, 1, 0, GFP_NOFS);
 860         while (start < end) {
 861                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
 862                 async_cow->inode = inode;
 863                 async_cow->root = root;
 864                 async_cow->locked_page = locked_page;
 865                 async_cow->start = start;
 866
 867                 if (btrfs_test_flag(inode, NOCOMPRESS))
 868                         cur_end = end;
 869                 else
 870                         cur_end = min(end, start + 512 * 1024 - 1);
 871
 872                 async_cow->end = cur_end;
 873                 INIT_LIST_HEAD(&async_cow->extents);
 874
 875                 async_cow->work.func = async_cow_start;
 876                 async_cow->work.ordered_func = async_cow_submit;
 877                 async_cow->work.ordered_free = async_cow_free;
 878                 async_cow->work.flags = 0;
 879
 880                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
 881                         PAGE_CACHE_SHIFT;
 882                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
 883
 884                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
 885                                    &async_cow->work);
 886
 887                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
 888                         wait_event(root->fs_info->async_submit_wait,
 889                            (atomic_read(&root->fs_info->async_delalloc_pages) <
 890                             limit));
 891                 }
 892
 893                 while (atomic_read(&root->fs_info->async_submit_draining) &&
 894                       atomic_read(&root->fs_info->async_delalloc_pages)) {
 895                         wait_event(root->fs_info->async_submit_wait,
 896                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
 897                            0));
 898                 }
 899
 900                 *nr_written += nr_pages;
 901                 start = cur_end + 1;
 902         }
 903         *page_started = 1;
 904         return 0;
 905 }
 906
 907 static noinline int csum_exist_in_range(struct btrfs_root *root,
 908                                         u64 bytenr, u64 num_bytes)
 909 {
 910         int ret;
 911         struct btrfs_ordered_sum *sums;
 912         LIST_HEAD(list);
 913
 914         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
 915                                        bytenr + num_bytes - 1, &list);
 916         if (ret == 0 && list_empty(&list))
 917                 return 0;
 918
 919         while (!list_empty(&list)) {
 920                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
 921                 list_del(&sums->list);
 922                 kfree(sums);
 923         }
 924         return 1;
 925 }
 926
 927 /*
 928  * when nowcow writeback call back.  This checks for snapshots or COW copies
 929  * of the extents that exist in the file, and COWs the file as required.
 930  *
 931  * If no cow copies or snapshots exist, we write directly to the existing
 932  * blocks on disk
 933  */
 934 static noinline int run_delalloc_nocow(struct inode *inode,
 935                                        struct page *locked_page,
 936                               u64 start, u64 end, int *page_started, int force,
 937                               unsigned long *nr_written)
 938 {
 939         struct btrfs_root *root = BTRFS_I(inode)->root;
 940         struct btrfs_trans_handle *trans;
 941         struct extent_buffer *leaf;
 942         struct btrfs_path *path;
 943         struct btrfs_file_extent_item *fi;
 944         struct btrfs_key found_key;
 945         u64 cow_start;
 946         u64 cur_offset;
 947         u64 extent_end;
 948         u64 disk_bytenr;
 949         u64 num_bytes;
 950         int extent_type;
 951         int ret;
 952         int type;
 953         int nocow;
 954         int check_prev = 1;
 955
 956         path = btrfs_alloc_path();
 957         BUG_ON(!path);
 958         trans = btrfs_join_transaction(root, 1);
 959         BUG_ON(!trans);
 960
 961         cow_start = (u64)-1;
 962         cur_offset = start;
 963         while (1) {
 964                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
 965                                                cur_offset, 0);
 966                 BUG_ON(ret < 0);
 967                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
 968                         leaf = path->nodes[0];
 969                         btrfs_item_key_to_cpu(leaf, &found_key,
 970                                               path->slots[0] - 1);
 971                         if (found_key.objectid == inode->i_ino &&
 972                             found_key.type == BTRFS_EXTENT_DATA_KEY)
 973                                 path->slots[0]--;
 974                 }
 975                 check_prev = 0;
 976 next_slot:
 977                 leaf = path->nodes[0];
 978                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 979                         ret = btrfs_next_leaf(root, path);
 980                         if (ret < 0)
 981                                 BUG_ON(1);
 982                         if (ret > 0)
 983                                 break;
 984                         leaf = path->nodes[0];
 985                 }
 986
 987                 nocow = 0;
 988                 disk_bytenr = 0;
 989                 num_bytes = 0;
 990                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 991
 992                 if (found_key.objectid > inode->i_ino ||
 993                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
 994                     found_key.offset > end)
 995                         break;
 996
 997                 if (found_key.offset > cur_offset) {
 998                         extent_end = found_key.offset;
 999                         goto out_check;
1000                 }
1001
1002                 fi = btrfs_item_ptr(leaf, path->slots[0],
1003                                     struct btrfs_file_extent_item);
1004                 extent_type = btrfs_file_extent_type(leaf, fi);
1005
1006                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1007                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1008                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1009                         extent_end = found_key.offset +
1010                                 btrfs_file_extent_num_bytes(leaf, fi);
1011                         if (extent_end <= start) {
1012                                 path->slots[0]++;
1013                                 goto next_slot;
1014                         }
1015                         if (disk_bytenr == 0)
1016                                 goto out_check;
1017                         if (btrfs_file_extent_compression(leaf, fi) ||
1018                             btrfs_file_extent_encryption(leaf, fi) ||
1019                             btrfs_file_extent_other_encoding(leaf, fi))
1020                                 goto out_check;
1021                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1022                                 goto out_check;
1023                         if (btrfs_extent_readonly(root, disk_bytenr))
1024                                 goto out_check;
1025                         if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1026                                                   disk_bytenr))
1027                                 goto out_check;
1028                         disk_bytenr += btrfs_file_extent_offset(leaf, fi);
1029                         disk_bytenr += cur_offset - found_key.offset;
1030                         num_bytes = min(end + 1, extent_end) - cur_offset;
1031                         /*
1032                          * force cow if csum exists in the range.
1033                          * this ensure that csum for a given extent are
1034                          * either valid or do not exist.
1035                          */
1036                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1037                                 goto out_check;
1038                         nocow = 1;
1039                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1040                         extent_end = found_key.offset +
1041                                 btrfs_file_extent_inline_len(leaf, fi);
1042                         extent_end = ALIGN(extent_end, root->sectorsize);
1043                 } else {
1044                         BUG_ON(1);
1045                 }
1046 out_check:
1047                 if (extent_end <= start) {
1048                         path->slots[0]++;
1049                         goto next_slot;
1050                 }
1051                 if (!nocow) {
1052                         if (cow_start == (u64)-1)
1053                                 cow_start = cur_offset;
1054                         cur_offset = extent_end;
1055                         if (cur_offset > end)
1056                                 break;
1057                         path->slots[0]++;
1058                         goto next_slot;
1059                 }
1060
1061                 btrfs_release_path(root, path);
1062                 if (cow_start != (u64)-1) {
1063                         ret = cow_file_range(inode, locked_page, cow_start,
1064                                         found_key.offset - 1, page_started,
1065                                         nr_written, 1);
1066                         BUG_ON(ret);
1067                         cow_start = (u64)-1;
1068                 }
1069
1070                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1071                         struct extent_map *em;
1072                         struct extent_map_tree *em_tree;
1073                         em_tree = &BTRFS_I(inode)->extent_tree;
1074                         em = alloc_extent_map(GFP_NOFS);
1075                         em->start = cur_offset;
1076                         em->orig_start = em->start;
1077                         em->len = num_bytes;
1078                         em->block_len = num_bytes;
1079                         em->block_start = disk_bytenr;
1080                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1081                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1082                         while (1) {
1083                                 spin_lock(&em_tree->lock);
1084                                 ret = add_extent_mapping(em_tree, em);
1085                                 spin_unlock(&em_tree->lock);
1086                                 if (ret != -EEXIST) {
1087                                         free_extent_map(em);
1088                                         break;
1089                                 }
1090                                 btrfs_drop_extent_cache(inode, em->start,
1091                                                 em->start + em->len - 1, 0);
1092                         }
1093                         type = BTRFS_ORDERED_PREALLOC;
1094                 } else {
1095                         type = BTRFS_ORDERED_NOCOW;
1096                 }
1097
1098                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1099                                                num_bytes, num_bytes, type);
1100                 BUG_ON(ret);
1101
1102                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1103                                         cur_offset, cur_offset + num_bytes - 1,
1104                                         locked_page, 1, 1, 1, 0, 0, 0);
1105                 cur_offset = extent_end;
1106                 if (cur_offset > end)
1107                         break;
1108         }
1109         btrfs_release_path(root, path);
1110
1111         if (cur_offset <= end && cow_start == (u64)-1)
1112                 cow_start = cur_offset;
1113         if (cow_start != (u64)-1) {
1114                 ret = cow_file_range(inode, locked_page, cow_start, end,
1115                                      page_started, nr_written, 1);
1116                 BUG_ON(ret);
1117         }
1118
1119         ret = btrfs_end_transaction(trans, root);
1120         BUG_ON(ret);
1121         btrfs_free_path(path);
1122         return 0;
1123 }
1124
1125 /*
1126  * extent_io.c call back to do delayed allocation processing
1127  */
1128 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1129                               u64 start, u64 end, int *page_started,
1130                               unsigned long *nr_written)
1131 {
1132         int ret;
1133         struct btrfs_root *root = BTRFS_I(inode)->root;
1134
1135         if (btrfs_test_flag(inode, NODATACOW))
1136                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1137                                          page_started, 1, nr_written);
1138         else if (btrfs_test_flag(inode, PREALLOC))
1139                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1140                                          page_started, 0, nr_written);
1141         else if (!btrfs_test_opt(root, COMPRESS))
1142                 ret = cow_file_range(inode, locked_page, start, end,
1143                                       page_started, nr_written, 1);
1144         else
1145                 ret = cow_file_range_async(inode, locked_page, start, end,
1146                                            page_started, nr_written);
1147         return ret;
1148 }
1149
1150 /*
1151  * extent_io.c set_bit_hook, used to track delayed allocation
1152  * bytes in this file, and to maintain the list of inodes that
1153  * have pending delalloc work to be done.
1154  */
1155 static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
1156                        unsigned long old, unsigned long bits)
1157 {
1158         /*
1159          * set_bit and clear bit hooks normally require _irqsave/restore
1160          * but in this case, we are only testeing for the DELALLOC
1161          * bit, which is only set or cleared with irqs on
1162          */
1163         if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1164                 struct btrfs_root *root = BTRFS_I(inode)->root;
1165                 btrfs_delalloc_reserve_space(root, inode, end - start + 1);
1166                 spin_lock(&root->fs_info->delalloc_lock);
1167                 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
1168                 root->fs_info->delalloc_bytes += end - start + 1;
1169                 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1170                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1171                                       &root->fs_info->delalloc_inodes);
1172                 }
1173                 spin_unlock(&root->fs_info->delalloc_lock);
1174         }
1175         return 0;
1176 }
1177
1178 /*
1179  * extent_io.c clear_bit_hook, see set_bit_hook for why
1180  */
1181 static int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
1182                          unsigned long old, unsigned long bits)
1183 {
1184         /*
1185          * set_bit and clear bit hooks normally require _irqsave/restore
1186          * but in this case, we are only testeing for the DELALLOC
1187          * bit, which is only set or cleared with irqs on
1188          */
1189         if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1190                 struct btrfs_root *root = BTRFS_I(inode)->root;
1191
1192                 spin_lock(&root->fs_info->delalloc_lock);
1193                 if (end - start + 1 > root->fs_info->delalloc_bytes) {
1194                         printk(KERN_INFO "btrfs warning: delalloc account "
1195                                "%llu %llu\n",
1196                                (unsigned long long)end - start + 1,
1197                                (unsigned long long)
1198                                root->fs_info->delalloc_bytes);
1199                         btrfs_delalloc_free_space(root, inode, (u64)-1);
1200                         root->fs_info->delalloc_bytes = 0;
1201                         BTRFS_I(inode)->delalloc_bytes = 0;
1202                 } else {
1203                         btrfs_delalloc_free_space(root, inode,
1204                                                   end - start + 1);
1205                         root->fs_info->delalloc_bytes -= end - start + 1;
1206                         BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
1207                 }
1208                 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
1209                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1210                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1211                 }
1212                 spin_unlock(&root->fs_info->delalloc_lock);
1213         }
1214         return 0;
1215 }
1216
1217 /*
1218  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1219  * we don't create bios that span stripes or chunks
1220  */
1221 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1222                          size_t size, struct bio *bio,
1223                          unsigned long bio_flags)
1224 {
1225         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1226         struct btrfs_mapping_tree *map_tree;
1227         u64 logical = (u64)bio->bi_sector << 9;
1228         u64 length = 0;
1229         u64 map_length;
1230         int ret;
1231
1232         if (bio_flags & EXTENT_BIO_COMPRESSED)
1233                 return 0;
1234
1235         length = bio->bi_size;
1236         map_tree = &root->fs_info->mapping_tree;
1237         map_length = length;
1238         ret = btrfs_map_block(map_tree, READ, logical,
1239                               &map_length, NULL, 0);
1240
1241         if (map_length < length + size)
1242                 return 1;
1243         return 0;
1244 }
1245
1246 /*
1247  * in order to insert checksums into the metadata in large chunks,
1248  * we wait until bio submission time.   All the pages in the bio are
1249  * checksummed and sums are attached onto the ordered extent record.
1250  *
1251  * At IO completion time the cums attached on the ordered extent record
1252  * are inserted into the btree
1253  */
1254 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1255                                     struct bio *bio, int mirror_num,
1256                                     unsigned long bio_flags)
1257 {
1258         struct btrfs_root *root = BTRFS_I(inode)->root;
1259         int ret = 0;
1260
1261         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1262         BUG_ON(ret);
1263         return 0;
1264 }
1265
1266 /*
1267  * in order to insert checksums into the metadata in large chunks,
1268  * we wait until bio submission time.   All the pages in the bio are
1269  * checksummed and sums are attached onto the ordered extent record.
1270  *
1271  * At IO completion time the cums attached on the ordered extent record
1272  * are inserted into the btree
1273  */
1274 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1275                           int mirror_num, unsigned long bio_flags)
1276 {
1277         struct btrfs_root *root = BTRFS_I(inode)->root;
1278         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1279 }
1280
1281 /*
1282  * extent_io.c submission hook. This does the right thing for csum calculation
1283  * on write, or reading the csums from the tree before a read
1284  */
1285 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1286                           int mirror_num, unsigned long bio_flags)
1287 {
1288         struct btrfs_root *root = BTRFS_I(inode)->root;
1289         int ret = 0;
1290         int skip_sum;
1291
1292         skip_sum = btrfs_test_flag(inode, NODATASUM);
1293
1294         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1295         BUG_ON(ret);
1296
1297         if (!(rw & (1 << BIO_RW))) {
1298                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1299                         return btrfs_submit_compressed_read(inode, bio,
1300                                                     mirror_num, bio_flags);
1301                 } else if (!skip_sum)
1302                         btrfs_lookup_bio_sums(root, inode, bio, NULL);
1303                 goto mapit;
1304         } else if (!skip_sum) {
1305                 /* csum items have already been cloned */
1306                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1307                         goto mapit;
1308                 /* we're doing a write, do the async checksumming */
1309                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1310                                    inode, rw, bio, mirror_num,
1311                                    bio_flags, __btrfs_submit_bio_start,
1312                                    __btrfs_submit_bio_done);
1313         }
1314
1315 mapit:
1316         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1317 }
1318
1319 /*
1320  * given a list of ordered sums record them in the inode.  This happens
1321  * at IO completion time based on sums calculated at bio submission time.
1322  */
1323 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1324                              struct inode *inode, u64 file_offset,
1325                              struct list_head *list)
1326 {
1327         struct btrfs_ordered_sum *sum;
1328
1329         btrfs_set_trans_block_group(trans, inode);
1330
1331         list_for_each_entry(sum, list, list) {
1332                 btrfs_csum_file_blocks(trans,
1333                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1334         }
1335         return 0;
1336 }
1337
1338 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end)
1339 {
1340         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1341                 WARN_ON(1);
1342         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1343                                    GFP_NOFS);
1344 }
1345
1346 /* see btrfs_writepage_start_hook for details on why this is required */
1347 struct btrfs_writepage_fixup {
1348         struct page *page;
1349         struct btrfs_work work;
1350 };
1351
1352 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1353 {
1354         struct btrfs_writepage_fixup *fixup;
1355         struct btrfs_ordered_extent *ordered;
1356         struct page *page;
1357         struct inode *inode;
1358         u64 page_start;
1359         u64 page_end;
1360
1361         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1362         page = fixup->page;
1363 again:
1364         lock_page(page);
1365         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1366                 ClearPageChecked(page);
1367                 goto out_page;
1368         }
1369
1370         inode = page->mapping->host;
1371         page_start = page_offset(page);
1372         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1373
1374         lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1375
1376         /* already ordered? We're done */
1377         if (test_range_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
1378                              EXTENT_ORDERED, 0)) {
1379                 goto out;
1380         }
1381
1382         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1383         if (ordered) {
1384                 unlock_extent(&BTRFS_I(inode)->io_tree, page_start,
1385                               page_end, GFP_NOFS);
1386                 unlock_page(page);
1387                 btrfs_start_ordered_extent(inode, ordered, 1);
1388                 goto again;
1389         }
1390
1391         btrfs_set_extent_delalloc(inode, page_start, page_end);
1392         ClearPageChecked(page);
1393 out:
1394         unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1395 out_page:
1396         unlock_page(page);
1397         page_cache_release(page);
1398 }
1399
1400 /*
1401  * There are a few paths in the higher layers of the kernel that directly
1402  * set the page dirty bit without asking the filesystem if it is a
1403  * good idea.  This causes problems because we want to make sure COW
1404  * properly happens and the data=ordered rules are followed.
