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/fs.h>
  20 #include <linux/blkdev.h>
  21 #include <linux/scatterlist.h>
  22 #include <linux/swap.h>
  23 #include <linux/radix-tree.h>
  24 #include <linux/writeback.h>
  25 #include <linux/buffer_head.h>
  26 #include <linux/workqueue.h>
  27 #include <linux/kthread.h>
  28 #include <linux/freezer.h>
  29 #include <linux/crc32c.h>
  30 #include "compat.h"
  31 #include "ctree.h"
  32 #include "disk-io.h"
  33 #include "transaction.h"
  34 #include "btrfs_inode.h"
  35 #include "volumes.h"
  36 #include "print-tree.h"
  37 #include "async-thread.h"
  38 #include "locking.h"
  39 #include "tree-log.h"
  40 #include "free-space-cache.h"
  41
  42 static struct extent_io_ops btree_extent_io_ops;
  43 static void end_workqueue_fn(struct btrfs_work *work);
  44 static void free_fs_root(struct btrfs_root *root);
  45
  46 static atomic_t btrfs_bdi_num = ATOMIC_INIT(0);
  47
  48 /*
  49  * end_io_wq structs are used to do processing in task context when an IO is
  50  * complete.  This is used during reads to verify checksums, and it is used
  51  * by writes to insert metadata for new file extents after IO is complete.
  52  */
  53 struct end_io_wq {
  54         struct bio *bio;
  55         bio_end_io_t *end_io;
  56         void *private;
  57         struct btrfs_fs_info *info;
  58         int error;
  59         int metadata;
  60         struct list_head list;
  61         struct btrfs_work work;
  62 };
  63
  64 /*
  65  * async submit bios are used to offload expensive checksumming
  66  * onto the worker threads.  They checksum file and metadata bios
  67  * just before they are sent down the IO stack.
  68  */
  69 struct async_submit_bio {
  70         struct inode *inode;
  71         struct bio *bio;
  72         struct list_head list;
  73         extent_submit_bio_hook_t *submit_bio_start;
  74         extent_submit_bio_hook_t *submit_bio_done;
  75         int rw;
  76         int mirror_num;
  77         unsigned long bio_flags;
  78         struct btrfs_work work;
  79 };
  80
  81 /* These are used to set the lockdep class on the extent buffer locks.
  82  * The class is set by the readpage_end_io_hook after the buffer has
  83  * passed csum validation but before the pages are unlocked.
  84  *
  85  * The lockdep class is also set by btrfs_init_new_buffer on freshly
  86  * allocated blocks.
  87  *
  88  * The class is based on the level in the tree block, which allows lockdep
  89  * to know that lower nodes nest inside the locks of higher nodes.
  90  *
  91  * We also add a check to make sure the highest level of the tree is
  92  * the same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this
  93  * code needs update as well.
  94  */
  95 #ifdef CONFIG_DEBUG_LOCK_ALLOC
  96 # if BTRFS_MAX_LEVEL != 8
  97 #  error
  98 # endif
  99 static struct lock_class_key btrfs_eb_class[BTRFS_MAX_LEVEL + 1];
 100 static const char *btrfs_eb_name[BTRFS_MAX_LEVEL + 1] = {
 101         /* leaf */
 102         "btrfs-extent-00",
 103         "btrfs-extent-01",
 104         "btrfs-extent-02",
 105         "btrfs-extent-03",
 106         "btrfs-extent-04",
 107         "btrfs-extent-05",
 108         "btrfs-extent-06",
 109         "btrfs-extent-07",
 110         /* highest possible level */
 111         "btrfs-extent-08",
 112 };
 113 #endif
 114
 115 /*
 116  * extents on the btree inode are pretty simple, there's one extent
 117  * that covers the entire device
 118  */
 119 static struct extent_map *btree_get_extent(struct inode *inode,
 120                 struct page *page, size_t page_offset, u64 start, u64 len,
 121                 int create)
 122 {
 123         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 124         struct extent_map *em;
 125         int ret;
 126
 127         read_lock(&em_tree->lock);
 128         em = lookup_extent_mapping(em_tree, start, len);
 129         if (em) {
 130                 em->bdev =
 131                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 132                 read_unlock(&em_tree->lock);
 133                 goto out;
 134         }
 135         read_unlock(&em_tree->lock);
 136
 137         em = alloc_extent_map(GFP_NOFS);
 138         if (!em) {
 139                 em = ERR_PTR(-ENOMEM);
 140                 goto out;
 141         }
 142         em->start = 0;
 143         em->len = (u64)-1;
 144         em->block_len = (u64)-1;
 145         em->block_start = 0;
 146         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 147
 148         write_lock(&em_tree->lock);
 149         ret = add_extent_mapping(em_tree, em);
 150         if (ret == -EEXIST) {
 151                 u64 failed_start = em->start;
 152                 u64 failed_len = em->len;
 153
 154                 free_extent_map(em);
 155                 em = lookup_extent_mapping(em_tree, start, len);
 156                 if (em) {
 157                         ret = 0;
 158                 } else {
 159                         em = lookup_extent_mapping(em_tree, failed_start,
 160                                                    failed_len);
 161                         ret = -EIO;
 162                 }
 163         } else if (ret) {
 164                 free_extent_map(em);
 165                 em = NULL;
 166         }
 167         write_unlock(&em_tree->lock);
 168
 169         if (ret)
 170                 em = ERR_PTR(ret);
 171 out:
 172         return em;
 173 }
 174
 175 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
 176 {
 177         return crc32c(seed, data, len);
 178 }
 179
 180 void btrfs_csum_final(u32 crc, char *result)
 181 {
 182         *(__le32 *)result = ~cpu_to_le32(crc);
 183 }
 184
 185 /*
 186  * compute the csum for a btree block, and either verify it or write it
 187  * into the csum field of the block.
