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 int btrfs_check_free_space(struct btrfs_root *root, u64 num_required,
  87                            int for_del)
  88 {
  89         u64 total;
  90         u64 used;
  91         u64 thresh;
  92         unsigned long flags;
  93         int ret = 0;
  94
  95         spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
  96         total = btrfs_super_total_bytes(&root->fs_info->super_copy);
  97         used = btrfs_super_bytes_used(&root->fs_info->super_copy);
  98         if (for_del)
  99                 thresh = total * 90;
 100         else
 101                 thresh = total * 85;
 102
 103         do_div(thresh, 100);
 104
 105         if (used + root->fs_info->delalloc_bytes + num_required > thresh)
 106                 ret = -ENOSPC;
 107         spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
 108         return ret;
 109 }
 110
 111 static int cow_file_range(struct inode *inode, u64 start, u64 end)
 112 {
 113         struct btrfs_root *root = BTRFS_I(inode)->root;
 114         struct btrfs_trans_handle *trans;
 115         u64 alloc_hint = 0;
 116         u64 num_bytes;
 117         u64 cur_alloc_size;
 118         u64 blocksize = root->sectorsize;
 119         u64 orig_num_bytes;
 120         struct btrfs_key ins;
 121         struct extent_map *em;
 122         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 123         int ret = 0;
 124
 125         trans = btrfs_join_transaction(root, 1);
 126         BUG_ON(!trans);
 127         btrfs_set_trans_block_group(trans, inode);
 128
 129         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 130         num_bytes = max(blocksize,  num_bytes);
 131         orig_num_bytes = num_bytes;
 132
 133         if (alloc_hint == EXTENT_MAP_INLINE)
 134                 goto out;
 135
 136         BUG_ON(num_bytes > btrfs_super_total_bytes(&root->fs_info->super_copy));
 137         mutex_lock(&BTRFS_I(inode)->extent_mutex);
 138         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1);
 139         mutex_unlock(&BTRFS_I(inode)->extent_mutex);
 140
 141         while(num_bytes > 0) {
 142                 cur_alloc_size = min(num_bytes, root->fs_info->max_extent);
 143                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 144                                            root->sectorsize, 0, alloc_hint,
 145                                            (u64)-1, &ins, 1);
 146                 if (ret) {
 147                         WARN_ON(1);
 148                         goto out;
 149                 }
 150                 em = alloc_extent_map(GFP_NOFS);
 151                 em->start = start;
 152                 em->len = ins.offset;
 153                 em->block_start = ins.objectid;
 154                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 155                 mutex_lock(&BTRFS_I(inode)->extent_mutex);
 156                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 157                 while(1) {
 158                         spin_lock(&em_tree->lock);
 159                         ret = add_extent_mapping(em_tree, em);
 160                         spin_unlock(&em_tree->lock);
 161                         if (ret != -EEXIST) {
 162                                 free_extent_map(em);
 163                                 break;
 164                         }
 165                         btrfs_drop_extent_cache(inode, start,
 166                                                 start + ins.offset - 1);
 167                 }
 168                 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
 169
 170                 cur_alloc_size = ins.offset;
 171                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 172                                                ins.offset, 0);
 173                 BUG_ON(ret);
 174                 if (num_bytes < cur_alloc_size) {
 175                         printk("num_bytes %Lu cur_alloc %Lu\n", num_bytes,
 176                                cur_alloc_size);
 177                         break;
 178                 }
 179                 num_bytes -= cur_alloc_size;
 180                 alloc_hint = ins.objectid + ins.offset;
 181                 start += cur_alloc_size;
 182         }
 183 out:
 184         btrfs_end_transaction(trans, root);
 185         return ret;
 186 }
 187
 188 static int run_delalloc_nocow(struct inode *inode, u64 start, u64 end)
 189 {
 190         u64 extent_start;
 191         u64 extent_end;
 192         u64 bytenr;
 193         u64 loops = 0;
 194         u64 total_fs_bytes;
 195         struct btrfs_root *root = BTRFS_I(inode)->root;
 196         struct btrfs_block_group_cache *block_group;
 197         struct btrfs_trans_handle *trans;
 198         struct extent_buffer *leaf;
 199         int found_type;
 200         struct btrfs_path *path;
 201         struct btrfs_file_extent_item *item;
 202         int ret;
 203         int err = 0;
 204         struct btrfs_key found_key;
 205
 206         total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
 207         path = btrfs_alloc_path();
 208         BUG_ON(!path);
 209         trans = btrfs_join_transaction(root, 1);
 210         BUG_ON(!trans);
 211 again:
 212         ret = btrfs_lookup_file_extent(NULL, root, path,
 213                                        inode->i_ino, start, 0);
 214         if (ret < 0) {
 215                 err = ret;
 216                 goto out;
 217         }
 218
 219         if (ret != 0) {
 220                 if (path->slots[0] == 0)
 221                         goto not_found;
 222                 path->slots[0]--;
 223         }
 224
 225         leaf = path->nodes[0];
 226         item = btrfs_item_ptr(leaf, path->slots[0],
 227                               struct btrfs_file_extent_item);
 228
 229         /* are we inside the extent that was found? */
 230         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 231         found_type = btrfs_key_type(&found_key);
 232         if (found_key.objectid != inode->i_ino ||
 233             found_type != BTRFS_EXTENT_DATA_KEY)
 234                 goto not_found;
 235
 236         found_type = btrfs_file_extent_type(leaf, item);
 237         extent_start = found_key.offset;
 238         if (found_type == BTRFS_FILE_EXTENT_REG) {
 239                 u64 extent_num_bytes;
 240
 241                 extent_num_bytes = btrfs_file_extent_num_bytes(leaf, item);
 242                 extent_end = extent_start + extent_num_bytes;
 243                 err = 0;
 244
 245                 if (loops && start != extent_start)
 246                         goto not_found;
 247
 248                 if (start < extent_start || start >= extent_end)
 249                         goto not_found;
 250
 251                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
 252                 if (bytenr == 0)
 253                         goto not_found;
 254
 255                 if (btrfs_cross_ref_exists(trans, root, &found_key, bytenr))
 256                         goto not_found;
 257                 /*
 258                  * we may be called by the resizer, make sure we're inside
 259                  * the limits of the FS
 260                  */
 261                 block_group = btrfs_lookup_block_group(root->fs_info,
 262                                                        bytenr);
 263                 if (!block_group || block_group->ro)
 264                         goto not_found;
 265
 266                 bytenr += btrfs_file_extent_offset(leaf, item);
 267                 extent_num_bytes = min(end + 1, extent_end) - start;
 268                 ret = btrfs_add_ordered_extent(inode, start, bytenr,
 269                                                 extent_num_bytes, 1);
 270                 if (ret) {
 271                         err = ret;
 272                         goto out;
 273                 }
 274
 275                 btrfs_release_path(root, path);
 276                 start = extent_end;
 277                 if (start <= end) {
 278                         loops++;
 279                         goto again;
 280                 }
 281         } else {
 282 not_found:
 283                 btrfs_end_transaction(trans, root);
 284                 btrfs_free_path(path);
 285                 return cow_file_range(inode, start, end);
 286         }
 287 out:
 288         WARN_ON(err);
 289         btrfs_end_transaction(trans, root);
 290         btrfs_free_path(path);
 291         return err;
 292 }
 293
 294 static int run_delalloc_range(struct inode *inode, u64 start, u64 end)
 295 {
 296         struct btrfs_root *root = BTRFS_I(inode)->root;
 297         int ret;
 298
 299         if (btrfs_test_opt(root, NODATACOW) ||
 300             btrfs_test_flag(inode, NODATACOW))
 301                 ret = run_delalloc_nocow(inode, start, end);
 302         else
 303                 ret = cow_file_range(inode, start, end);
 304
 305         return ret;
 306 }
 307
 308 int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
 309                        unsigned long old, unsigned long bits)
 310 {
 311         unsigned long flags;
 312         if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
 313                 struct btrfs_root *root = BTRFS_I(inode)->root;
 314                 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
 315                 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
 316                 root->fs_info->delalloc_bytes += end - start + 1;
 317                 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
 318                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
 319                                       &root->fs_info->delalloc_inodes);
 320                 }
 321                 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
 322         }
 323         return 0;
 324 }
 325
 326 int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
 327                          unsigned long old, unsigned long bits)
 328 {
 329         if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
 330                 struct btrfs_root *root = BTRFS_I(inode)->root;
 331                 unsigned long flags;
 332
 333                 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
 334                 if (end - start + 1 > root->fs_info->delalloc_bytes) {
 335                         printk("warning: delalloc account %Lu %Lu\n",
 336                                end - start + 1, root->fs_info->delalloc_bytes);
 337                         root->fs_info->delalloc_bytes = 0;
 338                         BTRFS_I(inode)->delalloc_bytes = 0;
 339                 } else {
 340                         root->fs_info->delalloc_bytes -= end - start + 1;
 341                         BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
 342                 }
 343                 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
 344                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
 345                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
 346                 }
 347                 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
 348         }
 349         return 0;
 350 }
 351
 352 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
 353                          size_t size, struct bio *bio)
 354 {
 355         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 356         struct btrfs_mapping_tree *map_tree;
 357         u64 logical = bio->bi_sector << 9;
 358         u64 length = 0;
 359         u64 map_length;
 360         int ret;
 361
 362         length = bio->bi_size;
 363         map_tree = &root->fs_info->mapping_tree;
 364         map_length = length;
 365         ret = btrfs_map_block(map_tree, READ, logical,
 366                               &map_length, NULL, 0);
 367
 368         if (map_length < length + size) {
 369                 return 1;
 370         }
 371         return 0;
 372 }
 373
 374 int __btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 375                           int mirror_num)
 376 {
 377         struct btrfs_root *root = BTRFS_I(inode)->root;
 378         int ret = 0;
 379
 380         ret = btrfs_csum_one_bio(root, inode, bio);
 381         BUG_ON(ret);
 382
 383         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
 384 }
 385
 386 int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 387                           int mirror_num)
 388 {
 389         struct btrfs_root *root = BTRFS_I(inode)->root;
 390         int ret = 0;
 391
 392         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
 393         BUG_ON(ret);
 394
 395         if (btrfs_test_opt(root, NODATASUM) ||
 396             btrfs_test_flag(inode, NODATASUM)) {
 397                 goto mapit;
 398         }
 399
 400         if (!(rw & (1 << BIO_RW))) {
 401                 btrfs_lookup_bio_sums(root, inode, bio);
 402                 goto mapit;
 403         }
 404         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 405                                    inode, rw, bio, mirror_num,
 406                                    __btrfs_submit_bio_hook);
 407 mapit:
 408         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
 409 }
 410
 411 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
 412                              struct inode *inode, u64 file_offset,
 413                              struct list_head *list)
 414 {
 415         struct list_head *cur;
 416         struct btrfs_ordered_sum *sum;
 417
 418         btrfs_set_trans_block_group(trans, inode);
 419         list_for_each(cur, list) {
 420                 sum = list_entry(cur, struct btrfs_ordered_sum, list);
 421                 btrfs_csum_file_blocks(trans, BTRFS_I(inode)->root,
 422                                        inode, sum);
 423         }
 424         return 0;
 425 }
 426
 427 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end)
 428 {
 429         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
 430                                    GFP_NOFS);
 431 }
 432
 433 struct btrfs_writepage_fixup {
 434         struct page *page;
 435         struct btrfs_work work;
 436 };
 437
 438 /* see btrfs_writepage_start_hook for details on why this is required */
 439 void btrfs_writepage_fixup_worker(struct btrfs_work *work)
 440 {
 441         struct btrfs_writepage_fixup *fixup;
 442         struct btrfs_ordered_extent *ordered;
 443         struct page *page;
 444         struct inode *inode;
 445         u64 page_start;
 446         u64 page_end;
 447
 448         fixup = container_of(work, struct btrfs_writepage_fixup, work);
 449         page = fixup->page;
 450 again:
 451         lock_page(page);
 452         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
 453                 ClearPageChecked(page);
 454                 goto out_page;
 455         }
 456
 457         inode = page->mapping->host;
 458         page_start = page_offset(page);
 459         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
 460
 461         lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
 462
 463         /* already ordered? We're done */
 464         if (test_range_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
 465                              EXTENT_ORDERED, 0)) {
 466                 goto out;
 467         }
 468
 469         ordered = btrfs_lookup_ordered_extent(inode, page_start);
 470         if (ordered) {
 471                 unlock_extent(&BTRFS_I(inode)->io_tree, page_start,
 472                               page_end, GFP_NOFS);
 473                 unlock_page(page);
 474                 btrfs_start_ordered_extent(inode, ordered, 1);
 475                 goto again;
 476         }
 477
 478         btrfs_set_extent_delalloc(inode, page_start, page_end);
 479         ClearPageChecked(page);
 480 out:
 481         unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
 482 out_page:
 483         unlock_page(page);
 484         page_cache_release(page);
 485 }
 486
 487 /*
 488  * There are a few paths in the higher layers of the kernel that directly
 489  * set the page dirty bit without asking the filesystem if it is a
 490  * good idea.  This causes problems because we want to make sure COW
 491  * properly happens and the data=ordered rules are followed.
 492  *
 493  * In our case any range that doesn't have the EXTENT_ORDERED bit set
 494  * hasn't been properly setup for IO.  We kick off an async process
 495  * to fix it up.  The async helper will wait for ordered extents, set
 496  * the delalloc bit and make it safe to write the page.