1405  *
1406  * In our case any range that doesn't have the ORDERED bit set
1407  * hasn't been properly setup for IO.  We kick off an async process
1408  * to fix it up.  The async helper will wait for ordered extents, set
1409  * the delalloc bit and make it safe to write the page.
1410  */
1411 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1412 {
1413         struct inode *inode = page->mapping->host;
1414         struct btrfs_writepage_fixup *fixup;
1415         struct btrfs_root *root = BTRFS_I(inode)->root;
1416         int ret;
1417
1418         ret = test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1419                              EXTENT_ORDERED, 0);
1420         if (ret)
1421                 return 0;
1422
1423         if (PageChecked(page))
1424                 return -EAGAIN;
1425
1426         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1427         if (!fixup)
1428                 return -EAGAIN;
1429
1430         SetPageChecked(page);
1431         page_cache_get(page);
1432         fixup->work.func = btrfs_writepage_fixup_worker;
1433         fixup->page = page;
1434         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1435         return -EAGAIN;
1436 }
1437
1438 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1439                                        struct inode *inode, u64 file_pos,
1440                                        u64 disk_bytenr, u64 disk_num_bytes,
1441                                        u64 num_bytes, u64 ram_bytes,
1442                                        u8 compression, u8 encryption,
1443                                        u16 other_encoding, int extent_type)
1444 {
1445         struct btrfs_root *root = BTRFS_I(inode)->root;
1446         struct btrfs_file_extent_item *fi;
1447         struct btrfs_path *path;
1448         struct extent_buffer *leaf;
1449         struct btrfs_key ins;
1450         u64 hint;
1451         int ret;
1452
1453         path = btrfs_alloc_path();
1454         BUG_ON(!path);
1455
1456         path->leave_spinning = 1;
1457         ret = btrfs_drop_extents(trans, root, inode, file_pos,
1458                                  file_pos + num_bytes, file_pos, &hint);
1459         BUG_ON(ret);
1460
1461         ins.objectid = inode->i_ino;
1462         ins.offset = file_pos;
1463         ins.type = BTRFS_EXTENT_DATA_KEY;
1464         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1465         BUG_ON(ret);
1466         leaf = path->nodes[0];
1467         fi = btrfs_item_ptr(leaf, path->slots[0],
1468                             struct btrfs_file_extent_item);
1469         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1470         btrfs_set_file_extent_type(leaf, fi, extent_type);
1471         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1472         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1473         btrfs_set_file_extent_offset(leaf, fi, 0);
1474         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1475         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1476         btrfs_set_file_extent_compression(leaf, fi, compression);
1477         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1478         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1479
1480         btrfs_unlock_up_safe(path, 1);
1481         btrfs_set_lock_blocking(leaf);
1482
1483         btrfs_mark_buffer_dirty(leaf);
1484
1485         inode_add_bytes(inode, num_bytes);
1486         btrfs_drop_extent_cache(inode, file_pos, file_pos + num_bytes - 1, 0);
1487
1488         ins.objectid = disk_bytenr;
1489         ins.offset = disk_num_bytes;
1490         ins.type = BTRFS_EXTENT_ITEM_KEY;
1491         ret = btrfs_alloc_reserved_extent(trans, root, leaf->start,
1492                                           root->root_key.objectid,
1493                                           trans->transid, inode->i_ino, &ins);
1494         BUG_ON(ret);
1495         btrfs_free_path(path);
1496
1497         return 0;
1498 }
1499
1500 /*
1501  * helper function for btrfs_finish_ordered_io, this
1502  * just reads in some of the csum leaves to prime them into ram
1503  * before we start the transaction.  It limits the amount of btree
1504  * reads required while inside the transaction.
1505  */
1506 static noinline void reada_csum(struct btrfs_root *root,
1507                                 struct btrfs_path *path,
1508                                 struct btrfs_ordered_extent *ordered_extent)
1509 {
1510         struct btrfs_ordered_sum *sum;
1511         u64 bytenr;
1512
1513         sum = list_entry(ordered_extent->list.next, struct btrfs_ordered_sum,
1514                          list);
1515         bytenr = sum->sums[0].bytenr;
1516
1517         /*
1518          * we don't care about the results, the point of this search is
1519          * just to get the btree leaves into ram
1520          */
1521         btrfs_lookup_csum(NULL, root->fs_info->csum_root, path, bytenr, 0);
1522 }
1523
1524 /* as ordered data IO finishes, this gets called so we can finish
1525  * an ordered extent if the range of bytes in the file it covers are
1526  * fully written.
1527  */
1528 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1529 {
1530         struct btrfs_root *root = BTRFS_I(inode)->root;
1531         struct btrfs_trans_handle *trans;
1532         struct btrfs_ordered_extent *ordered_extent = NULL;
1533         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1534         struct btrfs_path *path;
1535         int compressed = 0;
1536         int ret;
1537
1538         ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
1539         if (!ret)
1540                 return 0;
1541
1542         /*
1543          * before we join the transaction, try to do some of our IO.
1544          * This will limit the amount of IO that we have to do with
1545          * the transaction running.  We're unlikely to need to do any
1546          * IO if the file extents are new, the disk_i_size checks
1547          * covers the most common case.
1548          */
1549         if (start < BTRFS_I(inode)->disk_i_size) {
1550                 path = btrfs_alloc_path();
1551                 if (path) {
1552                         ret = btrfs_lookup_file_extent(NULL, root, path,
1553                                                        inode->i_ino,
1554                                                        start, 0);
1555                         ordered_extent = btrfs_lookup_ordered_extent(inode,
1556                                                                      start);
1557                         if (!list_empty(&ordered_extent->list)) {
1558                                 btrfs_release_path(root, path);
1559                                 reada_csum(root, path, ordered_extent);
1560                         }
1561                         btrfs_free_path(path);
1562                 }
1563         }
1564
1565         trans = btrfs_join_transaction(root, 1);
1566
1567         if (!ordered_extent)
1568                 ordered_extent = btrfs_lookup_ordered_extent(inode, start);
1569         BUG_ON(!ordered_extent);
1570         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
1571                 goto nocow;
1572
1573         lock_extent(io_tree, ordered_extent->file_offset,
1574                     ordered_extent->file_offset + ordered_extent->len - 1,
1575                     GFP_NOFS);
1576
1577         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1578                 compressed = 1;
1579         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1580                 BUG_ON(compressed);
1581                 ret = btrfs_mark_extent_written(trans, root, inode,
1582                                                 ordered_extent->file_offset,
1583                                                 ordered_extent->file_offset +
1584                                                 ordered_extent->len);
1585                 BUG_ON(ret);
1586         } else {
1587                 ret = insert_reserved_file_extent(trans, inode,
1588                                                 ordered_extent->file_offset,
1589                                                 ordered_extent->start,
1590                                                 ordered_extent->disk_len,
1591                                                 ordered_extent->len,
1592                                                 ordered_extent->len,
1593                                                 compressed, 0, 0,
1594                                                 BTRFS_FILE_EXTENT_REG);
1595                 BUG_ON(ret);
1596         }
1597         unlock_extent(io_tree, ordered_extent->file_offset,
1598                     ordered_extent->file_offset + ordered_extent->len - 1,
1599                     GFP_NOFS);
1600 nocow:
1601         add_pending_csums(trans, inode, ordered_extent->file_offset,
1602                           &ordered_extent->list);
1603
1604         mutex_lock(&BTRFS_I(inode)->extent_mutex);
1605         btrfs_ordered_update_i_size(inode, ordered_extent);
1606         btrfs_update_inode(trans, root, inode);
1607         btrfs_remove_ordered_extent(inode, ordered_extent);
1608         mutex_unlock(&BTRFS_I(inode)->extent_mutex);
1609
1610         /* once for us */
1611         btrfs_put_ordered_extent(ordered_extent);
1612         /* once for the tree */
1613         btrfs_put_ordered_extent(ordered_extent);
1614
1615         btrfs_end_transaction(trans, root);
1616         return 0;
1617 }
1618
1619 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1620                                 struct extent_state *state, int uptodate)
1621 {
1622         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1623 }
1624
1625 /*
1626  * When IO fails, either with EIO or csum verification fails, we
1627  * try other mirrors that might have a good copy of the data.  This
1628  * io_failure_record is used to record state as we go through all the
1629  * mirrors.  If another mirror has good data, the page is set up to date
1630  * and things continue.  If a good mirror can't be found, the original
1631  * bio end_io callback is called to indicate things have failed.
1632  */
1633 struct io_failure_record {
1634         struct page *page;
1635         u64 start;
1636         u64 len;
1637         u64 logical;
1638         unsigned long bio_flags;
1639         int last_mirror;
1640 };
1641
1642 static int btrfs_io_failed_hook(struct bio *failed_bio,
1643                          struct page *page, u64 start, u64 end,
1644                          struct extent_state *state)
1645 {
1646         struct io_failure_record *failrec = NULL;
1647         u64 private;
1648         struct extent_map *em;
1649         struct inode *inode = page->mapping->host;
1650         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1651         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1652         struct bio *bio;
1653         int num_copies;
1654         int ret;
1655         int rw;
1656         u64 logical;
1657
1658         ret = get_state_private(failure_tree, start, &private);
1659         if (ret) {
1660                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1661                 if (!failrec)
1662                         return -ENOMEM;
1663                 failrec->start = start;
1664                 failrec->len = end - start + 1;
1665                 failrec->last_mirror = 0;
1666                 failrec->bio_flags = 0;
1667
1668                 spin_lock(&em_tree->lock);
1669                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1670                 if (em->start > start || em->start + em->len < start) {
1671                         free_extent_map(em);
1672                         em = NULL;
1673                 }
1674                 spin_unlock(&em_tree->lock);
1675
1676                 if (!em || IS_ERR(em)) {
1677                         kfree(failrec);
1678                         return -EIO;
1679                 }
1680                 logical = start - em->start;
1681                 logical = em->block_start + logical;
1682                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1683                         logical = em->block_start;
1684                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1685                 }
1686                 failrec->logical = logical;
1687                 free_extent_map(em);
1688                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1689                                 EXTENT_DIRTY, GFP_NOFS);
1690                 set_state_private(failure_tree, start,
1691                                  (u64)(unsigned long)failrec);
1692         } else {
1693                 failrec = (struct io_failure_record *)(unsigned long)private;
1694         }
1695         num_copies = btrfs_num_copies(
1696                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1697                               failrec->logical, failrec->len);
1698         failrec->last_mirror++;
1699         if (!state) {
1700                 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1701                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1702                                                     failrec->start,
1703                                                     EXTENT_LOCKED);
1704                 if (state && state->start != failrec->start)
1705                         state = NULL;
1706                 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1707         }
1708         if (!state || failrec->last_mirror > num_copies) {
1709                 set_state_private(failure_tree, failrec->start, 0);
1710                 clear_extent_bits(failure_tree, failrec->start,
1711                                   failrec->start + failrec->len - 1,
1712                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1713                 kfree(failrec);
1714                 return -EIO;
1715         }
1716         bio = bio_alloc(GFP_NOFS, 1);
1717         bio->bi_private = state;
1718         bio->bi_end_io = failed_bio->bi_end_io;
1719         bio->bi_sector = failrec->logical >> 9;
1720         bio->bi_bdev = failed_bio->bi_bdev;
1721         bio->bi_size = 0;
1722
1723         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1724         if (failed_bio->bi_rw & (1 << BIO_RW))
1725                 rw = WRITE;
1726         else
1727                 rw = READ;
1728
1729         BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1730                                                       failrec->last_mirror,
1731                                                       failrec->bio_flags);
1732         return 0;
1733 }
1734
1735 /*
1736  * each time an IO finishes, we do a fast check in the IO failure tree
1737  * to see if we need to process or clean up an io_failure_record
1738  */
1739 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1740 {
1741         u64 private;
1742         u64 private_failure;
1743         struct io_failure_record *failure;
1744         int ret;
1745
1746         private = 0;
1747         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1748                              (u64)-1, 1, EXTENT_DIRTY)) {
1749                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1750                                         start, &private_failure);
1751                 if (ret == 0) {
1752                         failure = (struct io_failure_record *)(unsigned long)
1753                                    private_failure;
1754                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1755                                           failure->start, 0);
1756                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1757                                           failure->start,
1758                                           failure->start + failure->len - 1,
1759                                           EXTENT_DIRTY | EXTENT_LOCKED,
1760                                           GFP_NOFS);
1761                         kfree(failure);
1762                 }
1763         }
1764         return 0;
1765 }
1766
1767 /*
1768  * when reads are done, we need to check csums to verify the data is correct
1769  * if there's a match, we allow the bio to finish.  If not, we go through
1770  * the io_failure_record routines to find good copies
1771  */
1772 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1773                                struct extent_state *state)
1774 {
1775         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1776         struct inode *inode = page->mapping->host;
1777         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1778         char *kaddr;
1779         u64 private = ~(u32)0;
1780         int ret;
1781         struct btrfs_root *root = BTRFS_I(inode)->root;
1782         u32 csum = ~(u32)0;
1783
1784         if (PageChecked(page)) {
1785                 ClearPageChecked(page);
1786                 goto good;
1787         }
1788         if (btrfs_test_flag(inode, NODATASUM))
1789                 return 0;
1790
1791         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1792             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1)) {
1793                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1794                                   GFP_NOFS);
1795                 return 0;
1796         }
1797
1798         if (state && state->start == start) {
1799                 private = state->private;
1800                 ret = 0;
1801         } else {
1802                 ret = get_state_private(io_tree, start, &private);
1803         }
1804         kaddr = kmap_atomic(page, KM_USER0);
1805         if (ret)
1806                 goto zeroit;
1807
1808         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
1809         btrfs_csum_final(csum, (char *)&csum);
1810         if (csum != private)
1811                 goto zeroit;
1812
1813         kunmap_atomic(kaddr, KM_USER0);
1814 good:
1815         /* if the io failure tree for this inode is non-empty,
1816          * check to see if we've recovered from a failed IO
1817          */
1818         btrfs_clean_io_failures(inode, start);
1819         return 0;
1820
1821 zeroit:
1822         printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
1823                "private %llu\n", page->mapping->host->i_ino,
1824                (unsigned long long)start, csum,
1825                (unsigned long long)private);
1826         memset(kaddr + offset, 1, end - start + 1);
1827         flush_dcache_page(page);
1828         kunmap_atomic(kaddr, KM_USER0);
1829         if (private == 0)
1830                 return 0;
1831         return -EIO;
1832 }
1833
1834 /*
1835  * This creates an orphan entry for the given inode in case something goes
1836  * wrong in the middle of an unlink/truncate.
1837  */
1838 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
1839 {
1840         struct btrfs_root *root = BTRFS_I(inode)->root;
1841         int ret = 0;
1842
1843         spin_lock(&root->list_lock);
1844
1845         /* already on the orphan list, we're good */
1846         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
1847                 spin_unlock(&root->list_lock);
1848                 return 0;
1849         }
1850
1851         list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1852
1853         spin_unlock(&root->list_lock);
1854
1855         /*
1856          * insert an orphan item to track this unlinked/truncated file
1857          */
1858         ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
1859
1860         return ret;
1861 }
1862
1863 /*
1864  * We have done the truncate/delete so we can go ahead and remove the orphan
1865  * item for this particular inode.
1866  */
1867 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
1868 {
1869         struct btrfs_root *root = BTRFS_I(inode)->root;
1870         int ret = 0;
1871
1872         spin_lock(&root->list_lock);
1873
1874         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
1875                 spin_unlock(&root->list_lock);
1876                 return 0;
1877         }
1878
1879         list_del_init(&BTRFS_I(inode)->i_orphan);
1880         if (!trans) {
1881                 spin_unlock(&root->list_lock);
1882                 return 0;
1883         }
1884
1885         spin_unlock(&root->list_lock);
1886
1887         ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
1888
1889         return ret;
1890 }
1891
1892 /*
1893  * this cleans up any orphans that may be left on the list from the last use
1894  * of this root.
1895  */
1896 void btrfs_orphan_cleanup(struct btrfs_root *root)
1897 {
1898         struct btrfs_path *path;
1899         struct extent_buffer *leaf;
1900         struct btrfs_item *item;
1901         struct btrfs_key key, found_key;
1902         struct btrfs_trans_handle *trans;
1903         struct inode *inode;
1904         int ret = 0, nr_unlink = 0, nr_truncate = 0;
1905
1906         path = btrfs_alloc_path();
1907         if (!path)
1908                 return;
1909         path->reada = -1;
1910
1911         key.objectid = BTRFS_ORPHAN_OBJECTID;
1912         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1913         key.offset = (u64)-1;
1914
1915
1916         while (1) {
1917                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1918                 if (ret < 0) {
1919                         printk(KERN_ERR "Error searching slot for orphan: %d"
1920                                "\n", ret);
1921                         break;
1922                 }
1923
1924                 /*
1925                  * if ret == 0 means we found what we were searching for, which
1926                  * is weird, but possible, so only screw with path if we didnt
1927                  * find the key and see if we have stuff that matches
1928                  */
1929                 if (ret > 0) {
1930                         if (path->slots[0] == 0)
1931                                 break;
1932                         path->slots[0]--;
1933                 }
1934
1935                 /* pull out the item */
1936                 leaf = path->nodes[0];
1937                 item = btrfs_item_nr(leaf, path->slots[0]);
1938                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1939
1940                 /* make sure the item matches what we want */
1941                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
1942                         break;
1943                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
1944                         break;
1945
1946                 /* release the path since we're done with it */
1947                 btrfs_release_path(root, path);
1948
1949                 /*
1950                  * this is where we are basically btrfs_lookup, without the
1951                  * crossing root thing.  we store the inode number in the
1952                  * offset of the orphan item.