 188  */
 189 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 190                            int verify)
 191 {
 192         u16 csum_size =
 193                 btrfs_super_csum_size(&root->fs_info->super_copy);
 194         char *result = NULL;
 195         unsigned long len;
 196         unsigned long cur_len;
 197         unsigned long offset = BTRFS_CSUM_SIZE;
 198         char *map_token = NULL;
 199         char *kaddr;
 200         unsigned long map_start;
 201         unsigned long map_len;
 202         int err;
 203         u32 crc = ~(u32)0;
 204         unsigned long inline_result;
 205
 206         len = buf->len - offset;
 207         while (len > 0) {
 208                 err = map_private_extent_buffer(buf, offset, 32,
 209                                         &map_token, &kaddr,
 210                                         &map_start, &map_len, KM_USER0);
 211                 if (err)
 212                         return 1;
 213                 cur_len = min(len, map_len - (offset - map_start));
 214                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
 215                                       crc, cur_len);
 216                 len -= cur_len;
 217                 offset += cur_len;
 218                 unmap_extent_buffer(buf, map_token, KM_USER0);
 219         }
 220         if (csum_size > sizeof(inline_result)) {
 221                 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
 222                 if (!result)
 223                         return 1;
 224         } else {
 225                 result = (char *)&inline_result;
 226         }
 227
 228         btrfs_csum_final(crc, result);
 229
 230         if (verify) {
 231                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
 232                         u32 val;
 233                         u32 found = 0;
 234                         memcpy(&found, result, csum_size);
 235
 236                         read_extent_buffer(buf, &val, 0, csum_size);
 237                         if (printk_ratelimit()) {
 238                                 printk(KERN_INFO "btrfs: %s checksum verify "
 239                                        "failed on %llu wanted %X found %X "
 240                                        "level %d\n",
 241                                        root->fs_info->sb->s_id,
 242                                        (unsigned long long)buf->start, val, found,
 243                                        btrfs_header_level(buf));
 244                         }
 245                         if (result != (char *)&inline_result)
 246                                 kfree(result);
 247                         return 1;
 248                 }
 249         } else {
 250                 write_extent_buffer(buf, result, 0, csum_size);
 251         }
 252         if (result != (char *)&inline_result)
 253                 kfree(result);
 254         return 0;
 255 }
 256
 257 /*
 258  * we can't consider a given block up to date unless the transid of the
 259  * block matches the transid in the parent node's pointer.  This is how we
 260  * detect blocks that either didn't get written at all or got written
 261  * in the wrong place.
 262  */
 263 static int verify_parent_transid(struct extent_io_tree *io_tree,
 264                                  struct extent_buffer *eb, u64 parent_transid)
 265 {
 266         int ret;
 267
 268         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 269                 return 0;
 270
 271         lock_extent(io_tree, eb->start, eb->start + eb->len - 1, GFP_NOFS);
 272         if (extent_buffer_uptodate(io_tree, eb) &&
 273             btrfs_header_generation(eb) == parent_transid) {
 274                 ret = 0;
 275                 goto out;
 276         }
 277         if (printk_ratelimit()) {
 278                 printk("parent transid verify failed on %llu wanted %llu "
 279                        "found %llu\n",
 280                        (unsigned long long)eb->start,
 281                        (unsigned long long)parent_transid,
 282                        (unsigned long long)btrfs_header_generation(eb));
 283         }
 284         ret = 1;
 285         clear_extent_buffer_uptodate(io_tree, eb);
 286 out:
 287         unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
 288                       GFP_NOFS);
 289         return ret;
 290 }
 291
 292 /*
 293  * helper to read a given tree block, doing retries as required when
 294  * the checksums don't match and we have alternate mirrors to try.
 295  */
 296 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 297                                           struct extent_buffer *eb,
 298                                           u64 start, u64 parent_transid)
 299 {
 300         struct extent_io_tree *io_tree;
 301         int ret;
 302         int num_copies = 0;
 303         int mirror_num = 0;
 304
 305         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 306         while (1) {
 307                 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
 308                                                btree_get_extent, mirror_num);
 309                 if (!ret &&
 310                     !verify_parent_transid(io_tree, eb, parent_transid))
 311                         return ret;
 312
 313                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
 314                                               eb->start, eb->len);
 315                 if (num_copies == 1)
 316                         return ret;
 317
 318                 mirror_num++;
 319                 if (mirror_num > num_copies)
 320                         return ret;
 321         }
 322         return -EIO;
 323 }
 324
 325 /*
 326  * checksum a dirty tree block before IO.  This has extra checks to make sure
 327  * we only fill in the checksum field in the first page of a multi-page block
 328  */
 329
 330 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 331 {
 332         struct extent_io_tree *tree;
 333         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
 334         u64 found_start;
 335         int found_level;
 336         unsigned long len;
 337         struct extent_buffer *eb;
 338         int ret;
 339
 340         tree = &BTRFS_I(page->mapping->host)->io_tree;
 341
 342         if (page->private == EXTENT_PAGE_PRIVATE)
 343                 goto out;
 344         if (!page->private)
 345                 goto out;
 346         len = page->private >> 2;
 347         WARN_ON(len == 0);
 348
