Pileus Git - ~andy/linux/blob - fs/btrfs/inode.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/pagemap.h>
  25 #include <linux/highmem.h>
  26 #include <linux/time.h>
  27 #include <linux/init.h>
  28 #include <linux/string.h>
  29 #include <linux/smp_lock.h>
  30 #include <linux/backing-dev.h>
  31 #include <linux/mpage.h>
  32 #include <linux/swap.h>
  33 #include <linux/writeback.h>
  34 #include <linux/statfs.h>
  35 #include <linux/compat.h>
  36 #include <linux/bit_spinlock.h>
  37 #include <linux/version.h>
  38 #include <linux/xattr.h>
  39 #include <linux/posix_acl.h>
  40 #include <linux/falloc.h>
  41 #include "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 "compat.h"
  51 #include "tree-log.h"
  52 #include "ref-cache.h"
  53 #include "compression.h"
  54
  55 struct btrfs_iget_args {
  56         u64 ino;
  57         struct btrfs_root *root;
  58 };
  59
  60 static struct inode_operations btrfs_dir_inode_operations;
  61 static struct inode_operations btrfs_symlink_inode_operations;
  62 static struct inode_operations btrfs_dir_ro_inode_operations;
  63 static struct inode_operations btrfs_special_inode_operations;
  64 static struct inode_operations btrfs_file_inode_operations;
  65 static struct address_space_operations btrfs_aops;
  66 static struct address_space_operations btrfs_symlink_aops;
  67 static struct file_operations btrfs_dir_file_operations;
  68 static struct extent_io_ops btrfs_extent_io_ops;
  69
  70 static struct kmem_cache *btrfs_inode_cachep;
  71 struct kmem_cache *btrfs_trans_handle_cachep;
  72 struct kmem_cache *btrfs_transaction_cachep;
  73 struct kmem_cache *btrfs_bit_radix_cachep;
  74 struct kmem_cache *btrfs_path_cachep;
  75
  76 #define S_SHIFT 12
  77 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  78         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  79         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  80         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  81         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  82         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  83         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  84         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  85 };
  86
  87 static void btrfs_truncate(struct inode *inode);
  88 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
  89 static noinline int cow_file_range(struct inode *inode,
  90                                    struct page *locked_page,
  91                                    u64 start, u64 end, int *page_started,
  92                                    unsigned long *nr_written, int unlock);
  93
  94 /*
  95  * a very lame attempt at stopping writes when the FS is 85% full.  There
  96  * are countless ways this is incorrect, but it is better than nothing.
  97  */
  98 int btrfs_check_free_space(struct btrfs_root *root, u64 num_required,
  99                            int for_del)
 100 {
 101         u64 total;
 102         u64 used;
 103         u64 thresh;
 104         unsigned long flags;
 105         int ret = 0;
 106
 107         spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
 108         total = btrfs_super_total_bytes(&root->fs_info->super_copy);
 109         used = btrfs_super_bytes_used(&root->fs_info->super_copy);
 110         if (for_del)
 111                 thresh = total * 90;
 112         else
 113                 thresh = total * 85;
 114
 115         do_div(thresh, 100);
 116
 117         if (used + root->fs_info->delalloc_bytes + num_required > thresh)
 118                 ret = -ENOSPC;
 119         spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
 120         return ret;
 121 }
 122
 123 /*
 124  * this does all the hard work for inserting an inline extent into
 125  * the btree.  The caller should have done a btrfs_drop_extents so that
 126  * no overlapping inline items exist in the btree
 127  */
 128 static int noinline insert_inline_extent(struct btrfs_trans_handle *trans,
 129                                 struct btrfs_root *root, struct inode *inode,
 130                                 u64 start, size_t size, size_t compressed_size,
 131                                 struct page **compressed_pages)
 132 {
 133         struct btrfs_key key;
 134         struct btrfs_path *path;
 135         struct extent_buffer *leaf;
 136         struct page *page = NULL;
 137         char *kaddr;
 138         unsigned long ptr;
 139         struct btrfs_file_extent_item *ei;
 140         int err = 0;
 141         int ret;
 142         size_t cur_size = size;
 143         size_t datasize;
 144         unsigned long offset;
 145         int use_compress = 0;
 146
 147         if (compressed_size && compressed_pages) {
 148                 use_compress = 1;
 149                 cur_size = compressed_size;
 150         }
 151
 152         path = btrfs_alloc_path(); if (!path)
 153                 return -ENOMEM;
 154
 155         btrfs_set_trans_block_group(trans, inode);
 156
 157         key.objectid = inode->i_ino;
 158         key.offset = start;
 159         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 160         inode_add_bytes(inode, size);
 161         datasize = btrfs_file_extent_calc_inline_size(cur_size);
 162
 163         inode_add_bytes(inode, size);
 164         ret = btrfs_insert_empty_item(trans, root, path, &key,
 165                                       datasize);
 166         BUG_ON(ret);
 167         if (ret) {
 168                 err = ret;
 169                 printk("got bad ret %d\n", ret);
 170                 goto fail;
 171         }
 172         leaf = path->nodes[0];
 173         ei = btrfs_item_ptr(leaf, path->slots[0],
 174                             struct btrfs_file_extent_item);
 175         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 176         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 177         btrfs_set_file_extent_encryption(leaf, ei, 0);
 178         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 179         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 180         ptr = btrfs_file_extent_inline_start(ei);
 181
 182         if (use_compress) {
 183                 struct page *cpage;
 184                 int i = 0;
 185                 while(compressed_size > 0) {
 186                         cpage = compressed_pages[i];
 187                         cur_size = min_t(unsigned long, compressed_size,
 188                                        PAGE_CACHE_SIZE);
 189
 190                         kaddr = kmap(cpage);
 191                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 192                         kunmap(cpage);
 193
 194                         i++;
 195                         ptr += cur_size;
 196                         compressed_size -= cur_size;
 197                 }
 198                 btrfs_set_file_extent_compression(leaf, ei,
 199                                                   BTRFS_COMPRESS_ZLIB);
 200         } else {
 201                 page = find_get_page(inode->i_mapping,
 202                                      start >> PAGE_CACHE_SHIFT);
 203                 btrfs_set_file_extent_compression(leaf, ei, 0);
 204                 kaddr = kmap_atomic(page, KM_USER0);
 205                 offset = start & (PAGE_CACHE_SIZE - 1);
 206                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 207                 kunmap_atomic(kaddr, KM_USER0);
 208                 page_cache_release(page);
 209         }
 210         btrfs_mark_buffer_dirty(leaf);
 211         btrfs_free_path(path);
 212
 213         BTRFS_I(inode)->disk_i_size = inode->i_size;
 214         btrfs_update_inode(trans, root, inode);
 215         return 0;
 216 fail:
 217         btrfs_free_path(path);
 218         return err;
 219 }
 220
 221
 222 /*
 223  * conditionally insert an inline extent into the file.  This
 224  * does the checks required to make sure the data is small enough
 225  * to fit as an inline extent.
 226  */
 227 static int cow_file_range_inline(struct btrfs_trans_handle *trans,
 228                                  struct btrfs_root *root,
 229                                  struct inode *inode, u64 start, u64 end,
 230                                  size_t compressed_size,
 231                                  struct page **compressed_pages)
 232 {
 233         u64 isize = i_size_read(inode);
 234         u64 actual_end = min(end + 1, isize);
 235         u64 inline_len = actual_end - start;
 236         u64 aligned_end = (end + root->sectorsize - 1) &
 237                         ~((u64)root->sectorsize - 1);
 238         u64 hint_byte;
 239         u64 data_len = inline_len;
 240         int ret;
 241
 242         if (compressed_size)
 243                 data_len = compressed_size;
 244
 245         if (start > 0 ||
 246             actual_end >= PAGE_CACHE_SIZE ||
 247             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 248             (!compressed_size &&
 249             (actual_end & (root->sectorsize - 1)) == 0) ||
 250             end + 1 < isize ||
 251             data_len > root->fs_info->max_inline) {
 252                 return 1;
 253         }
 254
 255         ret = btrfs_drop_extents(trans, root, inode, start,
 256                                  aligned_end, start, &hint_byte);
 257         BUG_ON(ret);
 258
 259         if (isize > actual_end)
 260                 inline_len = min_t(u64, isize, actual_end);
 261         ret = insert_inline_extent(trans, root, inode, start,
 262                                    inline_len, compressed_size,
 263                                    compressed_pages);
 264         BUG_ON(ret);
 265         btrfs_drop_extent_cache(inode, start, aligned_end, 0);
 266         return 0;
 267 }
 268
 269 struct async_extent {
 270         u64 start;
 271         u64 ram_size;
 272         u64 compressed_size;
 273         struct page **pages;
 274         unsigned long nr_pages;
 275         struct list_head list;
 276 };
 277
 278 struct async_cow {
 279         struct inode *inode;
 280         struct btrfs_root *root;
 281         struct page *locked_page;
 282         u64 start;
 283         u64 end;
 284         struct list_head extents;
 285         struct btrfs_work work;
 286 };
 287
 288 static noinline int add_async_extent(struct async_cow *cow,
 289                                      u64 start, u64 ram_size,
 290                                      u64 compressed_size,
 291                                      struct page **pages,
 292                                      unsigned long nr_pages)
 293 {
 294         struct async_extent *async_extent;
 295
 296         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 297         async_extent->start = start;
 298         async_extent->ram_size = ram_size;
 299         async_extent->compressed_size = compressed_size;
 300         async_extent->pages = pages;
 301         async_extent->nr_pages = nr_pages;
 302         list_add_tail(&async_extent->list, &cow->extents);
 303         return 0;
 304 }
 305
 306 /*
 307  * we create compressed extents in two phases.  The first
 308  * phase compresses a range of pages that have already been
 309  * locked (both pages and state bits are locked).
 310  *
 311  * This is done inside an ordered work queue, and the compression
 312  * is spread across many cpus.  The actual IO submission is step
 313  * two, and the ordered work queue takes care of making sure that
 314  * happens in the same order things were put onto the queue by
 315  * writepages and friends.
 316  *
 317  * If this code finds it can't get good compression, it puts an
 318  * entry onto the work queue to write the uncompressed bytes.  This
 319  * makes sure that both compressed inodes and uncompressed inodes
 320  * are written in the same order that pdflush sent them down.
 321  */
 322 static noinline int compress_file_range(struct inode *inode,
 323                                         struct page *locked_page,
 324                                         u64 start, u64 end,
 325                                         struct async_cow *async_cow,
 326                                         int *num_added)
 327 {
 328         struct btrfs_root *root = BTRFS_I(inode)->root;
 329         struct btrfs_trans_handle *trans;
 330         u64 num_bytes;
 331         u64 orig_start;
 332         u64 disk_num_bytes;
 333         u64 blocksize = root->sectorsize;
 334         u64 actual_end;
 335         int ret = 0;
 336         struct page **pages = NULL;
 337         unsigned long nr_pages;
 338         unsigned long nr_pages_ret = 0;
 339         unsigned long total_compressed = 0;
 340         unsigned long total_in = 0;
 341         unsigned long max_compressed = 128 * 1024;
 342         unsigned long max_uncompressed = 128 * 1024;
 343         int i;
 344         int will_compress;
 345
 346         orig_start = start;
 347
 348 again:
 349         will_compress = 0;
 350         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 351         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 352
 353         actual_end = min_t(u64, i_size_read(inode), end + 1);
 354         total_compressed = actual_end - start;
 355
 356         /* we want to make sure that amount of ram required to uncompress
 357          * an extent is reasonable, so we limit the total size in ram
 358          * of a compressed extent to 128k.  This is a crucial number
 359          * because it also controls how easily we can spread reads across
 360          * cpus for decompression.
 361          *
 362          * We also want to make sure the amount of IO required to do
 363          * a random read is reasonably small, so we limit the size of
 364          * a compressed extent to 128k.
 365          */
 366         total_compressed = min(total_compressed, max_uncompressed);
 367         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 368         num_bytes = max(blocksize,  num_bytes);
 369         disk_num_bytes = num_bytes;
 370         total_in = 0;
 371         ret = 0;
 372
 373         /*
 374          * we do compression for mount -o compress and when the
 375          * inode has not been flagged as nocompress.  This flag can
 376          * change at any time if we discover bad compression ratios.
 377          */
 378         if (!btrfs_test_flag(inode, NOCOMPRESS) &&
 379             btrfs_test_opt(root, COMPRESS)) {
 380                 WARN_ON(pages);
 381                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 382
 383                 ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
 384                                                 total_compressed, pages,
 385                                                 nr_pages, &nr_pages_ret,
 386                                                 &total_in,
 387                                                 &total_compressed,
 388                                                 max_compressed);
 389
 390                 if (!ret) {
 391                         unsigned long offset = total_compressed &
 392                                 (PAGE_CACHE_SIZE - 1);
 393                         struct page *page = pages[nr_pages_ret - 1];
 394                         char *kaddr;
 395
 396                         /* zero the tail end of the last page, we might be
 397                          * sending it down to disk
 398                          */
 399                         if (offset) {
 400                                 kaddr = kmap_atomic(page, KM_USER0);
 401                                 memset(kaddr + offset, 0,
 402                                        PAGE_CACHE_SIZE - offset);
 403                                 kunmap_atomic(kaddr, KM_USER0);
 404                         }
 405                         will_compress = 1;
 406                 }
 407         }
 408         if (start == 0) {
 409                 trans = btrfs_join_transaction(root, 1);
 410                 BUG_ON(!trans);
 411                 btrfs_set_trans_block_group(trans, inode);
 412
 413                 /* lets try to make an inline extent */
 414                 if (ret || total_in < (actual_end - start)) {
 415                         /* we didn't compress the entire range, try
 416                          * to make an uncompressed inline extent.
 417                          */
 418                         ret = cow_file_range_inline(trans, root, inode,
 419                                                     start, end, 0, NULL);
 420                 } else {
 421                         /* try making a compressed inline extent */
 422                         ret = cow_file_range_inline(trans, root, inode,
 423                                                     start, end,
 424                                                     total_compressed, pages);
 425                 }
 426                 btrfs_end_transaction(trans, root);
 427                 if (ret == 0) {
 428                         /*
 429                          * inline extent creation worked, we don't need
 430                          * to create any more async work items.  Unlock
 431                          * and free up our temp pages.
 432                          */
 433                         extent_clear_unlock_delalloc(inode,
 434                                                      &BTRFS_I(inode)->io_tree,
 435                                                      start, end, NULL, 1, 0,
 436                                                      0, 1, 1, 1);
 437                         ret = 0;
 438                         goto free_pages_out;
 439                 }
 440         }
 441
 442         if (will_compress) {
 443                 /*
 444                  * we aren't doing an inline extent round the compressed size
 445                  * up to a block size boundary so the allocator does sane
 446                  * things
 447                  */
 448                 total_compressed = (total_compressed + blocksize - 1) &
 449                         ~(blocksize - 1);
 450
 451                 /*
 452                  * one last check to make sure the compression is really a
 453                  * win, compare the page count read with the blocks on disk
 454                  */
 455                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
 456                         ~(PAGE_CACHE_SIZE - 1);
 457                 if (total_compressed >= total_in) {
 458                         will_compress = 0;
 459                 } else {
 460                         disk_num_bytes = total_compressed;
 461                         num_bytes = total_in;
 462                 }
 463         }
 464         if (!will_compress && pages) {
 465                 /*
 466                  * the compression code ran but failed to make things smaller,
 467                  * free any pages it allocated and our page pointer array
 468                  */
 469                 for (i = 0; i < nr_pages_ret; i++) {
 470                         WARN_ON(pages[i]->mapping);
 471                         page_cache_release(pages[i]);
 472                 }
 473                 kfree(pages);
 474                 pages = NULL;
 475                 total_compressed = 0;
 476                 nr_pages_ret = 0;
 477
 478                 /* flag the file so we don't compress in the future */
 479                 btrfs_set_flag(inode, NOCOMPRESS);
 480         }
 481         if (will_compress) {
 482                 *num_added += 1;
 483
 484                 /* the async work queues will take care of doing actual
 485                  * allocation on disk for these compressed pages,
 486                  * and will submit them to the elevator.
 487                  */
 488                 add_async_extent(async_cow, start, num_bytes,
 489                                  total_compressed, pages, nr_pages_ret);
 490
 491                 if (start + num_bytes < end) {
 492                         start += num_bytes;
 493                         pages = NULL;
 494                         cond_resched();
 495                         goto again;
 496                 }
 497         } else {
 498                 /*
 499                  * No compression, but we still need to write the pages in
 500                  * the file we've been given so far.  redirty the locked
 501                  * page if it corresponds to our extent and set things up
 502                  * for the async work queue to run cow_file_range to do
 503                  * the normal delalloc dance
 504                  */
 505                 if (page_offset(locked_page) >= start &&
 506                     page_offset(locked_page) <= end) {
 507                         __set_page_dirty_nobuffers(locked_page);
 508                         /* unlocked later on in the async handlers */
 509                 }
 510                 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0);
 511                 *num_added += 1;
 512         }
 513
 514 out:
 515         return 0;
 516
 517 free_pages_out:
 518         for (i = 0; i < nr_pages_ret; i++) {
 519                 WARN_ON(pages[i]->mapping);
 520                 page_cache_release(pages[i]);
 521         }
 522         if (pages)
 523                 kfree(pages);
 524
 525         goto out;
 526 }
 527
 528 /*
 529  * phase two of compressed writeback.  This is the ordered portion
 530  * of the code, which only gets called in the order the work was
 531  * queued.  We walk all the async extents created by compress_file_range
 532  * and send them down to the disk.
 533  */
 534 static noinline int submit_compressed_extents(struct inode *inode,
 535                                               struct async_cow *async_cow)
 536 {
 537         struct async_extent *async_extent;
 538         u64 alloc_hint = 0;
 539         struct btrfs_trans_handle *trans;
 540         struct btrfs_key ins;
 541         struct extent_map *em;
 542         struct btrfs_root *root = BTRFS_I(inode)->root;
 543         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 544         struct extent_io_tree *io_tree;
 545         int ret;
 546
 547         if (list_empty(&async_cow->extents))
 548                 return 0;
 549
 550         trans = btrfs_join_transaction(root, 1);
 551
 552         while(!list_empty(&async_cow->extents)) {
 553                 async_extent = list_entry(async_cow->extents.next,
 554                                           struct async_extent, list);
 555                 list_del(&async_extent->list);
 556
 557                 io_tree = &BTRFS_I(inode)->io_tree;
 558
 559                 /* did the compression code fall back to uncompressed IO? */
 560                 if (!async_extent->pages) {
 561                         int page_started = 0;
 562                         unsigned long nr_written = 0;
 563
 564                         lock_extent(io_tree, async_extent->start,
 565                                     async_extent->start + async_extent->ram_size - 1,
 566                                     GFP_NOFS);
 567
 568                         /* allocate blocks */
 569                         cow_file_range(inode, async_cow->locked_page,
 570                                        async_extent->start,
 571                                        async_extent->start +
 572                                        async_extent->ram_size - 1,
 573                                        &page_started, &nr_written, 0);
 574
 575                         /*
 576                          * if page_started, cow_file_range inserted an
 577                          * inline extent and took care of all the unlocking
 578                          * and IO for us.  Otherwise, we need to submit
 579                          * all those pages down to the drive.
 580                          */
 581                         if (!page_started)
 582                                 extent_write_locked_range(io_tree,
 583                                                   inode, async_extent->start,
 584                                                   async_extent->start +
 585                                                   async_extent->ram_size - 1,
 586                                                   btrfs_get_extent,
 587                                                   WB_SYNC_ALL);
 588                         kfree(async_extent);
 589                         cond_resched();
 590                         continue;
 591                 }
 592
 593                 lock_extent(io_tree, async_extent->start,
 594                             async_extent->start + async_extent->ram_size - 1,
 595                             GFP_NOFS);
 596                 /*
 597                  * here we're doing allocation and writeback of the
 598                  * compressed pages
 599                  */
 600                 btrfs_drop_extent_cache(inode, async_extent->start,
 601                                         async_extent->start +
 602                                         async_extent->ram_size - 1, 0);
 603
 604                 ret = btrfs_reserve_extent(trans, root,
 605                                            async_extent->compressed_size,
 606                                            async_extent->compressed_size,
 607                                            0, alloc_hint,
 608                                            (u64)-1, &ins, 1);
 609                 BUG_ON(ret);
 610                 em = alloc_extent_map(GFP_NOFS);
 611                 em->start = async_extent->start;
 612                 em->len = async_extent->ram_size;
 613                 em->orig_start = em->start;
 614
 615                 em->block_start = ins.objectid;
 616                 em->block_len = ins.offset;
 617                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 618                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 619                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 620
 621                 while(1) {
 622                         spin_lock(&em_tree->lock);
 623                         ret = add_extent_mapping(em_tree, em);
 624                         spin_unlock(&em_tree->lock);
 625                         if (ret != -EEXIST) {
 626                                 free_extent_map(em);
 627                                 break;
 628                         }
 629                         btrfs_drop_extent_cache(inode, async_extent->start,
 630                                                 async_extent->start +
 631                                                 async_extent->ram_size - 1, 0);
 632                 }
 633
 634                 ret = btrfs_add_ordered_extent(inode, async_extent->start,
 635                                                ins.objectid,
 636                                                async_extent->ram_size,
 637                                                ins.offset,
 638                                                BTRFS_ORDERED_COMPRESSED);
 639                 BUG_ON(ret);
 640
 641                 btrfs_end_transaction(trans, root);
 642
 643                 /*
 644                  * clear dirty, set writeback and unlock the pages.
