fs/btrfs/inode.c

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