SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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/version.h>
  20 #include <linux/fs.h>
  21 #include <linux/blkdev.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 <linux/kthread.h>
  29 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,20)
  30 # include <linux/freezer.h>
  31 #else
  32 # include <linux/sched.h>
  33 #endif
  34 #include "crc32c.h"
  35 #include "ctree.h"
  36 #include "disk-io.h"
  37 #include "transaction.h"
  38 #include "btrfs_inode.h"
  39 #include "volumes.h"
  40 #include "print-tree.h"
  41 #include "async-thread.h"
  42 #include "locking.h"
  43 #include "ref-cache.h"
  44 #include "tree-log.h"
  45
  46 #if 0
  47 static int check_tree_block(struct btrfs_root *root, struct extent_buffer *buf)
  48 {
  49         if (extent_buffer_blocknr(buf) != btrfs_header_blocknr(buf)) {
  50                 printk(KERN_CRIT "buf blocknr(buf) is %llu, header is %llu\n",
  51                        (unsigned long long)extent_buffer_blocknr(buf),
  52                        (unsigned long long)btrfs_header_blocknr(buf));
  53                 return 1;
  54         }
  55         return 0;
  56 }
  57 #endif
  58
  59 static struct extent_io_ops btree_extent_io_ops;
  60 static void end_workqueue_fn(struct btrfs_work *work);
  61
  62 struct end_io_wq {
  63         struct bio *bio;
  64         bio_end_io_t *end_io;
  65         void *private;
  66         struct btrfs_fs_info *info;
  67         int error;
  68         int metadata;
  69         struct list_head list;
  70         struct btrfs_work work;
  71 };
  72
  73 struct async_submit_bio {
  74         struct inode *inode;
  75         struct bio *bio;
  76         struct list_head list;
  77         extent_submit_bio_hook_t *submit_bio_hook;
  78         int rw;
  79         int mirror_num;
  80         struct btrfs_work work;
  81 };
  82
  83 struct extent_map *btree_get_extent(struct inode *inode, struct page *page,
  84                                     size_t page_offset, u64 start, u64 len,
  85                                     int create)
  86 {
  87         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
  88         struct extent_map *em;
  89         int ret;
  90
  91         spin_lock(&em_tree->lock);
  92         em = lookup_extent_mapping(em_tree, start, len);
  93         if (em) {
  94                 em->bdev =
  95                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
  96                 spin_unlock(&em_tree->lock);
  97                 goto out;
  98         }
  99         spin_unlock(&em_tree->lock);
 100
 101         em = alloc_extent_map(GFP_NOFS);
 102         if (!em) {
 103                 em = ERR_PTR(-ENOMEM);
 104                 goto out;
 105         }
 106         em->start = 0;
 107         em->len = (u64)-1;
 108         em->block_start = 0;
 109         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 110
 111         spin_lock(&em_tree->lock);
 112         ret = add_extent_mapping(em_tree, em);
 113         if (ret == -EEXIST) {
 114                 u64 failed_start = em->start;
 115                 u64 failed_len = em->len;
 116
 117                 printk("failed to insert %Lu %Lu -> %Lu into tree\n",
 118                        em->start, em->len, em->block_start);
 119                 free_extent_map(em);
 120                 em = lookup_extent_mapping(em_tree, start, len);
 121                 if (em) {
 122                         printk("after failing, found %Lu %Lu %Lu\n",
 123                                em->start, em->len, em->block_start);
 124                         ret = 0;
 125                 } else {
 126                         em = lookup_extent_mapping(em_tree, failed_start,
 127                                                    failed_len);
 128                         if (em) {
 129                                 printk("double failure lookup gives us "
 130                                        "%Lu %Lu -> %Lu\n", em->start,
 131                                        em->len, em->block_start);
 132                                 free_extent_map(em);
 133                         }
 134                         ret = -EIO;
 135                 }
 136         } else if (ret) {
 137                 free_extent_map(em);
 138                 em = NULL;
 139         }
 140         spin_unlock(&em_tree->lock);
 141
 142         if (ret)
 143                 em = ERR_PTR(ret);
 144 out:
 145         return em;
 146 }
 147
 148 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
 149 {
 150         return btrfs_crc32c(seed, data, len);
 151 }
 152
 153 void btrfs_csum_final(u32 crc, char *result)
 154 {
 155         *(__le32 *)result = ~cpu_to_le32(crc);
 156 }
 157
 158 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 159                            int verify)
 160 {
 161         char result[BTRFS_CRC32_SIZE];
 162         unsigned long len;
 163         unsigned long cur_len;
 164         unsigned long offset = BTRFS_CSUM_SIZE;
 165         char *map_token = NULL;
 166         char *kaddr;
 167         unsigned long map_start;
 168         unsigned long map_len;
 169         int err;
 170         u32 crc = ~(u32)0;
 171
 172         len = buf->len - offset;
 173         while(len > 0) {
 174                 err = map_private_extent_buffer(buf, offset, 32,
 175                                         &map_token, &kaddr,
 176                                         &map_start, &map_len, KM_USER0);
 177                 if (err) {
 178                         printk("failed to map extent buffer! %lu\n",
 179                                offset);
 180                         return 1;
 181                 }
 182                 cur_len = min(len, map_len - (offset - map_start));
 183                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
 184                                       crc, cur_len);
 185                 len -= cur_len;
 186                 offset += cur_len;
 187                 unmap_extent_buffer(buf, map_token, KM_USER0);
 188         }
 189         btrfs_csum_final(crc, result);
 190
 191         if (verify) {
 192                 /* FIXME, this is not good */
 193                 if (memcmp_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE)) {
 194                         u32 val;
 195                         u32 found = 0;
 196                         memcpy(&found, result, BTRFS_CRC32_SIZE);
 197
 198                         read_extent_buffer(buf, &val, 0, BTRFS_CRC32_SIZE);
 199                         printk("btrfs: %s checksum verify failed on %llu "
 200                                "wanted %X found %X level %d\n",
 201                                root->fs_info->sb->s_id,
 202                                buf->start, val, found, btrfs_header_level(buf));
 203                         return 1;
 204                 }
 205         } else {
 206                 write_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE);
 207         }
 208         return 0;
 209 }
 210
 211 static int verify_parent_transid(struct extent_io_tree *io_tree,
 212                                  struct extent_buffer *eb, u64 parent_transid)
 213 {
 214         int ret;
 215
 216         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 217                 return 0;
 218
 219         lock_extent(io_tree, eb->start, eb->start + eb->len - 1, GFP_NOFS);
 220         if (extent_buffer_uptodate(io_tree, eb) &&
 221             btrfs_header_generation(eb) == parent_transid) {
 222                 ret = 0;
 223                 goto out;
 224         }
 225         printk("parent transid verify failed on %llu wanted %llu found %llu\n",
 226                (unsigned long long)eb->start,
 227                (unsigned long long)parent_transid,
 228                (unsigned long long)btrfs_header_generation(eb));
 229         ret = 1;
 230         clear_extent_buffer_uptodate(io_tree, eb);
 231 out:
 232         unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
 233                       GFP_NOFS);
 234         return ret;
 235
 236 }
 237
 238 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 239                                           struct extent_buffer *eb,
 240                                           u64 start, u64 parent_transid)
 241 {
 242         struct extent_io_tree *io_tree;
 243         int ret;
 244         int num_copies = 0;
 245         int mirror_num = 0;
 246
 247         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 248         while (1) {
 249                 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
 250                                                btree_get_extent, mirror_num);
 251                 if (!ret &&
 252                     !verify_parent_transid(io_tree, eb, parent_transid))
 253                         return ret;
 254 printk("read extent buffer pages failed with ret %d mirror no %d\n", ret, mirror_num);
 255                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
 256                                               eb->start, eb->len);
 257                 if (num_copies == 1)
 258                         return ret;
 259
 260                 mirror_num++;
 261                 if (mirror_num > num_copies)
 262                         return ret;
 263         }
 264         return -EIO;
 265 }
 266