 645                  */
 646                 extent_clear_unlock_delalloc(inode,
 647                                              &BTRFS_I(inode)->io_tree,
 648                                              async_extent->start,
 649                                              async_extent->start +
 650                                              async_extent->ram_size - 1,
 651                                              NULL, 1, 1, 0, 1, 1, 0);
 652
 653                 ret = btrfs_submit_compressed_write(inode,
 654                                          async_extent->start,
 655                                          async_extent->ram_size,
 656                                          ins.objectid,
 657                                          ins.offset, async_extent->pages,
 658                                          async_extent->nr_pages);
 659
 660                 BUG_ON(ret);
 661                 trans = btrfs_join_transaction(root, 1);
 662                 alloc_hint = ins.objectid + ins.offset;
 663                 kfree(async_extent);
 664                 cond_resched();
 665         }
 666
 667         btrfs_end_transaction(trans, root);
 668         return 0;
 669 }
 670
 671 /*
 672  * when extent_io.c finds a delayed allocation range in the file,
 673  * the call backs end up in this code.  The basic idea is to
 674  * allocate extents on disk for the range, and create ordered data structs
 675  * in ram to track those extents.
 676  *
 677  * locked_page is the page that writepage had locked already.  We use
 678  * it to make sure we don't do extra locks or unlocks.
 679  *
 680  * *page_started is set to one if we unlock locked_page and do everything
 681  * required to start IO on it.  It may be clean and already done with
 682  * IO when we return.
 683  */
 684 static noinline int cow_file_range(struct inode *inode,
 685                                    struct page *locked_page,
 686                                    u64 start, u64 end, int *page_started,
 687                                    unsigned long *nr_written,
 688                                    int unlock)
 689 {
 690         struct btrfs_root *root = BTRFS_I(inode)->root;
 691         struct btrfs_trans_handle *trans;
 692         u64 alloc_hint = 0;
 693         u64 num_bytes;
 694         unsigned long ram_size;
 695         u64 disk_num_bytes;
 696         u64 cur_alloc_size;
 697         u64 blocksize = root->sectorsize;
 698         u64 actual_end;
 699         struct btrfs_key ins;
 700         struct extent_map *em;
 701         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 702         int ret = 0;
 703
 704         trans = btrfs_join_transaction(root, 1);
 705         BUG_ON(!trans);
 706         btrfs_set_trans_block_group(trans, inode);
 707
 708         actual_end = min_t(u64, i_size_read(inode), end + 1);
 709
 710         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 711         num_bytes = max(blocksize,  num_bytes);
 712         disk_num_bytes = num_bytes;
 713         ret = 0;
 714
 715         if (start == 0) {
 716                 /* lets try to make an inline extent */
 717                 ret = cow_file_range_inline(trans, root, inode,
 718                                             start, end, 0, NULL);
 719                 if (ret == 0) {
 720                         extent_clear_unlock_delalloc(inode,
 721                                                      &BTRFS_I(inode)->io_tree,
 722                                                      start, end, NULL, 1, 1,
 723                                                      1, 1, 1, 1);
 724                         *nr_written = *nr_written +
 725                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 726                         *page_started = 1;
 727                         ret = 0;
 728                         goto out;
 729                 }
 730         }
 731
 732         BUG_ON(disk_num_bytes >
 733                btrfs_super_total_bytes(&root->fs_info->super_copy));
 734
 735         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 736
 737         while(disk_num_bytes > 0) {
 738                 cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent);
 739                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 740                                            root->sectorsize, 0, alloc_hint,
 741                                            (u64)-1, &ins, 1);
 742                 if (ret) {
 743                         BUG();
 744                 }
 745                 em = alloc_extent_map(GFP_NOFS);
 746                 em->start = start;
 747                 em->orig_start = em->start;
 748
 749                 ram_size = ins.offset;
 750                 em->len = ins.offset;
 751
 752                 em->block_start = ins.objectid;
 753                 em->block_len = ins.offset;
 754                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 755                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 756
 757                 while(1) {
 758                         spin_lock(&em_tree->lock);
 759                         ret = add_extent_mapping(em_tree, em);
 760                         spin_unlock(&em_tree->lock);
 761                         if (ret != -EEXIST) {
 762                                 free_extent_map(em);
 763                                 break;
 764                         }
 765                         btrfs_drop_extent_cache(inode, start,
 766                                                 start + ram_size - 1, 0);
 767                 }
 768
 769                 cur_alloc_size = ins.offset;
 770                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 771                                                ram_size, cur_alloc_size, 0);
 772                 BUG_ON(ret);
 773
 774                 if (disk_num_bytes < cur_alloc_size) {
 775                         printk("num_bytes %Lu cur_alloc %Lu\n", disk_num_bytes,
 776                                cur_alloc_size);
 777                         break;
 778                 }
 779                 /* we're not doing compressed IO, don't unlock the first
 780                  * page (which the caller expects to stay locked), don't
 781                  * clear any dirty bits and don't set any writeback bits
 782                  */
 783                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 784                                              start, start + ram_size - 1,
 785                                              locked_page, unlock, 1,
 786                                              1, 0, 0, 0);
 787                 disk_num_bytes -= cur_alloc_size;
 788                 num_bytes -= cur_alloc_size;
 789                 alloc_hint = ins.objectid + ins.offset;
 790                 start += cur_alloc_size;
 791         }
 792 out:
 793         ret = 0;
 794         btrfs_end_transaction(trans, root);
 795
 796         return ret;
 797 }
 798
 799 /*
 800  * work queue call back to started compression on a file and pages
 801  */
 802 static noinline void async_cow_start(struct btrfs_work *work)
 803 {
 804         struct async_cow *async_cow;
 805         int num_added = 0;
 806         async_cow = container_of(work, struct async_cow, work);
 807
 808         compress_file_range(async_cow->inode, async_cow->locked_page,
 809                             async_cow->start, async_cow->end, async_cow,
 810                             &num_added);
 811         if (num_added == 0)
 812                 async_cow->inode = NULL;
 813 }
 814
 815 /*
 816  * work queue call back to submit previously compressed pages
 817  */
 818 static noinline void async_cow_submit(struct btrfs_work *work)
 819 {
 820         struct async_cow *async_cow;
 821         struct btrfs_root *root;
 822         unsigned long nr_pages;
 823
 824         async_cow = container_of(work, struct async_cow, work);
 825
 826         root = async_cow->root;
 827         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
 828                 PAGE_CACHE_SHIFT;
 829
 830         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
 831
 832         if (atomic_read(&root->fs_info->async_delalloc_pages) <
 833             5 * 1042 * 1024 &&
 834             waitqueue_active(&root->fs_info->async_submit_wait))
 835                 wake_up(&root->fs_info->async_submit_wait);
 836
 837         if (async_cow->inode) {
 838                 submit_compressed_extents(async_cow->inode, async_cow);
 839         }
 840 }
 841
 842 static noinline void async_cow_free(struct btrfs_work *work)
 843 {
 844         struct async_cow *async_cow;
 845         async_cow = container_of(work, struct async_cow, work);
 846         kfree(async_cow);
 847 }
 848
 849 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
 850                                 u64 start, u64 end, int *page_started,
 851                                 unsigned long *nr_written)
 852 {
 853         struct async_cow *async_cow;
 854         struct btrfs_root *root = BTRFS_I(inode)->root;
 855         unsigned long nr_pages;
 856         u64 cur_end;
 857         int limit = 10 * 1024 * 1042;
 858
 859         if (!btrfs_test_opt(root, COMPRESS)) {
 860                 return cow_file_range(inode, locked_page, start, end,
 861                                       page_started, nr_written, 1);
 862         }
 863
 864         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED |
 865                          EXTENT_DELALLOC, 1, 0, GFP_NOFS);
 866         while(start < end) {
 867                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
 868                 async_cow->inode = inode;
 869                 async_cow->root = root;
 870                 async_cow->locked_page = locked_page;
 871                 async_cow->start = start;
 872
 873                 if (btrfs_test_flag(inode, NOCOMPRESS))
 874                         cur_end = end;
 875                 else
 876                         cur_end = min(end, start + 512 * 1024 - 1);
 877
 878                 async_cow->end = cur_end;
 879                 INIT_LIST_HEAD(&async_cow->extents);
 880
 881                 async_cow->work.func = async_cow_start;
 882                 async_cow->work.ordered_func = async_cow_submit;
 883                 async_cow->work.ordered_free = async_cow_free;
 884                 async_cow->work.flags = 0;
 885
 886                 while(atomic_read(&root->fs_info->async_submit_draining) &&
 887                       atomic_read(&root->fs_info->async_delalloc_pages)) {
 888                         wait_event(root->fs_info->async_submit_wait,
 889                              (atomic_read(&root->fs_info->async_delalloc_pages)
 890                               == 0));
 891                 }
 892
 893                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
 894                         PAGE_CACHE_SHIFT;
 895                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
 896
 897                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
 898                                    &async_cow->work);
 899
 900                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
 901                         wait_event(root->fs_info->async_submit_wait,
 902                            (atomic_read(&root->fs_info->async_delalloc_pages) <
 903                             limit));
 904                 }
 905
 906                 while(atomic_read(&root->fs_info->async_submit_draining) &&
 907                       atomic_read(&root->fs_info->async_delalloc_pages)) {
 908                         wait_event(root->fs_info->async_submit_wait,
 909                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
 910                            0));
 911                 }
 912
 913                 *nr_written += nr_pages;
 914                 start = cur_end + 1;
 915         }
 916         *page_started = 1;
 917         return 0;
 918 }
 919
 920 /*
 921  * when nowcow writeback call back.  This checks for snapshots or COW copies
 922  * of the extents that exist in the file, and COWs the file as required.
 923  *
 924  * If no cow copies or snapshots exist, we write directly to the existing
 925  * blocks on disk
 926  */
 927 static int run_delalloc_nocow(struct inode *inode, struct page *locked_page,
 928                               u64 start, u64 end, int *page_started, int force,
 929                               unsigned long *nr_written)
 930 {
 931         struct btrfs_root *root = BTRFS_I(inode)->root;
 932         struct btrfs_trans_handle *trans;
 933         struct extent_buffer *leaf;
 934         struct btrfs_path *path;
 935         struct btrfs_file_extent_item *fi;
 936         struct btrfs_key found_key;
 937         u64 cow_start;
 938         u64 cur_offset;
 939         u64 extent_end;
 940         u64 disk_bytenr;
 941         u64 num_bytes;
 942         int extent_type;
 943         int ret;
 944         int type;
 945         int nocow;
 946         int check_prev = 1;
 947
 948         path = btrfs_alloc_path();
 949         BUG_ON(!path);
 950         trans = btrfs_join_transaction(root, 1);
 951         BUG_ON(!trans);
 952
 953         cow_start = (u64)-1;
 954         cur_offset = start;
 955         while (1) {
 956                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
 957                                                cur_offset, 0);
 958                 BUG_ON(ret < 0);
 959                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
 960                         leaf = path->nodes[0];
 961                         btrfs_item_key_to_cpu(leaf, &found_key,
 962                                               path->slots[0] - 1);
 963                         if (found_key.objectid == inode->i_ino &&
 964                             found_key.type == BTRFS_EXTENT_DATA_KEY)
 965                                 path->slots[0]--;
 966                 }
 967                 check_prev = 0;
 968 next_slot:
 969                 leaf = path->nodes[0];
 970                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 971                         ret = btrfs_next_leaf(root, path);
 972                         if (ret < 0)
 973                                 BUG_ON(1);
 974                         if (ret > 0)
 975                                 break;
 976                         leaf = path->nodes[0];
 977                 }
 978
 979                 nocow = 0;
 980                 disk_bytenr = 0;
 981                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 982
 983                 if (found_key.objectid > inode->i_ino ||
 984                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
 985                     found_key.offset > end)
 986                         break;
 987
 988                 if (found_key.offset > cur_offset) {
 989                         extent_end = found_key.offset;
 990                         goto out_check;
 991                 }
 992
 993                 fi = btrfs_item_ptr(leaf, path->slots[0],
 994                                     struct btrfs_file_extent_item);
 995                 extent_type = btrfs_file_extent_type(leaf, fi);
 996
 997                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
 998                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
 999                         struct btrfs_block_group_cache *block_group;
1000                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1001                         extent_end = found_key.offset +
1002                                 btrfs_file_extent_num_bytes(leaf, fi);
1003                         if (extent_end <= start) {
1004                                 path->slots[0]++;
1005                                 goto next_slot;
1006                         }
1007                         if (btrfs_file_extent_compression(leaf, fi) ||
1008                             btrfs_file_extent_encryption(leaf, fi) ||
1009                             btrfs_file_extent_other_encoding(leaf, fi))
1010                                 goto out_check;
1011                         if (disk_bytenr == 0)
1012                                 goto out_check;
1013                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1014                                 goto out_check;
1015                         if (btrfs_cross_ref_exist(trans, root, disk_bytenr))
1016                                 goto out_check;
1017                         block_group = btrfs_lookup_block_group(root->fs_info,
1018                                                                disk_bytenr);
1019                         if (!block_group || block_group->ro)
1020                                 goto out_check;
1021                         disk_bytenr += btrfs_file_extent_offset(leaf, fi);
1022                         nocow = 1;
1023                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1024                         extent_end = found_key.offset +
1025                                 btrfs_file_extent_inline_len(leaf, fi);
1026                         extent_end = ALIGN(extent_end, root->sectorsize);
1027                 } else {
1028                         BUG_ON(1);
1029                 }
1030 out_check:
1031                 if (extent_end <= start) {
1032                         path->slots[0]++;
1033                         goto next_slot;
1034                 }
1035                 if (!nocow) {
1036                         if (cow_start == (u64)-1)
1037                                 cow_start = cur_offset;
1038                         cur_offset = extent_end;
1039                         if (cur_offset > end)
1040                                 break;
1041                         path->slots[0]++;
1042                         goto next_slot;
1043                 }
1044
1045                 btrfs_release_path(root, path);
1046                 if (cow_start != (u64)-1) {
1047                         ret = cow_file_range(inode, locked_page, cow_start,
1048                                         found_key.offset - 1, page_started,
1049                                         nr_written, 1);
1050                         BUG_ON(ret);
1051                         cow_start = (u64)-1;
1052                 }
1053
1054                 disk_bytenr += cur_offset - found_key.offset;
1055                 num_bytes = min(end + 1, extent_end) - cur_offset;
1056                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1057                         struct extent_map *em;
1058                         struct extent_map_tree *em_tree;
1059                         em_tree = &BTRFS_I(inode)->extent_tree;
1060                         em = alloc_extent_map(GFP_NOFS);
1061                         em->start = cur_offset;
1062                         em->orig_start = em->start;
1063                         em->len = num_bytes;
1064                         em->block_len = num_bytes;
1065                         em->block_start = disk_bytenr;
1066                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1067                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1068                         while (1) {
1069                                 spin_lock(&em_tree->lock);
1070                                 ret = add_extent_mapping(em_tree, em);
1071                                 spin_unlock(&em_tree->lock);
1072                                 if (ret != -EEXIST) {
1073                                         free_extent_map(em);
1074                                         break;
1075                                 }
1076                                 btrfs_drop_extent_cache(inode, em->start,
1077                                                 em->start + em->len - 1, 0);
1078                         }
1079                         type = BTRFS_ORDERED_PREALLOC;
1080                 } else {
1081                         type = BTRFS_ORDERED_NOCOW;
1082                 }
1083
1084                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1085                                                num_bytes, num_bytes, type);
1086                 BUG_ON(ret);
1087
1088                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1089                                         cur_offset, cur_offset + num_bytes - 1,
1090                                         locked_page, 1, 1, 1, 0, 0, 0);
1091                 cur_offset = extent_end;
1092                 if (cur_offset > end)
1093                         break;
1094         }
1095         btrfs_release_path(root, path);
1096
1097         if (cur_offset <= end && cow_start == (u64)-1)
1098                 cow_start = cur_offset;
1099         if (cow_start != (u64)-1) {
1100                 ret = cow_file_range(inode, locked_page, cow_start, end,
1101                                      page_started, nr_written, 1);
1102                 BUG_ON(ret);
1103         }
1104
1105         ret = btrfs_end_transaction(trans, root);
1106         BUG_ON(ret);
1107         btrfs_free_path(path);
1108         return 0;
1109 }
1110
1111 /*
1112  * extent_io.c call back to do delayed allocation processing
1113  */
1114 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1115                               u64 start, u64 end, int *page_started,
1116                               unsigned long *nr_written)
1117 {
1118         struct btrfs_root *root = BTRFS_I(inode)->root;
1119         int ret;
1120
1121         if (btrfs_test_opt(root, NODATACOW) ||
1122             btrfs_test_flag(inode, NODATACOW))
1123                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1124                                          page_started, 0, nr_written);
1125         else if (btrfs_test_flag(inode, PREALLOC))
1126                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1127                                          page_started, 1, nr_written);
1128         else
1129                 ret = cow_file_range_async(inode, locked_page, start, end,
1130                                      page_started, nr_written);
1131
1132         return ret;
1133 }
1134
1135 /*
1136  * extent_io.c set_bit_hook, used to track delayed allocation
1137  * bytes in this file, and to maintain the list of inodes that
1138  * have pending delalloc work to be done.
1139  */
1140 int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
1141                        unsigned long old, unsigned long bits)
1142 {
1143         unsigned long flags;
1144         if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1145                 struct btrfs_root *root = BTRFS_I(inode)->root;
1146                 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
1147                 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
1148                 root->fs_info->delalloc_bytes += end - start + 1;
1149                 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1150                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1151                                       &root->fs_info->delalloc_inodes);
1152                 }
1153                 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
1154         }
1155         return 0;
1156 }
1157
1158 /*
1159  * extent_io.c clear_bit_hook, see set_bit_hook for why
1160  */
1161 int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
1162                          unsigned long old, unsigned long bits)
1163 {
1164         if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1165                 struct btrfs_root *root = BTRFS_I(inode)->root;
1166                 unsigned long flags;
1167
1168                 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
1169                 if (end - start + 1 > root->fs_info->delalloc_bytes) {
1170                         printk("warning: delalloc account %Lu %Lu\n",
1171                                end - start + 1, root->fs_info->delalloc_bytes);
1172                         root->fs_info->delalloc_bytes = 0;
1173                         BTRFS_I(inode)->delalloc_bytes = 0;
1174                 } else {
1175                         root->fs_info->delalloc_bytes -= end - start + 1;
1176                         BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
1177                 }
1178                 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
1179                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1180                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1181                 }
1182                 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
1183         }
1184         return 0;
1185 }
1186
1187 /*
1188  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1189  * we don't create bios that span stripes or chunks
1190  */
1191 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1192                          size_t size, struct bio *bio,
1193                          unsigned long bio_flags)
1194 {
1195         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1196         struct btrfs_mapping_tree *map_tree;
1197         u64 logical = (u64)bio->bi_sector << 9;
1198         u64 length = 0;
1199         u64 map_length;
1200         int ret;
1201
1202         if (bio_flags & EXTENT_BIO_COMPRESSED)
1203                 return 0;
1204
1205         length = bio->bi_size;
1206         map_tree = &root->fs_info->mapping_tree;
1207         map_length = length;
1208         ret = btrfs_map_block(map_tree, READ, logical,
1209                               &map_length, NULL, 0);
1210
1211         if (map_length < length + size) {
1212                 return 1;
1213         }
1214         return 0;
1215 }
1216
1217 /*
1218  * in order to insert checksums into the metadata in large chunks,
1219  * we wait until bio submission time.   All the pages in the bio are
1220  * checksummed and sums are attached onto the ordered extent record.
1221  *
1222  * At IO completion time the cums attached on the ordered extent record
1223  * are inserted into the btree
1224  */
1225 int __btrfs_submit_bio_start(struct inode *inode, int rw, struct bio *bio,
1226                           int mirror_num, unsigned long bio_flags)
1227 {
1228         struct btrfs_root *root = BTRFS_I(inode)->root;
1229         int ret = 0;
1230
1231         ret = btrfs_csum_one_bio(root, inode, bio);
1232         BUG_ON(ret);
1233         return 0;
1234 }
1235
1236 /*
1237  * in order to insert checksums into the metadata in large chunks,
1238  * we wait until bio submission time.   All the pages in the bio are
1239  * checksummed and sums are attached onto the ordered extent record.
