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 (0 && 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         if (PageWriteback(page) || PageDirty(page))
 596             return 0;
 597
 598         tree = &BTRFS_I(page->mapping->host)->io_tree;
 599         map = &BTRFS_I(page->mapping->host)->extent_tree;
 600
 601         ret = try_release_extent_state(map, tree, page, gfp_flags);
 602         if (!ret) {
 603                 return 0;
 604         }
 605
 606         ret = try_release_extent_buffer(tree, page);
 607         if (ret == 1) {
 608                 ClearPagePrivate(page);
 609                 set_page_private(page, 0);
 610                 page_cache_release(page);
 611         }
 612
 613         return ret;
 614 }
 615
 616 static void btree_invalidatepage(struct page *page, unsigned long offset)
 617 {
 618         struct extent_io_tree *tree;
 619         tree = &BTRFS_I(page->mapping->host)->io_tree;
 620         extent_invalidatepage(tree, page, offset);
 621         btree_releasepage(page, GFP_NOFS);
 622         if (PagePrivate(page)) {
 623                 printk("warning page private not zero on page %Lu\n",
 624                        page_offset(page));
 625                 ClearPagePrivate(page);
 626                 set_page_private(page, 0);
 627                 page_cache_release(page);
 628         }
 629 }
 630
 631 #if 0
 632 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 633 {
 634         struct buffer_head *bh;
 635         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 636         struct buffer_head *head;
 637         if (!page_has_buffers(page)) {
 638                 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
 639                                         (1 << BH_Dirty)|(1 << BH_Uptodate));
 640         }
 641         head = page_buffers(page);
 642         bh = head;
 643         do {
 644                 if (buffer_dirty(bh))
 645                         csum_tree_block(root, bh, 0);
 646                 bh = bh->b_this_page;
 647         } while (bh != head);
 648         return block_write_full_page(page, btree_get_block, wbc);
 649 }
 650 #endif
 651
 652 static struct address_space_operations btree_aops = {
 653         .readpage       = btree_readpage,
 654         .writepage      = btree_writepage,
 655         .writepages     = btree_writepages,
 656         .releasepage    = btree_releasepage,
 657         .invalidatepage = btree_invalidatepage,
 658         .sync_page      = block_sync_page,
 659 };
 660
 661 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
 662                          u64 parent_transid)
 663 {
 664         struct extent_buffer *buf = NULL;
 665         struct inode *btree_inode = root->fs_info->btree_inode;
 666         int ret = 0;
 667
 668         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 669         if (!buf)
 670                 return 0;
 671         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
 672                                  buf, 0, 0, btree_get_extent, 0);
 673         free_extent_buffer(buf);
 674         return ret;
 675 }
 676
 677 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
 678                                             u64 bytenr, u32 blocksize)
 679 {
 680         struct inode *btree_inode = root->fs_info->btree_inode;
 681         struct extent_buffer *eb;
 682         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 683                                 bytenr, blocksize, GFP_NOFS);
 684         return eb;
 685 }
 686
 687 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
 688                                                  u64 bytenr, u32 blocksize)
 689 {
 690         struct inode *btree_inode = root->fs_info->btree_inode;
 691         struct extent_buffer *eb;
 692
 693         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 694                                  bytenr, blocksize, NULL, GFP_NOFS);
 695         return eb;
 696 }
 697
 698
 699 int btrfs_write_tree_block(struct extent_buffer *buf)
 700 {
 701         return btrfs_fdatawrite_range(buf->first_page->mapping, buf->start,
 702                                       buf->start + buf->len - 1, WB_SYNC_NONE);
 703 }
 704
 705 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
 706 {
 707         return btrfs_wait_on_page_writeback_range(buf->first_page->mapping,
 708                                   buf->start, buf->start + buf->len -1);
 709 }
 710
 711 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
 712                                       u32 blocksize, u64 parent_transid)
 713 {
 714         struct extent_buffer *buf = NULL;
 715         struct inode *btree_inode = root->fs_info->btree_inode;
 716         struct extent_io_tree *io_tree;
 717         int ret;
 718
 719         io_tree = &BTRFS_I(btree_inode)->io_tree;
 720
 721         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 722         if (!buf)
 723                 return NULL;
 724
 725         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
 726
 727         if (ret == 0) {
 728                 buf->flags |= EXTENT_UPTODATE;
 729         } else {
 730                 WARN_ON(1);
 731         }
 732         return buf;
 733
 734 }
 735
 736 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 737                      struct extent_buffer *buf)
 738 {
 739         struct inode *btree_inode = root->fs_info->btree_inode;
 740         if (btrfs_header_generation(buf) ==
 741             root->fs_info->running_transaction->transid) {
 742                 WARN_ON(!btrfs_tree_locked(buf));
 743                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
 744                                           buf);
 745         }
 746         return 0;
 747 }
 748
 749 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
 750                         u32 stripesize, struct btrfs_root *root,
 751                         struct btrfs_fs_info *fs_info,
 752                         u64 objectid)
 753 {
 754         root->node = NULL;
 755         root->inode = NULL;
 756         root->commit_root = NULL;
 757         root->ref_tree = NULL;
 758         root->sectorsize = sectorsize;
 759         root->nodesize = nodesize;
 760         root->leafsize = leafsize;
 761         root->stripesize = stripesize;
 762         root->ref_cows = 0;
 763         root->track_dirty = 0;
 764
 765         root->fs_info = fs_info;
 766         root->objectid = objectid;
 767         root->last_trans = 0;
 768         root->highest_inode = 0;
 769         root->last_inode_alloc = 0;
 770         root->name = NULL;
 771         root->in_sysfs = 0;
 772
 773         INIT_LIST_HEAD(&root->dirty_list);
 774         INIT_LIST_HEAD(&root->orphan_list);
 775         INIT_LIST_HEAD(&root->dead_list);
 776         spin_lock_init(&root->node_lock);
 777         spin_lock_init(&root->list_lock);
 778         mutex_init(&root->objectid_mutex);
 779         mutex_init(&root->log_mutex);
 780
 781         btrfs_leaf_ref_tree_init(&root->ref_tree_struct);
 782         root->ref_tree = &root->ref_tree_struct;