 349         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 350         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
 351                                              btrfs_header_generation(eb));
 352         BUG_ON(ret);
 353         found_start = btrfs_header_bytenr(eb);
 354         if (found_start != start) {
 355                 WARN_ON(1);
 356                 goto err;
 357         }
 358         if (eb->first_page != page) {
 359                 WARN_ON(1);
 360                 goto err;
 361         }
 362         if (!PageUptodate(page)) {
 363                 WARN_ON(1);
 364                 goto err;
 365         }
 366         found_level = btrfs_header_level(eb);
 367
 368         csum_tree_block(root, eb, 0);
 369 err:
 370         free_extent_buffer(eb);
 371 out:
 372         return 0;
 373 }
 374
 375 static int check_tree_block_fsid(struct btrfs_root *root,
 376                                  struct extent_buffer *eb)
 377 {
 378         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
 379         u8 fsid[BTRFS_UUID_SIZE];
 380         int ret = 1;
 381
 382         read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
 383                            BTRFS_FSID_SIZE);
 384         while (fs_devices) {
 385                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
 386                         ret = 0;
 387                         break;
 388                 }
 389                 fs_devices = fs_devices->seed;
 390         }
 391         return ret;
 392 }
 393
 394 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 395 void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level)
 396 {
 397         lockdep_set_class_and_name(&eb->lock,
 398                            &btrfs_eb_class[level],
 399                            btrfs_eb_name[level]);
 400 }
 401 #endif
 402
 403 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 404                                struct extent_state *state)
 405 {
 406         struct extent_io_tree *tree;
 407         u64 found_start;
 408         int found_level;
 409         unsigned long len;
 410         struct extent_buffer *eb;
 411         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 412         int ret = 0;
 413
 414         tree = &BTRFS_I(page->mapping->host)->io_tree;
 415         if (page->private == EXTENT_PAGE_PRIVATE)
 416                 goto out;
 417         if (!page->private)
 418                 goto out;
 419
 420         len = page->private >> 2;
 421         WARN_ON(len == 0);
 422
 423         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 424
 425         found_start = btrfs_header_bytenr(eb);
 426         if (found_start != start) {
 427                 if (printk_ratelimit()) {
 428                         printk(KERN_INFO "btrfs bad tree block start "
 429                                "%llu %llu\n",
 430                                (unsigned long long)found_start,
 431                                (unsigned long long)eb->start);
 432                 }
 433                 ret = -EIO;
 434                 goto err;
 435         }
 436         if (eb->first_page != page) {
 437                 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
 438                        eb->first_page->index, page->index);
 439                 WARN_ON(1);
 440                 ret = -EIO;
 441                 goto err;
 442         }
 443         if (check_tree_block_fsid(root, eb)) {
 444                 if (printk_ratelimit()) {
 445                         printk(KERN_INFO "btrfs bad fsid on block %llu\n",
 446                                (unsigned long long)eb->start);
 447                 }
 448                 ret = -EIO;
 449                 goto err;
 450         }
 451         found_level = btrfs_header_level(eb);
 452
 453         btrfs_set_buffer_lockdep_class(eb, found_level);
 454
 455         ret = csum_tree_block(root, eb, 1);
 456         if (ret)
 457                 ret = -EIO;
 458
 459         end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
 460         end = eb->start + end - 1;
 461 err:
 462         free_extent_buffer(eb);
 463 out:
 464         return ret;
 465 }
 466
 467 static void end_workqueue_bio(struct bio *bio, int err)
 468 {
 469         struct end_io_wq *end_io_wq = bio->bi_private;
 470         struct btrfs_fs_info *fs_info;
 471
 472         fs_info = end_io_wq->info;
 473         end_io_wq->error = err;
 474         end_io_wq->work.func = end_workqueue_fn;
 475         end_io_wq->work.flags = 0;
 476
 477         if (bio->bi_rw & (1 << BIO_RW)) {
 478                 if (end_io_wq->metadata)
 479                         btrfs_queue_worker(&fs_info->endio_meta_write_workers,
 480                                            &end_io_wq->work);
 481                 else
 482                         btrfs_queue_worker(&fs_info->endio_write_workers,
 483                                            &end_io_wq->work);
 484         } else {
 485                 if (end_io_wq->metadata)
 486                         btrfs_queue_worker(&fs_info->endio_meta_workers,
 487                                            &end_io_wq->work);
 488                 else
 489                         btrfs_queue_worker(&fs_info->endio_workers,
 490                                            &end_io_wq->work);
 491         }
 492 }
 493
 494 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 495                         int metadata)
 496 {
 497         struct end_io_wq *end_io_wq;
 498         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
 499         if (!end_io_wq)
 500                 return -ENOMEM;
 501
 502         end_io_wq->private = bio->bi_private;
 503         end_io_wq->end_io = bio->bi_end_io;
 504         end_io_wq->info = info;
 505         end_io_wq->error = 0;
 506         end_io_wq->bio = bio;
 507         end_io_wq->metadata = metadata;
 508
 509         bio->bi_private = end_io_wq;
 510         bio->bi_end_io = end_workqueue_bio;
 511         return 0;
 512 }
 513
 514 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
 515 {
 516         unsigned long limit = min_t(unsigned long,
 517                                     info->workers.max_workers,
 518                                     info->fs_devices->open_devices);
 519         return 256 * limit;
 520 }
 521
 522 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
 523 {
 524         return atomic_read(&info->nr_async_bios) >
 525                 btrfs_async_submit_limit(info);
 526 }
 527