1240  *
1241  * At IO completion time the cums attached on the ordered extent record
1242  * are inserted into the btree
1243  */
1244 int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1245                           int mirror_num, unsigned long bio_flags)
1246 {
1247         struct btrfs_root *root = BTRFS_I(inode)->root;
1248         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1249 }
1250
1251 /*
1252  * extent_io.c submission hook. This does the right thing for csum calculation on write,
1253  * or reading the csums from the tree before a read
1254  */
1255 int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1256                           int mirror_num, unsigned long bio_flags)
1257 {
1258         struct btrfs_root *root = BTRFS_I(inode)->root;
1259         int ret = 0;
1260         int skip_sum;
1261
1262         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1263         BUG_ON(ret);
1264
1265         skip_sum = btrfs_test_opt(root, NODATASUM) ||
1266                 btrfs_test_flag(inode, NODATASUM);
1267
1268         if (!(rw & (1 << BIO_RW))) {
1269
1270                 if (bio_flags & EXTENT_BIO_COMPRESSED)
1271                         return btrfs_submit_compressed_read(inode, bio,
1272                                                     mirror_num, bio_flags);
1273                 else if (!skip_sum)
1274                         btrfs_lookup_bio_sums(root, inode, bio);
1275                 goto mapit;
1276         } else if (!skip_sum) {
1277                 /* we're doing a write, do the async checksumming */
1278                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1279                                    inode, rw, bio, mirror_num,
1280                                    bio_flags, __btrfs_submit_bio_start,
1281                                    __btrfs_submit_bio_done);
1282         }
1283
1284 mapit:
1285         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1286 }
1287
1288 /*
1289  * given a list of ordered sums record them in the inode.  This happens
1290  * at IO completion time based on sums calculated at bio submission time.
1291  */
1292 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1293                              struct inode *inode, u64 file_offset,
1294                              struct list_head *list)
1295 {
1296         struct list_head *cur;
1297         struct btrfs_ordered_sum *sum;
1298
1299         btrfs_set_trans_block_group(trans, inode);
1300         list_for_each(cur, list) {
1301                 sum = list_entry(cur, struct btrfs_ordered_sum, list);
1302                 btrfs_csum_file_blocks(trans, BTRFS_I(inode)->root,
1303                                        inode, sum);
1304         }
1305         return 0;
1306 }
1307
1308 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end)
1309 {
1310         if ((end & (PAGE_CACHE_SIZE - 1)) == 0) {
1311                 WARN_ON(1);
1312         }
1313         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1314                                    GFP_NOFS);
1315 }
1316
1317 /* see btrfs_writepage_start_hook for details on why this is required */
1318 struct btrfs_writepage_fixup {
1319         struct page *page;
1320         struct btrfs_work work;
1321 };
1322
1323 void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1324 {
1325         struct btrfs_writepage_fixup *fixup;
1326         struct btrfs_ordered_extent *ordered;
1327         struct page *page;
1328         struct inode *inode;
1329         u64 page_start;
1330         u64 page_end;
1331
1332         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1333         page = fixup->page;
1334 again:
1335         lock_page(page);
1336         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1337                 ClearPageChecked(page);
1338                 goto out_page;
1339         }
1340
1341         inode = page->mapping->host;
1342         page_start = page_offset(page);
1343         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1344
1345         lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1346
1347         /* already ordered? We're done */
1348         if (test_range_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
1349                              EXTENT_ORDERED, 0)) {
1350                 goto out;
1351         }
1352
1353         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1354         if (ordered) {
1355                 unlock_extent(&BTRFS_I(inode)->io_tree, page_start,
1356                               page_end, GFP_NOFS);
1357                 unlock_page(page);
1358                 btrfs_start_ordered_extent(inode, ordered, 1);
1359                 goto again;
1360         }
1361
1362         btrfs_set_extent_delalloc(inode, page_start, page_end);
1363         ClearPageChecked(page);
1364 out:
1365         unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1366 out_page:
1367         unlock_page(page);
1368         page_cache_release(page);
1369 }
1370
1371 /*
1372  * There are a few paths in the higher layers of the kernel that directly
1373  * set the page dirty bit without asking the filesystem if it is a
1374  * good idea.  This causes problems because we want to make sure COW
1375  * properly happens and the data=ordered rules are followed.
1376  *
1377  * In our case any range that doesn't have the ORDERED bit set
1378  * hasn't been properly setup for IO.  We kick off an async process
1379  * to fix it up.  The async helper will wait for ordered extents, set
1380  * the delalloc bit and make it safe to write the page.
1381  */
1382 int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1383 {
1384         struct inode *inode = page->mapping->host;
1385         struct btrfs_writepage_fixup *fixup;
1386         struct btrfs_root *root = BTRFS_I(inode)->root;
1387         int ret;
1388
1389         ret = test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1390                              EXTENT_ORDERED, 0);
1391         if (ret)
1392                 return 0;
1393
1394         if (PageChecked(page))
1395                 return -EAGAIN;
1396
1397         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1398         if (!fixup)
1399                 return -EAGAIN;
1400
1401         SetPageChecked(page);
1402         page_cache_get(page);
1403         fixup->work.func = btrfs_writepage_fixup_worker;
1404         fixup->page = page;
1405         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1406         return -EAGAIN;
1407 }
1408
1409 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1410                                        struct inode *inode, u64 file_pos,
1411                                        u64 disk_bytenr, u64 disk_num_bytes,
1412                                        u64 num_bytes, u64 ram_bytes,
1413                                        u8 compression, u8 encryption,
1414                                        u16 other_encoding, int extent_type)
1415 {
1416         struct btrfs_root *root = BTRFS_I(inode)->root;
1417         struct btrfs_file_extent_item *fi;
1418         struct btrfs_path *path;
1419         struct extent_buffer *leaf;
1420         struct btrfs_key ins;
1421         u64 hint;
1422         int ret;
1423
1424         path = btrfs_alloc_path();
1425         BUG_ON(!path);
1426
1427         ret = btrfs_drop_extents(trans, root, inode, file_pos,
1428                                  file_pos + num_bytes, file_pos, &hint);
1429         BUG_ON(ret);
1430
1431         ins.objectid = inode->i_ino;
1432         ins.offset = file_pos;
1433         ins.type = BTRFS_EXTENT_DATA_KEY;
1434         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1435         BUG_ON(ret);
1436         leaf = path->nodes[0];
1437         fi = btrfs_item_ptr(leaf, path->slots[0],
1438                             struct btrfs_file_extent_item);
1439         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1440         btrfs_set_file_extent_type(leaf, fi, extent_type);
1441         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1442         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1443         btrfs_set_file_extent_offset(leaf, fi, 0);
1444         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1445         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1446         btrfs_set_file_extent_compression(leaf, fi, compression);
1447         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1448         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1449         btrfs_mark_buffer_dirty(leaf);
1450
1451         inode_add_bytes(inode, num_bytes);
1452         btrfs_drop_extent_cache(inode, file_pos, file_pos + num_bytes - 1, 0);
1453
1454         ins.objectid = disk_bytenr;
1455         ins.offset = disk_num_bytes;
1456         ins.type = BTRFS_EXTENT_ITEM_KEY;
1457         ret = btrfs_alloc_reserved_extent(trans, root, leaf->start,
1458                                           root->root_key.objectid,
1459                                           trans->transid, inode->i_ino, &ins);
1460         BUG_ON(ret);
1461
1462         btrfs_free_path(path);
1463         return 0;
1464 }
1465
1466 /* as ordered data IO finishes, this gets called so we can finish
1467  * an ordered extent if the range of bytes in the file it covers are
1468  * fully written.
1469  */
1470 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1471 {
1472         struct btrfs_root *root = BTRFS_I(inode)->root;
1473         struct btrfs_trans_handle *trans;
1474         struct btrfs_ordered_extent *ordered_extent;
1475         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1476         int compressed = 0;
1477         int ret;
1478
1479         ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
1480         if (!ret)
1481                 return 0;
1482
1483         trans = btrfs_join_transaction(root, 1);
1484
1485         ordered_extent = btrfs_lookup_ordered_extent(inode, start);
1486         BUG_ON(!ordered_extent);
1487         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
1488                 goto nocow;
1489
1490         lock_extent(io_tree, ordered_extent->file_offset,
1491                     ordered_extent->file_offset + ordered_extent->len - 1,
1492                     GFP_NOFS);
1493
1494         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1495                 compressed = 1;
1496         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1497                 BUG_ON(compressed);
1498                 ret = btrfs_mark_extent_written(trans, root, inode,
1499                                                 ordered_extent->file_offset,
1500                                                 ordered_extent->file_offset +
1501                                                 ordered_extent->len);
1502                 BUG_ON(ret);
1503         } else {
1504                 ret = insert_reserved_file_extent(trans, inode,
1505                                                 ordered_extent->file_offset,
1506                                                 ordered_extent->start,
1507                                                 ordered_extent->disk_len,
1508                                                 ordered_extent->len,
1509                                                 ordered_extent->len,
1510                                                 compressed, 0, 0,
1511                                                 BTRFS_FILE_EXTENT_REG);
1512                 BUG_ON(ret);
1513         }
1514         unlock_extent(io_tree, ordered_extent->file_offset,
1515                     ordered_extent->file_offset + ordered_extent->len - 1,
1516                     GFP_NOFS);
1517 nocow:
1518         add_pending_csums(trans, inode, ordered_extent->file_offset,
1519                           &ordered_extent->list);
1520
1521         mutex_lock(&BTRFS_I(inode)->extent_mutex);
1522         btrfs_ordered_update_i_size(inode, ordered_extent);
1523         btrfs_update_inode(trans, root, inode);
1524         btrfs_remove_ordered_extent(inode, ordered_extent);
1525         mutex_unlock(&BTRFS_I(inode)->extent_mutex);
1526
1527         /* once for us */
1528         btrfs_put_ordered_extent(ordered_extent);
1529         /* once for the tree */
1530         btrfs_put_ordered_extent(ordered_extent);
1531
1532         btrfs_end_transaction(trans, root);
1533         return 0;
1534 }
1535
1536 int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1537                                 struct extent_state *state, int uptodate)
1538 {
1539         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1540 }
1541
1542 /*
1543  * When IO fails, either with EIO or csum verification fails, we
1544  * try other mirrors that might have a good copy of the data.  This
1545  * io_failure_record is used to record state as we go through all the
1546  * mirrors.  If another mirror has good data, the page is set up to date
1547  * and things continue.  If a good mirror can't be found, the original
1548  * bio end_io callback is called to indicate things have failed.
1549  */
1550 struct io_failure_record {
1551         struct page *page;
1552         u64 start;
1553         u64 len;
1554         u64 logical;
1555         int last_mirror;
1556 };
1557
1558 int btrfs_io_failed_hook(struct bio *failed_bio,
1559                          struct page *page, u64 start, u64 end,
1560                          struct extent_state *state)
1561 {
1562         struct io_failure_record *failrec = NULL;
1563         u64 private;
1564         struct extent_map *em;
1565         struct inode *inode = page->mapping->host;
1566         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1567         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1568         struct bio *bio;
1569         int num_copies;
1570         int ret;
1571         int rw;
1572         u64 logical;
1573         unsigned long bio_flags = 0;
1574
1575         ret = get_state_private(failure_tree, start, &private);
1576         if (ret) {
1577                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1578                 if (!failrec)
1579                         return -ENOMEM;
1580                 failrec->start = start;
1581                 failrec->len = end - start + 1;
1582                 failrec->last_mirror = 0;
1583
1584                 spin_lock(&em_tree->lock);
1585                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1586                 if (em->start > start || em->start + em->len < start) {
1587                         free_extent_map(em);
1588                         em = NULL;
1589                 }
1590                 spin_unlock(&em_tree->lock);
1591
1592                 if (!em || IS_ERR(em)) {
1593                         kfree(failrec);
1594                         return -EIO;
1595                 }
1596                 logical = start - em->start;
1597                 logical = em->block_start + logical;
1598                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
1599                         bio_flags = EXTENT_BIO_COMPRESSED;
1600                 failrec->logical = logical;
1601                 free_extent_map(em);
1602                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1603                                 EXTENT_DIRTY, GFP_NOFS);
1604                 set_state_private(failure_tree, start,
1605                                  (u64)(unsigned long)failrec);
1606         } else {
1607                 failrec = (struct io_failure_record *)(unsigned long)private;
1608         }
1609         num_copies = btrfs_num_copies(
1610                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1611                               failrec->logical, failrec->len);
1612         failrec->last_mirror++;
1613         if (!state) {
1614                 spin_lock_irq(&BTRFS_I(inode)->io_tree.lock);
1615                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1616                                                     failrec->start,
1617                                                     EXTENT_LOCKED);
1618                 if (state && state->start != failrec->start)
1619                         state = NULL;
1620                 spin_unlock_irq(&BTRFS_I(inode)->io_tree.lock);
1621         }
1622         if (!state || failrec->last_mirror > num_copies) {
1623                 set_state_private(failure_tree, failrec->start, 0);
1624                 clear_extent_bits(failure_tree, failrec->start,
1625                                   failrec->start + failrec->len - 1,
1626                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1627                 kfree(failrec);
1628                 return -EIO;
1629         }
1630         bio = bio_alloc(GFP_NOFS, 1);
1631         bio->bi_private = state;
1632         bio->bi_end_io = failed_bio->bi_end_io;
1633         bio->bi_sector = failrec->logical >> 9;
1634         bio->bi_bdev = failed_bio->bi_bdev;
1635         bio->bi_size = 0;
1636         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1637         if (failed_bio->bi_rw & (1 << BIO_RW))
1638                 rw = WRITE;
1639         else
1640                 rw = READ;
1641
1642         BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1643                                                       failrec->last_mirror,
1644                                                       bio_flags);
1645         return 0;
1646 }
1647
1648 /*
1649  * each time an IO finishes, we do a fast check in the IO failure tree
1650  * to see if we need to process or clean up an io_failure_record
1651  */
1652 int btrfs_clean_io_failures(struct inode *inode, u64 start)
1653 {
1654         u64 private;
1655         u64 private_failure;
1656         struct io_failure_record *failure;
1657         int ret;
1658
1659         private = 0;
1660         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1661                              (u64)-1, 1, EXTENT_DIRTY)) {
1662                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1663                                         start, &private_failure);
1664                 if (ret == 0) {
1665                         failure = (struct io_failure_record *)(unsigned long)
1666                                    private_failure;
1667                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1668                                           failure->start, 0);
1669                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1670                                           failure->start,
1671                                           failure->start + failure->len - 1,
1672                                           EXTENT_DIRTY | EXTENT_LOCKED,
1673                                           GFP_NOFS);
1674                         kfree(failure);
1675                 }
1676         }
1677         return 0;
1678 }
1679
1680 /*
1681  * when reads are done, we need to check csums to verify the data is correct
1682  * if there's a match, we allow the bio to finish.  If not, we go through
1683  * the io_failure_record routines to find good copies
1684  */
1685 int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1686                                struct extent_state *state)
1687 {
1688         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1689         struct inode *inode = page->mapping->host;
1690         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1691         char *kaddr;
1692         u64 private = ~(u32)0;
1693         int ret;
1694         struct btrfs_root *root = BTRFS_I(inode)->root;
1695         u32 csum = ~(u32)0;
1696         unsigned long flags;
1697
1698         if (btrfs_test_opt(root, NODATASUM) ||
1699             btrfs_test_flag(inode, NODATASUM))
1700                 return 0;
1701         if (state && state->start == start) {
1702                 private = state->private;
1703                 ret = 0;
1704         } else {
1705                 ret = get_state_private(io_tree, start, &private);
1706         }
1707         local_irq_save(flags);
1708         kaddr = kmap_atomic(page, KM_IRQ0);
1709         if (ret) {
1710                 goto zeroit;
1711         }
1712         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
1713         btrfs_csum_final(csum, (char *)&csum);
1714         if (csum != private) {
1715                 goto zeroit;
1716         }
1717         kunmap_atomic(kaddr, KM_IRQ0);
1718         local_irq_restore(flags);
1719
1720         /* if the io failure tree for this inode is non-empty,
1721          * check to see if we've recovered from a failed IO
1722          */
1723         btrfs_clean_io_failures(inode, start);
1724         return 0;
1725
1726 zeroit:
1727         printk("btrfs csum failed ino %lu off %llu csum %u private %Lu\n",
1728                page->mapping->host->i_ino, (unsigned long long)start, csum,
1729                private);
1730         memset(kaddr + offset, 1, end - start + 1);
1731         flush_dcache_page(page);
1732         kunmap_atomic(kaddr, KM_IRQ0);
1733         local_irq_restore(flags);
1734         if (private == 0)
1735                 return 0;
1736         return -EIO;
1737 }
1738
1739 /*
1740  * This creates an orphan entry for the given inode in case something goes
1741  * wrong in the middle of an unlink/truncate.
1742  */
1743 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
1744 {
1745         struct btrfs_root *root = BTRFS_I(inode)->root;
1746         int ret = 0;
1747
1748         spin_lock(&root->list_lock);
1749
1750         /* already on the orphan list, we're good */
1751         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
1752                 spin_unlock(&root->list_lock);
1753                 return 0;
1754         }
1755
1756         list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1757
1758         spin_unlock(&root->list_lock);
1759
1760         /*
1761          * insert an orphan item to track this unlinked/truncated file
1762          */
1763         ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
1764
1765         return ret;
1766 }
1767
1768 /*
1769  * We have done the truncate/delete so we can go ahead and remove the orphan
1770  * item for this particular inode.
1771  */
1772 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
1773 {
1774         struct btrfs_root *root = BTRFS_I(inode)->root;
1775         int ret = 0;
1776
1777         spin_lock(&root->list_lock);
1778
1779         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
1780                 spin_unlock(&root->list_lock);
1781                 return 0;
1782         }
1783
1784         list_del_init(&BTRFS_I(inode)->i_orphan);
1785         if (!trans) {
1786                 spin_unlock(&root->list_lock);
1787                 return 0;
1788         }
1789
1790         spin_unlock(&root->list_lock);
1791
1792         ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
1793
1794         return ret;
1795 }
1796
1797 /*
1798  * this cleans up any orphans that may be left on the list from the last use
1799  * of this root.
1800  */
1801 void btrfs_orphan_cleanup(struct btrfs_root *root)
1802 {
1803         struct btrfs_path *path;
1804         struct extent_buffer *leaf;
1805         struct btrfs_item *item;
1806         struct btrfs_key key, found_key;
1807         struct btrfs_trans_handle *trans;
1808         struct inode *inode;
1809         int ret = 0, nr_unlink = 0, nr_truncate = 0;
1810
1811         path = btrfs_alloc_path();
1812         if (!path)
1813                 return;
1814         path->reada = -1;
1815
1816         key.objectid = BTRFS_ORPHAN_OBJECTID;
1817         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1818         key.offset = (u64)-1;
1819
1820
1821         while (1) {
1822                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1823                 if (ret < 0) {
1824                         printk(KERN_ERR "Error searching slot for orphan: %d"
1825                                "\n", ret);
1826                         break;
1827                 }
1828
1829                 /*
1830                  * if ret == 0 means we found what we were searching for, which
1831                  * is weird, but possible, so only screw with path if we didnt
1832                  * find the key and see if we have stuff that matches
1833                  */
1834                 if (ret > 0) {
1835                         if (path->slots[0] == 0)
1836                                 break;
1837                         path->slots[0]--;
1838                 }
1839
1840                 /* pull out the item */
1841                 leaf = path->nodes[0];
1842                 item = btrfs_item_nr(leaf, path->slots[0]);
1843                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1844
1845                 /* make sure the item matches what we want */
1846                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
1847                         break;
1848                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
1849                         break;
1850
1851                 /* release the path since we're done with it */
1852                 btrfs_release_path(root, path);
1853
1854                 /*
1855                  * this is where we are basically btrfs_lookup, without the
1856                  * crossing root thing.  we store the inode number in the
1857                  * offset of the orphan item.
