fs/btrfs/disk-io.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/fs.h>
  20 #include <linux/blkdev.h>
  21 #include <linux/crc32c.h>
  22 #include <linux/scatterlist.h>
  23 #include <linux/swap.h>
  24 #include <linux/radix-tree.h>
  25 #include <linux/writeback.h>
  26 #include <linux/buffer_head.h> // for block_sync_page
  27 #include <linux/workqueue.h>
  28 #include "ctree.h"
  29 #include "disk-io.h"
  30 #include "transaction.h"
  31 #include "btrfs_inode.h"
  32 #include "volumes.h"
  33 #include "print-tree.h"
  34
  35 #if 0
  36 static int check_tree_block(struct btrfs_root *root, struct extent_buffer *buf)
  37 {
  38         if (extent_buffer_blocknr(buf) != btrfs_header_blocknr(buf)) {
  39                 printk(KERN_CRIT "buf blocknr(buf) is %llu, header is %llu\n",
  40                        (unsigned long long)extent_buffer_blocknr(buf),
  41                        (unsigned long long)btrfs_header_blocknr(buf));
  42                 return 1;
  43         }
  44         return 0;
  45 }
  46 #endif
  47
  48 static struct extent_io_ops btree_extent_io_ops;
  49 static struct workqueue_struct *end_io_workqueue;
  50
  51 struct end_io_wq {
  52         struct bio *bio;
  53         bio_end_io_t *end_io;
  54         void *private;
  55         struct btrfs_fs_info *info;
  56         int error;
  57         int metadata;
  58         struct list_head list;
  59 };
  60
  61 struct extent_map *btree_get_extent(struct inode *inode, struct page *page,
  62                                     size_t page_offset, u64 start, u64 len,
  63                                     int create)
  64 {
  65         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
  66         struct extent_map *em;
  67         int ret;
  68
  69 again:
  70         spin_lock(&em_tree->lock);
  71         em = lookup_extent_mapping(em_tree, start, len);
  72         spin_unlock(&em_tree->lock);
  73         if (em) {
  74                 goto out;
  75         }
  76         em = alloc_extent_map(GFP_NOFS);
  77         if (!em) {
  78                 em = ERR_PTR(-ENOMEM);
  79                 goto out;
  80         }
  81         em->start = 0;
  82         em->len = i_size_read(inode);
  83         em->block_start = 0;
  84         em->bdev = inode->i_sb->s_bdev;
  85
  86         spin_lock(&em_tree->lock);
  87         ret = add_extent_mapping(em_tree, em);
  88         spin_unlock(&em_tree->lock);
  89
  90         if (ret == -EEXIST) {
  91                 free_extent_map(em);
  92                 em = NULL;
  93                 goto again;
  94         } else if (ret) {
  95                 em = ERR_PTR(ret);
  96         }
  97 out:
  98         return em;
  99 }
 100
 101 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
 102 {
 103         return crc32c(seed, data, len);
 104 }
 105
 106 void btrfs_csum_final(u32 crc, char *result)
 107 {
 108         *(__le32 *)result = ~cpu_to_le32(crc);
 109 }
 110
 111 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 112                            int verify)
 113 {
 114         char result[BTRFS_CRC32_SIZE];
 115         unsigned long len;
 116         unsigned long cur_len;
 117         unsigned long offset = BTRFS_CSUM_SIZE;
 118         char *map_token = NULL;
 119         char *kaddr;
 120         unsigned long map_start;
 121         unsigned long map_len;
 122         int err;
 123         u32 crc = ~(u32)0;
 124
 125         len = buf->len - offset;
 126         while(len > 0) {
 127                 err = map_private_extent_buffer(buf, offset, 32,
 128                                         &map_token, &kaddr,
 129                                         &map_start, &map_len, KM_USER0);
 130                 if (err) {
 131                         printk("failed to map extent buffer! %lu\n",
 132                                offset);
 133                         return 1;
 134                 }
 135                 cur_len = min(len, map_len - (offset - map_start));
 136                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
 137                                       crc, cur_len);
 138                 len -= cur_len;
 139                 offset += cur_len;
 140                 unmap_extent_buffer(buf, map_token, KM_USER0);
 141         }
 142         btrfs_csum_final(crc, result);
 143
 144         if (verify) {
 145                 int from_this_trans = 0;
 146
 147                 if (root->fs_info->running_transaction &&
 148                     btrfs_header_generation(buf) ==
 149                     root->fs_info->running_transaction->transid)
 150                         from_this_trans = 1;
 151
 152                 /* FIXME, this is not good */
 153                 if (memcmp_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE)) {
 154                         u32 val;
 155                         u32 found = 0;
 156                         memcpy(&found, result, BTRFS_CRC32_SIZE);
 157
 158                         read_extent_buffer(buf, &val, 0, BTRFS_CRC32_SIZE);
 159                         printk("btrfs: %s checksum verify failed on %llu "
 160                                "wanted %X found %X from_this_trans %d "
 161                                "level %d\n",
 162                                root->fs_info->sb->s_id,
 163                                buf->start, val, found, from_this_trans,
 164                                btrfs_header_level(buf));
 165                         return 1;
 166                 }
 167         } else {
 168                 write_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE);
 169         }
 170         return 0;
 171 }
 172
 173 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 174                                           struct extent_buffer *eb,
 175                                           u64 start)
 176 {
 177         struct extent_io_tree *io_tree;
 178         int ret;
 179         int num_copies = 0;
 180         int mirror_num = 0;
 181
 182         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 183         while (1) {
 184                 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
 185                                                btree_get_extent, mirror_num);
 186                 if (!ret) {
 187                         if (mirror_num)
 188 printk("good read %Lu mirror %d total %d\n", eb->start, mirror_num, num_copies);
 189                         return ret;
 190                 }
 191                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
 192                                               eb->start, eb->len);
 193 printk("failed to read %Lu mirror %d total %d\n", eb->start, mirror_num, num_copies);
 194                 if (num_copies == 1) {
 195 printk("reading %Lu failed only one copy\n", eb->start);
 196                         return ret;
 197                 }
 198                 mirror_num++;
 199                 if (mirror_num > num_copies) {
 200 printk("bailing at mirror %d of %d\n", mirror_num, num_copies);
 201                         return ret;
 202                 }
 203         }
 204 printk("read extent buffer page last\n");
 205         return -EIO;
 206 }
 207
 208 int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 209 {
 210         struct extent_io_tree *tree;
 211         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
 212         u64 found_start;
 213         int found_level;
 214         unsigned long len;
 215         struct extent_buffer *eb;
