SAFE public projects git trees. - safe/jmp/linux-2.6/blob - fs/btrfs/inode.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/pagemap.h>
  25 #include <linux/highmem.h>
  26 #include <linux/time.h>
  27 #include <linux/init.h>
  28 #include <linux/string.h>
  29 #include <linux/smp_lock.h>
  30 #include <linux/backing-dev.h>
  31 #include <linux/mpage.h>
  32 #include <linux/swap.h>
  33 #include <linux/writeback.h>
  34 #include <linux/statfs.h>
  35 #include <linux/compat.h>
  36 #include <linux/bit_spinlock.h>
  37 #include <linux/version.h>
  38 #include <linux/xattr.h>
  39 #include <linux/posix_acl.h>
  40 #include "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         u64 alloc_hint = 0;
 532         struct list_head list;
 533         struct btrfs_key ins;
 534         int ret;
 535
 536         ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
 537         if (!ret)
 538                 return 0;
 539
 540         trans = btrfs_join_transaction(root, 1);
 541
 542         ordered_extent = btrfs_lookup_ordered_extent(inode, start);
 543         BUG_ON(!ordered_extent);
 544         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
 545                 goto nocow;
 546
 547         lock_extent(io_tree, ordered_extent->file_offset,
 548                     ordered_extent->file_offset + ordered_extent->len - 1,
 549                     GFP_NOFS);
 550
 551         INIT_LIST_HEAD(&list);
 552
 553         ins.objectid = ordered_extent->start;
 554         ins.offset = ordered_extent->len;
 555         ins.type = BTRFS_EXTENT_ITEM_KEY;
 556
 557         ret = btrfs_alloc_reserved_extent(trans, root, root->root_key.objectid,
 558                                           trans->transid, inode->i_ino,
 559                                           ordered_extent->file_offset, &ins);
 560         BUG_ON(ret);
 561         mutex_lock(&BTRFS_I(inode)->extent_mutex);
 562
 563         ret = btrfs_drop_extents(trans, root, inode,
 564                                  ordered_extent->file_offset,
 565                                  ordered_extent->file_offset +
 566                                  ordered_extent->len,
 567                                  ordered_extent->file_offset, &alloc_hint);
 568         BUG_ON(ret);
 569         ret = btrfs_insert_file_extent(trans, root, inode->i_ino,
 570                                        ordered_extent->file_offset,
 571                                        ordered_extent->start,
 572                                        ordered_extent->len,
 573                                        ordered_extent->len, 0);
 574         BUG_ON(ret);
 575
 576         btrfs_drop_extent_cache(inode, ordered_extent->file_offset,
 577                                 ordered_extent->file_offset +
 578                                 ordered_extent->len - 1);
 579         mutex_unlock(&BTRFS_I(inode)->extent_mutex);
 580
 581         inode->i_blocks += ordered_extent->len >> 9;
 582         unlock_extent(io_tree, ordered_extent->file_offset,
 583                     ordered_extent->file_offset + ordered_extent->len - 1,
 584                     GFP_NOFS);
 585 nocow:
 586         add_pending_csums(trans, inode, ordered_extent->file_offset,
 587                           &ordered_extent->list);
 588
 589         btrfs_ordered_update_i_size(inode, ordered_extent);
 590         btrfs_update_inode(trans, root, inode);
 591         btrfs_remove_ordered_extent(inode, ordered_extent);
 592
 593         /* once for us */
 594         btrfs_put_ordered_extent(ordered_extent);
 595         /* once for the tree */
 596         btrfs_put_ordered_extent(ordered_extent);
 597
 598         btrfs_end_transaction(trans, root);
 599         return 0;
 600 }
 601
 602 int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
 603                                 struct extent_state *state, int uptodate)
 604 {
 605         return btrfs_finish_ordered_io(page->mapping->host, start, end);
 606 }
 607
 608 struct io_failure_record {
 609         struct page *page;
 610         u64 start;
 611         u64 len;
 612         u64 logical;
 613         int last_mirror;
 614 };
 615
 616 int btrfs_io_failed_hook(struct bio *failed_bio,
 617                          struct page *page, u64 start, u64 end,
 618                          struct extent_state *state)
 619 {
 620         struct io_failure_record *failrec = NULL;
 621         u64 private;
 622         struct extent_map *em;
 623         struct inode *inode = page->mapping->host;
 624         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
 625         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 626         struct bio *bio;
 627         int num_copies;
 628         int ret;
 629         int rw;
 630         u64 logical;
 631
 632         ret = get_state_private(failure_tree, start, &private);
 633         if (ret) {
 634                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
 635                 if (!failrec)
 636                         return -ENOMEM;
 637                 failrec->start = start;
 638                 failrec->len = end - start + 1;
 639                 failrec->last_mirror = 0;
 640
 641                 spin_lock(&em_tree->lock);
 642                 em = lookup_extent_mapping(em_tree, start, failrec->len);
 643                 if (em->start > start || em->start + em->len < start) {
 644                         free_extent_map(em);
 645                         em = NULL;
 646                 }
 647                 spin_unlock(&em_tree->lock);
 648
 649                 if (!em || IS_ERR(em)) {
 650                         kfree(failrec);
 651                         return -EIO;
 652                 }
 653                 logical = start - em->start;
 654                 logical = em->block_start + logical;
 655                 failrec->logical = logical;
 656                 free_extent_map(em);
 657                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
 658                                 EXTENT_DIRTY, GFP_NOFS);
 659                 set_state_private(failure_tree, start,
 660                                  (u64)(unsigned long)failrec);
 661         } else {
 662                 failrec = (struct io_failure_record *)(unsigned long)private;
 663         }
 664         num_copies = btrfs_num_copies(
 665                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
 666                               failrec->logical, failrec->len);
 667         failrec->last_mirror++;
 668         if (!state) {
 669                 spin_lock_irq(&BTRFS_I(inode)->io_tree.lock);
 670                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
 671                                                     failrec->start,
 672                                                     EXTENT_LOCKED);
 673                 if (state && state->start != failrec->start)
 674                         state = NULL;
 675                 spin_unlock_irq(&BTRFS_I(inode)->io_tree.lock);
 676         }
 677         if (!state || failrec->last_mirror > num_copies) {
 678                 set_state_private(failure_tree, failrec->start, 0);
 679                 clear_extent_bits(failure_tree, failrec->start,
 680                                   failrec->start + failrec->len - 1,
 681                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
 682                 kfree(failrec);
 683                 return -EIO;
 684         }
 685         bio = bio_alloc(GFP_NOFS, 1);
 686         bio->bi_private = state;
 687         bio->bi_end_io = failed_bio->bi_end_io;
 688         bio->bi_sector = failrec->logical >> 9;
 689         bio->bi_bdev = failed_bio->bi_bdev;
 690         bio->bi_size = 0;
 691         bio_add_page(bio, page, failrec->len, start - page_offset(page));
 692         if (failed_bio->bi_rw & (1 << BIO_RW))
 693                 rw = WRITE;
 694         else
 695                 rw = READ;
 696
 697         BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
 698                                                       failrec->last_mirror);
 699         return 0;
 700 }
 701
 702 int btrfs_clean_io_failures(struct inode *inode, u64 start)
 703 {
 704         u64 private;
 705         u64 private_failure;
 706         struct io_failure_record *failure;
 707         int ret;
 708
 709         private = 0;
 710         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
 711                              (u64)-1, 1, EXTENT_DIRTY)) {
 712                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
 713                                         start, &private_failure);
 714                 if (ret == 0) {
 715                         failure = (struct io_failure_record *)(unsigned long)
 716                                    private_failure;
 717                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
 718                                           failure->start, 0);
 719                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
 720                                           failure->start,
 721                                           failure->start + failure->len - 1,
 722                                           EXTENT_DIRTY | EXTENT_LOCKED,
 723                                           GFP_NOFS);
 724                         kfree(failure);
 725                 }
 726         }
 727         return 0;
 728 }
 729
 730 int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 731                                struct extent_state *state)
 732 {
 733         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
 734         struct inode *inode = page->mapping->host;
 735         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 736         char *kaddr;
 737         u64 private = ~(u32)0;
 738         int ret;
 739         struct btrfs_root *root = BTRFS_I(inode)->root;
 740         u32 csum = ~(u32)0;
 741         unsigned long flags;
 742
 743         if (btrfs_test_opt(root, NODATASUM) ||
 744             btrfs_test_flag(inode, NODATASUM))
 745                 return 0;
 746         if (state && state->start == start) {
 747                 private = state->private;
 748                 ret = 0;
 749         } else {
 750                 ret = get_state_private(io_tree, start, &private);
 751         }
 752         local_irq_save(flags);
 753         kaddr = kmap_atomic(page, KM_IRQ0);
 754         if (ret) {
 755                 goto zeroit;
 756         }
 757         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
 758         btrfs_csum_final(csum, (char *)&csum);
 759         if (csum != private) {
 760                 goto zeroit;
 761         }
 762         kunmap_atomic(kaddr, KM_IRQ0);
 763         local_irq_restore(flags);
 764
 765         /* if the io failure tree for this inode is non-empty,
 766          * check to see if we've recovered from a failed IO
 767          */
 768         btrfs_clean_io_failures(inode, start);
 769         return 0;
 770
 771 zeroit:
 772         printk("btrfs csum failed ino %lu off %llu csum %u private %Lu\n",
 773                page->mapping->host->i_ino, (unsigned long long)start, csum,
 774                private);
 775         memset(kaddr + offset, 1, end - start + 1);
 776         flush_dcache_page(page);
 777         kunmap_atomic(kaddr, KM_IRQ0);
 778         local_irq_restore(flags);
 779         if (private == 0)
 780                 return 0;
 781         return -EIO;
 782 }
 783
 784 /*
 785  * This creates an orphan entry for the given inode in case something goes
 786  * wrong in the middle of an unlink/truncate.
 787  */
 788 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
 789 {
 790         struct btrfs_root *root = BTRFS_I(inode)->root;
 791         int ret = 0;
 792
 793         spin_lock(&root->list_lock);
 794
 795         /* already on the orphan list, we're good */
 796         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
 797                 spin_unlock(&root->list_lock);
 798                 return 0;
 799         }
 800
 801         list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
 802
 803         spin_unlock(&root->list_lock);
 804
 805         /*
 806          * insert an orphan item to track this unlinked/truncated file
 807          */
 808         ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
 809
 810         return ret;
 811 }
 812
 813 /*
 814  * We have done the truncate/delete so we can go ahead and remove the orphan
 815  * item for this particular inode.
 816  */
 817 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
 818 {
 819         struct btrfs_root *root = BTRFS_I(inode)->root;
 820         int ret = 0;
 821
 822         spin_lock(&root->list_lock);
 823
 824         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
 825                 spin_unlock(&root->list_lock);
 826                 return 0;
 827         }
 828
 829         list_del_init(&BTRFS_I(inode)->i_orphan);
 830         if (!trans) {
 831                 spin_unlock(&root->list_lock);
 832                 return 0;
 833         }
 834
 835         spin_unlock(&root->list_lock);
 836
 837         ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
 838
 839         return ret;
 840 }
 841
 842 /*
 843  * this cleans up any orphans that may be left on the list from the last use
 844  * of this root.
 845  */
 846 void btrfs_orphan_cleanup(struct btrfs_root *root)
 847 {
 848         struct btrfs_path *path;
 849         struct extent_buffer *leaf;
 850         struct btrfs_item *item;
 851         struct btrfs_key key, found_key;
 852         struct btrfs_trans_handle *trans;
 853         struct inode *inode;
 854         int ret = 0, nr_unlink = 0, nr_truncate = 0;
 855
 856         /* don't do orphan cleanup if the fs is readonly. */
 857         if (root->inode->i_sb->s_flags & MS_RDONLY)
 858                 return;
 859
 860         path = btrfs_alloc_path();
 861         if (!path)
 862                 return;
 863         path->reada = -1;
 864
 865         key.objectid = BTRFS_ORPHAN_OBJECTID;
 866         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
 867         key.offset = (u64)-1;
 868
 869         trans = btrfs_start_transaction(root, 1);
 870         btrfs_set_trans_block_group(trans, root->inode);
 871
 872         while (1) {
 873                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 874                 if (ret < 0) {
 875                         printk(KERN_ERR "Error searching slot for orphan: %d"
 876                                "\n", ret);
 877                         break;
 878                 }
 879
 880                 /*
 881                  * if ret == 0 means we found what we were searching for, which
 882                  * is weird, but possible, so only screw with path if we didnt
 883                  * find the key and see if we have stuff that matches
 884                  */
 885                 if (ret > 0) {
 886                         if (path->slots[0] == 0)
 887                                 break;
 888                         path->slots[0]--;
 889                 }
 890
 891                 /* pull out the item */
 892                 leaf = path->nodes[0];
 893                 item = btrfs_item_nr(leaf, path->slots[0]);
 894                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 895
 896                 /* make sure the item matches what we want */
 897                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
 898                         break;
 899                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
 900                         break;
 901
 902                 /* release the path since we're done with it */
 903                 btrfs_release_path(root, path);
 904
 905                 /*
 906                  * this is where we are basically btrfs_lookup, without the
 907                  * crossing root thing.  we store the inode number in the
 908                  * offset of the orphan item.