1953                  */
1954                 inode = btrfs_iget_locked(root->fs_info->sb,
1955                                           found_key.offset, root);
1956                 if (!inode)
1957                         break;
1958
1959                 if (inode->i_state & I_NEW) {
1960                         BTRFS_I(inode)->root = root;
1961
1962                         /* have to set the location manually */
1963                         BTRFS_I(inode)->location.objectid = inode->i_ino;
1964                         BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
1965                         BTRFS_I(inode)->location.offset = 0;
1966
1967                         btrfs_read_locked_inode(inode);
1968                         unlock_new_inode(inode);
1969                 }
1970
1971                 /*
1972                  * add this inode to the orphan list so btrfs_orphan_del does
1973                  * the proper thing when we hit it
1974                  */
1975                 spin_lock(&root->list_lock);
1976                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1977                 spin_unlock(&root->list_lock);
1978
1979                 /*
1980                  * if this is a bad inode, means we actually succeeded in
1981                  * removing the inode, but not the orphan record, which means
1982                  * we need to manually delete the orphan since iput will just
1983                  * do a destroy_inode
1984                  */
1985                 if (is_bad_inode(inode)) {
1986                         trans = btrfs_start_transaction(root, 1);
1987                         btrfs_orphan_del(trans, inode);
1988                         btrfs_end_transaction(trans, root);
1989                         iput(inode);
1990                         continue;
1991                 }
1992
1993                 /* if we have links, this was a truncate, lets do that */
1994                 if (inode->i_nlink) {
1995                         nr_truncate++;
1996                         btrfs_truncate(inode);
1997                 } else {
1998                         nr_unlink++;
1999                 }
2000
2001                 /* this will do delete_inode and everything for us */
2002                 iput(inode);
2003         }
2004
2005         if (nr_unlink)
2006                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2007         if (nr_truncate)
2008                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2009
2010         btrfs_free_path(path);
2011 }
2012
2013 /*
2014  * read an inode from the btree into the in-memory inode
2015  */
2016 void btrfs_read_locked_inode(struct inode *inode)
2017 {
2018         struct btrfs_path *path;
2019         struct extent_buffer *leaf;
2020         struct btrfs_inode_item *inode_item;
2021         struct btrfs_timespec *tspec;
2022         struct btrfs_root *root = BTRFS_I(inode)->root;
2023         struct btrfs_key location;
2024         u64 alloc_group_block;
2025         u32 rdev;
2026         int ret;
2027
2028         path = btrfs_alloc_path();
2029         BUG_ON(!path);
2030         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2031
2032         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2033         if (ret)
2034                 goto make_bad;
2035
2036         leaf = path->nodes[0];
2037         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2038                                     struct btrfs_inode_item);
2039
2040         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2041         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2042         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2043         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2044         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2045
2046         tspec = btrfs_inode_atime(inode_item);
2047         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2048         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2049
2050         tspec = btrfs_inode_mtime(inode_item);
2051         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2052         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2053
2054         tspec = btrfs_inode_ctime(inode_item);
2055         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2056         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2057
2058         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2059         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2060         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2061         inode->i_generation = BTRFS_I(inode)->generation;
2062         inode->i_rdev = 0;
2063         rdev = btrfs_inode_rdev(leaf, inode_item);
2064
2065         BTRFS_I(inode)->index_cnt = (u64)-1;
2066         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2067
2068         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2069
2070         BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2071                                                 alloc_group_block, 0);
2072         btrfs_free_path(path);
2073         inode_item = NULL;
2074
2075         switch (inode->i_mode & S_IFMT) {
2076         case S_IFREG:
2077                 inode->i_mapping->a_ops = &btrfs_aops;
2078                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2079                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2080                 inode->i_fop = &btrfs_file_operations;
2081                 inode->i_op = &btrfs_file_inode_operations;
2082                 break;
2083         case S_IFDIR:
2084                 inode->i_fop = &btrfs_dir_file_operations;
2085                 if (root == root->fs_info->tree_root)
2086                         inode->i_op = &btrfs_dir_ro_inode_operations;
2087                 else
2088                         inode->i_op = &btrfs_dir_inode_operations;
2089                 break;
2090         case S_IFLNK:
2091                 inode->i_op = &btrfs_symlink_inode_operations;
2092                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2093                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2094                 break;
2095         default:
2096                 inode->i_op = &btrfs_special_inode_operations;
2097                 init_special_inode(inode, inode->i_mode, rdev);
2098                 break;
2099         }
2100         return;
2101
2102 make_bad:
2103         btrfs_free_path(path);
2104         make_bad_inode(inode);
2105 }
2106
2107 /*
2108  * given a leaf and an inode, copy the inode fields into the leaf
2109  */
2110 static void fill_inode_item(struct btrfs_trans_handle *trans,
2111                             struct extent_buffer *leaf,
2112                             struct btrfs_inode_item *item,
2113                             struct inode *inode)
2114 {
2115         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2116         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2117         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2118         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2119         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2120
2121         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2122                                inode->i_atime.tv_sec);
2123         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2124                                 inode->i_atime.tv_nsec);
2125
2126         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2127                                inode->i_mtime.tv_sec);
2128         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2129                                 inode->i_mtime.tv_nsec);
2130
2131         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2132                                inode->i_ctime.tv_sec);
2133         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2134                                 inode->i_ctime.tv_nsec);
2135
2136         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2137         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2138         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2139         btrfs_set_inode_transid(leaf, item, trans->transid);
2140         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2141         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2142         btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2143 }
2144
2145 /*
2146  * copy everything in the in-memory inode into the btree.
2147  */
2148 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2149                                 struct btrfs_root *root, struct inode *inode)
2150 {
2151         struct btrfs_inode_item *inode_item;
2152         struct btrfs_path *path;
2153         struct extent_buffer *leaf;
2154         int ret;
2155
2156         path = btrfs_alloc_path();
2157         BUG_ON(!path);
2158         path->leave_spinning = 1;
2159         ret = btrfs_lookup_inode(trans, root, path,
2160                                  &BTRFS_I(inode)->location, 1);
2161         if (ret) {
2162                 if (ret > 0)
2163                         ret = -ENOENT;
2164                 goto failed;
2165         }
2166
2167         btrfs_unlock_up_safe(path, 1);
2168         leaf = path->nodes[0];
2169         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2170                                   struct btrfs_inode_item);
2171
2172         fill_inode_item(trans, leaf, inode_item, inode);
2173         btrfs_mark_buffer_dirty(leaf);
2174         btrfs_set_inode_last_trans(trans, inode);
2175         ret = 0;
2176 failed:
2177         btrfs_free_path(path);
2178         return ret;
2179 }
2180
2181
2182 /*
2183  * unlink helper that gets used here in inode.c and in the tree logging
2184  * recovery code.  It remove a link in a directory with a given name, and
2185  * also drops the back refs in the inode to the directory
2186  */
2187 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2188                        struct btrfs_root *root,
2189                        struct inode *dir, struct inode *inode,
2190                        const char *name, int name_len)
2191 {
2192         struct btrfs_path *path;
2193         int ret = 0;
2194         struct extent_buffer *leaf;
2195         struct btrfs_dir_item *di;
2196         struct btrfs_key key;
2197         u64 index;
2198
2199         path = btrfs_alloc_path();
2200         if (!path) {
2201                 ret = -ENOMEM;
2202                 goto err;
2203         }
2204
2205         path->leave_spinning = 1;
2206         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2207                                     name, name_len, -1);
2208         if (IS_ERR(di)) {
2209                 ret = PTR_ERR(di);
2210                 goto err;
2211         }
2212         if (!di) {
2213                 ret = -ENOENT;
2214                 goto err;
2215         }
2216         leaf = path->nodes[0];
2217         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2218         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2219         if (ret)
2220                 goto err;
2221         btrfs_release_path(root, path);
2222
2223         ret = btrfs_del_inode_ref(trans, root, name, name_len,
2224                                   inode->i_ino,
2225                                   dir->i_ino, &index);
2226         if (ret) {
2227                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2228                        "inode %lu parent %lu\n", name_len, name,
2229                        inode->i_ino, dir->i_ino);
2230                 goto err;
2231         }
2232
2233         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2234                                          index, name, name_len, -1);
2235         if (IS_ERR(di)) {
2236                 ret = PTR_ERR(di);
2237                 goto err;
2238         }
2239         if (!di) {
2240                 ret = -ENOENT;
2241                 goto err;
2242         }
2243         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2244         btrfs_release_path(root, path);
2245
2246         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2247                                          inode, dir->i_ino);
2248         BUG_ON(ret != 0 && ret != -ENOENT);
2249         if (ret != -ENOENT)
2250                 BTRFS_I(dir)->log_dirty_trans = trans->transid;
2251
2252         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2253                                            dir, index);
2254         BUG_ON(ret);
2255 err:
2256         btrfs_free_path(path);
2257         if (ret)
2258                 goto out;
2259
2260         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2261         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2262         btrfs_update_inode(trans, root, dir);
2263         btrfs_drop_nlink(inode);
2264         ret = btrfs_update_inode(trans, root, inode);
2265         dir->i_sb->s_dirt = 1;
2266 out:
2267         return ret;
2268 }
2269
2270 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2271 {
2272         struct btrfs_root *root;
2273         struct btrfs_trans_handle *trans;
2274         struct inode *inode = dentry->d_inode;
2275         int ret;
2276         unsigned long nr = 0;
2277
2278         root = BTRFS_I(dir)->root;
2279
2280         trans = btrfs_start_transaction(root, 1);
2281
2282         btrfs_set_trans_block_group(trans, dir);
2283         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2284                                  dentry->d_name.name, dentry->d_name.len);
2285
2286         if (inode->i_nlink == 0)
2287                 ret = btrfs_orphan_add(trans, inode);
2288
2289         nr = trans->blocks_used;
2290
2291         btrfs_end_transaction_throttle(trans, root);
2292         btrfs_btree_balance_dirty(root, nr);
2293         return ret;
2294 }
2295
2296 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2297 {
2298         struct inode *inode = dentry->d_inode;
2299         int err = 0;
2300         int ret;
2301         struct btrfs_root *root = BTRFS_I(dir)->root;
2302         struct btrfs_trans_handle *trans;
2303         unsigned long nr = 0;
2304
2305         /*
2306          * the FIRST_FREE_OBJECTID check makes sure we don't try to rmdir
2307          * the root of a subvolume or snapshot
2308          */
2309         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2310             inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) {
2311                 return -ENOTEMPTY;
2312         }
2313
2314         trans = btrfs_start_transaction(root, 1);
2315         btrfs_set_trans_block_group(trans, dir);
2316
2317         err = btrfs_orphan_add(trans, inode);
2318         if (err)
2319                 goto fail_trans;
2320
2321         /* now the directory is empty */
2322         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2323                                  dentry->d_name.name, dentry->d_name.len);
2324         if (!err)
2325                 btrfs_i_size_write(inode, 0);
2326
2327 fail_trans:
2328         nr = trans->blocks_used;
2329         ret = btrfs_end_transaction_throttle(trans, root);
2330         btrfs_btree_balance_dirty(root, nr);
2331
2332         if (ret && !err)
2333                 err = ret;
2334         return err;
2335 }
2336
2337 #if 0
2338 /*
2339  * when truncating bytes in a file, it is possible to avoid reading
2340  * the leaves that contain only checksum items.  This can be the
2341  * majority of the IO required to delete a large file, but it must
2342  * be done carefully.
2343  *
2344  * The keys in the level just above the leaves are checked to make sure
2345  * the lowest key in a given leaf is a csum key, and starts at an offset
2346  * after the new  size.
2347  *
2348  * Then the key for the next leaf is checked to make sure it also has
2349  * a checksum item for the same file.  If it does, we know our target leaf
2350  * contains only checksum items, and it can be safely freed without reading
2351  * it.
2352  *
2353  * This is just an optimization targeted at large files.  It may do
2354  * nothing.  It will return 0 unless things went badly.
2355  */
2356 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
2357                                      struct btrfs_root *root,
2358                                      struct btrfs_path *path,
2359                                      struct inode *inode, u64 new_size)
2360 {
2361         struct btrfs_key key;
2362         int ret;
2363         int nritems;
2364         struct btrfs_key found_key;
2365         struct btrfs_key other_key;
2366         struct btrfs_leaf_ref *ref;
2367         u64 leaf_gen;
2368         u64 leaf_start;
2369
2370         path->lowest_level = 1;
2371         key.objectid = inode->i_ino;
2372         key.type = BTRFS_CSUM_ITEM_KEY;
2373         key.offset = new_size;
2374 again:
2375         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2376         if (ret < 0)
2377                 goto out;
2378
2379         if (path->nodes[1] == NULL) {
2380                 ret = 0;
2381                 goto out;
2382         }
2383         ret = 0;
2384         btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
2385         nritems = btrfs_header_nritems(path->nodes[1]);
2386
2387         if (!nritems)
2388                 goto out;
2389
2390         if (path->slots[1] >= nritems)
2391                 goto next_node;
2392
2393         /* did we find a key greater than anything we want to delete? */
2394         if (found_key.objectid > inode->i_ino ||
2395            (found_key.objectid == inode->i_ino && found_key.type > key.type))
2396                 goto out;
2397
2398         /* we check the next key in the node to make sure the leave contains
2399          * only checksum items.  This comparison doesn't work if our
2400          * leaf is the last one in the node
2401          */
2402         if (path->slots[1] + 1 >= nritems) {
2403 next_node:
2404                 /* search forward from the last key in the node, this
2405                  * will bring us into the next node in the tree
2406                  */
2407                 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
2408
2409                 /* unlikely, but we inc below, so check to be safe */
2410                 if (found_key.offset == (u64)-1)
2411                         goto out;
2412
2413                 /* search_forward needs a path with locks held, do the
2414                  * search again for the original key.  It is possible
2415                  * this will race with a balance and return a path that
2416                  * we could modify, but this drop is just an optimization
2417                  * and is allowed to miss some leaves.
2418                  */
2419                 btrfs_release_path(root, path);
2420                 found_key.offset++;
2421
2422                 /* setup a max key for search_forward */
2423                 other_key.offset = (u64)-1;
2424                 other_key.type = key.type;
2425                 other_key.objectid = key.objectid;
2426
2427                 path->keep_locks = 1;
2428                 ret = btrfs_search_forward(root, &found_key, &other_key,
2429                                            path, 0, 0);
2430                 path->keep_locks = 0;
2431                 if (ret || found_key.objectid != key.objectid ||
2432                     found_key.type != key.type) {
2433                         ret = 0;
2434                         goto out;
2435                 }
2436
2437                 key.offset = found_key.offset;
2438                 btrfs_release_path(root, path);
2439                 cond_resched();
2440                 goto again;
2441         }
2442
2443         /* we know there's one more slot after us in the tree,
2444          * read that key so we can verify it is also a checksum item
2445          */
2446         btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
2447
2448         if (found_key.objectid < inode->i_ino)
2449                 goto next_key;
2450
2451         if (found_key.type != key.type || found_key.offset < new_size)
2452                 goto next_key;
2453
2454         /*
2455          * if the key for the next leaf isn't a csum key from this objectid,
2456          * we can't be sure there aren't good items inside this leaf.
2457          * Bail out
2458          */
2459         if (other_key.objectid != inode->i_ino || other_key.type != key.type)
2460                 goto out;
2461
2462         leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
2463         leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
2464         /*
2465          * it is safe to delete this leaf, it contains only
2466          * csum items from this inode at an offset >= new_size
2467          */
2468         ret = btrfs_del_leaf(trans, root, path, leaf_start);
2469         BUG_ON(ret);
2470
2471         if (root->ref_cows && leaf_gen < trans->transid) {
2472                 ref = btrfs_alloc_leaf_ref(root, 0);
2473                 if (ref) {
2474                         ref->root_gen = root->root_key.offset;
2475                         ref->bytenr = leaf_start;
2476                         ref->owner = 0;
2477                         ref->generation = leaf_gen;
2478                         ref->nritems = 0;
2479
2480                         btrfs_sort_leaf_ref(ref);
2481
2482                         ret = btrfs_add_leaf_ref(root, ref, 0);
2483                         WARN_ON(ret);
2484                         btrfs_free_leaf_ref(root, ref);
2485                 } else {
2486                         WARN_ON(1);
2487                 }
2488         }
2489 next_key:
2490         btrfs_release_path(root, path);
2491
2492         if (other_key.objectid == inode->i_ino &&
2493             other_key.type == key.type && other_key.offset > key.offset) {
2494                 key.offset = other_key.offset;
2495                 cond_resched();
2496                 goto again;
2497         }
2498         ret = 0;
2499 out:
2500         /* fixup any changes we've made to the path */
2501         path->lowest_level = 0;
2502         path->keep_locks = 0;
2503         btrfs_release_path(root, path);
2504         return ret;
2505 }
2506
2507 #endif
2508
2509 /*
2510  * this can truncate away extent items, csum items and directory items.
2511  * It starts at a high offset and removes keys until it can't find
2512  * any higher than new_size
2513  *
2514  * csum items that cross the new i_size are truncated to the new size
2515  * as well.
2516  *
2517  * min_type is the minimum key type to truncate down to.  If set to 0, this
2518  * will kill all the items on this inode, including the INODE_ITEM_KEY.