 267 int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 268 {
 269         struct extent_io_tree *tree;
 270         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
 271         u64 found_start;
 272         int found_level;
 273         unsigned long len;
 274         struct extent_buffer *eb;
 275         int ret;
 276
 277         tree = &BTRFS_I(page->mapping->host)->io_tree;
 278
 279         if (page->private == EXTENT_PAGE_PRIVATE)
 280                 goto out;
 281         if (!page->private)
 282                 goto out;
 283         len = page->private >> 2;
 284         if (len == 0) {
 285                 WARN_ON(1);
 286         }
 287         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 288         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
 289                                              btrfs_header_generation(eb));
 290         BUG_ON(ret);
 291         found_start = btrfs_header_bytenr(eb);
 292         if (found_start != start) {
 293                 printk("warning: eb start incorrect %Lu buffer %Lu len %lu\n",
 294                        start, found_start, len);
 295                 WARN_ON(1);
 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                 goto err;
 303         }
 304         if (!PageUptodate(page)) {
 305                 printk("csum not up to date page %lu\n", page->index);
 306                 WARN_ON(1);
 307                 goto err;
 308         }
 309         found_level = btrfs_header_level(eb);
 310
 311         csum_tree_block(root, eb, 0);
 312 err:
 313         free_extent_buffer(eb);
 314 out:
 315         return 0;
 316 }
 317
 318 static int btree_writepage_io_hook(struct page *page, u64 start, u64 end)
 319 {
 320         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 321
 322         csum_dirty_buffer(root, page);
 323         return 0;
 324 }
 325
 326 int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 327                                struct extent_state *state)
 328 {
 329         struct extent_io_tree *tree;
 330         u64 found_start;
 331         int found_level;
 332         unsigned long len;
 333         struct extent_buffer *eb;
 334         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 335         int ret = 0;
 336
 337         tree = &BTRFS_I(page->mapping->host)->io_tree;
 338         if (page->private == EXTENT_PAGE_PRIVATE)
 339                 goto out;
 340         if (!page->private)
 341                 goto out;
 342         len = page->private >> 2;
 343         if (len == 0) {
 344                 WARN_ON(1);
 345         }
 346         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 347
 348         found_start = btrfs_header_bytenr(eb);
 349         if (found_start != start) {
 350                 printk("bad tree block start %llu %llu\n",
 351                        (unsigned long long)found_start,
 352                        (unsigned long long)eb->start);
 353                 ret = -EIO;
 354                 goto err;
 355         }
 356         if (eb->first_page != page) {
 357                 printk("bad first page %lu %lu\n", eb->first_page->index,
 358                        page->index);
 359                 WARN_ON(1);
 360                 ret = -EIO;
 361                 goto err;
 362         }
 363         if (memcmp_extent_buffer(eb, root->fs_info->fsid,
 364                                  (unsigned long)btrfs_header_fsid(eb),
 365                                  BTRFS_FSID_SIZE)) {
 366                 printk("bad fsid on block %Lu\n", eb->start);
 367                 ret = -EIO;
 368                 goto err;
 369         }
 370         found_level = btrfs_header_level(eb);
 371
 372         ret = csum_tree_block(root, eb, 1);
 373         if (ret)
 374                 ret = -EIO;
 375
 376         end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
 377         end = eb->start + end - 1;
 378 err:
 379         free_extent_buffer(eb);
 380 out:
 381         return ret;
 382 }
 383
 384 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
 385 static void end_workqueue_bio(struct bio *bio, int err)
 386 #else
 387 static int end_workqueue_bio(struct bio *bio,
 388                                    unsigned int bytes_done, int err)
 389 #endif
 390 {
 391         struct end_io_wq *end_io_wq = bio->bi_private;
 392         struct btrfs_fs_info *fs_info;
 393
 394 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
 395         if (bio->bi_size)
 396                 return 1;
 397 #endif
 398
 399         fs_info = end_io_wq->info;
 400         end_io_wq->error = err;
 401         end_io_wq->work.func = end_workqueue_fn;
 402         end_io_wq->work.flags = 0;
 403         if (bio->bi_rw & (1 << BIO_RW))
 404                 btrfs_queue_worker(&fs_info->endio_write_workers,
 405                                    &end_io_wq->work);
 406         else
 407                 btrfs_queue_worker(&fs_info->endio_workers, &end_io_wq->work);
 408
 409 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
 410         return 0;
 411 #endif
 412 }
 413
 414 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 415                         int metadata)
 416 {
 417         struct end_io_wq *end_io_wq;
 418         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
 419         if (!end_io_wq)
 420                 return -ENOMEM;
 421
 422         end_io_wq->private = bio->bi_private;
 423         end_io_wq->end_io = bio->bi_end_io;
 424         end_io_wq->info = info;
 425         end_io_wq->error = 0;
 426         end_io_wq->bio = bio;
 427         end_io_wq->metadata = metadata;
 428
 429         bio->bi_private = end_io_wq;
 430         bio->bi_end_io = end_workqueue_bio;
 431         return 0;
 432 }
 433
 434 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
 435 {
 436         unsigned long limit = min_t(unsigned long,
 437                                     info->workers.max_workers,
 438                                     info->fs_devices->open_devices);
 439         return 256 * limit;
 440 }
 441
 442 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
 443 {
 444         return atomic_read(&info->nr_async_bios) >
 445                 btrfs_async_submit_limit(info);
 446 }
 447
 448 static void run_one_async_submit(struct btrfs_work *work)
 449 {
 450         struct btrfs_fs_info *fs_info;
 451         struct async_submit_bio *async;
 452         int limit;
 453
 454         async = container_of(work, struct  async_submit_bio, work);
 455         fs_info = BTRFS_I(async->inode)->root->fs_info;
 456
 457         limit = btrfs_async_submit_limit(fs_info);
 458         limit = limit * 2 / 3;
 459
 460         atomic_dec(&fs_info->nr_async_submits);
 461
 462         if (atomic_read(&fs_info->nr_async_submits) < limit &&
 463             waitqueue_active(&fs_info->async_submit_wait))
 464                 wake_up(&fs_info->async_submit_wait);
 465
 466         async->submit_bio_hook(async->inode, async->rw, async->bio,
 467                                async->mirror_num);
 468         kfree(async);
 469 }
 470
 471 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 472                         int rw, struct bio *bio, int mirror_num,
 473                         extent_submit_bio_hook_t *submit_bio_hook)
 474 {
 475         struct async_submit_bio *async;
 476         int limit = btrfs_async_submit_limit(fs_info);
 477
 478         async = kmalloc(sizeof(*async), GFP_NOFS);
 479         if (!async)
 480                 return -ENOMEM;
 481
 482         async->inode = inode;
 483         async->rw = rw;
 484         async->bio = bio;
 485         async->mirror_num = mirror_num;
 486         async->submit_bio_hook = submit_bio_hook;
 487         async->work.func = run_one_async_submit;
 488         async->work.flags = 0;
 489         atomic_inc(&fs_info->nr_async_submits);
 490         btrfs_queue_worker(&fs_info->workers, &async->work);
 491
 492         if (atomic_read(&fs_info->nr_async_submits) > limit) {
 493                 wait_event_timeout(fs_info->async_submit_wait,
 494                            (atomic_read(&fs_info->nr_async_submits) < limit),
 495                            HZ/10);
 496
 497                 wait_event_timeout(fs_info->async_submit_wait,
 498                            (atomic_read(&fs_info->nr_async_bios) < limit),
 499                            HZ/10);
 500         }
 501         return 0;
 502 }
 503
 504 static int __btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 505                                  int mirror_num)
 506 {
 507         struct btrfs_root *root = BTRFS_I(inode)->root;
 508         u64 offset;
 509         int ret;
 510
 511         offset = bio->bi_sector << 9;
 512
 513         /*
 514          * when we're called for a write, we're already in the async
 515          * submission context.  Just jump into btrfs_map_bio
 516          */
 517         if (rw & (1 << BIO_RW)) {
 518                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 519                                      mirror_num, 1);
 520         }
 521
 522         /*
 523          * called for a read, do the setup so that checksum validation
 524          * can happen in the async kernel threads
 525          */
 526         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 1);
 527         BUG_ON(ret);
 528
 529         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
 530 }