 909                  */
 910                 inode = btrfs_iget_locked(root->inode->i_sb,
 911                                           found_key.offset, root);
 912                 if (!inode)
 913                         break;
 914
 915                 if (inode->i_state & I_NEW) {
 916                         BTRFS_I(inode)->root = root;
 917
 918                         /* have to set the location manually */
 919                         BTRFS_I(inode)->location.objectid = inode->i_ino;
 920                         BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
 921                         BTRFS_I(inode)->location.offset = 0;
 922
 923                         btrfs_read_locked_inode(inode);
 924                         unlock_new_inode(inode);
 925                 }
 926
 927                 /*
 928                  * add this inode to the orphan list so btrfs_orphan_del does
 929                  * the proper thing when we hit it
 930                  */
 931                 spin_lock(&root->list_lock);
 932                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
 933                 spin_unlock(&root->list_lock);
 934
 935                 /*
 936                  * if this is a bad inode, means we actually succeeded in
 937                  * removing the inode, but not the orphan record, which means
 938                  * we need to manually delete the orphan since iput will just
 939                  * do a destroy_inode
 940                  */
 941                 if (is_bad_inode(inode)) {
 942                         btrfs_orphan_del(trans, inode);
 943                         iput(inode);
 944                         continue;
 945                 }
 946
 947                 /* if we have links, this was a truncate, lets do that */
 948                 if (inode->i_nlink) {
 949                         nr_truncate++;
 950                         btrfs_truncate(inode);
 951                 } else {
 952                         nr_unlink++;
 953                 }
 954
 955                 /* this will do delete_inode and everything for us */
 956                 iput(inode);
 957         }
 958
 959         if (nr_unlink)
 960                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
 961         if (nr_truncate)
 962                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
 963
 964         btrfs_free_path(path);
 965         btrfs_end_transaction(trans, root);
 966 }
 967
 968 void btrfs_read_locked_inode(struct inode *inode)
 969 {
 970         struct btrfs_path *path;
 971         struct extent_buffer *leaf;
 972         struct btrfs_inode_item *inode_item;
 973         struct btrfs_timespec *tspec;
 974         struct btrfs_root *root = BTRFS_I(inode)->root;
 975         struct btrfs_key location;
 976         u64 alloc_group_block;
 977         u32 rdev;
 978         int ret;
 979
 980         path = btrfs_alloc_path();
 981         BUG_ON(!path);
 982         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
 983
 984         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
 985         if (ret)
 986                 goto make_bad;
 987
 988         leaf = path->nodes[0];
 989         inode_item = btrfs_item_ptr(leaf, path->slots[0],
 990                                     struct btrfs_inode_item);
 991
 992         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
 993         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
 994         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
 995         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
 996         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
 997
 998         tspec = btrfs_inode_atime(inode_item);
 999         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1000         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1001
1002         tspec = btrfs_inode_mtime(inode_item);
1003         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1004         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1005
1006         tspec = btrfs_inode_ctime(inode_item);
1007         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1008         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1009
1010         inode->i_blocks = btrfs_inode_nblocks(leaf, inode_item);
1011         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
1012         inode->i_generation = BTRFS_I(inode)->generation;
1013         inode->i_rdev = 0;
1014         rdev = btrfs_inode_rdev(leaf, inode_item);
1015
1016         BTRFS_I(inode)->index_cnt = (u64)-1;
1017
1018         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
1019         BTRFS_I(inode)->block_group = btrfs_lookup_block_group(root->fs_info,
1020                                                        alloc_group_block);
1021         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
1022         if (!BTRFS_I(inode)->block_group) {
1023                 BTRFS_I(inode)->block_group = btrfs_find_block_group(root,
1024                                                  NULL, 0,
1025                                                  BTRFS_BLOCK_GROUP_METADATA, 0);
1026         }
1027         btrfs_free_path(path);
1028         inode_item = NULL;
1029
1030         switch (inode->i_mode & S_IFMT) {
1031         case S_IFREG:
1032                 inode->i_mapping->a_ops = &btrfs_aops;
1033                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
1034                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
1035                 inode->i_fop = &btrfs_file_operations;
1036                 inode->i_op = &btrfs_file_inode_operations;
1037                 break;
1038         case S_IFDIR:
1039                 inode->i_fop = &btrfs_dir_file_operations;
1040                 if (root == root->fs_info->tree_root)
1041                         inode->i_op = &btrfs_dir_ro_inode_operations;
1042                 else
1043                         inode->i_op = &btrfs_dir_inode_operations;
1044                 break;
1045         case S_IFLNK:
1046                 inode->i_op = &btrfs_symlink_inode_operations;
1047                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
1048                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
1049                 break;
1050         default:
1051                 init_special_inode(inode, inode->i_mode, rdev);
1052                 break;
1053         }
1054         return;
1055
1056 make_bad:
1057         btrfs_free_path(path);
1058         make_bad_inode(inode);
1059 }
1060
1061 static void fill_inode_item(struct btrfs_trans_handle *trans,
1062                             struct extent_buffer *leaf,
1063                             struct btrfs_inode_item *item,
1064                             struct inode *inode)
1065 {
1066         btrfs_set_inode_uid(leaf, item, inode->i_uid);
1067         btrfs_set_inode_gid(leaf, item, inode->i_gid);
1068         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
1069         btrfs_set_inode_mode(leaf, item, inode->i_mode);
1070         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
1071
1072         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
1073                                inode->i_atime.tv_sec);
1074         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
1075                                 inode->i_atime.tv_nsec);
1076
1077         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
1078                                inode->i_mtime.tv_sec);
1079         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
1080                                 inode->i_mtime.tv_nsec);
1081
1082         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
1083                                inode->i_ctime.tv_sec);
1084         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
1085                                 inode->i_ctime.tv_nsec);
1086
1087         btrfs_set_inode_nblocks(leaf, item, inode->i_blocks);
1088         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
1089         btrfs_set_inode_transid(leaf, item, trans->transid);
1090         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
1091         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
1092         btrfs_set_inode_block_group(leaf, item,
1093                                     BTRFS_I(inode)->block_group->key.objectid);
1094 }
1095
1096 int noinline btrfs_update_inode(struct btrfs_trans_handle *trans,
1097                               struct btrfs_root *root,
1098                               struct inode *inode)
1099 {
1100         struct btrfs_inode_item *inode_item;
1101         struct btrfs_path *path;
1102         struct extent_buffer *leaf;
1103         int ret;
1104
1105         path = btrfs_alloc_path();
1106         BUG_ON(!path);
1107         ret = btrfs_lookup_inode(trans, root, path,
1108                                  &BTRFS_I(inode)->location, 1);
1109         if (ret) {
1110                 if (ret > 0)
1111                         ret = -ENOENT;
1112                 goto failed;
1113         }
1114
1115         leaf = path->nodes[0];
1116         inode_item = btrfs_item_ptr(leaf, path->slots[0],
1117                                   struct btrfs_inode_item);
1118
1119         fill_inode_item(trans, leaf, inode_item, inode);
1120         btrfs_mark_buffer_dirty(leaf);
1121         btrfs_set_inode_last_trans(trans, inode);
1122         ret = 0;
1123 failed:
1124         btrfs_free_path(path);
1125         return ret;
1126 }
1127
1128
1129 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
1130                        struct btrfs_root *root,
1131                        struct inode *dir, struct inode *inode,
1132                        const char *name, int name_len)
1133 {
1134         struct btrfs_path *path;
1135         int ret = 0;
1136         struct extent_buffer *leaf;
1137         struct btrfs_dir_item *di;
1138         struct btrfs_key key;
1139         u64 index;
1140
1141         path = btrfs_alloc_path();
1142         if (!path) {
1143                 ret = -ENOMEM;
1144                 goto err;
1145         }
1146
1147         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
1148                                     name, name_len, -1);
1149         if (IS_ERR(di)) {
1150                 ret = PTR_ERR(di);
1151                 goto err;
1152         }
1153         if (!di) {
1154                 ret = -ENOENT;
1155                 goto err;
1156         }
1157         leaf = path->nodes[0];
1158         btrfs_dir_item_key_to_cpu(leaf, di, &key);
1159         ret = btrfs_delete_one_dir_name(trans, root, path, di);
1160         if (ret)
1161                 goto err;
1162         btrfs_release_path(root, path);
1163
1164         ret = btrfs_del_inode_ref(trans, root, name, name_len,
1165                                   inode->i_ino,
1166                                   dir->i_ino, &index);
1167         if (ret) {
1168                 printk("failed to delete reference to %.*s, "
1169                        "inode %lu parent %lu\n", name_len, name,
1170                        inode->i_ino, dir->i_ino);
1171                 goto err;
1172         }
1173
1174         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
1175                                          index, name, name_len, -1);
1176         if (IS_ERR(di)) {
1177                 ret = PTR_ERR(di);
1178                 goto err;
1179         }
1180         if (!di) {
1181                 ret = -ENOENT;
1182                 goto err;
1183         }
1184         ret = btrfs_delete_one_dir_name(trans, root, path, di);
1185         btrfs_release_path(root, path);
1186
1187         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
1188                                          inode, dir->i_ino);
1189         BUG_ON(ret != 0 && ret != -ENOENT);
1190         if (ret != -ENOENT)
1191                 BTRFS_I(dir)->log_dirty_trans = trans->transid;
1192
1193         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
1194                                            dir, index);
1195         BUG_ON(ret);
1196 err:
1197         btrfs_free_path(path);
1198         if (ret)
1199                 goto out;
1200
1201         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
1202         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
1203         btrfs_update_inode(trans, root, dir);
1204         btrfs_drop_nlink(inode);
1205         ret = btrfs_update_inode(trans, root, inode);
1206         dir->i_sb->s_dirt = 1;
1207 out:
1208         return ret;
1209 }
1210
1211 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
1212 {
1213         struct btrfs_root *root;
1214         struct btrfs_trans_handle *trans;
1215         struct inode *inode = dentry->d_inode;
1216         int ret;
1217         unsigned long nr = 0;
1218
1219         root = BTRFS_I(dir)->root;
1220
1221         ret = btrfs_check_free_space(root, 1, 1);
1222         if (ret)
1223                 goto fail;
1224
1225         trans = btrfs_start_transaction(root, 1);
1226
1227         btrfs_set_trans_block_group(trans, dir);
1228         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
1229                                  dentry->d_name.name, dentry->d_name.len);
1230
1231         if (inode->i_nlink == 0)
1232                 ret = btrfs_orphan_add(trans, inode);
1233
1234         nr = trans->blocks_used;
1235
1236         btrfs_end_transaction_throttle(trans, root);
1237 fail:
1238         btrfs_btree_balance_dirty(root, nr);
1239         return ret;
1240 }
1241
1242 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
1243 {
1244         struct inode *inode = dentry->d_inode;
1245         int err = 0;
1246         int ret;
1247         struct btrfs_root *root = BTRFS_I(dir)->root;
1248         struct btrfs_trans_handle *trans;
1249         unsigned long nr = 0;
1250
1251         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
1252                 return -ENOTEMPTY;
1253         }
1254
1255         ret = btrfs_check_free_space(root, 1, 1);
1256         if (ret)
1257                 goto fail;
1258
1259         trans = btrfs_start_transaction(root, 1);
1260         btrfs_set_trans_block_group(trans, dir);
1261
1262         err = btrfs_orphan_add(trans, inode);
1263         if (err)
1264                 goto fail_trans;
1265
1266         /* now the directory is empty */
1267         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
1268                                  dentry->d_name.name, dentry->d_name.len);
1269         if (!err) {
1270                 btrfs_i_size_write(inode, 0);
1271         }
1272
1273 fail_trans:
1274         nr = trans->blocks_used;
1275         ret = btrfs_end_transaction_throttle(trans, root);
1276 fail:
1277         btrfs_btree_balance_dirty(root, nr);
1278
1279         if (ret && !err)
1280                 err = ret;
1281         return err;
1282 }
1283
1284 /*
1285  * this can truncate away extent items, csum items and directory items.
1286  * It starts at a high offset and removes keys until it can't find
1287  * any higher than i_size.
1288  *
1289  * csum items that cross the new i_size are truncated to the new size
1290  * as well.
1291  *
1292  * min_type is the minimum key type to truncate down to.  If set to 0, this
1293  * will kill all the items on this inode, including the INODE_ITEM_KEY.