 783
 784         memset(&root->root_key, 0, sizeof(root->root_key));
 785         memset(&root->root_item, 0, sizeof(root->root_item));
 786         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
 787         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
 788         root->defrag_trans_start = fs_info->generation;
 789         init_completion(&root->kobj_unregister);
 790         root->defrag_running = 0;
 791         root->defrag_level = 0;
 792         root->root_key.objectid = objectid;
 793         return 0;
 794 }
 795
 796 static int find_and_setup_root(struct btrfs_root *tree_root,
 797                                struct btrfs_fs_info *fs_info,
 798                                u64 objectid,
 799                                struct btrfs_root *root)
 800 {
 801         int ret;
 802         u32 blocksize;
 803
 804         __setup_root(tree_root->nodesize, tree_root->leafsize,
 805                      tree_root->sectorsize, tree_root->stripesize,
 806                      root, fs_info, objectid);
 807         ret = btrfs_find_last_root(tree_root, objectid,
 808                                    &root->root_item, &root->root_key);
 809         BUG_ON(ret);
 810
 811         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 812         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 813                                      blocksize, 0);
 814         BUG_ON(!root->node);
 815         return 0;
 816 }
 817
 818 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
 819                              struct btrfs_fs_info *fs_info)
 820 {
 821         struct extent_buffer *eb;
 822         int ret;
 823
 824         if (!fs_info->log_root_tree)
 825                 return 0;
 826
 827         eb = fs_info->log_root_tree->node;
 828
 829         WARN_ON(btrfs_header_level(eb) != 0);
 830         WARN_ON(btrfs_header_nritems(eb) != 0);
 831
 832         ret = btrfs_free_extent(trans, fs_info->tree_root,
 833                                 eb->start, eb->len,
 834                                 BTRFS_TREE_LOG_OBJECTID, 0, 0, 0, 1);
 835         BUG_ON(ret);
 836
 837         free_extent_buffer(eb);
 838         kfree(fs_info->log_root_tree);
 839         fs_info->log_root_tree = NULL;
 840         return 0;
 841 }
 842
 843 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
 844                              struct btrfs_fs_info *fs_info)
 845 {
 846         struct btrfs_root *root;
 847         struct btrfs_root *tree_root = fs_info->tree_root;
 848
 849         root = kzalloc(sizeof(*root), GFP_NOFS);
 850         if (!root)
 851                 return -ENOMEM;
 852
 853         __setup_root(tree_root->nodesize, tree_root->leafsize,
 854                      tree_root->sectorsize, tree_root->stripesize,
 855                      root, fs_info, BTRFS_TREE_LOG_OBJECTID);
 856
 857         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
 858         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
 859         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
 860         root->ref_cows = 0;
 861
 862         root->node = btrfs_alloc_free_block(trans, root, root->leafsize,
 863                                             BTRFS_TREE_LOG_OBJECTID,
 864                                             0, 0, 0, 0, 0);
 865
 866         btrfs_set_header_nritems(root->node, 0);
 867         btrfs_set_header_level(root->node, 0);
 868         btrfs_set_header_bytenr(root->node, root->node->start);
 869         btrfs_set_header_generation(root->node, trans->transid);
 870         btrfs_set_header_owner(root->node, BTRFS_TREE_LOG_OBJECTID);
 871
 872         write_extent_buffer(root->node, root->fs_info->fsid,
 873                             (unsigned long)btrfs_header_fsid(root->node),
 874                             BTRFS_FSID_SIZE);
 875         btrfs_mark_buffer_dirty(root->node);
 876         btrfs_tree_unlock(root->node);
 877         fs_info->log_root_tree = root;
 878         return 0;
 879 }
 880
 881 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
 882                                                struct btrfs_key *location)
 883 {
 884         struct btrfs_root *root;
 885         struct btrfs_fs_info *fs_info = tree_root->fs_info;
 886         struct btrfs_path *path;
 887         struct extent_buffer *l;
 888         u64 highest_inode;
 889         u32 blocksize;
 890         int ret = 0;
 891
 892         root = kzalloc(sizeof(*root), GFP_NOFS);
 893         if (!root)
 894                 return ERR_PTR(-ENOMEM);
 895         if (location->offset == (u64)-1) {
 896                 ret = find_and_setup_root(tree_root, fs_info,
 897                                           location->objectid, root);
 898                 if (ret) {
 899                         kfree(root);
 900                         return ERR_PTR(ret);
 901                 }
 902                 goto insert;
 903         }
 904
 905         __setup_root(tree_root->nodesize, tree_root->leafsize,
 906                      tree_root->sectorsize, tree_root->stripesize,
 907                      root, fs_info, location->objectid);
 908
 909         path = btrfs_alloc_path();
 910         BUG_ON(!path);
 911         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
 912         if (ret != 0) {
 913                 if (ret > 0)
 914                         ret = -ENOENT;
 915                 goto out;
 916         }
 917         l = path->nodes[0];
 918         read_extent_buffer(l, &root->root_item,
 919                btrfs_item_ptr_offset(l, path->slots[0]),
 920                sizeof(root->root_item));
 921         memcpy(&root->root_key, location, sizeof(*location));
 922         ret = 0;
 923 out:
 924         btrfs_release_path(root, path);
 925         btrfs_free_path(path);
 926         if (ret) {
 927                 kfree(root);
 928                 return ERR_PTR(ret);
 929         }
 930         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 931         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 932                                      blocksize, 0);
 933         BUG_ON(!root->node);
 934 insert:
 935         if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
 936                 root->ref_cows = 1;
 937                 ret = btrfs_find_highest_inode(root, &highest_inode);
 938                 if (ret == 0) {
 939                         root->highest_inode = highest_inode;
 940                         root->last_inode_alloc = highest_inode;
 941                 }
 942         }
 943         return root;
 944 }
 945
 946 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
 947                                         u64 root_objectid)
 948 {
 949         struct btrfs_root *root;
 950
 951         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
 952                 return fs_info->tree_root;
 953         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
 954                 return fs_info->extent_root;
 955
 956         root = radix_tree_lookup(&fs_info->fs_roots_radix,
