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 "ctree.h"
  41 #include "disk-io.h"
  42 #include "transaction.h"
  43 #include "btrfs_inode.h"
  44 #include "ioctl.h"
  45 #include "print-tree.h"
  46 #include "volumes.h"
  47 #include "ordered-data.h"
  48 #include "xattr.h"
  49 #include "compat.h"
  50 #include "tree-log.h"
  51
  52 struct btrfs_iget_args {
  53         u64 ino;
  54         struct btrfs_root *root;
  55 };
  56
  57 static struct inode_operations btrfs_dir_inode_operations;
  58 static struct inode_operations btrfs_symlink_inode_operations;
  59 static struct inode_operations btrfs_dir_ro_inode_operations;
  60 static struct inode_operations btrfs_special_inode_operations;
  61 static struct inode_operations btrfs_file_inode_operations;
  62 static struct address_space_operations btrfs_aops;
  63 static struct address_space_operations btrfs_symlink_aops;
  64 static struct file_operations btrfs_dir_file_operations;
  65 static struct extent_io_ops btrfs_extent_io_ops;
  66
  67 static struct kmem_cache *btrfs_inode_cachep;
  68 struct kmem_cache *btrfs_trans_handle_cachep;
  69 struct kmem_cache *btrfs_transaction_cachep;
  70 struct kmem_cache *btrfs_bit_radix_cachep;
  71 struct kmem_cache *btrfs_path_cachep;
  72
  73 #define S_SHIFT 12
  74 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  75         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  76         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  77         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  78         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  79         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  80         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  81         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  82 };
  83
  84 static void btrfs_truncate(struct inode *inode);
  85
  86 /*
  87  * a very lame attempt at stopping writes when the FS is 85% full.  There
  88  * are countless ways this is incorrect, but it is better than nothing.
  89  */
  90 int btrfs_check_free_space(struct btrfs_root *root, u64 num_required,
  91                            int for_del)
  92 {
  93         u64 total;
  94         u64 used;
  95         u64 thresh;
  96         unsigned long flags;
  97         int ret = 0;
  98
  99         spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
 100         total = btrfs_super_total_bytes(&root->fs_info->super_copy);
 101         used = btrfs_super_bytes_used(&root->fs_info->super_copy);
 102         if (for_del)
 103                 thresh = total * 90;
 104         else
 105                 thresh = total * 85;
 106
 107         do_div(thresh, 100);
 108
 109         if (used + root->fs_info->delalloc_bytes + num_required > thresh)
 110                 ret = -ENOSPC;
 111         spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
 112         return ret;
 113 }
 114
 115 /*
 116  * when extent_io.c finds a delayed allocation range in the file,
 117  * the call backs end up in this code.  The basic idea is to
 118  * allocate extents on disk for the range, and create ordered data structs
 119  * in ram to track those extents.
 120  */
 121 static int cow_file_range(struct inode *inode, u64 start, u64 end)
 122 {
 123         struct btrfs_root *root = BTRFS_I(inode)->root;
 124         struct btrfs_trans_handle *trans;
 125         u64 alloc_hint = 0;
 126         u64 num_bytes;
 127         u64 cur_alloc_size;
 128         u64 blocksize = root->sectorsize;
 129         u64 orig_num_bytes;
 130         struct btrfs_key ins;
 131         struct extent_map *em;
 132         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 133         int ret = 0;
 134
 135         trans = btrfs_join_transaction(root, 1);
 136         BUG_ON(!trans);
 137         btrfs_set_trans_block_group(trans, inode);
 138
 139         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 140         num_bytes = max(blocksize,  num_bytes);
 141         orig_num_bytes = num_bytes;
 142
 143         if (alloc_hint == EXTENT_MAP_INLINE)
 144                 goto out;
 145
 146         BUG_ON(num_bytes > btrfs_super_total_bytes(&root->fs_info->super_copy));
 147         mutex_lock(&BTRFS_I(inode)->extent_mutex);
 148         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 149         mutex_unlock(&BTRFS_I(inode)->extent_mutex);
 150
 151         while(num_bytes > 0) {
 152                 cur_alloc_size = min(num_bytes, root->fs_info->max_extent);
 153                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 154                                            root->sectorsize, 0, alloc_hint,
 155                                            (u64)-1, &ins, 1);
 156                 if (ret) {
 157                         WARN_ON(1);
 158                         goto out;
 159                 }
 160                 em = alloc_extent_map(GFP_NOFS);
 161                 em->start = start;
 162                 em->len = ins.offset;
 163                 em->block_start = ins.objectid;
 164                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 165                 mutex_lock(&BTRFS_I(inode)->extent_mutex);
 166                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 167                 while(1) {
 168                         spin_lock(&em_tree->lock);
 169                         ret = add_extent_mapping(em_tree, em);
 170                         spin_unlock(&em_tree->lock);
 171                         if (ret != -EEXIST) {
 172                                 free_extent_map(em);
 173                                 break;
 174                         }
 175                         btrfs_drop_extent_cache(inode, start,
 176                                                 start + ins.offset - 1, 0);
 177                 }
 178                 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
 179
 180                 cur_alloc_size = ins.offset;
 181                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 182                                                ins.offset, 0);
 183                 BUG_ON(ret);
 184                 if (num_bytes < cur_alloc_size) {
 185                         printk("num_bytes %Lu cur_alloc %Lu\n", num_bytes,
 186                                cur_alloc_size);
 187                         break;
 188                 }
 189                 num_bytes -= cur_alloc_size;
 190                 alloc_hint = ins.objectid + ins.offset;
 191                 start += cur_alloc_size;
 192         }
 193 out:
 194         btrfs_end_transaction(trans, root);
 195         return ret;
 196 }
 197
 198 /*
 199  * when nowcow writeback call back.  This checks for snapshots or COW copies
 200  * of the extents that exist in the file, and COWs the file as required.
 201  *
 202  * If no cow copies or snapshots exist, we write directly to the existing
 203  * blocks on disk
 204  */
 205 static int run_delalloc_nocow(struct inode *inode, u64 start, u64 end)
 206 {
 207         u64 extent_start;
 208         u64 extent_end;
 209         u64 bytenr;
 210         u64 loops = 0;
 211         u64 total_fs_bytes;
 212         struct btrfs_root *root = BTRFS_I(inode)->root;
 213         struct btrfs_block_group_cache *block_group;
 214         struct btrfs_trans_handle *trans;
 215         struct extent_buffer *leaf;
 216         int found_type;
 217         struct btrfs_path *path;
 218         struct btrfs_file_extent_item *item;
 219         int ret;
 220         int err = 0;
 221         struct btrfs_key found_key;
 222
 223         total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
 224         path = btrfs_alloc_path();
 225         BUG_ON(!path);
 226         trans = btrfs_join_transaction(root, 1);
 227         BUG_ON(!trans);
 228 again:
 229         ret = btrfs_lookup_file_extent(NULL, root, path,
 230                                        inode->i_ino, start, 0);
 231         if (ret < 0) {
 232                 err = ret;
 233                 goto out;
 234         }
 235
 236         if (ret != 0) {
 237                 if (path->slots[0] == 0)
 238                         goto not_found;
 239                 path->slots[0]--;
 240         }
 241
 242         leaf = path->nodes[0];
 243         item = btrfs_item_ptr(leaf, path->slots[0],
 244                               struct btrfs_file_extent_item);
 245
 246         /* are we inside the extent that was found? */
 247         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 248         found_type = btrfs_key_type(&found_key);
 249         if (found_key.objectid != inode->i_ino ||
 250             found_type != BTRFS_EXTENT_DATA_KEY)
 251                 goto not_found;
 252
 253         found_type = btrfs_file_extent_type(leaf, item);
 254         extent_start = found_key.offset;
 255         if (found_type == BTRFS_FILE_EXTENT_REG) {
 256                 u64 extent_num_bytes;
 257
 258                 extent_num_bytes = btrfs_file_extent_num_bytes(leaf, item);
 259                 extent_end = extent_start + extent_num_bytes;
 260                 err = 0;
 261
 262                 if (loops && start != extent_start)
 263                         goto not_found;
 264
 265                 if (start < extent_start || start >= extent_end)
 266                         goto not_found;
 267
 268                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
 269                 if (bytenr == 0)
 270                         goto not_found;
 271
 272                 if (btrfs_cross_ref_exists(trans, root, &found_key, bytenr))
 273                         goto not_found;
 274                 /*
 275                  * we may be called by the resizer, make sure we're inside
 276                  * the limits of the FS
 277                  */
 278                 block_group = btrfs_lookup_block_group(root->fs_info,
 279                                                        bytenr);
 280                 if (!block_group || block_group->ro)
 281                         goto not_found;
 282
 283                 bytenr += btrfs_file_extent_offset(leaf, item);
 284                 extent_num_bytes = min(end + 1, extent_end) - start;
 285                 ret = btrfs_add_ordered_extent(inode, start, bytenr,
 286                                                 extent_num_bytes, 1);
 287                 if (ret) {
 288                         err = ret;
 289                         goto out;
 290                 }
 291
 292                 btrfs_release_path(root, path);
 293                 start = extent_end;
 294                 if (start <= end) {
 295                         loops++;
 296                         goto again;
 297                 }
 298         } else {
 299 not_found:
 300                 btrfs_end_transaction(trans, root);
 301                 btrfs_free_path(path);
 302                 return cow_file_range(inode, start, end);
 303         }
 304 out:
 305         WARN_ON(err);
 306         btrfs_end_transaction(trans, root);
 307         btrfs_free_path(path);
 308         return err;
 309 }
 310
 311 /*
 312  * extent_io.c call back to do delayed allocation processing
 313  */
 314 static int run_delalloc_range(struct inode *inode, u64 start, u64 end)
 315 {
 316         struct btrfs_root *root = BTRFS_I(inode)->root;
 317         int ret;
 318
 319         if (btrfs_test_opt(root, NODATACOW) ||
 320             btrfs_test_flag(inode, NODATACOW))
 321                 ret = run_delalloc_nocow(inode, start, end);
 322         else
 323                 ret = cow_file_range(inode, start, end);
 324
 325         return ret;
 326 }
 327
 328 /*
 329  * extent_io.c set_bit_hook, used to track delayed allocation
 330  * bytes in this file, and to maintain the list of inodes that
 331  * have pending delalloc work to be done.
 332  */
 333 int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
 334                        unsigned long old, unsigned long bits)
 335 {
 336         unsigned long flags;
 337         if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
 338                 struct btrfs_root *root = BTRFS_I(inode)->root;
 339                 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
 340                 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
 341                 root->fs_info->delalloc_bytes += end - start + 1;
 342                 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
 343                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
 344                                       &root->fs_info->delalloc_inodes);
 345                 }
 346                 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
 347         }
 348         return 0;
 349 }
 350
 351 /*
 352  * extent_io.c clear_bit_hook, see set_bit_hook for why
 353  */
 354 int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
 355                          unsigned long old, unsigned long bits)
 356 {
 357         if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
 358                 struct btrfs_root *root = BTRFS_I(inode)->root;
 359                 unsigned long flags;
 360
 361                 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
 362                 if (end - start + 1 > root->fs_info->delalloc_bytes) {
 363                         printk("warning: delalloc account %Lu %Lu\n",
 364                                end - start + 1, root->fs_info->delalloc_bytes);
 365                         root->fs_info->delalloc_bytes = 0;
 366                         BTRFS_I(inode)->delalloc_bytes = 0;
 367                 } else {
 368                         root->fs_info->delalloc_bytes -= end - start + 1;
 369                         BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
 370                 }
 371                 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
 372                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
 373                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
 374                 }
 375                 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
 376         }
 377         return 0;
 378 }
 379
 380 /*
 381  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
 382  * we don't create bios that span stripes or chunks
 383  */
 384 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
 385                          size_t size, struct bio *bio)
 386 {
 387         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 388         struct btrfs_mapping_tree *map_tree;
 389         u64 logical = bio->bi_sector << 9;
 390         u64 length = 0;
 391         u64 map_length;
 392         int ret;
 393
 394         length = bio->bi_size;
 395         map_tree = &root->fs_info->mapping_tree;
 396         map_length = length;
 397         ret = btrfs_map_block(map_tree, READ, logical,
 398                               &map_length, NULL, 0);
 399
 400         if (map_length < length + size) {
 401                 return 1;
 402         }
 403         return 0;
 404 }
 405
 406 /*
 407  * in order to insert checksums into the metadata in large chunks,
 408  * we wait until bio submission time.   All the pages in the bio are
 409  * checksummed and sums are attached onto the ordered extent record.
 410  *
 411  * At IO completion time the cums attached on the ordered extent record
 412  * are inserted into the btree
 413  */
 414 int __btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 415                           int mirror_num)
 416 {
 417         struct btrfs_root *root = BTRFS_I(inode)->root;
 418         int ret = 0;
 419
 420         ret = btrfs_csum_one_bio(root, inode, bio);
 421         BUG_ON(ret);
 422
 423         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
 424 }
 425
 426 /*
 427  * extent_io.c submission hook. This does the right thing for csum calculation on write,
 428  * or reading the csums from the tree before a read
 429  */
 430 int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 431                           int mirror_num)
 432 {
 433         struct btrfs_root *root = BTRFS_I(inode)->root;
 434         int ret = 0;
 435
 436         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
 437         BUG_ON(ret);
 438
 439         if (btrfs_test_opt(root, NODATASUM) ||
 440             btrfs_test_flag(inode, NODATASUM)) {
 441                 goto mapit;
 442         }
 443
 444         if (!(rw & (1 << BIO_RW))) {
 445                 btrfs_lookup_bio_sums(root, inode, bio);
 446                 goto mapit;
 447         }
 448         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 449                                    inode, rw, bio, mirror_num,
 450                                    __btrfs_submit_bio_hook);
 451 mapit:
 452         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
 453 }
 454
 455 /*
 456  * given a list of ordered sums record them in the inode.  This happens
 457  * at IO completion time based on sums calculated at bio submission time.
 458  */
 459 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
 460                              struct inode *inode, u64 file_offset,
 461                              struct list_head *list)
 462 {
 463         struct list_head *cur;
 464         struct btrfs_ordered_sum *sum;
 465
 466         btrfs_set_trans_block_group(trans, inode);
 467         list_for_each(cur, list) {
 468                 sum = list_entry(cur, struct btrfs_ordered_sum, list);
 469                 btrfs_csum_file_blocks(trans, BTRFS_I(inode)->root,
 470                                        inode, sum);
 471         }
 472         return 0;
 473 }
 474
 475 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end)
 476 {
 477         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
 478                                    GFP_NOFS);
 479 }
 480
 481 /* see btrfs_writepage_start_hook for details on why this is required */
 482 struct btrfs_writepage_fixup {
 483         struct page *page;
 484         struct btrfs_work work;
 485 };
 486
 487 void btrfs_writepage_fixup_worker(struct btrfs_work *work)
 488 {
 489         struct btrfs_writepage_fixup *fixup;
 490         struct btrfs_ordered_extent *ordered;
 491         struct page *page;
 492         struct inode *inode;
 493         u64 page_start;
 494         u64 page_end;
 495
 496         fixup = container_of(work, struct btrfs_writepage_fixup, work);
 497         page = fixup->page;
 498 again:
 499         lock_page(page);
 500         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
 501                 ClearPageChecked(page);
 502                 goto out_page;
 503         }
 504
 505         inode = page->mapping->host;
 506         page_start = page_offset(page);
 507         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
 508
 509         lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
 510
 511         /* already ordered? We're done */
 512         if (test_range_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
 513                              EXTENT_ORDERED, 0)) {
 514                 goto out;
 515         }
 516
 517         ordered = btrfs_lookup_ordered_extent(inode, page_start);
 518         if (ordered) {
 519                 unlock_extent(&BTRFS_I(inode)->io_tree, page_start,
 520                               page_end, GFP_NOFS);
 521                 unlock_page(page);
 522                 btrfs_start_ordered_extent(inode, ordered, 1);
 523                 goto again;
 524         }
 525
 526         btrfs_set_extent_delalloc(inode, page_start, page_end);
 527         ClearPageChecked(page);
 528 out:
 529         unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
 530 out_page:
 531         unlock_page(page);
 532         page_cache_release(page);
 533 }
 534
 535 /*
 536  * There are a few paths in the higher layers of the kernel that directly
 537  * set the page dirty bit without asking the filesystem if it is a
 538  * good idea.  This causes problems because we want to make sure COW
 539  * properly happens and the data=ordered rules are followed.