1294  */
1295 noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
1296                                         struct btrfs_root *root,
1297                                         struct inode *inode,
1298                                         u64 new_size, u32 min_type)
1299 {
1300         int ret;
1301         struct btrfs_path *path;
1302         struct btrfs_key key;
1303         struct btrfs_key found_key;
1304         u32 found_type;
1305         struct extent_buffer *leaf;
1306         struct btrfs_file_extent_item *fi;
1307         u64 extent_start = 0;
1308         u64 extent_num_bytes = 0;
1309         u64 item_end = 0;
1310         u64 root_gen = 0;
1311         u64 root_owner = 0;
1312         int found_extent;
1313         int del_item;
1314         int pending_del_nr = 0;
1315         int pending_del_slot = 0;
1316         int extent_type = -1;
1317         u64 mask = root->sectorsize - 1;
1318
1319         if (root->ref_cows)
1320                 btrfs_drop_extent_cache(inode,
1321                                         new_size & (~mask), (u64)-1);
1322         path = btrfs_alloc_path();
1323         path->reada = -1;
1324         BUG_ON(!path);
1325
1326         /* FIXME, add redo link to tree so we don't leak on crash */
1327         key.objectid = inode->i_ino;
1328         key.offset = (u64)-1;
1329         key.type = (u8)-1;
1330
1331         btrfs_init_path(path);
1332 search_again:
1333         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1334         if (ret < 0) {
1335                 goto error;
1336         }
1337         if (ret > 0) {
1338                 /* there are no items in the tree for us to truncate, we're
1339                  * done
1340                  */
1341                 if (path->slots[0] == 0) {
1342                         ret = 0;
1343                         goto error;
1344                 }
1345                 path->slots[0]--;
1346         }
1347
1348         while(1) {
1349                 fi = NULL;
1350                 leaf = path->nodes[0];
1351                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1352                 found_type = btrfs_key_type(&found_key);
1353
1354                 if (found_key.objectid != inode->i_ino)
1355                         break;
1356
1357                 if (found_type < min_type)
1358                         break;
1359
1360                 item_end = found_key.offset;
1361                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
1362                         fi = btrfs_item_ptr(leaf, path->slots[0],
1363                                             struct btrfs_file_extent_item);
1364                         extent_type = btrfs_file_extent_type(leaf, fi);
1365                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
1366                                 item_end +=
1367                                     btrfs_file_extent_num_bytes(leaf, fi);
1368                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1369                                 struct btrfs_item *item = btrfs_item_nr(leaf,
1370                                                                 path->slots[0]);
1371                                 item_end += btrfs_file_extent_inline_len(leaf,
1372                                                                          item);
1373                         }
1374                         item_end--;
1375                 }
1376                 if (found_type == BTRFS_CSUM_ITEM_KEY) {
1377                         ret = btrfs_csum_truncate(trans, root, path,
1378                                                   new_size);
1379                         BUG_ON(ret);
1380                 }
1381                 if (item_end < new_size) {
1382                         if (found_type == BTRFS_DIR_ITEM_KEY) {
1383                                 found_type = BTRFS_INODE_ITEM_KEY;
1384                         } else if (found_type == BTRFS_EXTENT_ITEM_KEY) {
1385                                 found_type = BTRFS_CSUM_ITEM_KEY;
1386                         } else if (found_type == BTRFS_EXTENT_DATA_KEY) {
1387                                 found_type = BTRFS_XATTR_ITEM_KEY;
1388                         } else if (found_type == BTRFS_XATTR_ITEM_KEY) {
1389                                 found_type = BTRFS_INODE_REF_KEY;
1390                         } else if (found_type) {
1391                                 found_type--;
1392                         } else {
1393                                 break;
1394                         }
1395                         btrfs_set_key_type(&key, found_type);
1396                         goto next;
1397                 }
1398                 if (found_key.offset >= new_size)
1399                         del_item = 1;
1400                 else
1401                         del_item = 0;
1402                 found_extent = 0;
1403
1404                 /* FIXME, shrink the extent if the ref count is only 1 */
1405                 if (found_type != BTRFS_EXTENT_DATA_KEY)
1406                         goto delete;
1407
1408                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
1409                         u64 num_dec;
1410                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
1411                         if (!del_item) {
1412                                 u64 orig_num_bytes =
1413                                         btrfs_file_extent_num_bytes(leaf, fi);
1414                                 extent_num_bytes = new_size -
1415                                         found_key.offset + root->sectorsize - 1;
1416                                 extent_num_bytes = extent_num_bytes &
1417                                         ~((u64)root->sectorsize - 1);
1418                                 btrfs_set_file_extent_num_bytes(leaf, fi,
1419                                                          extent_num_bytes);
1420                                 num_dec = (orig_num_bytes -
1421                                            extent_num_bytes);
1422                                 if (root->ref_cows && extent_start != 0)
1423                                         dec_i_blocks(inode, num_dec);
1424                                 btrfs_mark_buffer_dirty(leaf);
1425                         } else {
1426                                 extent_num_bytes =
1427                                         btrfs_file_extent_disk_num_bytes(leaf,
1428                                                                          fi);
1429                                 /* FIXME blocksize != 4096 */
1430                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
1431                                 if (extent_start != 0) {
1432                                         found_extent = 1;
1433                                         if (root->ref_cows)
1434                                                 dec_i_blocks(inode, num_dec);
1435                                 }
1436                                 if (root->ref_cows) {
1437                                         root_gen =
1438                                                 btrfs_header_generation(leaf);
1439                                 }
1440                                 root_owner = btrfs_header_owner(leaf);
1441                         }
1442                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1443                         if (!del_item) {
1444                                 u32 size = new_size - found_key.offset;
1445
1446                                 if (root->ref_cows) {
1447                                         dec_i_blocks(inode, item_end + 1 -
1448                                                     found_key.offset - size);
1449                                 }
1450                                 size =
1451                                     btrfs_file_extent_calc_inline_size(size);
1452                                 ret = btrfs_truncate_item(trans, root, path,
1453                                                           size, 1);
1454                                 BUG_ON(ret);
1455                         } else if (root->ref_cows) {
1456                                 dec_i_blocks(inode, item_end + 1 -
1457                                              found_key.offset);
1458                         }
1459                 }
1460 delete:
1461                 if (del_item) {
1462                         if (!pending_del_nr) {
1463                                 /* no pending yet, add ourselves */
1464                                 pending_del_slot = path->slots[0];
1465                                 pending_del_nr = 1;
1466                         } else if (pending_del_nr &&
1467                                    path->slots[0] + 1 == pending_del_slot) {
1468                                 /* hop on the pending chunk */
1469                                 pending_del_nr++;
1470                                 pending_del_slot = path->slots[0];
1471                         } else {
1472                                 printk("bad pending slot %d pending_del_nr %d pending_del_slot %d\n", path->slots[0], pending_del_nr, pending_del_slot);
1473                         }
1474                 } else {
1475                         break;
1476                 }
1477                 if (found_extent) {
1478                         ret = btrfs_free_extent(trans, root, extent_start,
1479                                                 extent_num_bytes,
1480                                                 root_owner,
1481                                                 root_gen, inode->i_ino,
1482                                                 found_key.offset, 0);
1483                         BUG_ON(ret);
1484                 }
1485 next:
1486                 if (path->slots[0] == 0) {
1487                         if (pending_del_nr)
1488                                 goto del_pending;
1489                         btrfs_release_path(root, path);
1490                         goto search_again;
1491                 }
1492
1493                 path->slots[0]--;
1494                 if (pending_del_nr &&
1495                     path->slots[0] + 1 != pending_del_slot) {
1496                         struct btrfs_key debug;
1497 del_pending:
1498                         btrfs_item_key_to_cpu(path->nodes[0], &debug,
1499                                               pending_del_slot);
1500                         ret = btrfs_del_items(trans, root, path,
1501                                               pending_del_slot,
1502                                               pending_del_nr);
1503                         BUG_ON(ret);
1504                         pending_del_nr = 0;
1505                         btrfs_release_path(root, path);
1506                         goto search_again;
1507                 }
1508         }
1509         ret = 0;
1510 error:
1511         if (pending_del_nr) {
1512                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
1513                                       pending_del_nr);
1514         }
1515         btrfs_free_path(path);
1516         inode->i_sb->s_dirt = 1;
1517         return ret;
1518 }
1519
1520 /*
1521  * taken from block_truncate_page, but does cow as it zeros out
1522  * any bytes left in the last page in the file.
1523  */
1524 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
1525 {
1526         struct inode *inode = mapping->host;
1527         struct btrfs_root *root = BTRFS_I(inode)->root;
1528         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1529         struct btrfs_ordered_extent *ordered;
1530         char *kaddr;
1531         u32 blocksize = root->sectorsize;
1532         pgoff_t index = from >> PAGE_CACHE_SHIFT;
1533         unsigned offset = from & (PAGE_CACHE_SIZE-1);
1534         struct page *page;
1535         int ret = 0;
1536         u64 page_start;
1537         u64 page_end;
1538
1539         if ((offset & (blocksize - 1)) == 0)
1540                 goto out;
1541
1542         ret = -ENOMEM;
1543 again:
1544         page = grab_cache_page(mapping, index);
1545         if (!page)
1546                 goto out;
1547
1548         page_start = page_offset(page);
1549         page_end = page_start + PAGE_CACHE_SIZE - 1;
1550
1551         if (!PageUptodate(page)) {
1552                 ret = btrfs_readpage(NULL, page);
1553                 lock_page(page);
1554                 if (page->mapping != mapping) {
1555                         unlock_page(page);
1556                         page_cache_release(page);
1557                         goto again;
1558                 }
1559                 if (!PageUptodate(page)) {
1560                         ret = -EIO;
1561                         goto out_unlock;
1562                 }
1563         }
1564         wait_on_page_writeback(page);
1565
1566         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
1567         set_page_extent_mapped(page);
1568
1569         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1570         if (ordered) {
1571                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
1572                 unlock_page(page);
1573                 page_cache_release(page);
1574                 btrfs_start_ordered_extent(inode, ordered, 1);
1575                 btrfs_put_ordered_extent(ordered);
1576                 goto again;
1577         }
1578
1579         btrfs_set_extent_delalloc(inode, page_start, page_end);
1580         ret = 0;
1581         if (offset != PAGE_CACHE_SIZE) {
1582                 kaddr = kmap(page);
1583                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
1584                 flush_dcache_page(page);
1585                 kunmap(page);
1586         }
1587         ClearPageChecked(page);
1588         set_page_dirty(page);
1589         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
1590
1591 out_unlock:
1592         unlock_page(page);
1593         page_cache_release(page);
1594 out:
1595         return ret;
1596 }
1597
1598 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
1599 {
1600         struct inode *inode = dentry->d_inode;
1601         int err;
1602
1603         err = inode_change_ok(inode, attr);
1604         if (err)
1605                 return err;
1606
1607         if (S_ISREG(inode->i_mode) &&
1608             attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
1609                 struct btrfs_trans_handle *trans;
1610                 struct btrfs_root *root = BTRFS_I(inode)->root;
1611                 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1612
1613                 u64 mask = root->sectorsize - 1;
1614                 u64 hole_start = (inode->i_size + mask) & ~mask;
1615                 u64 block_end = (attr->ia_size + mask) & ~mask;
1616                 u64 hole_size;
1617                 u64 alloc_hint = 0;
1618
1619                 if (attr->ia_size <= hole_start)
1620                         goto out;
1621
1622                 err = btrfs_check_free_space(root, 1, 0);
1623                 if (err)
1624                         goto fail;
1625
1626                 btrfs_truncate_page(inode->i_mapping, inode->i_size);
1627
1628                 hole_size = block_end - hole_start;
1629                 while(1) {
1630                         struct btrfs_ordered_extent *ordered;
1631                         btrfs_wait_ordered_range(inode, hole_start, hole_size);
1632
1633                         lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
1634                         ordered = btrfs_lookup_ordered_extent(inode, hole_start);
1635                         if (ordered) {
1636                                 unlock_extent(io_tree, hole_start,
1637                                               block_end - 1, GFP_NOFS);
1638                                 btrfs_put_ordered_extent(ordered);
1639                         } else {
1640                                 break;
1641                         }
1642                 }
1643
1644                 trans = btrfs_start_transaction(root, 1);
1645                 btrfs_set_trans_block_group(trans, inode);
1646                 mutex_lock(&BTRFS_I(inode)->extent_mutex);
1647                 err = btrfs_drop_extents(trans, root, inode,
1648                                          hole_start, block_end, hole_start,
1649                                          &alloc_hint);
1650
1651                 if (alloc_hint != EXTENT_MAP_INLINE) {
1652                         err = btrfs_insert_file_extent(trans, root,
1653                                                        inode->i_ino,
1654                                                        hole_start, 0, 0,
1655                                                        hole_size, 0);
1656                         btrfs_drop_extent_cache(inode, hole_start,
1657                                                 (u64)-1);
1658                         btrfs_check_file(root, inode);
1659                 }
1660                 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
1661                 btrfs_end_transaction(trans, root);
1662                 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
1663                 if (err)
1664                         return err;
1665         }
1666 out:
1667         err = inode_setattr(inode, attr);
1668
1669         if (!err && ((attr->ia_valid & ATTR_MODE)))
1670                 err = btrfs_acl_chmod(inode);
1671 fail:
1672         return err;
1673 }
1674
1675 void btrfs_delete_inode(struct inode *inode)
1676 {
1677         struct btrfs_trans_handle *trans;
1678         struct btrfs_root *root = BTRFS_I(inode)->root;
1679         unsigned long nr;
1680         int ret;
1681
1682         truncate_inode_pages(&inode->i_data, 0);
1683         if (is_bad_inode(inode)) {
1684                 btrfs_orphan_del(NULL, inode);
1685                 goto no_delete;
1686         }
1687         btrfs_wait_ordered_range(inode, 0, (u64)-1);
1688
1689         btrfs_i_size_write(inode, 0);
1690         trans = btrfs_start_transaction(root, 1);
1691
1692         btrfs_set_trans_block_group(trans, inode);
1693         ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0);
1694         if (ret) {
1695                 btrfs_orphan_del(NULL, inode);
1696                 goto no_delete_lock;
1697         }
1698
1699         btrfs_orphan_del(trans, inode);
1700
1701         nr = trans->blocks_used;
1702         clear_inode(inode);
1703
1704         btrfs_end_transaction(trans, root);
1705         btrfs_btree_balance_dirty(root, nr);
1706         return;
1707
1708 no_delete_lock:
1709         nr = trans->blocks_used;
1710         btrfs_end_transaction(trans, root);
1711         btrfs_btree_balance_dirty(root, nr);
1712 no_delete:
1713         clear_inode(inode);
1714 }
1715
1716 /*
1717  * this returns the key found in the dir entry in the location pointer.
1718  * If no dir entries were found, location->objectid is 0.
1719  */
1720 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
1721                                struct btrfs_key *location)
1722 {
1723         const char *name = dentry->d_name.name;
1724         int namelen = dentry->d_name.len;
1725         struct btrfs_dir_item *di;
1726         struct btrfs_path *path;
1727         struct btrfs_root *root = BTRFS_I(dir)->root;
1728         int ret = 0;
1729
1730         path = btrfs_alloc_path();
1731         BUG_ON(!path);
1732
1733         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
1734                                     namelen, 0);
1735         if (IS_ERR(di))
1736                 ret = PTR_ERR(di);
1737         if (!di || IS_ERR(di)) {
1738                 goto out_err;
1739         }
1740         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
1741 out:
1742         btrfs_free_path(path);
1743         return ret;
1744 out_err:
1745         location->objectid = 0;
1746         goto out;
1747 }
1748
1749 /*
1750  * when we hit a tree root in a directory, the btrfs part of the inode
1751  * needs to be changed to reflect the root directory of the tree root.  This
1752  * is kind of like crossing a mount point.