2519  */
2520 noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2521                                         struct btrfs_root *root,
2522                                         struct inode *inode,
2523                                         u64 new_size, u32 min_type)
2524 {
2525         int ret;
2526         struct btrfs_path *path;
2527         struct btrfs_key key;
2528         struct btrfs_key found_key;
2529         u32 found_type = (u8)-1;
2530         struct extent_buffer *leaf;
2531         struct btrfs_file_extent_item *fi;
2532         u64 extent_start = 0;
2533         u64 extent_num_bytes = 0;
2534         u64 item_end = 0;
2535         u64 root_gen = 0;
2536         u64 root_owner = 0;
2537         int found_extent;
2538         int del_item;
2539         int pending_del_nr = 0;
2540         int pending_del_slot = 0;
2541         int extent_type = -1;
2542         int encoding;
2543         u64 mask = root->sectorsize - 1;
2544
2545         if (root->ref_cows)
2546                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
2547         path = btrfs_alloc_path();
2548         path->reada = -1;
2549         BUG_ON(!path);
2550
2551         /* FIXME, add redo link to tree so we don't leak on crash */
2552         key.objectid = inode->i_ino;
2553         key.offset = (u64)-1;
2554         key.type = (u8)-1;
2555
2556 search_again:
2557         path->leave_spinning = 1;
2558         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2559         if (ret < 0)
2560                 goto error;
2561
2562         if (ret > 0) {
2563                 /* there are no items in the tree for us to truncate, we're
2564                  * done
2565                  */
2566                 if (path->slots[0] == 0) {
2567                         ret = 0;
2568                         goto error;
2569                 }
2570                 path->slots[0]--;
2571         }
2572
2573         while (1) {
2574                 fi = NULL;
2575                 leaf = path->nodes[0];
2576                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2577                 found_type = btrfs_key_type(&found_key);
2578                 encoding = 0;
2579
2580                 if (found_key.objectid != inode->i_ino)
2581                         break;
2582
2583                 if (found_type < min_type)
2584                         break;
2585
2586                 item_end = found_key.offset;
2587                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
2588                         fi = btrfs_item_ptr(leaf, path->slots[0],
2589                                             struct btrfs_file_extent_item);
2590                         extent_type = btrfs_file_extent_type(leaf, fi);
2591                         encoding = btrfs_file_extent_compression(leaf, fi);
2592                         encoding |= btrfs_file_extent_encryption(leaf, fi);
2593                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
2594
2595                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2596                                 item_end +=
2597                                     btrfs_file_extent_num_bytes(leaf, fi);
2598                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2599                                 item_end += btrfs_file_extent_inline_len(leaf,
2600                                                                          fi);
2601                         }
2602                         item_end--;
2603                 }
2604                 if (item_end < new_size) {
2605                         if (found_type == BTRFS_DIR_ITEM_KEY)
2606                                 found_type = BTRFS_INODE_ITEM_KEY;
2607                         else if (found_type == BTRFS_EXTENT_ITEM_KEY)
2608                                 found_type = BTRFS_EXTENT_DATA_KEY;
2609                         else if (found_type == BTRFS_EXTENT_DATA_KEY)
2610                                 found_type = BTRFS_XATTR_ITEM_KEY;
2611                         else if (found_type == BTRFS_XATTR_ITEM_KEY)
2612                                 found_type = BTRFS_INODE_REF_KEY;
2613                         else if (found_type)
2614                                 found_type--;
2615                         else
2616                                 break;
2617                         btrfs_set_key_type(&key, found_type);
2618                         goto next;
2619                 }
2620                 if (found_key.offset >= new_size)
2621                         del_item = 1;
2622                 else
2623                         del_item = 0;
2624                 found_extent = 0;
2625
2626                 /* FIXME, shrink the extent if the ref count is only 1 */
2627                 if (found_type != BTRFS_EXTENT_DATA_KEY)
2628                         goto delete;
2629
2630                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2631                         u64 num_dec;
2632                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
2633                         if (!del_item && !encoding) {
2634                                 u64 orig_num_bytes =
2635                                         btrfs_file_extent_num_bytes(leaf, fi);
2636                                 extent_num_bytes = new_size -
2637                                         found_key.offset + root->sectorsize - 1;
2638                                 extent_num_bytes = extent_num_bytes &
2639                                         ~((u64)root->sectorsize - 1);
2640                                 btrfs_set_file_extent_num_bytes(leaf, fi,
2641                                                          extent_num_bytes);
2642                                 num_dec = (orig_num_bytes -
2643                                            extent_num_bytes);
2644                                 if (root->ref_cows && extent_start != 0)
2645                                         inode_sub_bytes(inode, num_dec);
2646                                 btrfs_mark_buffer_dirty(leaf);
2647                         } else {
2648                                 extent_num_bytes =
2649                                         btrfs_file_extent_disk_num_bytes(leaf,
2650                                                                          fi);
2651                                 /* FIXME blocksize != 4096 */
2652                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
2653                                 if (extent_start != 0) {
2654                                         found_extent = 1;
2655                                         if (root->ref_cows)
2656                                                 inode_sub_bytes(inode, num_dec);
2657                                 }
2658                                 root_gen = btrfs_header_generation(leaf);
2659                                 root_owner = btrfs_header_owner(leaf);
2660                         }
2661                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2662                         /*
2663                          * we can't truncate inline items that have had
2664                          * special encodings
2665                          */
2666                         if (!del_item &&
2667                             btrfs_file_extent_compression(leaf, fi) == 0 &&
2668                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
2669                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
2670                                 u32 size = new_size - found_key.offset;
2671
2672                                 if (root->ref_cows) {
2673                                         inode_sub_bytes(inode, item_end + 1 -
2674                                                         new_size);
2675                                 }
2676                                 size =
2677                                     btrfs_file_extent_calc_inline_size(size);
2678                                 ret = btrfs_truncate_item(trans, root, path,
2679                                                           size, 1);
2680                                 BUG_ON(ret);
2681                         } else if (root->ref_cows) {
2682                                 inode_sub_bytes(inode, item_end + 1 -
2683                                                 found_key.offset);
2684                         }
2685                 }
2686 delete:
2687                 if (del_item) {
2688                         if (!pending_del_nr) {
2689                                 /* no pending yet, add ourselves */
2690                                 pending_del_slot = path->slots[0];
2691                                 pending_del_nr = 1;
2692                         } else if (pending_del_nr &&
2693                                    path->slots[0] + 1 == pending_del_slot) {
2694                                 /* hop on the pending chunk */
2695                                 pending_del_nr++;
2696                                 pending_del_slot = path->slots[0];
2697                         } else {
2698                                 BUG();
2699                         }
2700                 } else {
2701                         break;
2702                 }
2703                 if (found_extent) {
2704                         btrfs_set_path_blocking(path);
2705                         ret = btrfs_free_extent(trans, root, extent_start,
2706                                                 extent_num_bytes,
2707                                                 leaf->start, root_owner,
2708                                                 root_gen, inode->i_ino, 0);
2709                         BUG_ON(ret);
2710                 }
2711 next:
2712                 if (path->slots[0] == 0) {
2713                         if (pending_del_nr)
2714                                 goto del_pending;
2715                         btrfs_release_path(root, path);
2716                         if (found_type == BTRFS_INODE_ITEM_KEY)
2717                                 break;
2718                         goto search_again;
2719                 }
2720
2721                 path->slots[0]--;
2722                 if (pending_del_nr &&
2723                     path->slots[0] + 1 != pending_del_slot) {
2724                         struct btrfs_key debug;
2725 del_pending:
2726                         btrfs_item_key_to_cpu(path->nodes[0], &debug,
2727                                               pending_del_slot);
2728                         ret = btrfs_del_items(trans, root, path,
2729                                               pending_del_slot,
2730                                               pending_del_nr);
2731                         BUG_ON(ret);
2732                         pending_del_nr = 0;
2733                         btrfs_release_path(root, path);
2734                         if (found_type == BTRFS_INODE_ITEM_KEY)
2735                                 break;
2736                         goto search_again;
2737                 }
2738         }
2739         ret = 0;
2740 error:
2741         if (pending_del_nr) {
2742                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
2743                                       pending_del_nr);
2744         }
2745         btrfs_free_path(path);
2746         inode->i_sb->s_dirt = 1;
2747         return ret;
2748 }
2749
2750 /*
2751  * taken from block_truncate_page, but does cow as it zeros out
2752  * any bytes left in the last page in the file.
2753  */
2754 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
2755 {
2756         struct inode *inode = mapping->host;
2757         struct btrfs_root *root = BTRFS_I(inode)->root;
2758         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2759         struct btrfs_ordered_extent *ordered;
2760         char *kaddr;
2761         u32 blocksize = root->sectorsize;
2762         pgoff_t index = from >> PAGE_CACHE_SHIFT;
2763         unsigned offset = from & (PAGE_CACHE_SIZE-1);
2764         struct page *page;
2765         int ret = 0;
2766         u64 page_start;
2767         u64 page_end;
2768
2769         if ((offset & (blocksize - 1)) == 0)
2770                 goto out;
2771
2772         ret = -ENOMEM;
2773 again:
2774         page = grab_cache_page(mapping, index);
2775         if (!page)
2776                 goto out;
2777
2778         page_start = page_offset(page);
2779         page_end = page_start + PAGE_CACHE_SIZE - 1;
2780
2781         if (!PageUptodate(page)) {
2782                 ret = btrfs_readpage(NULL, page);
2783                 lock_page(page);
2784                 if (page->mapping != mapping) {
2785                         unlock_page(page);
2786                         page_cache_release(page);
2787                         goto again;
2788                 }
2789                 if (!PageUptodate(page)) {
2790                         ret = -EIO;
2791                         goto out_unlock;
2792                 }
2793         }
2794         wait_on_page_writeback(page);
2795
2796         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2797         set_page_extent_mapped(page);
2798
2799         ordered = btrfs_lookup_ordered_extent(inode, page_start);
2800         if (ordered) {
2801                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2802                 unlock_page(page);
2803                 page_cache_release(page);
2804                 btrfs_start_ordered_extent(inode, ordered, 1);
2805                 btrfs_put_ordered_extent(ordered);
2806                 goto again;
2807         }
2808
2809         btrfs_set_extent_delalloc(inode, page_start, page_end);
2810         ret = 0;
2811         if (offset != PAGE_CACHE_SIZE) {
2812                 kaddr = kmap(page);
2813                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
2814                 flush_dcache_page(page);
2815                 kunmap(page);
2816         }
2817         ClearPageChecked(page);
2818         set_page_dirty(page);
2819         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2820
2821 out_unlock:
2822         unlock_page(page);
2823         page_cache_release(page);
2824 out:
2825         return ret;
2826 }
2827
2828 int btrfs_cont_expand(struct inode *inode, loff_t size)
2829 {
2830         struct btrfs_trans_handle *trans;
2831         struct btrfs_root *root = BTRFS_I(inode)->root;
2832         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2833         struct extent_map *em;
2834         u64 mask = root->sectorsize - 1;
2835         u64 hole_start = (inode->i_size + mask) & ~mask;
2836         u64 block_end = (size + mask) & ~mask;
2837         u64 last_byte;
2838         u64 cur_offset;
2839         u64 hole_size;
2840         int err;
2841
2842         if (size <= hole_start)
2843                 return 0;
2844
2845         err = btrfs_check_metadata_free_space(root);
2846         if (err)
2847                 return err;
2848
2849         btrfs_truncate_page(inode->i_mapping, inode->i_size);
2850
2851         while (1) {
2852                 struct btrfs_ordered_extent *ordered;
2853                 btrfs_wait_ordered_range(inode, hole_start,
2854                                          block_end - hole_start);
2855                 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2856                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
2857                 if (!ordered)
2858                         break;
2859                 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2860                 btrfs_put_ordered_extent(ordered);
2861         }
2862
2863         trans = btrfs_start_transaction(root, 1);
2864         btrfs_set_trans_block_group(trans, inode);
2865
2866         cur_offset = hole_start;
2867         while (1) {
2868                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2869                                 block_end - cur_offset, 0);
2870                 BUG_ON(IS_ERR(em) || !em);
2871                 last_byte = min(extent_map_end(em), block_end);
2872                 last_byte = (last_byte + mask) & ~mask;
2873                 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
2874                         u64 hint_byte = 0;
2875                         hole_size = last_byte - cur_offset;
2876                         err = btrfs_drop_extents(trans, root, inode,
2877                                                  cur_offset,
2878                                                  cur_offset + hole_size,
2879                                                  cur_offset, &hint_byte);
2880                         if (err)
2881                                 break;
2882                         err = btrfs_insert_file_extent(trans, root,
2883                                         inode->i_ino, cur_offset, 0,
2884                                         0, hole_size, 0, hole_size,
2885                                         0, 0, 0);
2886                         btrfs_drop_extent_cache(inode, hole_start,
2887                                         last_byte - 1, 0);
2888                 }
2889                 free_extent_map(em);
2890                 cur_offset = last_byte;
2891                 if (err || cur_offset >= block_end)
2892                         break;
2893         }
2894
2895         btrfs_end_transaction(trans, root);
2896         unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2897         return err;
2898 }
2899
2900 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
2901 {
2902         struct inode *inode = dentry->d_inode;
2903         int err;
2904
2905         err = inode_change_ok(inode, attr);
2906         if (err)
2907                 return err;
2908
2909         if (S_ISREG(inode->i_mode) &&
2910             attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
2911                 err = btrfs_cont_expand(inode, attr->ia_size);
2912                 if (err)
2913                         return err;
2914         }
2915
2916         err = inode_setattr(inode, attr);
2917
2918         if (!err && ((attr->ia_valid & ATTR_MODE)))
2919                 err = btrfs_acl_chmod(inode);
2920         return err;
2921 }
2922
2923 void btrfs_delete_inode(struct inode *inode)
2924 {
2925         struct btrfs_trans_handle *trans;
2926         struct btrfs_root *root = BTRFS_I(inode)->root;
2927         unsigned long nr;
2928         int ret;
2929
2930         truncate_inode_pages(&inode->i_data, 0);
2931         if (is_bad_inode(inode)) {
2932                 btrfs_orphan_del(NULL, inode);
2933                 goto no_delete;
2934         }
2935         btrfs_wait_ordered_range(inode, 0, (u64)-1);
2936
2937         btrfs_i_size_write(inode, 0);
2938         trans = btrfs_join_transaction(root, 1);
2939
2940         btrfs_set_trans_block_group(trans, inode);
2941         ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0);
2942         if (ret) {
2943                 btrfs_orphan_del(NULL, inode);
2944                 goto no_delete_lock;
2945         }
2946
2947         btrfs_orphan_del(trans, inode);
2948
2949         nr = trans->blocks_used;
2950         clear_inode(inode);
2951
2952         btrfs_end_transaction(trans, root);
2953         btrfs_btree_balance_dirty(root, nr);
2954         return;
2955
2956 no_delete_lock:
2957         nr = trans->blocks_used;
2958         btrfs_end_transaction(trans, root);
2959         btrfs_btree_balance_dirty(root, nr);
2960 no_delete:
2961         clear_inode(inode);
2962 }
2963
2964 /*
2965  * this returns the key found in the dir entry in the location pointer.
2966  * If no dir entries were found, location->objectid is 0.
2967  */
2968 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
2969                                struct btrfs_key *location)
2970 {
2971         const char *name = dentry->d_name.name;
2972         int namelen = dentry->d_name.len;
2973         struct btrfs_dir_item *di;
2974         struct btrfs_path *path;
2975         struct btrfs_root *root = BTRFS_I(dir)->root;
2976         int ret = 0;
2977
2978         path = btrfs_alloc_path();
2979         BUG_ON(!path);
2980
2981         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
2982                                     namelen, 0);
2983         if (IS_ERR(di))
2984                 ret = PTR_ERR(di);
2985
2986         if (!di || IS_ERR(di))
2987                 goto out_err;
2988
2989         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
2990 out:
2991         btrfs_free_path(path);
2992         return ret;
2993 out_err:
2994         location->objectid = 0;
2995         goto out;
2996 }
2997
2998 /*
2999  * when we hit a tree root in a directory, the btrfs part of the inode
3000  * needs to be changed to reflect the root directory of the tree root.  This
3001  * is kind of like crossing a mount point.
3002  */
3003 static int fixup_tree_root_location(struct btrfs_root *root,
3004                              struct btrfs_key *location,
3005                              struct btrfs_root **sub_root,
3006                              struct dentry *dentry)
3007 {
3008         struct btrfs_root_item *ri;
3009
3010         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
3011                 return 0;
3012         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
3013                 return 0;
3014
3015         *sub_root = btrfs_read_fs_root(root->fs_info, location,
3016                                         dentry->d_name.name,
3017                                         dentry->d_name.len);
3018         if (IS_ERR(*sub_root))
3019                 return PTR_ERR(*sub_root);
3020
3021         ri = &(*sub_root)->root_item;
3022         location->objectid = btrfs_root_dirid(ri);
3023         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
3024         location->offset = 0;
3025
3026         return 0;
3027 }
3028
3029 static noinline void init_btrfs_i(struct inode *inode)
3030 {
3031         struct btrfs_inode *bi = BTRFS_I(inode);
3032
3033         bi->i_acl = NULL;
3034         bi->i_default_acl = NULL;
3035
3036         bi->generation = 0;
3037         bi->sequence = 0;
3038         bi->last_trans = 0;
3039         bi->logged_trans = 0;
3040         bi->delalloc_bytes = 0;
3041         bi->reserved_bytes = 0;
3042         bi->disk_i_size = 0;
3043         bi->flags = 0;
3044         bi->index_cnt = (u64)-1;
3045         bi->log_dirty_trans = 0;
3046         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
3047         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
3048                              inode->i_mapping, GFP_NOFS);
3049         extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
3050                              inode->i_mapping, GFP_NOFS);
3051         INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
3052         btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
3053         mutex_init(&BTRFS_I(inode)->extent_mutex);
3054         mutex_init(&BTRFS_I(inode)->log_mutex);
3055 }
3056
3057 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3058 {
3059         struct btrfs_iget_args *args = p;
3060         inode->i_ino = args->ino;
3061         init_btrfs_i(inode);
3062         BTRFS_I(inode)->root = args->root;
3063         btrfs_set_inode_space_info(args->root, inode);
3064         return 0;
3065 }
3066
3067 static int btrfs_find_actor(struct inode *inode, void *opaque)
3068 {
3069         struct btrfs_iget_args *args = opaque;
3070         return args->ino == inode->i_ino &&
3071                 args->root == BTRFS_I(inode)->root;
3072 }
3073
3074 struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
3075                             struct btrfs_root *root, int wait)
3076 {
3077         struct inode *inode;
3078         struct btrfs_iget_args args;
3079         args.ino = objectid;
3080         args.root = root;
3081
3082         if (wait) {
3083                 inode = ilookup5(s, objectid, btrfs_find_actor,
3084                                  (void *)&args);
3085         } else {
3086                 inode = ilookup5_nowait(s, objectid, btrfs_find_actor,
3087                                         (void *)&args);
3088         }
3089         return inode;
3090 }
3091
3092 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
3093                                 struct btrfs_root *root)
3094 {
3095         struct inode *inode;
3096         struct btrfs_iget_args args;
3097         args.ino = objectid;
3098         args.root = root;
3099
3100         inode = iget5_locked(s, objectid, btrfs_find_actor,
3101                              btrfs_init_locked_inode,
3102                              (void *)&args);
3103         return inode;
3104 }
3105
3106 /* Get an inode object given its location and corresponding root.