 497  */
 498 int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
 499 {
 500         struct inode *inode = page->mapping->host;
 501         struct btrfs_writepage_fixup *fixup;
 502         struct btrfs_root *root = BTRFS_I(inode)->root;
 503         int ret;
 504
 505         ret = test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
 506                              EXTENT_ORDERED, 0);
 507         if (ret)
 508                 return 0;
 509
 510         if (PageChecked(page))
 511                 return -EAGAIN;
 512
 513         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
 514         if (!fixup)
 515                 return -EAGAIN;
 516
 517         SetPageChecked(page);
 518         page_cache_get(page);
 519         fixup->work.func = btrfs_writepage_fixup_worker;
 520         fixup->page = page;
 521         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
 522         return -EAGAIN;
 523 }
 524
 525 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
 526 {
 527         struct btrfs_root *root = BTRFS_I(inode)->root;
 528         struct btrfs_trans_handle *trans;
 529         struct btrfs_ordered_extent *ordered_extent;
 530         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 531         struct btrfs_file_extent_item *extent_item;
 532         struct btrfs_path *path = NULL;
 533         struct extent_buffer *leaf;
 534         u64 alloc_hint = 0;
 535         struct list_head list;
 536         struct btrfs_key ins;
 537         int ret;
 538
 539         ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
 540         if (!ret)
 541                 return 0;
 542
 543         trans = btrfs_join_transaction(root, 1);
 544
 545         ordered_extent = btrfs_lookup_ordered_extent(inode, start);
 546         BUG_ON(!ordered_extent);
 547         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
 548                 goto nocow;
 549
 550         path = btrfs_alloc_path();
 551         BUG_ON(!path);
 552
 553         lock_extent(io_tree, ordered_extent->file_offset,
 554                     ordered_extent->file_offset + ordered_extent->len - 1,
 555                     GFP_NOFS);
 556
 557         INIT_LIST_HEAD(&list);
 558
 559         mutex_lock(&BTRFS_I(inode)->extent_mutex);
 560
 561         ret = btrfs_drop_extents(trans, root, inode,
 562                                  ordered_extent->file_offset,
 563                                  ordered_extent->file_offset +
 564                                  ordered_extent->len,
 565                                  ordered_extent->file_offset, &alloc_hint);
 566         BUG_ON(ret);
 567
 568         ins.objectid = inode->i_ino;
 569         ins.offset = ordered_extent->file_offset;
 570         ins.type = BTRFS_EXTENT_DATA_KEY;
 571         ret = btrfs_insert_empty_item(trans, root, path, &ins,
 572                                       sizeof(*extent_item));
 573         BUG_ON(ret);
 574         leaf = path->nodes[0];
 575         extent_item = btrfs_item_ptr(leaf, path->slots[0],
 576                                      struct btrfs_file_extent_item);
 577         btrfs_set_file_extent_generation(leaf, extent_item, trans->transid);
 578         btrfs_set_file_extent_type(leaf, extent_item, BTRFS_FILE_EXTENT_REG);
 579         btrfs_set_file_extent_disk_bytenr(leaf, extent_item,
 580                                           ordered_extent->start);
 581         btrfs_set_file_extent_disk_num_bytes(leaf, extent_item,
 582                                              ordered_extent->len);
 583         btrfs_set_file_extent_offset(leaf, extent_item, 0);
 584         btrfs_set_file_extent_num_bytes(leaf, extent_item,
 585                                         ordered_extent->len);
 586         btrfs_mark_buffer_dirty(leaf);
 587
 588         btrfs_drop_extent_cache(inode, ordered_extent->file_offset,
 589                                 ordered_extent->file_offset +
 590                                 ordered_extent->len - 1);
 591         mutex_unlock(&BTRFS_I(inode)->extent_mutex);
 592
 593         ins.objectid = ordered_extent->start;
 594         ins.offset = ordered_extent->len;
 595         ins.type = BTRFS_EXTENT_ITEM_KEY;
 596         ret = btrfs_alloc_reserved_extent(trans, root, leaf->start,
 597                                           root->root_key.objectid,
 598                                           trans->transid, inode->i_ino,
 599                                           ordered_extent->file_offset, &ins);
 600         BUG_ON(ret);
 601         btrfs_release_path(root, path);
 602
 603         inode->i_blocks += ordered_extent->len >> 9;
 604         unlock_extent(io_tree, ordered_extent->file_offset,
 605                     ordered_extent->file_offset + ordered_extent->len - 1,
 606                     GFP_NOFS);
 607 nocow:
 608         add_pending_csums(trans, inode, ordered_extent->file_offset,
 609                           &ordered_extent->list);
 610
 611         btrfs_ordered_update_i_size(inode, ordered_extent);
 612         btrfs_update_inode(trans, root, inode);
 613         btrfs_remove_ordered_extent(inode, ordered_extent);
 614
 615         /* once for us */
 616         btrfs_put_ordered_extent(ordered_extent);
 617         /* once for the tree */
 618         btrfs_put_ordered_extent(ordered_extent);
 619
 620         btrfs_end_transaction(trans, root);
 621         if (path)
 622                 btrfs_free_path(path);
 623         return 0;
 624 }
 625
 626 int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
 627                                 struct extent_state *state, int uptodate)
 628 {
 629         return btrfs_finish_ordered_io(page->mapping->host, start, end);
 630 }
 631
 632 struct io_failure_record {
 633         struct page *page;
 634         u64 start;
 635         u64 len;
 636         u64 logical;
 637         int last_mirror;
 638 };
 639
 640 int btrfs_io_failed_hook(struct bio *failed_bio,
 641                          struct page *page, u64 start, u64 end,
 642                          struct extent_state *state)
 643 {
 644         struct io_failure_record *failrec = NULL;
 645         u64 private;
 646         struct extent_map *em;
 647         struct inode *inode = page->mapping->host;
 648         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
 649         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 650         struct bio *bio;
 651         int num_copies;
 652         int ret;
 653         int rw;
 654         u64 logical;
 655
 656         ret = get_state_private(failure_tree, start, &private);
 657         if (ret) {
 658                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
 659                 if (!failrec)
 660                         return -ENOMEM;
 661                 failrec->start = start;
 662                 failrec->len = end - start + 1;
 663                 failrec->last_mirror = 0;
 664
 665                 spin_lock(&em_tree->lock);
 666                 em = lookup_extent_mapping(em_tree, start, failrec->len);
 667                 if (em->start > start || em->start + em->len < start) {
 668                         free_extent_map(em);
 669                         em = NULL;
 670                 }
 671                 spin_unlock(&em_tree->lock);
 672
 673                 if (!em || IS_ERR(em)) {
 674                         kfree(failrec);
 675                         return -EIO;
 676                 }
 677                 logical = start - em->start;
 678                 logical = em->block_start + logical;
 679                 failrec->logical = logical;
 680                 free_extent_map(em);
 681                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
 682                                 EXTENT_DIRTY, GFP_NOFS);
 683                 set_state_private(failure_tree, start,
 684                                  (u64)(unsigned long)failrec);
 685         } else {
 686                 failrec = (struct io_failure_record *)(unsigned long)private;
 687         }
 688         num_copies = btrfs_num_copies(
 689                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
 690                               failrec->logical, failrec->len);
 691         failrec->last_mirror++;
 692         if (!state) {
 693                 spin_lock_irq(&BTRFS_I(inode)->io_tree.lock);
 694                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
 695                                                     failrec->start,
 696                                                     EXTENT_LOCKED);
 697                 if (state && state->start != failrec->start)
 698                         state = NULL;
 699                 spin_unlock_irq(&BTRFS_I(inode)->io_tree.lock);
 700         }
 701         if (!state || failrec->last_mirror > num_copies) {
 702                 set_state_private(failure_tree, failrec->start, 0);
 703                 clear_extent_bits(failure_tree, failrec->start,
 704                                   failrec->start + failrec->len - 1,
 705                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
 706                 kfree(failrec);
 707                 return -EIO;
 708         }
 709         bio = bio_alloc(GFP_NOFS, 1);
 710         bio->bi_private = state;
 711         bio->bi_end_io = failed_bio->bi_end_io;
 712         bio->bi_sector = failrec->logical >> 9;
 713         bio->bi_bdev = failed_bio->bi_bdev;
 714         bio->bi_size = 0;
 715         bio_add_page(bio, page, failrec->len, start - page_offset(page));
 716         if (failed_bio->bi_rw & (1 << BIO_RW))
 717                 rw = WRITE;
 718         else
 719                 rw = READ;
 720
 721         BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
 722                                                       failrec->last_mirror);
 723         return 0;
 724 }
 725
 726 int btrfs_clean_io_failures(struct inode *inode, u64 start)
 727 {
 728         u64 private;
 729         u64 private_failure;
 730         struct io_failure_record *failure;
 731         int ret;
 732
 733         private = 0;
 734         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
 735                              (u64)-1, 1, EXTENT_DIRTY)) {
 736                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
 737                                         start, &private_failure);
 738                 if (ret == 0) {
 739                         failure = (struct io_failure_record *)(unsigned long)
 740                                    private_failure;
 741                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
 742                                           failure->start, 0);
 743                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
 744                                           failure->start,
 745                                           failure->start + failure->len - 1,
 746                                           EXTENT_DIRTY | EXTENT_LOCKED,
 747                                           GFP_NOFS);
 748                         kfree(failure);
 749                 }
 750         }
 751         return 0;
 752 }
 753
 754 int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 755                                struct extent_state *state)
 756 {
 757         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
 758         struct inode *inode = page->mapping->host;
 759         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 760         char *kaddr;
 761         u64 private = ~(u32)0;
 762         int ret;
 763         struct btrfs_root *root = BTRFS_I(inode)->root;
 764         u32 csum = ~(u32)0;
 765         unsigned long flags;
 766
 767         if (btrfs_test_opt(root, NODATASUM) ||
 768             btrfs_test_flag(inode, NODATASUM))
 769                 return 0;
 770         if (state && state->start == start) {
 771                 private = state->private;
 772                 ret = 0;
 773         } else {
 774                 ret = get_state_private(io_tree, start, &private);
 775         }
 776         local_irq_save(flags);
 777         kaddr = kmap_atomic(page, KM_IRQ0);
 778         if (ret) {
 779                 goto zeroit;
 780         }
 781         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
 782         btrfs_csum_final(csum, (char *)&csum);
 783         if (csum != private) {
 784                 goto zeroit;
 785         }
 786         kunmap_atomic(kaddr, KM_IRQ0);
 787         local_irq_restore(flags);
 788
 789         /* if the io failure tree for this inode is non-empty,
 790          * check to see if we've recovered from a failed IO
 791          */
 792         btrfs_clean_io_failures(inode, start);
 793         return 0;
 794
 795 zeroit:
 796         printk("btrfs csum failed ino %lu off %llu csum %u private %Lu\n",
 797                page->mapping->host->i_ino, (unsigned long long)start, csum,
 798                private);
 799         memset(kaddr + offset, 1, end - start + 1);
 800         flush_dcache_page(page);
 801         kunmap_atomic(kaddr, KM_IRQ0);
 802         local_irq_restore(flags);
 803         if (private == 0)
 804                 return 0;
 805         return -EIO;
 806 }
 807
 808 /*
 809  * This creates an orphan entry for the given inode in case something goes
 810  * wrong in the middle of an unlink/truncate.
 811  */
 812 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
 813 {
 814         struct btrfs_root *root = BTRFS_I(inode)->root;
 815         int ret = 0;
 816
 817         spin_lock(&root->list_lock);
 818
 819         /* already on the orphan list, we're good */
 820         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
 821                 spin_unlock(&root->list_lock);
 822                 return 0;
 823         }
 824
 825         list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
 826
 827         spin_unlock(&root->list_lock);
 828
 829         /*
 830          * insert an orphan item to track this unlinked/truncated file
 831          */
 832         ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
 833
 834         return ret;
 835 }
 836
 837 /*
 838  * We have done the truncate/delete so we can go ahead and remove the orphan
 839  * item for this particular inode.
 840  */
 841 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
 842 {
 843         struct btrfs_root *root = BTRFS_I(inode)->root;
 844         int ret = 0;
 845
 846         spin_lock(&root->list_lock);
 847
 848         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
 849                 spin_unlock(&root->list_lock);
 850                 return 0;
 851         }
 852
 853         list_del_init(&BTRFS_I(inode)->i_orphan);
 854         if (!trans) {
 855                 spin_unlock(&root->list_lock);
 856                 return 0;
 857         }
 858
 859         spin_unlock(&root->list_lock);
 860
 861         ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
 862
 863         return ret;
 864 }
 865
 866 /*
 867  * this cleans up any orphans that may be left on the list from the last use
 868  * of this root.
 869  */
 870 void btrfs_orphan_cleanup(struct btrfs_root *root)
 871 {
 872         struct btrfs_path *path;
 873         struct extent_buffer *leaf;
 874         struct btrfs_item *item;
 875         struct btrfs_key key, found_key;
 876         struct btrfs_trans_handle *trans;
 877         struct inode *inode;
 878         int ret = 0, nr_unlink = 0, nr_truncate = 0;
 879
 880         /* don't do orphan cleanup if the fs is readonly. */
 881         if (root->inode->i_sb->s_flags & MS_RDONLY)
 882                 return;
 883
 884         path = btrfs_alloc_path();
 885         if (!path)
 886                 return;
 887         path->reada = -1;
 888
 889         key.objectid = BTRFS_ORPHAN_OBJECTID;
 890         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
 891         key.offset = (u64)-1;
 892
 893         trans = btrfs_start_transaction(root, 1);
 894         btrfs_set_trans_block_group(trans, root->inode);
 895
 896         while (1) {
 897                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 898                 if (ret < 0) {
 899                         printk(KERN_ERR "Error searching slot for orphan: %d"
 900                                "\n", ret);
 901                         break;
 902                 }
 903
 904                 /*
 905                  * if ret == 0 means we found what we were searching for, which
 906                  * is weird, but possible, so only screw with path if we didnt
 907                  * find the key and see if we have stuff that matches
 908                  */
 909                 if (ret > 0) {
 910                         if (path->slots[0] == 0)
 911                                 break;
 912                         path->slots[0]--;
 913                 }
 914
 915                 /* pull out the item */
 916                 leaf = path->nodes[0];
 917                 item = btrfs_item_nr(leaf, path->slots[0]);
 918                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 919
 920                 /* make sure the item matches what we want */
 921                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
 922                         break;
 923                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
 924                         break;
 925
 926                 /* release the path since we're done with it */
 927                 btrfs_release_path(root, path);
 928
 929                 /*
 930                  * this is where we are basically btrfs_lookup, without the
 931                  * crossing root thing.  we store the inode number in the
 932                  * offset of the orphan item.