1753  */
1754 static int fixup_tree_root_location(struct btrfs_root *root,
1755                              struct btrfs_key *location,
1756                              struct btrfs_root **sub_root,
1757                              struct dentry *dentry)
1758 {
1759         struct btrfs_root_item *ri;
1760
1761         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
1762                 return 0;
1763         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1764                 return 0;
1765
1766         *sub_root = btrfs_read_fs_root(root->fs_info, location,
1767                                         dentry->d_name.name,
1768                                         dentry->d_name.len);
1769         if (IS_ERR(*sub_root))
1770                 return PTR_ERR(*sub_root);
1771
1772         ri = &(*sub_root)->root_item;
1773         location->objectid = btrfs_root_dirid(ri);
1774         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1775         location->offset = 0;
1776
1777         return 0;
1778 }
1779
1780 static noinline void init_btrfs_i(struct inode *inode)
1781 {
1782         struct btrfs_inode *bi = BTRFS_I(inode);
1783
1784         bi->i_acl = NULL;
1785         bi->i_default_acl = NULL;
1786
1787         bi->generation = 0;
1788         bi->last_trans = 0;
1789         bi->logged_trans = 0;
1790         bi->delalloc_bytes = 0;
1791         bi->disk_i_size = 0;
1792         bi->flags = 0;
1793         bi->index_cnt = (u64)-1;
1794         bi->log_dirty_trans = 0;
1795         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
1796         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
1797                              inode->i_mapping, GFP_NOFS);
1798         extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
1799                              inode->i_mapping, GFP_NOFS);
1800         INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
1801         btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
1802         mutex_init(&BTRFS_I(inode)->csum_mutex);
1803         mutex_init(&BTRFS_I(inode)->extent_mutex);
1804         mutex_init(&BTRFS_I(inode)->log_mutex);
1805 }
1806
1807 static int btrfs_init_locked_inode(struct inode *inode, void *p)
1808 {
1809         struct btrfs_iget_args *args = p;
1810         inode->i_ino = args->ino;
1811         init_btrfs_i(inode);
1812         BTRFS_I(inode)->root = args->root;
1813         return 0;
1814 }
1815
1816 static int btrfs_find_actor(struct inode *inode, void *opaque)
1817 {
1818         struct btrfs_iget_args *args = opaque;
1819         return (args->ino == inode->i_ino &&
1820                 args->root == BTRFS_I(inode)->root);
1821 }
1822
1823 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
1824                                 struct btrfs_root *root)
1825 {
1826         struct inode *inode;
1827         struct btrfs_iget_args args;
1828         args.ino = objectid;
1829         args.root = root;
1830
1831         inode = iget5_locked(s, objectid, btrfs_find_actor,
1832                              btrfs_init_locked_inode,
1833                              (void *)&args);
1834         return inode;
1835 }
1836
1837 /* Get an inode object given its location and corresponding root.
1838  * Returns in *is_new if the inode was read from disk
1839  */
1840 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
1841                          struct btrfs_root *root, int *is_new)
1842 {
1843         struct inode *inode;
1844
1845         inode = btrfs_iget_locked(s, location->objectid, root);
1846         if (!inode)
1847                 return ERR_PTR(-EACCES);
1848
1849         if (inode->i_state & I_NEW) {
1850                 BTRFS_I(inode)->root = root;
1851                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
1852                 btrfs_read_locked_inode(inode);
1853                 unlock_new_inode(inode);
1854                 if (is_new)
1855                         *is_new = 1;
1856         } else {
1857                 if (is_new)
1858                         *is_new = 0;
1859         }
1860
1861         return inode;
1862 }
1863
1864 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
1865                                    struct nameidata *nd)
1866 {
1867         struct inode * inode;
1868         struct btrfs_inode *bi = BTRFS_I(dir);
1869         struct btrfs_root *root = bi->root;
1870         struct btrfs_root *sub_root = root;
1871         struct btrfs_key location;
1872         int ret, new, do_orphan = 0;
1873
1874         if (dentry->d_name.len > BTRFS_NAME_LEN)
1875                 return ERR_PTR(-ENAMETOOLONG);
1876
1877         ret = btrfs_inode_by_name(dir, dentry, &location);
1878
1879         if (ret < 0)
1880                 return ERR_PTR(ret);
1881
1882         inode = NULL;
1883         if (location.objectid) {
1884                 ret = fixup_tree_root_location(root, &location, &sub_root,
1885                                                 dentry);
1886                 if (ret < 0)
1887                         return ERR_PTR(ret);
1888                 if (ret > 0)
1889                         return ERR_PTR(-ENOENT);
1890                 inode = btrfs_iget(dir->i_sb, &location, sub_root, &new);
1891                 if (IS_ERR(inode))
1892                         return ERR_CAST(inode);
1893
1894                 /* the inode and parent dir are two different roots */
1895                 if (new && root != sub_root) {
1896                         igrab(inode);
1897                         sub_root->inode = inode;
1898                         do_orphan = 1;
1899                 }
1900         }
1901
1902         if (unlikely(do_orphan))
1903                 btrfs_orphan_cleanup(sub_root);
1904
1905         return d_splice_alias(inode, dentry);
1906 }
1907
1908 static unsigned char btrfs_filetype_table[] = {
1909         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
1910 };
1911
1912 static int btrfs_real_readdir(struct file *filp, void *dirent,
1913                               filldir_t filldir)
1914 {
1915         struct inode *inode = filp->f_dentry->d_inode;
1916         struct btrfs_root *root = BTRFS_I(inode)->root;
1917         struct btrfs_item *item;
1918         struct btrfs_dir_item *di;
1919         struct btrfs_key key;
1920         struct btrfs_key found_key;
1921         struct btrfs_path *path;
1922         int ret;
1923         u32 nritems;
1924         struct extent_buffer *leaf;
1925         int slot;
1926         int advance;
1927         unsigned char d_type;
1928         int over = 0;
1929         u32 di_cur;
1930         u32 di_total;
1931         u32 di_len;
1932         int key_type = BTRFS_DIR_INDEX_KEY;
1933         char tmp_name[32];
1934         char *name_ptr;
1935         int name_len;
1936
1937         /* FIXME, use a real flag for deciding about the key type */
1938         if (root->fs_info->tree_root == root)
1939                 key_type = BTRFS_DIR_ITEM_KEY;
1940
1941         /* special case for "." */
1942         if (filp->f_pos == 0) {
1943                 over = filldir(dirent, ".", 1,
1944                                1, inode->i_ino,
1945                                DT_DIR);
1946                 if (over)
1947                         return 0;
1948                 filp->f_pos = 1;
1949         }
1950         /* special case for .., just use the back ref */
1951         if (filp->f_pos == 1) {
1952                 u64 pino = parent_ino(filp->f_path.dentry);
1953                 over = filldir(dirent, "..", 2,
1954                                2, pino, DT_DIR);
1955                 if (over)
1956                         return 0;
1957                 filp->f_pos = 2;
1958         }
1959
1960         path = btrfs_alloc_path();
1961         path->reada = 2;
1962
1963         btrfs_set_key_type(&key, key_type);
1964         key.offset = filp->f_pos;
1965         key.objectid = inode->i_ino;
1966
1967         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1968         if (ret < 0)
1969                 goto err;
1970         advance = 0;
1971
1972         while (1) {
1973                 leaf = path->nodes[0];
1974                 nritems = btrfs_header_nritems(leaf);
1975                 slot = path->slots[0];
1976                 if (advance || slot >= nritems) {
1977                         if (slot >= nritems - 1) {
1978                                 ret = btrfs_next_leaf(root, path);
1979                                 if (ret)
1980                                         break;
1981                                 leaf = path->nodes[0];
1982                                 nritems = btrfs_header_nritems(leaf);
1983                                 slot = path->slots[0];
1984                         } else {
1985                                 slot++;
1986                                 path->slots[0]++;
1987                         }
1988                 }
1989                 advance = 1;
1990                 item = btrfs_item_nr(leaf, slot);
1991                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1992
1993                 if (found_key.objectid != key.objectid)
1994                         break;
1995                 if (btrfs_key_type(&found_key) != key_type)
1996                         break;
1997                 if (found_key.offset < filp->f_pos)
1998                         continue;
1999
2000                 filp->f_pos = found_key.offset;
2001
2002                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
2003                 di_cur = 0;
2004                 di_total = btrfs_item_size(leaf, item);
2005
2006                 while (di_cur < di_total) {
2007                         struct btrfs_key location;
2008
2009                         name_len = btrfs_dir_name_len(leaf, di);
2010                         if (name_len <= sizeof(tmp_name)) {
2011                                 name_ptr = tmp_name;
2012                         } else {
2013                                 name_ptr = kmalloc(name_len, GFP_NOFS);
2014                                 if (!name_ptr) {
2015                                         ret = -ENOMEM;
2016                                         goto err;
2017                                 }
2018                         }
2019                         read_extent_buffer(leaf, name_ptr,
2020                                            (unsigned long)(di + 1), name_len);
2021
2022                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
2023                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
2024                         over = filldir(dirent, name_ptr, name_len,
2025                                        found_key.offset, location.objectid,
2026                                        d_type);
2027
2028                         if (name_ptr != tmp_name)
2029                                 kfree(name_ptr);
2030
2031                         if (over)
2032                                 goto nopos;
2033
2034                         di_len = btrfs_dir_name_len(leaf, di) +
2035                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
2036                         di_cur += di_len;
2037                         di = (struct btrfs_dir_item *)((char *)di + di_len);
2038                 }
2039         }
2040
2041         /* Reached end of directory/root. Bump pos past the last item. */
2042         if (key_type == BTRFS_DIR_INDEX_KEY)
2043                 filp->f_pos = INT_LIMIT(typeof(filp->f_pos));
2044         else
2045                 filp->f_pos++;
2046 nopos:
2047         ret = 0;
2048 err:
2049         btrfs_free_path(path);
2050         return ret;
2051 }
2052
2053 /* Kernels earlier than 2.6.28 still have the NFS deadlock where nfsd
2054    will call the file system's ->lookup() method from within its
2055    filldir callback, which in turn was called from the file system's
2056    ->readdir() method. And will deadlock for many file systems. */
2057 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,28)
2058
2059 struct nfshack_dirent {
2060         u64             ino;
2061         loff_t          offset;
2062         int             namlen;
2063         unsigned int    d_type;
2064         char            name[];
2065 };
2066
2067 struct nfshack_readdir {
2068         char            *dirent;
2069         size_t          used;
2070         int             full;
2071 };
2072
2073
2074
2075 static int btrfs_nfshack_filldir(void *__buf, const char *name, int namlen,
2076                               loff_t offset, u64 ino, unsigned int d_type)
2077 {
2078         struct nfshack_readdir *buf = __buf;
2079         struct nfshack_dirent *de = (void *)(buf->dirent + buf->used);
2080         unsigned int reclen;
2081
2082         reclen = ALIGN(sizeof(struct nfshack_dirent) + namlen, sizeof(u64));
2083         if (buf->used + reclen > PAGE_SIZE) {
2084                 buf->full = 1;
2085                 return -EINVAL;
2086         }
2087
2088         de->namlen = namlen;
2089         de->offset = offset;
2090         de->ino = ino;
2091         de->d_type = d_type;
2092         memcpy(de->name, name, namlen);
2093         buf->used += reclen;
2094
2095         return 0;
2096 }
2097
2098 static int btrfs_nfshack_readdir(struct file *file, void *dirent,
2099                                  filldir_t filldir)
2100 {
2101         struct nfshack_readdir buf;
2102         struct nfshack_dirent *de;
2103         int err;
2104         int size;
2105         loff_t offset;
2106
2107         buf.dirent = (void *)__get_free_page(GFP_KERNEL);
2108         if (!buf.dirent)
2109                 return -ENOMEM;
2110
2111         offset = file->f_pos;
2112
2113         do {
2114                 unsigned int reclen;
2115
2116                 buf.used = 0;
2117                 buf.full = 0;
2118                 err = btrfs_real_readdir(file, &buf, btrfs_nfshack_filldir);
2119                 if (err)
2120                         break;
2121
2122                 size = buf.used;
2123
2124                 if (!size)
2125                         break;
2126
2127                 de = (struct nfshack_dirent *)buf.dirent;
2128                 while (size > 0) {
2129                         offset = de->offset;
2130
2131                         if (filldir(dirent, de->name, de->namlen, de->offset,
2132                                     de->ino, de->d_type))
2133                                 goto done;
2134                         offset = file->f_pos;
2135
2136                         reclen = ALIGN(sizeof(*de) + de->namlen,
2137                                        sizeof(u64));
2138                         size -= reclen;
2139                         de = (struct nfshack_dirent *)((char *)de + reclen);
2140                 }
2141         } while (buf.full);
2142
2143  done:
2144         free_page((unsigned long)buf.dirent);
2145         file->f_pos = offset;
2146
2147         return err;
2148 }
2149 #endif
2150
2151 int btrfs_write_inode(struct inode *inode, int wait)
2152 {
2153         struct btrfs_root *root = BTRFS_I(inode)->root;
2154         struct btrfs_trans_handle *trans;
2155         int ret = 0;
2156
2157         if (root->fs_info->closing > 1)
2158                 return 0;
2159
2160         if (wait) {
2161                 trans = btrfs_join_transaction(root, 1);
2162                 btrfs_set_trans_block_group(trans, inode);
2163                 ret = btrfs_commit_transaction(trans, root);
2164         }
2165         return ret;
2166 }
2167
2168 /*
2169  * This is somewhat expensive, updating the tree every time the
2170  * inode changes.  But, it is most likely to find the inode in cache.