 528 static void run_one_async_start(struct btrfs_work *work)
 529 {
 530         struct btrfs_fs_info *fs_info;
 531         struct async_submit_bio *async;
 532
 533         async = container_of(work, struct  async_submit_bio, work);
 534         fs_info = BTRFS_I(async->inode)->root->fs_info;
 535         async->submit_bio_start(async->inode, async->rw, async->bio,
 536                                async->mirror_num, async->bio_flags);
 537 }
 538
 539 static void run_one_async_done(struct btrfs_work *work)
 540 {
 541         struct btrfs_fs_info *fs_info;
 542         struct async_submit_bio *async;
 543         int limit;
 544
 545         async = container_of(work, struct  async_submit_bio, work);
 546         fs_info = BTRFS_I(async->inode)->root->fs_info;
 547
 548         limit = btrfs_async_submit_limit(fs_info);
 549         limit = limit * 2 / 3;
 550
 551         atomic_dec(&fs_info->nr_async_submits);
 552
 553         if (atomic_read(&fs_info->nr_async_submits) < limit &&
 554             waitqueue_active(&fs_info->async_submit_wait))
 555                 wake_up(&fs_info->async_submit_wait);
 556
 557         async->submit_bio_done(async->inode, async->rw, async->bio,
 558                                async->mirror_num, async->bio_flags);
 559 }
 560
 561 static void run_one_async_free(struct btrfs_work *work)
 562 {
 563         struct async_submit_bio *async;
 564
 565         async = container_of(work, struct  async_submit_bio, work);
 566         kfree(async);
 567 }
 568
 569 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 570                         int rw, struct bio *bio, int mirror_num,
 571                         unsigned long bio_flags,
 572                         extent_submit_bio_hook_t *submit_bio_start,
 573                         extent_submit_bio_hook_t *submit_bio_done)
 574 {
 575         struct async_submit_bio *async;
 576
 577         async = kmalloc(sizeof(*async), GFP_NOFS);
 578         if (!async)
 579                 return -ENOMEM;
 580
 581         async->inode = inode;
 582         async->rw = rw;
 583         async->bio = bio;
 584         async->mirror_num = mirror_num;
 585         async->submit_bio_start = submit_bio_start;
 586         async->submit_bio_done = submit_bio_done;
 587
 588         async->work.func = run_one_async_start;
 589         async->work.ordered_func = run_one_async_done;
 590         async->work.ordered_free = run_one_async_free;
 591
 592         async->work.flags = 0;
 593         async->bio_flags = bio_flags;
 594
 595         atomic_inc(&fs_info->nr_async_submits);
 596
 597         if (rw & (1 << BIO_RW_SYNCIO))
 598                 btrfs_set_work_high_prio(&async->work);
 599
 600         btrfs_queue_worker(&fs_info->workers, &async->work);
 601
 602         while (atomic_read(&fs_info->async_submit_draining) &&
 603               atomic_read(&fs_info->nr_async_submits)) {
 604                 wait_event(fs_info->async_submit_wait,
 605                            (atomic_read(&fs_info->nr_async_submits) == 0));
 606         }
 607
 608         return 0;
 609 }
 610
 611 static int btree_csum_one_bio(struct bio *bio)
 612 {
 613         struct bio_vec *bvec = bio->bi_io_vec;
 614         int bio_index = 0;
 615         struct btrfs_root *root;
 616
 617         WARN_ON(bio->bi_vcnt <= 0);
 618         while (bio_index < bio->bi_vcnt) {
 619                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 620                 csum_dirty_buffer(root, bvec->bv_page);
 621                 bio_index++;
 622                 bvec++;
 623         }
 624         return 0;
 625 }
 626
 627 static int __btree_submit_bio_start(struct inode *inode, int rw,
 628                                     struct bio *bio, int mirror_num,
 629                                     unsigned long bio_flags)
 630 {
 631         /*
 632          * when we're called for a write, we're already in the async
 633          * submission context.  Just jump into btrfs_map_bio
 634          */
 635         btree_csum_one_bio(bio);
 636         return 0;
 637 }
 638
 639 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
 640                                  int mirror_num, unsigned long bio_flags)
 641 {
 642         /*
 643          * when we're called for a write, we're already in the async
 644          * submission context.  Just jump into btrfs_map_bio
 645          */
 646         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
 647 }
 648
 649 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 650                                  int mirror_num, unsigned long bio_flags)
 651 {
 652         int ret;
 653
 654         ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
 655                                           bio, 1);
 656         BUG_ON(ret);
 657
 658         if (!(rw & (1 << BIO_RW))) {
 659                 /*
 660                  * called for a read, do the setup so that checksum validation
 661                  * can happen in the async kernel threads
 662                  */
 663                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 664                                      mirror_num, 0);
 665         }
 666
 667         /*
 668          * kthread helpers are used to submit writes so that checksumming
 669          * can happen in parallel across all CPUs
 670          */
 671         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 672                                    inode, rw, bio, mirror_num, 0,
 673                                    __btree_submit_bio_start,
 674                                    __btree_submit_bio_done);
 675 }
 676
 677 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 678 {
 679         struct extent_io_tree *tree;
 680         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 681         struct extent_buffer *eb;
 682         int was_dirty;
 683
 684         tree = &BTRFS_I(page->mapping->host)->io_tree;
 685         if (!(current->flags & PF_MEMALLOC)) {
 686                 return extent_write_full_page(tree, page,
 687                                               btree_get_extent, wbc);
 688         }
 689
 690         redirty_page_for_writepage(wbc, page);
 691         eb = btrfs_find_tree_block(root, page_offset(page),
 692                                       PAGE_CACHE_SIZE);
 693         WARN_ON(!eb);
 694
 695         was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
 696         if (!was_dirty) {
 697                 spin_lock(&root->fs_info->delalloc_lock);
 698                 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