 531
 532 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 533                                  int mirror_num)
 534 {
 535         /*
 536          * kthread helpers are used to submit writes so that checksumming
 537          * can happen in parallel across all CPUs
 538          */
 539         if (!(rw & (1 << BIO_RW))) {
 540                 return __btree_submit_bio_hook(inode, rw, bio, mirror_num);
 541         }
 542         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 543                                    inode, rw, bio, mirror_num,
 544                                    __btree_submit_bio_hook);
 545 }
 546
 547 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 548 {
 549         struct extent_io_tree *tree;
 550         tree = &BTRFS_I(page->mapping->host)->io_tree;
 551
 552         if (current->flags & PF_MEMALLOC) {
 553                 redirty_page_for_writepage(wbc, page);
 554                 unlock_page(page);
 555                 return 0;
 556         }
 557         return extent_write_full_page(tree, page, btree_get_extent, wbc);
 558 }
 559
 560 static int btree_writepages(struct address_space *mapping,
 561                             struct writeback_control *wbc)
 562 {
 563         struct extent_io_tree *tree;
 564         tree = &BTRFS_I(mapping->host)->io_tree;
 565         if (wbc->sync_mode == WB_SYNC_NONE) {
 566                 u64 num_dirty;
 567                 u64 start = 0;
 568                 unsigned long thresh = 8 * 1024 * 1024;
 569
 570                 if (wbc->for_kupdate)
 571                         return 0;
 572
 573                 num_dirty = count_range_bits(tree, &start, (u64)-1,
 574                                              thresh, EXTENT_DIRTY);
 575                 if (num_dirty < thresh) {
 576                         return 0;
 577                 }
 578         }
 579         return extent_writepages(tree, mapping, btree_get_extent, wbc);
 580 }
 581
 582 int btree_readpage(struct file *file, struct page *page)
 583 {
 584         struct extent_io_tree *tree;
 585         tree = &BTRFS_I(page->mapping->host)->io_tree;
 586         return extent_read_full_page(tree, page, btree_get_extent);
 587 }
 588
 589 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 590 {
 591         struct extent_io_tree *tree;
 592         struct extent_map_tree *map;
 593         int ret;
 594
 595         tree = &BTRFS_I(page->mapping->host)->io_tree;
 596         map = &BTRFS_I(page->mapping->host)->extent_tree;
 597
 598         ret = try_release_extent_state(map, tree, page, gfp_flags);
 599         if (!ret) {
 600                 return 0;
 601         }
 602
 603         ret = try_release_extent_buffer(tree, page);
 604         if (ret == 1) {
 605                 ClearPagePrivate(page);
 606                 set_page_private(page, 0);
 607                 page_cache_release(page);
 608         }
 609
 610         return ret;
 611 }
 612
 613 static void btree_invalidatepage(struct page *page, unsigned long offset)
 614 {
 615         struct extent_io_tree *tree;
 616         tree = &BTRFS_I(page->mapping->host)->io_tree;
 617         extent_invalidatepage(tree, page, offset);
 618         btree_releasepage(page, GFP_NOFS);
 619         if (PagePrivate(page)) {
 620                 printk("warning page private not zero on page %Lu\n",
 621                        page_offset(page));
 622                 ClearPagePrivate(page);
 623                 set_page_private(page, 0);
 624                 page_cache_release(page);
 625         }
 626 }
 627
 628 #if 0
 629 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 630 {
 631         struct buffer_head *bh;
 632         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 633         struct buffer_head *head;
 634         if (!page_has_buffers(page)) {
 635                 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
 636                                         (1 << BH_Dirty)|(1 << BH_Uptodate));
 637         }
 638         head = page_buffers(page);
 639         bh = head;
 640         do {
 641                 if (buffer_dirty(bh))
 642                         csum_tree_block(root, bh, 0);
 643                 bh = bh->b_this_page;
 644         } while (bh != head);
 645         return block_write_full_page(page, btree_get_block, wbc);
 646 }
 647 #endif
 648
 649 static struct address_space_operations btree_aops = {
 650         .readpage       = btree_readpage,
 651         .writepage      = btree_writepage,
 652         .writepages     = btree_writepages,
 653         .releasepage    = btree_releasepage,
 654         .invalidatepage = btree_invalidatepage,
 655         .sync_page      = block_sync_page,
 656 };
 657
 658 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
 659                          u64 parent_transid)
 660 {
 661         struct extent_buffer *buf = NULL;
 662         struct inode *btree_inode = root->fs_info->btree_inode;
 663         int ret = 0;
 664
 665         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 666         if (!buf)
 667                 return 0;
 668         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
 669                                  buf, 0, 0, btree_get_extent, 0);
 670         free_extent_buffer(buf);
 671         return ret;
 672 }
 673
 674 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
 675                                             u64 bytenr, u32 blocksize)
 676 {
 677         struct inode *btree_inode = root->fs_info->btree_inode;
 678         struct extent_buffer *eb;
 679         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 680                                 bytenr, blocksize, GFP_NOFS);
 681         return eb;
 682 }
 683
 684 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
 685                                                  u64 bytenr, u32 blocksize)
 686 {
 687         struct inode *btree_inode = root->fs_info->btree_inode;
 688         struct extent_buffer *eb;
 689
 690         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 691                                  bytenr, blocksize, NULL, GFP_NOFS);
 692         return eb;
 693 }
 694
 695
 696 int btrfs_write_tree_block(struct extent_buffer *buf)
 697 {
 698         return btrfs_fdatawrite_range(buf->first_page->mapping, buf->start,
 699                                       buf->start + buf->len - 1, WB_SYNC_NONE);
 700 }
 701
 702 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
 703 {
 704         return btrfs_wait_on_page_writeback_range(buf->first_page->mapping,
 705                                   buf->start, buf->start + buf->len -1);
 706 }
 707
 708 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
 709                                       u32 blocksize, u64 parent_transid)
 710 {
 711         struct extent_buffer *buf = NULL;
 712         struct inode *btree_inode = root->fs_info->btree_inode;
 713         struct extent_io_tree *io_tree;
 714         int ret;
 715
 716         io_tree = &BTRFS_I(btree_inode)->io_tree;
 717
 718         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 719         if (!buf)
 720                 return NULL;
 721
 722         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
 723
 724         if (ret == 0) {
 725                 buf->flags |= EXTENT_UPTODATE;
 726         } else {
 727                 WARN_ON(1);
 728         }
 729         return buf;
 730
 731 }
 732
 733 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 734                      struct extent_buffer *buf)
 735 {
 736         struct inode *btree_inode = root->fs_info->btree_inode;
 737         if (btrfs_header_generation(buf) ==
 738             root->fs_info->running_transaction->transid) {
 739                 WARN_ON(!btrfs_tree_locked(buf));
 740                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
 741                                           buf);
 742         }
 743         return 0;
 744 }
 745
 746 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
 747                         u32 stripesize, struct btrfs_root *root,
 748                         struct btrfs_fs_info *fs_info,
 749                         u64 objectid)
 750 {
 751         root->node = NULL;
 752         root->inode = NULL;
 753         root->commit_root = NULL;
 754         root->ref_tree = NULL;
 755         root->sectorsize = sectorsize;
 756         root->nodesize = nodesize;
 757         root->leafsize = leafsize;
 758         root->stripesize = stripesize;
 759         root->ref_cows = 0;
 760         root->track_dirty = 0;
 761
 762         root->fs_info = fs_info;
 763         root->objectid = objectid;
 764         root->last_trans = 0;
 765         root->highest_inode = 0;
 766         root->last_inode_alloc = 0;
 767         root->name = NULL;
 768         root->in_sysfs = 0;
 769
 770         INIT_LIST_HEAD(&root->dirty_list);
 771         INIT_LIST_HEAD(&root->orphan_list);
 772         INIT_LIST_HEAD(&root->dead_list);
 773         spin_lock_init(&root->node_lock);
 774         spin_lock_init(&root->list_lock);
 775         mutex_init(&root->objectid_mutex);
 776         mutex_init(&root->log_mutex);
 777
 778         btrfs_leaf_ref_tree_init(&root->ref_tree_struct);
 779         root->ref_tree = &root->ref_tree_struct;
 780
 781         memset(&root->root_key, 0, sizeof(root->root_key));
 782         memset(&root->root_item, 0, sizeof(root->root_item));