2171  * FIXME, needs more benchmarking...there are no reasons other than performance
2172  * to keep or drop this code.
2173  */
2174 void btrfs_dirty_inode(struct inode *inode)
2175 {
2176         struct btrfs_root *root = BTRFS_I(inode)->root;
2177         struct btrfs_trans_handle *trans;
2178
2179         trans = btrfs_join_transaction(root, 1);
2180         btrfs_set_trans_block_group(trans, inode);
2181         btrfs_update_inode(trans, root, inode);
2182         btrfs_end_transaction(trans, root);
2183 }
2184
2185 static int btrfs_set_inode_index_count(struct inode *inode)
2186 {
2187         struct btrfs_root *root = BTRFS_I(inode)->root;
2188         struct btrfs_key key, found_key;
2189         struct btrfs_path *path;
2190         struct extent_buffer *leaf;
2191         int ret;
2192
2193         key.objectid = inode->i_ino;
2194         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
2195         key.offset = (u64)-1;
2196
2197         path = btrfs_alloc_path();
2198         if (!path)
2199                 return -ENOMEM;
2200
2201         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2202         if (ret < 0)
2203                 goto out;
2204         /* FIXME: we should be able to handle this */
2205         if (ret == 0)
2206                 goto out;
2207         ret = 0;
2208
2209         /*
2210          * MAGIC NUMBER EXPLANATION:
2211          * since we search a directory based on f_pos we have to start at 2
2212          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
2213          * else has to start at 2
2214          */
2215         if (path->slots[0] == 0) {
2216                 BTRFS_I(inode)->index_cnt = 2;
2217                 goto out;
2218         }
2219
2220         path->slots[0]--;
2221
2222         leaf = path->nodes[0];
2223         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2224
2225         if (found_key.objectid != inode->i_ino ||
2226             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
2227                 BTRFS_I(inode)->index_cnt = 2;
2228                 goto out;
2229         }
2230
2231         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
2232 out:
2233         btrfs_free_path(path);
2234         return ret;
2235 }
2236
2237 static int btrfs_set_inode_index(struct inode *dir, struct inode *inode,
2238                                  u64 *index)
2239 {
2240         int ret = 0;
2241
2242         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
2243                 ret = btrfs_set_inode_index_count(dir);
2244                 if (ret) {
2245                         return ret;
2246                 }
2247         }
2248
2249         *index = BTRFS_I(dir)->index_cnt;
2250         BTRFS_I(dir)->index_cnt++;
2251
2252         return ret;
2253 }
2254
2255 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
2256                                      struct btrfs_root *root,
2257                                      struct inode *dir,
2258                                      const char *name, int name_len,
2259                                      u64 ref_objectid,
2260                                      u64 objectid,
2261                                      struct btrfs_block_group_cache *group,
2262                                      int mode, u64 *index)
2263 {
2264         struct inode *inode;
2265         struct btrfs_inode_item *inode_item;
2266         struct btrfs_block_group_cache *new_inode_group;
2267         struct btrfs_key *location;
2268         struct btrfs_path *path;
2269         struct btrfs_inode_ref *ref;
2270         struct btrfs_key key[2];
2271         u32 sizes[2];
2272         unsigned long ptr;
2273         int ret;
2274         int owner;
2275
2276         path = btrfs_alloc_path();
2277         BUG_ON(!path);
2278
2279         inode = new_inode(root->fs_info->sb);
2280         if (!inode)
2281                 return ERR_PTR(-ENOMEM);
2282
2283         if (dir) {
2284                 ret = btrfs_set_inode_index(dir, inode, index);
2285                 if (ret)
2286                         return ERR_PTR(ret);
2287         }
2288         /*
2289          * index_cnt is ignored for everything but a dir,
2290          * btrfs_get_inode_index_count has an explanation for the magic
2291          * number
2292          */
2293         init_btrfs_i(inode);
2294         BTRFS_I(inode)->index_cnt = 2;
2295         BTRFS_I(inode)->root = root;
2296         BTRFS_I(inode)->generation = trans->transid;
2297
2298         if (mode & S_IFDIR)
2299                 owner = 0;
2300         else
2301                 owner = 1;
2302         new_inode_group = btrfs_find_block_group(root, group, 0,
2303                                        BTRFS_BLOCK_GROUP_METADATA, owner);
2304         if (!new_inode_group) {
2305                 printk("find_block group failed\n");
2306                 new_inode_group = group;
2307         }
2308         BTRFS_I(inode)->block_group = new_inode_group;
2309
2310         key[0].objectid = objectid;
2311         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
2312         key[0].offset = 0;
2313
2314         key[1].objectid = objectid;
2315         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
2316         key[1].offset = ref_objectid;
2317
2318         sizes[0] = sizeof(struct btrfs_inode_item);
2319         sizes[1] = name_len + sizeof(*ref);
2320
2321         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
2322         if (ret != 0)
2323                 goto fail;
2324
2325         if (objectid > root->highest_inode)
2326                 root->highest_inode = objectid;
2327
2328         inode->i_uid = current->fsuid;
2329         inode->i_gid = current->fsgid;
2330         inode->i_mode = mode;
2331         inode->i_ino = objectid;
2332         inode->i_blocks = 0;
2333         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2334         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2335                                   struct btrfs_inode_item);
2336         fill_inode_item(trans, path->nodes[0], inode_item, inode);
2337
2338         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
2339                              struct btrfs_inode_ref);
2340         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
2341         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
2342         ptr = (unsigned long)(ref + 1);
2343         write_extent_buffer(path->nodes[0], name, ptr, name_len);
2344
2345         btrfs_mark_buffer_dirty(path->nodes[0]);
2346         btrfs_free_path(path);
2347
2348         location = &BTRFS_I(inode)->location;
2349         location->objectid = objectid;
2350         location->offset = 0;
2351         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
2352
2353         insert_inode_hash(inode);
2354         return inode;
2355 fail:
2356         if (dir)
2357                 BTRFS_I(dir)->index_cnt--;
2358         btrfs_free_path(path);
2359         return ERR_PTR(ret);
2360 }
2361
2362 static inline u8 btrfs_inode_type(struct inode *inode)
2363 {
2364         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
2365 }
2366
2367 int btrfs_add_link(struct btrfs_trans_handle *trans,
2368                    struct inode *parent_inode, struct inode *inode,
2369                    const char *name, int name_len, int add_backref, u64 index)
2370 {
2371         int ret;
2372         struct btrfs_key key;
2373         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
2374
2375         key.objectid = inode->i_ino;
2376         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
2377         key.offset = 0;
2378
2379         ret = btrfs_insert_dir_item(trans, root, name, name_len,
2380                                     parent_inode->i_ino,
2381                                     &key, btrfs_inode_type(inode),
2382                                     index);
2383         if (ret == 0) {
2384                 if (add_backref) {
2385                         ret = btrfs_insert_inode_ref(trans, root,
2386                                                      name, name_len,
2387                                                      inode->i_ino,
2388                                                      parent_inode->i_ino,
2389                                                      index);
2390                 }
2391                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
2392                                    name_len * 2);
2393                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
2394                 ret = btrfs_update_inode(trans, root, parent_inode);
2395         }
2396         return ret;
2397 }
2398
2399 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
2400                             struct dentry *dentry, struct inode *inode,
2401                             int backref, u64 index)
2402 {
2403         int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
2404                                  inode, dentry->d_name.name,
2405                                  dentry->d_name.len, backref, index);
2406         if (!err) {
2407                 d_instantiate(dentry, inode);
2408                 return 0;
2409         }
2410         if (err > 0)
2411                 err = -EEXIST;
2412         return err;
2413 }
2414
2415 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
2416                         int mode, dev_t rdev)
2417 {
2418         struct btrfs_trans_handle *trans;
2419         struct btrfs_root *root = BTRFS_I(dir)->root;
2420         struct inode *inode = NULL;
2421         int err;
2422         int drop_inode = 0;
2423         u64 objectid;
2424         unsigned long nr = 0;
2425         u64 index = 0;
2426
2427         if (!new_valid_dev(rdev))
2428                 return -EINVAL;
2429
2430         err = btrfs_check_free_space(root, 1, 0);
2431         if (err)
2432                 goto fail;
2433
2434         trans = btrfs_start_transaction(root, 1);
2435         btrfs_set_trans_block_group(trans, dir);
2436
2437         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2438         if (err) {
2439                 err = -ENOSPC;
2440                 goto out_unlock;
2441         }
2442
2443         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
2444                                 dentry->d_name.len,
2445                                 dentry->d_parent->d_inode->i_ino, objectid,
2446                                 BTRFS_I(dir)->block_group, mode, &index);
2447         err = PTR_ERR(inode);
2448         if (IS_ERR(inode))
2449                 goto out_unlock;
2450
2451         err = btrfs_init_acl(inode, dir);
2452         if (err) {
2453                 drop_inode = 1;
2454                 goto out_unlock;
2455         }
2456
2457         btrfs_set_trans_block_group(trans, inode);
2458         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
2459         if (err)
2460                 drop_inode = 1;
2461         else {
2462                 inode->i_op = &btrfs_special_inode_operations;
2463                 init_special_inode(inode, inode->i_mode, rdev);
2464                 btrfs_update_inode(trans, root, inode);
2465         }
2466         dir->i_sb->s_dirt = 1;
2467         btrfs_update_inode_block_group(trans, inode);
2468         btrfs_update_inode_block_group(trans, dir);
2469 out_unlock:
2470         nr = trans->blocks_used;
2471         btrfs_end_transaction_throttle(trans, root);
2472 fail:
2473         if (drop_inode) {
2474                 inode_dec_link_count(inode);
2475                 iput(inode);
2476         }
2477         btrfs_btree_balance_dirty(root, nr);
2478         return err;
2479 }
2480
2481 static int btrfs_create(struct inode *dir, struct dentry *dentry,
2482                         int mode, struct nameidata *nd)
2483 {
2484         struct btrfs_trans_handle *trans;
2485         struct btrfs_root *root = BTRFS_I(dir)->root;
2486         struct inode *inode = NULL;
2487         int err;
2488         int drop_inode = 0;
2489         unsigned long nr = 0;
2490         u64 objectid;
2491         u64 index = 0;
2492
2493         err = btrfs_check_free_space(root, 1, 0);
2494         if (err)
2495                 goto fail;
2496         trans = btrfs_start_transaction(root, 1);
2497         btrfs_set_trans_block_group(trans, dir);
2498
2499         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2500         if (err) {
2501                 err = -ENOSPC;
2502                 goto out_unlock;
2503         }
2504
2505         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
2506                                 dentry->d_name.len,
2507                                 dentry->d_parent->d_inode->i_ino,
2508                                 objectid, BTRFS_I(dir)->block_group, mode,
2509                                 &index);
2510         err = PTR_ERR(inode);
2511         if (IS_ERR(inode))
2512                 goto out_unlock;
2513
2514         err = btrfs_init_acl(inode, dir);
2515         if (err) {
2516                 drop_inode = 1;
2517                 goto out_unlock;
2518         }
2519
2520         btrfs_set_trans_block_group(trans, inode);
2521         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
2522         if (err)
2523                 drop_inode = 1;
2524         else {
2525                 inode->i_mapping->a_ops = &btrfs_aops;
2526                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2527                 inode->i_fop = &btrfs_file_operations;
2528                 inode->i_op = &btrfs_file_inode_operations;
2529                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2530         }
2531         dir->i_sb->s_dirt = 1;
2532         btrfs_update_inode_block_group(trans, inode);
2533         btrfs_update_inode_block_group(trans, dir);
2534 out_unlock:
2535         nr = trans->blocks_used;
2536         btrfs_end_transaction_throttle(trans, root);
2537 fail:
2538         if (drop_inode) {
2539                 inode_dec_link_count(inode);
2540                 iput(inode);
2541         }
2542         btrfs_btree_balance_dirty(root, nr);
2543         return err;
2544 }
2545
2546 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
2547                       struct dentry *dentry)
2548 {
2549         struct btrfs_trans_handle *trans;
2550         struct btrfs_root *root = BTRFS_I(dir)->root;
2551         struct inode *inode = old_dentry->d_inode;
2552         u64 index;
2553         unsigned long nr = 0;
2554         int err;
2555         int drop_inode = 0;
2556
2557         if (inode->i_nlink == 0)
2558                 return -ENOENT;
2559
2560         btrfs_inc_nlink(inode);
2561         err = btrfs_check_free_space(root, 1, 0);
2562         if (err)
2563                 goto fail;
2564         err = btrfs_set_inode_index(dir, inode, &index);
2565         if (err)
2566                 goto fail;
2567
2568         trans = btrfs_start_transaction(root, 1);
2569
2570         btrfs_set_trans_block_group(trans, dir);
2571         atomic_inc(&inode->i_count);
2572
2573         err = btrfs_add_nondir(trans, dentry, inode, 1, index);
2574
2575         if (err)
2576                 drop_inode = 1;
2577
2578         dir->i_sb->s_dirt = 1;
2579         btrfs_update_inode_block_group(trans, dir);
2580         err = btrfs_update_inode(trans, root, inode);
2581
2582         if (err)
2583                 drop_inode = 1;
2584
2585         nr = trans->blocks_used;
2586         btrfs_end_transaction_throttle(trans, root);
2587 fail:
2588         if (drop_inode) {
2589                 inode_dec_link_count(inode);
2590                 iput(inode);
2591         }