1858                  */
1859                 inode = btrfs_iget_locked(root->fs_info->sb,
1860                                           found_key.offset, root);
1861                 if (!inode)
1862                         break;
1863
1864                 if (inode->i_state & I_NEW) {
1865                         BTRFS_I(inode)->root = root;
1866
1867                         /* have to set the location manually */
1868                         BTRFS_I(inode)->location.objectid = inode->i_ino;
1869                         BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
1870                         BTRFS_I(inode)->location.offset = 0;
1871
1872                         btrfs_read_locked_inode(inode);
1873                         unlock_new_inode(inode);
1874                 }
1875
1876                 /*
1877                  * add this inode to the orphan list so btrfs_orphan_del does
1878                  * the proper thing when we hit it
1879                  */
1880                 spin_lock(&root->list_lock);
1881                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1882                 spin_unlock(&root->list_lock);
1883
1884                 /*
1885                  * if this is a bad inode, means we actually succeeded in
1886                  * removing the inode, but not the orphan record, which means
1887                  * we need to manually delete the orphan since iput will just
1888                  * do a destroy_inode
1889                  */
1890                 if (is_bad_inode(inode)) {
1891                         trans = btrfs_start_transaction(root, 1);
1892                         btrfs_orphan_del(trans, inode);
1893                         btrfs_end_transaction(trans, root);
1894                         iput(inode);
1895                         continue;
1896                 }
1897
1898                 /* if we have links, this was a truncate, lets do that */
1899                 if (inode->i_nlink) {
1900                         nr_truncate++;
1901                         btrfs_truncate(inode);
1902                 } else {
1903                         nr_unlink++;
1904                 }
1905
1906                 /* this will do delete_inode and everything for us */
1907                 iput(inode);
1908         }
1909
1910         if (nr_unlink)
1911                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
1912         if (nr_truncate)
1913                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
1914
1915         btrfs_free_path(path);
1916 }
1917
1918 /*
1919  * read an inode from the btree into the in-memory inode
1920  */
1921 void btrfs_read_locked_inode(struct inode *inode)
1922 {
1923         struct btrfs_path *path;
1924         struct extent_buffer *leaf;
1925         struct btrfs_inode_item *inode_item;
1926         struct btrfs_timespec *tspec;
1927         struct btrfs_root *root = BTRFS_I(inode)->root;
1928         struct btrfs_key location;
1929         u64 alloc_group_block;
1930         u32 rdev;
1931         int ret;
1932
1933         path = btrfs_alloc_path();
1934         BUG_ON(!path);
1935         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
1936
1937         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
1938         if (ret)
1939                 goto make_bad;
1940
1941         leaf = path->nodes[0];
1942         inode_item = btrfs_item_ptr(leaf, path->slots[0],
1943                                     struct btrfs_inode_item);
1944
1945         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
1946         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
1947         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
1948         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
1949         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
1950
1951         tspec = btrfs_inode_atime(inode_item);
1952         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1953         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1954
1955         tspec = btrfs_inode_mtime(inode_item);
1956         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1957         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1958
1959         tspec = btrfs_inode_ctime(inode_item);
1960         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1961         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1962
1963         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
1964         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
1965         inode->i_generation = BTRFS_I(inode)->generation;
1966         inode->i_rdev = 0;
1967         rdev = btrfs_inode_rdev(leaf, inode_item);
1968
1969         BTRFS_I(inode)->index_cnt = (u64)-1;
1970
1971         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
1972         BTRFS_I(inode)->block_group = btrfs_lookup_block_group(root->fs_info,
1973                                                        alloc_group_block);
1974         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
1975         if (!BTRFS_I(inode)->block_group) {
1976                 BTRFS_I(inode)->block_group = btrfs_find_block_group(root,
1977                                                  NULL, 0,
1978                                                  BTRFS_BLOCK_GROUP_METADATA, 0);
1979         }
1980         btrfs_free_path(path);
1981         inode_item = NULL;
1982
1983         switch (inode->i_mode & S_IFMT) {
1984         case S_IFREG:
1985                 inode->i_mapping->a_ops = &btrfs_aops;
1986                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
1987                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
1988                 inode->i_fop = &btrfs_file_operations;
1989                 inode->i_op = &btrfs_file_inode_operations;
1990                 break;
1991         case S_IFDIR:
1992                 inode->i_fop = &btrfs_dir_file_operations;
1993                 if (root == root->fs_info->tree_root)
1994                         inode->i_op = &btrfs_dir_ro_inode_operations;
1995                 else
1996                         inode->i_op = &btrfs_dir_inode_operations;
1997                 break;
1998         case S_IFLNK:
1999                 inode->i_op = &btrfs_symlink_inode_operations;
2000                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2001                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2002                 break;
2003         default:
2004                 init_special_inode(inode, inode->i_mode, rdev);
2005                 break;
2006         }
2007         return;
2008
2009 make_bad:
2010         btrfs_free_path(path);
2011         make_bad_inode(inode);
2012 }
2013
2014 /*
2015  * given a leaf and an inode, copy the inode fields into the leaf
2016  */
2017 static void fill_inode_item(struct btrfs_trans_handle *trans,
2018                             struct extent_buffer *leaf,
2019                             struct btrfs_inode_item *item,
2020                             struct inode *inode)
2021 {
2022         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2023         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2024         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2025         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2026         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2027
2028         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2029                                inode->i_atime.tv_sec);
2030         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2031                                 inode->i_atime.tv_nsec);
2032
2033         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2034                                inode->i_mtime.tv_sec);
2035         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2036                                 inode->i_mtime.tv_nsec);
2037
2038         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2039                                inode->i_ctime.tv_sec);
2040         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2041                                 inode->i_ctime.tv_nsec);
2042
2043         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2044         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2045         btrfs_set_inode_transid(leaf, item, trans->transid);
2046         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2047         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2048         btrfs_set_inode_block_group(leaf, item,
2049                                     BTRFS_I(inode)->block_group->key.objectid);
2050 }
2051
2052 /*
2053  * copy everything in the in-memory inode into the btree.
2054  */
2055 int noinline btrfs_update_inode(struct btrfs_trans_handle *trans,
2056                               struct btrfs_root *root,
2057                               struct inode *inode)
2058 {
2059         struct btrfs_inode_item *inode_item;
2060         struct btrfs_path *path;
2061         struct extent_buffer *leaf;
2062         int ret;
2063
2064         path = btrfs_alloc_path();
2065         BUG_ON(!path);
2066         ret = btrfs_lookup_inode(trans, root, path,
2067                                  &BTRFS_I(inode)->location, 1);
2068         if (ret) {
2069                 if (ret > 0)
2070                         ret = -ENOENT;
2071                 goto failed;
2072         }
2073
2074         leaf = path->nodes[0];
2075         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2076                                   struct btrfs_inode_item);
2077
2078         fill_inode_item(trans, leaf, inode_item, inode);
2079         btrfs_mark_buffer_dirty(leaf);
2080         btrfs_set_inode_last_trans(trans, inode);
2081         ret = 0;
2082 failed:
2083         btrfs_free_path(path);
2084         return ret;
2085 }
2086
2087
2088 /*
2089  * unlink helper that gets used here in inode.c and in the tree logging
2090  * recovery code.  It remove a link in a directory with a given name, and
2091  * also drops the back refs in the inode to the directory
2092  */
2093 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2094                        struct btrfs_root *root,
2095                        struct inode *dir, struct inode *inode,
2096                        const char *name, int name_len)
2097 {
2098         struct btrfs_path *path;
2099         int ret = 0;
2100         struct extent_buffer *leaf;
2101         struct btrfs_dir_item *di;
2102         struct btrfs_key key;
2103         u64 index;
2104
2105         path = btrfs_alloc_path();
2106         if (!path) {
2107                 ret = -ENOMEM;
2108                 goto err;
2109         }
2110
2111         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2112                                     name, name_len, -1);
2113         if (IS_ERR(di)) {
2114                 ret = PTR_ERR(di);
2115                 goto err;
2116         }
2117         if (!di) {
2118                 ret = -ENOENT;
2119                 goto err;
2120         }
2121         leaf = path->nodes[0];
2122         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2123         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2124         if (ret)
2125                 goto err;
2126         btrfs_release_path(root, path);
2127
2128         ret = btrfs_del_inode_ref(trans, root, name, name_len,
2129                                   inode->i_ino,
2130                                   dir->i_ino, &index);
2131         if (ret) {
2132                 printk("failed to delete reference to %.*s, "
2133                        "inode %lu parent %lu\n", name_len, name,
2134                        inode->i_ino, dir->i_ino);
2135                 goto err;
2136         }
2137
2138         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2139                                          index, name, name_len, -1);
2140         if (IS_ERR(di)) {
2141                 ret = PTR_ERR(di);
2142                 goto err;
2143         }
2144         if (!di) {
2145                 ret = -ENOENT;
2146                 goto err;
2147         }
2148         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2149         btrfs_release_path(root, path);
2150
2151         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2152                                          inode, dir->i_ino);
2153         BUG_ON(ret != 0 && ret != -ENOENT);
2154         if (ret != -ENOENT)
2155                 BTRFS_I(dir)->log_dirty_trans = trans->transid;
2156
2157         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2158                                            dir, index);
2159         BUG_ON(ret);
2160 err:
2161         btrfs_free_path(path);
2162         if (ret)
2163                 goto out;
2164
2165         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2166         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2167         btrfs_update_inode(trans, root, dir);
2168         btrfs_drop_nlink(inode);
2169         ret = btrfs_update_inode(trans, root, inode);
2170         dir->i_sb->s_dirt = 1;
2171 out:
2172         return ret;
2173 }
2174
2175 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2176 {
2177         struct btrfs_root *root;
2178         struct btrfs_trans_handle *trans;
2179         struct inode *inode = dentry->d_inode;
2180         int ret;
2181         unsigned long nr = 0;
2182
2183         root = BTRFS_I(dir)->root;
2184
2185         ret = btrfs_check_free_space(root, 1, 1);
2186         if (ret)
2187                 goto fail;
2188
2189         trans = btrfs_start_transaction(root, 1);
2190
2191         btrfs_set_trans_block_group(trans, dir);
2192         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2193                                  dentry->d_name.name, dentry->d_name.len);
2194
2195         if (inode->i_nlink == 0)
2196                 ret = btrfs_orphan_add(trans, inode);
2197
2198         nr = trans->blocks_used;
2199
2200         btrfs_end_transaction_throttle(trans, root);
2201 fail:
2202         btrfs_btree_balance_dirty(root, nr);
2203         return ret;
2204 }
2205
2206 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2207 {
2208         struct inode *inode = dentry->d_inode;
2209         int err = 0;
2210         int ret;
2211         struct btrfs_root *root = BTRFS_I(dir)->root;
2212         struct btrfs_trans_handle *trans;
2213         unsigned long nr = 0;
2214
2215         /*
2216          * the FIRST_FREE_OBJECTID check makes sure we don't try to rmdir
2217          * the root of a subvolume or snapshot
2218          */
2219         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2220             inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) {
2221                 return -ENOTEMPTY;
2222         }
2223
2224         ret = btrfs_check_free_space(root, 1, 1);
2225         if (ret)
2226                 goto fail;
2227
2228         trans = btrfs_start_transaction(root, 1);
2229         btrfs_set_trans_block_group(trans, dir);
2230
2231         err = btrfs_orphan_add(trans, inode);
2232         if (err)
2233                 goto fail_trans;
2234
2235         /* now the directory is empty */
2236         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2237                                  dentry->d_name.name, dentry->d_name.len);
2238         if (!err) {
2239                 btrfs_i_size_write(inode, 0);
2240         }
2241
2242 fail_trans:
2243         nr = trans->blocks_used;
2244         ret = btrfs_end_transaction_throttle(trans, root);
2245 fail:
2246         btrfs_btree_balance_dirty(root, nr);
2247
2248         if (ret && !err)
2249                 err = ret;
2250         return err;
2251 }
2252
2253 /*
2254  * when truncating bytes in a file, it is possible to avoid reading
2255  * the leaves that contain only checksum items.  This can be the
2256  * majority of the IO required to delete a large file, but it must
2257  * be done carefully.
2258  *
2259  * The keys in the level just above the leaves are checked to make sure
2260  * the lowest key in a given leaf is a csum key, and starts at an offset
2261  * after the new  size.
2262  *
2263  * Then the key for the next leaf is checked to make sure it also has
2264  * a checksum item for the same file.  If it does, we know our target leaf
2265  * contains only checksum items, and it can be safely freed without reading
2266  * it.
2267  *
2268  * This is just an optimization targeted at large files.  It may do
2269  * nothing.  It will return 0 unless things went badly.
2270  */
2271 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
2272                                      struct btrfs_root *root,
2273                                      struct btrfs_path *path,
2274                                      struct inode *inode, u64 new_size)
2275 {
2276         struct btrfs_key key;
2277         int ret;
2278         int nritems;
2279         struct btrfs_key found_key;
2280         struct btrfs_key other_key;
2281         struct btrfs_leaf_ref *ref;
2282         u64 leaf_gen;
2283         u64 leaf_start;
2284
2285         path->lowest_level = 1;
2286         key.objectid = inode->i_ino;
2287         key.type = BTRFS_CSUM_ITEM_KEY;
2288         key.offset = new_size;
2289 again:
2290         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2291         if (ret < 0)
2292                 goto out;
2293
2294         if (path->nodes[1] == NULL) {
2295                 ret = 0;
2296                 goto out;
2297         }
2298         ret = 0;
2299         btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
2300         nritems = btrfs_header_nritems(path->nodes[1]);
2301
2302         if (!nritems)
2303                 goto out;
2304
2305         if (path->slots[1] >= nritems)
2306                 goto next_node;
2307
2308         /* did we find a key greater than anything we want to delete? */
2309         if (found_key.objectid > inode->i_ino ||
2310            (found_key.objectid == inode->i_ino && found_key.type > key.type))
2311                 goto out;
2312
2313         /* we check the next key in the node to make sure the leave contains
2314          * only checksum items.  This comparison doesn't work if our
2315          * leaf is the last one in the node
2316          */
2317         if (path->slots[1] + 1 >= nritems) {
2318 next_node:
2319                 /* search forward from the last key in the node, this
2320                  * will bring us into the next node in the tree
2321                  */
2322                 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
2323
2324                 /* unlikely, but we inc below, so check to be safe */
2325                 if (found_key.offset == (u64)-1)
2326                         goto out;
2327
2328                 /* search_forward needs a path with locks held, do the
2329                  * search again for the original key.  It is possible
2330                  * this will race with a balance and return a path that
2331                  * we could modify, but this drop is just an optimization
2332                  * and is allowed to miss some leaves.
2333                  */
2334                 btrfs_release_path(root, path);
2335                 found_key.offset++;
2336
2337                 /* setup a max key for search_forward */
2338                 other_key.offset = (u64)-1;
2339                 other_key.type = key.type;
2340                 other_key.objectid = key.objectid;
2341
2342                 path->keep_locks = 1;
2343                 ret = btrfs_search_forward(root, &found_key, &other_key,
2344                                            path, 0, 0);
2345                 path->keep_locks = 0;
2346                 if (ret || found_key.objectid != key.objectid ||
2347                     found_key.type != key.type) {
2348                         ret = 0;
2349                         goto out;
2350                 }
2351
2352                 key.offset = found_key.offset;
2353                 btrfs_release_path(root, path);
2354                 cond_resched();
2355                 goto again;
2356         }
2357
2358         /* we know there's one more slot after us in the tree,
2359          * read that key so we can verify it is also a checksum item
2360          */
2361         btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
2362
2363         if (found_key.objectid < inode->i_ino)
2364                 goto next_key;
2365
2366         if (found_key.type != key.type || found_key.offset < new_size)
2367                 goto next_key;
2368
2369         /*
2370          * if the key for the next leaf isn't a csum key from this objectid,
2371          * we can't be sure there aren't good items inside this leaf.
2372          * Bail out
2373          */
2374         if (other_key.objectid != inode->i_ino || other_key.type != key.type)
2375                 goto out;
2376
2377         leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
2378         leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
2379         /*
2380          * it is safe to delete this leaf, it contains only
2381          * csum items from this inode at an offset >= new_size
2382          */
2383         ret = btrfs_del_leaf(trans, root, path, leaf_start);
2384         BUG_ON(ret);
2385
2386         if (root->ref_cows && leaf_gen < trans->transid) {
2387                 ref = btrfs_alloc_leaf_ref(root, 0);
2388                 if (ref) {
2389                         ref->root_gen = root->root_key.offset;
2390                         ref->bytenr = leaf_start;
2391                         ref->owner = 0;
2392                         ref->generation = leaf_gen;
2393                         ref->nritems = 0;
2394
2395                         ret = btrfs_add_leaf_ref(root, ref, 0);
2396                         WARN_ON(ret);
2397                         btrfs_free_leaf_ref(root, ref);
2398                 } else {
2399                         WARN_ON(1);
2400                 }
2401         }
2402 next_key:
2403         btrfs_release_path(root, path);
2404
2405         if (other_key.objectid == inode->i_ino &&
2406             other_key.type == key.type && other_key.offset > key.offset) {
2407                 key.offset = other_key.offset;
2408                 cond_resched();
2409                 goto again;
2410         }
2411         ret = 0;
2412 out:
2413         /* fixup any changes we've made to the path */
2414         path->lowest_level = 0;
2415         path->keep_locks = 0;
2416         btrfs_release_path(root, path);
2417         return ret;
2418 }
2419
2420 /*
2421  * this can truncate away extent items, csum items and directory items.
2422  * It starts at a high offset and removes keys until it can't find
2423  * any higher than new_size
2424  *
2425  * csum items that cross the new i_size are truncated to the new size
2426  * as well.
2427  *
2428  * min_type is the minimum key type to truncate down to.  If set to 0, this
2429  * will kill all the items on this inode, including the INODE_ITEM_KEY.
2430  */
2431 noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2432                                         struct btrfs_root *root,
2433                                         struct inode *inode,
2434                                         u64 new_size, u32 min_type)
2435 {
2436         int ret;
2437         struct btrfs_path *path;
2438         struct btrfs_key key;
2439         struct btrfs_key found_key;
2440         u32 found_type;
2441         struct extent_buffer *leaf;
2442         struct btrfs_file_extent_item *fi;
2443         u64 extent_start = 0;
2444         u64 extent_num_bytes = 0;
2445         u64 item_end = 0;
2446         u64 root_gen = 0;
2447         u64 root_owner = 0;
2448         int found_extent;
2449         int del_item;
2450         int pending_del_nr = 0;
2451         int pending_del_slot = 0;
2452         int extent_type = -1;
2453         int encoding;
2454         u64 mask = root->sectorsize - 1;
2455
2456         if (root->ref_cows)
2457                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
2458         path = btrfs_alloc_path();
2459         path->reada = -1;
2460         BUG_ON(!path);
2461
2462         /* FIXME, add redo link to tree so we don't leak on crash */
2463         key.objectid = inode->i_ino;
2464         key.offset = (u64)-1;
2465         key.type = (u8)-1;
2466
2467         btrfs_init_path(path);
2468
2469         ret = drop_csum_leaves(trans, root, path, inode, new_size);
2470         BUG_ON(ret);
2471
2472 search_again:
2473         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2474         if (ret < 0) {
2475                 goto error;
2476         }
2477         if (ret > 0) {
2478                 /* there are no items in the tree for us to truncate, we're
2479                  * done
2480                  */
2481                 if (path->slots[0] == 0) {
2482                         ret = 0;
2483                         goto error;
2484                 }
2485                 path->slots[0]--;
2486         }
2487
2488         while(1) {
2489                 fi = NULL;
2490                 leaf = path->nodes[0];
2491                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2492                 found_type = btrfs_key_type(&found_key);
2493                 encoding = 0;
2494
2495                 if (found_key.objectid != inode->i_ino)
2496                         break;
2497
2498                 if (found_type < min_type)
2499                         break;
2500
2501                 item_end = found_key.offset;
2502                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
2503                         fi = btrfs_item_ptr(leaf, path->slots[0],
2504                                             struct btrfs_file_extent_item);
2505                         extent_type = btrfs_file_extent_type(leaf, fi);
2506                         encoding = btrfs_file_extent_compression(leaf, fi);
2507                         encoding |= btrfs_file_extent_encryption(leaf, fi);
2508                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
2509
2510                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2511                                 item_end +=
2512                                     btrfs_file_extent_num_bytes(leaf, fi);
2513                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2514                                 item_end += btrfs_file_extent_inline_len(leaf,
2515                                                                          fi);
2516                         }
2517                         item_end--;
2518                 }
2519                 if (found_type == BTRFS_CSUM_ITEM_KEY) {
2520                         ret = btrfs_csum_truncate(trans, root, path,
2521                                                   new_size);
2522                         BUG_ON(ret);
2523                 }
2524                 if (item_end < new_size) {
2525                         if (found_type == BTRFS_DIR_ITEM_KEY) {
2526                                 found_type = BTRFS_INODE_ITEM_KEY;
2527                         } else if (found_type == BTRFS_EXTENT_ITEM_KEY) {
2528                                 found_type = BTRFS_CSUM_ITEM_KEY;
2529                         } else if (found_type == BTRFS_EXTENT_DATA_KEY) {
2530                                 found_type = BTRFS_XATTR_ITEM_KEY;
2531                         } else if (found_type == BTRFS_XATTR_ITEM_KEY) {
2532                                 found_type = BTRFS_INODE_REF_KEY;
2533                         } else if (found_type) {
2534                                 found_type--;
2535                         } else {
2536                                 break;
2537                         }
2538                         btrfs_set_key_type(&key, found_type);
2539                         goto next;
2540                 }
2541                 if (found_key.offset >= new_size)
2542                         del_item = 1;
2543                 else
2544                         del_item = 0;
2545                 found_extent = 0;
2546
2547                 /* FIXME, shrink the extent if the ref count is only 1 */
2548                 if (found_type != BTRFS_EXTENT_DATA_KEY)
2549                         goto delete;
2550
2551                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2552                         u64 num_dec;
2553                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
2554                         if (!del_item && !encoding) {
2555                                 u64 orig_num_bytes =
2556                                         btrfs_file_extent_num_bytes(leaf, fi);
2557                                 extent_num_bytes = new_size -
2558                                         found_key.offset + root->sectorsize - 1;
2559                                 extent_num_bytes = extent_num_bytes &
2560                                         ~((u64)root->sectorsize - 1);
2561                                 btrfs_set_file_extent_num_bytes(leaf, fi,
2562                                                          extent_num_bytes);
2563                                 num_dec = (orig_num_bytes -
2564                                            extent_num_bytes);
2565                                 if (root->ref_cows && extent_start != 0)
2566                                         inode_sub_bytes(inode, num_dec);
2567                                 btrfs_mark_buffer_dirty(leaf);
2568                         } else {
2569                                 extent_num_bytes =
2570                                         btrfs_file_extent_disk_num_bytes(leaf,
2571                                                                          fi);
2572                                 /* FIXME blocksize != 4096 */
2573                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
2574                                 if (extent_start != 0) {
2575                                         found_extent = 1;
2576                                         if (root->ref_cows)
2577                                                 inode_sub_bytes(inode, num_dec);
2578                                 }
2579                                 root_gen = btrfs_header_generation(leaf);
2580                                 root_owner = btrfs_header_owner(leaf);
2581                         }
2582                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2583                         /*
2584                          * we can't truncate inline items that have had
2585                          * special encodings
2586                          */
2587                         if (!del_item &&
2588                             btrfs_file_extent_compression(leaf, fi) == 0 &&
2589                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
2590                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
2591                                 u32 size = new_size - found_key.offset;
2592
2593                                 if (root->ref_cows) {
2594                                         inode_sub_bytes(inode, item_end + 1 -
2595                                                         new_size);
2596                                 }
2597                                 size =
2598                                     btrfs_file_extent_calc_inline_size(size);
2599                                 ret = btrfs_truncate_item(trans, root, path,
2600                                                           size, 1);
2601                                 BUG_ON(ret);
2602                         } else if (root->ref_cows) {
2603                                 inode_sub_bytes(inode, item_end + 1 -
2604                                                 found_key.offset);
2605                         }
2606                 }
2607 delete:
2608                 if (del_item) {
2609                         if (!pending_del_nr) {
2610                                 /* no pending yet, add ourselves */
2611                                 pending_del_slot = path->slots[0];
2612                                 pending_del_nr = 1;
2613                         } else if (pending_del_nr &&
2614                                    path->slots[0] + 1 == pending_del_slot) {
2615                                 /* hop on the pending chunk */
2616                                 pending_del_nr++;
2617                                 pending_del_slot = path->slots[0];
2618                         } else {
2619                                 printk("bad pending slot %d pending_del_nr %d pending_del_slot %d\n", path->slots[0], pending_del_nr, pending_del_slot);
2620                         }
2621                 } else {
2622                         break;
2623                 }
2624                 if (found_extent) {
2625                         ret = btrfs_free_extent(trans, root, extent_start,
2626                                                 extent_num_bytes,
2627                                                 leaf->start, root_owner,
2628                                                 root_gen, inode->i_ino, 0);
2629                         BUG_ON(ret);
2630                 }
2631 next:
2632                 if (path->slots[0] == 0) {
2633                         if (pending_del_nr)
2634                                 goto del_pending;
2635                         btrfs_release_path(root, path);
2636                         goto search_again;
2637                 }
2638
2639                 path->slots[0]--;
2640                 if (pending_del_nr &&
2641                     path->slots[0] + 1 != pending_del_slot) {
2642                         struct btrfs_key debug;
2643 del_pending:
2644                         btrfs_item_key_to_cpu(path->nodes[0], &debug,
2645                                               pending_del_slot);
2646                         ret = btrfs_del_items(trans, root, path,
2647                                               pending_del_slot,
2648                                               pending_del_nr);
2649                         BUG_ON(ret);
2650                         pending_del_nr = 0;
2651                         btrfs_release_path(root, path);
2652                         goto search_again;
2653                 }
2654         }
2655         ret = 0;
2656 error:
2657         if (pending_del_nr) {
2658                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
2659                                       pending_del_nr);
2660         }
2661         btrfs_free_path(path);
2662         inode->i_sb->s_dirt = 1;
2663         return ret;
2664 }
2665
2666 /*
2667  * taken from block_truncate_page, but does cow as it zeros out
2668  * any bytes left in the last page in the file.