 540  *
 541  * In our case any range that doesn't have the EXTENT_ORDERED bit set
 542  * hasn't been properly setup for IO.  We kick off an async process
 543  * to fix it up.  The async helper will wait for ordered extents, set
 544  * the delalloc bit and make it safe to write the page.
 545  */
 546 int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
 547 {
 548         struct inode *inode = page->mapping->host;
 549         struct btrfs_writepage_fixup *fixup;
 550         struct btrfs_root *root = BTRFS_I(inode)->root;
 551         int ret;
 552
 553         ret = test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
 554                              EXTENT_ORDERED, 0);
 555         if (ret)
 556                 return 0;
 557
 558         if (PageChecked(page))
 559                 return -EAGAIN;
 560
 561         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
 562         if (!fixup)
 563                 return -EAGAIN;
 564
 565         SetPageChecked(page);
 566         page_cache_get(page);
 567         fixup->work.func = btrfs_writepage_fixup_worker;
 568         fixup->page = page;
 569         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
 570         return -EAGAIN;
 571 }
 572
 573 /* as ordered data IO finishes, this gets called so we can finish
 574  * an ordered extent if the range of bytes in the file it covers are
 575  * fully written.
 576  */
 577 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
 578 {
 579         struct btrfs_root *root = BTRFS_I(inode)->root;
 580         struct btrfs_trans_handle *trans;
 581         struct btrfs_ordered_extent *ordered_extent;
 582         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 583         struct btrfs_file_extent_item *extent_item;
 584         struct btrfs_path *path = NULL;
 585         struct extent_buffer *leaf;
 586         u64 alloc_hint = 0;
 587         struct list_head list;
 588         struct btrfs_key ins;
 589         int ret;
 590
 591         ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
 592         if (!ret)
 593                 return 0;
 594
 595         trans = btrfs_join_transaction(root, 1);
 596
 597         ordered_extent = btrfs_lookup_ordered_extent(inode, start);
 598         BUG_ON(!ordered_extent);
 599         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
 600                 goto nocow;
 601
 602         path = btrfs_alloc_path();
 603         BUG_ON(!path);
 604
 605         lock_extent(io_tree, ordered_extent->file_offset,
 606                     ordered_extent->file_offset + ordered_extent->len - 1,
 607                     GFP_NOFS);
 608
 609         INIT_LIST_HEAD(&list);
 610
 611         mutex_lock(&BTRFS_I(inode)->extent_mutex);
 612
 613         ret = btrfs_drop_extents(trans, root, inode,
 614                                  ordered_extent->file_offset,
 615                                  ordered_extent->file_offset +
 616                                  ordered_extent->len,
 617                                  ordered_extent->file_offset, &alloc_hint);
 618         BUG_ON(ret);
 619
 620         ins.objectid = inode->i_ino;
 621         ins.offset = ordered_extent->file_offset;
 622         ins.type = BTRFS_EXTENT_DATA_KEY;
 623         ret = btrfs_insert_empty_item(trans, root, path, &ins,
 624                                       sizeof(*extent_item));
 625         BUG_ON(ret);
 626         leaf = path->nodes[0];
 627         extent_item = btrfs_item_ptr(leaf, path->slots[0],
 628                                      struct btrfs_file_extent_item);
 629         btrfs_set_file_extent_generation(leaf, extent_item, trans->transid);
 630         btrfs_set_file_extent_type(leaf, extent_item, BTRFS_FILE_EXTENT_REG);
 631         btrfs_set_file_extent_disk_bytenr(leaf, extent_item,
 632                                           ordered_extent->start);
 633         btrfs_set_file_extent_disk_num_bytes(leaf, extent_item,
 634                                              ordered_extent->len);
 635         btrfs_set_file_extent_offset(leaf, extent_item, 0);
 636         btrfs_set_file_extent_num_bytes(leaf, extent_item,
 637                                         ordered_extent->len);
 638         btrfs_mark_buffer_dirty(leaf);
 639
 640         btrfs_drop_extent_cache(inode, ordered_extent->file_offset,
 641                                 ordered_extent->file_offset +
 642                                 ordered_extent->len - 1, 0);
 643         mutex_unlock(&BTRFS_I(inode)->extent_mutex);
 644
 645         ins.objectid = ordered_extent->start;
 646         ins.offset = ordered_extent->len;
 647         ins.type = BTRFS_EXTENT_ITEM_KEY;
 648         ret = btrfs_alloc_reserved_extent(trans, root, leaf->start,
 649                                           root->root_key.objectid,
 650                                           trans->transid, inode->i_ino,
 651                                           ordered_extent->file_offset, &ins);
 652         BUG_ON(ret);
 653         btrfs_release_path(root, path);
 654
 655         inode->i_blocks += ordered_extent->len >> 9;
 656         unlock_extent(io_tree, ordered_extent->file_offset,
 657                     ordered_extent->file_offset + ordered_extent->len - 1,
 658                     GFP_NOFS);
 659 nocow:
 660         add_pending_csums(trans, inode, ordered_extent->file_offset,
 661                           &ordered_extent->list);
 662
 663         mutex_lock(&BTRFS_I(inode)->extent_mutex);
 664         btrfs_ordered_update_i_size(inode, ordered_extent);
 665         btrfs_update_inode(trans, root, inode);
 666         btrfs_remove_ordered_extent(inode, ordered_extent);
 667         mutex_unlock(&BTRFS_I(inode)->extent_mutex);
 668
 669         /* once for us */
 670         btrfs_put_ordered_extent(ordered_extent);
 671         /* once for the tree */
 672         btrfs_put_ordered_extent(ordered_extent);
 673
 674         btrfs_end_transaction(trans, root);
 675         if (path)
 676                 btrfs_free_path(path);
 677         return 0;
 678 }
 679
 680 int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
 681                                 struct extent_state *state, int uptodate)
 682 {
 683         return btrfs_finish_ordered_io(page->mapping->host, start, end);
 684 }
 685
 686 /*
 687  * When IO fails, either with EIO or csum verification fails, we
 688  * try other mirrors that might have a good copy of the data.  This
 689  * io_failure_record is used to record state as we go through all the
 690  * mirrors.  If another mirror has good data, the page is set up to date
 691  * and things continue.  If a good mirror can't be found, the original
 692  * bio end_io callback is called to indicate things have failed.
 693  */
 694 struct io_failure_record {
 695         struct page *page;
 696         u64 start;
 697         u64 len;
 698         u64 logical;
 699         int last_mirror;
 700 };
 701
 702 int btrfs_io_failed_hook(struct bio *failed_bio,
 703                          struct page *page, u64 start, u64 end,
 704                          struct extent_state *state)
 705 {
 706         struct io_failure_record *failrec = NULL;
 707         u64 private;
 708         struct extent_map *em;
 709         struct inode *inode = page->mapping->host;
 710         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
 711         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 712         struct bio *bio;
 713         int num_copies;
 714         int ret;
 715         int rw;
 716         u64 logical;
 717
 718         ret = get_state_private(failure_tree, start, &private);
 719         if (ret) {
 720                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
 721                 if (!failrec)
 722                         return -ENOMEM;
 723                 failrec->start = start;
 724                 failrec->len = end - start + 1;
 725                 failrec->last_mirror = 0;
 726
 727                 spin_lock(&em_tree->lock);
 728                 em = lookup_extent_mapping(em_tree, start, failrec->len);
 729                 if (em->start > start || em->start + em->len < start) {
 730                         free_extent_map(em);
 731                         em = NULL;
 732                 }
 733                 spin_unlock(&em_tree->lock);
 734
 735                 if (!em || IS_ERR(em)) {
 736                         kfree(failrec);
 737                         return -EIO;
 738                 }
 739                 logical = start - em->start;
 740                 logical = em->block_start + logical;
 741                 failrec->logical = logical;
 742                 free_extent_map(em);
 743                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
 744                                 EXTENT_DIRTY, GFP_NOFS);
 745                 set_state_private(failure_tree, start,
 746                                  (u64)(unsigned long)failrec);
 747         } else {
 748                 failrec = (struct io_failure_record *)(unsigned long)private;
 749         }
 750         num_copies = btrfs_num_copies(
 751                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
 752                               failrec->logical, failrec->len);
 753         failrec->last_mirror++;
 754         if (!state) {
 755                 spin_lock_irq(&BTRFS_I(inode)->io_tree.lock);
 756                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
 757                                                     failrec->start,
 758                                                     EXTENT_LOCKED);
 759                 if (state && state->start != failrec->start)
 760                         state = NULL;
 761                 spin_unlock_irq(&BTRFS_I(inode)->io_tree.lock);
 762         }
 763         if (!state || failrec->last_mirror > num_copies) {
 764                 set_state_private(failure_tree, failrec->start, 0);
 765                 clear_extent_bits(failure_tree, failrec->start,
 766                                   failrec->start + failrec->len - 1,
 767                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
 768                 kfree(failrec);
 769                 return -EIO;
 770         }
 771         bio = bio_alloc(GFP_NOFS, 1);
 772         bio->bi_private = state;
 773         bio->bi_end_io = failed_bio->bi_end_io;
 774         bio->bi_sector = failrec->logical >> 9;
 775         bio->bi_bdev = failed_bio->bi_bdev;
 776         bio->bi_size = 0;
 777         bio_add_page(bio, page, failrec->len, start - page_offset(page));
 778         if (failed_bio->bi_rw & (1 << BIO_RW))
 779                 rw = WRITE;
 780         else
 781                 rw = READ;
 782
 783         BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
 784                                                       failrec->last_mirror);
 785         return 0;
 786 }
 787
 788 /*
 789  * each time an IO finishes, we do a fast check in the IO failure tree
 790  * to see if we need to process or clean up an io_failure_record
 791  */
 792 int btrfs_clean_io_failures(struct inode *inode, u64 start)
 793 {
 794         u64 private;
 795         u64 private_failure;
 796         struct io_failure_record *failure;
 797         int ret;
 798
 799         private = 0;
 800         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
 801                              (u64)-1, 1, EXTENT_DIRTY)) {
 802                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
 803                                         start, &private_failure);
 804                 if (ret == 0) {
 805                         failure = (struct io_failure_record *)(unsigned long)
 806                                    private_failure;
 807                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
 808                                           failure->start, 0);
 809                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
 810                                           failure->start,
 811                                           failure->start + failure->len - 1,
 812                                           EXTENT_DIRTY | EXTENT_LOCKED,
 813                                           GFP_NOFS);
 814                         kfree(failure);
 815                 }
 816         }
 817         return 0;
 818 }
 819
 820 /*
 821  * when reads are done, we need to check csums to verify the data is correct
 822  * if there's a match, we allow the bio to finish.  If not, we go through
 823  * the io_failure_record routines to find good copies
 824  */
 825 int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 826                                struct extent_state *state)
 827 {
 828         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
 829         struct inode *inode = page->mapping->host;
 830         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 831         char *kaddr;
 832         u64 private = ~(u32)0;
 833         int ret;
 834         struct btrfs_root *root = BTRFS_I(inode)->root;
 835         u32 csum = ~(u32)0;
 836         unsigned long flags;
 837
 838         if (btrfs_test_opt(root, NODATASUM) ||
 839             btrfs_test_flag(inode, NODATASUM))
 840                 return 0;
 841         if (state && state->start == start) {
 842                 private = state->private;
 843                 ret = 0;
 844         } else {
 845                 ret = get_state_private(io_tree, start, &private);
 846         }
 847         local_irq_save(flags);
 848         kaddr = kmap_atomic(page, KM_IRQ0);
 849         if (ret) {
 850                 goto zeroit;
 851         }
 852         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
 853         btrfs_csum_final(csum, (char *)&csum);
 854         if (csum != private) {
 855                 goto zeroit;
 856         }
 857         kunmap_atomic(kaddr, KM_IRQ0);
 858         local_irq_restore(flags);
 859
 860         /* if the io failure tree for this inode is non-empty,
 861          * check to see if we've recovered from a failed IO
 862          */
 863         btrfs_clean_io_failures(inode, start);
 864         return 0;
 865
 866 zeroit:
 867         printk("btrfs csum failed ino %lu off %llu csum %u private %Lu\n",
 868                page->mapping->host->i_ino, (unsigned long long)start, csum,
 869                private);
 870         memset(kaddr + offset, 1, end - start + 1);
 871         flush_dcache_page(page);
 872         kunmap_atomic(kaddr, KM_IRQ0);
 873         local_irq_restore(flags);
 874         if (private == 0)
 875                 return 0;
 876         return -EIO;
 877 }
 878
 879 /*
 880  * This creates an orphan entry for the given inode in case something goes
 881  * wrong in the middle of an unlink/truncate.
 882  */
 883 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
 884 {
 885         struct btrfs_root *root = BTRFS_I(inode)->root;
 886         int ret = 0;
 887
 888         spin_lock(&root->list_lock);
 889
 890         /* already on the orphan list, we're good */
 891         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
 892                 spin_unlock(&root->list_lock);
 893                 return 0;
 894         }
 895
 896         list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
 897
 898         spin_unlock(&root->list_lock);
 899
 900         /*
 901          * insert an orphan item to track this unlinked/truncated file
 902          */
 903         ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
 904
 905         return ret;
 906 }
 907
 908 /*
 909  * We have done the truncate/delete so we can go ahead and remove the orphan
 910  * item for this particular inode.
 911  */
 912 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
 913 {
 914         struct btrfs_root *root = BTRFS_I(inode)->root;
 915         int ret = 0;
 916
 917         spin_lock(&root->list_lock);
 918
 919         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
 920                 spin_unlock(&root->list_lock);
 921                 return 0;
 922         }
 923
 924         list_del_init(&BTRFS_I(inode)->i_orphan);
 925         if (!trans) {
 926                 spin_unlock(&root->list_lock);
 927                 return 0;
 928         }
 929
 930         spin_unlock(&root->list_lock);
 931
 932         ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
 933
 934         return ret;
 935 }
 936
 937 /*
 938  * this cleans up any orphans that may be left on the list from the last use
 939  * of this root.