 933                  */
 934                 inode = btrfs_iget_locked(root->inode->i_sb,
 935                                           found_key.offset, root);
 936                 if (!inode)
 937                         break;
 938
 939                 if (inode->i_state & I_NEW) {
 940                         BTRFS_I(inode)->root = root;
 941
 942                         /* have to set the location manually */
 943                         BTRFS_I(inode)->location.objectid = inode->i_ino;
 944                         BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
 945                         BTRFS_I(inode)->location.offset = 0;
 946
 947                         btrfs_read_locked_inode(inode);
 948                         unlock_new_inode(inode);
 949                 }
 950
 951                 /*
 952                  * add this inode to the orphan list so btrfs_orphan_del does
 953                  * the proper thing when we hit it
 954                  */
 955                 spin_lock(&root->list_lock);
 956                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
 957                 spin_unlock(&root->list_lock);
 958
 959                 /*
 960                  * if this is a bad inode, means we actually succeeded in
 961                  * removing the inode, but not the orphan record, which means
 962                  * we need to manually delete the orphan since iput will just
 963                  * do a destroy_inode
 964                  */
 965                 if (is_bad_inode(inode)) {
 966                         btrfs_orphan_del(trans, inode);
 967                         iput(inode);
 968                         continue;
 969                 }
 970
 971                 /* if we have links, this was a truncate, lets do that */
 972                 if (inode->i_nlink) {
 973                         nr_truncate++;
 974                         btrfs_truncate(inode);
 975                 } else {
 976                         nr_unlink++;
 977                 }
 978
 979                 /* this will do delete_inode and everything for us */
 980                 iput(inode);
 981         }
 982
 983         if (nr_unlink)
 984                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
 985         if (nr_truncate)
 986                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
 987
 988         btrfs_free_path(path);
 989         btrfs_end_transaction(trans, root);
 990 }
 991
 992 void btrfs_read_locked_inode(struct inode *inode)
 993 {
 994         struct btrfs_path *path;
 995         struct extent_buffer *leaf;
 996         struct btrfs_inode_item *inode_item;
 997         struct btrfs_timespec *tspec;
 998         struct btrfs_root *root = BTRFS_I(inode)->root;
 999         struct btrfs_key location;
1000         u64 alloc_group_block;
1001         u32 rdev;
1002         int ret;
1003
1004         path = btrfs_alloc_path();
1005         BUG_ON(!path);
1006         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
1007
1008         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
1009         if (ret)
1010                 goto make_bad;
1011
1012         leaf = path->nodes[0];
1013         inode_item = btrfs_item_ptr(leaf, path->slots[0],
1014                                     struct btrfs_inode_item);
1015
1016         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
1017         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
1018         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
1019         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
1020         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
1021
1022         tspec = btrfs_inode_atime(inode_item);
1023         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1024         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1025
1026         tspec = btrfs_inode_mtime(inode_item);
1027         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1028         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1029
1030         tspec = btrfs_inode_ctime(inode_item);
1031         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1032         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1033
1034         inode->i_blocks = btrfs_inode_nblocks(leaf, inode_item);
1035         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
1036         inode->i_generation = BTRFS_I(inode)->generation;
1037         inode->i_rdev = 0;
1038         rdev = btrfs_inode_rdev(leaf, inode_item);
1039
1040         BTRFS_I(inode)->index_cnt = (u64)-1;
1041
1042         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
1043         BTRFS_I(inode)->block_group = btrfs_lookup_block_group(root->fs_info,
1044                                                        alloc_group_block);
1045         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
1046         if (!BTRFS_I(inode)->block_group) {
1047                 BTRFS_I(inode)->block_group = btrfs_find_block_group(root,
1048                                                  NULL, 0,
1049                                                  BTRFS_BLOCK_GROUP_METADATA, 0);
1050         }
1051         btrfs_free_path(path);
1052         inode_item = NULL;
1053
1054         switch (inode->i_mode & S_IFMT) {
1055         case S_IFREG:
1056                 inode->i_mapping->a_ops = &btrfs_aops;
1057                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
1058                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
1059                 inode->i_fop = &btrfs_file_operations;
1060                 inode->i_op = &btrfs_file_inode_operations;
1061                 break;
1062         case S_IFDIR:
1063                 inode->i_fop = &btrfs_dir_file_operations;
1064                 if (root == root->fs_info->tree_root)
1065                         inode->i_op = &btrfs_dir_ro_inode_operations;
1066                 else
1067                         inode->i_op = &btrfs_dir_inode_operations;
1068                 break;
1069         case S_IFLNK:
1070                 inode->i_op = &btrfs_symlink_inode_operations;
1071                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
1072                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
1073                 break;
1074         default:
1075                 init_special_inode(inode, inode->i_mode, rdev);
1076                 break;
1077         }
1078         return;
1079
1080 make_bad:
1081         btrfs_free_path(path);
1082         make_bad_inode(inode);
1083 }
1084
1085 static void fill_inode_item(struct btrfs_trans_handle *trans,
1086                             struct extent_buffer *leaf,
1087                             struct btrfs_inode_item *item,
1088                             struct inode *inode)
1089 {
1090         btrfs_set_inode_uid(leaf, item, inode->i_uid);
1091         btrfs_set_inode_gid(leaf, item, inode->i_gid);
1092         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
1093         btrfs_set_inode_mode(leaf, item, inode->i_mode);
1094         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
1095
1096         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
1097                                inode->i_atime.tv_sec);
1098         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
1099                                 inode->i_atime.tv_nsec);
1100
1101         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
1102                                inode->i_mtime.tv_sec);
1103         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
1104                                 inode->i_mtime.tv_nsec);
1105
1106         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
1107                                inode->i_ctime.tv_sec);
1108         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
1109                                 inode->i_ctime.tv_nsec);
1110
1111         btrfs_set_inode_nblocks(leaf, item, inode->i_blocks);
1112         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
1113         btrfs_set_inode_transid(leaf, item, trans->transid);
1114         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
1115         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
1116         btrfs_set_inode_block_group(leaf, item,
1117                                     BTRFS_I(inode)->block_group->key.objectid);
1118 }
1119
1120 int noinline btrfs_update_inode(struct btrfs_trans_handle *trans,
1121                               struct btrfs_root *root,
1122                               struct inode *inode)
1123 {
1124         struct btrfs_inode_item *inode_item;
1125         struct btrfs_path *path;
1126         struct extent_buffer *leaf;
1127         int ret;
1128
1129         path = btrfs_alloc_path();
1130         BUG_ON(!path);
1131         ret = btrfs_lookup_inode(trans, root, path,
1132                                  &BTRFS_I(inode)->location, 1);
1133         if (ret) {
1134                 if (ret > 0)
1135                         ret = -ENOENT;
1136                 goto failed;
1137         }
1138
1139         leaf = path->nodes[0];
1140         inode_item = btrfs_item_ptr(leaf, path->slots[0],
1141                                   struct btrfs_inode_item);
1142
1143         fill_inode_item(trans, leaf, inode_item, inode);
1144         btrfs_mark_buffer_dirty(leaf);
1145         btrfs_set_inode_last_trans(trans, inode);
1146         ret = 0;
1147 failed:
1148         btrfs_free_path(path);
1149         return ret;
1150 }
1151
1152
1153 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
1154                        struct btrfs_root *root,
1155                        struct inode *dir, struct inode *inode,
1156                        const char *name, int name_len)
1157 {
1158         struct btrfs_path *path;
1159         int ret = 0;
1160         struct extent_buffer *leaf;
1161         struct btrfs_dir_item *di;
1162         struct btrfs_key key;
1163         u64 index;
1164
1165         path = btrfs_alloc_path();
1166         if (!path) {
1167                 ret = -ENOMEM;
1168                 goto err;
1169         }
1170
1171         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
1172                                     name, name_len, -1);
1173         if (IS_ERR(di)) {
1174                 ret = PTR_ERR(di);
1175                 goto err;
1176         }
1177         if (!di) {
1178                 ret = -ENOENT;
1179                 goto err;
1180         }
1181         leaf = path->nodes[0];
1182         btrfs_dir_item_key_to_cpu(leaf, di, &key);
1183         ret = btrfs_delete_one_dir_name(trans, root, path, di);
1184         if (ret)
1185                 goto err;
1186         btrfs_release_path(root, path);
1187
1188         ret = btrfs_del_inode_ref(trans, root, name, name_len,
1189                                   inode->i_ino,
1190                                   dir->i_ino, &index);
1191         if (ret) {
1192                 printk("failed to delete reference to %.*s, "
1193                        "inode %lu parent %lu\n", name_len, name,
1194                        inode->i_ino, dir->i_ino);
1195                 goto err;
1196         }
1197
1198         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
1199                                          index, name, name_len, -1);
1200         if (IS_ERR(di)) {
1201                 ret = PTR_ERR(di);
1202                 goto err;
1203         }
1204         if (!di) {
1205                 ret = -ENOENT;
1206                 goto err;
1207         }
1208         ret = btrfs_delete_one_dir_name(trans, root, path, di);
1209         btrfs_release_path(root, path);
1210
1211         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
1212                                          inode, dir->i_ino);
1213         BUG_ON(ret != 0 && ret != -ENOENT);
1214         if (ret != -ENOENT)
1215                 BTRFS_I(dir)->log_dirty_trans = trans->transid;
1216
1217         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
1218                                            dir, index);
1219         BUG_ON(ret);
1220 err:
1221         btrfs_free_path(path);
1222         if (ret)
1223                 goto out;
1224
1225         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
1226         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
1227         btrfs_update_inode(trans, root, dir);
1228         btrfs_drop_nlink(inode);
1229         ret = btrfs_update_inode(trans, root, inode);
1230         dir->i_sb->s_dirt = 1;
1231 out:
1232         return ret;
1233 }
1234
1235 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
1236 {
1237         struct btrfs_root *root;
1238         struct btrfs_trans_handle *trans;
1239         struct inode *inode = dentry->d_inode;
1240         int ret;
1241         unsigned long nr = 0;
1242
1243         root = BTRFS_I(dir)->root;
1244
1245         ret = btrfs_check_free_space(root, 1, 1);
1246         if (ret)
1247                 goto fail;
1248
1249         trans = btrfs_start_transaction(root, 1);
1250
1251         btrfs_set_trans_block_group(trans, dir);
1252         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
1253                                  dentry->d_name.name, dentry->d_name.len);
1254
1255         if (inode->i_nlink == 0)
1256                 ret = btrfs_orphan_add(trans, inode);
1257
1258         nr = trans->blocks_used;
1259
1260         btrfs_end_transaction_throttle(trans, root);
1261 fail:
1262         btrfs_btree_balance_dirty(root, nr);
1263         return ret;
1264 }
1265
1266 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
1267 {
1268         struct inode *inode = dentry->d_inode;
1269         int err = 0;
1270         int ret;
1271         struct btrfs_root *root = BTRFS_I(dir)->root;
1272         struct btrfs_trans_handle *trans;
1273         unsigned long nr = 0;
1274
1275         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
1276                 return -ENOTEMPTY;
1277         }
1278
1279         ret = btrfs_check_free_space(root, 1, 1);
1280         if (ret)
1281                 goto fail;
1282
1283         trans = btrfs_start_transaction(root, 1);
1284         btrfs_set_trans_block_group(trans, dir);
1285
1286         err = btrfs_orphan_add(trans, inode);
1287         if (err)
1288                 goto fail_trans;
1289
1290         /* now the directory is empty */
1291         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
1292                                  dentry->d_name.name, dentry->d_name.len);
1293         if (!err) {
1294                 btrfs_i_size_write(inode, 0);
1295         }
1296
1297 fail_trans:
1298         nr = trans->blocks_used;
1299         ret = btrfs_end_transaction_throttle(trans, root);
1300 fail:
1301         btrfs_btree_balance_dirty(root, nr);
1302
1303         if (ret && !err)
1304                 err = ret;
1305         return err;
1306 }
1307
1308 /*
1309  * this can truncate away extent items, csum items and directory items.
1310  * It starts at a high offset and removes keys until it can't find
1311  * any higher than i_size.
1312  *
1313  * csum items that cross the new i_size are truncated to the new size
1314  * as well.
1315  *
1316  * min_type is the minimum key type to truncate down to.  If set to 0, this
1317  * will kill all the items on this inode, including the INODE_ITEM_KEY.
1318  */
1319 noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
1320                                         struct btrfs_root *root,
1321                                         struct inode *inode,
1322                                         u64 new_size, u32 min_type)
1323 {
1324         int ret;
1325         struct btrfs_path *path;
1326         struct btrfs_key key;
1327         struct btrfs_key found_key;
1328         u32 found_type;
1329         struct extent_buffer *leaf;
1330         struct btrfs_file_extent_item *fi;
1331         u64 extent_start = 0;
1332         u64 extent_num_bytes = 0;
1333         u64 item_end = 0;
1334         u64 root_gen = 0;
1335         u64 root_owner = 0;
1336         int found_extent;
1337         int del_item;
1338         int pending_del_nr = 0;
1339         int pending_del_slot = 0;
1340         int extent_type = -1;
1341         u64 mask = root->sectorsize - 1;
1342
1343         if (root->ref_cows)
1344                 btrfs_drop_extent_cache(inode,
1345                                         new_size & (~mask), (u64)-1);
1346         path = btrfs_alloc_path();
1347         path->reada = -1;
1348         BUG_ON(!path);
1349
1350         /* FIXME, add redo link to tree so we don't leak on crash */
1351         key.objectid = inode->i_ino;
1352         key.offset = (u64)-1;
1353         key.type = (u8)-1;
1354
1355         btrfs_init_path(path);
1356 search_again:
1357         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1358         if (ret < 0) {
1359                 goto error;
1360         }
1361         if (ret > 0) {
1362                 /* there are no items in the tree for us to truncate, we're
1363                  * done
1364                  */
1365                 if (path->slots[0] == 0) {
1366                         ret = 0;
1367                         goto error;
1368                 }
1369                 path->slots[0]--;
1370         }
1371
1372         while(1) {
1373                 fi = NULL;
1374                 leaf = path->nodes[0];
1375                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1376                 found_type = btrfs_key_type(&found_key);
1377
1378                 if (found_key.objectid != inode->i_ino)
1379                         break;
1380
1381                 if (found_type < min_type)
1382                         break;
1383
1384                 item_end = found_key.offset;
1385                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
1386                         fi = btrfs_item_ptr(leaf, path->slots[0],
1387                                             struct btrfs_file_extent_item);
1388                         extent_type = btrfs_file_extent_type(leaf, fi);
1389                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
1390                                 item_end +=
1391                                     btrfs_file_extent_num_bytes(leaf, fi);
1392                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1393                                 struct btrfs_item *item = btrfs_item_nr(leaf,
1394                                                                 path->slots[0]);
1395                                 item_end += btrfs_file_extent_inline_len(leaf,
1396                                                                          item);
1397                         }
1398                         item_end--;
1399                 }
1400                 if (found_type == BTRFS_CSUM_ITEM_KEY) {
1401                         ret = btrfs_csum_truncate(trans, root, path,
1402                                                   new_size);
1403                         BUG_ON(ret);
1404                 }
1405                 if (item_end < new_size) {
1406                         if (found_type == BTRFS_DIR_ITEM_KEY) {
1407                                 found_type = BTRFS_INODE_ITEM_KEY;
1408                         } else if (found_type == BTRFS_EXTENT_ITEM_KEY) {
1409                                 found_type = BTRFS_CSUM_ITEM_KEY;
1410                         } else if (found_type == BTRFS_EXTENT_DATA_KEY) {
1411                                 found_type = BTRFS_XATTR_ITEM_KEY;
1412                         } else if (found_type == BTRFS_XATTR_ITEM_KEY) {
1413                                 found_type = BTRFS_INODE_REF_KEY;
1414                         } else if (found_type) {
1415                                 found_type--;
1416                         } else {
1417                                 break;
1418                         }
1419                         btrfs_set_key_type(&key, found_type);
1420                         goto next;
1421                 }
1422                 if (found_key.offset >= new_size)
1423                         del_item = 1;
1424                 else
1425                         del_item = 0;
1426                 found_extent = 0;
1427
1428                 /* FIXME, shrink the extent if the ref count is only 1 */
1429                 if (found_type != BTRFS_EXTENT_DATA_KEY)
1430                         goto delete;
1431
1432                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
1433                         u64 num_dec;
1434                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
1435                         if (!del_item) {
1436                                 u64 orig_num_bytes =
1437                                         btrfs_file_extent_num_bytes(leaf, fi);
1438                                 extent_num_bytes = new_size -
1439                                         found_key.offset + root->sectorsize - 1;
1440                                 extent_num_bytes = extent_num_bytes &
1441                                         ~((u64)root->sectorsize - 1);
1442                                 btrfs_set_file_extent_num_bytes(leaf, fi,
1443                                                          extent_num_bytes);
1444                                 num_dec = (orig_num_bytes -
1445                                            extent_num_bytes);
1446                                 if (root->ref_cows && extent_start != 0)
1447                                         dec_i_blocks(inode, num_dec);
1448                                 btrfs_mark_buffer_dirty(leaf);
1449                         } else {
1450                                 extent_num_bytes =
1451                                         btrfs_file_extent_disk_num_bytes(leaf,
1452                                                                          fi);
1453                                 /* FIXME blocksize != 4096 */
1454                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
1455                                 if (extent_start != 0) {
1456                                         found_extent = 1;
1457                                         if (root->ref_cows)
1458                                                 dec_i_blocks(inode, num_dec);
1459                                 }
1460                                 root_gen = btrfs_header_generation(leaf);
1461                                 root_owner = btrfs_header_owner(leaf);
1462                         }
1463                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1464                         if (!del_item) {
1465                                 u32 size = new_size - found_key.offset;
1466
1467                                 if (root->ref_cows) {
1468                                         dec_i_blocks(inode, item_end + 1 -
1469                                                     found_key.offset - size);
1470                                 }
1471                                 size =
1472                                     btrfs_file_extent_calc_inline_size(size);
1473                                 ret = btrfs_truncate_item(trans, root, path,
1474                                                           size, 1);
1475                                 BUG_ON(ret);
1476                         } else if (root->ref_cows) {
1477                                 dec_i_blocks(inode, item_end + 1 -
1478                                              found_key.offset);
1479                         }
1480                 }
1481 delete:
1482                 if (del_item) {
1483                         if (!pending_del_nr) {
1484                                 /* no pending yet, add ourselves */
1485                                 pending_del_slot = path->slots[0];
1486                                 pending_del_nr = 1;
1487                         } else if (pending_del_nr &&
1488                                    path->slots[0] + 1 == pending_del_slot) {
1489                                 /* hop on the pending chunk */
1490                                 pending_del_nr++;
1491                                 pending_del_slot = path->slots[0];
1492                         } else {
1493                                 printk("bad pending slot %d pending_del_nr %d pending_del_slot %d\n", path->slots[0], pending_del_nr, pending_del_slot);
1494                         }
1495                 } else {
1496                         break;
1497                 }
1498                 if (found_extent) {
1499                         ret = btrfs_free_extent(trans, root, extent_start,
1500                                                 extent_num_bytes,
1501                                                 leaf->start, root_owner,
1502                                                 root_gen, inode->i_ino,
1503                                                 found_key.offset, 0);
1504                         BUG_ON(ret);
1505                 }
1506 next:
1507                 if (path->slots[0] == 0) {
1508                         if (pending_del_nr)
1509                                 goto del_pending;
1510                         btrfs_release_path(root, path);
1511                         goto search_again;
1512                 }
1513
1514                 path->slots[0]--;
1515                 if (pending_del_nr &&
1516                     path->slots[0] + 1 != pending_del_slot) {
1517                         struct btrfs_key debug;
1518 del_pending:
1519                         btrfs_item_key_to_cpu(path->nodes[0], &debug,
1520                                               pending_del_slot);
1521                         ret = btrfs_del_items(trans, root, path,
1522                                               pending_del_slot,
1523                                               pending_del_nr);
1524                         BUG_ON(ret);
1525                         pending_del_nr = 0;
1526                         btrfs_release_path(root, path);
1527                         goto search_again;
1528                 }
1529         }
1530         ret = 0;
1531 error:
1532         if (pending_del_nr) {
1533                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
1534                                       pending_del_nr);
1535         }
1536         btrfs_free_path(path);
1537         inode->i_sb->s_dirt = 1;
1538         return ret;
1539 }
1540
1541 /*
1542  * taken from block_truncate_page, but does cow as it zeros out
1543  * any bytes left in the last page in the file.