 783         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
 784         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
 785         root->defrag_trans_start = fs_info->generation;
 786         init_completion(&root->kobj_unregister);
 787         root->defrag_running = 0;
 788         root->defrag_level = 0;
 789         root->root_key.objectid = objectid;
 790         return 0;
 791 }
 792
 793 static int find_and_setup_root(struct btrfs_root *tree_root,
 794                                struct btrfs_fs_info *fs_info,
 795                                u64 objectid,
 796                                struct btrfs_root *root)
 797 {
 798         int ret;
 799         u32 blocksize;
 800
 801         __setup_root(tree_root->nodesize, tree_root->leafsize,
 802                      tree_root->sectorsize, tree_root->stripesize,
 803                      root, fs_info, objectid);
 804         ret = btrfs_find_last_root(tree_root, objectid,
 805                                    &root->root_item, &root->root_key);
 806         BUG_ON(ret);
 807
 808         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 809         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 810                                      blocksize, 0);
 811         BUG_ON(!root->node);
 812         return 0;
 813 }
 814
 815 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
 816                              struct btrfs_fs_info *fs_info)
 817 {
 818         struct extent_buffer *eb;
 819         int ret;
 820
 821         if (!fs_info->log_root_tree)
 822                 return 0;
 823
 824         eb = fs_info->log_root_tree->node;
 825
 826         WARN_ON(btrfs_header_level(eb) != 0);
 827         WARN_ON(btrfs_header_nritems(eb) != 0);
 828
 829         ret = btrfs_free_extent(trans, fs_info->tree_root,
 830                                 eb->start, eb->len,
 831                                 BTRFS_TREE_LOG_OBJECTID, 0, 0, 0, 1);
 832         BUG_ON(ret);
 833
 834         free_extent_buffer(eb);
 835         kfree(fs_info->log_root_tree);
 836         fs_info->log_root_tree = NULL;
 837         return 0;
 838 }
 839
 840 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
 841                              struct btrfs_fs_info *fs_info)
 842 {
 843         struct btrfs_root *root;
 844         struct btrfs_root *tree_root = fs_info->tree_root;
 845
 846         root = kzalloc(sizeof(*root), GFP_NOFS);
 847         if (!root)
 848                 return -ENOMEM;
 849
 850         __setup_root(tree_root->nodesize, tree_root->leafsize,
 851                      tree_root->sectorsize, tree_root->stripesize,
 852                      root, fs_info, BTRFS_TREE_LOG_OBJECTID);
 853
 854         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
 855         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
 856         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
 857         root->ref_cows = 0;
 858
 859         root->node = btrfs_alloc_free_block(trans, root, root->leafsize,
 860                                             BTRFS_TREE_LOG_OBJECTID,
 861                                             0, 0, 0, 0, 0);
 862
 863         btrfs_set_header_nritems(root->node, 0);
 864         btrfs_set_header_level(root->node, 0);
 865         btrfs_set_header_bytenr(root->node, root->node->start);
 866         btrfs_set_header_generation(root->node, trans->transid);
 867         btrfs_set_header_owner(root->node, BTRFS_TREE_LOG_OBJECTID);
 868
 869         write_extent_buffer(root->node, root->fs_info->fsid,
 870                             (unsigned long)btrfs_header_fsid(root->node),
 871                             BTRFS_FSID_SIZE);
 872         btrfs_mark_buffer_dirty(root->node);
 873         btrfs_tree_unlock(root->node);
 874         fs_info->log_root_tree = root;
 875         return 0;
 876 }
 877
 878 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
 879                                                struct btrfs_key *location)
 880 {
 881         struct btrfs_root *root;
 882         struct btrfs_fs_info *fs_info = tree_root->fs_info;
 883         struct btrfs_path *path;
 884         struct extent_buffer *l;
 885         u64 highest_inode;
 886         u32 blocksize;
 887         int ret = 0;
 888
 889         root = kzalloc(sizeof(*root), GFP_NOFS);
 890         if (!root)
 891                 return ERR_PTR(-ENOMEM);
 892         if (location->offset == (u64)-1) {
 893                 ret = find_and_setup_root(tree_root, fs_info,
 894                                           location->objectid, root);
 895                 if (ret) {
 896                         kfree(root);
 897                         return ERR_PTR(ret);
 898                 }
 899                 goto insert;
 900         }
 901
 902         __setup_root(tree_root->nodesize, tree_root->leafsize,
 903                      tree_root->sectorsize, tree_root->stripesize,
 904                      root, fs_info, location->objectid);
 905
 906         path = btrfs_alloc_path();
 907         BUG_ON(!path);
 908         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
 909         if (ret != 0) {
 910                 if (ret > 0)
 911                         ret = -ENOENT;
 912                 goto out;
 913         }
 914         l = path->nodes[0];
 915         read_extent_buffer(l, &root->root_item,
 916                btrfs_item_ptr_offset(l, path->slots[0]),
 917                sizeof(root->root_item));
 918         memcpy(&root->root_key, location, sizeof(*location));
 919         ret = 0;
 920 out:
 921         btrfs_release_path(root, path);
 922         btrfs_free_path(path);
 923         if (ret) {
 924                 kfree(root);
 925                 return ERR_PTR(ret);
 926         }
 927         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 928         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 929                                      blocksize, 0);
 930         BUG_ON(!root->node);
 931 insert:
 932         if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
 933                 root->ref_cows = 1;
 934                 ret = btrfs_find_highest_inode(root, &highest_inode);
 935                 if (ret == 0) {
 936                         root->highest_inode = highest_inode;
 937                         root->last_inode_alloc = highest_inode;
 938                 }
 939         }
 940         return root;
 941 }
 942
 943 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
 944                                         u64 root_objectid)
 945 {
 946         struct btrfs_root *root;
 947
 948         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
 949                 return fs_info->tree_root;
 950         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
 951                 return fs_info->extent_root;
 952
 953         root = radix_tree_lookup(&fs_info->fs_roots_radix,
 954                                  (unsigned long)root_objectid);
 955         return root;
 956 }
 957
 958 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
 959                                               struct btrfs_key *location)
 960 {
 961         struct btrfs_root *root;
 962         int ret;
 963
 964         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
 965                 return fs_info->tree_root;
 966         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
 967                 return fs_info->extent_root;
 968         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
 969                 return fs_info->chunk_root;
 970         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
 971                 return fs_info->dev_root;
 972
 973         root = radix_tree_lookup(&fs_info->fs_roots_radix,
 974                                  (unsigned long)location->objectid);
 975         if (root)
 976                 return root;
 977
 978         root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
 979         if (IS_ERR(root))
 980                 return root;
 981         ret = radix_tree_insert(&fs_info->fs_roots_radix,
 982                                 (unsigned long)root->root_key.objectid,
 983                                 root);
 984         if (ret) {
 985                 free_extent_buffer(root->node);
 986                 kfree(root);
 987                 return ERR_PTR(ret);
 988         }
 989         ret = btrfs_find_dead_roots(fs_info->tree_root,
 990                                     root->root_key.objectid, root);
 991         BUG_ON(ret);
 992
 993         return root;
 994 }
 995
 996 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
 997                                       struct btrfs_key *location,
 998                                       const char *name, int namelen)
 999 {
1000         struct btrfs_root *root;
1001         int ret;
1002
1003         root = btrfs_read_fs_root_no_name(fs_info, location);
1004         if (!root)
1005                 return NULL;
1006
1007         if (root->in_sysfs)
1008                 return root;
1009
1010         ret = btrfs_set_root_name(root, name, namelen);
1011         if (ret) {
1012                 free_extent_buffer(root->node);
1013                 kfree(root);
1014                 return ERR_PTR(ret);
1015         }
1016
1017         ret = btrfs_sysfs_add_root(root);
1018         if (ret) {