 940  */
 941 void btrfs_orphan_cleanup(struct btrfs_root *root)
 942 {
 943         struct btrfs_path *path;
 944         struct extent_buffer *leaf;
 945         struct btrfs_item *item;
 946         struct btrfs_key key, found_key;
 947         struct btrfs_trans_handle *trans;
 948         struct inode *inode;
 949         int ret = 0, nr_unlink = 0, nr_truncate = 0;
 950
 951         /* don't do orphan cleanup if the fs is readonly. */
 952         if (root->fs_info->sb->s_flags & MS_RDONLY)
 953                 return;
 954
 955         path = btrfs_alloc_path();
 956         if (!path)
 957                 return;
 958         path->reada = -1;
 959
 960         key.objectid = BTRFS_ORPHAN_OBJECTID;
 961         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
 962         key.offset = (u64)-1;
 963
 964
 965         while (1) {
 966                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 967                 if (ret < 0) {
 968                         printk(KERN_ERR "Error searching slot for orphan: %d"
 969                                "\n", ret);
 970                         break;
 971                 }
 972
 973                 /*
 974                  * if ret == 0 means we found what we were searching for, which
 975                  * is weird, but possible, so only screw with path if we didnt
 976                  * find the key and see if we have stuff that matches
 977                  */
 978                 if (ret > 0) {
 979                         if (path->slots[0] == 0)
 980                                 break;
 981                         path->slots[0]--;
 982                 }
 983
 984                 /* pull out the item */
 985                 leaf = path->nodes[0];
 986                 item = btrfs_item_nr(leaf, path->slots[0]);
 987                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 988
 989                 /* make sure the item matches what we want */
 990                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
 991                         break;
 992                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
 993                         break;
 994
 995                 /* release the path since we're done with it */
 996                 btrfs_release_path(root, path);
 997
 998                 /*
 999                  * this is where we are basically btrfs_lookup, without the
1000                  * crossing root thing.  we store the inode number in the
1001                  * offset of the orphan item.
1002                  */
1003                 inode = btrfs_iget_locked(root->fs_info->sb,
1004                                           found_key.offset, root);
1005                 if (!inode)
1006                         break;
1007
1008                 if (inode->i_state & I_NEW) {
1009                         BTRFS_I(inode)->root = root;
1010
1011                         /* have to set the location manually */
1012                         BTRFS_I(inode)->location.objectid = inode->i_ino;
1013                         BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
1014                         BTRFS_I(inode)->location.offset = 0;
1015
1016                         btrfs_read_locked_inode(inode);
1017                         unlock_new_inode(inode);
1018                 }
1019
1020                 /*
1021                  * add this inode to the orphan list so btrfs_orphan_del does
1022                  * the proper thing when we hit it
1023                  */
1024                 spin_lock(&root->list_lock);
1025                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1026                 spin_unlock(&root->list_lock);
1027
1028                 /*
1029                  * if this is a bad inode, means we actually succeeded in
1030                  * removing the inode, but not the orphan record, which means
1031                  * we need to manually delete the orphan since iput will just
1032                  * do a destroy_inode
1033                  */
1034                 if (is_bad_inode(inode)) {
1035                         trans = btrfs_start_transaction(root, 1);
1036                         btrfs_orphan_del(trans, inode);
1037                         btrfs_end_transaction(trans, root);
1038                         iput(inode);
1039                         continue;
1040                 }
1041
1042                 /* if we have links, this was a truncate, lets do that */
1043                 if (inode->i_nlink) {
1044                         nr_truncate++;
1045                         btrfs_truncate(inode);
1046                 } else {
1047                         nr_unlink++;
1048                 }
1049
1050                 /* this will do delete_inode and everything for us */
1051                 iput(inode);
1052         }
1053
1054         if (nr_unlink)
1055                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
1056         if (nr_truncate)
1057                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
1058
1059         btrfs_free_path(path);
1060 }
1061
1062 /*
1063  * read an inode from the btree into the in-memory inode
1064  */
1065 void btrfs_read_locked_inode(struct inode *inode)
1066 {
1067         struct btrfs_path *path;
1068         struct extent_buffer *leaf;
1069         struct btrfs_inode_item *inode_item;
1070         struct btrfs_timespec *tspec;
1071         struct btrfs_root *root = BTRFS_I(inode)->root;
1072         struct btrfs_key location;
1073         u64 alloc_group_block;
1074         u32 rdev;
1075         int ret;
1076
1077         path = btrfs_alloc_path();
1078         BUG_ON(!path);
1079         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
1080
1081         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
1082         if (ret)
1083                 goto make_bad;
1084
1085         leaf = path->nodes[0];
1086         inode_item = btrfs_item_ptr(leaf, path->slots[0],
1087                                     struct btrfs_inode_item);
1088
1089         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
1090         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
1091         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
1092         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
1093         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
1094
1095         tspec = btrfs_inode_atime(inode_item);
1096         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1097         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1098
1099         tspec = btrfs_inode_mtime(inode_item);
1100         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1101         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1102
1103         tspec = btrfs_inode_ctime(inode_item);
1104         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1105         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1106
1107         inode->i_blocks = btrfs_inode_nblocks(leaf, inode_item);
1108         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
1109         inode->i_generation = BTRFS_I(inode)->generation;
1110         inode->i_rdev = 0;
1111         rdev = btrfs_inode_rdev(leaf, inode_item);
1112
1113         BTRFS_I(inode)->index_cnt = (u64)-1;
1114
1115         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
1116         BTRFS_I(inode)->block_group = btrfs_lookup_block_group(root->fs_info,
1117                                                        alloc_group_block);
1118         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
1119         if (!BTRFS_I(inode)->block_group) {
1120                 BTRFS_I(inode)->block_group = btrfs_find_block_group(root,
1121                                                  NULL, 0,
1122                                                  BTRFS_BLOCK_GROUP_METADATA, 0);
1123         }
1124         btrfs_free_path(path);
1125         inode_item = NULL;
1126
1127         switch (inode->i_mode & S_IFMT) {
1128         case S_IFREG:
1129                 inode->i_mapping->a_ops = &btrfs_aops;
1130                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
1131                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
1132                 inode->i_fop = &btrfs_file_operations;
1133                 inode->i_op = &btrfs_file_inode_operations;
1134                 break;
1135         case S_IFDIR:
1136                 inode->i_fop = &btrfs_dir_file_operations;
1137                 if (root == root->fs_info->tree_root)
1138                         inode->i_op = &btrfs_dir_ro_inode_operations;
1139                 else
1140                         inode->i_op = &btrfs_dir_inode_operations;
1141                 break;
1142         case S_IFLNK:
1143                 inode->i_op = &btrfs_symlink_inode_operations;
1144                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
1145                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
1146                 break;
1147         default:
1148                 init_special_inode(inode, inode->i_mode, rdev);
1149                 break;
1150         }
1151         return;
1152
1153 make_bad:
1154         btrfs_free_path(path);
1155         make_bad_inode(inode);
1156 }
1157
1158 /*
1159  * given a leaf and an inode, copy the inode fields into the leaf
1160  */
1161 static void fill_inode_item(struct btrfs_trans_handle *trans,
1162                             struct extent_buffer *leaf,
1163                             struct btrfs_inode_item *item,
1164                             struct inode *inode)
1165 {
1166         btrfs_set_inode_uid(leaf, item, inode->i_uid);
1167         btrfs_set_inode_gid(leaf, item, inode->i_gid);
1168         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
1169         btrfs_set_inode_mode(leaf, item, inode->i_mode);
1170         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
1171
1172         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
1173                                inode->i_atime.tv_sec);
1174         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
1175                                 inode->i_atime.tv_nsec);
1176
1177         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
1178                                inode->i_mtime.tv_sec);
1179         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
1180                                 inode->i_mtime.tv_nsec);
1181
1182         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
1183                                inode->i_ctime.tv_sec);
1184         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
1185                                 inode->i_ctime.tv_nsec);
1186
1187         btrfs_set_inode_nblocks(leaf, item, inode->i_blocks);
1188         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
1189         btrfs_set_inode_transid(leaf, item, trans->transid);
1190         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
1191         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
1192         btrfs_set_inode_block_group(leaf, item,
1193                                     BTRFS_I(inode)->block_group->key.objectid);
1194 }
1195
1196 /*
1197  * copy everything in the in-memory inode into the btree.
1198  */
1199 int noinline btrfs_update_inode(struct btrfs_trans_handle *trans,
1200                               struct btrfs_root *root,
1201                               struct inode *inode)
1202 {
1203         struct btrfs_inode_item *inode_item;
1204         struct btrfs_path *path;
1205         struct extent_buffer *leaf;
1206         int ret;
1207
1208         path = btrfs_alloc_path();
1209         BUG_ON(!path);
1210         ret = btrfs_lookup_inode(trans, root, path,
1211                                  &BTRFS_I(inode)->location, 1);
1212         if (ret) {
1213                 if (ret > 0)
1214                         ret = -ENOENT;
1215                 goto failed;
1216         }
1217
1218         leaf = path->nodes[0];
1219         inode_item = btrfs_item_ptr(leaf, path->slots[0],
1220                                   struct btrfs_inode_item);
1221
1222         fill_inode_item(trans, leaf, inode_item, inode);
1223         btrfs_mark_buffer_dirty(leaf);
1224         btrfs_set_inode_last_trans(trans, inode);
1225         ret = 0;
1226 failed:
1227         btrfs_free_path(path);
1228         return ret;
1229 }
1230
1231
1232 /*
1233  * unlink helper that gets used here in inode.c and in the tree logging
1234  * recovery code.  It remove a link in a directory with a given name, and
1235  * also drops the back refs in the inode to the directory
1236  */
1237 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
1238                        struct btrfs_root *root,
1239                        struct inode *dir, struct inode *inode,
1240                        const char *name, int name_len)
1241 {
1242         struct btrfs_path *path;
1243         int ret = 0;
1244         struct extent_buffer *leaf;
1245         struct btrfs_dir_item *di;
1246         struct btrfs_key key;
1247         u64 index;
1248
1249         path = btrfs_alloc_path();
1250         if (!path) {
1251                 ret = -ENOMEM;
1252                 goto err;
1253         }
1254
1255         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
1256                                     name, name_len, -1);
1257         if (IS_ERR(di)) {
1258                 ret = PTR_ERR(di);
1259                 goto err;
1260         }
1261         if (!di) {
1262                 ret = -ENOENT;
1263                 goto err;
1264         }
1265         leaf = path->nodes[0];
1266         btrfs_dir_item_key_to_cpu(leaf, di, &key);
1267         ret = btrfs_delete_one_dir_name(trans, root, path, di);
1268         if (ret)
1269                 goto err;
1270         btrfs_release_path(root, path);
1271
1272         ret = btrfs_del_inode_ref(trans, root, name, name_len,
1273                                   inode->i_ino,
1274                                   dir->i_ino, &index);
1275         if (ret) {
1276                 printk("failed to delete reference to %.*s, "
1277                        "inode %lu parent %lu\n", name_len, name,
1278                        inode->i_ino, dir->i_ino);
1279                 goto err;
1280         }
1281
1282         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
1283                                          index, name, name_len, -1);
1284         if (IS_ERR(di)) {
1285                 ret = PTR_ERR(di);
1286                 goto err;
1287         }
1288         if (!di) {
1289                 ret = -ENOENT;
1290                 goto err;
1291         }
1292         ret = btrfs_delete_one_dir_name(trans, root, path, di);
1293         btrfs_release_path(root, path);
1294
1295         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
1296                                          inode, dir->i_ino);
1297         BUG_ON(ret != 0 && ret != -ENOENT);
1298         if (ret != -ENOENT)
1299                 BTRFS_I(dir)->log_dirty_trans = trans->transid;
1300
1301         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
1302                                            dir, index);
1303         BUG_ON(ret);
1304 err:
1305         btrfs_free_path(path);
1306         if (ret)
1307                 goto out;
1308
1309         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
1310         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
1311         btrfs_update_inode(trans, root, dir);
1312         btrfs_drop_nlink(inode);
1313         ret = btrfs_update_inode(trans, root, inode);
1314         dir->i_sb->s_dirt = 1;
1315 out:
1316         return ret;
1317 }
1318
1319 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
1320 {
1321         struct btrfs_root *root;
1322         struct btrfs_trans_handle *trans;
1323         struct inode *inode = dentry->d_inode;
1324         int ret;
1325         unsigned long nr = 0;
1326
1327         root = BTRFS_I(dir)->root;
1328
1329         ret = btrfs_check_free_space(root, 1, 1);
1330         if (ret)
1331                 goto fail;
1332
1333         trans = btrfs_start_transaction(root, 1);
1334
1335         btrfs_set_trans_block_group(trans, dir);
1336         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
1337                                  dentry->d_name.name, dentry->d_name.len);
1338
1339         if (inode->i_nlink == 0)
1340                 ret = btrfs_orphan_add(trans, inode);
1341
1342         nr = trans->blocks_used;
1343
1344         btrfs_end_transaction_throttle(trans, root);
1345 fail:
1346         btrfs_btree_balance_dirty(root, nr);
1347         return ret;
1348 }
1349
1350 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
1351 {
1352         struct inode *inode = dentry->d_inode;
1353         int err = 0;
1354         int ret;
1355         struct btrfs_root *root = BTRFS_I(dir)->root;
1356         struct btrfs_trans_handle *trans;
1357         unsigned long nr = 0;
1358
1359         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
1360                 return -ENOTEMPTY;
1361         }
1362
1363         ret = btrfs_check_free_space(root, 1, 1);
1364         if (ret)
1365                 goto fail;
1366
1367         trans = btrfs_start_transaction(root, 1);
1368         btrfs_set_trans_block_group(trans, dir);
1369
1370         err = btrfs_orphan_add(trans, inode);
1371         if (err)
1372                 goto fail_trans;
1373
1374         /* now the directory is empty */
1375         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
1376                                  dentry->d_name.name, dentry->d_name.len);
1377         if (!err) {
1378                 btrfs_i_size_write(inode, 0);
1379         }
1380
1381 fail_trans:
1382         nr = trans->blocks_used;
1383         ret = btrfs_end_transaction_throttle(trans, root);
1384 fail:
1385         btrfs_btree_balance_dirty(root, nr);
1386
1387         if (ret && !err)
1388                 err = ret;
1389         return err;
1390 }
1391
1392 /*
1393  * this can truncate away extent items, csum items and directory items.
1394  * It starts at a high offset and removes keys until it can't find
1395  * any higher than new_size
1396  *
1397  * csum items that cross the new i_size are truncated to the new size
1398  * as well.
1399  *
1400  * min_type is the minimum key type to truncate down to.  If set to 0, this
1401  * will kill all the items on this inode, including the INODE_ITEM_KEY.