1544  */
1545 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
1546 {
1547         struct inode *inode = mapping->host;
1548         struct btrfs_root *root = BTRFS_I(inode)->root;
1549         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1550         struct btrfs_ordered_extent *ordered;
1551         char *kaddr;
1552         u32 blocksize = root->sectorsize;
1553         pgoff_t index = from >> PAGE_CACHE_SHIFT;
1554         unsigned offset = from & (PAGE_CACHE_SIZE-1);
1555         struct page *page;
1556         int ret = 0;
1557         u64 page_start;
1558         u64 page_end;
1559
1560         if ((offset & (blocksize - 1)) == 0)
1561                 goto out;
1562
1563         ret = -ENOMEM;
1564 again:
1565         page = grab_cache_page(mapping, index);
1566         if (!page)
1567                 goto out;
1568
1569         page_start = page_offset(page);
1570         page_end = page_start + PAGE_CACHE_SIZE - 1;
1571
1572         if (!PageUptodate(page)) {
1573                 ret = btrfs_readpage(NULL, page);
1574                 lock_page(page);
1575                 if (page->mapping != mapping) {
1576                         unlock_page(page);
1577                         page_cache_release(page);
1578                         goto again;
1579                 }
1580                 if (!PageUptodate(page)) {
1581                         ret = -EIO;
1582                         goto out_unlock;
1583                 }
1584         }
1585         wait_on_page_writeback(page);
1586
1587         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
1588         set_page_extent_mapped(page);
1589
1590         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1591         if (ordered) {
1592                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
1593                 unlock_page(page);
1594                 page_cache_release(page);
1595                 btrfs_start_ordered_extent(inode, ordered, 1);
1596                 btrfs_put_ordered_extent(ordered);
1597                 goto again;
1598         }
1599
1600         btrfs_set_extent_delalloc(inode, page_start, page_end);
1601         ret = 0;
1602         if (offset != PAGE_CACHE_SIZE) {
1603                 kaddr = kmap(page);
1604                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
1605                 flush_dcache_page(page);
1606                 kunmap(page);
1607         }
1608         ClearPageChecked(page);
1609         set_page_dirty(page);
1610         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
1611
1612 out_unlock:
1613         unlock_page(page);
1614         page_cache_release(page);
1615 out:
1616         return ret;
1617 }
1618
1619 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
1620 {
1621         struct inode *inode = dentry->d_inode;
1622         int err;
1623
1624         err = inode_change_ok(inode, attr);
1625         if (err)
1626                 return err;
1627
1628         if (S_ISREG(inode->i_mode) &&
1629             attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
1630                 struct btrfs_trans_handle *trans;
1631                 struct btrfs_root *root = BTRFS_I(inode)->root;
1632                 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1633
1634                 u64 mask = root->sectorsize - 1;
1635                 u64 hole_start = (inode->i_size + mask) & ~mask;
1636                 u64 block_end = (attr->ia_size + mask) & ~mask;
1637                 u64 hole_size;
1638                 u64 alloc_hint = 0;
1639
1640                 if (attr->ia_size <= hole_start)
1641                         goto out;
1642
1643                 err = btrfs_check_free_space(root, 1, 0);
1644                 if (err)
1645                         goto fail;
1646
1647                 btrfs_truncate_page(inode->i_mapping, inode->i_size);
1648
1649                 hole_size = block_end - hole_start;
1650                 while(1) {
1651                         struct btrfs_ordered_extent *ordered;
1652                         btrfs_wait_ordered_range(inode, hole_start, hole_size);
1653
1654                         lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
1655                         ordered = btrfs_lookup_ordered_extent(inode, hole_start);
1656                         if (ordered) {
1657                                 unlock_extent(io_tree, hole_start,
1658                                               block_end - 1, GFP_NOFS);
1659                                 btrfs_put_ordered_extent(ordered);
1660                         } else {
1661                                 break;
1662                         }
1663                 }
1664
1665                 trans = btrfs_start_transaction(root, 1);
1666                 btrfs_set_trans_block_group(trans, inode);
1667                 mutex_lock(&BTRFS_I(inode)->extent_mutex);
1668                 err = btrfs_drop_extents(trans, root, inode,
1669                                          hole_start, block_end, hole_start,
1670                                          &alloc_hint);
1671
1672                 if (alloc_hint != EXTENT_MAP_INLINE) {
1673                         err = btrfs_insert_file_extent(trans, root,
1674                                                        inode->i_ino,
1675                                                        hole_start, 0, 0,
1676                                                        hole_size, 0);
1677                         btrfs_drop_extent_cache(inode, hole_start,
1678                                                 (u64)-1);
1679                         btrfs_check_file(root, inode);
1680                 }
1681                 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
1682                 btrfs_end_transaction(trans, root);
1683                 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
1684                 if (err)
1685                         return err;
1686         }
1687 out:
1688         err = inode_setattr(inode, attr);
1689
1690         if (!err && ((attr->ia_valid & ATTR_MODE)))
1691                 err = btrfs_acl_chmod(inode);
1692 fail:
1693         return err;
1694 }
1695
1696 void btrfs_delete_inode(struct inode *inode)
1697 {
1698         struct btrfs_trans_handle *trans;
1699         struct btrfs_root *root = BTRFS_I(inode)->root;
1700         unsigned long nr;
1701         int ret;
1702
1703         truncate_inode_pages(&inode->i_data, 0);
1704         if (is_bad_inode(inode)) {
1705                 btrfs_orphan_del(NULL, inode);
1706                 goto no_delete;
1707         }
1708         btrfs_wait_ordered_range(inode, 0, (u64)-1);
1709
1710         btrfs_i_size_write(inode, 0);
1711         trans = btrfs_start_transaction(root, 1);
1712
1713         btrfs_set_trans_block_group(trans, inode);
1714         ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0);
1715         if (ret) {
1716                 btrfs_orphan_del(NULL, inode);
1717                 goto no_delete_lock;
1718         }
1719
1720         btrfs_orphan_del(trans, inode);
1721
1722         nr = trans->blocks_used;
1723         clear_inode(inode);
1724
1725         btrfs_end_transaction(trans, root);
1726         btrfs_btree_balance_dirty(root, nr);
1727         return;
1728
1729 no_delete_lock:
1730         nr = trans->blocks_used;
1731         btrfs_end_transaction(trans, root);
1732         btrfs_btree_balance_dirty(root, nr);
1733 no_delete:
1734         clear_inode(inode);
1735 }
1736
1737 /*
1738  * this returns the key found in the dir entry in the location pointer.
1739  * If no dir entries were found, location->objectid is 0.
1740  */
1741 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
1742                                struct btrfs_key *location)
1743 {
1744         const char *name = dentry->d_name.name;
1745         int namelen = dentry->d_name.len;
1746         struct btrfs_dir_item *di;
1747         struct btrfs_path *path;
1748         struct btrfs_root *root = BTRFS_I(dir)->root;
1749         int ret = 0;
1750
1751         path = btrfs_alloc_path();
1752         BUG_ON(!path);
1753
1754         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
1755                                     namelen, 0);
1756         if (IS_ERR(di))
1757                 ret = PTR_ERR(di);
1758         if (!di || IS_ERR(di)) {
1759                 goto out_err;
1760         }
1761         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
1762 out:
1763         btrfs_free_path(path);
1764         return ret;
1765 out_err:
1766         location->objectid = 0;
1767         goto out;
1768 }
1769
1770 /*
1771  * when we hit a tree root in a directory, the btrfs part of the inode
1772  * needs to be changed to reflect the root directory of the tree root.  This
1773  * is kind of like crossing a mount point.
1774  */
1775 static int fixup_tree_root_location(struct btrfs_root *root,
1776                              struct btrfs_key *location,
1777                              struct btrfs_root **sub_root,
1778                              struct dentry *dentry)
1779 {
1780         struct btrfs_root_item *ri;
1781
1782         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
1783                 return 0;
1784         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1785                 return 0;
1786
1787         *sub_root = btrfs_read_fs_root(root->fs_info, location,
1788                                         dentry->d_name.name,
1789                                         dentry->d_name.len);
1790         if (IS_ERR(*sub_root))
1791                 return PTR_ERR(*sub_root);
1792
1793         ri = &(*sub_root)->root_item;
1794         location->objectid = btrfs_root_dirid(ri);
1795         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1796         location->offset = 0;
1797
1798         return 0;
1799 }
1800
1801 static noinline void init_btrfs_i(struct inode *inode)
1802 {
1803         struct btrfs_inode *bi = BTRFS_I(inode);
1804
1805         bi->i_acl = NULL;
1806         bi->i_default_acl = NULL;
1807
1808         bi->generation = 0;
1809         bi->last_trans = 0;
1810         bi->logged_trans = 0;
1811         bi->delalloc_bytes = 0;
1812         bi->disk_i_size = 0;
1813         bi->flags = 0;
1814         bi->index_cnt = (u64)-1;
1815         bi->log_dirty_trans = 0;
1816         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
1817         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
1818                              inode->i_mapping, GFP_NOFS);
1819         extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
1820                              inode->i_mapping, GFP_NOFS);
1821         INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
1822         btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
1823         mutex_init(&BTRFS_I(inode)->csum_mutex);
1824         mutex_init(&BTRFS_I(inode)->extent_mutex);
1825         mutex_init(&BTRFS_I(inode)->log_mutex);
1826 }
1827
1828 static int btrfs_init_locked_inode(struct inode *inode, void *p)
1829 {
1830         struct btrfs_iget_args *args = p;
1831         inode->i_ino = args->ino;
1832         init_btrfs_i(inode);
1833         BTRFS_I(inode)->root = args->root;
1834         return 0;
1835 }
1836
1837 static int btrfs_find_actor(struct inode *inode, void *opaque)
1838 {
1839         struct btrfs_iget_args *args = opaque;
1840         return (args->ino == inode->i_ino &&
1841                 args->root == BTRFS_I(inode)->root);
1842 }
1843
1844 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
1845                                 struct btrfs_root *root)
1846 {
1847         struct inode *inode;
1848         struct btrfs_iget_args args;
1849         args.ino = objectid;
1850         args.root = root;
1851
1852         inode = iget5_locked(s, objectid, btrfs_find_actor,
1853                              btrfs_init_locked_inode,
1854                              (void *)&args);
1855         return inode;
1856 }
1857
1858 /* Get an inode object given its location and corresponding root.