 216         int ret;
 217
 218         tree = &BTRFS_I(page->mapping->host)->io_tree;
 219
 220         if (page->private == EXTENT_PAGE_PRIVATE)
 221                 goto out;
 222         if (!page->private)
 223                 goto out;
 224         len = page->private >> 2;
 225         if (len == 0) {
 226                 WARN_ON(1);
 227         }
 228         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 229         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE);
 230         BUG_ON(ret);
 231         btrfs_clear_buffer_defrag(eb);
 232         found_start = btrfs_header_bytenr(eb);
 233         if (found_start != start) {
 234                 printk("warning: eb start incorrect %Lu buffer %Lu len %lu\n",
 235                        start, found_start, len);
 236                 WARN_ON(1);
 237                 goto err;
 238         }
 239         if (eb->first_page != page) {
 240                 printk("bad first page %lu %lu\n", eb->first_page->index,
 241                        page->index);
 242                 WARN_ON(1);
 243                 goto err;
 244         }
 245         if (!PageUptodate(page)) {
 246                 printk("csum not up to date page %lu\n", page->index);
 247                 WARN_ON(1);
 248                 goto err;
 249         }
 250         found_level = btrfs_header_level(eb);
 251         spin_lock(&root->fs_info->hash_lock);
 252         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
 253         spin_unlock(&root->fs_info->hash_lock);
 254         csum_tree_block(root, eb, 0);
 255 err:
 256         free_extent_buffer(eb);
 257 out:
 258         return 0;
 259 }
 260
 261 static int btree_writepage_io_hook(struct page *page, u64 start, u64 end)
 262 {
 263         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 264
 265         csum_dirty_buffer(root, page);
 266         return 0;
 267 }
 268
 269 int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 270                                struct extent_state *state)
 271 {
 272         struct extent_io_tree *tree;
 273         u64 found_start;
 274         int found_level;
 275         unsigned long len;
 276         struct extent_buffer *eb;
 277         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 278         int ret = 0;
 279
 280         tree = &BTRFS_I(page->mapping->host)->io_tree;
 281         if (page->private == EXTENT_PAGE_PRIVATE)
 282                 goto out;
 283         if (!page->private)
 284                 goto out;
 285         len = page->private >> 2;
 286         if (len == 0) {
 287                 WARN_ON(1);
 288         }
 289         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 290
 291         btrfs_clear_buffer_defrag(eb);
 292         found_start = btrfs_header_bytenr(eb);
 293         if (found_start != start) {
 294 printk("bad start on %Lu found %Lu\n", eb->start, found_start);
 295                 ret = -EIO;
 296                 goto err;
 297         }
 298         if (eb->first_page != page) {
 299                 printk("bad first page %lu %lu\n", eb->first_page->index,
 300                        page->index);
 301                 WARN_ON(1);
 302                 ret = -EIO;
 303                 goto err;
 304         }
 305         found_level = btrfs_header_level(eb);
 306
 307         ret = csum_tree_block(root, eb, 1);
 308         if (ret)
 309                 ret = -EIO;
 310
 311         end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
 312         end = eb->start + end - 1;
 313         release_extent_buffer_tail_pages(eb);
 314 err:
 315         free_extent_buffer(eb);
 316 out:
 317         return ret;
 318 }
 319
 320 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
 321 static void end_workqueue_bio(struct bio *bio, int err)
 322 #else
 323 static int end_workqueue_bio(struct bio *bio,
 324                                    unsigned int bytes_done, int err)
 325 #endif
 326 {
 327         struct end_io_wq *end_io_wq = bio->bi_private;
 328         struct btrfs_fs_info *fs_info;
 329         unsigned long flags;
 330
 331 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
 332         if (bio->bi_size)
 333                 return 1;
 334 #endif
 335
 336         fs_info = end_io_wq->info;
 337         spin_lock_irqsave(&fs_info->end_io_work_lock, flags);
 338         end_io_wq->error = err;
 339         list_add_tail(&end_io_wq->list, &fs_info->end_io_work_list);
 340         spin_unlock_irqrestore(&fs_info->end_io_work_lock, flags);
 341         queue_work(end_io_workqueue, &fs_info->end_io_work);
 342
 343 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
 344         return 0;
 345 #endif
 346 }
 347
 348 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 349                         int metadata)
 350 {
 351         struct end_io_wq *end_io_wq;
 352         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
 353         if (!end_io_wq)
 354                 return -ENOMEM;
 355
 356         end_io_wq->private = bio->bi_private;
 357         end_io_wq->end_io = bio->bi_end_io;
 358         end_io_wq->info = info;
 359         end_io_wq->error = 0;
 360         end_io_wq->bio = bio;
 361         end_io_wq->metadata = metadata;
 362
 363         bio->bi_private = end_io_wq;
 364         bio->bi_end_io = end_workqueue_bio;
 365         return 0;
 366 }
 367
 368 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 369                                  int mirror_num)
 370 {
 371         struct btrfs_root *root = BTRFS_I(inode)->root;
 372         u64 offset;
 373         int ret;
 374
 375         offset = bio->bi_sector << 9;
 376
 377         if (rw & (1 << BIO_RW)) {
 378                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num);
 379         }
 380
 381         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 1);
 382         BUG_ON(ret);
 383
 384         if (offset == BTRFS_SUPER_INFO_OFFSET) {
 385                 bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
 386                 submit_bio(rw, bio);
 387                 return 0;
 388         }
 389         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num);
 390 }
 391
 392 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 393 {
 394         struct extent_io_tree *tree;
 395         tree = &BTRFS_I(page->mapping->host)->io_tree;
 396         return extent_write_full_page(tree, page, btree_get_extent, wbc);
 397 }
 398
 399 static int btree_writepages(struct address_space *mapping,
 400                             struct writeback_control *wbc)
 401 {
 402         struct extent_io_tree *tree;
 403         tree = &BTRFS_I(mapping->host)->io_tree;
 404         if (wbc->sync_mode == WB_SYNC_NONE) {
 405                 u64 num_dirty;
 406                 u64 start = 0;
 407                 unsigned long thresh = 96 * 1024 * 1024;
 408
 409                 if (wbc->for_kupdate)
 410                         return 0;
 411
 412                 if (current_is_pdflush()) {
 413                         thresh = 96 * 1024 * 1024;
 414                 } else {
 415                         thresh = 8 * 1024 * 1024;
 416                 }
 417                 num_dirty = count_range_bits(tree, &start, (u64)-1,
 418                                              thresh, EXTENT_DIRTY);
 419                 if (num_dirty < thresh) {
 420                         return 0;
 421                 }
 422         }
 423         return extent_writepages(tree, mapping, btree_get_extent, wbc);