1402  */
1403 noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
1404                                         struct btrfs_root *root,
1405                                         struct inode *inode,
1406                                         u64 new_size, u32 min_type)
1407 {
1408         int ret;
1409         struct btrfs_path *path;
1410         struct btrfs_key key;
1411         struct btrfs_key found_key;
1412         u32 found_type;
1413         struct extent_buffer *leaf;
1414         struct btrfs_file_extent_item *fi;
1415         u64 extent_start = 0;
1416         u64 extent_num_bytes = 0;
1417         u64 item_end = 0;
1418         u64 root_gen = 0;
1419         u64 root_owner = 0;
1420         int found_extent;
1421         int del_item;
1422         int pending_del_nr = 0;
1423         int pending_del_slot = 0;
1424         int extent_type = -1;
1425         u64 mask = root->sectorsize - 1;
1426
1427         if (root->ref_cows)
1428                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
1429         path = btrfs_alloc_path();
1430         path->reada = -1;
1431         BUG_ON(!path);
1432
1433         /* FIXME, add redo link to tree so we don't leak on crash */
1434         key.objectid = inode->i_ino;
1435         key.offset = (u64)-1;
1436         key.type = (u8)-1;
1437
1438         btrfs_init_path(path);
1439 search_again:
1440         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1441         if (ret < 0) {
1442                 goto error;
1443         }
1444         if (ret > 0) {
1445                 /* there are no items in the tree for us to truncate, we're
1446                  * done
1447                  */
1448                 if (path->slots[0] == 0) {
1449                         ret = 0;
1450                         goto error;
1451                 }
1452                 path->slots[0]--;
1453         }
1454
1455         while(1) {
1456                 fi = NULL;
1457                 leaf = path->nodes[0];
1458                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1459                 found_type = btrfs_key_type(&found_key);
1460
1461                 if (found_key.objectid != inode->i_ino)
1462                         break;
1463
1464                 if (found_type < min_type)
1465                         break;
1466
1467                 item_end = found_key.offset;
1468                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
1469                         fi = btrfs_item_ptr(leaf, path->slots[0],
1470                                             struct btrfs_file_extent_item);
1471                         extent_type = btrfs_file_extent_type(leaf, fi);
1472                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
1473                                 item_end +=
1474                                     btrfs_file_extent_num_bytes(leaf, fi);
1475                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1476                                 struct btrfs_item *item = btrfs_item_nr(leaf,
1477                                                                 path->slots[0]);
1478                                 item_end += btrfs_file_extent_inline_len(leaf,
1479                                                                          item);
1480                         }
1481                         item_end--;
1482                 }
1483                 if (found_type == BTRFS_CSUM_ITEM_KEY) {
1484                         ret = btrfs_csum_truncate(trans, root, path,
1485                                                   new_size);
1486                         BUG_ON(ret);
1487                 }
1488                 if (item_end < new_size) {
1489                         if (found_type == BTRFS_DIR_ITEM_KEY) {
1490                                 found_type = BTRFS_INODE_ITEM_KEY;
1491                         } else if (found_type == BTRFS_EXTENT_ITEM_KEY) {
1492                                 found_type = BTRFS_CSUM_ITEM_KEY;
1493                         } else if (found_type == BTRFS_EXTENT_DATA_KEY) {
1494                                 found_type = BTRFS_XATTR_ITEM_KEY;
1495                         } else if (found_type == BTRFS_XATTR_ITEM_KEY) {
1496                                 found_type = BTRFS_INODE_REF_KEY;
1497                         } else if (found_type) {
1498                                 found_type--;
1499                         } else {
1500                                 break;
1501                         }
1502                         btrfs_set_key_type(&key, found_type);
1503                         goto next;
1504                 }
1505                 if (found_key.offset >= new_size)
1506                         del_item = 1;
1507                 else
1508                         del_item = 0;
1509                 found_extent = 0;
1510
1511                 /* FIXME, shrink the extent if the ref count is only 1 */
1512                 if (found_type != BTRFS_EXTENT_DATA_KEY)
1513                         goto delete;
1514
1515                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
1516                         u64 num_dec;
1517                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
1518                         if (!del_item) {
1519                                 u64 orig_num_bytes =
1520                                         btrfs_file_extent_num_bytes(leaf, fi);
1521                                 extent_num_bytes = new_size -
1522                                         found_key.offset + root->sectorsize - 1;
1523                                 extent_num_bytes = extent_num_bytes &
1524                                         ~((u64)root->sectorsize - 1);
1525                                 btrfs_set_file_extent_num_bytes(leaf, fi,
1526                                                          extent_num_bytes);
1527                                 num_dec = (orig_num_bytes -
1528                                            extent_num_bytes);
1529                                 if (root->ref_cows && extent_start != 0)
1530                                         dec_i_blocks(inode, num_dec);
1531                                 btrfs_mark_buffer_dirty(leaf);
1532                         } else {
1533                                 extent_num_bytes =
1534                                         btrfs_file_extent_disk_num_bytes(leaf,
1535                                                                          fi);
1536                                 /* FIXME blocksize != 4096 */
1537                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
1538                                 if (extent_start != 0) {
1539                                         found_extent = 1;
1540                                         if (root->ref_cows)
1541                                                 dec_i_blocks(inode, num_dec);
1542                                 }
1543                                 root_gen = btrfs_header_generation(leaf);
1544                                 root_owner = btrfs_header_owner(leaf);
1545                         }
1546                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1547                         if (!del_item) {
1548                                 u32 size = new_size - found_key.offset;
1549
1550                                 if (root->ref_cows) {
1551                                         dec_i_blocks(inode, item_end + 1 -
1552                                                     found_key.offset - size);
1553                                 }
1554                                 size =
1555                                     btrfs_file_extent_calc_inline_size(size);
1556                                 ret = btrfs_truncate_item(trans, root, path,
1557                                                           size, 1);
1558                                 BUG_ON(ret);
1559                         } else if (root->ref_cows) {
1560                                 dec_i_blocks(inode, item_end + 1 -
1561                                              found_key.offset);
1562                         }
1563                 }
1564 delete:
1565                 if (del_item) {
1566                         if (!pending_del_nr) {
1567                                 /* no pending yet, add ourselves */
1568                                 pending_del_slot = path->slots[0];
1569                                 pending_del_nr = 1;
1570                         } else if (pending_del_nr &&
1571                                    path->slots[0] + 1 == pending_del_slot) {
1572                                 /* hop on the pending chunk */
1573                                 pending_del_nr++;
1574                                 pending_del_slot = path->slots[0];
1575                         } else {
1576                                 printk("bad pending slot %d pending_del_nr %d pending_del_slot %d\n", path->slots[0], pending_del_nr, pending_del_slot);
1577                         }
1578                 } else {
1579                         break;
1580                 }
1581                 if (found_extent) {
1582                         ret = btrfs_free_extent(trans, root, extent_start,
1583                                                 extent_num_bytes,
1584                                                 leaf->start, root_owner,
1585                                                 root_gen, inode->i_ino,
1586                                                 found_key.offset, 0);
1587                         BUG_ON(ret);
1588                 }
1589 next:
1590                 if (path->slots[0] == 0) {
1591                         if (pending_del_nr)
1592                                 goto del_pending;
1593                         btrfs_release_path(root, path);
1594                         goto search_again;
1595                 }
1596
1597                 path->slots[0]--;
1598                 if (pending_del_nr &&
1599                     path->slots[0] + 1 != pending_del_slot) {
1600                         struct btrfs_key debug;
1601 del_pending:
1602                         btrfs_item_key_to_cpu(path->nodes[0], &debug,
1603                                               pending_del_slot);
1604                         ret = btrfs_del_items(trans, root, path,
1605                                               pending_del_slot,
1606                                               pending_del_nr);
1607                         BUG_ON(ret);
1608                         pending_del_nr = 0;
1609                         btrfs_release_path(root, path);
1610                         goto search_again;
1611                 }
1612         }
1613         ret = 0;
1614 error:
1615         if (pending_del_nr) {
1616                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
1617                                       pending_del_nr);
1618         }
1619         btrfs_free_path(path);
1620         inode->i_sb->s_dirt = 1;
1621         return ret;
1622 }
1623
1624 /*
1625  * taken from block_truncate_page, but does cow as it zeros out
1626  * any bytes left in the last page in the file.
1627  */
1628 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
1629 {
1630         struct inode *inode = mapping->host;
1631         struct btrfs_root *root = BTRFS_I(inode)->root;
1632         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1633         struct btrfs_ordered_extent *ordered;
1634         char *kaddr;
1635         u32 blocksize = root->sectorsize;
1636         pgoff_t index = from >> PAGE_CACHE_SHIFT;
1637         unsigned offset = from & (PAGE_CACHE_SIZE-1);
1638         struct page *page;
1639         int ret = 0;
1640         u64 page_start;
1641         u64 page_end;
1642
1643         if ((offset & (blocksize - 1)) == 0)
1644                 goto out;
1645
1646         ret = -ENOMEM;
1647 again:
1648         page = grab_cache_page(mapping, index);
1649         if (!page)
1650                 goto out;
1651
1652         page_start = page_offset(page);
1653         page_end = page_start + PAGE_CACHE_SIZE - 1;
1654
1655         if (!PageUptodate(page)) {
1656                 ret = btrfs_readpage(NULL, page);
1657                 lock_page(page);
1658                 if (page->mapping != mapping) {
1659                         unlock_page(page);
1660                         page_cache_release(page);
1661                         goto again;
1662                 }
1663                 if (!PageUptodate(page)) {
1664                         ret = -EIO;
1665                         goto out_unlock;
1666                 }
1667         }
1668         wait_on_page_writeback(page);
1669
1670         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
1671         set_page_extent_mapped(page);
1672
1673         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1674         if (ordered) {
1675                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
1676                 unlock_page(page);
1677                 page_cache_release(page);
1678                 btrfs_start_ordered_extent(inode, ordered, 1);
1679                 btrfs_put_ordered_extent(ordered);
1680                 goto again;
1681         }
1682
1683         btrfs_set_extent_delalloc(inode, page_start, page_end);
1684         ret = 0;
1685         if (offset != PAGE_CACHE_SIZE) {
1686                 kaddr = kmap(page);
1687                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
1688                 flush_dcache_page(page);
1689                 kunmap(page);
1690         }
1691         ClearPageChecked(page);
1692         set_page_dirty(page);
1693         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
1694
1695 out_unlock:
1696         unlock_page(page);
1697         page_cache_release(page);
1698 out:
1699         return ret;
1700 }
1701
1702 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
1703 {
1704         struct inode *inode = dentry->d_inode;
1705         int err;
1706
1707         err = inode_change_ok(inode, attr);
1708         if (err)
1709                 return err;
1710
1711         if (S_ISREG(inode->i_mode) &&
1712             attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
1713                 struct btrfs_trans_handle *trans;
1714                 struct btrfs_root *root = BTRFS_I(inode)->root;
1715                 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1716
1717                 u64 mask = root->sectorsize - 1;
1718                 u64 hole_start = (inode->i_size + mask) & ~mask;
1719                 u64 block_end = (attr->ia_size + mask) & ~mask;
1720                 u64 hole_size;
1721                 u64 alloc_hint = 0;
1722
1723                 if (attr->ia_size <= hole_start)
1724                         goto out;
1725
1726                 err = btrfs_check_free_space(root, 1, 0);
1727                 if (err)
1728                         goto fail;
1729
1730                 btrfs_truncate_page(inode->i_mapping, inode->i_size);
1731
1732                 hole_size = block_end - hole_start;
1733                 while(1) {
1734                         struct btrfs_ordered_extent *ordered;
1735                         btrfs_wait_ordered_range(inode, hole_start, hole_size);
1736
1737                         lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
1738                         ordered = btrfs_lookup_ordered_extent(inode, hole_start);
1739                         if (ordered) {
1740                                 unlock_extent(io_tree, hole_start,
1741                                               block_end - 1, GFP_NOFS);
1742                                 btrfs_put_ordered_extent(ordered);
1743                         } else {
1744                                 break;
1745                         }
1746                 }
1747
1748                 trans = btrfs_start_transaction(root, 1);
1749                 btrfs_set_trans_block_group(trans, inode);
1750                 mutex_lock(&BTRFS_I(inode)->extent_mutex);
1751                 err = btrfs_drop_extents(trans, root, inode,
1752                                          hole_start, block_end, hole_start,
1753                                          &alloc_hint);
1754
1755                 if (alloc_hint != EXTENT_MAP_INLINE) {
1756                         err = btrfs_insert_file_extent(trans, root,
1757                                                        inode->i_ino,
1758                                                        hole_start, 0, 0,
1759                                                        hole_size, 0);
1760                         btrfs_drop_extent_cache(inode, hole_start,
1761                                                 (u64)-1, 0);
1762                         btrfs_check_file(root, inode);
1763                 }
1764                 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
1765                 btrfs_end_transaction(trans, root);
1766                 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
1767                 if (err)
1768                         return err;
1769         }
1770 out:
1771         err = inode_setattr(inode, attr);
1772
1773         if (!err && ((attr->ia_valid & ATTR_MODE)))
1774                 err = btrfs_acl_chmod(inode);
1775 fail:
1776         return err;
1777 }
1778
1779 void btrfs_delete_inode(struct inode *inode)
1780 {
1781         struct btrfs_trans_handle *trans;
1782         struct btrfs_root *root = BTRFS_I(inode)->root;
1783         unsigned long nr;
1784         int ret;
1785
1786         truncate_inode_pages(&inode->i_data, 0);
1787         if (is_bad_inode(inode)) {
1788                 btrfs_orphan_del(NULL, inode);
1789                 goto no_delete;
1790         }
1791         btrfs_wait_ordered_range(inode, 0, (u64)-1);
1792
1793         btrfs_i_size_write(inode, 0);
1794         trans = btrfs_start_transaction(root, 1);
1795
1796         btrfs_set_trans_block_group(trans, inode);
1797         ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0);
1798         if (ret) {
1799                 btrfs_orphan_del(NULL, inode);
1800                 goto no_delete_lock;
1801         }
1802
1803         btrfs_orphan_del(trans, inode);
1804
1805         nr = trans->blocks_used;
1806         clear_inode(inode);
1807
1808         btrfs_end_transaction(trans, root);
1809         btrfs_btree_balance_dirty(root, nr);
1810         return;
1811
1812 no_delete_lock:
1813         nr = trans->blocks_used;
1814         btrfs_end_transaction(trans, root);
1815         btrfs_btree_balance_dirty(root, nr);
1816 no_delete:
1817         clear_inode(inode);
1818 }
1819
1820 /*
1821  * this returns the key found in the dir entry in the location pointer.
1822  * If no dir entries were found, location->objectid is 0.
1823  */
1824 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
1825                                struct btrfs_key *location)
1826 {
1827         const char *name = dentry->d_name.name;
1828         int namelen = dentry->d_name.len;
1829         struct btrfs_dir_item *di;
1830         struct btrfs_path *path;
1831         struct btrfs_root *root = BTRFS_I(dir)->root;
1832         int ret = 0;
1833
1834         path = btrfs_alloc_path();
1835         BUG_ON(!path);
1836
1837         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
1838                                     namelen, 0);
1839         if (IS_ERR(di))
1840                 ret = PTR_ERR(di);
1841         if (!di || IS_ERR(di)) {
1842                 goto out_err;
1843         }
1844         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
1845 out:
1846         btrfs_free_path(path);
1847         return ret;
1848 out_err:
1849         location->objectid = 0;
1850         goto out;
1851 }
1852
1853 /*
1854  * when we hit a tree root in a directory, the btrfs part of the inode
1855  * needs to be changed to reflect the root directory of the tree root.  This
1856  * is kind of like crossing a mount point.