1859  * Returns in *is_new if the inode was read from disk
1860  */
1861 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
1862                          struct btrfs_root *root, int *is_new)
1863 {
1864         struct inode *inode;
1865
1866         inode = btrfs_iget_locked(s, location->objectid, root);
1867         if (!inode)
1868                 return ERR_PTR(-EACCES);
1869
1870         if (inode->i_state & I_NEW) {
1871                 BTRFS_I(inode)->root = root;
1872                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
1873                 btrfs_read_locked_inode(inode);
1874                 unlock_new_inode(inode);
1875                 if (is_new)
1876                         *is_new = 1;
1877         } else {
1878                 if (is_new)
1879                         *is_new = 0;
1880         }
1881
1882         return inode;
1883 }
1884
1885 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
1886                                    struct nameidata *nd)
1887 {
1888         struct inode * inode;
1889         struct btrfs_inode *bi = BTRFS_I(dir);
1890         struct btrfs_root *root = bi->root;
1891         struct btrfs_root *sub_root = root;
1892         struct btrfs_key location;
1893         int ret, new, do_orphan = 0;
1894
1895         if (dentry->d_name.len > BTRFS_NAME_LEN)
1896                 return ERR_PTR(-ENAMETOOLONG);
1897
1898         ret = btrfs_inode_by_name(dir, dentry, &location);
1899
1900         if (ret < 0)
1901                 return ERR_PTR(ret);
1902
1903         inode = NULL;
1904         if (location.objectid) {
1905                 ret = fixup_tree_root_location(root, &location, &sub_root,
1906                                                 dentry);
1907                 if (ret < 0)
1908                         return ERR_PTR(ret);
1909                 if (ret > 0)
1910                         return ERR_PTR(-ENOENT);
1911                 inode = btrfs_iget(dir->i_sb, &location, sub_root, &new);
1912                 if (IS_ERR(inode))
1913                         return ERR_CAST(inode);
1914
1915                 /* the inode and parent dir are two different roots */
1916                 if (new && root != sub_root) {
1917                         igrab(inode);
1918                         sub_root->inode = inode;
1919                         do_orphan = 1;
1920                 }
1921         }
1922
1923         if (unlikely(do_orphan))
1924                 btrfs_orphan_cleanup(sub_root);
1925
1926         return d_splice_alias(inode, dentry);
1927 }
1928
1929 static unsigned char btrfs_filetype_table[] = {
1930         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
1931 };
1932
1933 static int btrfs_real_readdir(struct file *filp, void *dirent,
1934                               filldir_t filldir)
1935 {
1936         struct inode *inode = filp->f_dentry->d_inode;
1937         struct btrfs_root *root = BTRFS_I(inode)->root;
1938         struct btrfs_item *item;
1939         struct btrfs_dir_item *di;
1940         struct btrfs_key key;
1941         struct btrfs_key found_key;
1942         struct btrfs_path *path;
1943         int ret;
1944         u32 nritems;
1945         struct extent_buffer *leaf;
1946         int slot;
1947         int advance;
1948         unsigned char d_type;
1949         int over = 0;
1950         u32 di_cur;
1951         u32 di_total;
1952         u32 di_len;
1953         int key_type = BTRFS_DIR_INDEX_KEY;
1954         char tmp_name[32];
1955         char *name_ptr;
1956         int name_len;
1957
1958         /* FIXME, use a real flag for deciding about the key type */
1959         if (root->fs_info->tree_root == root)
1960                 key_type = BTRFS_DIR_ITEM_KEY;
1961
1962         /* special case for "." */
1963         if (filp->f_pos == 0) {
1964                 over = filldir(dirent, ".", 1,
1965                                1, inode->i_ino,
1966                                DT_DIR);
1967                 if (over)
1968                         return 0;
1969                 filp->f_pos = 1;
1970         }
1971         /* special case for .., just use the back ref */
1972         if (filp->f_pos == 1) {
1973                 u64 pino = parent_ino(filp->f_path.dentry);
1974                 over = filldir(dirent, "..", 2,
1975                                2, pino, DT_DIR);
1976                 if (over)
1977                         return 0;
1978                 filp->f_pos = 2;
1979         }
1980
1981         path = btrfs_alloc_path();
1982         path->reada = 2;
1983
1984         btrfs_set_key_type(&key, key_type);
1985         key.offset = filp->f_pos;
1986         key.objectid = inode->i_ino;
1987
1988         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1989         if (ret < 0)
1990                 goto err;
1991         advance = 0;
1992
1993         while (1) {
1994                 leaf = path->nodes[0];
1995                 nritems = btrfs_header_nritems(leaf);
1996                 slot = path->slots[0];
1997                 if (advance || slot >= nritems) {
1998                         if (slot >= nritems - 1) {
1999                                 ret = btrfs_next_leaf(root, path);
2000                                 if (ret)
2001                                         break;
2002                                 leaf = path->nodes[0];
2003                                 nritems = btrfs_header_nritems(leaf);
2004                                 slot = path->slots[0];
2005                         } else {
2006                                 slot++;
2007                                 path->slots[0]++;
2008                         }
2009                 }
2010                 advance = 1;
2011                 item = btrfs_item_nr(leaf, slot);
2012                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2013
2014                 if (found_key.objectid != key.objectid)
2015                         break;
2016                 if (btrfs_key_type(&found_key) != key_type)
2017                         break;
2018                 if (found_key.offset < filp->f_pos)
2019                         continue;
2020
2021                 filp->f_pos = found_key.offset;
2022
2023                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
2024                 di_cur = 0;
2025                 di_total = btrfs_item_size(leaf, item);
2026
2027                 while (di_cur < di_total) {
2028                         struct btrfs_key location;
2029
2030                         name_len = btrfs_dir_name_len(leaf, di);
2031                         if (name_len <= sizeof(tmp_name)) {
2032                                 name_ptr = tmp_name;
2033                         } else {
2034                                 name_ptr = kmalloc(name_len, GFP_NOFS);
2035                                 if (!name_ptr) {
2036                                         ret = -ENOMEM;
2037                                         goto err;
2038                                 }
2039                         }
2040                         read_extent_buffer(leaf, name_ptr,
2041                                            (unsigned long)(di + 1), name_len);
2042
2043                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
2044                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
2045                         over = filldir(dirent, name_ptr, name_len,
2046                                        found_key.offset, location.objectid,
2047                                        d_type);
2048
2049                         if (name_ptr != tmp_name)
2050                                 kfree(name_ptr);
2051
2052                         if (over)
2053                                 goto nopos;
2054
2055                         di_len = btrfs_dir_name_len(leaf, di) +
2056                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
2057                         di_cur += di_len;
2058                         di = (struct btrfs_dir_item *)((char *)di + di_len);
2059                 }
2060         }
2061
2062         /* Reached end of directory/root. Bump pos past the last item. */
2063         if (key_type == BTRFS_DIR_INDEX_KEY)
2064                 filp->f_pos = INT_LIMIT(typeof(filp->f_pos));
2065         else
2066                 filp->f_pos++;
2067 nopos:
2068         ret = 0;
2069 err:
2070         btrfs_free_path(path);
2071         return ret;
2072 }
2073
2074 /* Kernels earlier than 2.6.28 still have the NFS deadlock where nfsd
2075    will call the file system's ->lookup() method from within its
2076    filldir callback, which in turn was called from the file system's
2077    ->readdir() method. And will deadlock for many file systems. */
2078 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,28)
2079
2080 struct nfshack_dirent {
2081         u64             ino;
2082         loff_t          offset;
2083         int             namlen;
2084         unsigned int    d_type;
2085         char            name[];
2086 };
2087
2088 struct nfshack_readdir {
2089         char            *dirent;
2090         size_t          used;
2091         int             full;
2092 };
2093
2094
2095
2096 static int btrfs_nfshack_filldir(void *__buf, const char *name, int namlen,
2097                               loff_t offset, u64 ino, unsigned int d_type)
2098 {
2099         struct nfshack_readdir *buf = __buf;
2100         struct nfshack_dirent *de = (void *)(buf->dirent + buf->used);
2101         unsigned int reclen;
2102
2103         reclen = ALIGN(sizeof(struct nfshack_dirent) + namlen, sizeof(u64));
2104         if (buf->used + reclen > PAGE_SIZE) {
2105                 buf->full = 1;
2106                 return -EINVAL;
2107         }
2108
2109         de->namlen = namlen;
2110         de->offset = offset;
2111         de->ino = ino;
2112         de->d_type = d_type;
2113         memcpy(de->name, name, namlen);
2114         buf->used += reclen;
2115
2116         return 0;
2117 }
2118
2119 static int btrfs_nfshack_readdir(struct file *file, void *dirent,
2120                                  filldir_t filldir)
2121 {
2122         struct nfshack_readdir buf;
2123         struct nfshack_dirent *de;
2124         int err;
2125         int size;
2126         loff_t offset;
2127
2128         buf.dirent = (void *)__get_free_page(GFP_KERNEL);
2129         if (!buf.dirent)
2130                 return -ENOMEM;
2131
2132         offset = file->f_pos;
2133
2134         do {
2135                 unsigned int reclen;
2136
2137                 buf.used = 0;
2138                 buf.full = 0;
2139                 err = btrfs_real_readdir(file, &buf, btrfs_nfshack_filldir);
2140                 if (err)
2141                         break;
2142
2143                 size = buf.used;
2144
2145                 if (!size)
2146                         break;
2147
2148                 de = (struct nfshack_dirent *)buf.dirent;
2149                 while (size > 0) {
2150                         offset = de->offset;
2151
2152                         if (filldir(dirent, de->name, de->namlen, de->offset,
2153                                     de->ino, de->d_type))
2154                                 goto done;
2155                         offset = file->f_pos;
2156
2157                         reclen = ALIGN(sizeof(*de) + de->namlen,
2158                                        sizeof(u64));
2159                         size -= reclen;
2160                         de = (struct nfshack_dirent *)((char *)de + reclen);
2161                 }
2162         } while (buf.full);
2163
2164  done:
2165         free_page((unsigned long)buf.dirent);
2166         file->f_pos = offset;
2167
2168         return err;
2169 }
2170 #endif
2171
2172 int btrfs_write_inode(struct inode *inode, int wait)
2173 {
2174         struct btrfs_root *root = BTRFS_I(inode)->root;
2175         struct btrfs_trans_handle *trans;
2176         int ret = 0;
2177
2178         if (root->fs_info->closing > 1)
2179                 return 0;
2180
2181         if (wait) {
2182                 trans = btrfs_join_transaction(root, 1);
2183                 btrfs_set_trans_block_group(trans, inode);
2184                 ret = btrfs_commit_transaction(trans, root);
2185         }
2186         return ret;
2187 }
2188
2189 /*
2190  * This is somewhat expensive, updating the tree every time the
2191  * inode changes.  But, it is most likely to find the inode in cache.
2192  * FIXME, needs more benchmarking...there are no reasons other than performance
2193  * to keep or drop this code.
2194  */
2195 void btrfs_dirty_inode(struct inode *inode)
2196 {
2197         struct btrfs_root *root = BTRFS_I(inode)->root;
2198         struct btrfs_trans_handle *trans;
2199
2200         trans = btrfs_join_transaction(root, 1);
2201         btrfs_set_trans_block_group(trans, inode);
2202         btrfs_update_inode(trans, root, inode);
2203         btrfs_end_transaction(trans, root);
2204 }
2205
2206 static int btrfs_set_inode_index_count(struct inode *inode)
2207 {
2208         struct btrfs_root *root = BTRFS_I(inode)->root;
2209         struct btrfs_key key, found_key;
2210         struct btrfs_path *path;
2211         struct extent_buffer *leaf;
2212         int ret;
2213
2214         key.objectid = inode->i_ino;
2215         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
2216         key.offset = (u64)-1;
2217
2218         path = btrfs_alloc_path();
2219         if (!path)
2220                 return -ENOMEM;
2221
2222         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2223         if (ret < 0)
2224                 goto out;
2225         /* FIXME: we should be able to handle this */
2226         if (ret == 0)
2227                 goto out;
2228         ret = 0;
2229
2230         /*
2231          * MAGIC NUMBER EXPLANATION:
2232          * since we search a directory based on f_pos we have to start at 2
2233          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
2234          * else has to start at 2
2235          */
2236         if (path->slots[0] == 0) {
2237                 BTRFS_I(inode)->index_cnt = 2;
2238                 goto out;
2239         }
2240
2241         path->slots[0]--;
2242
2243         leaf = path->nodes[0];
2244         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2245
2246         if (found_key.objectid != inode->i_ino ||
2247             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
2248                 BTRFS_I(inode)->index_cnt = 2;
2249                 goto out;
2250         }
2251
2252         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
2253 out:
2254         btrfs_free_path(path);
2255         return ret;
2256 }
2257
2258 static int btrfs_set_inode_index(struct inode *dir, struct inode *inode,
2259                                  u64 *index)
2260 {
2261         int ret = 0;
2262
2263         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
2264                 ret = btrfs_set_inode_index_count(dir);
2265                 if (ret) {
2266                         return ret;
2267                 }
2268         }
2269
2270         *index = BTRFS_I(dir)->index_cnt;
2271         BTRFS_I(dir)->index_cnt++;
2272
2273         return ret;
2274 }
2275
2276 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
2277                                      struct btrfs_root *root,
2278                                      struct inode *dir,
2279                                      const char *name, int name_len,
2280                                      u64 ref_objectid,
2281                                      u64 objectid,
2282                                      struct btrfs_block_group_cache *group,
2283                                      int mode, u64 *index)
2284 {
2285         struct inode *inode;
2286         struct btrfs_inode_item *inode_item;
2287         struct btrfs_block_group_cache *new_inode_group;
2288         struct btrfs_key *location;
2289         struct btrfs_path *path;
2290         struct btrfs_inode_ref *ref;
2291         struct btrfs_key key[2];
2292         u32 sizes[2];
2293         unsigned long ptr;
2294         int ret;
2295         int owner;
2296
2297         path = btrfs_alloc_path();
2298         BUG_ON(!path);
2299
2300         inode = new_inode(root->fs_info->sb);
2301         if (!inode)
2302                 return ERR_PTR(-ENOMEM);
2303
2304         if (dir) {
2305                 ret = btrfs_set_inode_index(dir, inode, index);
2306                 if (ret)
2307                         return ERR_PTR(ret);
2308         }
2309         /*
2310          * index_cnt is ignored for everything but a dir,
2311          * btrfs_get_inode_index_count has an explanation for the magic
2312          * number
2313          */
2314         init_btrfs_i(inode);
2315         BTRFS_I(inode)->index_cnt = 2;
2316         BTRFS_I(inode)->root = root;
2317         BTRFS_I(inode)->generation = trans->transid;
2318
2319         if (mode & S_IFDIR)
2320                 owner = 0;
2321         else
2322                 owner = 1;
2323         new_inode_group = btrfs_find_block_group(root, group, 0,
2324                                        BTRFS_BLOCK_GROUP_METADATA, owner);
2325         if (!new_inode_group) {
2326                 printk("find_block group failed\n");
2327                 new_inode_group = group;
2328         }
2329         BTRFS_I(inode)->block_group = new_inode_group;
2330
2331         key[0].objectid = objectid;
2332         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
2333         key[0].offset = 0;
2334
2335         key[1].objectid = objectid;
2336         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
2337         key[1].offset = ref_objectid;
2338
2339         sizes[0] = sizeof(struct btrfs_inode_item);
2340         sizes[1] = name_len + sizeof(*ref);
2341
2342         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
2343         if (ret != 0)
2344                 goto fail;
2345
2346         if (objectid > root->highest_inode)