1019                 free_extent_buffer(root->node);
1020                 kfree(root->name);
1021                 kfree(root);
1022                 return ERR_PTR(ret);
1023         }
1024         root->in_sysfs = 1;
1025         return root;
1026 }
1027 #if 0
1028 static int add_hasher(struct btrfs_fs_info *info, char *type) {
1029         struct btrfs_hasher *hasher;
1030
1031         hasher = kmalloc(sizeof(*hasher), GFP_NOFS);
1032         if (!hasher)
1033                 return -ENOMEM;
1034         hasher->hash_tfm = crypto_alloc_hash(type, 0, CRYPTO_ALG_ASYNC);
1035         if (!hasher->hash_tfm) {
1036                 kfree(hasher);
1037                 return -EINVAL;
1038         }
1039         spin_lock(&info->hash_lock);
1040         list_add(&hasher->list, &info->hashers);
1041         spin_unlock(&info->hash_lock);
1042         return 0;
1043 }
1044 #endif
1045
1046 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1047 {
1048         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1049         int ret = 0;
1050         struct list_head *cur;
1051         struct btrfs_device *device;
1052         struct backing_dev_info *bdi;
1053
1054         if ((bdi_bits & (1 << BDI_write_congested)) &&
1055             btrfs_congested_async(info, 0))
1056                 return 1;
1057
1058         list_for_each(cur, &info->fs_devices->devices) {
1059                 device = list_entry(cur, struct btrfs_device, dev_list);
1060                 if (!device->bdev)
1061                         continue;
1062                 bdi = blk_get_backing_dev_info(device->bdev);
1063                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1064                         ret = 1;
1065                         break;
1066                 }
1067         }
1068         return ret;
1069 }
1070
1071 /*
1072  * this unplugs every device on the box, and it is only used when page
1073  * is null
1074  */
1075 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1076 {
1077         struct list_head *cur;
1078         struct btrfs_device *device;
1079         struct btrfs_fs_info *info;
1080
1081         info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1082         list_for_each(cur, &info->fs_devices->devices) {
1083                 device = list_entry(cur, struct btrfs_device, dev_list);
1084                 bdi = blk_get_backing_dev_info(device->bdev);
1085                 if (bdi->unplug_io_fn) {
1086                         bdi->unplug_io_fn(bdi, page);
1087                 }
1088         }
1089 }
1090
1091 void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1092 {
1093         struct inode *inode;
1094         struct extent_map_tree *em_tree;
1095         struct extent_map *em;
1096         struct address_space *mapping;
1097         u64 offset;
1098
1099         /* the generic O_DIRECT read code does this */
1100         if (!page) {
1101                 __unplug_io_fn(bdi, page);
1102                 return;
1103         }
1104
1105         /*
1106          * page->mapping may change at any time.  Get a consistent copy
1107          * and use that for everything below
1108          */
1109         smp_mb();
1110         mapping = page->mapping;
1111         if (!mapping)
1112                 return;
1113
1114         inode = mapping->host;
1115         offset = page_offset(page);
1116
1117         em_tree = &BTRFS_I(inode)->extent_tree;
1118         spin_lock(&em_tree->lock);
1119         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1120         spin_unlock(&em_tree->lock);
1121         if (!em) {
1122                 __unplug_io_fn(bdi, page);
1123                 return;
1124         }
1125
1126         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1127                 free_extent_map(em);
1128                 __unplug_io_fn(bdi, page);
1129                 return;
1130         }
1131         offset = offset - em->start;
1132         btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1133                           em->block_start + offset, page);
1134         free_extent_map(em);
1135 }
1136
1137 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1138 {
1139 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1140         bdi_init(bdi);
1141 #endif
1142         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1143         bdi->state              = 0;
1144         bdi->capabilities       = default_backing_dev_info.capabilities;
1145         bdi->unplug_io_fn       = btrfs_unplug_io_fn;
1146         bdi->unplug_io_data     = info;
1147         bdi->congested_fn       = btrfs_congested_fn;
1148         bdi->congested_data     = info;
1149         return 0;
1150 }
1151
1152 static int bio_ready_for_csum(struct bio *bio)
1153 {
1154         u64 length = 0;
1155         u64 buf_len = 0;
1156         u64 start = 0;
1157         struct page *page;
1158         struct extent_io_tree *io_tree = NULL;
1159         struct btrfs_fs_info *info = NULL;
1160         struct bio_vec *bvec;
1161         int i;
1162         int ret;
1163
1164         bio_for_each_segment(bvec, bio, i) {
1165                 page = bvec->bv_page;
1166                 if (page->private == EXTENT_PAGE_PRIVATE) {
1167                         length += bvec->bv_len;
1168                         continue;
1169                 }
1170                 if (!page->private) {
1171                         length += bvec->bv_len;
1172                         continue;
1173                 }
1174                 length = bvec->bv_len;
1175                 buf_len = page->private >> 2;
1176                 start = page_offset(page) + bvec->bv_offset;
1177                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1178                 info = BTRFS_I(page->mapping->host)->root->fs_info;
1179         }
1180         /* are we fully contained in this bio? */
1181         if (buf_len <= length)
1182                 return 1;
1183
1184         ret = extent_range_uptodate(io_tree, start + length,
1185                                     start + buf_len - 1);
1186         if (ret == 1)
1187                 return ret;
1188         return ret;
1189 }
1190
1191 /*
1192  * called by the kthread helper functions to finally call the bio end_io
1193  * functions.  This is where read checksum verification actually happens
1194  */
1195 static void end_workqueue_fn(struct btrfs_work *work)
1196 {
1197         struct bio *bio;
1198         struct end_io_wq *end_io_wq;
1199         struct btrfs_fs_info *fs_info;
1200         int error;
1201
1202         end_io_wq = container_of(work, struct end_io_wq, work);
1203         bio = end_io_wq->bio;
1204         fs_info = end_io_wq->info;
1205
1206         /* metadata bios are special because the whole tree block must
1207          * be checksummed at once.  This makes sure the entire block is in
1208          * ram and up to date before trying to verify things.  For
1209          * blocksize <= pagesize, it is basically a noop
1210          */
1211         if (end_io_wq->metadata && !bio_ready_for_csum(bio)) {
1212                 btrfs_queue_worker(&fs_info->endio_workers,
1213                                    &end_io_wq->work);
1214                 return;
1215         }
1216         error = end_io_wq->error;
1217         bio->bi_private = end_io_wq->private;
1218         bio->bi_end_io = end_io_wq->end_io;
1219         kfree(end_io_wq);
1220 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1221         bio_endio(bio, bio->bi_size, error);
1222 #else
1223         bio_endio(bio, error);
1224 #endif
1225 }
1226
1227 static int cleaner_kthread(void *arg)
1228 {
1229         struct btrfs_root *root = arg;
1230
1231         do {
1232                 smp_mb();
1233                 if (root->fs_info->closing)
1234                         break;
1235
1236                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1237                 mutex_lock(&root->fs_info->cleaner_mutex);
1238                 btrfs_clean_old_snapshots(root);
1239                 mutex_unlock(&root->fs_info->cleaner_mutex);
1240
1241                 if (freezing(current)) {
1242                         refrigerator();
1243                 } else {
1244                         smp_mb();
1245                         if (root->fs_info->closing)
1246                                 break;
1247                         set_current_state(TASK_INTERRUPTIBLE);
1248                         schedule();
1249                         __set_current_state(TASK_RUNNING);
1250                 }
1251         } while (!kthread_should_stop());
1252         return 0;
1253 }
1254
1255 static int transaction_kthread(void *arg)
1256 {
1257         struct btrfs_root *root = arg;
1258         struct btrfs_trans_handle *trans;
1259         struct btrfs_transaction *cur;
1260         unsigned long now;
1261         unsigned long delay;
1262         int ret;
1263
1264         do {
1265                 smp_mb();
1266                 if (root->fs_info->closing)
1267                         break;
1268
1269                 delay = HZ * 30;
1270                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1271                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1272
1273                 if (root->fs_info->total_ref_cache_size > 20 * 1024 * 1024) {
1274                         printk("btrfs: total reference cache size %Lu\n",
1275                                 root->fs_info->total_ref_cache_size);