1857  */
1858 static int fixup_tree_root_location(struct btrfs_root *root,
1859                              struct btrfs_key *location,
1860                              struct btrfs_root **sub_root,
1861                              struct dentry *dentry)
1862 {
1863         struct btrfs_root_item *ri;
1864
1865         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
1866                 return 0;
1867         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1868                 return 0;
1869
1870         *sub_root = btrfs_read_fs_root(root->fs_info, location,
1871                                         dentry->d_name.name,
1872                                         dentry->d_name.len);
1873         if (IS_ERR(*sub_root))
1874                 return PTR_ERR(*sub_root);
1875
1876         ri = &(*sub_root)->root_item;
1877         location->objectid = btrfs_root_dirid(ri);
1878         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1879         location->offset = 0;
1880
1881         return 0;
1882 }
1883
1884 static noinline void init_btrfs_i(struct inode *inode)
1885 {
1886         struct btrfs_inode *bi = BTRFS_I(inode);
1887
1888         bi->i_acl = NULL;
1889         bi->i_default_acl = NULL;
1890
1891         bi->generation = 0;
1892         bi->last_trans = 0;
1893         bi->logged_trans = 0;
1894         bi->delalloc_bytes = 0;
1895         bi->disk_i_size = 0;
1896         bi->flags = 0;
1897         bi->index_cnt = (u64)-1;
1898         bi->log_dirty_trans = 0;
1899         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
1900         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
1901                              inode->i_mapping, GFP_NOFS);
1902         extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
1903                              inode->i_mapping, GFP_NOFS);
1904         INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
1905         btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
1906         mutex_init(&BTRFS_I(inode)->csum_mutex);
1907         mutex_init(&BTRFS_I(inode)->extent_mutex);
1908         mutex_init(&BTRFS_I(inode)->log_mutex);
1909 }
1910
1911 static int btrfs_init_locked_inode(struct inode *inode, void *p)
1912 {
1913         struct btrfs_iget_args *args = p;
1914         inode->i_ino = args->ino;
1915         init_btrfs_i(inode);
1916         BTRFS_I(inode)->root = args->root;
1917         return 0;
1918 }
1919
1920 static int btrfs_find_actor(struct inode *inode, void *opaque)
1921 {
1922         struct btrfs_iget_args *args = opaque;
1923         return (args->ino == inode->i_ino &&
1924                 args->root == BTRFS_I(inode)->root);
1925 }
1926
1927 struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
1928                             struct btrfs_root *root, int wait)
1929 {
1930         struct inode *inode;
1931         struct btrfs_iget_args args;
1932         args.ino = objectid;
1933         args.root = root;
1934
1935         if (wait) {
1936                 inode = ilookup5(s, objectid, btrfs_find_actor,
1937                                  (void *)&args);
1938         } else {
1939                 inode = ilookup5_nowait(s, objectid, btrfs_find_actor,
1940                                         (void *)&args);
1941         }
1942         return inode;
1943 }
1944
1945 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
1946                                 struct btrfs_root *root)
1947 {
1948         struct inode *inode;
1949         struct btrfs_iget_args args;
1950         args.ino = objectid;
1951         args.root = root;
1952
1953         inode = iget5_locked(s, objectid, btrfs_find_actor,
1954                              btrfs_init_locked_inode,
1955                              (void *)&args);
1956         return inode;
1957 }
1958
1959 /* Get an inode object given its location and corresponding root.
1960  * Returns in *is_new if the inode was read from disk
1961  */
1962 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
1963                          struct btrfs_root *root, int *is_new)
1964 {
1965         struct inode *inode;
1966
1967         inode = btrfs_iget_locked(s, location->objectid, root);
1968         if (!inode)
1969                 return ERR_PTR(-EACCES);
1970
1971         if (inode->i_state & I_NEW) {
1972                 BTRFS_I(inode)->root = root;
1973                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
1974                 btrfs_read_locked_inode(inode);
1975                 unlock_new_inode(inode);
1976                 if (is_new)
1977                         *is_new = 1;
1978         } else {
1979                 if (is_new)
1980                         *is_new = 0;
1981         }
1982
1983         return inode;
1984 }
1985
1986 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
1987                                    struct nameidata *nd)
1988 {
1989         struct inode * inode;
1990         struct btrfs_inode *bi = BTRFS_I(dir);
1991         struct btrfs_root *root = bi->root;
1992         struct btrfs_root *sub_root = root;
1993         struct btrfs_key location;
1994         int ret, new, do_orphan = 0;
1995
1996         if (dentry->d_name.len > BTRFS_NAME_LEN)
1997                 return ERR_PTR(-ENAMETOOLONG);
1998
1999         ret = btrfs_inode_by_name(dir, dentry, &location);
2000
2001         if (ret < 0)
2002                 return ERR_PTR(ret);
2003
2004         inode = NULL;
2005         if (location.objectid) {
2006                 ret = fixup_tree_root_location(root, &location, &sub_root,
2007                                                 dentry);
2008                 if (ret < 0)
2009                         return ERR_PTR(ret);
2010                 if (ret > 0)
2011                         return ERR_PTR(-ENOENT);
2012                 inode = btrfs_iget(dir->i_sb, &location, sub_root, &new);
2013                 if (IS_ERR(inode))
2014                         return ERR_CAST(inode);
2015
2016                 /* the inode and parent dir are two different roots */
2017                 if (new && root != sub_root) {
2018                         igrab(inode);
2019                         sub_root->inode = inode;
2020                         do_orphan = 1;
2021                 }
2022         }
2023
2024         if (unlikely(do_orphan))
2025                 btrfs_orphan_cleanup(sub_root);
2026
2027         return d_splice_alias(inode, dentry);
2028 }
2029
2030 static unsigned char btrfs_filetype_table[] = {
2031         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
2032 };
2033
2034 static int btrfs_real_readdir(struct file *filp, void *dirent,
2035                               filldir_t filldir)
2036 {
2037         struct inode *inode = filp->f_dentry->d_inode;
2038         struct btrfs_root *root = BTRFS_I(inode)->root;
2039         struct btrfs_item *item;
2040         struct btrfs_dir_item *di;
2041         struct btrfs_key key;
2042         struct btrfs_key found_key;
2043         struct btrfs_path *path;
2044         int ret;
2045         u32 nritems;
2046         struct extent_buffer *leaf;
2047         int slot;
2048         int advance;
2049         unsigned char d_type;
2050         int over = 0;
2051         u32 di_cur;
2052         u32 di_total;
2053         u32 di_len;
2054         int key_type = BTRFS_DIR_INDEX_KEY;
2055         char tmp_name[32];
2056         char *name_ptr;
2057         int name_len;
2058
2059         /* FIXME, use a real flag for deciding about the key type */
2060         if (root->fs_info->tree_root == root)
2061                 key_type = BTRFS_DIR_ITEM_KEY;
2062
2063         /* special case for "." */
2064         if (filp->f_pos == 0) {
2065                 over = filldir(dirent, ".", 1,
2066                                1, inode->i_ino,
2067                                DT_DIR);
2068                 if (over)
2069                         return 0;
2070                 filp->f_pos = 1;
2071         }
2072         /* special case for .., just use the back ref */
2073         if (filp->f_pos == 1) {
2074                 u64 pino = parent_ino(filp->f_path.dentry);
2075                 over = filldir(dirent, "..", 2,
2076                                2, pino, DT_DIR);
2077                 if (over)
2078                         return 0;
2079                 filp->f_pos = 2;
2080         }
2081
2082         path = btrfs_alloc_path();
2083         path->reada = 2;
2084
2085         btrfs_set_key_type(&key, key_type);
2086         key.offset = filp->f_pos;
2087         key.objectid = inode->i_ino;
2088
2089         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2090         if (ret < 0)
2091                 goto err;
2092         advance = 0;
2093
2094         while (1) {
2095                 leaf = path->nodes[0];
2096                 nritems = btrfs_header_nritems(leaf);
2097                 slot = path->slots[0];
2098                 if (advance || slot >= nritems) {
2099                         if (slot >= nritems - 1) {
2100                                 ret = btrfs_next_leaf(root, path);
2101                                 if (ret)
2102                                         break;
2103                                 leaf = path->nodes[0];
2104                                 nritems = btrfs_header_nritems(leaf);
2105                                 slot = path->slots[0];
2106                         } else {
2107                                 slot++;
2108                                 path->slots[0]++;
2109                         }
2110                 }
2111                 advance = 1;
2112                 item = btrfs_item_nr(leaf, slot);
2113                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2114
2115                 if (found_key.objectid != key.objectid)
2116                         break;
2117                 if (btrfs_key_type(&found_key) != key_type)
2118                         break;
2119                 if (found_key.offset < filp->f_pos)
2120                         continue;
2121
2122                 filp->f_pos = found_key.offset;
2123
2124                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
2125                 di_cur = 0;
2126                 di_total = btrfs_item_size(leaf, item);
2127
2128                 while (di_cur < di_total) {
2129                         struct btrfs_key location;
2130
2131                         name_len = btrfs_dir_name_len(leaf, di);
2132                         if (name_len <= sizeof(tmp_name)) {
2133                                 name_ptr = tmp_name;
2134                         } else {
2135                                 name_ptr = kmalloc(name_len, GFP_NOFS);
2136                                 if (!name_ptr) {
2137                                         ret = -ENOMEM;
2138                                         goto err;
2139                                 }
2140                         }
2141                         read_extent_buffer(leaf, name_ptr,
2142                                            (unsigned long)(di + 1), name_len);
2143
2144                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
2145                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
2146                         over = filldir(dirent, name_ptr, name_len,
2147                                        found_key.offset, location.objectid,
2148                                        d_type);
2149
2150                         if (name_ptr != tmp_name)
2151                                 kfree(name_ptr);
2152
2153                         if (over)
2154                                 goto nopos;
2155
2156                         di_len = btrfs_dir_name_len(leaf, di) +
2157                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
2158                         di_cur += di_len;
2159                         di = (struct btrfs_dir_item *)((char *)di + di_len);
2160                 }
2161         }
2162
2163         /* Reached end of directory/root. Bump pos past the last item. */
2164         if (key_type == BTRFS_DIR_INDEX_KEY)
2165                 filp->f_pos = INT_LIMIT(typeof(filp->f_pos));
2166         else
2167                 filp->f_pos++;
2168 nopos:
2169         ret = 0;
2170 err:
2171         btrfs_free_path(path);
2172         return ret;
2173 }
2174
2175 int btrfs_write_inode(struct inode *inode, int wait)
2176 {
2177         struct btrfs_root *root = BTRFS_I(inode)->root;
2178         struct btrfs_trans_handle *trans;
2179         int ret = 0;
2180
2181         if (root->fs_info->closing > 1)
2182                 return 0;
2183
2184         if (wait) {
2185                 trans = btrfs_join_transaction(root, 1);
2186                 btrfs_set_trans_block_group(trans, inode);
2187                 ret = btrfs_commit_transaction(trans, root);
2188         }
2189         return ret;
2190 }
2191
2192 /*
2193  * This is somewhat expensive, updating the tree every time the
2194  * inode changes.  But, it is most likely to find the inode in cache.
2195  * FIXME, needs more benchmarking...there are no reasons other than performance
2196  * to keep or drop this code.
2197  */
2198 void btrfs_dirty_inode(struct inode *inode)
2199 {
2200         struct btrfs_root *root = BTRFS_I(inode)->root;
2201         struct btrfs_trans_handle *trans;
2202
2203         trans = btrfs_join_transaction(root, 1);
2204         btrfs_set_trans_block_group(trans, inode);
2205         btrfs_update_inode(trans, root, inode);
2206         btrfs_end_transaction(trans, root);
2207 }
2208
2209 /*
2210  * find the highest existing sequence number in a directory
2211  * and then set the in-memory index_cnt variable to reflect
2212  * free sequence numbers
2213  */
2214 static int btrfs_set_inode_index_count(struct inode *inode)
2215 {
2216         struct btrfs_root *root = BTRFS_I(inode)->root;
2217         struct btrfs_key key, found_key;
2218         struct btrfs_path *path;
2219         struct extent_buffer *leaf;
2220         int ret;
2221
2222         key.objectid = inode->i_ino;
2223         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
2224         key.offset = (u64)-1;
2225
2226         path = btrfs_alloc_path();
2227         if (!path)
2228                 return -ENOMEM;
2229
2230         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2231         if (ret < 0)
2232                 goto out;
2233         /* FIXME: we should be able to handle this */
2234         if (ret == 0)
2235                 goto out;
2236         ret = 0;
2237
2238         /*
2239          * MAGIC NUMBER EXPLANATION:
2240          * since we search a directory based on f_pos we have to start at 2
2241          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
2242          * else has to start at 2
2243          */
2244         if (path->slots[0] == 0) {
2245                 BTRFS_I(inode)->index_cnt = 2;
2246                 goto out;
2247         }
2248
2249         path->slots[0]--;
2250
2251         leaf = path->nodes[0];
2252         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2253
2254         if (found_key.objectid != inode->i_ino ||
2255             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
2256                 BTRFS_I(inode)->index_cnt = 2;
2257                 goto out;
2258         }
2259
2260         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
2261 out:
2262         btrfs_free_path(path);
2263         return ret;
2264 }
2265
2266 /*
2267  * helper to find a free sequence number in a given directory.  This current
2268  * code is very simple, later versions will do smarter things in the btree
2269  */
2270 static int btrfs_set_inode_index(struct inode *dir, struct inode *inode,
2271                                  u64 *index)
2272 {
2273         int ret = 0;
2274
2275         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
2276                 ret = btrfs_set_inode_index_count(dir);
2277                 if (ret) {
2278                         return ret;
2279                 }
2280         }
2281
2282         *index = BTRFS_I(dir)->index_cnt;
2283         BTRFS_I(dir)->index_cnt++;
2284
2285         return ret;
2286 }
2287
2288 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
2289                                      struct btrfs_root *root,
2290                                      struct inode *dir,
2291                                      const char *name, int name_len,
2292                                      u64 ref_objectid,
2293                                      u64 objectid,
2294                                      struct btrfs_block_group_cache *group,
2295                                      int mode, u64 *index)
2296 {
2297         struct inode *inode;
2298         struct btrfs_inode_item *inode_item;
2299         struct btrfs_block_group_cache *new_inode_group;
2300         struct btrfs_key *location;
2301         struct btrfs_path *path;
2302         struct btrfs_inode_ref *ref;
2303         struct btrfs_key key[2];
2304         u32 sizes[2];
2305         unsigned long ptr;
2306         int ret;
2307         int owner;
2308
2309         path = btrfs_alloc_path();
2310         BUG_ON(!path);
2311
2312         inode = new_inode(root->fs_info->sb);
2313         if (!inode)
2314                 return ERR_PTR(-ENOMEM);
2315
2316         if (dir) {
2317                 ret = btrfs_set_inode_index(dir, inode, index);
2318                 if (ret)
2319                         return ERR_PTR(ret);
2320         }
2321         /*
2322          * index_cnt is ignored for everything but a dir,
2323          * btrfs_get_inode_index_count has an explanation for the magic
2324          * number
2325          */
2326         init_btrfs_i(inode);
2327         BTRFS_I(inode)->index_cnt = 2;
2328         BTRFS_I(inode)->root = root;
2329         BTRFS_I(inode)->generation = trans->transid;
2330
2331         if (mode & S_IFDIR)
2332                 owner = 0;
2333         else
2334                 owner = 1;
2335         new_inode_group = btrfs_find_block_group(root, group, 0,
2336                                        BTRFS_BLOCK_GROUP_METADATA, owner);
2337         if (!new_inode_group) {
2338                 printk("find_block group failed\n");
2339                 new_inode_group = group;
2340         }
2341         BTRFS_I(inode)->block_group = new_inode_group;
2342
2343         key[0].objectid = objectid;
2344         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
2345         key[0].offset = 0;
2346
2347         key[1].objectid = objectid;
2348         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
2349         key[1].offset = ref_objectid;
2350
2351         sizes[0] = sizeof(struct btrfs_inode_item);
2352         sizes[1] = name_len + sizeof(*ref);
2353
2354         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
2355         if (ret != 0)
2356                 goto fail;
2357
2358         if (objectid > root->highest_inode)
2359                 root->highest_inode = objectid;
2360
2361         inode->i_uid = current->fsuid;
2362         inode->i_gid = current->fsgid;
2363         inode->i_mode = mode;
2364         inode->i_ino = objectid;
2365         inode->i_blocks = 0;
2366         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2367         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2368                                   struct btrfs_inode_item);
2369         fill_inode_item(trans, path->nodes[0], inode_item, inode);
2370
2371         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
2372                              struct btrfs_inode_ref);
2373         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
2374         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
2375         ptr = (unsigned long)(ref + 1);
2376         write_extent_buffer(path->nodes[0], name, ptr, name_len);
2377
2378         btrfs_mark_buffer_dirty(path->nodes[0]);
2379         btrfs_free_path(path);
2380
2381         location = &BTRFS_I(inode)->location;
2382         location->objectid = objectid;
2383         location->offset = 0;
2384         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
2385
2386         insert_inode_hash(inode);
2387         return inode;
2388 fail:
2389         if (dir)
2390                 BTRFS_I(dir)->index_cnt--;
2391         btrfs_free_path(path);
2392         return ERR_PTR(ret);
2393 }
2394
2395 static inline u8 btrfs_inode_type(struct inode *inode)
2396 {
2397         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
2398 }
2399
2400 /*
2401  * utility function to add 'inode' into 'parent_inode' with
2402  * a give name and a given sequence number.