3107  * Returns in *is_new if the inode was read from disk
3108  */
3109 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3110                          struct btrfs_root *root, int *is_new)
3111 {
3112         struct inode *inode;
3113
3114         inode = btrfs_iget_locked(s, location->objectid, root);
3115         if (!inode)
3116                 return ERR_PTR(-EACCES);
3117
3118         if (inode->i_state & I_NEW) {
3119                 BTRFS_I(inode)->root = root;
3120                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3121                 btrfs_read_locked_inode(inode);
3122                 unlock_new_inode(inode);
3123                 if (is_new)
3124                         *is_new = 1;
3125         } else {
3126                 if (is_new)
3127                         *is_new = 0;
3128         }
3129
3130         return inode;
3131 }
3132
3133 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3134 {
3135         struct inode *inode;
3136         struct btrfs_inode *bi = BTRFS_I(dir);
3137         struct btrfs_root *root = bi->root;
3138         struct btrfs_root *sub_root = root;
3139         struct btrfs_key location;
3140         int ret, new;
3141
3142         if (dentry->d_name.len > BTRFS_NAME_LEN)
3143                 return ERR_PTR(-ENAMETOOLONG);
3144
3145         ret = btrfs_inode_by_name(dir, dentry, &location);
3146
3147         if (ret < 0)
3148                 return ERR_PTR(ret);
3149
3150         inode = NULL;
3151         if (location.objectid) {
3152                 ret = fixup_tree_root_location(root, &location, &sub_root,
3153                                                 dentry);
3154                 if (ret < 0)
3155                         return ERR_PTR(ret);
3156                 if (ret > 0)
3157                         return ERR_PTR(-ENOENT);
3158                 inode = btrfs_iget(dir->i_sb, &location, sub_root, &new);
3159                 if (IS_ERR(inode))
3160                         return ERR_CAST(inode);
3161         }
3162         return inode;
3163 }
3164
3165 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3166                                    struct nameidata *nd)
3167 {
3168         struct inode *inode;
3169
3170         if (dentry->d_name.len > BTRFS_NAME_LEN)
3171                 return ERR_PTR(-ENAMETOOLONG);
3172
3173         inode = btrfs_lookup_dentry(dir, dentry);
3174         if (IS_ERR(inode))
3175                 return ERR_CAST(inode);
3176
3177         return d_splice_alias(inode, dentry);
3178 }
3179
3180 static unsigned char btrfs_filetype_table[] = {
3181         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
3182 };
3183
3184 static int btrfs_real_readdir(struct file *filp, void *dirent,
3185                               filldir_t filldir)
3186 {
3187         struct inode *inode = filp->f_dentry->d_inode;
3188         struct btrfs_root *root = BTRFS_I(inode)->root;
3189         struct btrfs_item *item;
3190         struct btrfs_dir_item *di;
3191         struct btrfs_key key;
3192         struct btrfs_key found_key;
3193         struct btrfs_path *path;
3194         int ret;
3195         u32 nritems;
3196         struct extent_buffer *leaf;
3197         int slot;
3198         int advance;
3199         unsigned char d_type;
3200         int over = 0;
3201         u32 di_cur;
3202         u32 di_total;
3203         u32 di_len;
3204         int key_type = BTRFS_DIR_INDEX_KEY;
3205         char tmp_name[32];
3206         char *name_ptr;
3207         int name_len;
3208
3209         /* FIXME, use a real flag for deciding about the key type */
3210         if (root->fs_info->tree_root == root)
3211                 key_type = BTRFS_DIR_ITEM_KEY;
3212
3213         /* special case for "." */
3214         if (filp->f_pos == 0) {
3215                 over = filldir(dirent, ".", 1,
3216                                1, inode->i_ino,
3217                                DT_DIR);
3218                 if (over)
3219                         return 0;
3220                 filp->f_pos = 1;
3221         }
3222         /* special case for .., just use the back ref */
3223         if (filp->f_pos == 1) {
3224                 u64 pino = parent_ino(filp->f_path.dentry);
3225                 over = filldir(dirent, "..", 2,
3226                                2, pino, DT_DIR);
3227                 if (over)
3228                         return 0;
3229                 filp->f_pos = 2;
3230         }
3231         path = btrfs_alloc_path();
3232         path->reada = 2;
3233
3234         btrfs_set_key_type(&key, key_type);
3235         key.offset = filp->f_pos;
3236         key.objectid = inode->i_ino;
3237
3238         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3239         if (ret < 0)
3240                 goto err;
3241         advance = 0;
3242
3243         while (1) {
3244                 leaf = path->nodes[0];
3245                 nritems = btrfs_header_nritems(leaf);
3246                 slot = path->slots[0];
3247                 if (advance || slot >= nritems) {
3248                         if (slot >= nritems - 1) {
3249                                 ret = btrfs_next_leaf(root, path);
3250                                 if (ret)
3251                                         break;
3252                                 leaf = path->nodes[0];
3253                                 nritems = btrfs_header_nritems(leaf);
3254                                 slot = path->slots[0];
3255                         } else {
3256                                 slot++;
3257                                 path->slots[0]++;
3258                         }
3259                 }
3260
3261                 advance = 1;
3262                 item = btrfs_item_nr(leaf, slot);
3263                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3264
3265                 if (found_key.objectid != key.objectid)
3266                         break;
3267                 if (btrfs_key_type(&found_key) != key_type)
3268                         break;
3269                 if (found_key.offset < filp->f_pos)
3270                         continue;
3271
3272                 filp->f_pos = found_key.offset;
3273
3274                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
3275                 di_cur = 0;
3276                 di_total = btrfs_item_size(leaf, item);
3277
3278                 while (di_cur < di_total) {
3279                         struct btrfs_key location;
3280
3281                         name_len = btrfs_dir_name_len(leaf, di);
3282                         if (name_len <= sizeof(tmp_name)) {
3283                                 name_ptr = tmp_name;
3284                         } else {
3285                                 name_ptr = kmalloc(name_len, GFP_NOFS);
3286                                 if (!name_ptr) {
3287                                         ret = -ENOMEM;
3288                                         goto err;
3289                                 }
3290                         }
3291                         read_extent_buffer(leaf, name_ptr,
3292                                            (unsigned long)(di + 1), name_len);
3293
3294                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
3295                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
3296
3297                         /* is this a reference to our own snapshot? If so
3298                          * skip it
3299                          */
3300                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
3301                             location.objectid == root->root_key.objectid) {
3302                                 over = 0;
3303                                 goto skip;
3304                         }
3305                         over = filldir(dirent, name_ptr, name_len,
3306                                        found_key.offset, location.objectid,
3307                                        d_type);
3308
3309 skip:
3310                         if (name_ptr != tmp_name)
3311                                 kfree(name_ptr);
3312
3313                         if (over)
3314                                 goto nopos;
3315                         di_len = btrfs_dir_name_len(leaf, di) +
3316                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
3317                         di_cur += di_len;
3318                         di = (struct btrfs_dir_item *)((char *)di + di_len);
3319                 }
3320         }
3321
3322         /* Reached end of directory/root. Bump pos past the last item. */
3323         if (key_type == BTRFS_DIR_INDEX_KEY)
3324                 filp->f_pos = INT_LIMIT(off_t);
3325         else
3326                 filp->f_pos++;
3327 nopos:
3328         ret = 0;
3329 err:
3330         btrfs_free_path(path);
3331         return ret;
3332 }
3333
3334 int btrfs_write_inode(struct inode *inode, int wait)
3335 {
3336         struct btrfs_root *root = BTRFS_I(inode)->root;
3337         struct btrfs_trans_handle *trans;
3338         int ret = 0;
3339
3340         if (root->fs_info->btree_inode == inode)
3341                 return 0;
3342
3343         if (wait) {
3344                 trans = btrfs_join_transaction(root, 1);
3345                 btrfs_set_trans_block_group(trans, inode);
3346                 ret = btrfs_commit_transaction(trans, root);
3347         }
3348         return ret;
3349 }
3350
3351 /*
3352  * This is somewhat expensive, updating the tree every time the
3353  * inode changes.  But, it is most likely to find the inode in cache.
3354  * FIXME, needs more benchmarking...there are no reasons other than performance
3355  * to keep or drop this code.
3356  */
3357 void btrfs_dirty_inode(struct inode *inode)
3358 {
3359         struct btrfs_root *root = BTRFS_I(inode)->root;
3360         struct btrfs_trans_handle *trans;
3361
3362         trans = btrfs_join_transaction(root, 1);
3363         btrfs_set_trans_block_group(trans, inode);
3364         btrfs_update_inode(trans, root, inode);
3365         btrfs_end_transaction(trans, root);
3366 }
3367
3368 /*
3369  * find the highest existing sequence number in a directory
3370  * and then set the in-memory index_cnt variable to reflect
3371  * free sequence numbers
3372  */
3373 static int btrfs_set_inode_index_count(struct inode *inode)
3374 {
3375         struct btrfs_root *root = BTRFS_I(inode)->root;
3376         struct btrfs_key key, found_key;
3377         struct btrfs_path *path;
3378         struct extent_buffer *leaf;
3379         int ret;
3380
3381         key.objectid = inode->i_ino;
3382         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
3383         key.offset = (u64)-1;
3384
3385         path = btrfs_alloc_path();
3386         if (!path)
3387                 return -ENOMEM;
3388
3389         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3390         if (ret < 0)
3391                 goto out;
3392         /* FIXME: we should be able to handle this */
3393         if (ret == 0)
3394                 goto out;
3395         ret = 0;
3396
3397         /*
3398          * MAGIC NUMBER EXPLANATION:
3399          * since we search a directory based on f_pos we have to start at 2
3400          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
3401          * else has to start at 2
3402          */
3403         if (path->slots[0] == 0) {
3404                 BTRFS_I(inode)->index_cnt = 2;
3405                 goto out;
3406         }
3407
3408         path->slots[0]--;
3409
3410         leaf = path->nodes[0];
3411         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3412
3413         if (found_key.objectid != inode->i_ino ||
3414             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
3415                 BTRFS_I(inode)->index_cnt = 2;
3416                 goto out;
3417         }
3418
3419         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
3420 out:
3421         btrfs_free_path(path);
3422         return ret;
3423 }
3424
3425 /*
3426  * helper to find a free sequence number in a given directory.  This current
3427  * code is very simple, later versions will do smarter things in the btree
3428  */
3429 int btrfs_set_inode_index(struct inode *dir, u64 *index)
3430 {
3431         int ret = 0;
3432
3433         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
3434                 ret = btrfs_set_inode_index_count(dir);
3435                 if (ret)
3436                         return ret;
3437         }
3438
3439         *index = BTRFS_I(dir)->index_cnt;
3440         BTRFS_I(dir)->index_cnt++;
3441
3442         return ret;
3443 }
3444
3445 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
3446                                      struct btrfs_root *root,
3447                                      struct inode *dir,
3448                                      const char *name, int name_len,
3449                                      u64 ref_objectid, u64 objectid,
3450                                      u64 alloc_hint, int mode, u64 *index)
3451 {
3452         struct inode *inode;
3453         struct btrfs_inode_item *inode_item;
3454         struct btrfs_key *location;
3455         struct btrfs_path *path;
3456         struct btrfs_inode_ref *ref;
3457         struct btrfs_key key[2];
3458         u32 sizes[2];
3459         unsigned long ptr;
3460         int ret;
3461         int owner;
3462
3463         path = btrfs_alloc_path();
3464         BUG_ON(!path);
3465
3466         inode = new_inode(root->fs_info->sb);
3467         if (!inode)
3468                 return ERR_PTR(-ENOMEM);
3469
3470         if (dir) {
3471                 ret = btrfs_set_inode_index(dir, index);
3472                 if (ret)
3473                         return ERR_PTR(ret);
3474         }
3475         /*
3476          * index_cnt is ignored for everything but a dir,
3477          * btrfs_get_inode_index_count has an explanation for the magic
3478          * number
3479          */
3480         init_btrfs_i(inode);
3481         BTRFS_I(inode)->index_cnt = 2;
3482         BTRFS_I(inode)->root = root;
3483         BTRFS_I(inode)->generation = trans->transid;
3484         btrfs_set_inode_space_info(root, inode);
3485
3486         if (mode & S_IFDIR)
3487                 owner = 0;
3488         else
3489                 owner = 1;
3490         BTRFS_I(inode)->block_group =
3491                         btrfs_find_block_group(root, 0, alloc_hint, owner);
3492         if ((mode & S_IFREG)) {
3493                 if (btrfs_test_opt(root, NODATASUM))
3494                         btrfs_set_flag(inode, NODATASUM);
3495                 if (btrfs_test_opt(root, NODATACOW))
3496                         btrfs_set_flag(inode, NODATACOW);
3497         }
3498
3499         key[0].objectid = objectid;
3500         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
3501         key[0].offset = 0;
3502
3503         key[1].objectid = objectid;
3504         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
3505         key[1].offset = ref_objectid;
3506
3507         sizes[0] = sizeof(struct btrfs_inode_item);
3508         sizes[1] = name_len + sizeof(*ref);
3509
3510         path->leave_spinning = 1;
3511         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
3512         if (ret != 0)
3513                 goto fail;
3514
3515         if (objectid > root->highest_inode)
3516                 root->highest_inode = objectid;
3517
3518         inode->i_uid = current_fsuid();
3519
3520         if (dir && (dir->i_mode & S_ISGID)) {
3521                 inode->i_gid = dir->i_gid;
3522                 if (S_ISDIR(mode))
3523                         mode |= S_ISGID;
3524         } else
3525                 inode->i_gid = current_fsgid();
3526
3527         inode->i_mode = mode;
3528         inode->i_ino = objectid;
3529         inode_set_bytes(inode, 0);
3530         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3531         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3532                                   struct btrfs_inode_item);
3533         fill_inode_item(trans, path->nodes[0], inode_item, inode);
3534
3535         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3536                              struct btrfs_inode_ref);
3537         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
3538         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
3539         ptr = (unsigned long)(ref + 1);
3540         write_extent_buffer(path->nodes[0], name, ptr, name_len);
3541
3542         btrfs_mark_buffer_dirty(path->nodes[0]);
3543         btrfs_free_path(path);
3544
3545         location = &BTRFS_I(inode)->location;
3546         location->objectid = objectid;
3547         location->offset = 0;
3548         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
3549
3550         insert_inode_hash(inode);
3551         return inode;
3552 fail:
3553         if (dir)
3554                 BTRFS_I(dir)->index_cnt--;
3555         btrfs_free_path(path);
3556         return ERR_PTR(ret);
3557 }
3558
3559 static inline u8 btrfs_inode_type(struct inode *inode)
3560 {
3561         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
3562 }
3563
3564 /*
3565  * utility function to add 'inode' into 'parent_inode' with
3566  * a give name and a given sequence number.
3567  * if 'add_backref' is true, also insert a backref from the
3568  * inode to the parent directory.