2347                 root->highest_inode = objectid;
2348
2349         inode->i_uid = current->fsuid;
2350         inode->i_gid = current->fsgid;
2351         inode->i_mode = mode;
2352         inode->i_ino = objectid;
2353         inode->i_blocks = 0;
2354         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2355         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2356                                   struct btrfs_inode_item);
2357         fill_inode_item(trans, path->nodes[0], inode_item, inode);
2358
2359         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
2360                              struct btrfs_inode_ref);
2361         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
2362         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
2363         ptr = (unsigned long)(ref + 1);
2364         write_extent_buffer(path->nodes[0], name, ptr, name_len);
2365
2366         btrfs_mark_buffer_dirty(path->nodes[0]);
2367         btrfs_free_path(path);
2368
2369         location = &BTRFS_I(inode)->location;
2370         location->objectid = objectid;
2371         location->offset = 0;
2372         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
2373
2374         insert_inode_hash(inode);
2375         return inode;
2376 fail:
2377         if (dir)
2378                 BTRFS_I(dir)->index_cnt--;
2379         btrfs_free_path(path);
2380         return ERR_PTR(ret);
2381 }
2382
2383 static inline u8 btrfs_inode_type(struct inode *inode)
2384 {
2385         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
2386 }
2387
2388 int btrfs_add_link(struct btrfs_trans_handle *trans,
2389                    struct inode *parent_inode, struct inode *inode,
2390                    const char *name, int name_len, int add_backref, u64 index)
2391 {
2392         int ret;
2393         struct btrfs_key key;
2394         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
2395
2396         key.objectid = inode->i_ino;
2397         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
2398         key.offset = 0;
2399
2400         ret = btrfs_insert_dir_item(trans, root, name, name_len,
2401                                     parent_inode->i_ino,
2402                                     &key, btrfs_inode_type(inode),
2403                                     index);
2404         if (ret == 0) {
2405                 if (add_backref) {
2406                         ret = btrfs_insert_inode_ref(trans, root,
2407                                                      name, name_len,
2408                                                      inode->i_ino,
2409                                                      parent_inode->i_ino,
2410                                                      index);
2411                 }
2412                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
2413                                    name_len * 2);
2414                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
2415                 ret = btrfs_update_inode(trans, root, parent_inode);
2416         }
2417         return ret;
2418 }
2419
2420 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
2421                             struct dentry *dentry, struct inode *inode,
2422                             int backref, u64 index)
2423 {
2424         int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
2425                                  inode, dentry->d_name.name,
2426                                  dentry->d_name.len, backref, index);
2427         if (!err) {
2428                 d_instantiate(dentry, inode);
2429                 return 0;
2430         }
2431         if (err > 0)
2432                 err = -EEXIST;
2433         return err;
2434 }
2435
2436 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
2437                         int mode, dev_t rdev)
2438 {
2439         struct btrfs_trans_handle *trans;
2440         struct btrfs_root *root = BTRFS_I(dir)->root;
2441         struct inode *inode = NULL;
2442         int err;
2443         int drop_inode = 0;
2444         u64 objectid;
2445         unsigned long nr = 0;
2446         u64 index = 0;
2447
2448         if (!new_valid_dev(rdev))
2449                 return -EINVAL;
2450
2451         err = btrfs_check_free_space(root, 1, 0);
2452         if (err)
2453                 goto fail;
2454
2455         trans = btrfs_start_transaction(root, 1);
2456         btrfs_set_trans_block_group(trans, dir);
2457
2458         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2459         if (err) {
2460                 err = -ENOSPC;
2461                 goto out_unlock;
2462         }
2463
2464         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
2465                                 dentry->d_name.len,
2466                                 dentry->d_parent->d_inode->i_ino, objectid,
2467                                 BTRFS_I(dir)->block_group, mode, &index);
2468         err = PTR_ERR(inode);
2469         if (IS_ERR(inode))
2470                 goto out_unlock;
2471
2472         err = btrfs_init_acl(inode, dir);
2473         if (err) {
2474                 drop_inode = 1;
2475                 goto out_unlock;
2476         }
2477
2478         btrfs_set_trans_block_group(trans, inode);
2479         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
2480         if (err)
2481                 drop_inode = 1;
2482         else {
2483                 inode->i_op = &btrfs_special_inode_operations;
2484                 init_special_inode(inode, inode->i_mode, rdev);
2485                 btrfs_update_inode(trans, root, inode);
2486         }
2487         dir->i_sb->s_dirt = 1;
2488         btrfs_update_inode_block_group(trans, inode);
2489         btrfs_update_inode_block_group(trans, dir);
2490 out_unlock:
2491         nr = trans->blocks_used;
2492         btrfs_end_transaction_throttle(trans, root);
2493 fail:
2494         if (drop_inode) {
2495                 inode_dec_link_count(inode);
2496                 iput(inode);
2497         }
2498         btrfs_btree_balance_dirty(root, nr);
2499         return err;
2500 }
2501
2502 static int btrfs_create(struct inode *dir, struct dentry *dentry,
2503                         int mode, struct nameidata *nd)
2504 {
2505         struct btrfs_trans_handle *trans;
2506         struct btrfs_root *root = BTRFS_I(dir)->root;
2507         struct inode *inode = NULL;
2508         int err;
2509         int drop_inode = 0;
2510         unsigned long nr = 0;
2511         u64 objectid;
2512         u64 index = 0;
2513
2514         err = btrfs_check_free_space(root, 1, 0);
2515         if (err)
2516                 goto fail;
2517         trans = btrfs_start_transaction(root, 1);
2518         btrfs_set_trans_block_group(trans, dir);
2519
2520         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2521         if (err) {
2522                 err = -ENOSPC;
2523                 goto out_unlock;
2524         }
2525
2526         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
2527                                 dentry->d_name.len,
2528                                 dentry->d_parent->d_inode->i_ino,
2529                                 objectid, BTRFS_I(dir)->block_group, mode,
2530                                 &index);
2531         err = PTR_ERR(inode);
2532         if (IS_ERR(inode))
2533                 goto out_unlock;
2534
2535         err = btrfs_init_acl(inode, dir);
2536         if (err) {
2537                 drop_inode = 1;
2538                 goto out_unlock;
2539         }
2540
2541         btrfs_set_trans_block_group(trans, inode);
2542         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
2543         if (err)
2544                 drop_inode = 1;
2545         else {
2546                 inode->i_mapping->a_ops = &btrfs_aops;
2547                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2548                 inode->i_fop = &btrfs_file_operations;
2549                 inode->i_op = &btrfs_file_inode_operations;
2550                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2551         }
2552         dir->i_sb->s_dirt = 1;
2553         btrfs_update_inode_block_group(trans, inode);
2554         btrfs_update_inode_block_group(trans, dir);
2555 out_unlock:
2556         nr = trans->blocks_used;
2557         btrfs_end_transaction_throttle(trans, root);
2558 fail:
2559         if (drop_inode) {
2560                 inode_dec_link_count(inode);
2561                 iput(inode);
2562         }
2563         btrfs_btree_balance_dirty(root, nr);
2564         return err;
2565 }
2566
2567 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
2568                       struct dentry *dentry)
2569 {
2570         struct btrfs_trans_handle *trans;
2571         struct btrfs_root *root = BTRFS_I(dir)->root;
2572         struct inode *inode = old_dentry->d_inode;
2573         u64 index;
2574         unsigned long nr = 0;
2575         int err;
2576         int drop_inode = 0;
2577
2578         if (inode->i_nlink == 0)
2579                 return -ENOENT;
2580
2581         btrfs_inc_nlink(inode);
2582         err = btrfs_check_free_space(root, 1, 0);
2583         if (err)
2584                 goto fail;
2585         err = btrfs_set_inode_index(dir, inode, &index);
2586         if (err)
2587                 goto fail;
2588
2589         trans = btrfs_start_transaction(root, 1);
2590
2591         btrfs_set_trans_block_group(trans, dir);
2592         atomic_inc(&inode->i_count);
2593
2594         err = btrfs_add_nondir(trans, dentry, inode, 1, index);
2595
2596         if (err)
2597                 drop_inode = 1;
2598
2599         dir->i_sb->s_dirt = 1;
2600         btrfs_update_inode_block_group(trans, dir);
2601         err = btrfs_update_inode(trans, root, inode);
2602
2603         if (err)
2604                 drop_inode = 1;
2605
2606         nr = trans->blocks_used;
2607         btrfs_end_transaction_throttle(trans, root);
2608 fail:
2609         if (drop_inode) {
2610                 inode_dec_link_count(inode);
2611                 iput(inode);
2612         }
2613         btrfs_btree_balance_dirty(root, nr);
2614         return err;
2615 }
2616
2617 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2618 {
2619         struct inode *inode = NULL;
2620         struct btrfs_trans_handle *trans;
2621         struct btrfs_root *root = BTRFS_I(dir)->root;
2622         int err = 0;
2623         int drop_on_err = 0;
2624         u64 objectid = 0;
2625         u64 index = 0;
2626         unsigned long nr = 1;
2627
2628         err = btrfs_check_free_space(root, 1, 0);
2629         if (err)
2630                 goto out_unlock;
2631
2632         trans = btrfs_start_transaction(root, 1);
2633         btrfs_set_trans_block_group(trans, dir);
2634
2635         if (IS_ERR(trans)) {
2636                 err = PTR_ERR(trans);
2637                 goto out_unlock;
2638         }
2639
2640         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2641         if (err) {
2642                 err = -ENOSPC;
2643                 goto out_unlock;
2644         }
2645
2646         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
2647                                 dentry->d_name.len,
2648                                 dentry->d_parent->d_inode->i_ino, objectid,
2649                                 BTRFS_I(dir)->block_group, S_IFDIR | mode,
2650                                 &index);
2651         if (IS_ERR(inode)) {
2652                 err = PTR_ERR(inode);
2653                 goto out_fail;
2654         }
2655
2656         drop_on_err = 1;
2657
2658         err = btrfs_init_acl(inode, dir);
2659         if (err)
2660                 goto out_fail;
2661
2662         inode->i_op = &btrfs_dir_inode_operations;
2663         inode->i_fop = &btrfs_dir_file_operations;
2664         btrfs_set_trans_block_group(trans, inode);
2665
2666         btrfs_i_size_write(inode, 0);
2667         err = btrfs_update_inode(trans, root, inode);
2668         if (err)
2669                 goto out_fail;
2670
2671         err = btrfs_add_link(trans, dentry->d_parent->d_inode,
2672                                  inode, dentry->d_name.name,
2673                                  dentry->d_name.len, 0, index);
2674         if (err)
2675                 goto out_fail;
2676
2677         d_instantiate(dentry, inode);
2678         drop_on_err = 0;
2679         dir->i_sb->s_dirt = 1;
2680         btrfs_update_inode_block_group(trans, inode);
2681         btrfs_update_inode_block_group(trans, dir);
2682
2683 out_fail:
2684         nr = trans->blocks_used;
2685         btrfs_end_transaction_throttle(trans, root);
2686
2687 out_unlock:
2688         if (drop_on_err)
2689                 iput(inode);
2690         btrfs_btree_balance_dirty(root, nr);
2691         return err;
2692 }
2693
2694 static int merge_extent_mapping(struct extent_map_tree *em_tree,
2695                                 struct extent_map *existing,
2696                                 struct extent_map *em,
2697                                 u64 map_start, u64 map_len)
2698 {
2699         u64 start_diff;
2700
2701         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
2702         start_diff = map_start - em->start;
2703         em->start = map_start;
2704         em->len = map_len;
2705         if (em->block_start < EXTENT_MAP_LAST_BYTE)
2706                 em->block_start += start_diff;
2707         return add_extent_mapping(em_tree, em);
2708 }
2709
2710 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
2711                                     size_t pg_offset, u64 start, u64 len,
2712                                     int create)
2713 {
2714         int ret;
2715         int err = 0;
2716         u64 bytenr;
2717         u64 extent_start = 0;
2718         u64 extent_end = 0;
2719         u64 objectid = inode->i_ino;
2720         u32 found_type;
2721         struct btrfs_path *path = NULL;
2722         struct btrfs_root *root = BTRFS_I(inode)->root;
2723         struct btrfs_file_extent_item *item;
2724         struct extent_buffer *leaf;
2725         struct btrfs_key found_key;
2726         struct extent_map *em = NULL;
2727         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2728         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2729         struct btrfs_trans_handle *trans = NULL;
2730
2731 again:
2732         spin_lock(&em_tree->lock);
2733         em = lookup_extent_mapping(em_tree, start, len);
2734         if (em)
2735                 em->bdev = root->fs_info->fs_devices->latest_bdev;
2736         spin_unlock(&em_tree->lock);
2737
2738         if (em) {
2739                 if (em->start > start || em->start + em->len <= start)
2740                         free_extent_map(em);
2741                 else if (em->block_start == EXTENT_MAP_INLINE && page)
2742                         free_extent_map(em);
2743                 else
2744                         goto out;
2745         }
2746         em = alloc_extent_map(GFP_NOFS);
2747         if (!em) {
2748                 err = -ENOMEM;
2749                 goto out;
2750         }
2751         em->bdev = root->fs_info->fs_devices->latest_bdev;
2752         em->start = EXTENT_MAP_HOLE;
2753         em->len = (u64)-1;
2754
2755         if (!path) {
2756                 path = btrfs_alloc_path();
2757                 BUG_ON(!path);
2758         }
2759
2760         ret = btrfs_lookup_file_extent(trans, root, path,
2761                                        objectid, start, trans != NULL);
2762         if (ret < 0) {
2763                 err = ret;
2764                 goto out;
2765         }
2766
2767         if (ret != 0) {
2768                 if (path->slots[0] == 0)
2769                         goto not_found;
2770                 path->slots[0]--;
2771         }
2772
2773         leaf = path->nodes[0];
2774         item = btrfs_item_ptr(leaf, path->slots[0],
2775                               struct btrfs_file_extent_item);
2776         /* are we inside the extent that was found? */
2777         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2778         found_type = btrfs_key_type(&found_key);
2779         if (found_key.objectid != objectid ||
2780             found_type != BTRFS_EXTENT_DATA_KEY) {
2781                 goto not_found;
2782         }
2783
2784         found_type = btrfs_file_extent_type(leaf, item);
2785         extent_start = found_key.offset;
2786         if (found_type == BTRFS_FILE_EXTENT_REG) {
2787                 extent_end = extent_start +
2788                        btrfs_file_extent_num_bytes(leaf, item);
2789                 err = 0;
2790                 if (start < extent_start || start >= extent_end) {
2791                         em->start = start;
2792                         if (start < extent_start) {
2793                                 if (start + len <= extent_start)
2794                                         goto not_found;
2795                                 em->len = extent_end - extent_start;
2796                         } else {
2797                                 em->len = len;
2798                         }
2799                         goto not_found_em;
2800                 }
2801                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
2802                 if (bytenr == 0) {
2803                         em->start = extent_start;
2804                         em->len = extent_end - extent_start;
2805                         em->block_start = EXTENT_MAP_HOLE;
2806                         goto insert;
2807                 }
2808                 bytenr += btrfs_file_extent_offset(leaf, item);
2809                 em->block_start = bytenr;
2810                 em->start = extent_start;
2811                 em->len = extent_end - extent_start;
2812                 goto insert;
2813         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
2814                 u64 page_start;
2815                 unsigned long ptr;
2816                 char *map;