2403  * if 'add_backref' is true, also insert a backref from the
2404  * inode to the parent directory.
2405  */
2406 int btrfs_add_link(struct btrfs_trans_handle *trans,
2407                    struct inode *parent_inode, struct inode *inode,
2408                    const char *name, int name_len, int add_backref, u64 index)
2409 {
2410         int ret;
2411         struct btrfs_key key;
2412         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
2413
2414         key.objectid = inode->i_ino;
2415         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
2416         key.offset = 0;
2417
2418         ret = btrfs_insert_dir_item(trans, root, name, name_len,
2419                                     parent_inode->i_ino,
2420                                     &key, btrfs_inode_type(inode),
2421                                     index);
2422         if (ret == 0) {
2423                 if (add_backref) {
2424                         ret = btrfs_insert_inode_ref(trans, root,
2425                                                      name, name_len,
2426                                                      inode->i_ino,
2427                                                      parent_inode->i_ino,
2428                                                      index);
2429                 }
2430                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
2431                                    name_len * 2);
2432                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
2433                 ret = btrfs_update_inode(trans, root, parent_inode);
2434         }
2435         return ret;
2436 }
2437
2438 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
2439                             struct dentry *dentry, struct inode *inode,
2440                             int backref, u64 index)
2441 {
2442         int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
2443                                  inode, dentry->d_name.name,
2444                                  dentry->d_name.len, backref, index);
2445         if (!err) {
2446                 d_instantiate(dentry, inode);
2447                 return 0;
2448         }
2449         if (err > 0)
2450                 err = -EEXIST;
2451         return err;
2452 }
2453
2454 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
2455                         int mode, dev_t rdev)
2456 {
2457         struct btrfs_trans_handle *trans;
2458         struct btrfs_root *root = BTRFS_I(dir)->root;
2459         struct inode *inode = NULL;
2460         int err;
2461         int drop_inode = 0;
2462         u64 objectid;
2463         unsigned long nr = 0;
2464         u64 index = 0;
2465
2466         if (!new_valid_dev(rdev))
2467                 return -EINVAL;
2468
2469         err = btrfs_check_free_space(root, 1, 0);
2470         if (err)
2471                 goto fail;
2472
2473         trans = btrfs_start_transaction(root, 1);
2474         btrfs_set_trans_block_group(trans, dir);
2475
2476         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2477         if (err) {
2478                 err = -ENOSPC;
2479                 goto out_unlock;
2480         }
2481
2482         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
2483                                 dentry->d_name.len,
2484                                 dentry->d_parent->d_inode->i_ino, objectid,
2485                                 BTRFS_I(dir)->block_group, mode, &index);
2486         err = PTR_ERR(inode);
2487         if (IS_ERR(inode))
2488                 goto out_unlock;
2489
2490         err = btrfs_init_acl(inode, dir);
2491         if (err) {
2492                 drop_inode = 1;
2493                 goto out_unlock;
2494         }
2495
2496         btrfs_set_trans_block_group(trans, inode);
2497         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
2498         if (err)
2499                 drop_inode = 1;
2500         else {
2501                 inode->i_op = &btrfs_special_inode_operations;
2502                 init_special_inode(inode, inode->i_mode, rdev);
2503                 btrfs_update_inode(trans, root, inode);
2504         }
2505         dir->i_sb->s_dirt = 1;
2506         btrfs_update_inode_block_group(trans, inode);
2507         btrfs_update_inode_block_group(trans, dir);
2508 out_unlock:
2509         nr = trans->blocks_used;
2510         btrfs_end_transaction_throttle(trans, root);
2511 fail:
2512         if (drop_inode) {
2513                 inode_dec_link_count(inode);
2514                 iput(inode);
2515         }
2516         btrfs_btree_balance_dirty(root, nr);
2517         return err;
2518 }
2519
2520 static int btrfs_create(struct inode *dir, struct dentry *dentry,
2521                         int mode, struct nameidata *nd)
2522 {
2523         struct btrfs_trans_handle *trans;
2524         struct btrfs_root *root = BTRFS_I(dir)->root;
2525         struct inode *inode = NULL;
2526         int err;
2527         int drop_inode = 0;
2528         unsigned long nr = 0;
2529         u64 objectid;
2530         u64 index = 0;
2531
2532         err = btrfs_check_free_space(root, 1, 0);
2533         if (err)
2534                 goto fail;
2535         trans = btrfs_start_transaction(root, 1);
2536         btrfs_set_trans_block_group(trans, dir);
2537
2538         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2539         if (err) {
2540                 err = -ENOSPC;
2541                 goto out_unlock;
2542         }
2543
2544         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
2545                                 dentry->d_name.len,
2546                                 dentry->d_parent->d_inode->i_ino,
2547                                 objectid, BTRFS_I(dir)->block_group, mode,
2548                                 &index);
2549         err = PTR_ERR(inode);
2550         if (IS_ERR(inode))
2551                 goto out_unlock;
2552
2553         err = btrfs_init_acl(inode, dir);
2554         if (err) {
2555                 drop_inode = 1;
2556                 goto out_unlock;
2557         }
2558
2559         btrfs_set_trans_block_group(trans, inode);
2560         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
2561         if (err)
2562                 drop_inode = 1;
2563         else {
2564                 inode->i_mapping->a_ops = &btrfs_aops;
2565                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2566                 inode->i_fop = &btrfs_file_operations;
2567                 inode->i_op = &btrfs_file_inode_operations;
2568                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2569         }
2570         dir->i_sb->s_dirt = 1;
2571         btrfs_update_inode_block_group(trans, inode);
2572         btrfs_update_inode_block_group(trans, dir);
2573 out_unlock:
2574         nr = trans->blocks_used;
2575         btrfs_end_transaction_throttle(trans, root);
2576 fail:
2577         if (drop_inode) {
2578                 inode_dec_link_count(inode);
2579                 iput(inode);
2580         }
2581         btrfs_btree_balance_dirty(root, nr);
2582         return err;
2583 }
2584
2585 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
2586                       struct dentry *dentry)
2587 {
2588         struct btrfs_trans_handle *trans;
2589         struct btrfs_root *root = BTRFS_I(dir)->root;
2590         struct inode *inode = old_dentry->d_inode;
2591         u64 index;
2592         unsigned long nr = 0;
2593         int err;
2594         int drop_inode = 0;
2595
2596         if (inode->i_nlink == 0)
2597                 return -ENOENT;
2598
2599         btrfs_inc_nlink(inode);
2600         err = btrfs_check_free_space(root, 1, 0);
2601         if (err)
2602                 goto fail;
2603         err = btrfs_set_inode_index(dir, inode, &index);
2604         if (err)
2605                 goto fail;
2606
2607         trans = btrfs_start_transaction(root, 1);
2608
2609         btrfs_set_trans_block_group(trans, dir);
2610         atomic_inc(&inode->i_count);
2611
2612         err = btrfs_add_nondir(trans, dentry, inode, 1, index);
2613
2614         if (err)
2615                 drop_inode = 1;
2616
2617         dir->i_sb->s_dirt = 1;
2618         btrfs_update_inode_block_group(trans, dir);
2619         err = btrfs_update_inode(trans, root, inode);
2620
2621         if (err)
2622                 drop_inode = 1;
2623
2624         nr = trans->blocks_used;
2625         btrfs_end_transaction_throttle(trans, root);
2626 fail:
2627         if (drop_inode) {
2628                 inode_dec_link_count(inode);
2629                 iput(inode);
2630         }
2631         btrfs_btree_balance_dirty(root, nr);
2632         return err;
2633 }
2634
2635 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2636 {
2637         struct inode *inode = NULL;
2638         struct btrfs_trans_handle *trans;
2639         struct btrfs_root *root = BTRFS_I(dir)->root;
2640         int err = 0;
2641         int drop_on_err = 0;
2642         u64 objectid = 0;
2643         u64 index = 0;
2644         unsigned long nr = 1;
2645
2646         err = btrfs_check_free_space(root, 1, 0);
2647         if (err)
2648                 goto out_unlock;
2649
2650         trans = btrfs_start_transaction(root, 1);
2651         btrfs_set_trans_block_group(trans, dir);
2652
2653         if (IS_ERR(trans)) {
2654                 err = PTR_ERR(trans);
2655                 goto out_unlock;
2656         }
2657
2658         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2659         if (err) {
2660                 err = -ENOSPC;
2661                 goto out_unlock;
2662         }
2663
2664         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
2665                                 dentry->d_name.len,
2666                                 dentry->d_parent->d_inode->i_ino, objectid,
2667                                 BTRFS_I(dir)->block_group, S_IFDIR | mode,
2668                                 &index);
2669         if (IS_ERR(inode)) {
2670                 err = PTR_ERR(inode);
2671                 goto out_fail;
2672         }
2673
2674         drop_on_err = 1;
2675
2676         err = btrfs_init_acl(inode, dir);
2677         if (err)
2678                 goto out_fail;
2679
2680         inode->i_op = &btrfs_dir_inode_operations;
2681         inode->i_fop = &btrfs_dir_file_operations;
2682         btrfs_set_trans_block_group(trans, inode);
2683
2684         btrfs_i_size_write(inode, 0);
2685         err = btrfs_update_inode(trans, root, inode);
2686         if (err)
2687                 goto out_fail;
2688
2689         err = btrfs_add_link(trans, dentry->d_parent->d_inode,
2690                                  inode, dentry->d_name.name,
2691                                  dentry->d_name.len, 0, index);
2692         if (err)
2693                 goto out_fail;
2694
2695         d_instantiate(dentry, inode);
2696         drop_on_err = 0;
2697         dir->i_sb->s_dirt = 1;
2698         btrfs_update_inode_block_group(trans, inode);
2699         btrfs_update_inode_block_group(trans, dir);
2700
2701 out_fail:
2702         nr = trans->blocks_used;
2703         btrfs_end_transaction_throttle(trans, root);
2704
2705 out_unlock:
2706         if (drop_on_err)
2707                 iput(inode);
2708         btrfs_btree_balance_dirty(root, nr);
2709         return err;
2710 }
2711
2712 /* helper for btfs_get_extent.  Given an existing extent in the tree,
2713  * and an extent that you want to insert, deal with overlap and insert
2714  * the new extent into the tree.
2715  */
2716 static int merge_extent_mapping(struct extent_map_tree *em_tree,
2717                                 struct extent_map *existing,
2718                                 struct extent_map *em,
2719                                 u64 map_start, u64 map_len)
2720 {
2721         u64 start_diff;
2722
2723         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
2724         start_diff = map_start - em->start;
2725         em->start = map_start;
2726         em->len = map_len;
2727         if (em->block_start < EXTENT_MAP_LAST_BYTE)
2728                 em->block_start += start_diff;
2729         return add_extent_mapping(em_tree, em);
2730 }
2731
2732 /*
2733  * a bit scary, this does extent mapping from logical file offset to the disk.
2734  * the ugly parts come from merging extents from the disk with the
2735  * in-ram representation.  This gets more complex because of the data=ordered code,
2736  * where the in-ram extents might be locked pending data=ordered completion.
2737  *
2738  * This also copies inline extents directly into the page.