 957                                  (unsigned long)root_objectid);
 958         return root;
 959 }
 960
 961 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
 962                                               struct btrfs_key *location)
 963 {
 964         struct btrfs_root *root;
 965         int ret;
 966
 967         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
 968                 return fs_info->tree_root;
 969         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
 970                 return fs_info->extent_root;
 971         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
 972                 return fs_info->chunk_root;
 973         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
 974                 return fs_info->dev_root;
 975
 976         root = radix_tree_lookup(&fs_info->fs_roots_radix,
 977                                  (unsigned long)location->objectid);
 978         if (root)
 979                 return root;
 980
 981         root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
 982         if (IS_ERR(root))
 983                 return root;
 984         ret = radix_tree_insert(&fs_info->fs_roots_radix,
 985                                 (unsigned long)root->root_key.objectid,
 986                                 root);
 987         if (ret) {
 988                 free_extent_buffer(root->node);
 989                 kfree(root);
 990                 return ERR_PTR(ret);
 991         }
 992         ret = btrfs_find_dead_roots(fs_info->tree_root,
 993                                     root->root_key.objectid, root);
 994         BUG_ON(ret);
 995
 996         return root;
 997 }
 998
 999 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1000                                       struct btrfs_key *location,
1001                                       const char *name, int namelen)
1002 {
1003         struct btrfs_root *root;
1004         int ret;
1005
1006         root = btrfs_read_fs_root_no_name(fs_info, location);
1007         if (!root)
1008                 return NULL;
1009
1010         if (root->in_sysfs)
1011                 return root;
1012
1013         ret = btrfs_set_root_name(root, name, namelen);
1014         if (ret) {
1015                 free_extent_buffer(root->node);
1016                 kfree(root);
1017                 return ERR_PTR(ret);
1018         }
1019
1020         ret = btrfs_sysfs_add_root(root);
1021         if (ret) {
1022                 free_extent_buffer(root->node);
1023                 kfree(root->name);
1024                 kfree(root);
1025                 return ERR_PTR(ret);
1026         }
1027         root->in_sysfs = 1;
1028         return root;
1029 }
1030 #if 0
1031 static int add_hasher(struct btrfs_fs_info *info, char *type) {
1032         struct btrfs_hasher *hasher;
1033
1034         hasher = kmalloc(sizeof(*hasher), GFP_NOFS);
1035         if (!hasher)
1036                 return -ENOMEM;
1037         hasher->hash_tfm = crypto_alloc_hash(type, 0, CRYPTO_ALG_ASYNC);
1038         if (!hasher->hash_tfm) {
1039                 kfree(hasher);
1040                 return -EINVAL;
1041         }
1042         spin_lock(&info->hash_lock);
1043         list_add(&hasher->list, &info->hashers);
1044         spin_unlock(&info->hash_lock);
1045         return 0;
1046 }
1047 #endif
1048
1049 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1050 {
1051         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1052         int ret = 0;
1053         struct list_head *cur;
1054         struct btrfs_device *device;
1055         struct backing_dev_info *bdi;
1056
1057         if ((bdi_bits & (1 << BDI_write_congested)) &&
1058             btrfs_congested_async(info, 0))
1059                 return 1;
1060
1061         list_for_each(cur, &info->fs_devices->devices) {
1062                 device = list_entry(cur, struct btrfs_device, dev_list);
1063                 if (!device->bdev)
1064                         continue;
1065                 bdi = blk_get_backing_dev_info(device->bdev);
1066                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1067                         ret = 1;
1068                         break;
1069                 }
1070         }
1071         return ret;
1072 }
1073
1074 /*
1075  * this unplugs every device on the box, and it is only used when page
1076  * is null
1077  */
1078 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1079 {
1080         struct list_head *cur;
1081         struct btrfs_device *device;
1082         struct btrfs_fs_info *info;
1083
1084         info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1085         list_for_each(cur, &info->fs_devices->devices) {
1086                 device = list_entry(cur, struct btrfs_device, dev_list);
1087                 bdi = blk_get_backing_dev_info(device->bdev);
1088                 if (bdi->unplug_io_fn) {
1089                         bdi->unplug_io_fn(bdi, page);
1090                 }
1091         }
1092 }
1093
1094 void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1095 {
1096         struct inode *inode;
1097         struct extent_map_tree *em_tree;
1098         struct extent_map *em;
1099         struct address_space *mapping;
1100         u64 offset;
1101
1102         /* the generic O_DIRECT read code does this */
1103         if (!page) {
1104                 __unplug_io_fn(bdi, page);
1105                 return;
1106         }
1107
1108         /*
1109          * page->mapping may change at any time.  Get a consistent copy
1110          * and use that for everything below
1111          */
1112         smp_mb();
1113         mapping = page->mapping;
1114         if (!mapping)
1115                 return;
1116
1117         inode = mapping->host;
1118         offset = page_offset(page);
1119
1120         em_tree = &BTRFS_I(inode)->extent_tree;
1121         spin_lock(&em_tree->lock);
1122         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1123         spin_unlock(&em_tree->lock);
1124         if (!em) {
1125                 __unplug_io_fn(bdi, page);
1126                 return;
1127         }
1128
1129         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1130                 free_extent_map(em);
1131                 __unplug_io_fn(bdi, page);
1132                 return;
1133         }
1134         offset = offset - em->start;
1135         btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1136                           em->block_start + offset, page);
1137         free_extent_map(em);
1138 }
1139
1140 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1141 {
1142 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1143         bdi_init(bdi);
1144 #endif
1145         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1146         bdi->state              = 0;
1147         bdi->capabilities       = default_backing_dev_info.capabilities;
1148         bdi->unplug_io_fn       = btrfs_unplug_io_fn;
1149         bdi->unplug_io_data     = info;
1150         bdi->congested_fn       = btrfs_congested_fn;
1151         bdi->congested_data     = info;
1152         return 0;