 424 }
 425
 426 int btree_readpage(struct file *file, struct page *page)
 427 {
 428         struct extent_io_tree *tree;
 429         tree = &BTRFS_I(page->mapping->host)->io_tree;
 430         return extent_read_full_page(tree, page, btree_get_extent);
 431 }
 432
 433 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 434 {
 435         struct extent_io_tree *tree;
 436         struct extent_map_tree *map;
 437         int ret;
 438
 439         if (page_count(page) > 3) {
 440                 /* once for page->private, once for the caller, once
 441                  * once for the page cache
 442                  */
 443                 return 0;
 444         }
 445         tree = &BTRFS_I(page->mapping->host)->io_tree;
 446         map = &BTRFS_I(page->mapping->host)->extent_tree;
 447         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
 448         if (ret == 1) {
 449                 invalidate_extent_lru(tree, page_offset(page), PAGE_CACHE_SIZE);
 450                 ClearPagePrivate(page);
 451                 set_page_private(page, 0);
 452                 page_cache_release(page);
 453         }
 454         return ret;
 455 }
 456
 457 static void btree_invalidatepage(struct page *page, unsigned long offset)
 458 {
 459         struct extent_io_tree *tree;
 460         tree = &BTRFS_I(page->mapping->host)->io_tree;
 461         extent_invalidatepage(tree, page, offset);
 462         btree_releasepage(page, GFP_NOFS);
 463 }
 464
 465 #if 0
 466 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 467 {
 468         struct buffer_head *bh;
 469         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 470         struct buffer_head *head;
 471         if (!page_has_buffers(page)) {
 472                 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
 473                                         (1 << BH_Dirty)|(1 << BH_Uptodate));
 474         }
 475         head = page_buffers(page);
 476         bh = head;
 477         do {
 478                 if (buffer_dirty(bh))
 479                         csum_tree_block(root, bh, 0);
 480                 bh = bh->b_this_page;
 481         } while (bh != head);
 482         return block_write_full_page(page, btree_get_block, wbc);
 483 }
 484 #endif
 485
 486 static struct address_space_operations btree_aops = {
 487         .readpage       = btree_readpage,
 488         .writepage      = btree_writepage,
 489         .writepages     = btree_writepages,
 490         .releasepage    = btree_releasepage,
 491         .invalidatepage = btree_invalidatepage,
 492         .sync_page      = block_sync_page,
 493 };
 494
 495 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize)
 496 {
 497         struct extent_buffer *buf = NULL;
 498         struct inode *btree_inode = root->fs_info->btree_inode;
 499         int ret = 0;
 500
 501         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 502         if (!buf)
 503                 return 0;
 504         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
 505                                  buf, 0, 0, btree_get_extent, 0);
 506         free_extent_buffer(buf);
 507         return ret;
 508 }
 509
 510 static int close_all_devices(struct btrfs_fs_info *fs_info)
 511 {
 512         struct list_head *list;
 513         struct list_head *next;
 514         struct btrfs_device *device;
 515
 516         list = &fs_info->fs_devices->devices;
 517         list_for_each(next, list) {
 518                 device = list_entry(next, struct btrfs_device, dev_list);
 519                 if (device->bdev && device->bdev != fs_info->sb->s_bdev)
 520                         close_bdev_excl(device->bdev);
 521                 device->bdev = NULL;
 522         }
 523         return 0;
 524 }
 525
 526 int btrfs_verify_block_csum(struct btrfs_root *root,
 527                             struct extent_buffer *buf)
 528 {
 529         return btrfs_buffer_uptodate(buf);
 530 }
 531
 532 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
 533                                             u64 bytenr, u32 blocksize)
 534 {
 535         struct inode *btree_inode = root->fs_info->btree_inode;
 536         struct extent_buffer *eb;
 537         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 538                                 bytenr, blocksize, GFP_NOFS);
 539         return eb;
 540 }
 541
 542 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
 543                                                  u64 bytenr, u32 blocksize)
 544 {
 545         struct inode *btree_inode = root->fs_info->btree_inode;
 546         struct extent_buffer *eb;
 547
 548         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 549                                  bytenr, blocksize, NULL, GFP_NOFS);
 550         return eb;
 551 }
 552
 553
 554 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
 555                                       u32 blocksize)
 556 {
 557         struct extent_buffer *buf = NULL;
 558         struct inode *btree_inode = root->fs_info->btree_inode;
 559         struct extent_io_tree *io_tree;
 560         int ret;
 561
 562         io_tree = &BTRFS_I(btree_inode)->io_tree;
 563
 564         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 565         if (!buf)
 566                 return NULL;
 567
 568         ret = btree_read_extent_buffer_pages(root, buf, 0);
 569
 570         if (ret == 0) {
 571                 buf->flags |= EXTENT_UPTODATE;
 572         }
 573         return buf;
 574
 575 }
 576
 577 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 578                      struct extent_buffer *buf)
 579 {
 580         struct inode *btree_inode = root->fs_info->btree_inode;
 581         if (btrfs_header_generation(buf) ==
 582             root->fs_info->running_transaction->transid)
 583                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
 584                                           buf);
 585         return 0;
 586 }
 587
 588 int wait_on_tree_block_writeback(struct btrfs_root *root,
 589                                  struct extent_buffer *buf)
 590 {
 591         struct inode *btree_inode = root->fs_info->btree_inode;
 592         wait_on_extent_buffer_writeback(&BTRFS_I(btree_inode)->io_tree,
 593                                         buf);
 594         return 0;
 595 }
 596
 597 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
 598                         u32 stripesize, struct btrfs_root *root,
 599                         struct btrfs_fs_info *fs_info,
 600                         u64 objectid)
 601 {
 602         root->node = NULL;
 603         root->inode = NULL;
 604         root->commit_root = NULL;
 605         root->sectorsize = sectorsize;
 606         root->nodesize = nodesize;
 607         root->leafsize = leafsize;
 608         root->stripesize = stripesize;
 609         root->ref_cows = 0;
 610         root->track_dirty = 0;
 611
 612         root->fs_info = fs_info;
 613         root->objectid = objectid;
 614         root->last_trans = 0;
 615         root->highest_inode = 0;
 616         root->last_inode_alloc = 0;
 617         root->name = NULL;
 618         root->in_sysfs = 0;
 619
 620         INIT_LIST_HEAD(&root->dirty_list);
 621         memset(&root->root_key, 0, sizeof(root->root_key));
 622         memset(&root->root_item, 0, sizeof(root->root_item));
 623         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