 699                 spin_unlock(&root->fs_info->delalloc_lock);
 700         }
 701         free_extent_buffer(eb);
 702
 703         unlock_page(page);
 704         return 0;
 705 }
 706
 707 static int btree_writepages(struct address_space *mapping,
 708                             struct writeback_control *wbc)
 709 {
 710         struct extent_io_tree *tree;
 711         tree = &BTRFS_I(mapping->host)->io_tree;
 712         if (wbc->sync_mode == WB_SYNC_NONE) {
 713                 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
 714                 u64 num_dirty;
 715                 unsigned long thresh = 32 * 1024 * 1024;
 716
 717                 if (wbc->for_kupdate)
 718                         return 0;
 719
 720                 /* this is a bit racy, but that's ok */
 721                 num_dirty = root->fs_info->dirty_metadata_bytes;
 722                 if (num_dirty < thresh)
 723                         return 0;
 724         }
 725         return extent_writepages(tree, mapping, btree_get_extent, wbc);
 726 }
 727
 728 static int btree_readpage(struct file *file, struct page *page)
 729 {
 730         struct extent_io_tree *tree;
 731         tree = &BTRFS_I(page->mapping->host)->io_tree;
 732         return extent_read_full_page(tree, page, btree_get_extent);
 733 }
 734
 735 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 736 {
 737         struct extent_io_tree *tree;
 738         struct extent_map_tree *map;
 739         int ret;
 740
 741         if (PageWriteback(page) || PageDirty(page))
 742                 return 0;
 743
 744         tree = &BTRFS_I(page->mapping->host)->io_tree;
 745         map = &BTRFS_I(page->mapping->host)->extent_tree;
 746
 747         ret = try_release_extent_state(map, tree, page, gfp_flags);
 748         if (!ret)
 749                 return 0;
 750
 751         ret = try_release_extent_buffer(tree, page);
 752         if (ret == 1) {
 753                 ClearPagePrivate(page);
 754                 set_page_private(page, 0);
 755                 page_cache_release(page);
 756         }
 757
 758         return ret;
 759 }
 760
 761 static void btree_invalidatepage(struct page *page, unsigned long offset)
 762 {
 763         struct extent_io_tree *tree;
 764         tree = &BTRFS_I(page->mapping->host)->io_tree;
 765         extent_invalidatepage(tree, page, offset);
 766         btree_releasepage(page, GFP_NOFS);
 767         if (PagePrivate(page)) {
 768                 printk(KERN_WARNING "btrfs warning page private not zero "
 769                        "on page %llu\n", (unsigned long long)page_offset(page));
 770                 ClearPagePrivate(page);
 771                 set_page_private(page, 0);
 772                 page_cache_release(page);
 773         }
 774 }
 775
 776 static const struct address_space_operations btree_aops = {
 777         .readpage       = btree_readpage,
 778         .writepage      = btree_writepage,
 779         .writepages     = btree_writepages,
 780         .releasepage    = btree_releasepage,
 781         .invalidatepage = btree_invalidatepage,
 782         .sync_page      = block_sync_page,
 783 };
 784
 785 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
 786                          u64 parent_transid)
 787 {
 788         struct extent_buffer *buf = NULL;
 789         struct inode *btree_inode = root->fs_info->btree_inode;
 790         int ret = 0;
 791
 792         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 793         if (!buf)
 794                 return 0;
 795         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
 796                                  buf, 0, 0, btree_get_extent, 0);
 797         free_extent_buffer(buf);
 798         return ret;
 799 }
 800
 801 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
 802                                             u64 bytenr, u32 blocksize)
 803 {
 804         struct inode *btree_inode = root->fs_info->btree_inode;
 805         struct extent_buffer *eb;
 806         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 807                                 bytenr, blocksize, GFP_NOFS);
 808         return eb;
 809 }
 810
 811 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
 812                                                  u64 bytenr, u32 blocksize)
 813 {
 814         struct inode *btree_inode = root->fs_info->btree_inode;
 815         struct extent_buffer *eb;
 816
 817         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 818                                  bytenr, blocksize, NULL, GFP_NOFS);
 819         return eb;
 820 }
 821
 822
 823 int btrfs_write_tree_block(struct extent_buffer *buf)
 824 {
 825         return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
 826                                         buf->start + buf->len - 1);
 827 }
 828
 829 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
 830 {
 831         return filemap_fdatawait_range(buf->first_page->mapping,
 832                                        buf->start, buf->start + buf->len - 1);
 833 }
 834
 835 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
 836                                       u32 blocksize, u64 parent_transid)
 837 {
 838         struct extent_buffer *buf = NULL;
 839         struct inode *btree_inode = root->fs_info->btree_inode;
 840         struct extent_io_tree *io_tree;
 841         int ret;
 842
 843         io_tree = &BTRFS_I(btree_inode)->io_tree;
 844
 845         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 846         if (!buf)
 847                 return NULL;
 848
 849         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
 850
 851         if (ret == 0)
 852                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
 853         return buf;
 854
 855 }
 856
 857 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 858                      struct extent_buffer *buf)
 859 {
 860         struct inode *btree_inode = root->fs_info->btree_inode;
 861         if (btrfs_header_generation(buf) ==
 862             root->fs_info->running_transaction->transid) {
 863                 btrfs_assert_tree_locked(buf);
 864
 865                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
 866                         spin_lock(&root->fs_info->delalloc_lock);
 867                         if (root->fs_info->dirty_metadata_bytes >= buf->len)
 868                                 root->fs_info->dirty_metadata_bytes -= buf->len;
 869                         else