1153 }
1154
1155 static int bio_ready_for_csum(struct bio *bio)
1156 {
1157         u64 length = 0;
1158         u64 buf_len = 0;
1159         u64 start = 0;
1160         struct page *page;
1161         struct extent_io_tree *io_tree = NULL;
1162         struct btrfs_fs_info *info = NULL;
1163         struct bio_vec *bvec;
1164         int i;
1165         int ret;
1166
1167         bio_for_each_segment(bvec, bio, i) {
1168                 page = bvec->bv_page;
1169                 if (page->private == EXTENT_PAGE_PRIVATE) {
1170                         length += bvec->bv_len;
1171                         continue;
1172                 }
1173                 if (!page->private) {
1174                         length += bvec->bv_len;
1175                         continue;
1176                 }
1177                 length = bvec->bv_len;
1178                 buf_len = page->private >> 2;
1179                 start = page_offset(page) + bvec->bv_offset;
1180                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1181                 info = BTRFS_I(page->mapping->host)->root->fs_info;
1182         }
1183         /* are we fully contained in this bio? */
1184         if (buf_len <= length)
1185                 return 1;
1186
1187         ret = extent_range_uptodate(io_tree, start + length,
1188                                     start + buf_len - 1);
1189         if (ret == 1)
1190                 return ret;
1191         return ret;
1192 }
1193
1194 /*
1195  * called by the kthread helper functions to finally call the bio end_io
1196  * functions.  This is where read checksum verification actually happens
1197  */
1198 static void end_workqueue_fn(struct btrfs_work *work)
1199 {
1200         struct bio *bio;
1201         struct end_io_wq *end_io_wq;
1202         struct btrfs_fs_info *fs_info;
1203         int error;
1204
1205         end_io_wq = container_of(work, struct end_io_wq, work);
1206         bio = end_io_wq->bio;
1207         fs_info = end_io_wq->info;
1208
1209         /* metadata bios are special because the whole tree block must
1210          * be checksummed at once.  This makes sure the entire block is in
1211          * ram and up to date before trying to verify things.  For
1212          * blocksize <= pagesize, it is basically a noop
1213          */
1214         if (end_io_wq->metadata && !bio_ready_for_csum(bio)) {
1215                 btrfs_queue_worker(&fs_info->endio_workers,
1216                                    &end_io_wq->work);
1217                 return;
1218         }
1219         error = end_io_wq->error;
1220         bio->bi_private = end_io_wq->private;
1221         bio->bi_end_io = end_io_wq->end_io;
1222         kfree(end_io_wq);
1223 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1224         bio_endio(bio, bio->bi_size, error);
1225 #else
1226         bio_endio(bio, error);
1227 #endif
1228 }
1229
1230 static int cleaner_kthread(void *arg)
1231 {
1232         struct btrfs_root *root = arg;
1233
1234         do {
1235                 smp_mb();
1236                 if (root->fs_info->closing)
1237                         break;
1238
1239                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1240                 mutex_lock(&root->fs_info->cleaner_mutex);
1241                 btrfs_clean_old_snapshots(root);
1242                 mutex_unlock(&root->fs_info->cleaner_mutex);
1243
1244                 if (freezing(current)) {
1245                         refrigerator();
1246                 } else {
1247                         smp_mb();
1248                         if (root->fs_info->closing)
1249                                 break;
1250                         set_current_state(TASK_INTERRUPTIBLE);
1251                         schedule();
1252                         __set_current_state(TASK_RUNNING);
1253                 }
1254         } while (!kthread_should_stop());
1255         return 0;
1256 }
1257
1258 static int transaction_kthread(void *arg)
1259 {
1260         struct btrfs_root *root = arg;
1261         struct btrfs_trans_handle *trans;
1262         struct btrfs_transaction *cur;
1263         unsigned long now;
1264         unsigned long delay;
1265         int ret;
1266
1267         do {
1268                 smp_mb();
1269                 if (root->fs_info->closing)
1270                         break;
1271
1272                 delay = HZ * 30;
1273                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1274                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1275
1276                 if (root->fs_info->total_ref_cache_size > 20 * 1024 * 1024) {
1277                         printk("btrfs: total reference cache size %Lu\n",
1278                                 root->fs_info->total_ref_cache_size);
1279                 }
1280
1281                 mutex_lock(&root->fs_info->trans_mutex);
1282                 cur = root->fs_info->running_transaction;
1283                 if (!cur) {
1284                         mutex_unlock(&root->fs_info->trans_mutex);
1285                         goto sleep;
1286                 }
1287
1288                 now = get_seconds();
1289                 if (now < cur->start_time || now - cur->start_time < 30) {
1290                         mutex_unlock(&root->fs_info->trans_mutex);
1291                         delay = HZ * 5;
1292                         goto sleep;
1293                 }
1294                 mutex_unlock(&root->fs_info->trans_mutex);
1295                 trans = btrfs_start_transaction(root, 1);
1296                 ret = btrfs_commit_transaction(trans, root);
1297 sleep:
1298                 wake_up_process(root->fs_info->cleaner_kthread);
1299                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1300
1301                 if (freezing(current)) {
1302                         refrigerator();
1303                 } else {
1304                         if (root->fs_info->closing)
1305                                 break;
1306                         set_current_state(TASK_INTERRUPTIBLE);
1307                         schedule_timeout(delay);
1308                         __set_current_state(TASK_RUNNING);
1309                 }
1310         } while (!kthread_should_stop());
1311         return 0;
1312 }
1313
1314 struct btrfs_root *open_ctree(struct super_block *sb,
1315                               struct btrfs_fs_devices *fs_devices,
1316                               char *options)
1317 {
1318         u32 sectorsize;
1319         u32 nodesize;
1320         u32 leafsize;
1321         u32 blocksize;
1322         u32 stripesize;
1323         struct buffer_head *bh;
1324         struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1325                                                  GFP_NOFS);
1326         struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
1327                                                GFP_NOFS);
1328         struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1329                                                 GFP_NOFS);