2669  */
2670 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
2671 {
2672         struct inode *inode = mapping->host;
2673         struct btrfs_root *root = BTRFS_I(inode)->root;
2674         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2675         struct btrfs_ordered_extent *ordered;
2676         char *kaddr;
2677         u32 blocksize = root->sectorsize;
2678         pgoff_t index = from >> PAGE_CACHE_SHIFT;
2679         unsigned offset = from & (PAGE_CACHE_SIZE-1);
2680         struct page *page;
2681         int ret = 0;
2682         u64 page_start;
2683         u64 page_end;
2684
2685         if ((offset & (blocksize - 1)) == 0)
2686                 goto out;
2687
2688         ret = -ENOMEM;
2689 again:
2690         page = grab_cache_page(mapping, index);
2691         if (!page)
2692                 goto out;
2693
2694         page_start = page_offset(page);
2695         page_end = page_start + PAGE_CACHE_SIZE - 1;
2696
2697         if (!PageUptodate(page)) {
2698                 ret = btrfs_readpage(NULL, page);
2699                 lock_page(page);
2700                 if (page->mapping != mapping) {
2701                         unlock_page(page);
2702                         page_cache_release(page);
2703                         goto again;
2704                 }
2705                 if (!PageUptodate(page)) {
2706                         ret = -EIO;
2707                         goto out_unlock;
2708                 }
2709         }
2710         wait_on_page_writeback(page);
2711
2712         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2713         set_page_extent_mapped(page);
2714
2715         ordered = btrfs_lookup_ordered_extent(inode, page_start);
2716         if (ordered) {
2717                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2718                 unlock_page(page);
2719                 page_cache_release(page);
2720                 btrfs_start_ordered_extent(inode, ordered, 1);
2721                 btrfs_put_ordered_extent(ordered);
2722                 goto again;
2723         }
2724
2725         btrfs_set_extent_delalloc(inode, page_start, page_end);
2726         ret = 0;
2727         if (offset != PAGE_CACHE_SIZE) {
2728                 kaddr = kmap(page);
2729                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
2730                 flush_dcache_page(page);
2731                 kunmap(page);
2732         }
2733         ClearPageChecked(page);
2734         set_page_dirty(page);
2735         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2736
2737 out_unlock:
2738         unlock_page(page);
2739         page_cache_release(page);
2740 out:
2741         return ret;
2742 }
2743
2744 int btrfs_cont_expand(struct inode *inode, loff_t size)
2745 {
2746         struct btrfs_trans_handle *trans;
2747         struct btrfs_root *root = BTRFS_I(inode)->root;
2748         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2749         struct extent_map *em;
2750         u64 mask = root->sectorsize - 1;
2751         u64 hole_start = (inode->i_size + mask) & ~mask;
2752         u64 block_end = (size + mask) & ~mask;
2753         u64 last_byte;
2754         u64 cur_offset;
2755         u64 hole_size;
2756         int err;
2757
2758         if (size <= hole_start)
2759                 return 0;
2760
2761         err = btrfs_check_free_space(root, 1, 0);
2762         if (err)
2763                 return err;
2764
2765         btrfs_truncate_page(inode->i_mapping, inode->i_size);
2766
2767         while (1) {
2768                 struct btrfs_ordered_extent *ordered;
2769                 btrfs_wait_ordered_range(inode, hole_start,
2770                                          block_end - hole_start);
2771                 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2772                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
2773                 if (!ordered)
2774                         break;
2775                 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2776                 btrfs_put_ordered_extent(ordered);
2777         }
2778
2779         trans = btrfs_start_transaction(root, 1);
2780         btrfs_set_trans_block_group(trans, inode);
2781
2782         cur_offset = hole_start;
2783         while (1) {
2784                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2785                                 block_end - cur_offset, 0);
2786                 BUG_ON(IS_ERR(em) || !em);
2787                 last_byte = min(extent_map_end(em), block_end);
2788                 last_byte = (last_byte + mask) & ~mask;
2789                 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
2790                         u64 hint_byte = 0;
2791                         hole_size = last_byte - cur_offset;
2792                         err = btrfs_drop_extents(trans, root, inode,
2793                                                  cur_offset,
2794                                                  cur_offset + hole_size,
2795                                                  cur_offset, &hint_byte);
2796                         if (err)
2797                                 break;
2798                         err = btrfs_insert_file_extent(trans, root,
2799                                         inode->i_ino, cur_offset, 0,
2800                                         0, hole_size, 0, hole_size,
2801                                         0, 0, 0);
2802                         btrfs_drop_extent_cache(inode, hole_start,
2803                                         last_byte - 1, 0);
2804                 }
2805                 free_extent_map(em);
2806                 cur_offset = last_byte;
2807                 if (err || cur_offset >= block_end)
2808                         break;
2809         }
2810
2811         btrfs_end_transaction(trans, root);
2812         unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2813         return err;
2814 }
2815
2816 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
2817 {
2818         struct inode *inode = dentry->d_inode;
2819         int err;
2820
2821         err = inode_change_ok(inode, attr);
2822         if (err)
2823                 return err;
2824
2825         if (S_ISREG(inode->i_mode) &&
2826             attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
2827                 err = btrfs_cont_expand(inode, attr->ia_size);
2828                 if (err)
2829                         return err;
2830         }
2831
2832         err = inode_setattr(inode, attr);
2833
2834         if (!err && ((attr->ia_valid & ATTR_MODE)))
2835                 err = btrfs_acl_chmod(inode);
2836         return err;
2837 }
2838
2839 void btrfs_delete_inode(struct inode *inode)
2840 {
2841         struct btrfs_trans_handle *trans;
2842         struct btrfs_root *root = BTRFS_I(inode)->root;
2843         unsigned long nr;
2844         int ret;
2845
2846         truncate_inode_pages(&inode->i_data, 0);
2847         if (is_bad_inode(inode)) {
2848                 btrfs_orphan_del(NULL, inode);
2849                 goto no_delete;
2850         }
2851         btrfs_wait_ordered_range(inode, 0, (u64)-1);
2852
2853         btrfs_i_size_write(inode, 0);
2854         trans = btrfs_start_transaction(root, 1);
2855
2856         btrfs_set_trans_block_group(trans, inode);
2857         ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0);
2858         if (ret) {
2859                 btrfs_orphan_del(NULL, inode);
2860                 goto no_delete_lock;
2861         }
2862
2863         btrfs_orphan_del(trans, inode);
2864
2865         nr = trans->blocks_used;
2866         clear_inode(inode);
2867
2868         btrfs_end_transaction(trans, root);
2869         btrfs_btree_balance_dirty(root, nr);
2870         return;
2871
2872 no_delete_lock:
2873         nr = trans->blocks_used;
2874         btrfs_end_transaction(trans, root);
2875         btrfs_btree_balance_dirty(root, nr);
2876 no_delete:
2877         clear_inode(inode);
2878 }
2879
2880 /*
2881  * this returns the key found in the dir entry in the location pointer.
2882  * If no dir entries were found, location->objectid is 0.
2883  */
2884 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
2885                                struct btrfs_key *location)
2886 {
2887         const char *name = dentry->d_name.name;
2888         int namelen = dentry->d_name.len;
2889         struct btrfs_dir_item *di;
2890         struct btrfs_path *path;
2891         struct btrfs_root *root = BTRFS_I(dir)->root;
2892         int ret = 0;
2893
2894         path = btrfs_alloc_path();
2895         BUG_ON(!path);
2896
2897         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
2898                                     namelen, 0);
2899         if (IS_ERR(di))
2900                 ret = PTR_ERR(di);
2901         if (!di || IS_ERR(di)) {
2902                 goto out_err;
2903         }
2904         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
2905 out:
2906         btrfs_free_path(path);
2907         return ret;
2908 out_err:
2909         location->objectid = 0;
2910         goto out;
2911 }
2912
2913 /*
2914  * when we hit a tree root in a directory, the btrfs part of the inode
2915  * needs to be changed to reflect the root directory of the tree root.  This
2916  * is kind of like crossing a mount point.
2917  */
2918 static int fixup_tree_root_location(struct btrfs_root *root,
2919                              struct btrfs_key *location,
2920                              struct btrfs_root **sub_root,
2921                              struct dentry *dentry)
2922 {
2923         struct btrfs_root_item *ri;
2924
2925         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
2926                 return 0;
2927         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
2928                 return 0;
2929
2930         *sub_root = btrfs_read_fs_root(root->fs_info, location,
2931                                         dentry->d_name.name,
2932                                         dentry->d_name.len);
2933         if (IS_ERR(*sub_root))
2934                 return PTR_ERR(*sub_root);
2935
2936         ri = &(*sub_root)->root_item;
2937         location->objectid = btrfs_root_dirid(ri);
2938         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
2939         location->offset = 0;
2940
2941         return 0;
2942 }
2943
2944 static noinline void init_btrfs_i(struct inode *inode)
2945 {
2946         struct btrfs_inode *bi = BTRFS_I(inode);
2947
2948         bi->i_acl = NULL;
2949         bi->i_default_acl = NULL;
2950
2951         bi->generation = 0;
2952         bi->last_trans = 0;
2953         bi->logged_trans = 0;
2954         bi->delalloc_bytes = 0;
2955         bi->disk_i_size = 0;
2956         bi->flags = 0;
2957         bi->index_cnt = (u64)-1;
2958         bi->log_dirty_trans = 0;
2959         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
2960         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
2961                              inode->i_mapping, GFP_NOFS);
2962         extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
2963                              inode->i_mapping, GFP_NOFS);
2964         INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
2965         btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
2966         mutex_init(&BTRFS_I(inode)->csum_mutex);
2967         mutex_init(&BTRFS_I(inode)->extent_mutex);
2968         mutex_init(&BTRFS_I(inode)->log_mutex);
2969 }
2970
2971 static int btrfs_init_locked_inode(struct inode *inode, void *p)
2972 {
2973         struct btrfs_iget_args *args = p;
2974         inode->i_ino = args->ino;
2975         init_btrfs_i(inode);
2976         BTRFS_I(inode)->root = args->root;
2977         return 0;
2978 }
2979
2980 static int btrfs_find_actor(struct inode *inode, void *opaque)
2981 {
2982         struct btrfs_iget_args *args = opaque;
2983         return (args->ino == inode->i_ino &&
2984                 args->root == BTRFS_I(inode)->root);
2985 }
2986
2987 struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
2988                             struct btrfs_root *root, int wait)
2989 {
2990         struct inode *inode;
2991         struct btrfs_iget_args args;
2992         args.ino = objectid;
2993         args.root = root;
2994
2995         if (wait) {
2996                 inode = ilookup5(s, objectid, btrfs_find_actor,
2997                                  (void *)&args);
2998         } else {
2999                 inode = ilookup5_nowait(s, objectid, btrfs_find_actor,
3000                                         (void *)&args);
3001         }
3002         return inode;
3003 }
3004
3005 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
3006                                 struct btrfs_root *root)
3007 {
3008         struct inode *inode;
3009         struct btrfs_iget_args args;
3010         args.ino = objectid;
3011         args.root = root;
3012
3013         inode = iget5_locked(s, objectid, btrfs_find_actor,
3014                              btrfs_init_locked_inode,
3015                              (void *)&args);
3016         return inode;
3017 }
3018
3019 /* Get an inode object given its location and corresponding root.
3020  * Returns in *is_new if the inode was read from disk
3021  */
3022 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3023                          struct btrfs_root *root, int *is_new)
3024 {
3025         struct inode *inode;
3026
3027         inode = btrfs_iget_locked(s, location->objectid, root);
3028         if (!inode)
3029                 return ERR_PTR(-EACCES);
3030
3031         if (inode->i_state & I_NEW) {
3032                 BTRFS_I(inode)->root = root;
3033                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3034                 btrfs_read_locked_inode(inode);
3035                 unlock_new_inode(inode);
3036                 if (is_new)
3037                         *is_new = 1;
3038         } else {
3039                 if (is_new)
3040                         *is_new = 0;
3041         }
3042
3043         return inode;
3044 }
3045
3046 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3047 {
3048         struct inode * inode;
3049         struct btrfs_inode *bi = BTRFS_I(dir);
3050         struct btrfs_root *root = bi->root;
3051         struct btrfs_root *sub_root = root;
3052         struct btrfs_key location;
3053         int ret, new;
3054
3055         if (dentry->d_name.len > BTRFS_NAME_LEN)
3056                 return ERR_PTR(-ENAMETOOLONG);
3057
3058         ret = btrfs_inode_by_name(dir, dentry, &location);
3059
3060         if (ret < 0)
3061                 return ERR_PTR(ret);
3062
3063         inode = NULL;
3064         if (location.objectid) {
3065                 ret = fixup_tree_root_location(root, &location, &sub_root,
3066                                                 dentry);
3067                 if (ret < 0)
3068                         return ERR_PTR(ret);
3069                 if (ret > 0)
3070                         return ERR_PTR(-ENOENT);
3071                 inode = btrfs_iget(dir->i_sb, &location, sub_root, &new);
3072                 if (IS_ERR(inode))
3073                         return ERR_CAST(inode);
3074         }
3075         return inode;
3076 }
3077
3078 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3079                                    struct nameidata *nd)
3080 {
3081         struct inode *inode;
3082
3083         if (dentry->d_name.len > BTRFS_NAME_LEN)
3084                 return ERR_PTR(-ENAMETOOLONG);
3085
3086         inode = btrfs_lookup_dentry(dir, dentry);
3087         if (IS_ERR(inode))
3088                 return ERR_CAST(inode);
3089
3090         return d_splice_alias(inode, dentry);
3091 }
3092
3093 static unsigned char btrfs_filetype_table[] = {
3094         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
3095 };
3096
3097 static int btrfs_real_readdir(struct file *filp, void *dirent,
3098                               filldir_t filldir)
3099 {
3100         struct inode *inode = filp->f_dentry->d_inode;
3101         struct btrfs_root *root = BTRFS_I(inode)->root;
3102         struct btrfs_item *item;
3103         struct btrfs_dir_item *di;
3104         struct btrfs_key key;
3105         struct btrfs_key found_key;
3106         struct btrfs_path *path;
3107         int ret;
3108         u32 nritems;
3109         struct extent_buffer *leaf;
3110         int slot;
3111         int advance;
3112         unsigned char d_type;
3113         int over = 0;
3114         u32 di_cur;
3115         u32 di_total;
3116         u32 di_len;
3117         int key_type = BTRFS_DIR_INDEX_KEY;
3118         char tmp_name[32];
3119         char *name_ptr;
3120         int name_len;
3121
3122         /* FIXME, use a real flag for deciding about the key type */
3123         if (root->fs_info->tree_root == root)
3124                 key_type = BTRFS_DIR_ITEM_KEY;
3125
3126         /* special case for "." */
3127         if (filp->f_pos == 0) {
3128                 over = filldir(dirent, ".", 1,
3129                                1, inode->i_ino,
3130                                DT_DIR);
3131                 if (over)
3132                         return 0;
3133                 filp->f_pos = 1;
3134         }
3135         /* special case for .., just use the back ref */
3136         if (filp->f_pos == 1) {
3137                 u64 pino = parent_ino(filp->f_path.dentry);
3138                 over = filldir(dirent, "..", 2,
3139                                2, pino, DT_DIR);
3140                 if (over)
3141                         return 0;
3142                 filp->f_pos = 2;
3143         }
3144         path = btrfs_alloc_path();
3145         path->reada = 2;
3146
3147         btrfs_set_key_type(&key, key_type);
3148         key.offset = filp->f_pos;
3149         key.objectid = inode->i_ino;
3150
3151         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3152         if (ret < 0)
3153                 goto err;
3154         advance = 0;
3155
3156         while (1) {
3157                 leaf = path->nodes[0];
3158                 nritems = btrfs_header_nritems(leaf);
3159                 slot = path->slots[0];
3160                 if (advance || slot >= nritems) {
3161                         if (slot >= nritems - 1) {
3162                                 ret = btrfs_next_leaf(root, path);
3163                                 if (ret)
3164                                         break;
3165                                 leaf = path->nodes[0];
3166                                 nritems = btrfs_header_nritems(leaf);
3167                                 slot = path->slots[0];
3168                         } else {
3169                                 slot++;
3170                                 path->slots[0]++;
3171                         }
3172                 }
3173
3174                 advance = 1;
3175                 item = btrfs_item_nr(leaf, slot);
3176                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3177
3178                 if (found_key.objectid != key.objectid)
3179                         break;
3180                 if (btrfs_key_type(&found_key) != key_type)
3181                         break;
3182                 if (found_key.offset < filp->f_pos)
3183                         continue;
3184
3185                 filp->f_pos = found_key.offset;
3186
3187                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
3188                 di_cur = 0;
3189                 di_total = btrfs_item_size(leaf, item);
3190
3191                 while (di_cur < di_total) {
3192                         struct btrfs_key location;
3193
3194                         name_len = btrfs_dir_name_len(leaf, di);
3195                         if (name_len <= sizeof(tmp_name)) {
3196                                 name_ptr = tmp_name;
3197                         } else {
3198                                 name_ptr = kmalloc(name_len, GFP_NOFS);
3199                                 if (!name_ptr) {
3200                                         ret = -ENOMEM;
3201                                         goto err;
3202                                 }
3203                         }
3204                         read_extent_buffer(leaf, name_ptr,
3205                                            (unsigned long)(di + 1), name_len);
3206
3207                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
3208                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
3209
3210                         /* is this a reference to our own snapshot? If so
3211                          * skip it
3212                          */
3213                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
3214                             location.objectid == root->root_key.objectid) {
3215                                 over = 0;
3216                                 goto skip;
3217                         }
3218                         over = filldir(dirent, name_ptr, name_len,
3219                                        found_key.offset, location.objectid,
3220                                        d_type);
3221
3222 skip:
3223                         if (name_ptr != tmp_name)
3224                                 kfree(name_ptr);
3225
3226                         if (over)
3227                                 goto nopos;
3228                         di_len = btrfs_dir_name_len(leaf, di) +
3229                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
3230                         di_cur += di_len;
3231                         di = (struct btrfs_dir_item *)((char *)di + di_len);
3232                 }
3233         }
3234
3235         /* Reached end of directory/root. Bump pos past the last item. */
3236         if (key_type == BTRFS_DIR_INDEX_KEY)
3237                 filp->f_pos = INT_LIMIT(typeof(filp->f_pos));
3238         else
3239                 filp->f_pos++;
3240 nopos:
3241         ret = 0;
3242 err:
3243         btrfs_free_path(path);
3244         return ret;
3245 }
3246
3247 int btrfs_write_inode(struct inode *inode, int wait)
3248 {
3249         struct btrfs_root *root = BTRFS_I(inode)->root;
3250         struct btrfs_trans_handle *trans;
3251         int ret = 0;
3252
3253         if (root->fs_info->btree_inode == inode)
3254                 return 0;
3255
3256         if (wait) {
3257                 trans = btrfs_join_transaction(root, 1);
3258                 btrfs_set_trans_block_group(trans, inode);
3259                 ret = btrfs_commit_transaction(trans, root);
3260         }
3261         return ret;
3262 }
3263
3264 /*
3265  * This is somewhat expensive, updating the tree every time the
3266  * inode changes.  But, it is most likely to find the inode in cache.