2739  */
2740 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
2741                                     size_t pg_offset, u64 start, u64 len,
2742                                     int create)
2743 {
2744         int ret;
2745         int err = 0;
2746         u64 bytenr;
2747         u64 extent_start = 0;
2748         u64 extent_end = 0;
2749         u64 objectid = inode->i_ino;
2750         u32 found_type;
2751         struct btrfs_path *path = NULL;
2752         struct btrfs_root *root = BTRFS_I(inode)->root;
2753         struct btrfs_file_extent_item *item;
2754         struct extent_buffer *leaf;
2755         struct btrfs_key found_key;
2756         struct extent_map *em = NULL;
2757         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2758         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2759         struct btrfs_trans_handle *trans = NULL;
2760
2761 again:
2762         spin_lock(&em_tree->lock);
2763         em = lookup_extent_mapping(em_tree, start, len);
2764         if (em)
2765                 em->bdev = root->fs_info->fs_devices->latest_bdev;
2766         spin_unlock(&em_tree->lock);
2767
2768         if (em) {
2769                 if (em->start > start || em->start + em->len <= start)
2770                         free_extent_map(em);
2771                 else if (em->block_start == EXTENT_MAP_INLINE && page)
2772                         free_extent_map(em);
2773                 else
2774                         goto out;
2775         }
2776         em = alloc_extent_map(GFP_NOFS);
2777         if (!em) {
2778                 err = -ENOMEM;
2779                 goto out;
2780         }
2781         em->bdev = root->fs_info->fs_devices->latest_bdev;
2782         em->start = EXTENT_MAP_HOLE;
2783         em->len = (u64)-1;
2784
2785         if (!path) {
2786                 path = btrfs_alloc_path();
2787                 BUG_ON(!path);
2788         }
2789
2790         ret = btrfs_lookup_file_extent(trans, root, path,
2791                                        objectid, start, trans != NULL);
2792         if (ret < 0) {
2793                 err = ret;
2794                 goto out;
2795         }
2796
2797         if (ret != 0) {
2798                 if (path->slots[0] == 0)
2799                         goto not_found;
2800                 path->slots[0]--;
2801         }
2802
2803         leaf = path->nodes[0];
2804         item = btrfs_item_ptr(leaf, path->slots[0],
2805                               struct btrfs_file_extent_item);
2806         /* are we inside the extent that was found? */
2807         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2808         found_type = btrfs_key_type(&found_key);
2809         if (found_key.objectid != objectid ||
2810             found_type != BTRFS_EXTENT_DATA_KEY) {
2811                 goto not_found;
2812         }
2813
2814         found_type = btrfs_file_extent_type(leaf, item);
2815         extent_start = found_key.offset;
2816         if (found_type == BTRFS_FILE_EXTENT_REG) {
2817                 extent_end = extent_start +
2818                        btrfs_file_extent_num_bytes(leaf, item);
2819                 err = 0;
2820                 if (start < extent_start || start >= extent_end) {
2821                         em->start = start;
2822                         if (start < extent_start) {
2823                                 if (start + len <= extent_start)
2824                                         goto not_found;
2825                                 em->len = extent_end - extent_start;
2826                         } else {
2827                                 em->len = len;
2828                         }
2829                         goto not_found_em;
2830                 }
2831                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
2832                 if (bytenr == 0) {
2833                         em->start = extent_start;
2834                         em->len = extent_end - extent_start;
2835                         em->block_start = EXTENT_MAP_HOLE;
2836                         goto insert;
2837                 }
2838                 bytenr += btrfs_file_extent_offset(leaf, item);
2839                 em->block_start = bytenr;
2840                 em->start = extent_start;
2841                 em->len = extent_end - extent_start;
2842                 goto insert;
2843         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
2844                 u64 page_start;
2845                 unsigned long ptr;
2846                 char *map;
2847                 size_t size;
2848                 size_t extent_offset;
2849                 size_t copy_size;
2850
2851                 size = btrfs_file_extent_inline_len(leaf, btrfs_item_nr(leaf,
2852                                                     path->slots[0]));
2853                 extent_end = (extent_start + size + root->sectorsize - 1) &
2854                         ~((u64)root->sectorsize - 1);
2855                 if (start < extent_start || start >= extent_end) {
2856                         em->start = start;
2857                         if (start < extent_start) {
2858                                 if (start + len <= extent_start)
2859                                         goto not_found;
2860                                 em->len = extent_end - extent_start;
2861                         } else {
2862                                 em->len = len;
2863                         }
2864                         goto not_found_em;
2865                 }
2866                 em->block_start = EXTENT_MAP_INLINE;
2867
2868                 if (!page) {
2869                         em->start = extent_start;
2870                         em->len = size;
2871                         goto out;
2872                 }
2873
2874                 page_start = page_offset(page) + pg_offset;
2875                 extent_offset = page_start - extent_start;
2876                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
2877                                 size - extent_offset);
2878                 em->start = extent_start + extent_offset;
2879                 em->len = (copy_size + root->sectorsize - 1) &
2880                         ~((u64)root->sectorsize - 1);
2881                 map = kmap(page);
2882                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
2883                 if (create == 0 && !PageUptodate(page)) {
2884                         read_extent_buffer(leaf, map + pg_offset, ptr,
2885                                            copy_size);
2886                         flush_dcache_page(page);
2887                 } else if (create && PageUptodate(page)) {
2888                         if (!trans) {
2889                                 kunmap(page);
2890                                 free_extent_map(em);
2891                                 em = NULL;
2892                                 btrfs_release_path(root, path);
2893                                 trans = btrfs_join_transaction(root, 1);
2894                                 goto again;
2895                         }
2896                         write_extent_buffer(leaf, map + pg_offset, ptr,
2897                                             copy_size);
2898                         btrfs_mark_buffer_dirty(leaf);
2899                 }
2900                 kunmap(page);
2901                 set_extent_uptodate(io_tree, em->start,
2902                                     extent_map_end(em) - 1, GFP_NOFS);
2903                 goto insert;
2904         } else {
2905                 printk("unkknown found_type %d\n", found_type);
2906                 WARN_ON(1);
2907         }
2908 not_found:
2909         em->start = start;
2910         em->len = len;
2911 not_found_em:
2912         em->block_start = EXTENT_MAP_HOLE;
2913 insert:
2914         btrfs_release_path(root, path);
2915         if (em->start > start || extent_map_end(em) <= start) {
2916                 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->len, start, len);
2917                 err = -EIO;
2918                 goto out;
2919         }
2920
2921         err = 0;
2922         spin_lock(&em_tree->lock);
2923         ret = add_extent_mapping(em_tree, em);
2924         /* it is possible that someone inserted the extent into the tree
2925          * while we had the lock dropped.  It is also possible that
2926          * an overlapping map exists in the tree
2927          */
2928         if (ret == -EEXIST) {
2929                 struct extent_map *existing;
2930
2931                 ret = 0;
2932
2933                 existing = lookup_extent_mapping(em_tree, start, len);
2934                 if (existing && (existing->start > start ||
2935                     existing->start + existing->len <= start)) {
2936                         free_extent_map(existing);
2937                         existing = NULL;
2938                 }
2939                 if (!existing) {
2940                         existing = lookup_extent_mapping(em_tree, em->start,
2941                                                          em->len);
2942                         if (existing) {
2943                                 err = merge_extent_mapping(em_tree, existing,
2944                                                            em, start,
2945                                                            root->sectorsize);
2946                                 free_extent_map(existing);
2947                                 if (err) {
2948                                         free_extent_map(em);
2949                                         em = NULL;
2950                                 }
2951                         } else {
2952                                 err = -EIO;
2953                                 printk("failing to insert %Lu %Lu\n",
2954                                        start, len);
2955                                 free_extent_map(em);
2956                                 em = NULL;
2957                         }
2958                 } else {
2959                         free_extent_map(em);
2960                         em = existing;
2961                         err = 0;
2962                 }
2963         }
2964         spin_unlock(&em_tree->lock);
2965 out:
2966         if (path)
2967                 btrfs_free_path(path);
2968         if (trans) {
2969                 ret = btrfs_end_transaction(trans, root);
2970                 if (!err) {
2971                         err = ret;
2972                 }
2973         }
2974         if (err) {
2975                 free_extent_map(em);
2976                 WARN_ON(1);
2977                 return ERR_PTR(err);
2978         }
2979         return em;
2980 }
2981
2982 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
2983                         const struct iovec *iov, loff_t offset,
2984                         unsigned long nr_segs)
2985 {
2986         return -EINVAL;
2987 }
2988
2989 static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
2990 {
2991         return extent_bmap(mapping, iblock, btrfs_get_extent);
2992 }
2993
2994 int btrfs_readpage(struct file *file, struct page *page)
2995 {
2996         struct extent_io_tree *tree;
2997         tree = &BTRFS_I(page->mapping->host)->io_tree;
2998         return extent_read_full_page(tree, page, btrfs_get_extent);
2999 }
3000
3001 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
3002 {
3003         struct extent_io_tree *tree;
3004
3005
3006         if (current->flags & PF_MEMALLOC) {
3007                 redirty_page_for_writepage(wbc, page);
3008                 unlock_page(page);
3009                 return 0;
3010         }
3011         tree = &BTRFS_I(page->mapping->host)->io_tree;
3012         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
3013 }
3014
3015 int btrfs_writepages(struct address_space *mapping,
3016                      struct writeback_control *wbc)
3017 {
3018         struct extent_io_tree *tree;
3019         tree = &BTRFS_I(mapping->host)->io_tree;
3020         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
3021 }
3022
3023 static int
3024 btrfs_readpages(struct file *file, struct address_space *mapping,
3025                 struct list_head *pages, unsigned nr_pages)
3026 {
3027         struct extent_io_tree *tree;
3028         tree = &BTRFS_I(mapping->host)->io_tree;
3029         return extent_readpages(tree, mapping, pages, nr_pages,
3030                                 btrfs_get_extent);
3031 }
3032 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
3033 {
3034         struct extent_io_tree *tree;
3035         struct extent_map_tree *map;
3036         int ret;
3037
3038         tree = &BTRFS_I(page->mapping->host)->io_tree;
3039         map = &BTRFS_I(page->mapping->host)->extent_tree;
3040         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
3041         if (ret == 1) {
3042                 ClearPagePrivate(page);
3043                 set_page_private(page, 0);
3044                 page_cache_release(page);
3045         }
3046         return ret;
3047 }
3048
3049 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
3050 {
3051         if (PageWriteback(page) || PageDirty(page))
3052                 return 0;
3053         return __btrfs_releasepage(page, gfp_flags);
3054 }
3055
3056 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
3057 {
3058         struct extent_io_tree *tree;
3059         struct btrfs_ordered_extent *ordered;
3060         u64 page_start = page_offset(page);
3061         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
3062
3063         wait_on_page_writeback(page);
3064         tree = &BTRFS_I(page->mapping->host)->io_tree;
3065         if (offset) {
3066                 btrfs_releasepage(page, GFP_NOFS);
3067                 return;
3068         }
3069
3070         lock_extent(tree, page_start, page_end, GFP_NOFS);
3071         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
3072                                            page_offset(page));
3073         if (ordered) {
3074                 /*
3075                  * IO on this page will never be started, so we need
3076                  * to account for any ordered extents now
3077                  */
3078                 clear_extent_bit(tree, page_start, page_end,
3079                                  EXTENT_DIRTY | EXTENT_DELALLOC |
3080                                  EXTENT_LOCKED, 1, 0, GFP_NOFS);
3081                 btrfs_finish_ordered_io(page->mapping->host,
3082                                         page_start, page_end);
3083                 btrfs_put_ordered_extent(ordered);
3084                 lock_extent(tree, page_start, page_end, GFP_NOFS);
3085         }
3086         clear_extent_bit(tree, page_start, page_end,
3087                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
3088                  EXTENT_ORDERED,
3089                  1, 1, GFP_NOFS);
3090         __btrfs_releasepage(page, GFP_NOFS);
3091
3092         ClearPageChecked(page);
3093         if (PagePrivate(page)) {
3094                 ClearPagePrivate(page);
3095                 set_page_private(page, 0);
3096                 page_cache_release(page);
3097         }
3098 }
3099
3100 /*
3101  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
3102  * called from a page fault handler when a page is first dirtied. Hence we must
3103  * be careful to check for EOF conditions here. We set the page up correctly
3104  * for a written page which means we get ENOSPC checking when writing into
3105  * holes and correct delalloc and unwritten extent mapping on filesystems that
3106  * support these features.
3107  *
3108  * We are not allowed to take the i_mutex here so we have to play games to
3109  * protect against truncate races as the page could now be beyond EOF.  Because
3110  * vmtruncate() writes the inode size before removing pages, once we have the
3111  * page lock we can determine safely if the page is beyond EOF. If it is not
3112  * beyond EOF, then the page is guaranteed safe against truncation until we
3113  * unlock the page.
3114  */
3115 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
3116 {
3117         struct inode *inode = fdentry(vma->vm_file)->d_inode;
3118         struct btrfs_root *root = BTRFS_I(inode)->root;
3119         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3120         struct btrfs_ordered_extent *ordered;
3121         char *kaddr;
3122         unsigned long zero_start;
3123         loff_t size;
3124         int ret;
3125         u64 page_start;
3126         u64 page_end;
3127
3128         ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
3129         if (ret)
3130                 goto out;
3131
3132         ret = -EINVAL;
3133 again:
3134         lock_page(page);
3135         size = i_size_read(inode);
3136         page_start = page_offset(page);
3137         page_end = page_start + PAGE_CACHE_SIZE - 1;
3138
3139         if ((page->mapping != inode->i_mapping) ||
3140             (page_start >= size)) {
3141                 /* page got truncated out from underneath us */
3142                 goto out_unlock;
3143         }
3144         wait_on_page_writeback(page);
3145
3146         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
3147         set_page_extent_mapped(page);
3148
3149         /*
3150          * we can't set the delalloc bits if there are pending ordered
3151          * extents.  Drop our locks and wait for them to finish
3152          */
3153         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3154         if (ordered) {
3155                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3156                 unlock_page(page);
3157                 btrfs_start_ordered_extent(inode, ordered, 1);
3158                 btrfs_put_ordered_extent(ordered);
3159                 goto again;
3160         }
3161
3162         btrfs_set_extent_delalloc(inode, page_start, page_end);
3163         ret = 0;
3164
3165         /* page is wholly or partially inside EOF */
3166         if (page_start + PAGE_CACHE_SIZE > size)
3167                 zero_start = size & ~PAGE_CACHE_MASK;
3168         else
3169                 zero_start = PAGE_CACHE_SIZE;
3170
3171         if (zero_start != PAGE_CACHE_SIZE) {
3172                 kaddr = kmap(page);
3173                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
3174                 flush_dcache_page(page);
3175                 kunmap(page);
3176         }
3177         ClearPageChecked(page);
3178         set_page_dirty(page);
3179         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3180
3181 out_unlock:
3182         unlock_page(page);
3183 out:
3184         return ret;
3185 }
3186
3187 static void btrfs_truncate(struct inode *inode)
3188 {
3189         struct btrfs_root *root = BTRFS_I(inode)->root;
3190         int ret;
3191         struct btrfs_trans_handle *trans;
3192         unsigned long nr;
3193         u64 mask = root->sectorsize - 1;
3194
3195         if (!S_ISREG(inode->i_mode))
3196                 return;
3197         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3198                 return;
3199
3200         btrfs_truncate_page(inode->i_mapping, inode->i_size);
3201         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
3202
3203         trans = btrfs_start_transaction(root, 1);
3204         btrfs_set_trans_block_group(trans, inode);
3205         btrfs_i_size_write(inode, inode->i_size);
3206
3207         ret = btrfs_orphan_add(trans, inode);
3208         if (ret)
3209                 goto out;
3210         /* FIXME, add redo link to tree so we don't leak on crash */
3211         ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size,
3212                                       BTRFS_EXTENT_DATA_KEY);
3213         btrfs_update_inode(trans, root, inode);
3214
3215         ret = btrfs_orphan_del(trans, inode);
3216         BUG_ON(ret);
3217
3218 out:
3219         nr = trans->blocks_used;
3220         ret = btrfs_end_transaction_throttle(trans, root);
3221         BUG_ON(ret);
3222         btrfs_btree_balance_dirty(root, nr);
3223 }
3224
3225 /*
3226  * Invalidate a single dcache entry at the root of the filesystem.
3227  * Needed after creation of snapshot or subvolume.
3228  */
3229 void btrfs_invalidate_dcache_root(struct btrfs_root *root, char *name,
3230                                   int namelen)
3231 {
3232         struct dentry *alias, *entry;
3233         struct qstr qstr;
3234
3235         alias = d_find_alias(root->fs_info->sb->s_root->d_inode);
3236         if (alias) {
3237                 qstr.name = name;
3238                 qstr.len = namelen;
3239                 /* change me if btrfs ever gets a d_hash operation */
3240                 qstr.hash = full_name_hash(qstr.name, qstr.len);
3241                 entry = d_lookup(alias, &qstr);
3242                 dput(alias);
3243                 if (entry) {
3244                         d_invalidate(entry);
3245                         dput(entry);
3246                 }
3247         }
3248 }
3249
3250 /*
3251  * create a new subvolume directory/inode (helper for the ioctl).