1330         struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1331                                                 GFP_NOFS);
1332         struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1333                                               GFP_NOFS);
1334         struct btrfs_root *log_tree_root;
1335
1336         int ret;
1337         int err = -EINVAL;
1338
1339         struct btrfs_super_block *disk_super;
1340
1341         if (!extent_root || !tree_root || !fs_info) {
1342                 err = -ENOMEM;
1343                 goto fail;
1344         }
1345         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1346         INIT_LIST_HEAD(&fs_info->trans_list);
1347         INIT_LIST_HEAD(&fs_info->dead_roots);
1348         INIT_LIST_HEAD(&fs_info->hashers);
1349         INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1350         spin_lock_init(&fs_info->hash_lock);
1351         spin_lock_init(&fs_info->delalloc_lock);
1352         spin_lock_init(&fs_info->new_trans_lock);
1353         spin_lock_init(&fs_info->ref_cache_lock);
1354
1355         init_completion(&fs_info->kobj_unregister);
1356         fs_info->tree_root = tree_root;
1357         fs_info->extent_root = extent_root;
1358         fs_info->chunk_root = chunk_root;
1359         fs_info->dev_root = dev_root;
1360         fs_info->fs_devices = fs_devices;
1361         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1362         INIT_LIST_HEAD(&fs_info->space_info);
1363         btrfs_mapping_init(&fs_info->mapping_tree);
1364         atomic_set(&fs_info->nr_async_submits, 0);
1365         atomic_set(&fs_info->nr_async_bios, 0);
1366         atomic_set(&fs_info->throttles, 0);
1367         atomic_set(&fs_info->throttle_gen, 0);
1368         fs_info->sb = sb;
1369         fs_info->max_extent = (u64)-1;
1370         fs_info->max_inline = 8192 * 1024;
1371         setup_bdi(fs_info, &fs_info->bdi);
1372         fs_info->btree_inode = new_inode(sb);
1373         fs_info->btree_inode->i_ino = 1;
1374         fs_info->btree_inode->i_nlink = 1;
1375         fs_info->thread_pool_size = min(num_online_cpus() + 2, 8);
1376
1377         INIT_LIST_HEAD(&fs_info->ordered_extents);
1378         spin_lock_init(&fs_info->ordered_extent_lock);
1379
1380         sb->s_blocksize = 4096;
1381         sb->s_blocksize_bits = blksize_bits(4096);
1382
1383         /*
1384          * we set the i_size on the btree inode to the max possible int.
1385          * the real end of the address space is determined by all of
1386          * the devices in the system
1387          */
1388         fs_info->btree_inode->i_size = OFFSET_MAX;
1389         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1390         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1391
1392         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1393                              fs_info->btree_inode->i_mapping,
1394                              GFP_NOFS);
1395         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1396                              GFP_NOFS);
1397
1398         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1399
1400         extent_io_tree_init(&fs_info->free_space_cache,
1401                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1402         extent_io_tree_init(&fs_info->block_group_cache,
1403                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1404         extent_io_tree_init(&fs_info->pinned_extents,
1405                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1406         extent_io_tree_init(&fs_info->pending_del,
1407                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1408         extent_io_tree_init(&fs_info->extent_ins,
1409                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1410         fs_info->do_barriers = 1;
1411
1412         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1413         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1414                sizeof(struct btrfs_key));
1415         insert_inode_hash(fs_info->btree_inode);
1416         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1417
1418         mutex_init(&fs_info->trans_mutex);
1419         mutex_init(&fs_info->tree_log_mutex);
1420         mutex_init(&fs_info->drop_mutex);
1421         mutex_init(&fs_info->alloc_mutex);
1422         mutex_init(&fs_info->chunk_mutex);
1423         mutex_init(&fs_info->transaction_kthread_mutex);
1424         mutex_init(&fs_info->cleaner_mutex);
1425         mutex_init(&fs_info->volume_mutex);
1426         init_waitqueue_head(&fs_info->transaction_throttle);
1427         init_waitqueue_head(&fs_info->transaction_wait);
1428         init_waitqueue_head(&fs_info->async_submit_wait);
1429         init_waitqueue_head(&fs_info->tree_log_wait);
1430         atomic_set(&fs_info->tree_log_commit, 0);
1431         atomic_set(&fs_info->tree_log_writers, 0);
1432         fs_info->tree_log_transid = 0;
1433
1434 #if 0
1435         ret = add_hasher(fs_info, "crc32c");
1436         if (ret) {
1437                 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
1438                 err = -ENOMEM;
1439                 goto fail_iput;
1440         }
1441 #endif
1442         __setup_root(4096, 4096, 4096, 4096, tree_root,
1443                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1444
1445
1446         bh = __bread(fs_devices->latest_bdev,
1447                      BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
1448         if (!bh)
1449                 goto fail_iput;
1450
1451         memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1452         brelse(bh);
1453
1454         memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1455
1456         disk_super = &fs_info->super_copy;
1457         if (!btrfs_super_root(disk_super))
1458                 goto fail_sb_buffer;
1459
1460         err = btrfs_parse_options(tree_root, options);
1461         if (err)
1462                 goto fail_sb_buffer;
1463
1464         /*
1465          * we need to start all the end_io workers up front because the
1466          * queue work function gets called at interrupt time, and so it
1467          * cannot dynamically grow.
1468          */
1469         btrfs_init_workers(&fs_info->workers, "worker",
1470                            fs_info->thread_pool_size);
1471         btrfs_init_workers(&fs_info->submit_workers, "submit",
1472                            min_t(u64, fs_devices->num_devices,
1473                            fs_info->thread_pool_size));
1474
1475         /* a higher idle thresh on the submit workers makes it much more
1476          * likely that bios will be send down in a sane order to the
1477          * devices
1478          */
1479         fs_info->submit_workers.idle_thresh = 64;
1480
1481         /* fs_info->workers is responsible for checksumming file data
1482          * blocks and metadata.  Using a larger idle thresh allows each
1483          * worker thread to operate on things in roughly the order they
1484          * were sent by the writeback daemons, improving overall locality
1485          * of the IO going down the pipe.