2592         btrfs_btree_balance_dirty(root, nr);
2593         return err;
2594 }
2595
2596 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2597 {
2598         struct inode *inode = NULL;
2599         struct btrfs_trans_handle *trans;
2600         struct btrfs_root *root = BTRFS_I(dir)->root;
2601         int err = 0;
2602         int drop_on_err = 0;
2603         u64 objectid = 0;
2604         u64 index = 0;
2605         unsigned long nr = 1;
2606
2607         err = btrfs_check_free_space(root, 1, 0);
2608         if (err)
2609                 goto out_unlock;
2610
2611         trans = btrfs_start_transaction(root, 1);
2612         btrfs_set_trans_block_group(trans, dir);
2613
2614         if (IS_ERR(trans)) {
2615                 err = PTR_ERR(trans);
2616                 goto out_unlock;
2617         }
2618
2619         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2620         if (err) {
2621                 err = -ENOSPC;
2622                 goto out_unlock;
2623         }
2624
2625         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
2626                                 dentry->d_name.len,
2627                                 dentry->d_parent->d_inode->i_ino, objectid,
2628                                 BTRFS_I(dir)->block_group, S_IFDIR | mode,
2629                                 &index);
2630         if (IS_ERR(inode)) {
2631                 err = PTR_ERR(inode);
2632                 goto out_fail;
2633         }
2634
2635         drop_on_err = 1;
2636
2637         err = btrfs_init_acl(inode, dir);
2638         if (err)
2639                 goto out_fail;
2640
2641         inode->i_op = &btrfs_dir_inode_operations;
2642         inode->i_fop = &btrfs_dir_file_operations;
2643         btrfs_set_trans_block_group(trans, inode);
2644
2645         btrfs_i_size_write(inode, 0);
2646         err = btrfs_update_inode(trans, root, inode);
2647         if (err)
2648                 goto out_fail;
2649
2650         err = btrfs_add_link(trans, dentry->d_parent->d_inode,
2651                                  inode, dentry->d_name.name,
2652                                  dentry->d_name.len, 0, index);
2653         if (err)
2654                 goto out_fail;
2655
2656         d_instantiate(dentry, inode);
2657         drop_on_err = 0;
2658         dir->i_sb->s_dirt = 1;
2659         btrfs_update_inode_block_group(trans, inode);
2660         btrfs_update_inode_block_group(trans, dir);
2661
2662 out_fail:
2663         nr = trans->blocks_used;
2664         btrfs_end_transaction_throttle(trans, root);
2665
2666 out_unlock:
2667         if (drop_on_err)
2668                 iput(inode);
2669         btrfs_btree_balance_dirty(root, nr);
2670         return err;
2671 }
2672
2673 static int merge_extent_mapping(struct extent_map_tree *em_tree,
2674                                 struct extent_map *existing,
2675                                 struct extent_map *em,
2676                                 u64 map_start, u64 map_len)
2677 {
2678         u64 start_diff;
2679
2680         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
2681         start_diff = map_start - em->start;
2682         em->start = map_start;
2683         em->len = map_len;
2684         if (em->block_start < EXTENT_MAP_LAST_BYTE)
2685                 em->block_start += start_diff;
2686         return add_extent_mapping(em_tree, em);
2687 }
2688
2689 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
2690                                     size_t pg_offset, u64 start, u64 len,
2691                                     int create)
2692 {
2693         int ret;
2694         int err = 0;
2695         u64 bytenr;
2696         u64 extent_start = 0;
2697         u64 extent_end = 0;
2698         u64 objectid = inode->i_ino;
2699         u32 found_type;
2700         struct btrfs_path *path = NULL;
2701         struct btrfs_root *root = BTRFS_I(inode)->root;
2702         struct btrfs_file_extent_item *item;
2703         struct extent_buffer *leaf;
2704         struct btrfs_key found_key;
2705         struct extent_map *em = NULL;
2706         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2707         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2708         struct btrfs_trans_handle *trans = NULL;
2709
2710 again:
2711         spin_lock(&em_tree->lock);
2712         em = lookup_extent_mapping(em_tree, start, len);
2713         if (em)
2714                 em->bdev = root->fs_info->fs_devices->latest_bdev;
2715         spin_unlock(&em_tree->lock);
2716
2717         if (em) {
2718                 if (em->start > start || em->start + em->len <= start)
2719                         free_extent_map(em);
2720                 else if (em->block_start == EXTENT_MAP_INLINE && page)
2721                         free_extent_map(em);
2722                 else
2723                         goto out;
2724         }
2725         em = alloc_extent_map(GFP_NOFS);
2726         if (!em) {
2727                 err = -ENOMEM;
2728                 goto out;
2729         }
2730         em->bdev = root->fs_info->fs_devices->latest_bdev;
2731         em->start = EXTENT_MAP_HOLE;
2732         em->len = (u64)-1;
2733
2734         if (!path) {
2735                 path = btrfs_alloc_path();
2736                 BUG_ON(!path);
2737         }
2738
2739         ret = btrfs_lookup_file_extent(trans, root, path,
2740                                        objectid, start, trans != NULL);
2741         if (ret < 0) {
2742                 err = ret;
2743                 goto out;
2744         }
2745
2746         if (ret != 0) {
2747                 if (path->slots[0] == 0)
2748                         goto not_found;
2749                 path->slots[0]--;
2750         }
2751
2752         leaf = path->nodes[0];
2753         item = btrfs_item_ptr(leaf, path->slots[0],
2754                               struct btrfs_file_extent_item);
2755         /* are we inside the extent that was found? */
2756         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2757         found_type = btrfs_key_type(&found_key);
2758         if (found_key.objectid != objectid ||
2759             found_type != BTRFS_EXTENT_DATA_KEY) {
2760                 goto not_found;
2761         }
2762
2763         found_type = btrfs_file_extent_type(leaf, item);
2764         extent_start = found_key.offset;
2765         if (found_type == BTRFS_FILE_EXTENT_REG) {
2766                 extent_end = extent_start +
2767                        btrfs_file_extent_num_bytes(leaf, item);
2768                 err = 0;
2769                 if (start < extent_start || start >= extent_end) {
2770                         em->start = start;
2771                         if (start < extent_start) {
2772                                 if (start + len <= extent_start)
2773                                         goto not_found;
2774                                 em->len = extent_end - extent_start;
2775                         } else {
2776                                 em->len = len;
2777                         }
2778                         goto not_found_em;
2779                 }
2780                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
2781                 if (bytenr == 0) {
2782                         em->start = extent_start;
2783                         em->len = extent_end - extent_start;
2784                         em->block_start = EXTENT_MAP_HOLE;
2785                         goto insert;
2786                 }
2787                 bytenr += btrfs_file_extent_offset(leaf, item);
2788                 em->block_start = bytenr;
2789                 em->start = extent_start;
2790                 em->len = extent_end - extent_start;
2791                 goto insert;
2792         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
2793                 u64 page_start;
2794                 unsigned long ptr;
2795                 char *map;
2796                 size_t size;
2797                 size_t extent_offset;
2798                 size_t copy_size;
2799
2800                 size = btrfs_file_extent_inline_len(leaf, btrfs_item_nr(leaf,
2801                                                     path->slots[0]));
2802                 extent_end = (extent_start + size + root->sectorsize - 1) &
2803                         ~((u64)root->sectorsize - 1);
2804                 if (start < extent_start || start >= extent_end) {
2805                         em->start = start;
2806                         if (start < extent_start) {
2807                                 if (start + len <= extent_start)
2808                                         goto not_found;
2809                                 em->len = extent_end - extent_start;
2810                         } else {
2811                                 em->len = len;
2812                         }
2813                         goto not_found_em;
2814                 }
2815                 em->block_start = EXTENT_MAP_INLINE;
2816
2817                 if (!page) {
2818                         em->start = extent_start;
2819                         em->len = size;
2820                         goto out;
2821                 }
2822
2823                 page_start = page_offset(page) + pg_offset;
2824                 extent_offset = page_start - extent_start;
2825                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
2826                                 size - extent_offset);
2827                 em->start = extent_start + extent_offset;
2828                 em->len = (copy_size + root->sectorsize - 1) &
2829                         ~((u64)root->sectorsize - 1);
2830                 map = kmap(page);
2831                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
2832                 if (create == 0 && !PageUptodate(page)) {
2833                         read_extent_buffer(leaf, map + pg_offset, ptr,
2834                                            copy_size);
2835                         flush_dcache_page(page);
2836                 } else if (create && PageUptodate(page)) {
2837                         if (!trans) {
2838                                 kunmap(page);
2839                                 free_extent_map(em);
2840                                 em = NULL;
2841                                 btrfs_release_path(root, path);
2842                                 trans = btrfs_join_transaction(root, 1);
2843                                 goto again;
2844                         }
2845                         write_extent_buffer(leaf, map + pg_offset, ptr,
2846                                             copy_size);
2847                         btrfs_mark_buffer_dirty(leaf);
2848                 }
2849                 kunmap(page);
2850                 set_extent_uptodate(io_tree, em->start,
2851                                     extent_map_end(em) - 1, GFP_NOFS);
2852                 goto insert;
2853         } else {
2854                 printk("unkknown found_type %d\n", found_type);
2855                 WARN_ON(1);
2856         }
2857 not_found:
2858         em->start = start;
2859         em->len = len;
2860 not_found_em:
2861         em->block_start = EXTENT_MAP_HOLE;
2862 insert:
2863         btrfs_release_path(root, path);
2864         if (em->start > start || extent_map_end(em) <= start) {
2865                 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->len, start, len);
2866                 err = -EIO;
2867                 goto out;
2868         }
2869
2870         err = 0;
2871         spin_lock(&em_tree->lock);
2872         ret = add_extent_mapping(em_tree, em);
2873         /* it is possible that someone inserted the extent into the tree
2874          * while we had the lock dropped.  It is also possible that
2875          * an overlapping map exists in the tree
2876          */
2877         if (ret == -EEXIST) {
2878                 struct extent_map *existing;
2879
2880                 ret = 0;
2881
2882                 existing = lookup_extent_mapping(em_tree, start, len);
2883                 if (existing && (existing->start > start ||
2884                     existing->start + existing->len <= start)) {
2885                         free_extent_map(existing);
2886                         existing = NULL;
2887                 }
2888                 if (!existing) {
2889                         existing = lookup_extent_mapping(em_tree, em->start,
2890                                                          em->len);
2891                         if (existing) {
2892                                 err = merge_extent_mapping(em_tree, existing,
2893                                                            em, start,
2894                                                            root->sectorsize);
2895                                 free_extent_map(existing);
2896                                 if (err) {
2897                                         free_extent_map(em);
2898                                         em = NULL;
2899                                 }
2900                         } else {
2901                                 err = -EIO;
2902                                 printk("failing to insert %Lu %Lu\n",
2903                                        start, len);
2904                                 free_extent_map(em);
2905                                 em = NULL;
2906                         }
2907                 } else {
2908                         free_extent_map(em);
2909                         em = existing;
2910                         err = 0;
2911                 }
2912         }
2913         spin_unlock(&em_tree->lock);
2914 out:
2915         if (path)
2916                 btrfs_free_path(path);
2917         if (trans) {
2918                 ret = btrfs_end_transaction(trans, root);
2919                 if (!err) {
2920                         err = ret;
2921                 }
2922         }
2923         if (err) {
2924                 free_extent_map(em);
2925                 WARN_ON(1);
2926                 return ERR_PTR(err);
2927         }
2928         return em;
2929 }
2930
2931 #if 0 /* waiting for O_DIRECT reads */
2932 static int btrfs_get_block(struct inode *inode, sector_t iblock,
2933                         struct buffer_head *bh_result, int create)
2934 {
2935         struct extent_map *em;
2936         u64 start = (u64)iblock << inode->i_blkbits;
2937         struct btrfs_multi_bio *multi = NULL;
2938         struct btrfs_root *root = BTRFS_I(inode)->root;
2939         u64 len;
2940         u64 logical;
2941         u64 map_length;
2942         int ret = 0;
2943
2944         em = btrfs_get_extent(inode, NULL, 0, start, bh_result->b_size, 0);
2945
2946         if (!em || IS_ERR(em))
2947                 goto out;
2948
2949         if (em->start > start || em->start + em->len <= start) {
2950             goto out;
2951         }
2952
2953         if (em->block_start == EXTENT_MAP_INLINE) {
2954                 ret = -EINVAL;
2955                 goto out;
2956         }
2957
2958         len = em->start + em->len - start;
2959         len = min_t(u64, len, INT_LIMIT(typeof(bh_result->b_size)));
2960
2961         if (em->block_start == EXTENT_MAP_HOLE ||
2962             em->block_start == EXTENT_MAP_DELALLOC) {
2963                 bh_result->b_size = len;
2964                 goto out;
2965         }
2966
2967         logical = start - em->start;
2968         logical = em->block_start + logical;
2969
2970         map_length = len;
2971         ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
2972                               logical, &map_length, &multi, 0);
2973         BUG_ON(ret);