 870                                 WARN_ON(1);
 871                         spin_unlock(&root->fs_info->delalloc_lock);
 872                 }
 873
 874                 /* ugh, clear_extent_buffer_dirty needs to lock the page */
 875                 btrfs_set_lock_blocking(buf);
 876                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
 877                                           buf);
 878         }
 879         return 0;
 880 }
 881
 882 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
 883                         u32 stripesize, struct btrfs_root *root,
 884                         struct btrfs_fs_info *fs_info,
 885                         u64 objectid)
 886 {
 887         root->node = NULL;
 888         root->commit_root = NULL;
 889         root->sectorsize = sectorsize;
 890         root->nodesize = nodesize;
 891         root->leafsize = leafsize;
 892         root->stripesize = stripesize;
 893         root->ref_cows = 0;
 894         root->track_dirty = 0;
 895
 896         root->fs_info = fs_info;
 897         root->objectid = objectid;
 898         root->last_trans = 0;
 899         root->highest_objectid = 0;
 900         root->name = NULL;
 901         root->in_sysfs = 0;
 902         root->inode_tree.rb_node = NULL;
 903
 904         INIT_LIST_HEAD(&root->dirty_list);
 905         INIT_LIST_HEAD(&root->orphan_list);
 906         INIT_LIST_HEAD(&root->root_list);
 907         spin_lock_init(&root->node_lock);
 908         spin_lock_init(&root->list_lock);
 909         spin_lock_init(&root->inode_lock);
 910         mutex_init(&root->objectid_mutex);
 911         mutex_init(&root->log_mutex);
 912         init_waitqueue_head(&root->log_writer_wait);
 913         init_waitqueue_head(&root->log_commit_wait[0]);
 914         init_waitqueue_head(&root->log_commit_wait[1]);
 915         atomic_set(&root->log_commit[0], 0);
 916         atomic_set(&root->log_commit[1], 0);
 917         atomic_set(&root->log_writers, 0);
 918         root->log_batch = 0;
 919         root->log_transid = 0;
 920         extent_io_tree_init(&root->dirty_log_pages,
 921                              fs_info->btree_inode->i_mapping, GFP_NOFS);
 922
 923         memset(&root->root_key, 0, sizeof(root->root_key));
 924         memset(&root->root_item, 0, sizeof(root->root_item));
 925         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
 926         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
 927         root->defrag_trans_start = fs_info->generation;
 928         init_completion(&root->kobj_unregister);
 929         root->defrag_running = 0;
 930         root->defrag_level = 0;
 931         root->root_key.objectid = objectid;
 932         root->anon_super.s_root = NULL;
 933         root->anon_super.s_dev = 0;
 934         INIT_LIST_HEAD(&root->anon_super.s_list);
 935         INIT_LIST_HEAD(&root->anon_super.s_instances);
 936         init_rwsem(&root->anon_super.s_umount);
 937
 938         return 0;
 939 }
 940
 941 static int find_and_setup_root(struct btrfs_root *tree_root,
 942                                struct btrfs_fs_info *fs_info,
 943                                u64 objectid,
 944                                struct btrfs_root *root)
 945 {
 946         int ret;
 947         u32 blocksize;
 948         u64 generation;
 949
 950         __setup_root(tree_root->nodesize, tree_root->leafsize,
 951                      tree_root->sectorsize, tree_root->stripesize,
 952                      root, fs_info, objectid);
 953         ret = btrfs_find_last_root(tree_root, objectid,
 954                                    &root->root_item, &root->root_key);
 955         if (ret > 0)
 956                 return -ENOENT;
 957         BUG_ON(ret);
 958
 959         generation = btrfs_root_generation(&root->root_item);
 960         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 961         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 962                                      blocksize, generation);
 963         BUG_ON(!root->node);
 964         root->commit_root = btrfs_root_node(root);
 965         return 0;
 966 }
 967
 968 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
 969                              struct btrfs_fs_info *fs_info)
 970 {
 971         struct extent_buffer *eb;
 972         struct btrfs_root *log_root_tree = fs_info->log_root_tree;
 973         u64 start = 0;
 974         u64 end = 0;
 975         int ret;
 976
 977         if (!log_root_tree)
 978                 return 0;
 979
 980         while (1) {
 981                 ret = find_first_extent_bit(&log_root_tree->dirty_log_pages,
 982                                     0, &start, &end, EXTENT_DIRTY);
 983                 if (ret)
 984                         break;
 985
 986                 clear_extent_dirty(&log_root_tree->dirty_log_pages,
 987                                    start, end, GFP_NOFS);
 988         }
 989         eb = fs_info->log_root_tree->node;
 990
 991         WARN_ON(btrfs_header_level(eb) != 0);
 992         WARN_ON(btrfs_header_nritems(eb) != 0);
 993
 994         ret = btrfs_free_reserved_extent(fs_info->tree_root,
 995                                 eb->start, eb->len);
 996         BUG_ON(ret);
 997
 998         free_extent_buffer(eb);
 999         kfree(fs_info->log_root_tree);
1000         fs_info->log_root_tree = NULL;
1001         return 0;
1002 }
1003
1004 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1005                                          struct btrfs_fs_info *fs_info)
1006 {
1007         struct btrfs_root *root;
1008         struct btrfs_root *tree_root = fs_info->tree_root;
1009         struct extent_buffer *leaf;
1010
1011         root = kzalloc(sizeof(*root), GFP_NOFS);
1012         if (!root)
1013                 return ERR_PTR(-ENOMEM);
1014
1015         __setup_root(tree_root->nodesize, tree_root->leafsize,
1016                      tree_root->sectorsize, tree_root->stripesize,
1017                      root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1018
1019         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1020         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1021         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1022         /*
1023          * log trees do not get reference counted because they go away
1024          * before a real commit is actually done.  They do store pointers
1025          * to file data extents, and those reference counts still get
1026          * updated (along with back refs to the log tree).