3267  * FIXME, needs more benchmarking...there are no reasons other than performance
3268  * to keep or drop this code.
3269  */
3270 void btrfs_dirty_inode(struct inode *inode)
3271 {
3272         struct btrfs_root *root = BTRFS_I(inode)->root;
3273         struct btrfs_trans_handle *trans;
3274
3275         trans = btrfs_join_transaction(root, 1);
3276         btrfs_set_trans_block_group(trans, inode);
3277         btrfs_update_inode(trans, root, inode);
3278         btrfs_end_transaction(trans, root);
3279 }
3280
3281 /*
3282  * find the highest existing sequence number in a directory
3283  * and then set the in-memory index_cnt variable to reflect
3284  * free sequence numbers
3285  */
3286 static int btrfs_set_inode_index_count(struct inode *inode)
3287 {
3288         struct btrfs_root *root = BTRFS_I(inode)->root;
3289         struct btrfs_key key, found_key;
3290         struct btrfs_path *path;
3291         struct extent_buffer *leaf;
3292         int ret;
3293
3294         key.objectid = inode->i_ino;
3295         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
3296         key.offset = (u64)-1;
3297
3298         path = btrfs_alloc_path();
3299         if (!path)
3300                 return -ENOMEM;
3301
3302         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3303         if (ret < 0)
3304                 goto out;
3305         /* FIXME: we should be able to handle this */
3306         if (ret == 0)
3307                 goto out;
3308         ret = 0;
3309
3310         /*
3311          * MAGIC NUMBER EXPLANATION:
3312          * since we search a directory based on f_pos we have to start at 2
3313          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
3314          * else has to start at 2
3315          */
3316         if (path->slots[0] == 0) {
3317                 BTRFS_I(inode)->index_cnt = 2;
3318                 goto out;
3319         }
3320
3321         path->slots[0]--;
3322
3323         leaf = path->nodes[0];
3324         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3325
3326         if (found_key.objectid != inode->i_ino ||
3327             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
3328                 BTRFS_I(inode)->index_cnt = 2;
3329                 goto out;
3330         }
3331
3332         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
3333 out:
3334         btrfs_free_path(path);
3335         return ret;
3336 }
3337
3338 /*
3339  * helper to find a free sequence number in a given directory.  This current
3340  * code is very simple, later versions will do smarter things in the btree
3341  */
3342 int btrfs_set_inode_index(struct inode *dir, u64 *index)
3343 {
3344         int ret = 0;
3345
3346         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
3347                 ret = btrfs_set_inode_index_count(dir);
3348                 if (ret) {
3349                         return ret;
3350                 }
3351         }
3352
3353         *index = BTRFS_I(dir)->index_cnt;
3354         BTRFS_I(dir)->index_cnt++;
3355
3356         return ret;
3357 }
3358
3359 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
3360                                      struct btrfs_root *root,
3361                                      struct inode *dir,
3362                                      const char *name, int name_len,
3363                                      u64 ref_objectid,
3364                                      u64 objectid,
3365                                      struct btrfs_block_group_cache *group,
3366                                      int mode, u64 *index)
3367 {
3368         struct inode *inode;
3369         struct btrfs_inode_item *inode_item;
3370         struct btrfs_block_group_cache *new_inode_group;
3371         struct btrfs_key *location;
3372         struct btrfs_path *path;
3373         struct btrfs_inode_ref *ref;
3374         struct btrfs_key key[2];
3375         u32 sizes[2];
3376         unsigned long ptr;
3377         int ret;
3378         int owner;
3379
3380         path = btrfs_alloc_path();
3381         BUG_ON(!path);
3382
3383         inode = new_inode(root->fs_info->sb);
3384         if (!inode)
3385                 return ERR_PTR(-ENOMEM);
3386
3387         if (dir) {
3388                 ret = btrfs_set_inode_index(dir, index);
3389                 if (ret)
3390                         return ERR_PTR(ret);
3391         }
3392         /*
3393          * index_cnt is ignored for everything but a dir,
3394          * btrfs_get_inode_index_count has an explanation for the magic
3395          * number
3396          */
3397         init_btrfs_i(inode);
3398         BTRFS_I(inode)->index_cnt = 2;
3399         BTRFS_I(inode)->root = root;
3400         BTRFS_I(inode)->generation = trans->transid;
3401
3402         if (mode & S_IFDIR)
3403                 owner = 0;
3404         else
3405                 owner = 1;
3406         new_inode_group = btrfs_find_block_group(root, group, 0,
3407                                        BTRFS_BLOCK_GROUP_METADATA, owner);
3408         if (!new_inode_group) {
3409                 printk("find_block group failed\n");
3410                 new_inode_group = group;
3411         }
3412         BTRFS_I(inode)->block_group = new_inode_group;
3413
3414         key[0].objectid = objectid;
3415         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
3416         key[0].offset = 0;
3417
3418         key[1].objectid = objectid;
3419         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
3420         key[1].offset = ref_objectid;
3421
3422         sizes[0] = sizeof(struct btrfs_inode_item);
3423         sizes[1] = name_len + sizeof(*ref);
3424
3425         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
3426         if (ret != 0)
3427                 goto fail;
3428
3429         if (objectid > root->highest_inode)
3430                 root->highest_inode = objectid;
3431
3432         inode->i_uid = current->fsuid;
3433         inode->i_gid = current->fsgid;
3434         inode->i_mode = mode;
3435         inode->i_ino = objectid;
3436         inode_set_bytes(inode, 0);
3437         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3438         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3439                                   struct btrfs_inode_item);
3440         fill_inode_item(trans, path->nodes[0], inode_item, inode);
3441
3442         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3443                              struct btrfs_inode_ref);
3444         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
3445         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
3446         ptr = (unsigned long)(ref + 1);
3447         write_extent_buffer(path->nodes[0], name, ptr, name_len);
3448
3449         btrfs_mark_buffer_dirty(path->nodes[0]);
3450         btrfs_free_path(path);
3451
3452         location = &BTRFS_I(inode)->location;
3453         location->objectid = objectid;
3454         location->offset = 0;
3455         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
3456
3457         insert_inode_hash(inode);
3458         return inode;
3459 fail:
3460         if (dir)
3461                 BTRFS_I(dir)->index_cnt--;
3462         btrfs_free_path(path);
3463         return ERR_PTR(ret);
3464 }
3465
3466 static inline u8 btrfs_inode_type(struct inode *inode)
3467 {
3468         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
3469 }
3470
3471 /*
3472  * utility function to add 'inode' into 'parent_inode' with
3473  * a give name and a given sequence number.
3474  * if 'add_backref' is true, also insert a backref from the
3475  * inode to the parent directory.
3476  */
3477 int btrfs_add_link(struct btrfs_trans_handle *trans,
3478                    struct inode *parent_inode, struct inode *inode,
3479                    const char *name, int name_len, int add_backref, u64 index)
3480 {
3481         int ret;
3482         struct btrfs_key key;
3483         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
3484
3485         key.objectid = inode->i_ino;
3486         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
3487         key.offset = 0;
3488
3489         ret = btrfs_insert_dir_item(trans, root, name, name_len,
3490                                     parent_inode->i_ino,
3491                                     &key, btrfs_inode_type(inode),
3492                                     index);
3493         if (ret == 0) {
3494                 if (add_backref) {
3495                         ret = btrfs_insert_inode_ref(trans, root,
3496                                                      name, name_len,
3497                                                      inode->i_ino,
3498                                                      parent_inode->i_ino,
3499                                                      index);
3500                 }
3501                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
3502                                    name_len * 2);
3503                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
3504                 ret = btrfs_update_inode(trans, root, parent_inode);
3505         }
3506         return ret;
3507 }
3508
3509 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
3510                             struct dentry *dentry, struct inode *inode,
3511                             int backref, u64 index)
3512 {
3513         int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3514                                  inode, dentry->d_name.name,
3515                                  dentry->d_name.len, backref, index);
3516         if (!err) {
3517                 d_instantiate(dentry, inode);
3518                 return 0;
3519         }
3520         if (err > 0)
3521                 err = -EEXIST;
3522         return err;
3523 }
3524
3525 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
3526                         int mode, dev_t rdev)
3527 {
3528         struct btrfs_trans_handle *trans;
3529         struct btrfs_root *root = BTRFS_I(dir)->root;
3530         struct inode *inode = NULL;
3531         int err;
3532         int drop_inode = 0;
3533         u64 objectid;
3534         unsigned long nr = 0;
3535         u64 index = 0;
3536
3537         if (!new_valid_dev(rdev))
3538                 return -EINVAL;
3539
3540         err = btrfs_check_free_space(root, 1, 0);
3541         if (err)
3542                 goto fail;
3543
3544         trans = btrfs_start_transaction(root, 1);
3545         btrfs_set_trans_block_group(trans, dir);
3546
3547         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3548         if (err) {
3549                 err = -ENOSPC;
3550                 goto out_unlock;
3551         }
3552
3553         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3554                                 dentry->d_name.len,
3555                                 dentry->d_parent->d_inode->i_ino, objectid,
3556                                 BTRFS_I(dir)->block_group, mode, &index);
3557         err = PTR_ERR(inode);
3558         if (IS_ERR(inode))
3559                 goto out_unlock;
3560
3561         err = btrfs_init_acl(inode, dir);
3562         if (err) {
3563                 drop_inode = 1;
3564                 goto out_unlock;
3565         }
3566
3567         btrfs_set_trans_block_group(trans, inode);
3568         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3569         if (err)
3570                 drop_inode = 1;
3571         else {
3572                 inode->i_op = &btrfs_special_inode_operations;
3573                 init_special_inode(inode, inode->i_mode, rdev);
3574                 btrfs_update_inode(trans, root, inode);
3575         }
3576         dir->i_sb->s_dirt = 1;
3577         btrfs_update_inode_block_group(trans, inode);
3578         btrfs_update_inode_block_group(trans, dir);
3579 out_unlock:
3580         nr = trans->blocks_used;
3581         btrfs_end_transaction_throttle(trans, root);
3582 fail:
3583         if (drop_inode) {
3584                 inode_dec_link_count(inode);
3585                 iput(inode);
3586         }
3587         btrfs_btree_balance_dirty(root, nr);
3588         return err;
3589 }
3590
3591 static int btrfs_create(struct inode *dir, struct dentry *dentry,
3592                         int mode, struct nameidata *nd)
3593 {
3594         struct btrfs_trans_handle *trans;
3595         struct btrfs_root *root = BTRFS_I(dir)->root;
3596         struct inode *inode = NULL;
3597         int err;
3598         int drop_inode = 0;
3599         unsigned long nr = 0;
3600         u64 objectid;
3601         u64 index = 0;
3602
3603         err = btrfs_check_free_space(root, 1, 0);
3604         if (err)
3605                 goto fail;
3606         trans = btrfs_start_transaction(root, 1);
3607         btrfs_set_trans_block_group(trans, dir);
3608
3609         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3610         if (err) {
3611                 err = -ENOSPC;
3612                 goto out_unlock;
3613         }
3614
3615         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3616                                 dentry->d_name.len,
3617                                 dentry->d_parent->d_inode->i_ino,
3618                                 objectid, BTRFS_I(dir)->block_group, mode,
3619                                 &index);
3620         err = PTR_ERR(inode);
3621         if (IS_ERR(inode))
3622                 goto out_unlock;
3623
3624         err = btrfs_init_acl(inode, dir);
3625         if (err) {
3626                 drop_inode = 1;
3627                 goto out_unlock;
3628         }
3629
3630         btrfs_set_trans_block_group(trans, inode);
3631         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3632         if (err)
3633                 drop_inode = 1;
3634         else {
3635                 inode->i_mapping->a_ops = &btrfs_aops;
3636                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3637                 inode->i_fop = &btrfs_file_operations;
3638                 inode->i_op = &btrfs_file_inode_operations;
3639                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3640         }
3641         dir->i_sb->s_dirt = 1;
3642         btrfs_update_inode_block_group(trans, inode);
3643         btrfs_update_inode_block_group(trans, dir);
3644 out_unlock:
3645         nr = trans->blocks_used;
3646         btrfs_end_transaction_throttle(trans, root);
3647 fail:
3648         if (drop_inode) {
3649                 inode_dec_link_count(inode);
3650                 iput(inode);
3651         }
3652         btrfs_btree_balance_dirty(root, nr);
3653         return err;
3654 }
3655
3656 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
3657                       struct dentry *dentry)
3658 {
3659         struct btrfs_trans_handle *trans;
3660         struct btrfs_root *root = BTRFS_I(dir)->root;
3661         struct inode *inode = old_dentry->d_inode;
3662         u64 index;
3663         unsigned long nr = 0;
3664         int err;
3665         int drop_inode = 0;
3666
3667         if (inode->i_nlink == 0)
3668                 return -ENOENT;
3669
3670         btrfs_inc_nlink(inode);
3671         err = btrfs_check_free_space(root, 1, 0);
3672         if (err)
3673                 goto fail;
3674         err = btrfs_set_inode_index(dir, &index);
3675         if (err)
3676                 goto fail;
3677
3678         trans = btrfs_start_transaction(root, 1);
3679
3680         btrfs_set_trans_block_group(trans, dir);
3681         atomic_inc(&inode->i_count);
3682
3683         err = btrfs_add_nondir(trans, dentry, inode, 1, index);
3684
3685         if (err)
3686                 drop_inode = 1;
3687
3688         dir->i_sb->s_dirt = 1;
3689         btrfs_update_inode_block_group(trans, dir);
3690         err = btrfs_update_inode(trans, root, inode);
3691
3692         if (err)
3693                 drop_inode = 1;
3694
3695         nr = trans->blocks_used;
3696         btrfs_end_transaction_throttle(trans, root);
3697 fail:
3698         if (drop_inode) {
3699                 inode_dec_link_count(inode);
3700                 iput(inode);
3701         }
3702         btrfs_btree_balance_dirty(root, nr);
3703         return err;
3704 }
3705
3706 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
3707 {
3708         struct inode *inode = NULL;
3709         struct btrfs_trans_handle *trans;
3710         struct btrfs_root *root = BTRFS_I(dir)->root;
3711         int err = 0;
3712         int drop_on_err = 0;
3713         u64 objectid = 0;
3714         u64 index = 0;
3715         unsigned long nr = 1;
3716
3717         err = btrfs_check_free_space(root, 1, 0);
3718         if (err)
3719                 goto out_unlock;
3720
3721         trans = btrfs_start_transaction(root, 1);
3722         btrfs_set_trans_block_group(trans, dir);
3723
3724         if (IS_ERR(trans)) {
3725                 err = PTR_ERR(trans);
3726                 goto out_unlock;
3727         }
3728
3729         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3730         if (err) {
3731                 err = -ENOSPC;
3732                 goto out_unlock;
3733         }
3734
3735         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3736                                 dentry->d_name.len,
3737                                 dentry->d_parent->d_inode->i_ino, objectid,
3738                                 BTRFS_I(dir)->block_group, S_IFDIR | mode,
3739                                 &index);
3740         if (IS_ERR(inode)) {
3741                 err = PTR_ERR(inode);
3742                 goto out_fail;
3743         }
3744
3745         drop_on_err = 1;
3746
3747         err = btrfs_init_acl(inode, dir);
3748         if (err)
3749                 goto out_fail;
3750
3751         inode->i_op = &btrfs_dir_inode_operations;
3752         inode->i_fop = &btrfs_dir_file_operations;
3753         btrfs_set_trans_block_group(trans, inode);
3754
3755         btrfs_i_size_write(inode, 0);
3756         err = btrfs_update_inode(trans, root, inode);
3757         if (err)
3758                 goto out_fail;
3759
3760         err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3761                                  inode, dentry->d_name.name,
3762                                  dentry->d_name.len, 0, index);
3763         if (err)
3764                 goto out_fail;
3765
3766         d_instantiate(dentry, inode);
3767         drop_on_err = 0;
3768         dir->i_sb->s_dirt = 1;
3769         btrfs_update_inode_block_group(trans, inode);
3770         btrfs_update_inode_block_group(trans, dir);
3771
3772 out_fail:
3773         nr = trans->blocks_used;
3774         btrfs_end_transaction_throttle(trans, root);
3775
3776 out_unlock:
3777         if (drop_on_err)
3778                 iput(inode);
3779         btrfs_btree_balance_dirty(root, nr);
3780         return err;
3781 }
3782
3783 /* helper for btfs_get_extent.  Given an existing extent in the tree,
3784  * and an extent that you want to insert, deal with overlap and insert
3785  * the new extent into the tree.
3786  */
3787 static int merge_extent_mapping(struct extent_map_tree *em_tree,
3788                                 struct extent_map *existing,
3789                                 struct extent_map *em,
3790                                 u64 map_start, u64 map_len)
3791 {
3792         u64 start_diff;
3793
3794         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
3795         start_diff = map_start - em->start;
3796         em->start = map_start;
3797         em->len = map_len;
3798         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
3799             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
3800                 em->block_start += start_diff;
3801                 em->block_len -= start_diff;
3802         }
3803         return add_extent_mapping(em_tree, em);
3804 }
3805
3806 static noinline int uncompress_inline(struct btrfs_path *path,
3807                                       struct inode *inode, struct page *page,
3808                                       size_t pg_offset, u64 extent_offset,
3809                                       struct btrfs_file_extent_item *item)
3810 {
3811         int ret;
3812         struct extent_buffer *leaf = path->nodes[0];
3813         char *tmp;
3814         size_t max_size;
3815         unsigned long inline_size;
3816         unsigned long ptr;
3817
3818         WARN_ON(pg_offset != 0);
3819         max_size = btrfs_file_extent_ram_bytes(leaf, item);
3820         inline_size = btrfs_file_extent_inline_item_len(leaf,
3821                                         btrfs_item_nr(leaf, path->slots[0]));
3822         tmp = kmalloc(inline_size, GFP_NOFS);
3823         ptr = btrfs_file_extent_inline_start(item);
3824
3825         read_extent_buffer(leaf, tmp, ptr, inline_size);
3826
3827         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
3828         ret = btrfs_zlib_decompress(tmp, page, extent_offset,
3829                                     inline_size, max_size);
3830         if (ret) {
3831                 char *kaddr = kmap_atomic(page, KM_USER0);
3832                 unsigned long copy_size = min_t(u64,
3833                                   PAGE_CACHE_SIZE - pg_offset,
3834                                   max_size - extent_offset);
3835                 memset(kaddr + pg_offset, 0, copy_size);
3836                 kunmap_atomic(kaddr, KM_USER0);
3837         }
3838         kfree(tmp);
3839         return 0;
3840 }
3841
3842 /*
3843  * a bit scary, this does extent mapping from logical file offset to the disk.