2974         bh_result->b_blocknr = multi->stripes[0].physical >> inode->i_blkbits;
2975         bh_result->b_size = min(map_length, len);
2976
2977         bh_result->b_bdev = multi->stripes[0].dev->bdev;
2978         set_buffer_mapped(bh_result);
2979         kfree(multi);
2980 out:
2981         free_extent_map(em);
2982         return ret;
2983 }
2984 #endif
2985
2986 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
2987                         const struct iovec *iov, loff_t offset,
2988                         unsigned long nr_segs)
2989 {
2990         return -EINVAL;
2991 #if 0
2992         struct file *file = iocb->ki_filp;
2993         struct inode *inode = file->f_mapping->host;
2994
2995         if (rw == WRITE)
2996                 return -EINVAL;
2997
2998         return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
2999                                   offset, nr_segs, btrfs_get_block, NULL);
3000 #endif
3001 }
3002
3003 static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
3004 {
3005         return extent_bmap(mapping, iblock, btrfs_get_extent);
3006 }
3007
3008 int btrfs_readpage(struct file *file, struct page *page)
3009 {
3010         struct extent_io_tree *tree;
3011         tree = &BTRFS_I(page->mapping->host)->io_tree;
3012         return extent_read_full_page(tree, page, btrfs_get_extent);
3013 }
3014
3015 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
3016 {
3017         struct extent_io_tree *tree;
3018
3019
3020         if (current->flags & PF_MEMALLOC) {
3021                 redirty_page_for_writepage(wbc, page);
3022                 unlock_page(page);
3023                 return 0;
3024         }
3025         tree = &BTRFS_I(page->mapping->host)->io_tree;
3026         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
3027 }
3028
3029 int btrfs_writepages(struct address_space *mapping,
3030                      struct writeback_control *wbc)
3031 {
3032         struct extent_io_tree *tree;
3033         tree = &BTRFS_I(mapping->host)->io_tree;
3034         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
3035 }
3036
3037 static int
3038 btrfs_readpages(struct file *file, struct address_space *mapping,
3039                 struct list_head *pages, unsigned nr_pages)
3040 {
3041         struct extent_io_tree *tree;
3042         tree = &BTRFS_I(mapping->host)->io_tree;
3043         return extent_readpages(tree, mapping, pages, nr_pages,
3044                                 btrfs_get_extent);
3045 }
3046 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
3047 {
3048         struct extent_io_tree *tree;
3049         struct extent_map_tree *map;
3050         int ret;
3051
3052         tree = &BTRFS_I(page->mapping->host)->io_tree;
3053         map = &BTRFS_I(page->mapping->host)->extent_tree;
3054         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
3055         if (ret == 1) {
3056                 ClearPagePrivate(page);
3057                 set_page_private(page, 0);
3058                 page_cache_release(page);
3059         }
3060         return ret;
3061 }
3062
3063 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
3064 {
3065         if (PageWriteback(page) || PageDirty(page))
3066                 return 0;
3067         return __btrfs_releasepage(page, gfp_flags);
3068 }
3069
3070 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
3071 {
3072         struct extent_io_tree *tree;
3073         struct btrfs_ordered_extent *ordered;
3074         u64 page_start = page_offset(page);
3075         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
3076
3077         wait_on_page_writeback(page);
3078         tree = &BTRFS_I(page->mapping->host)->io_tree;
3079         if (offset) {
3080                 btrfs_releasepage(page, GFP_NOFS);
3081                 return;
3082         }
3083
3084         lock_extent(tree, page_start, page_end, GFP_NOFS);
3085         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
3086                                            page_offset(page));
3087         if (ordered) {
3088                 /*
3089                  * IO on this page will never be started, so we need
3090                  * to account for any ordered extents now
3091                  */
3092                 clear_extent_bit(tree, page_start, page_end,
3093                                  EXTENT_DIRTY | EXTENT_DELALLOC |
3094                                  EXTENT_LOCKED, 1, 0, GFP_NOFS);
3095                 btrfs_finish_ordered_io(page->mapping->host,
3096                                         page_start, page_end);
3097                 btrfs_put_ordered_extent(ordered);
3098                 lock_extent(tree, page_start, page_end, GFP_NOFS);
3099         }
3100         clear_extent_bit(tree, page_start, page_end,
3101                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
3102                  EXTENT_ORDERED,
3103                  1, 1, GFP_NOFS);
3104         __btrfs_releasepage(page, GFP_NOFS);
3105
3106         ClearPageChecked(page);
3107         if (PagePrivate(page)) {
3108                 ClearPagePrivate(page);
3109                 set_page_private(page, 0);
3110                 page_cache_release(page);
3111         }
3112 }
3113
3114 /*
3115  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
3116  * called from a page fault handler when a page is first dirtied. Hence we must
3117  * be careful to check for EOF conditions here. We set the page up correctly
3118  * for a written page which means we get ENOSPC checking when writing into
3119  * holes and correct delalloc and unwritten extent mapping on filesystems that
3120  * support these features.
3121  *
3122  * We are not allowed to take the i_mutex here so we have to play games to
3123  * protect against truncate races as the page could now be beyond EOF.  Because
3124  * vmtruncate() writes the inode size before removing pages, once we have the
3125  * page lock we can determine safely if the page is beyond EOF. If it is not
3126  * beyond EOF, then the page is guaranteed safe against truncation until we
3127  * unlock the page.
3128  */
3129 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
3130 {
3131         struct inode *inode = fdentry(vma->vm_file)->d_inode;
3132         struct btrfs_root *root = BTRFS_I(inode)->root;
3133         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3134         struct btrfs_ordered_extent *ordered;
3135         char *kaddr;
3136         unsigned long zero_start;
3137         loff_t size;
3138         int ret;
3139         u64 page_start;
3140         u64 page_end;
3141
3142         ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
3143         if (ret)
3144                 goto out;
3145
3146         ret = -EINVAL;
3147 again:
3148         lock_page(page);
3149         size = i_size_read(inode);
3150         page_start = page_offset(page);
3151         page_end = page_start + PAGE_CACHE_SIZE - 1;
3152
3153         if ((page->mapping != inode->i_mapping) ||
3154             (page_start >= size)) {
3155                 /* page got truncated out from underneath us */
3156                 goto out_unlock;
3157         }
3158         wait_on_page_writeback(page);
3159
3160         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
3161         set_page_extent_mapped(page);
3162
3163         /*
3164          * we can't set the delalloc bits if there are pending ordered
3165          * extents.  Drop our locks and wait for them to finish
3166          */
3167         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3168         if (ordered) {
3169                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3170                 unlock_page(page);
3171                 btrfs_start_ordered_extent(inode, ordered, 1);
3172                 btrfs_put_ordered_extent(ordered);
3173                 goto again;
3174         }
3175
3176         btrfs_set_extent_delalloc(inode, page_start, page_end);
3177         ret = 0;
3178
3179         /* page is wholly or partially inside EOF */
3180         if (page_start + PAGE_CACHE_SIZE > size)
3181                 zero_start = size & ~PAGE_CACHE_MASK;
3182         else
3183                 zero_start = PAGE_CACHE_SIZE;
3184
3185         if (zero_start != PAGE_CACHE_SIZE) {
3186                 kaddr = kmap(page);
3187                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
3188                 flush_dcache_page(page);
3189                 kunmap(page);
3190         }
3191         ClearPageChecked(page);
3192         set_page_dirty(page);
3193         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3194
3195 out_unlock:
3196         unlock_page(page);
3197 out:
3198         return ret;
3199 }
3200
3201 static void btrfs_truncate(struct inode *inode)
3202 {
3203         struct btrfs_root *root = BTRFS_I(inode)->root;
3204         int ret;
3205         struct btrfs_trans_handle *trans;
3206         unsigned long nr;
3207         u64 mask = root->sectorsize - 1;
3208
3209         if (!S_ISREG(inode->i_mode))
3210                 return;
3211         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3212                 return;
3213
3214         btrfs_truncate_page(inode->i_mapping, inode->i_size);
3215         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
3216
3217         trans = btrfs_start_transaction(root, 1);
3218         btrfs_set_trans_block_group(trans, inode);
3219         btrfs_i_size_write(inode, inode->i_size);
3220
3221         ret = btrfs_orphan_add(trans, inode);
3222         if (ret)
3223                 goto out;
3224         /* FIXME, add redo link to tree so we don't leak on crash */
3225         ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size,
3226                                       BTRFS_EXTENT_DATA_KEY);
3227         btrfs_update_inode(trans, root, inode);
3228
3229         ret = btrfs_orphan_del(trans, inode);
3230         BUG_ON(ret);
3231
3232 out:
3233         nr = trans->blocks_used;
3234         ret = btrfs_end_transaction_throttle(trans, root);
3235         BUG_ON(ret);
3236         btrfs_btree_balance_dirty(root, nr);
3237 }
3238
3239 /*
3240  * Invalidate a single dcache entry at the root of the filesystem.
3241  * Needed after creation of snapshot or subvolume.
3242  */
3243 void btrfs_invalidate_dcache_root(struct btrfs_root *root, char *name,
3244                                   int namelen)
3245 {
3246         struct dentry *alias, *entry;
3247         struct qstr qstr;
3248
3249         alias = d_find_alias(root->fs_info->sb->s_root->d_inode);
3250         if (alias) {
3251                 qstr.name = name;
3252                 qstr.len = namelen;
3253                 /* change me if btrfs ever gets a d_hash operation */
3254                 qstr.hash = full_name_hash(qstr.name, qstr.len);
3255                 entry = d_lookup(alias, &qstr);
3256                 dput(alias);
3257                 if (entry) {
3258                         d_invalidate(entry);
3259                         dput(entry);
3260                 }
3261         }
3262 }
3263
3264 int btrfs_create_subvol_root(struct btrfs_root *new_root,
3265                 struct btrfs_trans_handle *trans, u64 new_dirid,
3266                 struct btrfs_block_group_cache *block_group)
3267 {
3268         struct inode *inode;
3269         u64 index = 0;
3270
3271         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
3272                                 new_dirid, block_group, S_IFDIR | 0700, &index);
3273         if (IS_ERR(inode))
3274                 return PTR_ERR(inode);
3275         inode->i_op = &btrfs_dir_inode_operations;
3276         inode->i_fop = &btrfs_dir_file_operations;
3277         new_root->inode = inode;
3278
3279         inode->i_nlink = 1;
3280         btrfs_i_size_write(inode, 0);
3281
3282         return btrfs_update_inode(trans, new_root, inode);
3283 }
3284
3285 unsigned long btrfs_force_ra(struct address_space *mapping,
3286                               struct file_ra_state *ra, struct file *file,
3287                               pgoff_t offset, pgoff_t last_index)
3288 {
3289         pgoff_t req_size = last_index - offset + 1;
3290
3291 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
3292         offset = page_cache_readahead(mapping, ra, file, offset, req_size);
3293         return offset;
3294 #else
3295         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
3296         return offset + req_size;
3297 #endif
3298 }
3299
3300 struct inode *btrfs_alloc_inode(struct super_block *sb)
3301 {
3302         struct btrfs_inode *ei;
3303
3304         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
3305         if (!ei)
3306                 return NULL;
3307         ei->last_trans = 0;
3308         ei->logged_trans = 0;
3309         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
3310         ei->i_acl = BTRFS_ACL_NOT_CACHED;
3311         ei->i_default_acl = BTRFS_ACL_NOT_CACHED;
3312         INIT_LIST_HEAD(&ei->i_orphan);
3313         return &ei->vfs_inode;
3314 }
3315
3316 void btrfs_destroy_inode(struct inode *inode)
3317 {
3318         struct btrfs_ordered_extent *ordered;
3319         WARN_ON(!list_empty(&inode->i_dentry));
3320         WARN_ON(inode->i_data.nrpages);
3321
3322         if (BTRFS_I(inode)->i_acl &&
3323             BTRFS_I(inode)->i_acl != BTRFS_ACL_NOT_CACHED)
3324                 posix_acl_release(BTRFS_I(inode)->i_acl);
3325         if (BTRFS_I(inode)->i_default_acl &&
3326             BTRFS_I(inode)->i_default_acl != BTRFS_ACL_NOT_CACHED)
3327                 posix_acl_release(BTRFS_I(inode)->i_default_acl);
3328
3329         spin_lock(&BTRFS_I(inode)->root->list_lock);
3330         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
3331                 printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
3332                        " list\n", inode->i_ino);
3333                 dump_stack();
3334         }
3335         spin_unlock(&BTRFS_I(inode)->root->list_lock);
3336
3337         while(1) {
3338                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
3339                 if (!ordered)
3340                         break;
3341                 else {
3342                         printk("found ordered extent %Lu %Lu\n",
3343                                ordered->file_offset, ordered->len);
3344                         btrfs_remove_ordered_extent(inode, ordered);
3345                         btrfs_put_ordered_extent(ordered);
3346                         btrfs_put_ordered_extent(ordered);
3347                 }
3348         }
3349         btrfs_drop_extent_cache(inode, 0, (u64)-1);
3350         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
3351 }
3352
3353 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,26)
3354 static void init_once(void *foo)
3355 #elif LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
3356 static void init_once(struct kmem_cache * cachep, void *foo)
3357 #else
3358 static void init_once(void * foo, struct kmem_cache * cachep,
3359                       unsigned long flags)
3360 #endif
3361 {
3362         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
3363
3364         inode_init_once(&ei->vfs_inode);
3365 }
3366
3367 void btrfs_destroy_cachep(void)
3368 {
3369         if (btrfs_inode_cachep)
3370                 kmem_cache_destroy(btrfs_inode_cachep);
3371         if (btrfs_trans_handle_cachep)
3372                 kmem_cache_destroy(btrfs_trans_handle_cachep);
3373         if (btrfs_transaction_cachep)
3374                 kmem_cache_destroy(btrfs_transaction_cachep);
3375         if (btrfs_bit_radix_cachep)
3376                 kmem_cache_destroy(btrfs_bit_radix_cachep);