 624         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
 625         init_completion(&root->kobj_unregister);
 626         root->defrag_running = 0;
 627         root->defrag_level = 0;
 628         root->root_key.objectid = objectid;
 629         return 0;
 630 }
 631
 632 static int find_and_setup_root(struct btrfs_root *tree_root,
 633                                struct btrfs_fs_info *fs_info,
 634                                u64 objectid,
 635                                struct btrfs_root *root)
 636 {
 637         int ret;
 638         u32 blocksize;
 639
 640         __setup_root(tree_root->nodesize, tree_root->leafsize,
 641                      tree_root->sectorsize, tree_root->stripesize,
 642                      root, fs_info, objectid);
 643         ret = btrfs_find_last_root(tree_root, objectid,
 644                                    &root->root_item, &root->root_key);
 645         BUG_ON(ret);
 646
 647         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 648         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 649                                      blocksize);
 650         BUG_ON(!root->node);
 651         return 0;
 652 }
 653
 654 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_fs_info *fs_info,
 655                                                struct btrfs_key *location)
 656 {
 657         struct btrfs_root *root;
 658         struct btrfs_root *tree_root = fs_info->tree_root;
 659         struct btrfs_path *path;
 660         struct extent_buffer *l;
 661         u64 highest_inode;
 662         u32 blocksize;
 663         int ret = 0;
 664
 665         root = kzalloc(sizeof(*root), GFP_NOFS);
 666         if (!root)
 667                 return ERR_PTR(-ENOMEM);
 668         if (location->offset == (u64)-1) {
 669                 ret = find_and_setup_root(tree_root, fs_info,
 670                                           location->objectid, root);
 671                 if (ret) {
 672                         kfree(root);
 673                         return ERR_PTR(ret);
 674                 }
 675                 goto insert;
 676         }
 677
 678         __setup_root(tree_root->nodesize, tree_root->leafsize,
 679                      tree_root->sectorsize, tree_root->stripesize,
 680                      root, fs_info, location->objectid);
 681
 682         path = btrfs_alloc_path();
 683         BUG_ON(!path);
 684         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
 685         if (ret != 0) {
 686                 if (ret > 0)
 687                         ret = -ENOENT;
 688                 goto out;
 689         }
 690         l = path->nodes[0];
 691         read_extent_buffer(l, &root->root_item,
 692                btrfs_item_ptr_offset(l, path->slots[0]),
 693                sizeof(root->root_item));
 694         memcpy(&root->root_key, location, sizeof(*location));
 695         ret = 0;
 696 out:
 697         btrfs_release_path(root, path);
 698         btrfs_free_path(path);
 699         if (ret) {
 700                 kfree(root);
 701                 return ERR_PTR(ret);
 702         }
 703         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 704         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 705                                      blocksize);
 706         BUG_ON(!root->node);
 707 insert:
 708         root->ref_cows = 1;
 709         ret = btrfs_find_highest_inode(root, &highest_inode);
 710         if (ret == 0) {
 711                 root->highest_inode = highest_inode;
 712                 root->last_inode_alloc = highest_inode;
 713         }
 714         return root;
 715 }
 716
 717 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
 718                                         u64 root_objectid)
 719 {
 720         struct btrfs_root *root;
 721
 722         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
 723                 return fs_info->tree_root;
 724         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
 725                 return fs_info->extent_root;
 726
 727         root = radix_tree_lookup(&fs_info->fs_roots_radix,
 728                                  (unsigned long)root_objectid);
 729         return root;
 730 }
 731
 732 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
 733                                               struct btrfs_key *location)
 734 {
 735         struct btrfs_root *root;
 736         int ret;
 737
 738         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
 739                 return fs_info->tree_root;
 740         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
 741                 return fs_info->extent_root;
 742
 743         root = radix_tree_lookup(&fs_info->fs_roots_radix,
 744                                  (unsigned long)location->objectid);
 745         if (root)
 746                 return root;
 747
 748         root = btrfs_read_fs_root_no_radix(fs_info, location);
 749         if (IS_ERR(root))
 750                 return root;
 751         ret = radix_tree_insert(&fs_info->fs_roots_radix,
 752                                 (unsigned long)root->root_key.objectid,
 753                                 root);
 754         if (ret) {
 755                 free_extent_buffer(root->node);
 756                 kfree(root);
 757                 return ERR_PTR(ret);
 758         }
 759         ret = btrfs_find_dead_roots(fs_info->tree_root,
 760                                     root->root_key.objectid, root);
 761         BUG_ON(ret);
 762
 763         return root;
 764 }
 765
 766 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
 767                                       struct btrfs_key *location,
 768                                       const char *name, int namelen)
 769 {
 770         struct btrfs_root *root;
 771         int ret;
 772
 773         root = btrfs_read_fs_root_no_name(fs_info, location);
 774         if (!root)
 775                 return NULL;
 776
 777         if (root->in_sysfs)
 778                 return root;
 779
 780         ret = btrfs_set_root_name(root, name, namelen);
 781         if (ret) {
 782                 free_extent_buffer(root->node);
 783                 kfree(root);
 784                 return ERR_PTR(ret);
 785         }
 786
 787         ret = btrfs_sysfs_add_root(root);
 788         if (ret) {
 789                 free_extent_buffer(root->node);
 790                 kfree(root->name);
 791                 kfree(root);
 792                 return ERR_PTR(ret);
 793         }
 794         root->in_sysfs = 1;
 795         return root;
 796 }
 797 #if 0
 798 static int add_hasher(struct btrfs_fs_info *info, char *type) {
 799         struct btrfs_hasher *hasher;
 800
 801         hasher = kmalloc(sizeof(*hasher), GFP_NOFS);
 802         if (!hasher)
 803                 return -ENOMEM;
 804         hasher->hash_tfm = crypto_alloc_hash(type, 0, CRYPTO_ALG_ASYNC);
 805         if (!hasher->hash_tfm) {
 806                 kfree(hasher);
 807                 return -EINVAL;
 808         }
 809         spin_lock(&info->hash_lock);
 810         list_add(&hasher->list, &info->hashers);
 811         spin_unlock(&info->hash_lock);
 812         return 0;
 813 }
 814 #endif
 815
 816 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
 817 {
 818         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
 819         int ret = 0;
 820         struct list_head *cur;
 821         struct btrfs_device *device;
 822         struct backing_dev_info *bdi;