2817                 size_t size;
2818                 size_t extent_offset;
2819                 size_t copy_size;
2820
2821                 size = btrfs_file_extent_inline_len(leaf, btrfs_item_nr(leaf,
2822                                                     path->slots[0]));
2823                 extent_end = (extent_start + size + root->sectorsize - 1) &
2824                         ~((u64)root->sectorsize - 1);
2825                 if (start < extent_start || start >= extent_end) {
2826                         em->start = start;
2827                         if (start < extent_start) {
2828                                 if (start + len <= extent_start)
2829                                         goto not_found;
2830                                 em->len = extent_end - extent_start;
2831                         } else {
2832                                 em->len = len;
2833                         }
2834                         goto not_found_em;
2835                 }
2836                 em->block_start = EXTENT_MAP_INLINE;
2837
2838                 if (!page) {
2839                         em->start = extent_start;
2840                         em->len = size;
2841                         goto out;
2842                 }
2843
2844                 page_start = page_offset(page) + pg_offset;
2845                 extent_offset = page_start - extent_start;
2846                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
2847                                 size - extent_offset);
2848                 em->start = extent_start + extent_offset;
2849                 em->len = (copy_size + root->sectorsize - 1) &
2850                         ~((u64)root->sectorsize - 1);
2851                 map = kmap(page);
2852                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
2853                 if (create == 0 && !PageUptodate(page)) {
2854                         read_extent_buffer(leaf, map + pg_offset, ptr,
2855                                            copy_size);
2856                         flush_dcache_page(page);
2857                 } else if (create && PageUptodate(page)) {
2858                         if (!trans) {
2859                                 kunmap(page);
2860                                 free_extent_map(em);
2861                                 em = NULL;
2862                                 btrfs_release_path(root, path);
2863                                 trans = btrfs_join_transaction(root, 1);
2864                                 goto again;
2865                         }
2866                         write_extent_buffer(leaf, map + pg_offset, ptr,
2867                                             copy_size);
2868                         btrfs_mark_buffer_dirty(leaf);
2869                 }
2870                 kunmap(page);
2871                 set_extent_uptodate(io_tree, em->start,
2872                                     extent_map_end(em) - 1, GFP_NOFS);
2873                 goto insert;
2874         } else {
2875                 printk("unkknown found_type %d\n", found_type);
2876                 WARN_ON(1);
2877         }
2878 not_found:
2879         em->start = start;
2880         em->len = len;
2881 not_found_em:
2882         em->block_start = EXTENT_MAP_HOLE;
2883 insert:
2884         btrfs_release_path(root, path);
2885         if (em->start > start || extent_map_end(em) <= start) {
2886                 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->len, start, len);
2887                 err = -EIO;
2888                 goto out;
2889         }
2890
2891         err = 0;
2892         spin_lock(&em_tree->lock);
2893         ret = add_extent_mapping(em_tree, em);
2894         /* it is possible that someone inserted the extent into the tree
2895          * while we had the lock dropped.  It is also possible that
2896          * an overlapping map exists in the tree
2897          */
2898         if (ret == -EEXIST) {
2899                 struct extent_map *existing;
2900
2901                 ret = 0;
2902
2903                 existing = lookup_extent_mapping(em_tree, start, len);
2904                 if (existing && (existing->start > start ||
2905                     existing->start + existing->len <= start)) {
2906                         free_extent_map(existing);
2907                         existing = NULL;
2908                 }
2909                 if (!existing) {
2910                         existing = lookup_extent_mapping(em_tree, em->start,
2911                                                          em->len);
2912                         if (existing) {
2913                                 err = merge_extent_mapping(em_tree, existing,
2914                                                            em, start,
2915                                                            root->sectorsize);
2916                                 free_extent_map(existing);
2917                                 if (err) {
2918                                         free_extent_map(em);
2919                                         em = NULL;
2920                                 }
2921                         } else {
2922                                 err = -EIO;
2923                                 printk("failing to insert %Lu %Lu\n",
2924                                        start, len);
2925                                 free_extent_map(em);
2926                                 em = NULL;
2927                         }
2928                 } else {
2929                         free_extent_map(em);
2930                         em = existing;
2931                         err = 0;
2932                 }
2933         }
2934         spin_unlock(&em_tree->lock);
2935 out:
2936         if (path)
2937                 btrfs_free_path(path);
2938         if (trans) {
2939                 ret = btrfs_end_transaction(trans, root);
2940                 if (!err) {
2941                         err = ret;
2942                 }
2943         }
2944         if (err) {
2945                 free_extent_map(em);
2946                 WARN_ON(1);
2947                 return ERR_PTR(err);
2948         }
2949         return em;
2950 }
2951
2952 #if 0 /* waiting for O_DIRECT reads */
2953 static int btrfs_get_block(struct inode *inode, sector_t iblock,
2954                         struct buffer_head *bh_result, int create)
2955 {
2956         struct extent_map *em;
2957         u64 start = (u64)iblock << inode->i_blkbits;
2958         struct btrfs_multi_bio *multi = NULL;
2959         struct btrfs_root *root = BTRFS_I(inode)->root;
2960         u64 len;
2961         u64 logical;
2962         u64 map_length;
2963         int ret = 0;
2964
2965         em = btrfs_get_extent(inode, NULL, 0, start, bh_result->b_size, 0);
2966
2967         if (!em || IS_ERR(em))
2968                 goto out;
2969
2970         if (em->start > start || em->start + em->len <= start) {
2971             goto out;
2972         }
2973
2974         if (em->block_start == EXTENT_MAP_INLINE) {
2975                 ret = -EINVAL;
2976                 goto out;
2977         }
2978
2979         len = em->start + em->len - start;
2980         len = min_t(u64, len, INT_LIMIT(typeof(bh_result->b_size)));
2981
2982         if (em->block_start == EXTENT_MAP_HOLE ||
2983             em->block_start == EXTENT_MAP_DELALLOC) {
2984                 bh_result->b_size = len;
2985                 goto out;
2986         }
2987
2988         logical = start - em->start;
2989         logical = em->block_start + logical;
2990
2991         map_length = len;
2992         ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
2993                               logical, &map_length, &multi, 0);
2994         BUG_ON(ret);
2995         bh_result->b_blocknr = multi->stripes[0].physical >> inode->i_blkbits;
2996         bh_result->b_size = min(map_length, len);
2997
2998         bh_result->b_bdev = multi->stripes[0].dev->bdev;
2999         set_buffer_mapped(bh_result);
3000         kfree(multi);
3001 out:
3002         free_extent_map(em);
3003         return ret;
3004 }
3005 #endif
3006
3007 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
3008                         const struct iovec *iov, loff_t offset,
3009                         unsigned long nr_segs)
3010 {
3011         return -EINVAL;
3012 #if 0
3013         struct file *file = iocb->ki_filp;
3014         struct inode *inode = file->f_mapping->host;
3015
3016         if (rw == WRITE)
3017                 return -EINVAL;
3018
3019         return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
3020                                   offset, nr_segs, btrfs_get_block, NULL);
3021 #endif
3022 }
3023
3024 static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
3025 {
3026         return extent_bmap(mapping, iblock, btrfs_get_extent);
3027 }
3028
3029 int btrfs_readpage(struct file *file, struct page *page)
3030 {
3031         struct extent_io_tree *tree;
3032         tree = &BTRFS_I(page->mapping->host)->io_tree;
3033         return extent_read_full_page(tree, page, btrfs_get_extent);
3034 }
3035
3036 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
3037 {
3038         struct extent_io_tree *tree;
3039
3040
3041         if (current->flags & PF_MEMALLOC) {
3042                 redirty_page_for_writepage(wbc, page);
3043                 unlock_page(page);
3044                 return 0;
3045         }
3046         tree = &BTRFS_I(page->mapping->host)->io_tree;
3047         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
3048 }
3049
3050 int btrfs_writepages(struct address_space *mapping,
3051                      struct writeback_control *wbc)
3052 {
3053         struct extent_io_tree *tree;
3054         tree = &BTRFS_I(mapping->host)->io_tree;
3055         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
3056 }
3057
3058 static int
3059 btrfs_readpages(struct file *file, struct address_space *mapping,
3060                 struct list_head *pages, unsigned nr_pages)
3061 {
3062         struct extent_io_tree *tree;
3063         tree = &BTRFS_I(mapping->host)->io_tree;
3064         return extent_readpages(tree, mapping, pages, nr_pages,
3065                                 btrfs_get_extent);
3066 }
3067 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
3068 {
3069         struct extent_io_tree *tree;
3070         struct extent_map_tree *map;
3071         int ret;
3072
3073         tree = &BTRFS_I(page->mapping->host)->io_tree;
3074         map = &BTRFS_I(page->mapping->host)->extent_tree;
3075         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
3076         if (ret == 1) {
3077                 ClearPagePrivate(page);
3078                 set_page_private(page, 0);
3079                 page_cache_release(page);
3080         }
3081         return ret;
3082 }
3083
3084 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
3085 {
3086         if (PageWriteback(page) || PageDirty(page))
3087                 return 0;
3088         return __btrfs_releasepage(page, gfp_flags);
3089 }
3090
3091 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
3092 {
3093         struct extent_io_tree *tree;
3094         struct btrfs_ordered_extent *ordered;
3095         u64 page_start = page_offset(page);
3096         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
3097
3098         wait_on_page_writeback(page);
3099         tree = &BTRFS_I(page->mapping->host)->io_tree;
3100         if (offset) {
3101                 btrfs_releasepage(page, GFP_NOFS);
3102                 return;
3103         }
3104
3105         lock_extent(tree, page_start, page_end, GFP_NOFS);
3106         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
3107                                            page_offset(page));
3108         if (ordered) {
3109                 /*
3110                  * IO on this page will never be started, so we need
3111                  * to account for any ordered extents now
3112                  */
3113                 clear_extent_bit(tree, page_start, page_end,
3114                                  EXTENT_DIRTY | EXTENT_DELALLOC |
3115                                  EXTENT_LOCKED, 1, 0, GFP_NOFS);
3116                 btrfs_finish_ordered_io(page->mapping->host,
3117                                         page_start, page_end);
3118                 btrfs_put_ordered_extent(ordered);
3119                 lock_extent(tree, page_start, page_end, GFP_NOFS);
3120         }
3121         clear_extent_bit(tree, page_start, page_end,
3122                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
3123                  EXTENT_ORDERED,
3124                  1, 1, GFP_NOFS);
3125         __btrfs_releasepage(page, GFP_NOFS);
3126
3127         ClearPageChecked(page);
3128         if (PagePrivate(page)) {
3129                 ClearPagePrivate(page);
3130                 set_page_private(page, 0);
3131                 page_cache_release(page);
3132         }
3133 }
3134
3135 /*
3136  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
3137  * called from a page fault handler when a page is first dirtied. Hence we must
3138  * be careful to check for EOF conditions here. We set the page up correctly
3139  * for a written page which means we get ENOSPC checking when writing into
3140  * holes and correct delalloc and unwritten extent mapping on filesystems that
3141  * support these features.
3142  *
3143  * We are not allowed to take the i_mutex here so we have to play games to
3144  * protect against truncate races as the page could now be beyond EOF.  Because
3145  * vmtruncate() writes the inode size before removing pages, once we have the
3146  * page lock we can determine safely if the page is beyond EOF. If it is not
3147  * beyond EOF, then the page is guaranteed safe against truncation until we
3148  * unlock the page.
3149  */
3150 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
3151 {
3152         struct inode *inode = fdentry(vma->vm_file)->d_inode;
3153         struct btrfs_root *root = BTRFS_I(inode)->root;
3154         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3155         struct btrfs_ordered_extent *ordered;
3156         char *kaddr;
3157         unsigned long zero_start;
3158         loff_t size;
3159         int ret;
3160         u64 page_start;
3161         u64 page_end;
3162
3163         ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
3164         if (ret)
3165                 goto out;
3166
3167         ret = -EINVAL;
3168 again:
3169         lock_page(page);
3170         size = i_size_read(inode);
3171         page_start = page_offset(page);
3172         page_end = page_start + PAGE_CACHE_SIZE - 1;
3173
3174         if ((page->mapping != inode->i_mapping) ||
3175             (page_start >= size)) {
3176                 /* page got truncated out from underneath us */
3177                 goto out_unlock;
3178         }
3179         wait_on_page_writeback(page);
3180
3181         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
3182         set_page_extent_mapped(page);
3183
3184         /*
3185          * we can't set the delalloc bits if there are pending ordered
3186          * extents.  Drop our locks and wait for them to finish
3187          */
3188         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3189         if (ordered) {
3190                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3191                 unlock_page(page);
3192                 btrfs_start_ordered_extent(inode, ordered, 1);
3193                 btrfs_put_ordered_extent(ordered);
3194                 goto again;
3195         }
3196
3197         btrfs_set_extent_delalloc(inode, page_start, page_end);
3198         ret = 0;
3199
3200         /* page is wholly or partially inside EOF */
3201         if (page_start + PAGE_CACHE_SIZE > size)
3202                 zero_start = size & ~PAGE_CACHE_MASK;
3203         else
3204                 zero_start = PAGE_CACHE_SIZE;
3205
3206         if (zero_start != PAGE_CACHE_SIZE) {
3207                 kaddr = kmap(page);
3208                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
3209                 flush_dcache_page(page);
3210                 kunmap(page);
3211         }
3212         ClearPageChecked(page);
3213         set_page_dirty(page);
3214         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3215
3216 out_unlock:
3217         unlock_page(page);
3218 out:
3219         return ret;
3220 }
3221
3222 static void btrfs_truncate(struct inode *inode)
3223 {
3224         struct btrfs_root *root = BTRFS_I(inode)->root;
3225         int ret;
3226         struct btrfs_trans_handle *trans;
3227         unsigned long nr;
3228         u64 mask = root->sectorsize - 1;
3229
3230         if (!S_ISREG(inode->i_mode))
3231                 return;
3232         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3233                 return;
3234
3235         btrfs_truncate_page(inode->i_mapping, inode->i_size);
3236         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
3237
3238         trans = btrfs_start_transaction(root, 1);
3239         btrfs_set_trans_block_group(trans, inode);
3240         btrfs_i_size_write(inode, inode->i_size);
3241
3242         ret = btrfs_orphan_add(trans, inode);
3243         if (ret)
3244                 goto out;
3245         /* FIXME, add redo link to tree so we don't leak on crash */
3246         ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size,
3247                                       BTRFS_EXTENT_DATA_KEY);
3248         btrfs_update_inode(trans, root, inode);
3249
3250         ret = btrfs_orphan_del(trans, inode);
3251         BUG_ON(ret);
3252
3253 out:
3254         nr = trans->blocks_used;
3255         ret = btrfs_end_transaction_throttle(trans, root);
3256         BUG_ON(ret);
3257         btrfs_btree_balance_dirty(root, nr);
3258 }
3259
3260 /*
3261  * Invalidate a single dcache entry at the root of the filesystem.
3262  * Needed after creation of snapshot or subvolume.