3377         if (btrfs_path_cachep)
3378                 kmem_cache_destroy(btrfs_path_cachep);
3379 }
3380
3381 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
3382                                        unsigned long extra_flags,
3383 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,26)
3384                                        void (*ctor)(void *)
3385 #elif LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
3386                                        void (*ctor)(struct kmem_cache *, void *)
3387 #else
3388                                        void (*ctor)(void *, struct kmem_cache *,
3389                                                     unsigned long)
3390 #endif
3391                                      )
3392 {
3393         return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
3394                                  SLAB_MEM_SPREAD | extra_flags), ctor
3395 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
3396                                  ,NULL
3397 #endif
3398                                 );
3399 }
3400
3401 int btrfs_init_cachep(void)
3402 {
3403         btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
3404                                           sizeof(struct btrfs_inode),
3405                                           0, init_once);
3406         if (!btrfs_inode_cachep)
3407                 goto fail;
3408         btrfs_trans_handle_cachep =
3409                         btrfs_cache_create("btrfs_trans_handle_cache",
3410                                            sizeof(struct btrfs_trans_handle),
3411                                            0, NULL);
3412         if (!btrfs_trans_handle_cachep)
3413                 goto fail;
3414         btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
3415                                              sizeof(struct btrfs_transaction),
3416                                              0, NULL);
3417         if (!btrfs_transaction_cachep)
3418                 goto fail;
3419         btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
3420                                          sizeof(struct btrfs_path),
3421                                          0, NULL);
3422         if (!btrfs_path_cachep)
3423                 goto fail;
3424         btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
3425                                               SLAB_DESTROY_BY_RCU, NULL);
3426         if (!btrfs_bit_radix_cachep)
3427                 goto fail;
3428         return 0;
3429 fail:
3430         btrfs_destroy_cachep();
3431         return -ENOMEM;
3432 }
3433
3434 static int btrfs_getattr(struct vfsmount *mnt,
3435                          struct dentry *dentry, struct kstat *stat)
3436 {
3437         struct inode *inode = dentry->d_inode;
3438         generic_fillattr(inode, stat);
3439         stat->blksize = PAGE_CACHE_SIZE;
3440         stat->blocks = inode->i_blocks + (BTRFS_I(inode)->delalloc_bytes >> 9);
3441         return 0;
3442 }
3443
3444 static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
3445                            struct inode * new_dir,struct dentry *new_dentry)
3446 {
3447         struct btrfs_trans_handle *trans;
3448         struct btrfs_root *root = BTRFS_I(old_dir)->root;
3449         struct inode *new_inode = new_dentry->d_inode;
3450         struct inode *old_inode = old_dentry->d_inode;
3451         struct timespec ctime = CURRENT_TIME;
3452         u64 index = 0;
3453         int ret;
3454
3455         if (S_ISDIR(old_inode->i_mode) && new_inode &&
3456             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
3457                 return -ENOTEMPTY;
3458         }
3459
3460         ret = btrfs_check_free_space(root, 1, 0);
3461         if (ret)
3462                 goto out_unlock;
3463
3464         trans = btrfs_start_transaction(root, 1);
3465
3466         btrfs_set_trans_block_group(trans, new_dir);
3467
3468         btrfs_inc_nlink(old_dentry->d_inode);
3469         old_dir->i_ctime = old_dir->i_mtime = ctime;
3470         new_dir->i_ctime = new_dir->i_mtime = ctime;
3471         old_inode->i_ctime = ctime;
3472
3473         ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode,
3474                                  old_dentry->d_name.name,
3475                                  old_dentry->d_name.len);
3476         if (ret)
3477                 goto out_fail;
3478
3479         if (new_inode) {
3480                 new_inode->i_ctime = CURRENT_TIME;
3481                 ret = btrfs_unlink_inode(trans, root, new_dir,
3482                                          new_dentry->d_inode,
3483                                          new_dentry->d_name.name,
3484                                          new_dentry->d_name.len);
3485                 if (ret)
3486                         goto out_fail;
3487                 if (new_inode->i_nlink == 0) {
3488                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
3489                         if (ret)
3490                                 goto out_fail;
3491                 }
3492
3493         }
3494         ret = btrfs_set_inode_index(new_dir, old_inode, &index);
3495         if (ret)
3496                 goto out_fail;
3497
3498         ret = btrfs_add_link(trans, new_dentry->d_parent->d_inode,
3499                              old_inode, new_dentry->d_name.name,
3500                              new_dentry->d_name.len, 1, index);
3501         if (ret)
3502                 goto out_fail;
3503
3504 out_fail:
3505         btrfs_end_transaction_throttle(trans, root);
3506 out_unlock:
3507         return ret;
3508 }
3509
3510 int btrfs_start_delalloc_inodes(struct btrfs_root *root)
3511 {
3512         struct list_head *head = &root->fs_info->delalloc_inodes;
3513         struct btrfs_inode *binode;
3514         unsigned long flags;
3515
3516         spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
3517         while(!list_empty(head)) {
3518                 binode = list_entry(head->next, struct btrfs_inode,
3519                                     delalloc_inodes);
3520                 atomic_inc(&binode->vfs_inode.i_count);
3521                 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
3522                 filemap_write_and_wait(binode->vfs_inode.i_mapping);
3523                 iput(&binode->vfs_inode);
3524                 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
3525         }
3526         spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
3527         return 0;
3528 }
3529
3530 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
3531                          const char *symname)
3532 {
3533         struct btrfs_trans_handle *trans;
3534         struct btrfs_root *root = BTRFS_I(dir)->root;
3535         struct btrfs_path *path;
3536         struct btrfs_key key;
3537         struct inode *inode = NULL;
3538         int err;
3539         int drop_inode = 0;
3540         u64 objectid;
3541         u64 index = 0 ;
3542         int name_len;
3543         int datasize;
3544         unsigned long ptr;
3545         struct btrfs_file_extent_item *ei;
3546         struct extent_buffer *leaf;
3547         unsigned long nr = 0;
3548
3549         name_len = strlen(symname) + 1;
3550         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
3551                 return -ENAMETOOLONG;
3552
3553         err = btrfs_check_free_space(root, 1, 0);
3554         if (err)
3555                 goto out_fail;
3556
3557         trans = btrfs_start_transaction(root, 1);
3558         btrfs_set_trans_block_group(trans, dir);
3559
3560         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3561         if (err) {
3562                 err = -ENOSPC;
3563                 goto out_unlock;
3564         }
3565
3566         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3567                                 dentry->d_name.len,
3568                                 dentry->d_parent->d_inode->i_ino, objectid,
3569                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
3570                                 &index);
3571         err = PTR_ERR(inode);
3572         if (IS_ERR(inode))
3573                 goto out_unlock;
3574
3575         err = btrfs_init_acl(inode, dir);
3576         if (err) {
3577                 drop_inode = 1;
3578                 goto out_unlock;
3579         }
3580
3581         btrfs_set_trans_block_group(trans, inode);
3582         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3583         if (err)
3584                 drop_inode = 1;
3585         else {
3586                 inode->i_mapping->a_ops = &btrfs_aops;
3587                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3588                 inode->i_fop = &btrfs_file_operations;
3589                 inode->i_op = &btrfs_file_inode_operations;
3590                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3591         }
3592         dir->i_sb->s_dirt = 1;
3593         btrfs_update_inode_block_group(trans, inode);
3594         btrfs_update_inode_block_group(trans, dir);
3595         if (drop_inode)
3596                 goto out_unlock;
3597
3598         path = btrfs_alloc_path();
3599         BUG_ON(!path);
3600         key.objectid = inode->i_ino;
3601         key.offset = 0;
3602         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
3603         datasize = btrfs_file_extent_calc_inline_size(name_len);
3604         err = btrfs_insert_empty_item(trans, root, path, &key,
3605                                       datasize);
3606         if (err) {
3607                 drop_inode = 1;
3608                 goto out_unlock;
3609         }
3610         leaf = path->nodes[0];
3611         ei = btrfs_item_ptr(leaf, path->slots[0],
3612                             struct btrfs_file_extent_item);
3613         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
3614         btrfs_set_file_extent_type(leaf, ei,
3615                                    BTRFS_FILE_EXTENT_INLINE);
3616         ptr = btrfs_file_extent_inline_start(ei);
3617         write_extent_buffer(leaf, symname, ptr, name_len);
3618         btrfs_mark_buffer_dirty(leaf);
3619         btrfs_free_path(path);
3620
3621         inode->i_op = &btrfs_symlink_inode_operations;
3622         inode->i_mapping->a_ops = &btrfs_symlink_aops;
3623         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3624         btrfs_i_size_write(inode, name_len - 1);
3625         err = btrfs_update_inode(trans, root, inode);
3626         if (err)
3627                 drop_inode = 1;
3628
3629 out_unlock:
3630         nr = trans->blocks_used;
3631         btrfs_end_transaction_throttle(trans, root);
3632 out_fail:
3633         if (drop_inode) {
3634                 inode_dec_link_count(inode);
3635                 iput(inode);
3636         }
3637         btrfs_btree_balance_dirty(root, nr);
3638         return err;
3639 }
3640
3641 static int btrfs_set_page_dirty(struct page *page)
3642 {
3643         return __set_page_dirty_nobuffers(page);
3644 }
3645
3646 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,26)
3647 static int btrfs_permission(struct inode *inode, int mask)
3648 #else
3649 static int btrfs_permission(struct inode *inode, int mask,
3650                             struct nameidata *nd)
3651 #endif
3652 {
3653         if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
3654                 return -EACCES;
3655         return generic_permission(inode, mask, btrfs_check_acl);
3656 }
3657
3658 static struct inode_operations btrfs_dir_inode_operations = {
3659         .lookup         = btrfs_lookup,
3660         .create         = btrfs_create,
3661         .unlink         = btrfs_unlink,
3662         .link           = btrfs_link,
3663         .mkdir          = btrfs_mkdir,
3664         .rmdir          = btrfs_rmdir,
3665         .rename         = btrfs_rename,
3666         .symlink        = btrfs_symlink,
3667         .setattr        = btrfs_setattr,
3668         .mknod          = btrfs_mknod,
3669         .setxattr       = btrfs_setxattr,
3670         .getxattr       = btrfs_getxattr,
3671         .listxattr      = btrfs_listxattr,
3672         .removexattr    = btrfs_removexattr,
3673         .permission     = btrfs_permission,
3674 };
3675 static struct inode_operations btrfs_dir_ro_inode_operations = {
3676         .lookup         = btrfs_lookup,
3677         .permission     = btrfs_permission,
3678 };
3679 static struct file_operations btrfs_dir_file_operations = {
3680         .llseek         = generic_file_llseek,
3681         .read           = generic_read_dir,
3682 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,28)
3683         .readdir        = btrfs_nfshack_readdir,
3684 #else /* NFSd readdir/lookup deadlock is fixed */
3685         .readdir        = btrfs_real_readdir,
3686 #endif
3687         .unlocked_ioctl = btrfs_ioctl,
3688 #ifdef CONFIG_COMPAT
3689         .compat_ioctl   = btrfs_ioctl,
3690 #endif
3691         .release        = btrfs_release_file,
3692         .fsync          = btrfs_sync_file,
3693 };
3694
3695 static struct extent_io_ops btrfs_extent_io_ops = {
3696         .fill_delalloc = run_delalloc_range,
3697         .submit_bio_hook = btrfs_submit_bio_hook,
3698         .merge_bio_hook = btrfs_merge_bio_hook,
3699         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
3700         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
3701         .writepage_start_hook = btrfs_writepage_start_hook,
3702         .readpage_io_failed_hook = btrfs_io_failed_hook,
3703         .set_bit_hook = btrfs_set_bit_hook,
3704         .clear_bit_hook = btrfs_clear_bit_hook,
3705 };
3706
3707 static struct address_space_operations btrfs_aops = {
3708         .readpage       = btrfs_readpage,
3709         .writepage      = btrfs_writepage,
3710         .writepages     = btrfs_writepages,
3711         .readpages      = btrfs_readpages,
3712         .sync_page      = block_sync_page,
3713         .bmap           = btrfs_bmap,
3714         .direct_IO      = btrfs_direct_IO,
3715         .invalidatepage = btrfs_invalidatepage,
3716         .releasepage    = btrfs_releasepage,
3717         .set_page_dirty = btrfs_set_page_dirty,
3718 };
3719
3720 static struct address_space_operations btrfs_symlink_aops = {
3721         .readpage       = btrfs_readpage,
3722         .writepage      = btrfs_writepage,
3723         .invalidatepage = btrfs_invalidatepage,
3724         .releasepage    = btrfs_releasepage,
3725 };
3726
3727 static struct inode_operations btrfs_file_inode_operations = {
3728         .truncate       = btrfs_truncate,
3729         .getattr        = btrfs_getattr,
3730         .setattr        = btrfs_setattr,
3731         .setxattr       = btrfs_setxattr,
3732         .getxattr       = btrfs_getxattr,
3733         .listxattr      = btrfs_listxattr,
3734         .removexattr    = btrfs_removexattr,
3735         .permission     = btrfs_permission,
3736 };
3737 static struct inode_operations btrfs_special_inode_operations = {
3738         .getattr        = btrfs_getattr,
3739         .setattr        = btrfs_setattr,
3740         .permission     = btrfs_permission,
3741         .setxattr       = btrfs_setxattr,
3742         .getxattr       = btrfs_getxattr,
3743         .listxattr      = btrfs_listxattr,
3744         .removexattr    = btrfs_removexattr,
3745 };
3746 static struct inode_operations btrfs_symlink_inode_operations = {
3747         .readlink       = generic_readlink,
3748         .follow_link    = page_follow_link_light,
3749         .put_link       = page_put_link,
3750         .permission     = btrfs_permission,
3751 };