 823
 824         list_for_each(cur, &info->fs_devices->devices) {
 825                 device = list_entry(cur, struct btrfs_device, dev_list);
 826                 bdi = blk_get_backing_dev_info(device->bdev);
 827                 if (bdi && bdi_congested(bdi, bdi_bits)) {
 828                         ret = 1;
 829                         break;
 830                 }
 831         }
 832         return ret;
 833 }
 834
 835 void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
 836 {
 837         struct list_head *cur;
 838         struct btrfs_device *device;
 839         struct btrfs_fs_info *info;
 840
 841         info = (struct btrfs_fs_info *)bdi->unplug_io_data;
 842         list_for_each(cur, &info->fs_devices->devices) {
 843                 device = list_entry(cur, struct btrfs_device, dev_list);
 844                 bdi = blk_get_backing_dev_info(device->bdev);
 845                 if (bdi->unplug_io_fn) {
 846                         bdi->unplug_io_fn(bdi, page);
 847                 }
 848         }
 849 }
 850
 851 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
 852 {
 853         bdi_init(bdi);
 854         bdi->ra_pages   = default_backing_dev_info.ra_pages * 4;
 855         bdi->state              = 0;
 856         bdi->capabilities       = default_backing_dev_info.capabilities;
 857         bdi->unplug_io_fn       = btrfs_unplug_io_fn;
 858         bdi->unplug_io_data     = info;
 859         bdi->congested_fn       = btrfs_congested_fn;
 860         bdi->congested_data     = info;
 861         return 0;
 862 }
 863
 864 static int bio_ready_for_csum(struct bio *bio)
 865 {
 866         u64 length = 0;
 867         u64 buf_len = 0;
 868         u64 start = 0;
 869         struct page *page;
 870         struct extent_io_tree *io_tree = NULL;
 871         struct btrfs_fs_info *info = NULL;
 872         struct bio_vec *bvec;
 873         int i;
 874         int ret;
 875
 876         bio_for_each_segment(bvec, bio, i) {
 877                 page = bvec->bv_page;
 878                 if (page->private == EXTENT_PAGE_PRIVATE) {
 879                         length += bvec->bv_len;
 880                         continue;
 881                 }
 882                 if (!page->private) {
 883                         length += bvec->bv_len;
 884                         continue;
 885                 }
 886                 length = bvec->bv_len;
 887                 buf_len = page->private >> 2;
 888                 start = page_offset(page) + bvec->bv_offset;
 889                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
 890                 info = BTRFS_I(page->mapping->host)->root->fs_info;
 891         }
 892         /* are we fully contained in this bio? */
 893         if (buf_len <= length)
 894                 return 1;
 895
 896         ret = extent_range_uptodate(io_tree, start + length,
 897                                     start + buf_len - 1);
 898         if (ret == 1)
 899                 return ret;
 900         return ret;
 901 }
 902
 903 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
 904 void btrfs_end_io_csum(void *p)
 905 #else
 906 void btrfs_end_io_csum(struct work_struct *work)
 907 #endif
 908 {
 909 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
 910         struct btrfs_fs_info *fs_info = p;
 911 #else
 912         struct btrfs_fs_info *fs_info = container_of(work,
 913                                                      struct btrfs_fs_info,
 914                                                      end_io_work);
 915 #endif
 916         unsigned long flags;
 917         struct end_io_wq *end_io_wq;
 918         struct bio *bio;
 919         struct list_head *next;
 920         int error;
 921         int was_empty;
 922
 923         while(1) {
 924                 spin_lock_irqsave(&fs_info->end_io_work_lock, flags);
 925                 if (list_empty(&fs_info->end_io_work_list)) {
 926                         spin_unlock_irqrestore(&fs_info->end_io_work_lock,
 927                                                flags);
 928                         return;
 929                 }
 930                 next = fs_info->end_io_work_list.next;
 931                 list_del(next);
 932                 spin_unlock_irqrestore(&fs_info->end_io_work_lock, flags);
 933
 934                 end_io_wq = list_entry(next, struct end_io_wq, list);
 935
 936                 bio = end_io_wq->bio;
 937                 if (end_io_wq->metadata && !bio_ready_for_csum(bio)) {
 938                         spin_lock_irqsave(&fs_info->end_io_work_lock, flags);
 939                         was_empty = list_empty(&fs_info->end_io_work_list);
 940                         list_add_tail(&end_io_wq->list,
 941                                       &fs_info->end_io_work_list);
 942                         spin_unlock_irqrestore(&fs_info->end_io_work_lock,
 943                                                flags);
 944                         if (was_empty)
 945                                 return;
 946                         continue;
 947                 }
 948                 error = end_io_wq->error;
 949                 bio->bi_private = end_io_wq->private;
 950                 bio->bi_end_io = end_io_wq->end_io;
 951                 kfree(end_io_wq);
 952                 bio_endio(bio, error);
 953         }
 954 }
 955
 956
 957 struct btrfs_root *open_ctree(struct super_block *sb,
 958                               struct btrfs_fs_devices *fs_devices)
 959 {
 960         u32 sectorsize;
 961         u32 nodesize;
 962         u32 leafsize;
 963         u32 blocksize;
 964         u32 stripesize;
 965         struct btrfs_root *extent_root = kmalloc(sizeof(struct btrfs_root),
 966                                                  GFP_NOFS);
 967         struct btrfs_root *tree_root = kmalloc(sizeof(struct btrfs_root),
 968                                                GFP_NOFS);
 969         struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
 970                                                 GFP_NOFS);
 971         struct btrfs_root *chunk_root = kmalloc(sizeof(struct btrfs_root),
 972                                                 GFP_NOFS);
 973         struct btrfs_root *dev_root = kmalloc(sizeof(struct btrfs_root),
 974                                               GFP_NOFS);
 975         int ret;
 976         int err = -EINVAL;
 977         struct btrfs_super_block *disk_super;
 978
 979         if (!extent_root || !tree_root || !fs_info) {
 980                 err = -ENOMEM;
 981                 goto fail;
 982         }
 983         end_io_workqueue = create_workqueue("btrfs-end-io");
 984         BUG_ON(!end_io_workqueue);
 985
 986         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
 987         INIT_LIST_HEAD(&fs_info->trans_list);
 988         INIT_LIST_HEAD(&fs_info->dead_roots);
 989         INIT_LIST_HEAD(&fs_info->hashers);
 990         INIT_LIST_HEAD(&fs_info->end_io_work_list);
 991         spin_lock_init(&fs_info->hash_lock);
 992         spin_lock_init(&fs_info->end_io_work_lock);
 993         spin_lock_init(&fs_info->delalloc_lock);
 994         spin_lock_init(&fs_info->new_trans_lock);
 995
 996         init_completion(&fs_info->kobj_unregister);
 997         sb_set_blocksize(sb, BTRFS_SUPER_INFO_SIZE);
 998         fs_info->tree_root = tree_root;
 999         fs_info->extent_root = extent_root;
1000         fs_info->chunk_root = chunk_root;
1001         fs_info->dev_root = dev_root;