3569  */
3570 int btrfs_add_link(struct btrfs_trans_handle *trans,
3571                    struct inode *parent_inode, struct inode *inode,
3572                    const char *name, int name_len, int add_backref, u64 index)
3573 {
3574         int ret;
3575         struct btrfs_key key;
3576         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
3577
3578         key.objectid = inode->i_ino;
3579         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
3580         key.offset = 0;
3581
3582         ret = btrfs_insert_dir_item(trans, root, name, name_len,
3583                                     parent_inode->i_ino,
3584                                     &key, btrfs_inode_type(inode),
3585                                     index);
3586         if (ret == 0) {
3587                 if (add_backref) {
3588                         ret = btrfs_insert_inode_ref(trans, root,
3589                                                      name, name_len,
3590                                                      inode->i_ino,
3591                                                      parent_inode->i_ino,
3592                                                      index);
3593                 }
3594                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
3595                                    name_len * 2);
3596                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
3597                 ret = btrfs_update_inode(trans, root, parent_inode);
3598         }
3599         return ret;
3600 }
3601
3602 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
3603                             struct dentry *dentry, struct inode *inode,
3604                             int backref, u64 index)
3605 {
3606         int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3607                                  inode, dentry->d_name.name,
3608                                  dentry->d_name.len, backref, index);
3609         if (!err) {
3610                 d_instantiate(dentry, inode);
3611                 return 0;
3612         }
3613         if (err > 0)
3614                 err = -EEXIST;
3615         return err;
3616 }
3617
3618 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
3619                         int mode, dev_t rdev)
3620 {
3621         struct btrfs_trans_handle *trans;
3622         struct btrfs_root *root = BTRFS_I(dir)->root;
3623         struct inode *inode = NULL;
3624         int err;
3625         int drop_inode = 0;
3626         u64 objectid;
3627         unsigned long nr = 0;
3628         u64 index = 0;
3629
3630         if (!new_valid_dev(rdev))
3631                 return -EINVAL;
3632
3633         err = btrfs_check_metadata_free_space(root);
3634         if (err)
3635                 goto fail;
3636
3637         trans = btrfs_start_transaction(root, 1);
3638         btrfs_set_trans_block_group(trans, dir);
3639
3640         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3641         if (err) {
3642                 err = -ENOSPC;
3643                 goto out_unlock;
3644         }
3645
3646         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3647                                 dentry->d_name.len,
3648                                 dentry->d_parent->d_inode->i_ino, objectid,
3649                                 BTRFS_I(dir)->block_group, mode, &index);
3650         err = PTR_ERR(inode);
3651         if (IS_ERR(inode))
3652                 goto out_unlock;
3653
3654         err = btrfs_init_inode_security(inode, dir);
3655         if (err) {
3656                 drop_inode = 1;
3657                 goto out_unlock;
3658         }
3659
3660         btrfs_set_trans_block_group(trans, inode);
3661         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3662         if (err)
3663                 drop_inode = 1;
3664         else {
3665                 inode->i_op = &btrfs_special_inode_operations;
3666                 init_special_inode(inode, inode->i_mode, rdev);
3667                 btrfs_update_inode(trans, root, inode);
3668         }
3669         dir->i_sb->s_dirt = 1;
3670         btrfs_update_inode_block_group(trans, inode);
3671         btrfs_update_inode_block_group(trans, dir);
3672 out_unlock:
3673         nr = trans->blocks_used;
3674         btrfs_end_transaction_throttle(trans, root);
3675 fail:
3676         if (drop_inode) {
3677                 inode_dec_link_count(inode);
3678                 iput(inode);
3679         }
3680         btrfs_btree_balance_dirty(root, nr);
3681         return err;
3682 }
3683
3684 static int btrfs_create(struct inode *dir, struct dentry *dentry,
3685                         int mode, struct nameidata *nd)
3686 {
3687         struct btrfs_trans_handle *trans;
3688         struct btrfs_root *root = BTRFS_I(dir)->root;
3689         struct inode *inode = NULL;
3690         int err;
3691         int drop_inode = 0;
3692         unsigned long nr = 0;
3693         u64 objectid;
3694         u64 index = 0;
3695
3696         err = btrfs_check_metadata_free_space(root);
3697         if (err)
3698                 goto fail;
3699         trans = btrfs_start_transaction(root, 1);
3700         btrfs_set_trans_block_group(trans, dir);
3701
3702         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3703         if (err) {
3704                 err = -ENOSPC;
3705                 goto out_unlock;
3706         }
3707
3708         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3709                                 dentry->d_name.len,
3710                                 dentry->d_parent->d_inode->i_ino,
3711                                 objectid, BTRFS_I(dir)->block_group, mode,
3712                                 &index);
3713         err = PTR_ERR(inode);
3714         if (IS_ERR(inode))
3715                 goto out_unlock;
3716
3717         err = btrfs_init_inode_security(inode, dir);
3718         if (err) {
3719                 drop_inode = 1;
3720                 goto out_unlock;
3721         }
3722
3723         btrfs_set_trans_block_group(trans, inode);
3724         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3725         if (err)
3726                 drop_inode = 1;
3727         else {
3728                 inode->i_mapping->a_ops = &btrfs_aops;
3729                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3730                 inode->i_fop = &btrfs_file_operations;
3731                 inode->i_op = &btrfs_file_inode_operations;
3732                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3733         }
3734         dir->i_sb->s_dirt = 1;
3735         btrfs_update_inode_block_group(trans, inode);
3736         btrfs_update_inode_block_group(trans, dir);
3737 out_unlock:
3738         nr = trans->blocks_used;
3739         btrfs_end_transaction_throttle(trans, root);
3740 fail:
3741         if (drop_inode) {
3742                 inode_dec_link_count(inode);
3743                 iput(inode);
3744         }
3745         btrfs_btree_balance_dirty(root, nr);
3746         return err;
3747 }
3748
3749 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
3750                       struct dentry *dentry)
3751 {
3752         struct btrfs_trans_handle *trans;
3753         struct btrfs_root *root = BTRFS_I(dir)->root;
3754         struct inode *inode = old_dentry->d_inode;
3755         u64 index;
3756         unsigned long nr = 0;
3757         int err;
3758         int drop_inode = 0;
3759
3760         if (inode->i_nlink == 0)
3761                 return -ENOENT;
3762
3763         btrfs_inc_nlink(inode);
3764         err = btrfs_check_metadata_free_space(root);
3765         if (err)
3766                 goto fail;
3767         err = btrfs_set_inode_index(dir, &index);
3768         if (err)
3769                 goto fail;
3770
3771         trans = btrfs_start_transaction(root, 1);
3772
3773         btrfs_set_trans_block_group(trans, dir);
3774         atomic_inc(&inode->i_count);
3775
3776         err = btrfs_add_nondir(trans, dentry, inode, 1, index);
3777
3778         if (err)
3779                 drop_inode = 1;
3780
3781         dir->i_sb->s_dirt = 1;
3782         btrfs_update_inode_block_group(trans, dir);
3783         err = btrfs_update_inode(trans, root, inode);
3784
3785         if (err)
3786                 drop_inode = 1;
3787
3788         nr = trans->blocks_used;
3789         btrfs_end_transaction_throttle(trans, root);
3790 fail:
3791         if (drop_inode) {
3792                 inode_dec_link_count(inode);
3793                 iput(inode);
3794         }
3795         btrfs_btree_balance_dirty(root, nr);
3796         return err;
3797 }
3798
3799 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
3800 {
3801         struct inode *inode = NULL;
3802         struct btrfs_trans_handle *trans;
3803         struct btrfs_root *root = BTRFS_I(dir)->root;
3804         int err = 0;
3805         int drop_on_err = 0;
3806         u64 objectid = 0;
3807         u64 index = 0;
3808         unsigned long nr = 1;
3809
3810         err = btrfs_check_metadata_free_space(root);
3811         if (err)
3812                 goto out_unlock;
3813
3814         trans = btrfs_start_transaction(root, 1);
3815         btrfs_set_trans_block_group(trans, dir);
3816
3817         if (IS_ERR(trans)) {
3818                 err = PTR_ERR(trans);
3819                 goto out_unlock;
3820         }
3821
3822         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3823         if (err) {
3824                 err = -ENOSPC;
3825                 goto out_unlock;
3826         }
3827
3828         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3829                                 dentry->d_name.len,
3830                                 dentry->d_parent->d_inode->i_ino, objectid,
3831                                 BTRFS_I(dir)->block_group, S_IFDIR | mode,
3832                                 &index);
3833         if (IS_ERR(inode)) {
3834                 err = PTR_ERR(inode);
3835                 goto out_fail;
3836         }
3837
3838         drop_on_err = 1;
3839
3840         err = btrfs_init_inode_security(inode, dir);
3841         if (err)
3842                 goto out_fail;
3843
3844         inode->i_op = &btrfs_dir_inode_operations;
3845         inode->i_fop = &btrfs_dir_file_operations;
3846         btrfs_set_trans_block_group(trans, inode);
3847
3848         btrfs_i_size_write(inode, 0);
3849         err = btrfs_update_inode(trans, root, inode);
3850         if (err)
3851                 goto out_fail;
3852
3853         err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3854                                  inode, dentry->d_name.name,
3855                                  dentry->d_name.len, 0, index);
3856         if (err)
3857                 goto out_fail;
3858
3859         d_instantiate(dentry, inode);
3860         drop_on_err = 0;
3861         dir->i_sb->s_dirt = 1;
3862         btrfs_update_inode_block_group(trans, inode);
3863         btrfs_update_inode_block_group(trans, dir);
3864
3865 out_fail:
3866         nr = trans->blocks_used;
3867         btrfs_end_transaction_throttle(trans, root);
3868
3869 out_unlock:
3870         if (drop_on_err)
3871                 iput(inode);
3872         btrfs_btree_balance_dirty(root, nr);
3873         return err;
3874 }
3875
3876 /* helper for btfs_get_extent.  Given an existing extent in the tree,
3877  * and an extent that you want to insert, deal with overlap and insert
3878  * the new extent into the tree.
3879  */
3880 static int merge_extent_mapping(struct extent_map_tree *em_tree,
3881                                 struct extent_map *existing,
3882                                 struct extent_map *em,
3883                                 u64 map_start, u64 map_len)
3884 {
3885         u64 start_diff;
3886
3887         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
3888         start_diff = map_start - em->start;
3889         em->start = map_start;
3890         em->len = map_len;
3891         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
3892             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
3893                 em->block_start += start_diff;
3894                 em->block_len -= start_diff;
3895         }
3896         return add_extent_mapping(em_tree, em);
3897 }
3898
3899 static noinline int uncompress_inline(struct btrfs_path *path,
3900                                       struct inode *inode, struct page *page,
3901                                       size_t pg_offset, u64 extent_offset,
3902                                       struct btrfs_file_extent_item *item)
3903 {
3904         int ret;
3905         struct extent_buffer *leaf = path->nodes[0];
3906         char *tmp;
3907         size_t max_size;
3908         unsigned long inline_size;
3909         unsigned long ptr;
3910
3911         WARN_ON(pg_offset != 0);
3912         max_size = btrfs_file_extent_ram_bytes(leaf, item);
3913         inline_size = btrfs_file_extent_inline_item_len(leaf,
3914                                         btrfs_item_nr(leaf, path->slots[0]));
3915         tmp = kmalloc(inline_size, GFP_NOFS);
3916         ptr = btrfs_file_extent_inline_start(item);
3917
3918         read_extent_buffer(leaf, tmp, ptr, inline_size);
3919
3920         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
3921         ret = btrfs_zlib_decompress(tmp, page, extent_offset,
3922                                     inline_size, max_size);
3923         if (ret) {
3924                 char *kaddr = kmap_atomic(page, KM_USER0);
3925                 unsigned long copy_size = min_t(u64,
3926                                   PAGE_CACHE_SIZE - pg_offset,
3927                                   max_size - extent_offset);
3928                 memset(kaddr + pg_offset, 0, copy_size);
3929                 kunmap_atomic(kaddr, KM_USER0);
3930         }
3931         kfree(tmp);
3932         return 0;
3933 }
3934
3935 /*
3936  * a bit scary, this does extent mapping from logical file offset to the disk.
3937  * the ugly parts come from merging extents from the disk with the in-ram
3938  * representation.  This gets more complex because of the data=ordered code,
3939  * where the in-ram extents might be locked pending data=ordered completion.
3940  *
3941  * This also copies inline extents directly into the page.
3942  */
3943
3944 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
3945                                     size_t pg_offset, u64 start, u64 len,
3946                                     int create)
3947 {
3948         int ret;
3949         int err = 0;
3950         u64 bytenr;
3951         u64 extent_start = 0;
3952         u64 extent_end = 0;
3953         u64 objectid = inode->i_ino;
3954         u32 found_type;
3955         struct btrfs_path *path = NULL;
3956         struct btrfs_root *root = BTRFS_I(inode)->root;
3957         struct btrfs_file_extent_item *item;
3958         struct extent_buffer *leaf;
3959         struct btrfs_key found_key;
3960         struct extent_map *em = NULL;
3961         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3962         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3963         struct btrfs_trans_handle *trans = NULL;
3964         int compressed;
3965
3966 again:
3967         spin_lock(&em_tree->lock);
3968         em = lookup_extent_mapping(em_tree, start, len);
3969         if (em)
3970                 em->bdev = root->fs_info->fs_devices->latest_bdev;
3971         spin_unlock(&em_tree->lock);
3972
3973         if (em) {
3974                 if (em->start > start || em->start + em->len <= start)
3975                         free_extent_map(em);
3976                 else if (em->block_start == EXTENT_MAP_INLINE && page)
3977                         free_extent_map(em);
3978                 else
3979                         goto out;
3980         }
3981         em = alloc_extent_map(GFP_NOFS);
3982         if (!em) {
3983                 err = -ENOMEM;
3984                 goto out;
3985         }
3986         em->bdev = root->fs_info->fs_devices->latest_bdev;
3987         em->start = EXTENT_MAP_HOLE;
3988         em->orig_start = EXTENT_MAP_HOLE;
3989         em->len = (u64)-1;
3990         em->block_len = (u64)-1;
3991
3992         if (!path) {
3993                 path = btrfs_alloc_path();
3994                 BUG_ON(!path);
3995         }
3996
3997         ret = btrfs_lookup_file_extent(trans, root, path,
3998                                        objectid, start, trans != NULL);
3999         if (ret < 0) {
4000                 err = ret;
4001                 goto out;
4002         }
4003
4004         if (ret != 0) {
4005                 if (path->slots[0] == 0)
4006                         goto not_found;
4007                 path->slots[0]--;
4008         }
4009
4010         leaf = path->nodes[0];
4011         item = btrfs_item_ptr(leaf, path->slots[0],
4012                               struct btrfs_file_extent_item);
4013         /* are we inside the extent that was found? */
4014         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4015         found_type = btrfs_key_type(&found_key);
4016         if (found_key.objectid != objectid ||
4017             found_type != BTRFS_EXTENT_DATA_KEY) {
4018                 goto not_found;
4019         }
4020
4021         found_type = btrfs_file_extent_type(leaf, item);
4022         extent_start = found_key.offset;
4023         compressed = btrfs_file_extent_compression(leaf, item);
4024         if (found_type == BTRFS_FILE_EXTENT_REG ||
4025             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4026                 extent_end = extent_start +
4027                        btrfs_file_extent_num_bytes(leaf, item);
4028         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4029                 size_t size;
4030                 size = btrfs_file_extent_inline_len(leaf, item);
4031                 extent_end = (extent_start + size + root->sectorsize - 1) &
4032                         ~((u64)root->sectorsize - 1);
4033         }
4034
4035         if (start >= extent_end) {
4036                 path->slots[0]++;
4037                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
4038                         ret = btrfs_next_leaf(root, path);
4039                         if (ret < 0) {
4040                                 err = ret;
4041                                 goto out;
4042                         }
4043                         if (ret > 0)
4044                                 goto not_found;
4045                         leaf = path->nodes[0];
4046                 }
4047                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4048                 if (found_key.objectid != objectid ||
4049                     found_key.type != BTRFS_EXTENT_DATA_KEY)
4050                         goto not_found;
4051                 if (start + len <= found_key.offset)
4052                         goto not_found;
4053                 em->start = start;
4054                 em->len = found_key.offset - start;
4055                 goto not_found_em;
4056         }
4057
4058         if (found_type == BTRFS_FILE_EXTENT_REG ||
4059             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4060                 em->start = extent_start;
4061                 em->len = extent_end - extent_start;
4062                 em->orig_start = extent_start -
4063                                  btrfs_file_extent_offset(leaf, item);
4064                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
4065                 if (bytenr == 0) {
4066                         em->block_start = EXTENT_MAP_HOLE;
4067                         goto insert;
4068                 }
4069                 if (compressed) {
4070                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4071                         em->block_start = bytenr;
4072                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
4073                                                                          item);
4074                 } else {
4075                         bytenr += btrfs_file_extent_offset(leaf, item);
4076                         em->block_start = bytenr;
4077                         em->block_len = em->len;
4078                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
4079                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
4080                 }
4081                 goto insert;
4082         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4083                 unsigned long ptr;
4084                 char *map;
4085                 size_t size;
4086                 size_t extent_offset;
4087                 size_t copy_size;
4088
4089                 em->block_start = EXTENT_MAP_INLINE;
4090                 if (!page || create) {
4091                         em->start = extent_start;
4092                         em->len = extent_end - extent_start;
4093                         goto out;
4094                 }
4095
4096                 size = btrfs_file_extent_inline_len(leaf, item);
4097                 extent_offset = page_offset(page) + pg_offset - extent_start;
4098                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
4099                                 size - extent_offset);
4100                 em->start = extent_start + extent_offset;
4101                 em->len = (copy_size + root->sectorsize - 1) &
4102                         ~((u64)root->sectorsize - 1);
4103                 em->orig_start = EXTENT_MAP_INLINE;
4104                 if (compressed)
4105                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4106                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
4107                 if (create == 0 && !PageUptodate(page)) {
4108                         if (btrfs_file_extent_compression(leaf, item) ==
4109                             BTRFS_COMPRESS_ZLIB) {
4110                                 ret = uncompress_inline(path, inode, page,
4111                                                         pg_offset,
4112                                                         extent_offset, item);
4113                                 BUG_ON(ret);
4114                         } else {
4115                                 map = kmap(page);
4116                                 read_extent_buffer(leaf, map + pg_offset, ptr,
4117                                                    copy_size);
4118                                 kunmap(page);
4119                         }
4120                         flush_dcache_page(page);
4121                 } else if (create && PageUptodate(page)) {
4122                         if (!trans) {
4123                                 kunmap(page);
4124                                 free_extent_map(em);
4125                                 em = NULL;
4126                                 btrfs_release_path(root, path);
4127                                 trans = btrfs_join_transaction(root, 1);
4128                                 goto again;
4129                         }
4130                         map = kmap(page);
4131                         write_extent_buffer(leaf, map + pg_offset, ptr,
4132                                             copy_size);
4133                         kunmap(page);
4134                         btrfs_mark_buffer_dirty(leaf);
4135                 }