3844  * the ugly parts come from merging extents from the disk with the
3845  * in-ram representation.  This gets more complex because of the data=ordered code,
3846  * where the in-ram extents might be locked pending data=ordered completion.
3847  *
3848  * This also copies inline extents directly into the page.
3849  */
3850 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
3851                                     size_t pg_offset, u64 start, u64 len,
3852                                     int create)
3853 {
3854         int ret;
3855         int err = 0;
3856         u64 bytenr;
3857         u64 extent_start = 0;
3858         u64 extent_end = 0;
3859         u64 objectid = inode->i_ino;
3860         u32 found_type;
3861         struct btrfs_path *path = NULL;
3862         struct btrfs_root *root = BTRFS_I(inode)->root;
3863         struct btrfs_file_extent_item *item;
3864         struct extent_buffer *leaf;
3865         struct btrfs_key found_key;
3866         struct extent_map *em = NULL;
3867         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3868         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3869         struct btrfs_trans_handle *trans = NULL;
3870         int compressed;
3871
3872 again:
3873         spin_lock(&em_tree->lock);
3874         em = lookup_extent_mapping(em_tree, start, len);
3875         if (em)
3876                 em->bdev = root->fs_info->fs_devices->latest_bdev;
3877         spin_unlock(&em_tree->lock);
3878
3879         if (em) {
3880                 if (em->start > start || em->start + em->len <= start)
3881                         free_extent_map(em);
3882                 else if (em->block_start == EXTENT_MAP_INLINE && page)
3883                         free_extent_map(em);
3884                 else
3885                         goto out;
3886         }
3887         em = alloc_extent_map(GFP_NOFS);
3888         if (!em) {
3889                 err = -ENOMEM;
3890                 goto out;
3891         }
3892         em->bdev = root->fs_info->fs_devices->latest_bdev;
3893         em->start = EXTENT_MAP_HOLE;
3894         em->orig_start = EXTENT_MAP_HOLE;
3895         em->len = (u64)-1;
3896         em->block_len = (u64)-1;
3897
3898         if (!path) {
3899                 path = btrfs_alloc_path();
3900                 BUG_ON(!path);
3901         }
3902
3903         ret = btrfs_lookup_file_extent(trans, root, path,
3904                                        objectid, start, trans != NULL);
3905         if (ret < 0) {
3906                 err = ret;
3907                 goto out;
3908         }
3909
3910         if (ret != 0) {
3911                 if (path->slots[0] == 0)
3912                         goto not_found;
3913                 path->slots[0]--;
3914         }
3915
3916         leaf = path->nodes[0];
3917         item = btrfs_item_ptr(leaf, path->slots[0],
3918                               struct btrfs_file_extent_item);
3919         /* are we inside the extent that was found? */
3920         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3921         found_type = btrfs_key_type(&found_key);
3922         if (found_key.objectid != objectid ||
3923             found_type != BTRFS_EXTENT_DATA_KEY) {
3924                 goto not_found;
3925         }
3926
3927         found_type = btrfs_file_extent_type(leaf, item);
3928         extent_start = found_key.offset;
3929         compressed = btrfs_file_extent_compression(leaf, item);
3930         if (found_type == BTRFS_FILE_EXTENT_REG ||
3931             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
3932                 extent_end = extent_start +
3933                        btrfs_file_extent_num_bytes(leaf, item);
3934         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
3935                 size_t size;
3936                 size = btrfs_file_extent_inline_len(leaf, item);
3937                 extent_end = (extent_start + size + root->sectorsize - 1) &
3938                         ~((u64)root->sectorsize - 1);
3939         }
3940
3941         if (start >= extent_end) {
3942                 path->slots[0]++;
3943                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3944                         ret = btrfs_next_leaf(root, path);
3945                         if (ret < 0) {
3946                                 err = ret;
3947                                 goto out;
3948                         }
3949                         if (ret > 0)
3950                                 goto not_found;
3951                         leaf = path->nodes[0];
3952                 }
3953                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3954                 if (found_key.objectid != objectid ||
3955                     found_key.type != BTRFS_EXTENT_DATA_KEY)
3956                         goto not_found;
3957                 if (start + len <= found_key.offset)
3958                         goto not_found;
3959                 em->start = start;
3960                 em->len = found_key.offset - start;
3961                 goto not_found_em;
3962         }
3963
3964         if (found_type == BTRFS_FILE_EXTENT_REG ||
3965             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
3966                 em->start = extent_start;
3967                 em->len = extent_end - extent_start;
3968                 em->orig_start = extent_start -
3969                                  btrfs_file_extent_offset(leaf, item);
3970                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
3971                 if (bytenr == 0) {
3972                         em->block_start = EXTENT_MAP_HOLE;
3973                         goto insert;
3974                 }
3975                 if (compressed) {
3976                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3977                         em->block_start = bytenr;
3978                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
3979                                                                          item);
3980                 } else {
3981                         bytenr += btrfs_file_extent_offset(leaf, item);
3982                         em->block_start = bytenr;
3983                         em->block_len = em->len;
3984                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
3985                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
3986                 }
3987                 goto insert;
3988         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
3989                 unsigned long ptr;
3990                 char *map;
3991                 size_t size;
3992                 size_t extent_offset;
3993                 size_t copy_size;
3994
3995                 em->block_start = EXTENT_MAP_INLINE;
3996                 if (!page || create) {
3997                         em->start = extent_start;
3998                         em->len = extent_end - extent_start;
3999                         goto out;
4000                 }
4001
4002                 size = btrfs_file_extent_inline_len(leaf, item);
4003                 extent_offset = page_offset(page) + pg_offset - extent_start;
4004                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
4005                                 size - extent_offset);
4006                 em->start = extent_start + extent_offset;
4007                 em->len = (copy_size + root->sectorsize - 1) &
4008                         ~((u64)root->sectorsize - 1);
4009                 em->orig_start = EXTENT_MAP_INLINE;
4010                 if (compressed)
4011                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4012                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
4013                 if (create == 0 && !PageUptodate(page)) {
4014                         if (btrfs_file_extent_compression(leaf, item) ==
4015                             BTRFS_COMPRESS_ZLIB) {
4016                                 ret = uncompress_inline(path, inode, page,
4017                                                         pg_offset,
4018                                                         extent_offset, item);
4019                                 BUG_ON(ret);
4020                         } else {
4021                                 map = kmap(page);
4022                                 read_extent_buffer(leaf, map + pg_offset, ptr,
4023                                                    copy_size);
4024                                 kunmap(page);
4025                         }
4026                         flush_dcache_page(page);
4027                 } else if (create && PageUptodate(page)) {
4028                         if (!trans) {
4029                                 kunmap(page);
4030                                 free_extent_map(em);
4031                                 em = NULL;
4032                                 btrfs_release_path(root, path);
4033                                 trans = btrfs_join_transaction(root, 1);
4034                                 goto again;
4035                         }
4036                         map = kmap(page);
4037                         write_extent_buffer(leaf, map + pg_offset, ptr,
4038                                             copy_size);
4039                         kunmap(page);
4040                         btrfs_mark_buffer_dirty(leaf);
4041                 }
4042                 set_extent_uptodate(io_tree, em->start,
4043                                     extent_map_end(em) - 1, GFP_NOFS);
4044                 goto insert;
4045         } else {
4046                 printk("unkknown found_type %d\n", found_type);
4047                 WARN_ON(1);
4048         }
4049 not_found:
4050         em->start = start;
4051         em->len = len;
4052 not_found_em:
4053         em->block_start = EXTENT_MAP_HOLE;
4054         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
4055 insert:
4056         btrfs_release_path(root, path);
4057         if (em->start > start || extent_map_end(em) <= start) {
4058                 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->len, start, len);
4059                 err = -EIO;
4060                 goto out;
4061         }
4062
4063         err = 0;
4064         spin_lock(&em_tree->lock);
4065         ret = add_extent_mapping(em_tree, em);
4066         /* it is possible that someone inserted the extent into the tree
4067          * while we had the lock dropped.  It is also possible that
4068          * an overlapping map exists in the tree
4069          */
4070         if (ret == -EEXIST) {
4071                 struct extent_map *existing;
4072
4073                 ret = 0;
4074
4075                 existing = lookup_extent_mapping(em_tree, start, len);
4076                 if (existing && (existing->start > start ||
4077                     existing->start + existing->len <= start)) {
4078                         free_extent_map(existing);
4079                         existing = NULL;
4080                 }
4081                 if (!existing) {
4082                         existing = lookup_extent_mapping(em_tree, em->start,
4083                                                          em->len);
4084                         if (existing) {
4085                                 err = merge_extent_mapping(em_tree, existing,
4086                                                            em, start,
4087                                                            root->sectorsize);
4088                                 free_extent_map(existing);
4089                                 if (err) {
4090                                         free_extent_map(em);
4091                                         em = NULL;
4092                                 }
4093                         } else {
4094                                 err = -EIO;
4095                                 printk("failing to insert %Lu %Lu\n",
4096                                        start, len);
4097                                 free_extent_map(em);
4098                                 em = NULL;
4099                         }
4100                 } else {
4101                         free_extent_map(em);
4102                         em = existing;
4103                         err = 0;
4104                 }
4105         }
4106         spin_unlock(&em_tree->lock);
4107 out:
4108         if (path)
4109                 btrfs_free_path(path);
4110         if (trans) {
4111                 ret = btrfs_end_transaction(trans, root);
4112                 if (!err) {
4113                         err = ret;
4114                 }
4115         }
4116         if (err) {
4117                 free_extent_map(em);
4118                 WARN_ON(1);
4119                 return ERR_PTR(err);
4120         }
4121         return em;
4122 }
4123
4124 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
4125                         const struct iovec *iov, loff_t offset,
4126                         unsigned long nr_segs)
4127 {
4128         return -EINVAL;
4129 }
4130
4131 static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
4132 {
4133         return extent_bmap(mapping, iblock, btrfs_get_extent);
4134 }
4135
4136 int btrfs_readpage(struct file *file, struct page *page)
4137 {
4138         struct extent_io_tree *tree;
4139         tree = &BTRFS_I(page->mapping->host)->io_tree;
4140         return extent_read_full_page(tree, page, btrfs_get_extent);
4141 }
4142
4143 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
4144 {
4145         struct extent_io_tree *tree;
4146
4147
4148         if (current->flags & PF_MEMALLOC) {
4149                 redirty_page_for_writepage(wbc, page);
4150                 unlock_page(page);
4151                 return 0;
4152         }
4153         tree = &BTRFS_I(page->mapping->host)->io_tree;
4154         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
4155 }
4156
4157 int btrfs_writepages(struct address_space *mapping,
4158                      struct writeback_control *wbc)
4159 {
4160         struct extent_io_tree *tree;
4161
4162         tree = &BTRFS_I(mapping->host)->io_tree;
4163         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
4164 }
4165
4166 static int
4167 btrfs_readpages(struct file *file, struct address_space *mapping,
4168                 struct list_head *pages, unsigned nr_pages)
4169 {
4170         struct extent_io_tree *tree;
4171         tree = &BTRFS_I(mapping->host)->io_tree;
4172         return extent_readpages(tree, mapping, pages, nr_pages,
4173                                 btrfs_get_extent);
4174 }
4175 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4176 {
4177         struct extent_io_tree *tree;
4178         struct extent_map_tree *map;
4179         int ret;
4180
4181         tree = &BTRFS_I(page->mapping->host)->io_tree;
4182         map = &BTRFS_I(page->mapping->host)->extent_tree;
4183         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
4184         if (ret == 1) {
4185                 ClearPagePrivate(page);
4186                 set_page_private(page, 0);
4187                 page_cache_release(page);
4188         }
4189         return ret;
4190 }
4191
4192 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4193 {
4194         if (PageWriteback(page) || PageDirty(page))
4195                 return 0;
4196         return __btrfs_releasepage(page, gfp_flags);
4197 }
4198
4199 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
4200 {
4201         struct extent_io_tree *tree;
4202         struct btrfs_ordered_extent *ordered;
4203         u64 page_start = page_offset(page);
4204         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
4205
4206         wait_on_page_writeback(page);
4207         tree = &BTRFS_I(page->mapping->host)->io_tree;
4208         if (offset) {
4209                 btrfs_releasepage(page, GFP_NOFS);
4210                 return;
4211         }
4212
4213         lock_extent(tree, page_start, page_end, GFP_NOFS);
4214         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
4215                                            page_offset(page));
4216         if (ordered) {
4217                 /*
4218                  * IO on this page will never be started, so we need
4219                  * to account for any ordered extents now
4220                  */
4221                 clear_extent_bit(tree, page_start, page_end,
4222                                  EXTENT_DIRTY | EXTENT_DELALLOC |
4223                                  EXTENT_LOCKED, 1, 0, GFP_NOFS);
4224                 btrfs_finish_ordered_io(page->mapping->host,
4225                                         page_start, page_end);
4226                 btrfs_put_ordered_extent(ordered);
4227                 lock_extent(tree, page_start, page_end, GFP_NOFS);
4228         }
4229         clear_extent_bit(tree, page_start, page_end,
4230                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4231                  EXTENT_ORDERED,
4232                  1, 1, GFP_NOFS);
4233         __btrfs_releasepage(page, GFP_NOFS);
4234
4235         ClearPageChecked(page);
4236         if (PagePrivate(page)) {
4237                 ClearPagePrivate(page);
4238                 set_page_private(page, 0);
4239                 page_cache_release(page);
4240         }
4241 }
4242
4243 /*
4244  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
4245  * called from a page fault handler when a page is first dirtied. Hence we must
4246  * be careful to check for EOF conditions here. We set the page up correctly
4247  * for a written page which means we get ENOSPC checking when writing into
4248  * holes and correct delalloc and unwritten extent mapping on filesystems that
4249  * support these features.
4250  *
4251  * We are not allowed to take the i_mutex here so we have to play games to
4252  * protect against truncate races as the page could now be beyond EOF.  Because
4253  * vmtruncate() writes the inode size before removing pages, once we have the
4254  * page lock we can determine safely if the page is beyond EOF. If it is not
4255  * beyond EOF, then the page is guaranteed safe against truncation until we
4256  * unlock the page.
4257  */
4258 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
4259 {
4260         struct inode *inode = fdentry(vma->vm_file)->d_inode;
4261         struct btrfs_root *root = BTRFS_I(inode)->root;
4262         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4263         struct btrfs_ordered_extent *ordered;
4264         char *kaddr;
4265         unsigned long zero_start;
4266         loff_t size;
4267         int ret;
4268         u64 page_start;
4269         u64 page_end;
4270
4271         ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
4272         if (ret)
4273                 goto out;
4274
4275         ret = -EINVAL;
4276 again:
4277         lock_page(page);
4278         size = i_size_read(inode);
4279         page_start = page_offset(page);
4280         page_end = page_start + PAGE_CACHE_SIZE - 1;
4281
4282         if ((page->mapping != inode->i_mapping) ||
4283             (page_start >= size)) {
4284                 /* page got truncated out from underneath us */
4285                 goto out_unlock;
4286         }
4287         wait_on_page_writeback(page);
4288
4289         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
4290         set_page_extent_mapped(page);
4291
4292         /*
4293          * we can't set the delalloc bits if there are pending ordered
4294          * extents.  Drop our locks and wait for them to finish
4295          */
4296         ordered = btrfs_lookup_ordered_extent(inode, page_start);
4297         if (ordered) {
4298                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4299                 unlock_page(page);
4300                 btrfs_start_ordered_extent(inode, ordered, 1);
4301                 btrfs_put_ordered_extent(ordered);
4302                 goto again;
4303         }
4304
4305         btrfs_set_extent_delalloc(inode, page_start, page_end);
4306         ret = 0;
4307
4308         /* page is wholly or partially inside EOF */
4309         if (page_start + PAGE_CACHE_SIZE > size)
4310                 zero_start = size & ~PAGE_CACHE_MASK;
4311         else
4312                 zero_start = PAGE_CACHE_SIZE;
4313
4314         if (zero_start != PAGE_CACHE_SIZE) {
4315                 kaddr = kmap(page);
4316                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
4317                 flush_dcache_page(page);
4318                 kunmap(page);
4319         }
4320         ClearPageChecked(page);
4321         set_page_dirty(page);
4322         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4323
4324 out_unlock:
4325         unlock_page(page);
4326 out:
4327         return ret;
4328 }
4329
4330 static void btrfs_truncate(struct inode *inode)
4331 {
4332         struct btrfs_root *root = BTRFS_I(inode)->root;
4333         int ret;
4334         struct btrfs_trans_handle *trans;
4335         unsigned long nr;
4336         u64 mask = root->sectorsize - 1;
4337
4338         if (!S_ISREG(inode->i_mode))
4339                 return;
4340         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4341                 return;
4342
4343         btrfs_truncate_page(inode->i_mapping, inode->i_size);
4344         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
4345
4346         trans = btrfs_start_transaction(root, 1);
4347         btrfs_set_trans_block_group(trans, inode);
4348         btrfs_i_size_write(inode, inode->i_size);
4349
4350         ret = btrfs_orphan_add(trans, inode);
4351         if (ret)
4352                 goto out;
4353         /* FIXME, add redo link to tree so we don't leak on crash */
4354         ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size,
4355                                       BTRFS_EXTENT_DATA_KEY);
4356         btrfs_update_inode(trans, root, inode);
4357
4358         ret = btrfs_orphan_del(trans, inode);
4359         BUG_ON(ret);
4360
4361 out:
4362         nr = trans->blocks_used;
4363         ret = btrfs_end_transaction_throttle(trans, root);
4364         BUG_ON(ret);
4365         btrfs_btree_balance_dirty(root, nr);
4366 }
4367
4368 /*
4369  * Invalidate a single dcache entry at the root of the filesystem.
4370  * Needed after creation of snapshot or subvolume.
4371  */
4372 void btrfs_invalidate_dcache_root(struct inode *dir, char *name,
4373                                   int namelen)
4374 {
4375         struct dentry *alias, *entry;
4376         struct qstr qstr;
4377
4378         alias = d_find_alias(dir);
4379         if (alias) {
4380                 qstr.name = name;
4381                 qstr.len = namelen;
4382                 /* change me if btrfs ever gets a d_hash operation */
4383                 qstr.hash = full_name_hash(qstr.name, qstr.len);
4384                 entry = d_lookup(alias, &qstr);
4385                 dput(alias);
4386                 if (entry) {
4387                         d_invalidate(entry);
4388                         dput(entry);
4389                 }
4390         }
4391 }
4392
4393 /*
4394  * create a new subvolume directory/inode (helper for the ioctl).