1002         fs_info->fs_devices = fs_devices;
1003         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1004         INIT_LIST_HEAD(&fs_info->space_info);
1005         btrfs_mapping_init(&fs_info->mapping_tree);
1006         fs_info->sb = sb;
1007         fs_info->max_extent = (u64)-1;
1008         fs_info->max_inline = 8192 * 1024;
1009         setup_bdi(fs_info, &fs_info->bdi);
1010         fs_info->btree_inode = new_inode(sb);
1011         fs_info->btree_inode->i_ino = 1;
1012         fs_info->btree_inode->i_nlink = 1;
1013         fs_info->btree_inode->i_size = sb->s_bdev->bd_inode->i_size;
1014         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1015         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1016
1017         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1018                              fs_info->btree_inode->i_mapping,
1019                              GFP_NOFS);
1020         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1021                              GFP_NOFS);
1022
1023         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1024
1025         extent_io_tree_init(&fs_info->free_space_cache,
1026                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1027         extent_io_tree_init(&fs_info->block_group_cache,
1028                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1029         extent_io_tree_init(&fs_info->pinned_extents,
1030                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1031         extent_io_tree_init(&fs_info->pending_del,
1032                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1033         extent_io_tree_init(&fs_info->extent_ins,
1034                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1035         fs_info->do_barriers = 1;
1036
1037         INIT_WORK(&fs_info->end_io_work, btrfs_end_io_csum);
1038 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1039         INIT_WORK(&fs_info->trans_work, btrfs_transaction_cleaner, fs_info);
1040 #else
1041         INIT_DELAYED_WORK(&fs_info->trans_work, btrfs_transaction_cleaner);
1042 #endif
1043         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1044         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1045                sizeof(struct btrfs_key));
1046         insert_inode_hash(fs_info->btree_inode);
1047         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1048
1049         mutex_init(&fs_info->trans_mutex);
1050         mutex_init(&fs_info->fs_mutex);
1051
1052 #if 0
1053         ret = add_hasher(fs_info, "crc32c");
1054         if (ret) {
1055                 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
1056                 err = -ENOMEM;
1057                 goto fail_iput;
1058         }
1059 #endif
1060         __setup_root(4096, 4096, 4096, 4096, tree_root,
1061                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1062
1063         fs_info->sb_buffer = read_tree_block(tree_root,
1064                                              BTRFS_SUPER_INFO_OFFSET,
1065                                              4096);
1066
1067         if (!fs_info->sb_buffer)
1068                 goto fail_iput;
1069
1070         read_extent_buffer(fs_info->sb_buffer, &fs_info->super_copy, 0,
1071                            sizeof(fs_info->super_copy));
1072
1073         read_extent_buffer(fs_info->sb_buffer, fs_info->fsid,
1074                            (unsigned long)btrfs_super_fsid(fs_info->sb_buffer),
1075                            BTRFS_FSID_SIZE);
1076
1077         disk_super = &fs_info->super_copy;
1078         if (!btrfs_super_root(disk_super))
1079                 goto fail_sb_buffer;
1080
1081         if (btrfs_super_num_devices(disk_super) != fs_devices->num_devices) {
1082                 printk("Btrfs: wanted %llu devices, but found %llu\n",
1083                        (unsigned long long)btrfs_super_num_devices(disk_super),
1084                        (unsigned long long)fs_devices->num_devices);
1085                 goto fail_sb_buffer;
1086         }
1087         nodesize = btrfs_super_nodesize(disk_super);
1088         leafsize = btrfs_super_leafsize(disk_super);
1089         sectorsize = btrfs_super_sectorsize(disk_super);
1090         stripesize = btrfs_super_stripesize(disk_super);
1091         tree_root->nodesize = nodesize;
1092         tree_root->leafsize = leafsize;
1093         tree_root->sectorsize = sectorsize;
1094         tree_root->stripesize = stripesize;
1095         sb_set_blocksize(sb, sectorsize);
1096
1097         i_size_write(fs_info->btree_inode,
1098                      btrfs_super_total_bytes(disk_super));
1099
1100         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1101                     sizeof(disk_super->magic))) {
1102                 printk("btrfs: valid FS not found on %s\n", sb->s_id);
1103                 goto fail_sb_buffer;
1104         }
1105
1106         mutex_lock(&fs_info->fs_mutex);
1107
1108         ret = btrfs_read_sys_array(tree_root);
1109         BUG_ON(ret);
1110
1111         blocksize = btrfs_level_size(tree_root,
1112                                      btrfs_super_chunk_root_level(disk_super));
1113
1114         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1115                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1116
1117         chunk_root->node = read_tree_block(chunk_root,
1118                                            btrfs_super_chunk_root(disk_super),
1119                                            blocksize);
1120         BUG_ON(!chunk_root->node);
1121
1122         ret = btrfs_read_chunk_tree(chunk_root);
1123         BUG_ON(ret);
1124
1125         blocksize = btrfs_level_size(tree_root,
1126                                      btrfs_super_root_level(disk_super));
1127
1128
1129         tree_root->node = read_tree_block(tree_root,
1130                                           btrfs_super_root(disk_super),
1131                                           blocksize);
1132         if (!tree_root->node)
1133                 goto fail_sb_buffer;
1134
1135
1136         ret = find_and_setup_root(tree_root, fs_info,
1137                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1138         if (ret)
1139                 goto fail_tree_root;
1140         extent_root->track_dirty = 1;
1141
1142         ret = find_and_setup_root(tree_root, fs_info,
1143                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
1144         dev_root->track_dirty = 1;
1145
1146         if (ret)
1147                 goto fail_extent_root;
1148
1149         btrfs_read_block_groups(extent_root);
1150
1151         fs_info->generation = btrfs_super_generation(disk_super) + 1;
1152         fs_info->data_alloc_profile = (u64)-1;
1153         fs_info->metadata_alloc_profile = (u64)-1;
1154         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1155
1156         mutex_unlock(&fs_info->fs_mutex);
1157         return tree_root;
1158
1159 fail_extent_root:
1160         free_extent_buffer(extent_root->node);
1161 fail_tree_root:
1162         mutex_unlock(&fs_info->fs_mutex);
1163         free_extent_buffer(tree_root->node);
1164 fail_sb_buffer:
1165         free_extent_buffer(fs_info->sb_buffer);
1166         extent_io_tree_empty_lru(&BTRFS_I(fs_info->btree_inode)->io_tree);
1167 fail_iput:
1168         iput(fs_info->btree_inode);
1169 fail:
1170         close_all_devices(fs_info);
1171         kfree(extent_root);