1027          */
1028         root->ref_cows = 0;
1029
1030         leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1031                                       BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
1032         if (IS_ERR(leaf)) {
1033                 kfree(root);
1034                 return ERR_CAST(leaf);
1035         }
1036
1037         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1038         btrfs_set_header_bytenr(leaf, leaf->start);
1039         btrfs_set_header_generation(leaf, trans->transid);
1040         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1041         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1042         root->node = leaf;
1043
1044         write_extent_buffer(root->node, root->fs_info->fsid,
1045                             (unsigned long)btrfs_header_fsid(root->node),
1046                             BTRFS_FSID_SIZE);
1047         btrfs_mark_buffer_dirty(root->node);
1048         btrfs_tree_unlock(root->node);
1049         return root;
1050 }
1051
1052 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1053                              struct btrfs_fs_info *fs_info)
1054 {
1055         struct btrfs_root *log_root;
1056
1057         log_root = alloc_log_tree(trans, fs_info);
1058         if (IS_ERR(log_root))
1059                 return PTR_ERR(log_root);
1060         WARN_ON(fs_info->log_root_tree);
1061         fs_info->log_root_tree = log_root;
1062         return 0;
1063 }
1064
1065 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1066                        struct btrfs_root *root)
1067 {
1068         struct btrfs_root *log_root;
1069         struct btrfs_inode_item *inode_item;
1070
1071         log_root = alloc_log_tree(trans, root->fs_info);
1072         if (IS_ERR(log_root))
1073                 return PTR_ERR(log_root);
1074
1075         log_root->last_trans = trans->transid;
1076         log_root->root_key.offset = root->root_key.objectid;
1077
1078         inode_item = &log_root->root_item.inode;
1079         inode_item->generation = cpu_to_le64(1);
1080         inode_item->size = cpu_to_le64(3);
1081         inode_item->nlink = cpu_to_le32(1);
1082         inode_item->nbytes = cpu_to_le64(root->leafsize);
1083         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1084
1085         btrfs_set_root_node(&log_root->root_item, log_root->node);
1086
1087         WARN_ON(root->log_root);
1088         root->log_root = log_root;
1089         root->log_transid = 0;
1090         return 0;
1091 }
1092
1093 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1094                                                struct btrfs_key *location)
1095 {
1096         struct btrfs_root *root;
1097         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1098         struct btrfs_path *path;
1099         struct extent_buffer *l;
1100         u64 generation;
1101         u32 blocksize;
1102         int ret = 0;
1103
1104         root = kzalloc(sizeof(*root), GFP_NOFS);
1105         if (!root)
1106                 return ERR_PTR(-ENOMEM);
1107         if (location->offset == (u64)-1) {
1108                 ret = find_and_setup_root(tree_root, fs_info,
1109                                           location->objectid, root);
1110                 if (ret) {
1111                         kfree(root);
1112                         return ERR_PTR(ret);
1113                 }
1114                 goto out;
1115         }
1116
1117         __setup_root(tree_root->nodesize, tree_root->leafsize,
1118                      tree_root->sectorsize, tree_root->stripesize,
1119                      root, fs_info, location->objectid);
1120
1121         path = btrfs_alloc_path();
1122         BUG_ON(!path);
1123         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1124         if (ret == 0) {
1125                 l = path->nodes[0];
1126                 read_extent_buffer(l, &root->root_item,
1127                                 btrfs_item_ptr_offset(l, path->slots[0]),
1128                                 sizeof(root->root_item));
1129                 memcpy(&root->root_key, location, sizeof(*location));
1130         }
1131         btrfs_free_path(path);
1132         if (ret) {
1133                 if (ret > 0)
1134                         ret = -ENOENT;
1135                 return ERR_PTR(ret);
1136         }
1137
1138         generation = btrfs_root_generation(&root->root_item);
1139         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1140         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1141                                      blocksize, generation);
1142         root->commit_root = btrfs_root_node(root);
1143         BUG_ON(!root->node);
1144 out:
1145         if (location->objectid != BTRFS_TREE_LOG_OBJECTID)
1146                 root->ref_cows = 1;
1147
1148         return root;
1149 }
1150
1151 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1152                                         u64 root_objectid)
1153 {
1154         struct btrfs_root *root;
1155
1156         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1157                 return fs_info->tree_root;
1158         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1159                 return fs_info->extent_root;
1160
1161         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1162                                  (unsigned long)root_objectid);
1163         return root;
1164 }
1165
1166 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1167                                               struct btrfs_key *location)
1168 {
1169         struct btrfs_root *root;
1170         int ret;
1171
1172         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1173                 return fs_info->tree_root;
1174         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1175                 return fs_info->extent_root;
1176         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1177                 return fs_info->chunk_root;
1178         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1179                 return fs_info->dev_root;
1180         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1181                 return fs_info->csum_root;
1182 again:
1183         spin_lock(&fs_info->fs_roots_radix_lock);
1184         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1185                                  (unsigned long)location->objectid);
1186         spin_unlock(&fs_info->fs_roots_radix_lock);
1187         if (root)
1188                 return root;
1189
1190         ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1191         if (ret == 0)
1192                 ret = -ENOENT;
1193         if (ret < 0)
1194                 return ERR_PTR(ret);
1195
1196         root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1197         if (IS_ERR(root))
1198                 return root;
1199
1200         WARN_ON(btrfs_root_refs(&root->root_item) == 0);
1201         set_anon_super(&root->anon_super, NULL);
1202
1203         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1204         if (ret)
1205                 goto fail;
1206
1207         spin_lock(&fs_info->fs_roots_radix_lock);