4395  */
4396 int btrfs_create_subvol_root(struct btrfs_root *new_root, struct dentry *dentry,
4397                 struct btrfs_trans_handle *trans, u64 new_dirid,
4398                 struct btrfs_block_group_cache *block_group)
4399 {
4400         struct inode *inode;
4401         int error;
4402         u64 index = 0;
4403
4404         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
4405                                 new_dirid, block_group, S_IFDIR | 0700, &index);
4406         if (IS_ERR(inode))
4407                 return PTR_ERR(inode);
4408         inode->i_op = &btrfs_dir_inode_operations;
4409         inode->i_fop = &btrfs_dir_file_operations;
4410
4411         inode->i_nlink = 1;
4412         btrfs_i_size_write(inode, 0);
4413
4414         error = btrfs_update_inode(trans, new_root, inode);
4415         if (error)
4416                 return error;
4417
4418         d_instantiate(dentry, inode);
4419         return 0;
4420 }
4421
4422 /* helper function for file defrag and space balancing.  This
4423  * forces readahead on a given range of bytes in an inode
4424  */
4425 unsigned long btrfs_force_ra(struct address_space *mapping,
4426                               struct file_ra_state *ra, struct file *file,
4427                               pgoff_t offset, pgoff_t last_index)
4428 {
4429         pgoff_t req_size = last_index - offset + 1;
4430
4431         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
4432         return offset + req_size;
4433 }
4434
4435 struct inode *btrfs_alloc_inode(struct super_block *sb)
4436 {
4437         struct btrfs_inode *ei;
4438
4439         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
4440         if (!ei)
4441                 return NULL;
4442         ei->last_trans = 0;
4443         ei->logged_trans = 0;
4444         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
4445         ei->i_acl = BTRFS_ACL_NOT_CACHED;
4446         ei->i_default_acl = BTRFS_ACL_NOT_CACHED;
4447         INIT_LIST_HEAD(&ei->i_orphan);
4448         return &ei->vfs_inode;
4449 }
4450
4451 void btrfs_destroy_inode(struct inode *inode)
4452 {
4453         struct btrfs_ordered_extent *ordered;
4454         WARN_ON(!list_empty(&inode->i_dentry));
4455         WARN_ON(inode->i_data.nrpages);
4456
4457         if (BTRFS_I(inode)->i_acl &&
4458             BTRFS_I(inode)->i_acl != BTRFS_ACL_NOT_CACHED)
4459                 posix_acl_release(BTRFS_I(inode)->i_acl);
4460         if (BTRFS_I(inode)->i_default_acl &&
4461             BTRFS_I(inode)->i_default_acl != BTRFS_ACL_NOT_CACHED)
4462                 posix_acl_release(BTRFS_I(inode)->i_default_acl);
4463
4464         spin_lock(&BTRFS_I(inode)->root->list_lock);
4465         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
4466                 printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
4467                        " list\n", inode->i_ino);
4468                 dump_stack();
4469         }
4470         spin_unlock(&BTRFS_I(inode)->root->list_lock);
4471
4472         while(1) {
4473                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
4474                 if (!ordered)
4475                         break;
4476                 else {
4477                         printk("found ordered extent %Lu %Lu\n",
4478                                ordered->file_offset, ordered->len);
4479                         btrfs_remove_ordered_extent(inode, ordered);
4480                         btrfs_put_ordered_extent(ordered);
4481                         btrfs_put_ordered_extent(ordered);
4482                 }
4483         }
4484         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
4485         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
4486 }
4487
4488 static void init_once(void *foo)
4489 {
4490         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
4491
4492         inode_init_once(&ei->vfs_inode);
4493 }
4494
4495 void btrfs_destroy_cachep(void)
4496 {
4497         if (btrfs_inode_cachep)
4498                 kmem_cache_destroy(btrfs_inode_cachep);
4499         if (btrfs_trans_handle_cachep)
4500                 kmem_cache_destroy(btrfs_trans_handle_cachep);
4501         if (btrfs_transaction_cachep)
4502                 kmem_cache_destroy(btrfs_transaction_cachep);
4503         if (btrfs_bit_radix_cachep)
4504                 kmem_cache_destroy(btrfs_bit_radix_cachep);
4505         if (btrfs_path_cachep)
4506                 kmem_cache_destroy(btrfs_path_cachep);
4507 }
4508
4509 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
4510                                        unsigned long extra_flags,
4511                                        void (*ctor)(void *))
4512 {
4513         return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
4514                                  SLAB_MEM_SPREAD | extra_flags), ctor);
4515 }
4516
4517 int btrfs_init_cachep(void)
4518 {
4519         btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
4520                                           sizeof(struct btrfs_inode),
4521                                           0, init_once);
4522         if (!btrfs_inode_cachep)
4523                 goto fail;
4524         btrfs_trans_handle_cachep =
4525                         btrfs_cache_create("btrfs_trans_handle_cache",
4526                                            sizeof(struct btrfs_trans_handle),
4527                                            0, NULL);
4528         if (!btrfs_trans_handle_cachep)
4529                 goto fail;
4530         btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
4531                                              sizeof(struct btrfs_transaction),
4532                                              0, NULL);
4533         if (!btrfs_transaction_cachep)
4534                 goto fail;
4535         btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
4536                                          sizeof(struct btrfs_path),
4537                                          0, NULL);
4538         if (!btrfs_path_cachep)
4539                 goto fail;
4540         btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
4541                                               SLAB_DESTROY_BY_RCU, NULL);
4542         if (!btrfs_bit_radix_cachep)
4543                 goto fail;
4544         return 0;
4545 fail:
4546         btrfs_destroy_cachep();
4547         return -ENOMEM;
4548 }
4549
4550 static int btrfs_getattr(struct vfsmount *mnt,
4551                          struct dentry *dentry, struct kstat *stat)
4552 {
4553         struct inode *inode = dentry->d_inode;
4554         generic_fillattr(inode, stat);
4555         stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
4556         stat->blksize = PAGE_CACHE_SIZE;
4557         stat->blocks = (inode_get_bytes(inode) +
4558                         BTRFS_I(inode)->delalloc_bytes) >> 9;
4559         return 0;
4560 }
4561
4562 static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
4563                            struct inode * new_dir,struct dentry *new_dentry)
4564 {
4565         struct btrfs_trans_handle *trans;
4566         struct btrfs_root *root = BTRFS_I(old_dir)->root;
4567         struct inode *new_inode = new_dentry->d_inode;
4568         struct inode *old_inode = old_dentry->d_inode;
4569         struct timespec ctime = CURRENT_TIME;
4570         u64 index = 0;
4571         int ret;
4572
4573         /* we're not allowed to rename between subvolumes */
4574         if (BTRFS_I(old_inode)->root->root_key.objectid !=
4575             BTRFS_I(new_dir)->root->root_key.objectid)
4576                 return -EXDEV;
4577
4578         if (S_ISDIR(old_inode->i_mode) && new_inode &&
4579             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
4580                 return -ENOTEMPTY;
4581         }
4582
4583         ret = btrfs_check_free_space(root, 1, 0);
4584         if (ret)
4585                 goto out_unlock;
4586
4587         trans = btrfs_start_transaction(root, 1);
4588
4589         btrfs_set_trans_block_group(trans, new_dir);
4590
4591         btrfs_inc_nlink(old_dentry->d_inode);
4592         old_dir->i_ctime = old_dir->i_mtime = ctime;
4593         new_dir->i_ctime = new_dir->i_mtime = ctime;
4594         old_inode->i_ctime = ctime;
4595
4596         ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode,
4597                                  old_dentry->d_name.name,
4598                                  old_dentry->d_name.len);
4599         if (ret)
4600                 goto out_fail;
4601
4602         if (new_inode) {
4603                 new_inode->i_ctime = CURRENT_TIME;
4604                 ret = btrfs_unlink_inode(trans, root, new_dir,
4605                                          new_dentry->d_inode,
4606                                          new_dentry->d_name.name,
4607                                          new_dentry->d_name.len);
4608                 if (ret)
4609                         goto out_fail;
4610                 if (new_inode->i_nlink == 0) {
4611                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
4612                         if (ret)
4613                                 goto out_fail;
4614                 }
4615
4616         }
4617         ret = btrfs_set_inode_index(new_dir, &index);
4618         if (ret)
4619                 goto out_fail;
4620
4621         ret = btrfs_add_link(trans, new_dentry->d_parent->d_inode,
4622                              old_inode, new_dentry->d_name.name,
4623                              new_dentry->d_name.len, 1, index);
4624         if (ret)
4625                 goto out_fail;
4626
4627 out_fail:
4628         btrfs_end_transaction_throttle(trans, root);
4629 out_unlock:
4630         return ret;
4631 }
4632
4633 /*
4634  * some fairly slow code that needs optimization. This walks the list
4635  * of all the inodes with pending delalloc and forces them to disk.
4636  */
4637 int btrfs_start_delalloc_inodes(struct btrfs_root *root)
4638 {
4639         struct list_head *head = &root->fs_info->delalloc_inodes;
4640         struct btrfs_inode *binode;
4641         struct inode *inode;
4642         unsigned long flags;
4643
4644         if (root->fs_info->sb->s_flags & MS_RDONLY)
4645                 return -EROFS;
4646
4647         spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
4648         while(!list_empty(head)) {
4649                 binode = list_entry(head->next, struct btrfs_inode,
4650                                     delalloc_inodes);
4651                 inode = igrab(&binode->vfs_inode);
4652                 if (!inode)
4653                         list_del_init(&binode->delalloc_inodes);
4654                 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
4655                 if (inode) {
4656                         filemap_flush(inode->i_mapping);
4657                         iput(inode);
4658                 }
4659                 cond_resched();
4660                 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
4661         }
4662         spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
4663
4664         /* the filemap_flush will queue IO into the worker threads, but
4665          * we have to make sure the IO is actually started and that
4666          * ordered extents get created before we return
4667          */
4668         atomic_inc(&root->fs_info->async_submit_draining);
4669         while(atomic_read(&root->fs_info->nr_async_submits) ||
4670               atomic_read(&root->fs_info->async_delalloc_pages)) {
4671                 wait_event(root->fs_info->async_submit_wait,
4672                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
4673                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
4674         }
4675         atomic_dec(&root->fs_info->async_submit_draining);
4676         return 0;
4677 }
4678
4679 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
4680                          const char *symname)
4681 {
4682         struct btrfs_trans_handle *trans;
4683         struct btrfs_root *root = BTRFS_I(dir)->root;
4684         struct btrfs_path *path;
4685         struct btrfs_key key;
4686         struct inode *inode = NULL;
4687         int err;
4688         int drop_inode = 0;
4689         u64 objectid;
4690         u64 index = 0 ;
4691         int name_len;
4692         int datasize;
4693         unsigned long ptr;
4694         struct btrfs_file_extent_item *ei;
4695         struct extent_buffer *leaf;
4696         unsigned long nr = 0;
4697
4698         name_len = strlen(symname) + 1;
4699         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
4700                 return -ENAMETOOLONG;
4701
4702         err = btrfs_check_free_space(root, 1, 0);
4703         if (err)
4704                 goto out_fail;
4705
4706         trans = btrfs_start_transaction(root, 1);
4707         btrfs_set_trans_block_group(trans, dir);
4708
4709         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4710         if (err) {
4711                 err = -ENOSPC;
4712                 goto out_unlock;
4713         }
4714
4715         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4716                                 dentry->d_name.len,
4717                                 dentry->d_parent->d_inode->i_ino, objectid,
4718                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
4719                                 &index);
4720         err = PTR_ERR(inode);
4721         if (IS_ERR(inode))
4722                 goto out_unlock;
4723
4724         err = btrfs_init_acl(inode, dir);
4725         if (err) {
4726                 drop_inode = 1;
4727                 goto out_unlock;
4728         }
4729
4730         btrfs_set_trans_block_group(trans, inode);
4731         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
4732         if (err)
4733                 drop_inode = 1;
4734         else {
4735                 inode->i_mapping->a_ops = &btrfs_aops;
4736                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4737                 inode->i_fop = &btrfs_file_operations;
4738                 inode->i_op = &btrfs_file_inode_operations;
4739                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4740         }
4741         dir->i_sb->s_dirt = 1;
4742         btrfs_update_inode_block_group(trans, inode);
4743         btrfs_update_inode_block_group(trans, dir);
4744         if (drop_inode)
4745                 goto out_unlock;
4746
4747         path = btrfs_alloc_path();
4748         BUG_ON(!path);
4749         key.objectid = inode->i_ino;
4750         key.offset = 0;
4751         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
4752         datasize = btrfs_file_extent_calc_inline_size(name_len);
4753         err = btrfs_insert_empty_item(trans, root, path, &key,
4754                                       datasize);
4755         if (err) {
4756                 drop_inode = 1;
4757                 goto out_unlock;
4758         }
4759         leaf = path->nodes[0];
4760         ei = btrfs_item_ptr(leaf, path->slots[0],
4761                             struct btrfs_file_extent_item);
4762         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
4763         btrfs_set_file_extent_type(leaf, ei,
4764                                    BTRFS_FILE_EXTENT_INLINE);
4765         btrfs_set_file_extent_encryption(leaf, ei, 0);
4766         btrfs_set_file_extent_compression(leaf, ei, 0);
4767         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
4768         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
4769
4770         ptr = btrfs_file_extent_inline_start(ei);
4771         write_extent_buffer(leaf, symname, ptr, name_len);
4772         btrfs_mark_buffer_dirty(leaf);
4773         btrfs_free_path(path);
4774
4775         inode->i_op = &btrfs_symlink_inode_operations;
4776         inode->i_mapping->a_ops = &btrfs_symlink_aops;
4777         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4778         inode_set_bytes(inode, name_len);
4779         btrfs_i_size_write(inode, name_len - 1);
4780         err = btrfs_update_inode(trans, root, inode);
4781         if (err)
4782                 drop_inode = 1;
4783
4784 out_unlock:
4785         nr = trans->blocks_used;
4786         btrfs_end_transaction_throttle(trans, root);
4787 out_fail:
4788         if (drop_inode) {
4789                 inode_dec_link_count(inode);
4790                 iput(inode);
4791         }
4792         btrfs_btree_balance_dirty(root, nr);
4793         return err;
4794 }
4795
4796 static int prealloc_file_range(struct inode *inode, u64 start, u64 end,
4797                                u64 alloc_hint, int mode)
4798 {
4799         struct btrfs_trans_handle *trans;
4800         struct btrfs_root *root = BTRFS_I(inode)->root;
4801         struct btrfs_key ins;
4802         u64 alloc_size;
4803         u64 cur_offset = start;
4804         u64 num_bytes = end - start;
4805         int ret = 0;
4806
4807         trans = btrfs_join_transaction(root, 1);
4808         BUG_ON(!trans);
4809         btrfs_set_trans_block_group(trans, inode);
4810
4811         while (num_bytes > 0) {
4812                 alloc_size = min(num_bytes, root->fs_info->max_extent);
4813                 ret = btrfs_reserve_extent(trans, root, alloc_size,
4814                                            root->sectorsize, 0, alloc_hint,
4815                                            (u64)-1, &ins, 1);
4816                 if (ret) {
4817                         WARN_ON(1);
4818                         goto out;
4819                 }
4820                 ret = insert_reserved_file_extent(trans, inode,
4821                                                   cur_offset, ins.objectid,
4822                                                   ins.offset, ins.offset,
4823                                                   ins.offset, 0, 0, 0,
4824                                                   BTRFS_FILE_EXTENT_PREALLOC);
4825                 BUG_ON(ret);
4826                 num_bytes -= ins.offset;
4827                 cur_offset += ins.offset;
4828                 alloc_hint = ins.objectid + ins.offset;
4829         }
4830 out:
4831         if (cur_offset > start) {
4832                 inode->i_ctime = CURRENT_TIME;
4833                 btrfs_set_flag(inode, PREALLOC);
4834                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4835                     cur_offset > i_size_read(inode))
4836                         btrfs_i_size_write(inode, cur_offset);
4837                 ret = btrfs_update_inode(trans, root, inode);
4838                 BUG_ON(ret);
4839         }
4840
4841         btrfs_end_transaction(trans, root);
4842         return ret;
4843 }
4844
4845 static long btrfs_fallocate(struct inode *inode, int mode,
4846                             loff_t offset, loff_t len)
4847 {
4848         u64 cur_offset;
4849         u64 last_byte;
4850         u64 alloc_start;
4851         u64 alloc_end;
4852         u64 alloc_hint = 0;
4853         u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
4854         struct extent_map *em;
4855         int ret;
4856
4857         alloc_start = offset & ~mask;
4858         alloc_end =  (offset + len + mask) & ~mask;
4859
4860         mutex_lock(&inode->i_mutex);
4861         if (alloc_start > inode->i_size) {
4862                 ret = btrfs_cont_expand(inode, alloc_start);
4863                 if (ret)
4864                         goto out;
4865         }
4866
4867         while (1) {
4868                 struct btrfs_ordered_extent *ordered;
4869                 lock_extent(&BTRFS_I(inode)->io_tree, alloc_start,
4870                             alloc_end - 1, GFP_NOFS);
4871                 ordered = btrfs_lookup_first_ordered_extent(inode,
4872                                                             alloc_end - 1);
4873                 if (ordered &&
4874                     ordered->file_offset + ordered->len > alloc_start &&
4875                     ordered->file_offset < alloc_end) {
4876                         btrfs_put_ordered_extent(ordered);
4877                         unlock_extent(&BTRFS_I(inode)->io_tree,
4878                                       alloc_start, alloc_end - 1, GFP_NOFS);
4879                         btrfs_wait_ordered_range(inode, alloc_start,
4880                                                  alloc_end - alloc_start);
4881                 } else {
4882                         if (ordered)
4883                                 btrfs_put_ordered_extent(ordered);
4884                         break;
4885                 }
4886         }
4887
4888         cur_offset = alloc_start;
4889         while (1) {
4890                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4891                                       alloc_end - cur_offset, 0);
4892                 BUG_ON(IS_ERR(em) || !em);
4893                 last_byte = min(extent_map_end(em), alloc_end);
4894                 last_byte = (last_byte + mask) & ~mask;
4895                 if (em->block_start == EXTENT_MAP_HOLE) {
4896                         ret = prealloc_file_range(inode, cur_offset,
4897                                         last_byte, alloc_hint, mode);
4898                         if (ret < 0) {
4899                                 free_extent_map(em);
4900                                 break;
4901                         }
4902                 }
4903                 if (em->block_start <= EXTENT_MAP_LAST_BYTE)
4904                         alloc_hint = em->block_start;
4905                 free_extent_map(em);
4906
4907                 cur_offset = last_byte;
4908                 if (cur_offset >= alloc_end) {
4909                         ret = 0;
4910                         break;
4911                 }
4912         }
4913         unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, alloc_end - 1,
4914                       GFP_NOFS);
4915 out:
4916         mutex_unlock(&inode->i_mutex);
4917         return ret;
4918 }
4919
4920 static int btrfs_set_page_dirty(struct page *page)
4921 {
4922         return __set_page_dirty_nobuffers(page);
4923 }
4924
4925 static int btrfs_permission(struct inode *inode, int mask)
4926 {
4927         if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
4928                 return -EACCES;
4929         return generic_permission(inode, mask, btrfs_check_acl);
4930 }
4931
4932 static struct inode_operations btrfs_dir_inode_operations = {
4933         .getattr        = btrfs_getattr,
4934         .lookup         = btrfs_lookup,
4935         .create         = btrfs_create,
4936         .unlink         = btrfs_unlink,
4937         .link           = btrfs_link,
4938         .mkdir          = btrfs_mkdir,
4939         .rmdir          = btrfs_rmdir,
4940         .rename         = btrfs_rename,
4941         .symlink        = btrfs_symlink,
4942         .setattr        = btrfs_setattr,
4943         .mknod          = btrfs_mknod,
4944         .setxattr       = btrfs_setxattr,
4945         .getxattr       = btrfs_getxattr,
4946         .listxattr      = btrfs_listxattr,
4947         .removexattr    = btrfs_removexattr,
4948         .permission     = btrfs_permission,
4949 };
4950 static struct inode_operations btrfs_dir_ro_inode_operations = {
4951         .lookup         = btrfs_lookup,
4952         .permission     = btrfs_permission,
4953 };
4954 static struct file_operations btrfs_dir_file_operations = {
4955         .llseek         = generic_file_llseek,
4956         .read           = generic_read_dir,
4957         .readdir        = btrfs_real_readdir,
4958         .unlocked_ioctl = btrfs_ioctl,
4959 #ifdef CONFIG_COMPAT
4960         .compat_ioctl   = btrfs_ioctl,
4961 #endif
4962         .release        = btrfs_release_file,
4963         .fsync          = btrfs_sync_file,
4964 };
4965
4966 static struct extent_io_ops btrfs_extent_io_ops = {
4967         .fill_delalloc = run_delalloc_range,
4968         .submit_bio_hook = btrfs_submit_bio_hook,
4969         .merge_bio_hook = btrfs_merge_bio_hook,
4970         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
4971         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
4972         .writepage_start_hook = btrfs_writepage_start_hook,
4973         .readpage_io_failed_hook = btrfs_io_failed_hook,
4974         .set_bit_hook = btrfs_set_bit_hook,
4975         .clear_bit_hook = btrfs_clear_bit_hook,
4976 };
4977
4978 static struct address_space_operations btrfs_aops = {
4979         .readpage       = btrfs_readpage,
4980         .writepage      = btrfs_writepage,
4981         .writepages     = btrfs_writepages,
4982         .readpages      = btrfs_readpages,
4983         .sync_page      = block_sync_page,
4984         .bmap           = btrfs_bmap,
4985         .direct_IO      = btrfs_direct_IO,
4986         .invalidatepage = btrfs_invalidatepage,
4987         .releasepage    = btrfs_releasepage,
4988         .set_page_dirty = btrfs_set_page_dirty,
4989 };
4990
4991 static struct address_space_operations btrfs_symlink_aops = {
4992         .readpage       = btrfs_readpage,
4993         .writepage      = btrfs_writepage,
4994         .invalidatepage = btrfs_invalidatepage,
4995         .releasepage    = btrfs_releasepage,
4996 };
4997
4998 static struct inode_operations btrfs_file_inode_operations = {
4999         .truncate       = btrfs_truncate,
5000         .getattr        = btrfs_getattr,
5001         .setattr        = btrfs_setattr,
5002         .setxattr       = btrfs_setxattr,
5003         .getxattr       = btrfs_getxattr,
5004         .listxattr      = btrfs_listxattr,
5005         .removexattr    = btrfs_removexattr,
5006         .permission     = btrfs_permission,
5007         .fallocate      = btrfs_fallocate,
5008 };
5009 static struct inode_operations btrfs_special_inode_operations = {
5010         .getattr        = btrfs_getattr,
5011         .setattr        = btrfs_setattr,
5012         .permission     = btrfs_permission,
5013         .setxattr       = btrfs_setxattr,
5014         .getxattr       = btrfs_getxattr,
5015         .listxattr      = btrfs_listxattr,
5016         .removexattr    = btrfs_removexattr,
5017 };
5018 static struct inode_operations btrfs_symlink_inode_operations = {
5019         .readlink       = generic_readlink,
5020         .follow_link    = page_follow_link_light,
5021         .put_link       = page_put_link,
5022         .permission     = btrfs_permission,
5023 };