1172         kfree(tree_root);
1173         bdi_destroy(&fs_info->bdi);
1174         kfree(fs_info);
1175         return ERR_PTR(err);
1176 }
1177
1178 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1179 {
1180         char b[BDEVNAME_SIZE];
1181
1182         if (uptodate) {
1183                 set_buffer_uptodate(bh);
1184         } else {
1185                 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1186                         printk(KERN_WARNING "lost page write due to "
1187                                         "I/O error on %s\n",
1188                                        bdevname(bh->b_bdev, b));
1189                 }
1190                 set_buffer_write_io_error(bh);
1191                 clear_buffer_uptodate(bh);
1192         }
1193         unlock_buffer(bh);
1194         put_bh(bh);
1195 }
1196
1197 int write_all_supers(struct btrfs_root *root)
1198 {
1199         struct list_head *cur;
1200         struct list_head *head = &root->fs_info->fs_devices->devices;
1201         struct btrfs_device *dev;
1202         struct extent_buffer *sb;
1203         struct btrfs_dev_item *dev_item;
1204         struct buffer_head *bh;
1205         int ret;
1206         int do_barriers;
1207
1208         do_barriers = !btrfs_test_opt(root, NOBARRIER);
1209
1210         sb = root->fs_info->sb_buffer;
1211         dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
1212                                                       dev_item);
1213         list_for_each(cur, head) {
1214                 dev = list_entry(cur, struct btrfs_device, dev_list);
1215                 btrfs_set_device_type(sb, dev_item, dev->type);
1216                 btrfs_set_device_id(sb, dev_item, dev->devid);
1217                 btrfs_set_device_total_bytes(sb, dev_item, dev->total_bytes);
1218                 btrfs_set_device_bytes_used(sb, dev_item, dev->bytes_used);
1219                 btrfs_set_device_io_align(sb, dev_item, dev->io_align);
1220                 btrfs_set_device_io_width(sb, dev_item, dev->io_width);
1221                 btrfs_set_device_sector_size(sb, dev_item, dev->sector_size);
1222                 write_extent_buffer(sb, dev->uuid,
1223                                     (unsigned long)btrfs_device_uuid(dev_item),
1224                                     BTRFS_DEV_UUID_SIZE);
1225
1226                 btrfs_set_header_flag(sb, BTRFS_HEADER_FLAG_WRITTEN);
1227                 csum_tree_block(root, sb, 0);
1228
1229                 bh = __getblk(dev->bdev, BTRFS_SUPER_INFO_OFFSET /
1230                               root->fs_info->sb->s_blocksize,
1231                               BTRFS_SUPER_INFO_SIZE);
1232
1233                 read_extent_buffer(sb, bh->b_data, 0, BTRFS_SUPER_INFO_SIZE);
1234                 dev->pending_io = bh;
1235
1236                 get_bh(bh);
1237                 set_buffer_uptodate(bh);
1238                 lock_buffer(bh);
1239                 bh->b_end_io = btrfs_end_buffer_write_sync;
1240
1241                 if (do_barriers && dev->barriers) {
1242                         ret = submit_bh(WRITE_BARRIER, bh);
1243                         if (ret == -EOPNOTSUPP) {
1244                                 printk("btrfs: disabling barriers on dev %s\n",
1245                                        dev->name);
1246                                 set_buffer_uptodate(bh);
1247                                 dev->barriers = 0;
1248                                 get_bh(bh);
1249                                 lock_buffer(bh);
1250                                 ret = submit_bh(WRITE, bh);
1251                         }
1252                 } else {
1253                         ret = submit_bh(WRITE, bh);
1254                 }
1255                 BUG_ON(ret);
1256         }
1257
1258         list_for_each(cur, head) {
1259                 dev = list_entry(cur, struct btrfs_device, dev_list);
1260                 BUG_ON(!dev->pending_io);
1261                 bh = dev->pending_io;
1262                 wait_on_buffer(bh);
1263                 if (!buffer_uptodate(dev->pending_io)) {
1264                         if (do_barriers && dev->barriers) {
1265                                 printk("btrfs: disabling barriers on dev %s\n",
1266                                        dev->name);
1267                                 set_buffer_uptodate(bh);
1268                                 get_bh(bh);
1269                                 lock_buffer(bh);
1270                                 dev->barriers = 0;
1271                                 ret = submit_bh(WRITE, bh);
1272                                 BUG_ON(ret);
1273                                 wait_on_buffer(bh);
1274                                 BUG_ON(!buffer_uptodate(bh));
1275                         } else {
1276                                 BUG();
1277                         }
1278
1279                 }
1280                 dev->pending_io = NULL;
1281                 brelse(bh);
1282         }
1283         return 0;
1284 }
1285
1286 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
1287                       *root)
1288 {
1289         int ret;
1290
1291         ret = write_all_supers(root);
1292 #if 0
1293         if (!btrfs_test_opt(root, NOBARRIER))
1294                 blkdev_issue_flush(sb->s_bdev, NULL);
1295         set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, super);
1296         ret = sync_page_range_nolock(btree_inode, btree_inode->i_mapping,
1297                                      super->start, super->len);
1298         if (!btrfs_test_opt(root, NOBARRIER))
1299                 blkdev_issue_flush(sb->s_bdev, NULL);
1300 #endif
1301         return ret;
1302 }
1303
1304 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
1305 {
1306         radix_tree_delete(&fs_info->fs_roots_radix,
1307                           (unsigned long)root->root_key.objectid);
1308         if (root->in_sysfs)
1309                 btrfs_sysfs_del_root(root);
1310         if (root->inode)
1311                 iput(root->inode);
1312         if (root->node)
1313                 free_extent_buffer(root->node);
1314         if (root->commit_root)
1315                 free_extent_buffer(root->commit_root);
1316         if (root->name)
1317                 kfree(root->name);
1318         kfree(root);
1319         return 0;
1320 }
1321
1322 static int del_fs_roots(struct btrfs_fs_info *fs_info)
1323 {
1324         int ret;
1325         struct btrfs_root *gang[8];
1326         int i;
1327
1328         while(1) {
1329                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1330                                              (void **)gang, 0,
1331                                              ARRAY_SIZE(gang));
1332                 if (!ret)
1333                         break;
1334                 for (i = 0; i < ret; i++)
1335                         btrfs_free_fs_root(fs_info, gang[i]);
1336         }
1337         return 0;
1338 }
1339
1340 int close_ctree(struct btrfs_root *root)
1341 {
1342         int ret;
1343         struct btrfs_trans_handle *trans;
1344         struct btrfs_fs_info *fs_info = root->fs_info;
1345
1346         fs_info->closing = 1;
1347         btrfs_transaction_flush_work(root);
1348         mutex_lock(&fs_info->fs_mutex);
1349         btrfs_defrag_dirty_roots(root->fs_info);
1350         trans = btrfs_start_transaction(root, 1);
1351         ret = btrfs_commit_transaction(trans, root);
1352         /* run commit again to  drop the original snapshot */