3252  */
3253 int btrfs_create_subvol_root(struct btrfs_root *new_root,
3254                 struct btrfs_trans_handle *trans, u64 new_dirid,
3255                 struct btrfs_block_group_cache *block_group)
3256 {
3257         struct inode *inode;
3258         u64 index = 0;
3259
3260         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
3261                                 new_dirid, block_group, S_IFDIR | 0700, &index);
3262         if (IS_ERR(inode))
3263                 return PTR_ERR(inode);
3264         inode->i_op = &btrfs_dir_inode_operations;
3265         inode->i_fop = &btrfs_dir_file_operations;
3266         new_root->inode = inode;
3267
3268         inode->i_nlink = 1;
3269         btrfs_i_size_write(inode, 0);
3270
3271         return btrfs_update_inode(trans, new_root, inode);
3272 }
3273
3274 /* helper function for file defrag and space balancing.  This
3275  * forces readahead on a given range of bytes in an inode
3276  */
3277 unsigned long btrfs_force_ra(struct address_space *mapping,
3278                               struct file_ra_state *ra, struct file *file,
3279                               pgoff_t offset, pgoff_t last_index)
3280 {
3281         pgoff_t req_size = last_index - offset + 1;
3282
3283         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
3284         return offset + req_size;
3285 }
3286
3287 struct inode *btrfs_alloc_inode(struct super_block *sb)
3288 {
3289         struct btrfs_inode *ei;
3290
3291         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
3292         if (!ei)
3293                 return NULL;
3294         ei->last_trans = 0;
3295         ei->logged_trans = 0;
3296         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
3297         ei->i_acl = BTRFS_ACL_NOT_CACHED;
3298         ei->i_default_acl = BTRFS_ACL_NOT_CACHED;
3299         INIT_LIST_HEAD(&ei->i_orphan);
3300         return &ei->vfs_inode;
3301 }
3302
3303 void btrfs_destroy_inode(struct inode *inode)
3304 {
3305         struct btrfs_ordered_extent *ordered;
3306         WARN_ON(!list_empty(&inode->i_dentry));
3307         WARN_ON(inode->i_data.nrpages);
3308
3309         if (BTRFS_I(inode)->i_acl &&
3310             BTRFS_I(inode)->i_acl != BTRFS_ACL_NOT_CACHED)
3311                 posix_acl_release(BTRFS_I(inode)->i_acl);
3312         if (BTRFS_I(inode)->i_default_acl &&
3313             BTRFS_I(inode)->i_default_acl != BTRFS_ACL_NOT_CACHED)
3314                 posix_acl_release(BTRFS_I(inode)->i_default_acl);
3315
3316         spin_lock(&BTRFS_I(inode)->root->list_lock);
3317         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
3318                 printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
3319                        " list\n", inode->i_ino);
3320                 dump_stack();
3321         }
3322         spin_unlock(&BTRFS_I(inode)->root->list_lock);
3323
3324         while(1) {
3325                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
3326                 if (!ordered)
3327                         break;
3328                 else {
3329                         printk("found ordered extent %Lu %Lu\n",
3330                                ordered->file_offset, ordered->len);
3331                         btrfs_remove_ordered_extent(inode, ordered);
3332                         btrfs_put_ordered_extent(ordered);
3333                         btrfs_put_ordered_extent(ordered);
3334                 }
3335         }
3336         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
3337         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
3338 }
3339
3340 static void init_once(void *foo)
3341 {
3342         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
3343
3344         inode_init_once(&ei->vfs_inode);
3345 }
3346
3347 void btrfs_destroy_cachep(void)
3348 {
3349         if (btrfs_inode_cachep)
3350                 kmem_cache_destroy(btrfs_inode_cachep);
3351         if (btrfs_trans_handle_cachep)
3352                 kmem_cache_destroy(btrfs_trans_handle_cachep);
3353         if (btrfs_transaction_cachep)
3354                 kmem_cache_destroy(btrfs_transaction_cachep);
3355         if (btrfs_bit_radix_cachep)
3356                 kmem_cache_destroy(btrfs_bit_radix_cachep);
3357         if (btrfs_path_cachep)
3358                 kmem_cache_destroy(btrfs_path_cachep);
3359 }
3360
3361 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
3362                                        unsigned long extra_flags,
3363                                        void (*ctor)(void *))
3364 {
3365         return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
3366                                  SLAB_MEM_SPREAD | extra_flags), ctor);
3367 }
3368
3369 int btrfs_init_cachep(void)
3370 {
3371         btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
3372                                           sizeof(struct btrfs_inode),
3373                                           0, init_once);
3374         if (!btrfs_inode_cachep)
3375                 goto fail;
3376         btrfs_trans_handle_cachep =
3377                         btrfs_cache_create("btrfs_trans_handle_cache",
3378                                            sizeof(struct btrfs_trans_handle),
3379                                            0, NULL);
3380         if (!btrfs_trans_handle_cachep)
3381                 goto fail;
3382         btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
3383                                              sizeof(struct btrfs_transaction),
3384                                              0, NULL);
3385         if (!btrfs_transaction_cachep)
3386                 goto fail;
3387         btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
3388                                          sizeof(struct btrfs_path),
3389                                          0, NULL);
3390         if (!btrfs_path_cachep)
3391                 goto fail;
3392         btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
3393                                               SLAB_DESTROY_BY_RCU, NULL);
3394         if (!btrfs_bit_radix_cachep)
3395                 goto fail;
3396         return 0;
3397 fail:
3398         btrfs_destroy_cachep();
3399         return -ENOMEM;
3400 }
3401
3402 static int btrfs_getattr(struct vfsmount *mnt,
3403                          struct dentry *dentry, struct kstat *stat)
3404 {
3405         struct inode *inode = dentry->d_inode;
3406         generic_fillattr(inode, stat);
3407         stat->blksize = PAGE_CACHE_SIZE;
3408         stat->blocks = inode->i_blocks + (BTRFS_I(inode)->delalloc_bytes >> 9);
3409         return 0;
3410 }
3411
3412 static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
3413                            struct inode * new_dir,struct dentry *new_dentry)
3414 {
3415         struct btrfs_trans_handle *trans;
3416         struct btrfs_root *root = BTRFS_I(old_dir)->root;
3417         struct inode *new_inode = new_dentry->d_inode;
3418         struct inode *old_inode = old_dentry->d_inode;
3419         struct timespec ctime = CURRENT_TIME;
3420         u64 index = 0;
3421         int ret;
3422
3423         if (S_ISDIR(old_inode->i_mode) && new_inode &&
3424             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
3425                 return -ENOTEMPTY;
3426         }
3427
3428         ret = btrfs_check_free_space(root, 1, 0);
3429         if (ret)
3430                 goto out_unlock;
3431
3432         trans = btrfs_start_transaction(root, 1);
3433
3434         btrfs_set_trans_block_group(trans, new_dir);
3435
3436         btrfs_inc_nlink(old_dentry->d_inode);
3437         old_dir->i_ctime = old_dir->i_mtime = ctime;
3438         new_dir->i_ctime = new_dir->i_mtime = ctime;
3439         old_inode->i_ctime = ctime;
3440
3441         ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode,
3442                                  old_dentry->d_name.name,
3443                                  old_dentry->d_name.len);
3444         if (ret)
3445                 goto out_fail;
3446
3447         if (new_inode) {
3448                 new_inode->i_ctime = CURRENT_TIME;
3449                 ret = btrfs_unlink_inode(trans, root, new_dir,
3450                                          new_dentry->d_inode,
3451                                          new_dentry->d_name.name,
3452                                          new_dentry->d_name.len);
3453                 if (ret)
3454                         goto out_fail;
3455                 if (new_inode->i_nlink == 0) {
3456                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
3457                         if (ret)
3458                                 goto out_fail;
3459                 }
3460
3461         }
3462         ret = btrfs_set_inode_index(new_dir, old_inode, &index);
3463         if (ret)
3464                 goto out_fail;
3465
3466         ret = btrfs_add_link(trans, new_dentry->d_parent->d_inode,
3467                              old_inode, new_dentry->d_name.name,
3468                              new_dentry->d_name.len, 1, index);
3469         if (ret)
3470                 goto out_fail;
3471
3472 out_fail:
3473         btrfs_end_transaction_throttle(trans, root);
3474 out_unlock:
3475         return ret;
3476 }
3477
3478 /*
3479  * some fairly slow code that needs optimization. This walks the list
3480  * of all the inodes with pending delalloc and forces them to disk.
3481  */
3482 int btrfs_start_delalloc_inodes(struct btrfs_root *root)
3483 {
3484         struct list_head *head = &root->fs_info->delalloc_inodes;
3485         struct btrfs_inode *binode;
3486         struct inode *inode;
3487         unsigned long flags;
3488
3489         spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
3490         while(!list_empty(head)) {
3491                 binode = list_entry(head->next, struct btrfs_inode,
3492                                     delalloc_inodes);
3493                 inode = igrab(&binode->vfs_inode);
3494                 if (!inode)
3495                         list_del_init(&binode->delalloc_inodes);
3496                 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
3497                 if (inode) {
3498                         filemap_flush(inode->i_mapping);
3499                         iput(inode);
3500                 }
3501                 cond_resched();
3502                 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
3503         }
3504         spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
3505
3506         /* the filemap_flush will queue IO into the worker threads, but
3507          * we have to make sure the IO is actually started and that
3508          * ordered extents get created before we return
3509          */
3510         atomic_inc(&root->fs_info->async_submit_draining);
3511         while(atomic_read(&root->fs_info->nr_async_submits)) {
3512                 wait_event(root->fs_info->async_submit_wait,
3513                    (atomic_read(&root->fs_info->nr_async_submits) == 0));
3514         }
3515         atomic_dec(&root->fs_info->async_submit_draining);
3516         return 0;
3517 }
3518
3519 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
3520                          const char *symname)
3521 {
3522         struct btrfs_trans_handle *trans;
3523         struct btrfs_root *root = BTRFS_I(dir)->root;
3524         struct btrfs_path *path;
3525         struct btrfs_key key;
3526         struct inode *inode = NULL;
3527         int err;
3528         int drop_inode = 0;
3529         u64 objectid;
3530         u64 index = 0 ;
3531         int name_len;
3532         int datasize;
3533         unsigned long ptr;
3534         struct btrfs_file_extent_item *ei;
3535         struct extent_buffer *leaf;
3536         unsigned long nr = 0;
3537
3538         name_len = strlen(symname) + 1;
3539         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
3540                 return -ENAMETOOLONG;
3541
3542         err = btrfs_check_free_space(root, 1, 0);
3543         if (err)
3544                 goto out_fail;
3545
3546         trans = btrfs_start_transaction(root, 1);
3547         btrfs_set_trans_block_group(trans, dir);
3548
3549         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3550         if (err) {
3551                 err = -ENOSPC;
3552                 goto out_unlock;
3553         }
3554
3555         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3556                                 dentry->d_name.len,
3557                                 dentry->d_parent->d_inode->i_ino, objectid,
3558                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
3559                                 &index);
3560         err = PTR_ERR(inode);
3561         if (IS_ERR(inode))
3562                 goto out_unlock;
3563
3564         err = btrfs_init_acl(inode, dir);
3565         if (err) {
3566                 drop_inode = 1;
3567                 goto out_unlock;
3568         }
3569
3570         btrfs_set_trans_block_group(trans, inode);
3571         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3572         if (err)
3573                 drop_inode = 1;
3574         else {
3575                 inode->i_mapping->a_ops = &btrfs_aops;
3576                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3577                 inode->i_fop = &btrfs_file_operations;
3578                 inode->i_op = &btrfs_file_inode_operations;
3579                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3580         }
3581         dir->i_sb->s_dirt = 1;
3582         btrfs_update_inode_block_group(trans, inode);
3583         btrfs_update_inode_block_group(trans, dir);
3584         if (drop_inode)
3585                 goto out_unlock;
3586
3587         path = btrfs_alloc_path();
3588         BUG_ON(!path);
3589         key.objectid = inode->i_ino;
3590         key.offset = 0;
3591         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
3592         datasize = btrfs_file_extent_calc_inline_size(name_len);
3593         err = btrfs_insert_empty_item(trans, root, path, &key,
3594                                       datasize);
3595         if (err) {
3596                 drop_inode = 1;
3597                 goto out_unlock;
3598         }
3599         leaf = path->nodes[0];
3600         ei = btrfs_item_ptr(leaf, path->slots[0],
3601                             struct btrfs_file_extent_item);
3602         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
3603         btrfs_set_file_extent_type(leaf, ei,
3604                                    BTRFS_FILE_EXTENT_INLINE);
3605         ptr = btrfs_file_extent_inline_start(ei);
3606         write_extent_buffer(leaf, symname, ptr, name_len);
3607         btrfs_mark_buffer_dirty(leaf);
3608         btrfs_free_path(path);
3609
3610         inode->i_op = &btrfs_symlink_inode_operations;
3611         inode->i_mapping->a_ops = &btrfs_symlink_aops;
3612         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3613         btrfs_i_size_write(inode, name_len - 1);
3614         err = btrfs_update_inode(trans, root, inode);
3615         if (err)
3616                 drop_inode = 1;
3617
3618 out_unlock:
3619         nr = trans->blocks_used;
3620         btrfs_end_transaction_throttle(trans, root);
3621 out_fail:
3622         if (drop_inode) {
3623                 inode_dec_link_count(inode);
3624                 iput(inode);
3625         }
3626         btrfs_btree_balance_dirty(root, nr);
3627         return err;
3628 }
3629
3630 static int btrfs_set_page_dirty(struct page *page)
3631 {
3632         return __set_page_dirty_nobuffers(page);
3633 }
3634
3635 static int btrfs_permission(struct inode *inode, int mask)
3636 {
3637         if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
3638                 return -EACCES;
3639         return generic_permission(inode, mask, btrfs_check_acl);
3640 }
3641
3642 static struct inode_operations btrfs_dir_inode_operations = {
3643         .lookup         = btrfs_lookup,
3644         .create         = btrfs_create,
3645         .unlink         = btrfs_unlink,
3646         .link           = btrfs_link,
3647         .mkdir          = btrfs_mkdir,
3648         .rmdir          = btrfs_rmdir,
3649         .rename         = btrfs_rename,
3650         .symlink        = btrfs_symlink,
3651         .setattr        = btrfs_setattr,
3652         .mknod          = btrfs_mknod,
3653         .setxattr       = btrfs_setxattr,
3654         .getxattr       = btrfs_getxattr,
3655         .listxattr      = btrfs_listxattr,
3656         .removexattr    = btrfs_removexattr,
3657         .permission     = btrfs_permission,
3658 };
3659 static struct inode_operations btrfs_dir_ro_inode_operations = {
3660         .lookup         = btrfs_lookup,
3661         .permission     = btrfs_permission,
3662 };
3663 static struct file_operations btrfs_dir_file_operations = {
3664         .llseek         = generic_file_llseek,
3665         .read           = generic_read_dir,
3666         .readdir        = btrfs_real_readdir,
3667         .unlocked_ioctl = btrfs_ioctl,
3668 #ifdef CONFIG_COMPAT
3669         .compat_ioctl   = btrfs_ioctl,
3670 #endif
3671         .release        = btrfs_release_file,
3672         .fsync          = btrfs_sync_file,
3673 };
3674
3675 static struct extent_io_ops btrfs_extent_io_ops = {
3676         .fill_delalloc = run_delalloc_range,
3677         .submit_bio_hook = btrfs_submit_bio_hook,
3678         .merge_bio_hook = btrfs_merge_bio_hook,
3679         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
3680         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
3681         .writepage_start_hook = btrfs_writepage_start_hook,
3682         .readpage_io_failed_hook = btrfs_io_failed_hook,
3683         .set_bit_hook = btrfs_set_bit_hook,
3684         .clear_bit_hook = btrfs_clear_bit_hook,
3685 };
3686
3687 static struct address_space_operations btrfs_aops = {
3688         .readpage       = btrfs_readpage,
3689         .writepage      = btrfs_writepage,
3690         .writepages     = btrfs_writepages,
3691         .readpages      = btrfs_readpages,
3692         .sync_page      = block_sync_page,
3693         .bmap           = btrfs_bmap,
3694         .direct_IO      = btrfs_direct_IO,
3695         .invalidatepage = btrfs_invalidatepage,
3696         .releasepage    = btrfs_releasepage,
3697         .set_page_dirty = btrfs_set_page_dirty,
3698 };
3699
3700 static struct address_space_operations btrfs_symlink_aops = {
3701         .readpage       = btrfs_readpage,
3702         .writepage      = btrfs_writepage,
3703         .invalidatepage = btrfs_invalidatepage,
3704         .releasepage    = btrfs_releasepage,
3705 };
3706
3707 static struct inode_operations btrfs_file_inode_operations = {
3708         .truncate       = btrfs_truncate,
3709         .getattr        = btrfs_getattr,
3710         .setattr        = btrfs_setattr,
3711         .setxattr       = btrfs_setxattr,
3712         .getxattr       = btrfs_getxattr,
3713         .listxattr      = btrfs_listxattr,
3714         .removexattr    = btrfs_removexattr,
3715         .permission     = btrfs_permission,
3716 };
3717 static struct inode_operations btrfs_special_inode_operations = {
3718         .getattr        = btrfs_getattr,
3719         .setattr        = btrfs_setattr,
3720         .permission     = btrfs_permission,
3721         .setxattr       = btrfs_setxattr,
3722         .getxattr       = btrfs_getxattr,
3723         .listxattr      = btrfs_listxattr,
3724         .removexattr    = btrfs_removexattr,
3725 };
3726 static struct inode_operations btrfs_symlink_inode_operations = {
3727         .readlink       = generic_readlink,
3728         .follow_link    = page_follow_link_light,
3729         .put_link       = page_put_link,
3730         .permission     = btrfs_permission,
3731 };