1276                 }
1277
1278                 mutex_lock(&root->fs_info->trans_mutex);
1279                 cur = root->fs_info->running_transaction;
1280                 if (!cur) {
1281                         mutex_unlock(&root->fs_info->trans_mutex);
1282                         goto sleep;
1283                 }
1284
1285                 now = get_seconds();
1286                 if (now < cur->start_time || now - cur->start_time < 30) {
1287                         mutex_unlock(&root->fs_info->trans_mutex);
1288                         delay = HZ * 5;
1289                         goto sleep;
1290                 }
1291                 mutex_unlock(&root->fs_info->trans_mutex);
1292                 trans = btrfs_start_transaction(root, 1);
1293                 ret = btrfs_commit_transaction(trans, root);
1294 sleep:
1295                 wake_up_process(root->fs_info->cleaner_kthread);
1296                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1297
1298                 if (freezing(current)) {
1299                         refrigerator();
1300                 } else {
1301                         if (root->fs_info->closing)
1302                                 break;
1303                         set_current_state(TASK_INTERRUPTIBLE);
1304                         schedule_timeout(delay);
1305                         __set_current_state(TASK_RUNNING);
1306                 }
1307         } while (!kthread_should_stop());
1308         return 0;
1309 }
1310
1311 struct btrfs_root *open_ctree(struct super_block *sb,
1312                               struct btrfs_fs_devices *fs_devices,
1313                               char *options)
1314 {
1315         u32 sectorsize;
1316         u32 nodesize;
1317         u32 leafsize;
1318         u32 blocksize;
1319         u32 stripesize;
1320         struct buffer_head *bh;
1321         struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1322                                                  GFP_NOFS);
1323         struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
1324                                                GFP_NOFS);
1325         struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1326                                                 GFP_NOFS);
1327         struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1328                                                 GFP_NOFS);
1329         struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1330                                               GFP_NOFS);
1331         struct btrfs_root *log_tree_root;
1332
1333         int ret;
1334         int err = -EINVAL;
1335
1336         struct btrfs_super_block *disk_super;
1337
1338         if (!extent_root || !tree_root || !fs_info) {
1339                 err = -ENOMEM;
1340                 goto fail;
1341         }
1342         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1343         INIT_LIST_HEAD(&fs_info->trans_list);
1344         INIT_LIST_HEAD(&fs_info->dead_roots);
1345         INIT_LIST_HEAD(&fs_info->hashers);
1346         INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1347         spin_lock_init(&fs_info->hash_lock);
1348         spin_lock_init(&fs_info->delalloc_lock);
1349         spin_lock_init(&fs_info->new_trans_lock);
1350         spin_lock_init(&fs_info->ref_cache_lock);
1351
1352         init_completion(&fs_info->kobj_unregister);
1353         fs_info->tree_root = tree_root;
1354         fs_info->extent_root = extent_root;
1355         fs_info->chunk_root = chunk_root;
1356         fs_info->dev_root = dev_root;
1357         fs_info->fs_devices = fs_devices;
1358         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1359         INIT_LIST_HEAD(&fs_info->space_info);
1360         btrfs_mapping_init(&fs_info->mapping_tree);
1361         atomic_set(&fs_info->nr_async_submits, 0);
1362         atomic_set(&fs_info->nr_async_bios, 0);
1363         atomic_set(&fs_info->throttles, 0);
1364         atomic_set(&fs_info->throttle_gen, 0);
1365         fs_info->sb = sb;
1366         fs_info->max_extent = (u64)-1;
1367         fs_info->max_inline = 8192 * 1024;
1368         setup_bdi(fs_info, &fs_info->bdi);
1369         fs_info->btree_inode = new_inode(sb);
1370         fs_info->btree_inode->i_ino = 1;
1371         fs_info->btree_inode->i_nlink = 1;
1372         fs_info->thread_pool_size = min(num_online_cpus() + 2, 8);
1373
1374         INIT_LIST_HEAD(&fs_info->ordered_extents);
1375         spin_lock_init(&fs_info->ordered_extent_lock);
1376
1377         sb->s_blocksize = 4096;
1378         sb->s_blocksize_bits = blksize_bits(4096);
1379
1380         /*
1381          * we set the i_size on the btree inode to the max possible int.
1382          * the real end of the address space is determined by all of
1383          * the devices in the system
1384          */
1385         fs_info->btree_inode->i_size = OFFSET_MAX;
1386         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1387         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1388
1389         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1390                              fs_info->btree_inode->i_mapping,
1391                              GFP_NOFS);
1392         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1393                              GFP_NOFS);
1394
1395         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1396
1397         extent_io_tree_init(&fs_info->free_space_cache,
1398                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1399         extent_io_tree_init(&fs_info->block_group_cache,
1400                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1401         extent_io_tree_init(&fs_info->pinned_extents,
1402                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1403         extent_io_tree_init(&fs_info->pending_del,
1404                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1405         extent_io_tree_init(&fs_info->extent_ins,
1406                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1407         fs_info->do_barriers = 1;
1408
1409         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1410         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1411                sizeof(struct btrfs_key));
1412         insert_inode_hash(fs_info->btree_inode);
1413         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1414
1415         mutex_init(&fs_info->trans_mutex);
1416         mutex_init(&fs_info->tree_log_mutex);
1417         mutex_init(&fs_info->drop_mutex);
1418         mutex_init(&fs_info->alloc_mutex);
1419         mutex_init(&fs_info->chunk_mutex);
1420         mutex_init(&fs_info->transaction_kthread_mutex);
1421         mutex_init(&fs_info->cleaner_mutex);
1422         mutex_init(&fs_info->volume_mutex);
1423         init_waitqueue_head(&fs_info->transaction_throttle);
1424         init_waitqueue_head(&fs_info->transaction_wait);
1425         init_waitqueue_head(&fs_info->async_submit_wait);
1426         init_waitqueue_head(&fs_info->tree_log_wait);
1427         atomic_set(&fs_info->tree_log_commit, 0);
1428         atomic_set(&fs_info->tree_log_writers, 0);
1429         fs_info->tree_log_transid = 0;
1430
1431 #if 0
1432         ret = add_hasher(fs_info, "crc32c");
1433         if (ret) {
1434                 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
1435                 err = -ENOMEM;
1436                 goto fail_iput;
1437         }
1438 #endif
1439         __setup_root(4096, 4096, 4096, 4096, tree_root,
1440                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1441
1442
1443         bh = __bread(fs_devices->latest_bdev,
1444                      BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
1445         if (!bh)
1446                 goto fail_iput;
1447
1448         memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1449         brelse(bh);
1450
1451         memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1452
1453         disk_super = &fs_info->super_copy;
1454         if (!btrfs_super_root(disk_super))
1455                 goto fail_sb_buffer;
1456
1457         err = btrfs_parse_options(tree_root, options);
1458         if (err)
1459                 goto fail_sb_buffer;
1460
1461         /*
1462          * we need to start all the end_io workers up front because the
1463          * queue work function gets called at interrupt time, and so it
1464          * cannot dynamically grow.
1465          */
1466         btrfs_init_workers(&fs_info->workers, "worker",
1467                            fs_info->thread_pool_size);
1468         btrfs_init_workers(&fs_info->submit_workers, "submit",
1469                            min_t(u64, fs_devices->num_devices,
1470                            fs_info->thread_pool_size));
1471
1472         /* a higher idle thresh on the submit workers makes it much more
1473          * likely that bios will be send down in a sane order to the
1474          * devices
1475          */
1476         fs_info->submit_workers.idle_thresh = 64;
1477
1478         /* fs_info->workers is responsible for checksumming file data
1479          * blocks and metadata.  Using a larger idle thresh allows each
1480          * worker thread to operate on things in roughly the order they
1481          * were sent by the writeback daemons, improving overall locality
1482          * of the IO going down the pipe.