1353         trans = btrfs_start_transaction(root, 1);
1354         btrfs_commit_transaction(trans, root);
1355         ret = btrfs_write_and_wait_transaction(NULL, root);
1356         BUG_ON(ret);
1357         write_ctree_super(NULL, root);
1358         mutex_unlock(&fs_info->fs_mutex);
1359
1360         if (fs_info->delalloc_bytes) {
1361                 printk("btrfs: at unmount delalloc count %Lu\n",
1362                        fs_info->delalloc_bytes);
1363         }
1364         if (fs_info->extent_root->node)
1365                 free_extent_buffer(fs_info->extent_root->node);
1366
1367         if (fs_info->tree_root->node)
1368                 free_extent_buffer(fs_info->tree_root->node);
1369
1370         if (root->fs_info->chunk_root->node);
1371                 free_extent_buffer(root->fs_info->chunk_root->node);
1372
1373         if (root->fs_info->dev_root->node);
1374                 free_extent_buffer(root->fs_info->dev_root->node);
1375
1376         free_extent_buffer(fs_info->sb_buffer);
1377
1378         btrfs_free_block_groups(root->fs_info);
1379         del_fs_roots(fs_info);
1380
1381         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1382
1383         extent_io_tree_empty_lru(&fs_info->free_space_cache);
1384         extent_io_tree_empty_lru(&fs_info->block_group_cache);
1385         extent_io_tree_empty_lru(&fs_info->pinned_extents);
1386         extent_io_tree_empty_lru(&fs_info->pending_del);
1387         extent_io_tree_empty_lru(&fs_info->extent_ins);
1388         extent_io_tree_empty_lru(&BTRFS_I(fs_info->btree_inode)->io_tree);
1389
1390         truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
1391         flush_workqueue(end_io_workqueue);
1392         destroy_workqueue(end_io_workqueue);
1393
1394         iput(fs_info->btree_inode);
1395 #if 0
1396         while(!list_empty(&fs_info->hashers)) {
1397                 struct btrfs_hasher *hasher;
1398                 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
1399                                     hashers);
1400                 list_del(&hasher->hashers);
1401                 crypto_free_hash(&fs_info->hash_tfm);
1402                 kfree(hasher);
1403         }
1404 #endif
1405         close_all_devices(fs_info);
1406         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1407         bdi_destroy(&fs_info->bdi);
1408
1409         kfree(fs_info->extent_root);
1410         kfree(fs_info->tree_root);
1411         kfree(fs_info->chunk_root);
1412         kfree(fs_info->dev_root);
1413         return 0;
1414 }
1415
1416 int btrfs_buffer_uptodate(struct extent_buffer *buf)
1417 {
1418         struct inode *btree_inode = buf->first_page->mapping->host;
1419         return extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
1420 }
1421
1422 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
1423 {
1424         struct inode *btree_inode = buf->first_page->mapping->host;
1425         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
1426                                           buf);
1427 }
1428
1429 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
1430 {
1431         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1432         u64 transid = btrfs_header_generation(buf);
1433         struct inode *btree_inode = root->fs_info->btree_inode;
1434
1435         if (transid != root->fs_info->generation) {
1436                 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
1437                         (unsigned long long)buf->start,
1438                         transid, root->fs_info->generation);
1439                 WARN_ON(1);
1440         }
1441         set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
1442 }
1443
1444 void btrfs_throttle(struct btrfs_root *root)
1445 {
1446         struct backing_dev_info *bdi;
1447
1448         bdi = root->fs_info->sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
1449         if (root->fs_info->throttles && bdi_write_congested(bdi)) {
1450 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,18)
1451                 congestion_wait(WRITE, HZ/20);
1452 #else
1453                 blk_congestion_wait(WRITE, HZ/20);
1454 #endif
1455         }
1456 }
1457
1458 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
1459 {
1460         balance_dirty_pages_ratelimited_nr(
1461                                    root->fs_info->btree_inode->i_mapping, 1);
1462 }
1463
1464 void btrfs_set_buffer_defrag(struct extent_buffer *buf)
1465 {
1466         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1467         struct inode *btree_inode = root->fs_info->btree_inode;
1468         set_extent_bits(&BTRFS_I(btree_inode)->io_tree, buf->start,
1469                         buf->start + buf->len - 1, EXTENT_DEFRAG, GFP_NOFS);
1470 }
1471
1472 void btrfs_set_buffer_defrag_done(struct extent_buffer *buf)
1473 {
1474         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1475         struct inode *btree_inode = root->fs_info->btree_inode;
1476         set_extent_bits(&BTRFS_I(btree_inode)->io_tree, buf->start,
1477                         buf->start + buf->len - 1, EXTENT_DEFRAG_DONE,
1478                         GFP_NOFS);
1479 }
1480
1481 int btrfs_buffer_defrag(struct extent_buffer *buf)
1482 {
1483         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1484         struct inode *btree_inode = root->fs_info->btree_inode;
1485         return test_range_bit(&BTRFS_I(btree_inode)->io_tree,
1486                      buf->start, buf->start + buf->len - 1, EXTENT_DEFRAG, 0);
1487 }
1488
1489 int btrfs_buffer_defrag_done(struct extent_buffer *buf)
1490 {
1491         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1492         struct inode *btree_inode = root->fs_info->btree_inode;
1493         return test_range_bit(&BTRFS_I(btree_inode)->io_tree,
1494                      buf->start, buf->start + buf->len - 1,
1495                      EXTENT_DEFRAG_DONE, 0);
1496 }
1497
1498 int btrfs_clear_buffer_defrag_done(struct extent_buffer *buf)
1499 {
1500         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1501         struct inode *btree_inode = root->fs_info->btree_inode;
1502         return clear_extent_bits(&BTRFS_I(btree_inode)->io_tree,
1503                      buf->start, buf->start + buf->len - 1,
1504                      EXTENT_DEFRAG_DONE, GFP_NOFS);
1505 }
1506
1507 int btrfs_clear_buffer_defrag(struct extent_buffer *buf)
1508 {
1509         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1510         struct inode *btree_inode = root->fs_info->btree_inode;
1511         return clear_extent_bits(&BTRFS_I(btree_inode)->io_tree,
1512                      buf->start, buf->start + buf->len - 1,
1513                      EXTENT_DEFRAG, GFP_NOFS);
1514 }
1515
1516 int btrfs_read_buffer(struct extent_buffer *buf)
1517 {
1518         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1519         int ret;
1520         ret = btree_read_extent_buffer_pages(root, buf, 0);
1521         if (ret == 0) {
1522                 buf->flags |= EXTENT_UPTODATE;
1523         }
1524         return ret;
1525 }
1526
1527 static struct extent_io_ops btree_extent_io_ops = {
1528         .writepage_io_hook = btree_writepage_io_hook,
1529         .readpage_end_io_hook = btree_readpage_end_io_hook,
1530         .submit_bio_hook = btree_submit_bio_hook,
1531         /* note we're sharing with inode.c for the merge bio hook */
1532         .merge_bio_hook = btrfs_merge_bio_hook,
1533 };