3263  */
3264 void btrfs_invalidate_dcache_root(struct btrfs_root *root, char *name,
3265                                   int namelen)
3266 {
3267         struct dentry *alias, *entry;
3268         struct qstr qstr;
3269
3270         alias = d_find_alias(root->fs_info->sb->s_root->d_inode);
3271         if (alias) {
3272                 qstr.name = name;
3273                 qstr.len = namelen;
3274                 /* change me if btrfs ever gets a d_hash operation */
3275                 qstr.hash = full_name_hash(qstr.name, qstr.len);
3276                 entry = d_lookup(alias, &qstr);
3277                 dput(alias);
3278                 if (entry) {
3279                         d_invalidate(entry);
3280                         dput(entry);
3281                 }
3282         }
3283 }
3284
3285 int btrfs_create_subvol_root(struct btrfs_root *new_root,
3286                 struct btrfs_trans_handle *trans, u64 new_dirid,
3287                 struct btrfs_block_group_cache *block_group)
3288 {
3289         struct inode *inode;
3290         u64 index = 0;
3291
3292         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
3293                                 new_dirid, block_group, S_IFDIR | 0700, &index);
3294         if (IS_ERR(inode))
3295                 return PTR_ERR(inode);
3296         inode->i_op = &btrfs_dir_inode_operations;
3297         inode->i_fop = &btrfs_dir_file_operations;
3298         new_root->inode = inode;
3299
3300         inode->i_nlink = 1;
3301         btrfs_i_size_write(inode, 0);
3302
3303         return btrfs_update_inode(trans, new_root, inode);
3304 }
3305
3306 unsigned long btrfs_force_ra(struct address_space *mapping,
3307                               struct file_ra_state *ra, struct file *file,
3308                               pgoff_t offset, pgoff_t last_index)
3309 {
3310         pgoff_t req_size = last_index - offset + 1;
3311
3312 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
3313         offset = page_cache_readahead(mapping, ra, file, offset, req_size);
3314         return offset;
3315 #else
3316         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
3317         return offset + req_size;
3318 #endif
3319 }
3320
3321 struct inode *btrfs_alloc_inode(struct super_block *sb)
3322 {
3323         struct btrfs_inode *ei;
3324
3325         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
3326         if (!ei)
3327                 return NULL;
3328         ei->last_trans = 0;
3329         ei->logged_trans = 0;
3330         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
3331         ei->i_acl = BTRFS_ACL_NOT_CACHED;
3332         ei->i_default_acl = BTRFS_ACL_NOT_CACHED;
3333         INIT_LIST_HEAD(&ei->i_orphan);
3334         return &ei->vfs_inode;
3335 }
3336
3337 void btrfs_destroy_inode(struct inode *inode)
3338 {
3339         struct btrfs_ordered_extent *ordered;
3340         WARN_ON(!list_empty(&inode->i_dentry));
3341         WARN_ON(inode->i_data.nrpages);
3342
3343         if (BTRFS_I(inode)->i_acl &&
3344             BTRFS_I(inode)->i_acl != BTRFS_ACL_NOT_CACHED)
3345                 posix_acl_release(BTRFS_I(inode)->i_acl);
3346         if (BTRFS_I(inode)->i_default_acl &&
3347             BTRFS_I(inode)->i_default_acl != BTRFS_ACL_NOT_CACHED)
3348                 posix_acl_release(BTRFS_I(inode)->i_default_acl);
3349
3350         spin_lock(&BTRFS_I(inode)->root->list_lock);
3351         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
3352                 printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
3353                        " list\n", inode->i_ino);
3354                 dump_stack();
3355         }
3356         spin_unlock(&BTRFS_I(inode)->root->list_lock);
3357
3358         while(1) {
3359                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
3360                 if (!ordered)
3361                         break;
3362                 else {
3363                         printk("found ordered extent %Lu %Lu\n",
3364                                ordered->file_offset, ordered->len);
3365                         btrfs_remove_ordered_extent(inode, ordered);
3366                         btrfs_put_ordered_extent(ordered);
3367                         btrfs_put_ordered_extent(ordered);
3368                 }
3369         }
3370         btrfs_drop_extent_cache(inode, 0, (u64)-1);
3371         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
3372 }
3373
3374 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,26)
3375 static void init_once(void *foo)
3376 #elif LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
3377 static void init_once(struct kmem_cache * cachep, void *foo)
3378 #else
3379 static void init_once(void * foo, struct kmem_cache * cachep,
3380                       unsigned long flags)
3381 #endif
3382 {
3383         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
3384
3385         inode_init_once(&ei->vfs_inode);
3386 }
3387
3388 void btrfs_destroy_cachep(void)
3389 {
3390         if (btrfs_inode_cachep)
3391                 kmem_cache_destroy(btrfs_inode_cachep);
3392         if (btrfs_trans_handle_cachep)
3393                 kmem_cache_destroy(btrfs_trans_handle_cachep);
3394         if (btrfs_transaction_cachep)
3395                 kmem_cache_destroy(btrfs_transaction_cachep);
3396         if (btrfs_bit_radix_cachep)
3397                 kmem_cache_destroy(btrfs_bit_radix_cachep);
3398         if (btrfs_path_cachep)
3399                 kmem_cache_destroy(btrfs_path_cachep);
3400 }
3401
3402 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
3403                                        unsigned long extra_flags,
3404 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,26)
3405                                        void (*ctor)(void *)
3406 #elif LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
3407                                        void (*ctor)(struct kmem_cache *, void *)
3408 #else
3409                                        void (*ctor)(void *, struct kmem_cache *,
3410                                                     unsigned long)
3411 #endif
3412                                      )
3413 {
3414         return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
3415                                  SLAB_MEM_SPREAD | extra_flags), ctor
3416 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
3417                                  ,NULL
3418 #endif
3419                                 );
3420 }
3421
3422 int btrfs_init_cachep(void)
3423 {
3424         btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
3425                                           sizeof(struct btrfs_inode),
3426                                           0, init_once);
3427         if (!btrfs_inode_cachep)
3428                 goto fail;
3429         btrfs_trans_handle_cachep =
3430                         btrfs_cache_create("btrfs_trans_handle_cache",
3431                                            sizeof(struct btrfs_trans_handle),
3432                                            0, NULL);
3433         if (!btrfs_trans_handle_cachep)
3434                 goto fail;
3435         btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
3436                                              sizeof(struct btrfs_transaction),
3437                                              0, NULL);
3438         if (!btrfs_transaction_cachep)
3439                 goto fail;
3440         btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
3441                                          sizeof(struct btrfs_path),
3442                                          0, NULL);
3443         if (!btrfs_path_cachep)
3444                 goto fail;
3445         btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
3446                                               SLAB_DESTROY_BY_RCU, NULL);
3447         if (!btrfs_bit_radix_cachep)
3448                 goto fail;
3449         return 0;
3450 fail:
3451         btrfs_destroy_cachep();
3452         return -ENOMEM;
3453 }
3454
3455 static int btrfs_getattr(struct vfsmount *mnt,
3456                          struct dentry *dentry, struct kstat *stat)
3457 {
3458         struct inode *inode = dentry->d_inode;
3459         generic_fillattr(inode, stat);
3460         stat->blksize = PAGE_CACHE_SIZE;
3461         stat->blocks = inode->i_blocks + (BTRFS_I(inode)->delalloc_bytes >> 9);
3462         return 0;
3463 }
3464
3465 static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
3466                            struct inode * new_dir,struct dentry *new_dentry)
3467 {
3468         struct btrfs_trans_handle *trans;
3469         struct btrfs_root *root = BTRFS_I(old_dir)->root;
3470         struct inode *new_inode = new_dentry->d_inode;
3471         struct inode *old_inode = old_dentry->d_inode;
3472         struct timespec ctime = CURRENT_TIME;
3473         u64 index = 0;
3474         int ret;
3475
3476         if (S_ISDIR(old_inode->i_mode) && new_inode &&
3477             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
3478                 return -ENOTEMPTY;
3479         }
3480
3481         ret = btrfs_check_free_space(root, 1, 0);
3482         if (ret)
3483                 goto out_unlock;
3484
3485         trans = btrfs_start_transaction(root, 1);
3486
3487         btrfs_set_trans_block_group(trans, new_dir);
3488
3489         btrfs_inc_nlink(old_dentry->d_inode);
3490         old_dir->i_ctime = old_dir->i_mtime = ctime;
3491         new_dir->i_ctime = new_dir->i_mtime = ctime;
3492         old_inode->i_ctime = ctime;
3493
3494         ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode,
3495                                  old_dentry->d_name.name,
3496                                  old_dentry->d_name.len);
3497         if (ret)
3498                 goto out_fail;
3499
3500         if (new_inode) {
3501                 new_inode->i_ctime = CURRENT_TIME;
3502                 ret = btrfs_unlink_inode(trans, root, new_dir,
3503                                          new_dentry->d_inode,
3504                                          new_dentry->d_name.name,
3505                                          new_dentry->d_name.len);
3506                 if (ret)
3507                         goto out_fail;
3508                 if (new_inode->i_nlink == 0) {
3509                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
3510                         if (ret)
3511                                 goto out_fail;
3512                 }
3513
3514         }
3515         ret = btrfs_set_inode_index(new_dir, old_inode, &index);
3516         if (ret)
3517                 goto out_fail;
3518
3519         ret = btrfs_add_link(trans, new_dentry->d_parent->d_inode,
3520                              old_inode, new_dentry->d_name.name,
3521                              new_dentry->d_name.len, 1, index);
3522         if (ret)
3523                 goto out_fail;
3524
3525 out_fail:
3526         btrfs_end_transaction_throttle(trans, root);
3527 out_unlock:
3528         return ret;
3529 }
3530
3531 int btrfs_start_delalloc_inodes(struct btrfs_root *root)
3532 {
3533         struct list_head *head = &root->fs_info->delalloc_inodes;
3534         struct btrfs_inode *binode;
3535         unsigned long flags;
3536
3537         spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
3538         while(!list_empty(head)) {
3539                 binode = list_entry(head->next, struct btrfs_inode,
3540                                     delalloc_inodes);
3541                 atomic_inc(&binode->vfs_inode.i_count);
3542                 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
3543                 filemap_write_and_wait(binode->vfs_inode.i_mapping);
3544                 iput(&binode->vfs_inode);
3545                 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
3546         }
3547         spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
3548         return 0;
3549 }
3550
3551 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
3552                          const char *symname)
3553 {
3554         struct btrfs_trans_handle *trans;
3555         struct btrfs_root *root = BTRFS_I(dir)->root;
3556         struct btrfs_path *path;
3557         struct btrfs_key key;
3558         struct inode *inode = NULL;
3559         int err;
3560         int drop_inode = 0;
3561         u64 objectid;
3562         u64 index = 0 ;
3563         int name_len;
3564         int datasize;
3565         unsigned long ptr;
3566         struct btrfs_file_extent_item *ei;
3567         struct extent_buffer *leaf;
3568         unsigned long nr = 0;
3569
3570         name_len = strlen(symname) + 1;
3571         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
3572                 return -ENAMETOOLONG;
3573
3574         err = btrfs_check_free_space(root, 1, 0);
3575         if (err)
3576                 goto out_fail;
3577
3578         trans = btrfs_start_transaction(root, 1);
3579         btrfs_set_trans_block_group(trans, dir);
3580
3581         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3582         if (err) {
3583                 err = -ENOSPC;
3584                 goto out_unlock;
3585         }
3586
3587         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3588                                 dentry->d_name.len,
3589                                 dentry->d_parent->d_inode->i_ino, objectid,
3590                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
3591                                 &index);
3592         err = PTR_ERR(inode);
3593         if (IS_ERR(inode))
3594                 goto out_unlock;
3595
3596         err = btrfs_init_acl(inode, dir);
3597         if (err) {
3598                 drop_inode = 1;
3599                 goto out_unlock;
3600         }
3601
3602         btrfs_set_trans_block_group(trans, inode);
3603         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3604         if (err)
3605                 drop_inode = 1;
3606         else {
3607                 inode->i_mapping->a_ops = &btrfs_aops;
3608                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3609                 inode->i_fop = &btrfs_file_operations;
3610                 inode->i_op = &btrfs_file_inode_operations;
3611                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3612         }
3613         dir->i_sb->s_dirt = 1;
3614         btrfs_update_inode_block_group(trans, inode);
3615         btrfs_update_inode_block_group(trans, dir);
3616         if (drop_inode)
3617                 goto out_unlock;
3618
3619         path = btrfs_alloc_path();
3620         BUG_ON(!path);
3621         key.objectid = inode->i_ino;
3622         key.offset = 0;
3623         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
3624         datasize = btrfs_file_extent_calc_inline_size(name_len);
3625         err = btrfs_insert_empty_item(trans, root, path, &key,
3626                                       datasize);
3627         if (err) {
3628                 drop_inode = 1;
3629                 goto out_unlock;
3630         }
3631         leaf = path->nodes[0];
3632         ei = btrfs_item_ptr(leaf, path->slots[0],
3633                             struct btrfs_file_extent_item);
3634         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
3635         btrfs_set_file_extent_type(leaf, ei,
3636                                    BTRFS_FILE_EXTENT_INLINE);
3637         ptr = btrfs_file_extent_inline_start(ei);
3638         write_extent_buffer(leaf, symname, ptr, name_len);
3639         btrfs_mark_buffer_dirty(leaf);
3640         btrfs_free_path(path);
3641
3642         inode->i_op = &btrfs_symlink_inode_operations;
3643         inode->i_mapping->a_ops = &btrfs_symlink_aops;
3644         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3645         btrfs_i_size_write(inode, name_len - 1);
3646         err = btrfs_update_inode(trans, root, inode);
3647         if (err)
3648                 drop_inode = 1;
3649
3650 out_unlock:
3651         nr = trans->blocks_used;
3652         btrfs_end_transaction_throttle(trans, root);
3653 out_fail:
3654         if (drop_inode) {
3655                 inode_dec_link_count(inode);
3656                 iput(inode);
3657         }
3658         btrfs_btree_balance_dirty(root, nr);
3659         return err;
3660 }
3661
3662 static int btrfs_set_page_dirty(struct page *page)
3663 {
3664         return __set_page_dirty_nobuffers(page);
3665 }
3666
3667 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,26)
3668 static int btrfs_permission(struct inode *inode, int mask)
3669 #else
3670 static int btrfs_permission(struct inode *inode, int mask,
3671                             struct nameidata *nd)
3672 #endif
3673 {
3674         if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
3675                 return -EACCES;
3676         return generic_permission(inode, mask, btrfs_check_acl);
3677 }
3678
3679 static struct inode_operations btrfs_dir_inode_operations = {
3680         .lookup         = btrfs_lookup,
3681         .create         = btrfs_create,
3682         .unlink         = btrfs_unlink,
3683         .link           = btrfs_link,
3684         .mkdir          = btrfs_mkdir,
3685         .rmdir          = btrfs_rmdir,
3686         .rename         = btrfs_rename,
3687         .symlink        = btrfs_symlink,
3688         .setattr        = btrfs_setattr,
3689         .mknod          = btrfs_mknod,
3690         .setxattr       = btrfs_setxattr,
3691         .getxattr       = btrfs_getxattr,
3692         .listxattr      = btrfs_listxattr,
3693         .removexattr    = btrfs_removexattr,
3694         .permission     = btrfs_permission,
3695 };
3696 static struct inode_operations btrfs_dir_ro_inode_operations = {
3697         .lookup         = btrfs_lookup,
3698         .permission     = btrfs_permission,
3699 };
3700 static struct file_operations btrfs_dir_file_operations = {
3701         .llseek         = generic_file_llseek,
3702         .read           = generic_read_dir,
3703 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,28)
3704         .readdir        = btrfs_nfshack_readdir,
3705 #else /* NFSd readdir/lookup deadlock is fixed */
3706         .readdir        = btrfs_real_readdir,
3707 #endif
3708         .unlocked_ioctl = btrfs_ioctl,
3709 #ifdef CONFIG_COMPAT
3710         .compat_ioctl   = btrfs_ioctl,
3711 #endif
3712         .release        = btrfs_release_file,
3713         .fsync          = btrfs_sync_file,
3714 };
3715
3716 static struct extent_io_ops btrfs_extent_io_ops = {
3717         .fill_delalloc = run_delalloc_range,
3718         .submit_bio_hook = btrfs_submit_bio_hook,
3719         .merge_bio_hook = btrfs_merge_bio_hook,
3720         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
3721         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
3722         .writepage_start_hook = btrfs_writepage_start_hook,
3723         .readpage_io_failed_hook = btrfs_io_failed_hook,
3724         .set_bit_hook = btrfs_set_bit_hook,
3725         .clear_bit_hook = btrfs_clear_bit_hook,
3726 };
3727
3728 static struct address_space_operations btrfs_aops = {
3729         .readpage       = btrfs_readpage,
3730         .writepage      = btrfs_writepage,
3731         .writepages     = btrfs_writepages,
3732         .readpages      = btrfs_readpages,
3733         .sync_page      = block_sync_page,
3734         .bmap           = btrfs_bmap,
3735         .direct_IO      = btrfs_direct_IO,
3736         .invalidatepage = btrfs_invalidatepage,
3737         .releasepage    = btrfs_releasepage,
3738         .set_page_dirty = btrfs_set_page_dirty,
3739 };
3740
3741 static struct address_space_operations btrfs_symlink_aops = {
3742         .readpage       = btrfs_readpage,
3743         .writepage      = btrfs_writepage,
3744         .invalidatepage = btrfs_invalidatepage,
3745         .releasepage    = btrfs_releasepage,
3746 };
3747
3748 static struct inode_operations btrfs_file_inode_operations = {
3749         .truncate       = btrfs_truncate,
3750         .getattr        = btrfs_getattr,
3751         .setattr        = btrfs_setattr,
3752         .setxattr       = btrfs_setxattr,
3753         .getxattr       = btrfs_getxattr,
3754         .listxattr      = btrfs_listxattr,
3755         .removexattr    = btrfs_removexattr,
3756         .permission     = btrfs_permission,
3757 };
3758 static struct inode_operations btrfs_special_inode_operations = {
3759         .getattr        = btrfs_getattr,
3760         .setattr        = btrfs_setattr,
3761         .permission     = btrfs_permission,
3762         .setxattr       = btrfs_setxattr,
3763         .getxattr       = btrfs_getxattr,
3764         .listxattr      = btrfs_listxattr,
3765         .removexattr    = btrfs_removexattr,
3766 };
3767 static struct inode_operations btrfs_symlink_inode_operations = {
3768         .readlink       = generic_readlink,
3769         .follow_link    = page_follow_link_light,
3770         .put_link       = page_put_link,
3771         .permission     = btrfs_permission,
3772 };