4136                 set_extent_uptodate(io_tree, em->start,
4137                                     extent_map_end(em) - 1, GFP_NOFS);
4138                 goto insert;
4139         } else {
4140                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
4141                 WARN_ON(1);
4142         }
4143 not_found:
4144         em->start = start;
4145         em->len = len;
4146 not_found_em:
4147         em->block_start = EXTENT_MAP_HOLE;
4148         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
4149 insert:
4150         btrfs_release_path(root, path);
4151         if (em->start > start || extent_map_end(em) <= start) {
4152                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
4153                        "[%llu %llu]\n", (unsigned long long)em->start,
4154                        (unsigned long long)em->len,
4155                        (unsigned long long)start,
4156                        (unsigned long long)len);
4157                 err = -EIO;
4158                 goto out;
4159         }
4160
4161         err = 0;
4162         spin_lock(&em_tree->lock);
4163         ret = add_extent_mapping(em_tree, em);
4164         /* it is possible that someone inserted the extent into the tree
4165          * while we had the lock dropped.  It is also possible that
4166          * an overlapping map exists in the tree
4167          */
4168         if (ret == -EEXIST) {
4169                 struct extent_map *existing;
4170
4171                 ret = 0;
4172
4173                 existing = lookup_extent_mapping(em_tree, start, len);
4174                 if (existing && (existing->start > start ||
4175                     existing->start + existing->len <= start)) {
4176                         free_extent_map(existing);
4177                         existing = NULL;
4178                 }
4179                 if (!existing) {
4180                         existing = lookup_extent_mapping(em_tree, em->start,
4181                                                          em->len);
4182                         if (existing) {
4183                                 err = merge_extent_mapping(em_tree, existing,
4184                                                            em, start,
4185                                                            root->sectorsize);
4186                                 free_extent_map(existing);
4187                                 if (err) {
4188                                         free_extent_map(em);
4189                                         em = NULL;
4190                                 }
4191                         } else {
4192                                 err = -EIO;
4193                                 free_extent_map(em);
4194                                 em = NULL;
4195                         }
4196                 } else {
4197                         free_extent_map(em);
4198                         em = existing;
4199                         err = 0;
4200                 }
4201         }
4202         spin_unlock(&em_tree->lock);
4203 out:
4204         if (path)
4205                 btrfs_free_path(path);
4206         if (trans) {
4207                 ret = btrfs_end_transaction(trans, root);
4208                 if (!err)
4209                         err = ret;
4210         }
4211         if (err) {
4212                 free_extent_map(em);
4213                 WARN_ON(1);
4214                 return ERR_PTR(err);
4215         }
4216         return em;
4217 }
4218
4219 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
4220                         const struct iovec *iov, loff_t offset,
4221                         unsigned long nr_segs)
4222 {
4223         return -EINVAL;
4224 }
4225
4226 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4227                 __u64 start, __u64 len)
4228 {
4229         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent);
4230 }
4231
4232 int btrfs_readpage(struct file *file, struct page *page)
4233 {
4234         struct extent_io_tree *tree;
4235         tree = &BTRFS_I(page->mapping->host)->io_tree;
4236         return extent_read_full_page(tree, page, btrfs_get_extent);
4237 }
4238
4239 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
4240 {
4241         struct extent_io_tree *tree;
4242
4243
4244         if (current->flags & PF_MEMALLOC) {
4245                 redirty_page_for_writepage(wbc, page);
4246                 unlock_page(page);
4247                 return 0;
4248         }
4249         tree = &BTRFS_I(page->mapping->host)->io_tree;
4250         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
4251 }
4252
4253 int btrfs_writepages(struct address_space *mapping,
4254                      struct writeback_control *wbc)
4255 {
4256         struct extent_io_tree *tree;
4257
4258         tree = &BTRFS_I(mapping->host)->io_tree;
4259         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
4260 }
4261
4262 static int
4263 btrfs_readpages(struct file *file, struct address_space *mapping,
4264                 struct list_head *pages, unsigned nr_pages)
4265 {
4266         struct extent_io_tree *tree;
4267         tree = &BTRFS_I(mapping->host)->io_tree;
4268         return extent_readpages(tree, mapping, pages, nr_pages,
4269                                 btrfs_get_extent);
4270 }
4271 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4272 {
4273         struct extent_io_tree *tree;
4274         struct extent_map_tree *map;
4275         int ret;
4276
4277         tree = &BTRFS_I(page->mapping->host)->io_tree;
4278         map = &BTRFS_I(page->mapping->host)->extent_tree;
4279         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
4280         if (ret == 1) {
4281                 ClearPagePrivate(page);
4282                 set_page_private(page, 0);
4283                 page_cache_release(page);
4284         }
4285         return ret;
4286 }
4287
4288 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4289 {
4290         if (PageWriteback(page) || PageDirty(page))
4291                 return 0;
4292         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
4293 }
4294
4295 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
4296 {
4297         struct extent_io_tree *tree;
4298         struct btrfs_ordered_extent *ordered;
4299         u64 page_start = page_offset(page);
4300         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
4301
4302         wait_on_page_writeback(page);
4303         tree = &BTRFS_I(page->mapping->host)->io_tree;
4304         if (offset) {
4305                 btrfs_releasepage(page, GFP_NOFS);
4306                 return;
4307         }
4308
4309         lock_extent(tree, page_start, page_end, GFP_NOFS);
4310         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
4311                                            page_offset(page));
4312         if (ordered) {
4313                 /*
4314                  * IO on this page will never be started, so we need
4315                  * to account for any ordered extents now
4316                  */
4317                 clear_extent_bit(tree, page_start, page_end,
4318                                  EXTENT_DIRTY | EXTENT_DELALLOC |
4319                                  EXTENT_LOCKED, 1, 0, GFP_NOFS);
4320                 btrfs_finish_ordered_io(page->mapping->host,
4321                                         page_start, page_end);
4322                 btrfs_put_ordered_extent(ordered);
4323                 lock_extent(tree, page_start, page_end, GFP_NOFS);
4324         }
4325         clear_extent_bit(tree, page_start, page_end,
4326                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4327                  EXTENT_ORDERED,
4328                  1, 1, GFP_NOFS);
4329         __btrfs_releasepage(page, GFP_NOFS);
4330
4331         ClearPageChecked(page);
4332         if (PagePrivate(page)) {
4333                 ClearPagePrivate(page);
4334                 set_page_private(page, 0);
4335                 page_cache_release(page);
4336         }
4337 }
4338
4339 /*
4340  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
4341  * called from a page fault handler when a page is first dirtied. Hence we must
4342  * be careful to check for EOF conditions here. We set the page up correctly
4343  * for a written page which means we get ENOSPC checking when writing into
4344  * holes and correct delalloc and unwritten extent mapping on filesystems that
4345  * support these features.
4346  *
4347  * We are not allowed to take the i_mutex here so we have to play games to
4348  * protect against truncate races as the page could now be beyond EOF.  Because
4349  * vmtruncate() writes the inode size before removing pages, once we have the
4350  * page lock we can determine safely if the page is beyond EOF. If it is not
4351  * beyond EOF, then the page is guaranteed safe against truncation until we
4352  * unlock the page.
4353  */
4354 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
4355 {
4356         struct inode *inode = fdentry(vma->vm_file)->d_inode;
4357         struct btrfs_root *root = BTRFS_I(inode)->root;
4358         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4359         struct btrfs_ordered_extent *ordered;
4360         char *kaddr;
4361         unsigned long zero_start;
4362         loff_t size;
4363         int ret;
4364         u64 page_start;
4365         u64 page_end;
4366
4367         ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE);
4368         if (ret)
4369                 goto out;
4370
4371         ret = -EINVAL;
4372 again:
4373         lock_page(page);
4374         size = i_size_read(inode);
4375         page_start = page_offset(page);
4376         page_end = page_start + PAGE_CACHE_SIZE - 1;
4377
4378         if ((page->mapping != inode->i_mapping) ||
4379             (page_start >= size)) {
4380                 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
4381                 /* page got truncated out from underneath us */
4382                 goto out_unlock;
4383         }
4384         wait_on_page_writeback(page);
4385
4386         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
4387         set_page_extent_mapped(page);
4388
4389         /*
4390          * we can't set the delalloc bits if there are pending ordered
4391          * extents.  Drop our locks and wait for them to finish
4392          */
4393         ordered = btrfs_lookup_ordered_extent(inode, page_start);
4394         if (ordered) {
4395                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4396                 unlock_page(page);
4397                 btrfs_start_ordered_extent(inode, ordered, 1);
4398                 btrfs_put_ordered_extent(ordered);
4399                 goto again;
4400         }
4401
4402         btrfs_set_extent_delalloc(inode, page_start, page_end);
4403         ret = 0;
4404
4405         /* page is wholly or partially inside EOF */
4406         if (page_start + PAGE_CACHE_SIZE > size)
4407                 zero_start = size & ~PAGE_CACHE_MASK;
4408         else
4409                 zero_start = PAGE_CACHE_SIZE;
4410
4411         if (zero_start != PAGE_CACHE_SIZE) {
4412                 kaddr = kmap(page);
4413                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
4414                 flush_dcache_page(page);
4415                 kunmap(page);
4416         }
4417         ClearPageChecked(page);
4418         set_page_dirty(page);
4419         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4420
4421 out_unlock:
4422         unlock_page(page);
4423 out:
4424         return ret;
4425 }
4426
4427 static void btrfs_truncate(struct inode *inode)
4428 {
4429         struct btrfs_root *root = BTRFS_I(inode)->root;
4430         int ret;
4431         struct btrfs_trans_handle *trans;
4432         unsigned long nr;
4433         u64 mask = root->sectorsize - 1;
4434
4435         if (!S_ISREG(inode->i_mode))
4436                 return;
4437         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4438                 return;
4439
4440         btrfs_truncate_page(inode->i_mapping, inode->i_size);
4441         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
4442
4443         trans = btrfs_start_transaction(root, 1);
4444         btrfs_set_trans_block_group(trans, inode);
4445         btrfs_i_size_write(inode, inode->i_size);
4446
4447         ret = btrfs_orphan_add(trans, inode);
4448         if (ret)
4449                 goto out;
4450         /* FIXME, add redo link to tree so we don't leak on crash */
4451         ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size,
4452                                       BTRFS_EXTENT_DATA_KEY);
4453         btrfs_update_inode(trans, root, inode);
4454
4455         ret = btrfs_orphan_del(trans, inode);
4456         BUG_ON(ret);
4457
4458 out:
4459         nr = trans->blocks_used;
4460         ret = btrfs_end_transaction_throttle(trans, root);
4461         BUG_ON(ret);
4462         btrfs_btree_balance_dirty(root, nr);
4463 }
4464
4465 /*
4466  * create a new subvolume directory/inode (helper for the ioctl).
4467  */
4468 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
4469                              struct btrfs_root *new_root, struct dentry *dentry,
4470                              u64 new_dirid, u64 alloc_hint)
4471 {
4472         struct inode *inode;
4473         int error;
4474         u64 index = 0;
4475
4476         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
4477                                 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
4478         if (IS_ERR(inode))
4479                 return PTR_ERR(inode);
4480         inode->i_op = &btrfs_dir_inode_operations;
4481         inode->i_fop = &btrfs_dir_file_operations;
4482
4483         inode->i_nlink = 1;
4484         btrfs_i_size_write(inode, 0);
4485
4486         error = btrfs_update_inode(trans, new_root, inode);
4487         if (error)
4488                 return error;
4489
4490         d_instantiate(dentry, inode);
4491         return 0;
4492 }
4493
4494 /* helper function for file defrag and space balancing.  This
4495  * forces readahead on a given range of bytes in an inode
4496  */
4497 unsigned long btrfs_force_ra(struct address_space *mapping,
4498                               struct file_ra_state *ra, struct file *file,
4499                               pgoff_t offset, pgoff_t last_index)
4500 {
4501         pgoff_t req_size = last_index - offset + 1;
4502
4503         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
4504         return offset + req_size;
4505 }
4506
4507 struct inode *btrfs_alloc_inode(struct super_block *sb)
4508 {
4509         struct btrfs_inode *ei;
4510
4511         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
4512         if (!ei)
4513                 return NULL;
4514         ei->last_trans = 0;
4515         ei->logged_trans = 0;
4516         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
4517         ei->i_acl = BTRFS_ACL_NOT_CACHED;
4518         ei->i_default_acl = BTRFS_ACL_NOT_CACHED;
4519         INIT_LIST_HEAD(&ei->i_orphan);
4520         return &ei->vfs_inode;
4521 }
4522
4523 void btrfs_destroy_inode(struct inode *inode)
4524 {
4525         struct btrfs_ordered_extent *ordered;
4526         WARN_ON(!list_empty(&inode->i_dentry));
4527         WARN_ON(inode->i_data.nrpages);
4528
4529         if (BTRFS_I(inode)->i_acl &&
4530             BTRFS_I(inode)->i_acl != BTRFS_ACL_NOT_CACHED)
4531                 posix_acl_release(BTRFS_I(inode)->i_acl);
4532         if (BTRFS_I(inode)->i_default_acl &&
4533             BTRFS_I(inode)->i_default_acl != BTRFS_ACL_NOT_CACHED)
4534                 posix_acl_release(BTRFS_I(inode)->i_default_acl);
4535
4536         spin_lock(&BTRFS_I(inode)->root->list_lock);
4537         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
4538                 printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
4539                        " list\n", inode->i_ino);
4540                 dump_stack();
4541         }
4542         spin_unlock(&BTRFS_I(inode)->root->list_lock);
4543
4544         while (1) {
4545                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
4546                 if (!ordered)
4547                         break;
4548                 else {
4549                         printk(KERN_ERR "btrfs found ordered "
4550                                "extent %llu %llu on inode cleanup\n",
4551                                (unsigned long long)ordered->file_offset,
4552                                (unsigned long long)ordered->len);
4553                         btrfs_remove_ordered_extent(inode, ordered);
4554                         btrfs_put_ordered_extent(ordered);
4555                         btrfs_put_ordered_extent(ordered);
4556                 }
4557         }
4558         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
4559         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
4560 }
4561
4562 static void init_once(void *foo)
4563 {
4564         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
4565
4566         inode_init_once(&ei->vfs_inode);
4567 }
4568
4569 void btrfs_destroy_cachep(void)
4570 {
4571         if (btrfs_inode_cachep)
4572                 kmem_cache_destroy(btrfs_inode_cachep);
4573         if (btrfs_trans_handle_cachep)
4574                 kmem_cache_destroy(btrfs_trans_handle_cachep);
4575         if (btrfs_transaction_cachep)
4576                 kmem_cache_destroy(btrfs_transaction_cachep);
4577         if (btrfs_bit_radix_cachep)
4578                 kmem_cache_destroy(btrfs_bit_radix_cachep);
4579         if (btrfs_path_cachep)
4580                 kmem_cache_destroy(btrfs_path_cachep);
4581 }
4582
4583 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
4584                                        unsigned long extra_flags,
4585                                        void (*ctor)(void *))
4586 {
4587         return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
4588                                  SLAB_MEM_SPREAD | extra_flags), ctor);
4589 }
4590
4591 int btrfs_init_cachep(void)
4592 {
4593         btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
4594                                           sizeof(struct btrfs_inode),
4595                                           0, init_once);
4596         if (!btrfs_inode_cachep)
4597                 goto fail;
4598         btrfs_trans_handle_cachep =
4599                         btrfs_cache_create("btrfs_trans_handle_cache",
4600                                            sizeof(struct btrfs_trans_handle),
4601                                            0, NULL);
4602         if (!btrfs_trans_handle_cachep)
4603                 goto fail;
4604         btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
4605                                              sizeof(struct btrfs_transaction),
4606                                              0, NULL);
4607         if (!btrfs_transaction_cachep)
4608                 goto fail;
4609         btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
4610                                          sizeof(struct btrfs_path),
4611                                          0, NULL);
4612         if (!btrfs_path_cachep)
4613                 goto fail;
4614         btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
4615                                               SLAB_DESTROY_BY_RCU, NULL);
4616         if (!btrfs_bit_radix_cachep)
4617                 goto fail;
4618         return 0;
4619 fail:
4620         btrfs_destroy_cachep();
4621         return -ENOMEM;
4622 }
4623
4624 static int btrfs_getattr(struct vfsmount *mnt,
4625                          struct dentry *dentry, struct kstat *stat)
4626 {
4627         struct inode *inode = dentry->d_inode;
4628         generic_fillattr(inode, stat);
4629         stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
4630         stat->blksize = PAGE_CACHE_SIZE;
4631         stat->blocks = (inode_get_bytes(inode) +
4632                         BTRFS_I(inode)->delalloc_bytes) >> 9;
4633         return 0;
4634 }
4635
4636 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4637                            struct inode *new_dir, struct dentry *new_dentry)
4638 {
4639         struct btrfs_trans_handle *trans;
4640         struct btrfs_root *root = BTRFS_I(old_dir)->root;
4641         struct inode *new_inode = new_dentry->d_inode;
4642         struct inode *old_inode = old_dentry->d_inode;
4643         struct timespec ctime = CURRENT_TIME;
4644         u64 index = 0;
4645         int ret;
4646
4647         /* we're not allowed to rename between subvolumes */
4648         if (BTRFS_I(old_inode)->root->root_key.objectid !=
4649             BTRFS_I(new_dir)->root->root_key.objectid)
4650                 return -EXDEV;
4651
4652         if (S_ISDIR(old_inode->i_mode) && new_inode &&
4653             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
4654                 return -ENOTEMPTY;
4655         }
4656
4657         /* to rename a snapshot or subvolume, we need to juggle the
4658          * backrefs.  This isn't coded yet
4659          */
4660         if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
4661                 return -EXDEV;
4662
4663         ret = btrfs_check_metadata_free_space(root);
4664         if (ret)
4665                 goto out_unlock;
4666
4667         trans = btrfs_start_transaction(root, 1);
4668
4669         btrfs_set_trans_block_group(trans, new_dir);
4670
4671         btrfs_inc_nlink(old_dentry->d_inode);
4672         old_dir->i_ctime = old_dir->i_mtime = ctime;
4673         new_dir->i_ctime = new_dir->i_mtime = ctime;
4674         old_inode->i_ctime = ctime;
4675
4676         ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode,
4677                                  old_dentry->d_name.name,
4678                                  old_dentry->d_name.len);
4679         if (ret)
4680                 goto out_fail;
4681
4682         if (new_inode) {
4683                 new_inode->i_ctime = CURRENT_TIME;
4684                 ret = btrfs_unlink_inode(trans, root, new_dir,
4685                                          new_dentry->d_inode,
4686                                          new_dentry->d_name.name,
4687                                          new_dentry->d_name.len);
4688                 if (ret)
4689                         goto out_fail;
4690                 if (new_inode->i_nlink == 0) {
4691                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
4692                         if (ret)
4693                                 goto out_fail;
4694                 }
4695
4696         }
4697         ret = btrfs_set_inode_index(new_dir, &index);
4698         if (ret)
4699                 goto out_fail;
4700
4701         ret = btrfs_add_link(trans, new_dentry->d_parent->d_inode,
4702                              old_inode, new_dentry->d_name.name,
4703                              new_dentry->d_name.len, 1, index);
4704         if (ret)
4705                 goto out_fail;
4706
4707 out_fail:
4708         btrfs_end_transaction_throttle(trans, root);
4709 out_unlock:
4710         return ret;
4711 }
4712
4713 /*
4714  * some fairly slow code that needs optimization. This walks the list
4715  * of all the inodes with pending delalloc and forces them to disk.