1486          */
1487         fs_info->workers.idle_thresh = 128;
1488
1489         btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1);
1490         btrfs_init_workers(&fs_info->endio_workers, "endio",
1491                            fs_info->thread_pool_size);
1492         btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1493                            fs_info->thread_pool_size);
1494
1495         /*
1496          * endios are largely parallel and should have a very
1497          * low idle thresh
1498          */
1499         fs_info->endio_workers.idle_thresh = 4;
1500         fs_info->endio_write_workers.idle_thresh = 64;
1501
1502         btrfs_start_workers(&fs_info->workers, 1);
1503         btrfs_start_workers(&fs_info->submit_workers, 1);
1504         btrfs_start_workers(&fs_info->fixup_workers, 1);
1505         btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
1506         btrfs_start_workers(&fs_info->endio_write_workers,
1507                             fs_info->thread_pool_size);
1508
1509         err = -EINVAL;
1510         if (btrfs_super_num_devices(disk_super) > fs_devices->open_devices) {
1511                 printk("Btrfs: wanted %llu devices, but found %llu\n",
1512                        (unsigned long long)btrfs_super_num_devices(disk_super),
1513                        (unsigned long long)fs_devices->open_devices);
1514                 if (btrfs_test_opt(tree_root, DEGRADED))
1515                         printk("continuing in degraded mode\n");
1516                 else {
1517                         goto fail_sb_buffer;
1518                 }
1519         }
1520
1521         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1522
1523         nodesize = btrfs_super_nodesize(disk_super);
1524         leafsize = btrfs_super_leafsize(disk_super);
1525         sectorsize = btrfs_super_sectorsize(disk_super);
1526         stripesize = btrfs_super_stripesize(disk_super);
1527         tree_root->nodesize = nodesize;
1528         tree_root->leafsize = leafsize;
1529         tree_root->sectorsize = sectorsize;
1530         tree_root->stripesize = stripesize;
1531
1532         sb->s_blocksize = sectorsize;
1533         sb->s_blocksize_bits = blksize_bits(sectorsize);
1534
1535         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1536                     sizeof(disk_super->magic))) {
1537                 printk("btrfs: valid FS not found on %s\n", sb->s_id);
1538                 goto fail_sb_buffer;
1539         }
1540
1541         mutex_lock(&fs_info->chunk_mutex);
1542         ret = btrfs_read_sys_array(tree_root);
1543         mutex_unlock(&fs_info->chunk_mutex);
1544         if (ret) {
1545                 printk("btrfs: failed to read the system array on %s\n",
1546                        sb->s_id);
1547                 goto fail_sys_array;
1548         }
1549
1550         blocksize = btrfs_level_size(tree_root,
1551                                      btrfs_super_chunk_root_level(disk_super));
1552
1553         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1554                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1555
1556         chunk_root->node = read_tree_block(chunk_root,
1557                                            btrfs_super_chunk_root(disk_super),
1558                                            blocksize, 0);
1559         BUG_ON(!chunk_root->node);
1560
1561         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1562                  (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1563                  BTRFS_UUID_SIZE);
1564
1565         mutex_lock(&fs_info->chunk_mutex);
1566         ret = btrfs_read_chunk_tree(chunk_root);
1567         mutex_unlock(&fs_info->chunk_mutex);
1568         BUG_ON(ret);
1569
1570         btrfs_close_extra_devices(fs_devices);
1571
1572         blocksize = btrfs_level_size(tree_root,
1573                                      btrfs_super_root_level(disk_super));
1574
1575
1576         tree_root->node = read_tree_block(tree_root,
1577                                           btrfs_super_root(disk_super),
1578                                           blocksize, 0);
1579         if (!tree_root->node)
1580                 goto fail_sb_buffer;
1581
1582
1583         ret = find_and_setup_root(tree_root, fs_info,
1584                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1585         if (ret)
1586                 goto fail_tree_root;
1587         extent_root->track_dirty = 1;
1588
1589         ret = find_and_setup_root(tree_root, fs_info,
1590                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
1591         dev_root->track_dirty = 1;
1592
1593         if (ret)
1594                 goto fail_extent_root;
1595
1596         btrfs_read_block_groups(extent_root);
1597
1598         fs_info->generation = btrfs_super_generation(disk_super) + 1;
1599         fs_info->data_alloc_profile = (u64)-1;
1600         fs_info->metadata_alloc_profile = (u64)-1;
1601         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1602         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1603                                                "btrfs-cleaner");
1604         if (!fs_info->cleaner_kthread)
1605                 goto fail_extent_root;
1606
1607         fs_info->transaction_kthread = kthread_run(transaction_kthread,
1608                                                    tree_root,
1609                                                    "btrfs-transaction");
1610         if (!fs_info->transaction_kthread)
1611                 goto fail_cleaner;
1612
1613         if (btrfs_super_log_root(disk_super) != 0) {
1614                 u32 blocksize;
1615                 u64 bytenr = btrfs_super_log_root(disk_super);
1616
1617                 blocksize =
1618                      btrfs_level_size(tree_root,
1619                                       btrfs_super_log_root_level(disk_super));
1620
1621                 log_tree_root = kzalloc(sizeof(struct btrfs_root),
1622                                                       GFP_NOFS);
1623
1624                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1625                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1626
1627                 log_tree_root->node = read_tree_block(tree_root, bytenr,
1628                                                       blocksize, 0);
1629                 ret = btrfs_recover_log_trees(log_tree_root);
1630                 BUG_ON(ret);
1631         }
1632         fs_info->last_trans_committed = btrfs_super_generation(disk_super);
1633         return tree_root;
1634
1635 fail_cleaner:
1636         kthread_stop(fs_info->cleaner_kthread);
1637 fail_extent_root:
1638         free_extent_buffer(extent_root->node);
1639 fail_tree_root:
1640         free_extent_buffer(tree_root->node);
1641 fail_sys_array:
1642 fail_sb_buffer:
1643         btrfs_stop_workers(&fs_info->fixup_workers);
1644         btrfs_stop_workers(&fs_info->workers);
1645         btrfs_stop_workers(&fs_info->endio_workers);
1646         btrfs_stop_workers(&fs_info->endio_write_workers);
1647         btrfs_stop_workers(&fs_info->submit_workers);
1648 fail_iput:
1649         iput(fs_info->btree_inode);
1650 fail:
1651         btrfs_close_devices(fs_info->fs_devices);
1652         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1653
1654         kfree(extent_root);
1655         kfree(tree_root);
1656 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1657         bdi_destroy(&fs_info->bdi);
