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