4716  */
4717 int btrfs_start_delalloc_inodes(struct btrfs_root *root)
4718 {
4719         struct list_head *head = &root->fs_info->delalloc_inodes;
4720         struct btrfs_inode *binode;
4721         struct inode *inode;
4722
4723         if (root->fs_info->sb->s_flags & MS_RDONLY)
4724                 return -EROFS;
4725
4726         spin_lock(&root->fs_info->delalloc_lock);
4727         while (!list_empty(head)) {
4728                 binode = list_entry(head->next, struct btrfs_inode,
4729                                     delalloc_inodes);
4730                 inode = igrab(&binode->vfs_inode);
4731                 if (!inode)
4732                         list_del_init(&binode->delalloc_inodes);
4733                 spin_unlock(&root->fs_info->delalloc_lock);
4734                 if (inode) {
4735                         filemap_flush(inode->i_mapping);
4736                         iput(inode);
4737                 }
4738                 cond_resched();
4739                 spin_lock(&root->fs_info->delalloc_lock);
4740         }
4741         spin_unlock(&root->fs_info->delalloc_lock);
4742
4743         /* the filemap_flush will queue IO into the worker threads, but
4744          * we have to make sure the IO is actually started and that
4745          * ordered extents get created before we return
4746          */
4747         atomic_inc(&root->fs_info->async_submit_draining);
4748         while (atomic_read(&root->fs_info->nr_async_submits) ||
4749               atomic_read(&root->fs_info->async_delalloc_pages)) {
4750                 wait_event(root->fs_info->async_submit_wait,
4751                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
4752                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
4753         }
4754         atomic_dec(&root->fs_info->async_submit_draining);
4755         return 0;
4756 }
4757
4758 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
4759                          const char *symname)
4760 {
4761         struct btrfs_trans_handle *trans;
4762         struct btrfs_root *root = BTRFS_I(dir)->root;
4763         struct btrfs_path *path;
4764         struct btrfs_key key;
4765         struct inode *inode = NULL;
4766         int err;
4767         int drop_inode = 0;
4768         u64 objectid;
4769         u64 index = 0 ;
4770         int name_len;
4771         int datasize;
4772         unsigned long ptr;
4773         struct btrfs_file_extent_item *ei;
4774         struct extent_buffer *leaf;
4775         unsigned long nr = 0;
4776
4777         name_len = strlen(symname) + 1;
4778         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
4779                 return -ENAMETOOLONG;
4780
4781         err = btrfs_check_metadata_free_space(root);
4782         if (err)
4783                 goto out_fail;
4784
4785         trans = btrfs_start_transaction(root, 1);
4786         btrfs_set_trans_block_group(trans, dir);
4787
4788         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4789         if (err) {
4790                 err = -ENOSPC;
4791                 goto out_unlock;
4792         }
4793
4794         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4795                                 dentry->d_name.len,
4796                                 dentry->d_parent->d_inode->i_ino, objectid,
4797                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
4798                                 &index);
4799         err = PTR_ERR(inode);
4800         if (IS_ERR(inode))
4801                 goto out_unlock;
4802
4803         err = btrfs_init_inode_security(inode, dir);
4804         if (err) {
4805                 drop_inode = 1;
4806                 goto out_unlock;
4807         }
4808
4809         btrfs_set_trans_block_group(trans, inode);
4810         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
4811         if (err)
4812                 drop_inode = 1;
4813         else {
4814                 inode->i_mapping->a_ops = &btrfs_aops;
4815                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4816                 inode->i_fop = &btrfs_file_operations;
4817                 inode->i_op = &btrfs_file_inode_operations;
4818                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4819         }
4820         dir->i_sb->s_dirt = 1;
4821         btrfs_update_inode_block_group(trans, inode);
4822         btrfs_update_inode_block_group(trans, dir);
4823         if (drop_inode)
4824                 goto out_unlock;
4825
4826         path = btrfs_alloc_path();
4827         BUG_ON(!path);
4828         key.objectid = inode->i_ino;
4829         key.offset = 0;
4830         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
4831         datasize = btrfs_file_extent_calc_inline_size(name_len);
4832         err = btrfs_insert_empty_item(trans, root, path, &key,
4833                                       datasize);
4834         if (err) {
4835                 drop_inode = 1;
4836                 goto out_unlock;
4837         }
4838         leaf = path->nodes[0];
4839         ei = btrfs_item_ptr(leaf, path->slots[0],
4840                             struct btrfs_file_extent_item);
4841         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
4842         btrfs_set_file_extent_type(leaf, ei,
4843                                    BTRFS_FILE_EXTENT_INLINE);
4844         btrfs_set_file_extent_encryption(leaf, ei, 0);
4845         btrfs_set_file_extent_compression(leaf, ei, 0);
4846         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
4847         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
4848
4849         ptr = btrfs_file_extent_inline_start(ei);
4850         write_extent_buffer(leaf, symname, ptr, name_len);
4851         btrfs_mark_buffer_dirty(leaf);
4852         btrfs_free_path(path);
4853
4854         inode->i_op = &btrfs_symlink_inode_operations;
4855         inode->i_mapping->a_ops = &btrfs_symlink_aops;
4856         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4857         inode_set_bytes(inode, name_len);
4858         btrfs_i_size_write(inode, name_len - 1);
4859         err = btrfs_update_inode(trans, root, inode);
4860         if (err)
4861                 drop_inode = 1;
4862
4863 out_unlock:
4864         nr = trans->blocks_used;
4865         btrfs_end_transaction_throttle(trans, root);
4866 out_fail:
4867         if (drop_inode) {
4868                 inode_dec_link_count(inode);
4869                 iput(inode);
4870         }
4871         btrfs_btree_balance_dirty(root, nr);
4872         return err;
4873 }
4874
4875 static int prealloc_file_range(struct inode *inode, u64 start, u64 end,
4876                                u64 alloc_hint, int mode)
4877 {
4878         struct btrfs_trans_handle *trans;
4879         struct btrfs_root *root = BTRFS_I(inode)->root;
4880         struct btrfs_key ins;
4881         u64 alloc_size;
4882         u64 cur_offset = start;
4883         u64 num_bytes = end - start;
4884         int ret = 0;
4885
4886         trans = btrfs_join_transaction(root, 1);
4887         BUG_ON(!trans);
4888         btrfs_set_trans_block_group(trans, inode);
4889
4890         while (num_bytes > 0) {
4891                 alloc_size = min(num_bytes, root->fs_info->max_extent);
4892                 ret = btrfs_reserve_extent(trans, root, alloc_size,
4893                                            root->sectorsize, 0, alloc_hint,
4894                                            (u64)-1, &ins, 1);
4895                 if (ret) {
4896                         WARN_ON(1);
4897                         goto out;
4898                 }
4899                 ret = insert_reserved_file_extent(trans, inode,
4900                                                   cur_offset, ins.objectid,
4901                                                   ins.offset, ins.offset,
4902                                                   ins.offset, 0, 0, 0,
4903                                                   BTRFS_FILE_EXTENT_PREALLOC);
4904                 BUG_ON(ret);
4905                 num_bytes -= ins.offset;
4906                 cur_offset += ins.offset;
4907                 alloc_hint = ins.objectid + ins.offset;
4908         }
4909 out:
4910         if (cur_offset > start) {
4911                 inode->i_ctime = CURRENT_TIME;
4912                 btrfs_set_flag(inode, PREALLOC);
4913                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4914                     cur_offset > i_size_read(inode))
4915                         btrfs_i_size_write(inode, cur_offset);
4916                 ret = btrfs_update_inode(trans, root, inode);
4917                 BUG_ON(ret);
4918         }
4919
4920         btrfs_end_transaction(trans, root);
4921         return ret;
4922 }
4923
4924 static long btrfs_fallocate(struct inode *inode, int mode,
4925                             loff_t offset, loff_t len)
4926 {
4927         u64 cur_offset;
4928         u64 last_byte;
4929         u64 alloc_start;
4930         u64 alloc_end;
4931         u64 alloc_hint = 0;
4932         u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
4933         struct extent_map *em;
4934         int ret;
4935
4936         alloc_start = offset & ~mask;
4937         alloc_end =  (offset + len + mask) & ~mask;
4938
4939         mutex_lock(&inode->i_mutex);
4940         if (alloc_start > inode->i_size) {
4941                 ret = btrfs_cont_expand(inode, alloc_start);
4942                 if (ret)
4943                         goto out;
4944         }
4945
4946         while (1) {
4947                 struct btrfs_ordered_extent *ordered;
4948                 lock_extent(&BTRFS_I(inode)->io_tree, alloc_start,
4949                             alloc_end - 1, GFP_NOFS);
4950                 ordered = btrfs_lookup_first_ordered_extent(inode,
4951                                                             alloc_end - 1);
4952                 if (ordered &&
4953                     ordered->file_offset + ordered->len > alloc_start &&
4954                     ordered->file_offset < alloc_end) {
4955                         btrfs_put_ordered_extent(ordered);
4956                         unlock_extent(&BTRFS_I(inode)->io_tree,
4957                                       alloc_start, alloc_end - 1, GFP_NOFS);
4958                         btrfs_wait_ordered_range(inode, alloc_start,
4959                                                  alloc_end - alloc_start);
4960                 } else {
4961                         if (ordered)
4962                                 btrfs_put_ordered_extent(ordered);
4963                         break;
4964                 }
4965         }
4966
4967         cur_offset = alloc_start;
4968         while (1) {
4969                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4970                                       alloc_end - cur_offset, 0);
4971                 BUG_ON(IS_ERR(em) || !em);
4972                 last_byte = min(extent_map_end(em), alloc_end);
4973                 last_byte = (last_byte + mask) & ~mask;
4974                 if (em->block_start == EXTENT_MAP_HOLE) {
4975                         ret = prealloc_file_range(inode, cur_offset,
4976                                         last_byte, alloc_hint, mode);
4977                         if (ret < 0) {
4978                                 free_extent_map(em);
4979                                 break;
4980                         }
4981                 }
4982                 if (em->block_start <= EXTENT_MAP_LAST_BYTE)
4983                         alloc_hint = em->block_start;
4984                 free_extent_map(em);
4985
4986                 cur_offset = last_byte;
4987                 if (cur_offset >= alloc_end) {
4988                         ret = 0;
4989                         break;
4990                 }
4991         }
4992         unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, alloc_end - 1,
4993                       GFP_NOFS);
4994 out:
4995         mutex_unlock(&inode->i_mutex);
4996         return ret;
4997 }
4998
4999 static int btrfs_set_page_dirty(struct page *page)
5000 {
5001         return __set_page_dirty_nobuffers(page);
5002 }
5003
5004 static int btrfs_permission(struct inode *inode, int mask)
5005 {
5006         if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
5007                 return -EACCES;
5008         return generic_permission(inode, mask, btrfs_check_acl);
5009 }
5010
5011 static struct inode_operations btrfs_dir_inode_operations = {
5012         .getattr        = btrfs_getattr,
5013         .lookup         = btrfs_lookup,
5014         .create         = btrfs_create,
5015         .unlink         = btrfs_unlink,
5016         .link           = btrfs_link,
5017         .mkdir          = btrfs_mkdir,
5018         .rmdir          = btrfs_rmdir,
5019         .rename         = btrfs_rename,
5020         .symlink        = btrfs_symlink,
5021         .setattr        = btrfs_setattr,
5022         .mknod          = btrfs_mknod,
5023         .setxattr       = btrfs_setxattr,
5024         .getxattr       = btrfs_getxattr,
5025         .listxattr      = btrfs_listxattr,
5026         .removexattr    = btrfs_removexattr,
5027         .permission     = btrfs_permission,
5028 };
5029 static struct inode_operations btrfs_dir_ro_inode_operations = {
5030         .lookup         = btrfs_lookup,
5031         .permission     = btrfs_permission,
5032 };
5033 static struct file_operations btrfs_dir_file_operations = {
5034         .llseek         = generic_file_llseek,
5035         .read           = generic_read_dir,
5036         .readdir        = btrfs_real_readdir,
5037         .unlocked_ioctl = btrfs_ioctl,
5038 #ifdef CONFIG_COMPAT
5039         .compat_ioctl   = btrfs_ioctl,
5040 #endif
5041         .release        = btrfs_release_file,
5042         .fsync          = btrfs_sync_file,
5043 };
5044
5045 static struct extent_io_ops btrfs_extent_io_ops = {
5046         .fill_delalloc = run_delalloc_range,
5047         .submit_bio_hook = btrfs_submit_bio_hook,
5048         .merge_bio_hook = btrfs_merge_bio_hook,
5049         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
5050         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
5051         .writepage_start_hook = btrfs_writepage_start_hook,
5052         .readpage_io_failed_hook = btrfs_io_failed_hook,
5053         .set_bit_hook = btrfs_set_bit_hook,
5054         .clear_bit_hook = btrfs_clear_bit_hook,
5055 };
5056
5057 /*
5058  * btrfs doesn't support the bmap operation because swapfiles
5059  * use bmap to make a mapping of extents in the file.  They assume
5060  * these extents won't change over the life of the file and they
5061  * use the bmap result to do IO directly to the drive.
5062  *
5063  * the btrfs bmap call would return logical addresses that aren't
5064  * suitable for IO and they also will change frequently as COW
5065  * operations happen.  So, swapfile + btrfs == corruption.
5066  *
5067  * For now we're avoiding this by dropping bmap.
5068  */
5069 static struct address_space_operations btrfs_aops = {
5070         .readpage       = btrfs_readpage,
5071         .writepage      = btrfs_writepage,
5072         .writepages     = btrfs_writepages,
5073         .readpages      = btrfs_readpages,
5074         .sync_page      = block_sync_page,
5075         .direct_IO      = btrfs_direct_IO,
5076         .invalidatepage = btrfs_invalidatepage,
5077         .releasepage    = btrfs_releasepage,
5078         .set_page_dirty = btrfs_set_page_dirty,
5079 };
5080
5081 static struct address_space_operations btrfs_symlink_aops = {
5082         .readpage       = btrfs_readpage,
5083         .writepage      = btrfs_writepage,
5084         .invalidatepage = btrfs_invalidatepage,
5085         .releasepage    = btrfs_releasepage,
5086 };
5087
5088 static struct inode_operations btrfs_file_inode_operations = {
5089         .truncate       = btrfs_truncate,
5090         .getattr        = btrfs_getattr,
5091         .setattr        = btrfs_setattr,
5092         .setxattr       = btrfs_setxattr,
5093         .getxattr       = btrfs_getxattr,
5094         .listxattr      = btrfs_listxattr,
5095         .removexattr    = btrfs_removexattr,
5096         .permission     = btrfs_permission,
5097         .fallocate      = btrfs_fallocate,
5098         .fiemap         = btrfs_fiemap,
5099 };
5100 static struct inode_operations btrfs_special_inode_operations = {
5101         .getattr        = btrfs_getattr,
5102         .setattr        = btrfs_setattr,
5103         .permission     = btrfs_permission,
5104         .setxattr       = btrfs_setxattr,
5105         .getxattr       = btrfs_getxattr,
5106         .listxattr      = btrfs_listxattr,
5107         .removexattr    = btrfs_removexattr,
5108 };
5109 static struct inode_operations btrfs_symlink_inode_operations = {
5110         .readlink       = generic_readlink,
5111         .follow_link    = page_follow_link_light,
5112         .put_link       = page_put_link,
5113         .permission     = btrfs_permission,
5114         .setxattr       = btrfs_setxattr,
5115         .getxattr       = btrfs_getxattr,
5116         .listxattr      = btrfs_listxattr,
5117         .removexattr    = btrfs_removexattr,
5118 };