1483          */
1484         fs_info->workers.idle_thresh = 128;
1485
1486         btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1);
1487         btrfs_init_workers(&fs_info->endio_workers, "endio",
1488                            fs_info->thread_pool_size);
1489         btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1490                            fs_info->thread_pool_size);
1491
1492         /*
1493          * endios are largely parallel and should have a very
1494          * low idle thresh
1495          */
1496         fs_info->endio_workers.idle_thresh = 4;
1497         fs_info->endio_write_workers.idle_thresh = 4;
1498
1499         btrfs_start_workers(&fs_info->workers, 1);
1500         btrfs_start_workers(&fs_info->submit_workers, 1);
1501         btrfs_start_workers(&fs_info->fixup_workers, 1);
1502         btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
1503         btrfs_start_workers(&fs_info->endio_write_workers,
1504                             fs_info->thread_pool_size);
1505
1506         err = -EINVAL;
1507         if (btrfs_super_num_devices(disk_super) > fs_devices->open_devices) {
1508                 printk("Btrfs: wanted %llu devices, but found %llu\n",
1509                        (unsigned long long)btrfs_super_num_devices(disk_super),
1510                        (unsigned long long)fs_devices->open_devices);
1511                 if (btrfs_test_opt(tree_root, DEGRADED))
1512                         printk("continuing in degraded mode\n");
1513                 else {
1514                         goto fail_sb_buffer;
1515                 }
1516         }
1517
1518         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1519
1520         nodesize = btrfs_super_nodesize(disk_super);
1521         leafsize = btrfs_super_leafsize(disk_super);
1522         sectorsize = btrfs_super_sectorsize(disk_super);
1523         stripesize = btrfs_super_stripesize(disk_super);
1524         tree_root->nodesize = nodesize;
1525         tree_root->leafsize = leafsize;
1526         tree_root->sectorsize = sectorsize;
1527         tree_root->stripesize = stripesize;
1528
1529         sb->s_blocksize = sectorsize;
1530         sb->s_blocksize_bits = blksize_bits(sectorsize);
1531
1532         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1533                     sizeof(disk_super->magic))) {
1534                 printk("btrfs: valid FS not found on %s\n", sb->s_id);
1535                 goto fail_sb_buffer;
1536         }
1537
1538         mutex_lock(&fs_info->chunk_mutex);
1539         ret = btrfs_read_sys_array(tree_root);
1540         mutex_unlock(&fs_info->chunk_mutex);
1541         if (ret) {
1542                 printk("btrfs: failed to read the system array on %s\n",
1543                        sb->s_id);
1544                 goto fail_sys_array;
1545         }
1546
1547         blocksize = btrfs_level_size(tree_root,
1548                                      btrfs_super_chunk_root_level(disk_super));
1549
1550         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1551                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1552
1553         chunk_root->node = read_tree_block(chunk_root,
1554                                            btrfs_super_chunk_root(disk_super),
1555                                            blocksize, 0);
1556         BUG_ON(!chunk_root->node);
1557
1558         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1559                  (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1560                  BTRFS_UUID_SIZE);
1561
1562         mutex_lock(&fs_info->chunk_mutex);
1563         ret = btrfs_read_chunk_tree(chunk_root);
1564         mutex_unlock(&fs_info->chunk_mutex);
1565         BUG_ON(ret);
1566
1567         btrfs_close_extra_devices(fs_devices);
1568
1569         blocksize = btrfs_level_size(tree_root,
1570                                      btrfs_super_root_level(disk_super));
1571
1572
1573         tree_root->node = read_tree_block(tree_root,
1574                                           btrfs_super_root(disk_super),
1575                                           blocksize, 0);
1576         if (!tree_root->node)
1577                 goto fail_sb_buffer;
1578
1579
1580         ret = find_and_setup_root(tree_root, fs_info,
1581                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1582         if (ret)
1583                 goto fail_tree_root;
1584         extent_root->track_dirty = 1;
1585
1586         ret = find_and_setup_root(tree_root, fs_info,
1587                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
1588         dev_root->track_dirty = 1;
1589
1590         if (ret)
1591                 goto fail_extent_root;
1592
1593         btrfs_read_block_groups(extent_root);
1594
1595         fs_info->generation = btrfs_super_generation(disk_super) + 1;
1596         fs_info->data_alloc_profile = (u64)-1;
1597         fs_info->metadata_alloc_profile = (u64)-1;
1598         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1599         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1600                                                "btrfs-cleaner");
1601         if (!fs_info->cleaner_kthread)
1602                 goto fail_extent_root;
1603
1604         fs_info->transaction_kthread = kthread_run(transaction_kthread,
1605                                                    tree_root,
1606                                                    "btrfs-transaction");
1607         if (!fs_info->transaction_kthread)
1608                 goto fail_cleaner;
1609
1610         if (btrfs_super_log_root(disk_super) != 0) {
1611                 u32 blocksize;
1612                 u64 bytenr = btrfs_super_log_root(disk_super);
1613
1614                 blocksize =
1615                      btrfs_level_size(tree_root,
1616                                       btrfs_super_log_root_level(disk_super));
1617
1618                 log_tree_root = kzalloc(sizeof(struct btrfs_root),
1619                                                       GFP_NOFS);
1620
1621                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1622                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1623
1624                 log_tree_root->node = read_tree_block(tree_root, bytenr,
1625                                                       blocksize, 0);
1626                 ret = btrfs_recover_log_trees(log_tree_root);
1627                 BUG_ON(ret);
1628         }
1629         fs_info->last_trans_committed = btrfs_super_generation(disk_super);
1630         return tree_root;
1631
1632 fail_cleaner:
1633         kthread_stop(fs_info->cleaner_kthread);
1634 fail_extent_root:
1635         free_extent_buffer(extent_root->node);
1636 fail_tree_root:
1637         free_extent_buffer(tree_root->node);
1638 fail_sys_array:
1639 fail_sb_buffer:
1640         btrfs_stop_workers(&fs_info->fixup_workers);
1641         btrfs_stop_workers(&fs_info->workers);
1642         btrfs_stop_workers(&fs_info->endio_workers);
1643         btrfs_stop_workers(&fs_info->endio_write_workers);
1644         btrfs_stop_workers(&fs_info->submit_workers);
1645 fail_iput:
1646         iput(fs_info->btree_inode);
1647 fail:
1648         btrfs_close_devices(fs_info->fs_devices);
1649         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1650
1651         kfree(extent_root);
1652         kfree(tree_root);
1653 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1654         bdi_destroy(&fs_info->bdi);
1655 #endif
1656         kfree(fs_info);
1657         return ERR_PTR(err);
1658 }
1659
1660 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1661 {
1662         char b[BDEVNAME_SIZE];
1663
1664         if (uptodate) {
1665                 set_buffer_uptodate(bh);
1666         } else {
1667                 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1668                         printk(KERN_WARNING "lost page write due to "
1669                                         "I/O error on %s\n",
1670                                        bdevname(bh->b_bdev, b));
1671                 }
1672                 /* note, we dont' set_buffer_write_io_error because we have
1673                  * our own ways of dealing with the IO errors
1674                  */
1675                 clear_buffer_uptodate(bh);
1676         }
1677         unlock_buffer(bh);
1678         put_bh(bh);
1679 }
1680
1681 int write_all_supers(struct btrfs_root *root)
1682 {
1683         struct list_head *cur;
1684         struct list_head *head = &root->fs_info->fs_devices->devices;
1685         struct btrfs_device *dev;
1686         struct btrfs_super_block *sb;
1687         struct btrfs_dev_item *dev_item;
1688         struct buffer_head *bh;
1689         int ret;
1690         int do_barriers;
1691         int max_errors;
1692         int total_errors = 0;
1693         u32 crc;
1694         u64 flags;
1695
1696         max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1697         do_barriers = !btrfs_test_opt(root, NOBARRIER);
1698
1699         sb = &root->fs_info->super_for_commit;
1700         dev_item = &sb->dev_item;
1701         list_for_each(cur, head) {
1702                 dev = list_entry(cur, struct btrfs_device, dev_list);
1703                 if (!dev->bdev) {
1704                         total_errors++;
1705                         continue;
1706                 }
1707                 if (!dev->in_fs_metadata)
1708                         continue;
1709
1710                 btrfs_set_stack_device_type(dev_item, dev->type);
1711                 btrfs_set_stack_device_id(dev_item, dev->devid);
1712                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
1713                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
1714                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
1715                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
1716                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
1717                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
1718                 flags = btrfs_super_flags(sb);
1719                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
1720
1721
1722                 crc = ~(u32)0;
1723                 crc = btrfs_csum_data(root, (char *)sb + BTRFS_CSUM_SIZE, crc,
1724                                       BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
1725                 btrfs_csum_final(crc, sb->csum);
1726
1727                 bh = __getblk(dev->bdev, BTRFS_SUPER_INFO_OFFSET / 4096,
1728                               BTRFS_SUPER_INFO_SIZE);
1729
1730                 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
1731                 dev->pending_io = bh;
1732
1733                 get_bh(bh);
1734                 set_buffer_uptodate(bh);
1735                 lock_buffer(bh);
1736                 bh->b_end_io = btrfs_end_buffer_write_sync;
1737
1738                 if (do_barriers && dev->barriers) {
1739                         ret = submit_bh(WRITE_BARRIER, bh);
1740                         if (ret == -EOPNOTSUPP) {
1741                                 printk("btrfs: disabling barriers on dev %s\n",