1658 #endif
1659         kfree(fs_info);
1660         return ERR_PTR(err);
1661 }
1662
1663 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1664 {
1665         char b[BDEVNAME_SIZE];
1666
1667         if (uptodate) {
1668                 set_buffer_uptodate(bh);
1669         } else {
1670                 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1671                         printk(KERN_WARNING "lost page write due to "
1672                                         "I/O error on %s\n",
1673                                        bdevname(bh->b_bdev, b));
1674                 }
1675                 /* note, we dont' set_buffer_write_io_error because we have
1676                  * our own ways of dealing with the IO errors
1677                  */
1678                 clear_buffer_uptodate(bh);
1679         }
1680         unlock_buffer(bh);
1681         put_bh(bh);
1682 }
1683
1684 int write_all_supers(struct btrfs_root *root)
1685 {
1686         struct list_head *cur;
1687         struct list_head *head = &root->fs_info->fs_devices->devices;
1688         struct btrfs_device *dev;
1689         struct btrfs_super_block *sb;
1690         struct btrfs_dev_item *dev_item;
1691         struct buffer_head *bh;
1692         int ret;
1693         int do_barriers;
1694         int max_errors;
1695         int total_errors = 0;
1696         u32 crc;
1697         u64 flags;
1698
1699         max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1700         do_barriers = !btrfs_test_opt(root, NOBARRIER);
1701
1702         sb = &root->fs_info->super_for_commit;
1703         dev_item = &sb->dev_item;
1704         list_for_each(cur, head) {
1705                 dev = list_entry(cur, struct btrfs_device, dev_list);
1706                 if (!dev->bdev) {
1707                         total_errors++;
1708                         continue;
1709                 }
1710                 if (!dev->in_fs_metadata)
1711                         continue;
1712
1713                 btrfs_set_stack_device_type(dev_item, dev->type);
1714                 btrfs_set_stack_device_id(dev_item, dev->devid);
1715                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
1716                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
1717                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
1718                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
1719                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
1720                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
1721                 flags = btrfs_super_flags(sb);
1722                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
1723
1724
1725                 crc = ~(u32)0;
1726                 crc = btrfs_csum_data(root, (char *)sb + BTRFS_CSUM_SIZE, crc,
1727                                       BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
1728                 btrfs_csum_final(crc, sb->csum);
1729
1730                 bh = __getblk(dev->bdev, BTRFS_SUPER_INFO_OFFSET / 4096,
1731                               BTRFS_SUPER_INFO_SIZE);
1732
1733                 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
1734                 dev->pending_io = bh;
1735
1736                 get_bh(bh);
1737                 set_buffer_uptodate(bh);
1738                 lock_buffer(bh);
1739                 bh->b_end_io = btrfs_end_buffer_write_sync;
1740
1741                 if (do_barriers && dev->barriers) {
1742                         ret = submit_bh(WRITE_BARRIER, bh);
1743                         if (ret == -EOPNOTSUPP) {
1744                                 printk("btrfs: disabling barriers on dev %s\n",
1745                                        dev->name);
1746                                 set_buffer_uptodate(bh);
1747                                 dev->barriers = 0;
1748                                 get_bh(bh);
1749                                 lock_buffer(bh);
1750                                 ret = submit_bh(WRITE, bh);
1751                         }
1752                 } else {
1753                         ret = submit_bh(WRITE, bh);
1754                 }
1755                 if (ret)
1756                         total_errors++;
1757         }
1758         if (total_errors > max_errors) {
1759                 printk("btrfs: %d errors while writing supers\n", total_errors);
1760                 BUG();
1761         }
1762         total_errors = 0;
1763
1764         list_for_each(cur, head) {
1765                 dev = list_entry(cur, struct btrfs_device, dev_list);
1766                 if (!dev->bdev)
1767                         continue;
1768                 if (!dev->in_fs_metadata)
1769                         continue;
1770
1771                 BUG_ON(!dev->pending_io);
1772                 bh = dev->pending_io;
1773                 wait_on_buffer(bh);
1774                 if (!buffer_uptodate(dev->pending_io)) {
1775                         if (do_barriers && dev->barriers) {
1776                                 printk("btrfs: disabling barriers on dev %s\n",
1777                                        dev->name);
1778                                 set_buffer_uptodate(bh);
1779                                 get_bh(bh);
1780                                 lock_buffer(bh);
1781                                 dev->barriers = 0;
1782                                 ret = submit_bh(WRITE, bh);
1783                                 BUG_ON(ret);
1784                                 wait_on_buffer(bh);
1785                                 if (!buffer_uptodate(bh))
1786                                         total_errors++;
1787                         } else {
1788                                 total_errors++;
1789                         }
1790
1791                 }
1792                 dev->pending_io = NULL;
1793                 brelse(bh);
1794         }
1795         if (total_errors > max_errors) {
1796                 printk("btrfs: %d errors while writing supers\n", total_errors);
1797                 BUG();
1798         }
1799         return 0;
1800 }
1801
1802 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
1803                       *root)
1804 {
1805         int ret;
1806
1807         ret = write_all_supers(root);
1808         return ret;
1809 }
1810
1811 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
1812 {
1813         radix_tree_delete(&fs_info->fs_roots_radix,
1814                           (unsigned long)root->root_key.objectid);
1815         if (root->in_sysfs)
1816                 btrfs_sysfs_del_root(root);
1817         if (root->inode)
1818                 iput(root->inode);
1819         if (root->node)
1820                 free_extent_buffer(root->node);
1821         if (root->commit_root)
1822                 free_extent_buffer(root->commit_root);
1823         if (root->name)
1824                 kfree(root->name);
1825         kfree(root);
1826         return 0;
1827 }
1828
1829 static int del_fs_roots(struct btrfs_fs_info *fs_info)
1830 {
1831         int ret;
1832         struct btrfs_root *gang[8];
1833         int i;
1834
1835         while(1) {
1836                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1837                                              (void **)gang, 0,
1838                                              ARRAY_SIZE(gang));
1839                 if (!ret)
1840                         break;