1742                                        dev->name);
1743                                 set_buffer_uptodate(bh);
1744                                 dev->barriers = 0;
1745                                 get_bh(bh);
1746                                 lock_buffer(bh);
1747                                 ret = submit_bh(WRITE, bh);
1748                         }
1749                 } else {
1750                         ret = submit_bh(WRITE, bh);
1751                 }
1752                 if (ret)
1753                         total_errors++;
1754         }
1755         if (total_errors > max_errors) {
1756                 printk("btrfs: %d errors while writing supers\n", total_errors);
1757                 BUG();
1758         }
1759         total_errors = 0;
1760
1761         list_for_each(cur, head) {
1762                 dev = list_entry(cur, struct btrfs_device, dev_list);
1763                 if (!dev->bdev)
1764                         continue;
1765                 if (!dev->in_fs_metadata)
1766                         continue;
1767
1768                 BUG_ON(!dev->pending_io);
1769                 bh = dev->pending_io;
1770                 wait_on_buffer(bh);
1771                 if (!buffer_uptodate(dev->pending_io)) {
1772                         if (do_barriers && dev->barriers) {
1773                                 printk("btrfs: disabling barriers on dev %s\n",
1774                                        dev->name);
1775                                 set_buffer_uptodate(bh);
1776                                 get_bh(bh);
1777                                 lock_buffer(bh);
1778                                 dev->barriers = 0;
1779                                 ret = submit_bh(WRITE, bh);
1780                                 BUG_ON(ret);
1781                                 wait_on_buffer(bh);
1782                                 if (!buffer_uptodate(bh))
1783                                         total_errors++;
1784                         } else {
1785                                 total_errors++;
1786                         }
1787
1788                 }
1789                 dev->pending_io = NULL;
1790                 brelse(bh);
1791         }
1792         if (total_errors > max_errors) {
1793                 printk("btrfs: %d errors while writing supers\n", total_errors);
1794                 BUG();
1795         }
1796         return 0;
1797 }
1798
1799 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
1800                       *root)
1801 {
1802         int ret;
1803
1804         ret = write_all_supers(root);
1805         return ret;
1806 }
1807
1808 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
1809 {
1810         radix_tree_delete(&fs_info->fs_roots_radix,
1811                           (unsigned long)root->root_key.objectid);
1812         if (root->in_sysfs)
1813                 btrfs_sysfs_del_root(root);
1814         if (root->inode)
1815                 iput(root->inode);
1816         if (root->node)
1817                 free_extent_buffer(root->node);
1818         if (root->commit_root)
1819                 free_extent_buffer(root->commit_root);
1820         if (root->name)
1821                 kfree(root->name);
1822         kfree(root);
1823         return 0;
1824 }
1825
1826 static int del_fs_roots(struct btrfs_fs_info *fs_info)
1827 {
1828         int ret;
1829         struct btrfs_root *gang[8];
1830         int i;
1831
1832         while(1) {
1833                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1834                                              (void **)gang, 0,
1835                                              ARRAY_SIZE(gang));
1836                 if (!ret)
1837                         break;
1838                 for (i = 0; i < ret; i++)
1839                         btrfs_free_fs_root(fs_info, gang[i]);
1840         }
1841         return 0;
1842 }
1843
1844 int close_ctree(struct btrfs_root *root)
1845 {
1846         int ret;
1847         struct btrfs_trans_handle *trans;
1848         struct btrfs_fs_info *fs_info = root->fs_info;
1849
1850         fs_info->closing = 1;
1851         smp_mb();
1852
1853         kthread_stop(root->fs_info->transaction_kthread);
1854         kthread_stop(root->fs_info->cleaner_kthread);
1855
1856         btrfs_clean_old_snapshots(root);
1857         trans = btrfs_start_transaction(root, 1);
1858         ret = btrfs_commit_transaction(trans, root);
1859         /* run commit again to  drop the original snapshot */
1860         trans = btrfs_start_transaction(root, 1);
1861         btrfs_commit_transaction(trans, root);
1862         ret = btrfs_write_and_wait_transaction(NULL, root);
1863         BUG_ON(ret);
1864
1865         write_ctree_super(NULL, root);
1866
1867         if (fs_info->delalloc_bytes) {
1868                 printk("btrfs: at unmount delalloc count %Lu\n",
1869                        fs_info->delalloc_bytes);
1870         }
1871         if (fs_info->total_ref_cache_size) {
1872                 printk("btrfs: at umount reference cache size %Lu\n",
1873                         fs_info->total_ref_cache_size);
1874         }
1875
1876         if (fs_info->extent_root->node)
1877                 free_extent_buffer(fs_info->extent_root->node);
1878
1879         if (fs_info->tree_root->node)
1880                 free_extent_buffer(fs_info->tree_root->node);
1881
1882         if (root->fs_info->chunk_root->node);
1883                 free_extent_buffer(root->fs_info->chunk_root->node);
1884
1885         if (root->fs_info->dev_root->node);
1886                 free_extent_buffer(root->fs_info->dev_root->node);
1887
1888         btrfs_free_block_groups(root->fs_info);
1889         fs_info->closing = 2;
1890         del_fs_roots(fs_info);
1891
1892         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1893
1894         truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
1895
1896         btrfs_stop_workers(&fs_info->fixup_workers);
1897         btrfs_stop_workers(&fs_info->workers);
1898         btrfs_stop_workers(&fs_info->endio_workers);
1899         btrfs_stop_workers(&fs_info->endio_write_workers);
1900         btrfs_stop_workers(&fs_info->submit_workers);
1901
1902         iput(fs_info->btree_inode);
1903 #if 0
1904         while(!list_empty(&fs_info->hashers)) {
1905                 struct btrfs_hasher *hasher;
1906                 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
1907                                     hashers);
1908                 list_del(&hasher->hashers);
1909                 crypto_free_hash(&fs_info->hash_tfm);
1910                 kfree(hasher);
1911         }
1912 #endif
1913         btrfs_close_devices(fs_info->fs_devices);
1914         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1915
1916 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1917         bdi_destroy(&fs_info->bdi);
1918 #endif
1919
1920         kfree(fs_info->extent_root);
1921         kfree(fs_info->tree_root);
1922         kfree(fs_info->chunk_root);
1923         kfree(fs_info->dev_root);
1924         return 0;
1925 }
1926
1927 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
1928 {
1929         int ret;
1930         struct inode *btree_inode = buf->first_page->mapping->host;
1931
1932         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
1933         if (!ret)
1934                 return ret;
1935
1936         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
1937                                     parent_transid);
1938         return !ret;
1939 }
1940
1941 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
1942 {
1943         struct inode *btree_inode = buf->first_page->mapping->host;
1944         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
1945                                           buf);
1946 }
1947
1948 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
1949 {
1950         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1951         u64 transid = btrfs_header_generation(buf);
1952         struct inode *btree_inode = root->fs_info->btree_inode;
1953
1954         WARN_ON(!btrfs_tree_locked(buf));
1955         if (transid != root->fs_info->generation) {
1956                 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
1957                         (unsigned long long)buf->start,
1958                         transid, root->fs_info->generation);
1959                 WARN_ON(1);
1960         }
1961         set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
1962 }
1963
1964 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
1965 {
1966         /*
1967          * looks as though older kernels can get into trouble with
1968          * this code, they end up stuck in balance_dirty_pages forever
1969          */
1970         struct extent_io_tree *tree;
1971         u64 num_dirty;
1972         u64 start = 0;
1973         unsigned long thresh = 96 * 1024 * 1024;
1974         tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
1975
1976         if (current_is_pdflush() || current->flags & PF_MEMALLOC)
1977                 return;
1978
1979         num_dirty = count_range_bits(tree, &start, (u64)-1,
1980                                      thresh, EXTENT_DIRTY);
1981         if (num_dirty > thresh) {
1982                 balance_dirty_pages_ratelimited_nr(
1983                                    root->fs_info->btree_inode->i_mapping, 1);
1984         }
1985         return;
1986 }
1987
1988 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
1989 {
1990         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1991         int ret;
1992         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1993         if (ret == 0) {
1994                 buf->flags |= EXTENT_UPTODATE;
1995         }
1996         return ret;
1997 }
1998
1999 int btree_lock_page_hook(struct page *page)
2000 {
2001         struct inode *inode = page->mapping->host;
2002         struct btrfs_root *root = BTRFS_I(inode)->root;
2003         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2004         struct extent_buffer *eb;
2005         unsigned long len;
2006         u64 bytenr = page_offset(page);
2007
2008         if (page->private == EXTENT_PAGE_PRIVATE)
2009                 goto out;
2010
2011         len = page->private >> 2;
2012         eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2013         if (!eb)
2014                 goto out;
2015
2016         btrfs_tree_lock(eb);
2017         spin_lock(&root->fs_info->hash_lock);
2018         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2019         spin_unlock(&root->fs_info->hash_lock);
2020         btrfs_tree_unlock(eb);
2021         free_extent_buffer(eb);
2022 out:
2023         lock_page(page);
2024         return 0;
2025 }
2026
2027 static struct extent_io_ops btree_extent_io_ops = {
2028         .write_cache_pages_lock_hook = btree_lock_page_hook,
2029         .writepage_io_hook = btree_writepage_io_hook,
2030         .readpage_end_io_hook = btree_readpage_end_io_hook,
2031         .submit_bio_hook = btree_submit_bio_hook,
2032         /* note we're sharing with inode.c for the merge bio hook */
2033         .merge_bio_hook = btrfs_merge_bio_hook,
2034 };