1841                 for (i = 0; i < ret; i++)
1842                         btrfs_free_fs_root(fs_info, gang[i]);
1843         }
1844         return 0;
1845 }
1846
1847 int close_ctree(struct btrfs_root *root)
1848 {
1849         int ret;
1850         struct btrfs_trans_handle *trans;
1851         struct btrfs_fs_info *fs_info = root->fs_info;
1852
1853         fs_info->closing = 1;
1854         smp_mb();
1855
1856         kthread_stop(root->fs_info->transaction_kthread);
1857         kthread_stop(root->fs_info->cleaner_kthread);
1858
1859         btrfs_clean_old_snapshots(root);
1860         trans = btrfs_start_transaction(root, 1);
1861         ret = btrfs_commit_transaction(trans, root);
1862         /* run commit again to  drop the original snapshot */
1863         trans = btrfs_start_transaction(root, 1);
1864         btrfs_commit_transaction(trans, root);
1865         ret = btrfs_write_and_wait_transaction(NULL, root);
1866         BUG_ON(ret);
1867
1868         write_ctree_super(NULL, root);
1869
1870         if (fs_info->delalloc_bytes) {
1871                 printk("btrfs: at unmount delalloc count %Lu\n",
1872                        fs_info->delalloc_bytes);
1873         }
1874         if (fs_info->total_ref_cache_size) {
1875                 printk("btrfs: at umount reference cache size %Lu\n",
1876                         fs_info->total_ref_cache_size);
1877         }
1878
1879         if (fs_info->extent_root->node)
1880                 free_extent_buffer(fs_info->extent_root->node);
1881
1882         if (fs_info->tree_root->node)
1883                 free_extent_buffer(fs_info->tree_root->node);
1884
1885         if (root->fs_info->chunk_root->node);
1886                 free_extent_buffer(root->fs_info->chunk_root->node);
1887
1888         if (root->fs_info->dev_root->node);
1889                 free_extent_buffer(root->fs_info->dev_root->node);
1890
1891         btrfs_free_block_groups(root->fs_info);
1892         fs_info->closing = 2;
1893         del_fs_roots(fs_info);
1894
1895         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1896
1897         truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
1898
1899         btrfs_stop_workers(&fs_info->fixup_workers);
1900         btrfs_stop_workers(&fs_info->workers);
1901         btrfs_stop_workers(&fs_info->endio_workers);
1902         btrfs_stop_workers(&fs_info->endio_write_workers);
1903         btrfs_stop_workers(&fs_info->submit_workers);
1904
1905         iput(fs_info->btree_inode);
1906 #if 0
1907         while(!list_empty(&fs_info->hashers)) {
1908                 struct btrfs_hasher *hasher;
1909                 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
1910                                     hashers);
1911                 list_del(&hasher->hashers);
1912                 crypto_free_hash(&fs_info->hash_tfm);
1913                 kfree(hasher);
1914         }
1915 #endif
1916         btrfs_close_devices(fs_info->fs_devices);
1917         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1918
1919 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1920         bdi_destroy(&fs_info->bdi);
1921 #endif
1922
1923         kfree(fs_info->extent_root);
1924         kfree(fs_info->tree_root);
1925         kfree(fs_info->chunk_root);
1926         kfree(fs_info->dev_root);
1927         return 0;
1928 }
1929
1930 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
1931 {
1932         int ret;
1933         struct inode *btree_inode = buf->first_page->mapping->host;
1934
1935         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
1936         if (!ret)
1937                 return ret;
1938
1939         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
1940                                     parent_transid);
1941         return !ret;
1942 }
1943
1944 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
1945 {
1946         struct inode *btree_inode = buf->first_page->mapping->host;
1947         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
1948                                           buf);
1949 }
1950
1951 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
1952 {
1953         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1954         u64 transid = btrfs_header_generation(buf);
1955         struct inode *btree_inode = root->fs_info->btree_inode;
1956
1957         WARN_ON(!btrfs_tree_locked(buf));
1958         if (transid != root->fs_info->generation) {
1959                 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
1960                         (unsigned long long)buf->start,
1961                         transid, root->fs_info->generation);
1962                 WARN_ON(1);
1963         }
1964         set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
1965 }
1966
1967 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
1968 {
1969         /*
1970          * looks as though older kernels can get into trouble with
1971          * this code, they end up stuck in balance_dirty_pages forever
1972          */
1973         struct extent_io_tree *tree;
1974         u64 num_dirty;
1975         u64 start = 0;
1976         unsigned long thresh = 96 * 1024 * 1024;
1977         tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
1978
1979         if (current_is_pdflush() || current->flags & PF_MEMALLOC)
1980                 return;
1981
1982         num_dirty = count_range_bits(tree, &start, (u64)-1,
1983                                      thresh, EXTENT_DIRTY);
1984         if (num_dirty > thresh) {
1985                 balance_dirty_pages_ratelimited_nr(
1986                                    root->fs_info->btree_inode->i_mapping, 1);
1987         }
1988         return;
1989 }
1990
1991 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
1992 {
1993         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1994         int ret;
1995         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1996         if (ret == 0) {
1997                 buf->flags |= EXTENT_UPTODATE;
1998         }
1999         return ret;
2000 }
2001
2002 int btree_lock_page_hook(struct page *page)
2003 {
2004         struct inode *inode = page->mapping->host;
2005         struct btrfs_root *root = BTRFS_I(inode)->root;
2006         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2007         struct extent_buffer *eb;
2008         unsigned long len;
2009         u64 bytenr = page_offset(page);
2010
2011         if (page->private == EXTENT_PAGE_PRIVATE)
2012                 goto out;
2013
2014         len = page->private >> 2;
2015         eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2016         if (!eb)
2017                 goto out;
2018
2019         btrfs_tree_lock(eb);
2020         spin_lock(&root->fs_info->hash_lock);
2021         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2022         spin_unlock(&root->fs_info->hash_lock);
2023         btrfs_tree_unlock(eb);
2024         free_extent_buffer(eb);
2025 out:
2026         lock_page(page);
2027         return 0;
2028 }
2029
2030 static struct extent_io_ops btree_extent_io_ops = {
2031         .write_cache_pages_lock_hook = btree_lock_page_hook,
2032         .writepage_io_hook = btree_writepage_io_hook,
2033         .readpage_end_io_hook = btree_readpage_end_io_hook,
2034         .submit_bio_hook = btree_submit_bio_hook,
2035         /* note we're sharing with inode.c for the merge bio hook */
2036         .merge_bio_hook = btrfs_merge_bio_hook,
2037 };