1208         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1209                                 (unsigned long)root->root_key.objectid,
1210                                 root);
1211         if (ret == 0)
1212                 root->in_radix = 1;
1213         spin_unlock(&fs_info->fs_roots_radix_lock);
1214         radix_tree_preload_end();
1215         if (ret) {
1216                 if (ret == -EEXIST) {
1217                         free_fs_root(root);
1218                         goto again;
1219                 }
1220                 goto fail;
1221         }
1222
1223         ret = btrfs_find_dead_roots(fs_info->tree_root,
1224                                     root->root_key.objectid);
1225         WARN_ON(ret);
1226
1227         if (!(fs_info->sb->s_flags & MS_RDONLY))
1228                 btrfs_orphan_cleanup(root);
1229
1230         return root;
1231 fail:
1232         free_fs_root(root);
1233         return ERR_PTR(ret);
1234 }
1235
1236 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1237                                       struct btrfs_key *location,
1238                                       const char *name, int namelen)
1239 {
1240         return btrfs_read_fs_root_no_name(fs_info, location);
1241 #if 0
1242         struct btrfs_root *root;
1243         int ret;
1244
1245         root = btrfs_read_fs_root_no_name(fs_info, location);
1246         if (!root)
1247                 return NULL;
1248
1249         if (root->in_sysfs)
1250                 return root;
1251
1252         ret = btrfs_set_root_name(root, name, namelen);
1253         if (ret) {
1254                 free_extent_buffer(root->node);
1255                 kfree(root);
1256                 return ERR_PTR(ret);
1257         }
1258
1259         ret = btrfs_sysfs_add_root(root);
1260         if (ret) {
1261                 free_extent_buffer(root->node);
1262                 kfree(root->name);
1263                 kfree(root);
1264                 return ERR_PTR(ret);
1265         }
1266         root->in_sysfs = 1;
1267         return root;
1268 #endif
1269 }
1270
1271 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1272 {
1273         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1274         int ret = 0;
1275         struct btrfs_device *device;
1276         struct backing_dev_info *bdi;
1277
1278         list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1279                 if (!device->bdev)
1280                         continue;
1281                 bdi = blk_get_backing_dev_info(device->bdev);
1282                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1283                         ret = 1;
1284                         break;
1285                 }
1286         }
1287         return ret;
1288 }
1289
1290 /*
1291  * this unplugs every device on the box, and it is only used when page
1292  * is null
1293  */
1294 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1295 {
1296         struct btrfs_device *device;
1297         struct btrfs_fs_info *info;
1298
1299         info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1300         list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1301                 if (!device->bdev)
1302                         continue;
1303
1304                 bdi = blk_get_backing_dev_info(device->bdev);
1305                 if (bdi->unplug_io_fn)
1306                         bdi->unplug_io_fn(bdi, page);
1307         }
1308 }
1309
1310 static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1311 {
1312         struct inode *inode;
1313         struct extent_map_tree *em_tree;
1314         struct extent_map *em;
1315         struct address_space *mapping;
1316         u64 offset;
1317
1318         /* the generic O_DIRECT read code does this */
1319         if (1 || !page) {
1320                 __unplug_io_fn(bdi, page);
1321                 return;
1322         }
1323
1324         /*
1325          * page->mapping may change at any time.  Get a consistent copy
1326          * and use that for everything below
1327          */
1328         smp_mb();
1329         mapping = page->mapping;
1330         if (!mapping)
1331                 return;
1332
1333         inode = mapping->host;
1334
1335         /*
1336          * don't do the expensive searching for a small number of
1337          * devices
1338          */
1339         if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1340                 __unplug_io_fn(bdi, page);
1341                 return;
1342         }
1343
1344         offset = page_offset(page);
1345
1346         em_tree = &BTRFS_I(inode)->extent_tree;
1347         read_lock(&em_tree->lock);
1348         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1349         read_unlock(&em_tree->lock);
1350         if (!em) {
1351                 __unplug_io_fn(bdi, page);
1352                 return;
1353         }
1354
1355         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1356                 free_extent_map(em);
1357                 __unplug_io_fn(bdi, page);
1358                 return;
1359         }
1360         offset = offset - em->start;
1361         btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1362                           em->block_start + offset, page);
1363         free_extent_map(em);
1364 }
1365
1366 /*
1367  * If this fails, caller must call bdi_destroy() to get rid of the
1368  * bdi again.
1369  */
1370 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1371 {
1372         int err;
1373
1374         bdi->name = "btrfs";
1375         bdi->capabilities = BDI_CAP_MAP_COPY;
1376         err = bdi_init(bdi);
1377         if (err)
1378                 return err;
1379
1380         err = bdi_register(bdi, NULL, "btrfs-%d",
1381                                 atomic_inc_return(&btrfs_bdi_num));
1382         if (err) {
1383                 bdi_destroy(bdi);
1384                 return err;
1385         }
1386
1387         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1388         bdi->unplug_io_fn       = btrfs_unplug_io_fn;
1389         bdi->unplug_io_data     = info;
1390         bdi->congested_fn       = btrfs_congested_fn;
1391         bdi->congested_data     = info;
1392         return 0;
1393 }
1394
1395 static int bio_ready_for_csum(struct bio *bio)
1396 {
1397         u64 length = 0;
1398         u64 buf_len = 0;
1399         u64 start = 0;
1400         struct page *page;
1401         struct extent_io_tree *io_tree = NULL;
1402         struct btrfs_fs_info *info = NULL;
1403         struct bio_vec *bvec;
1404         int i;
1405         int ret;
1406
1407         bio_for_each_segment(bvec, bio, i) {
1408                 page = bvec->bv_page;
1409                 if (page->private == EXTENT_PAGE_PRIVATE) {
1410                         length += bvec->bv_len;
1411                         continue;
1412                 }
1413                 if (!page->private) {
1414                         length += bvec->bv_len;
1415                         continue;
1416                 }
1417                 length = bvec->bv_len;
1418                 buf_len = page->private >> 2;
1419                 start = page_offset(page) + bvec->bv_offset;
1420                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1421                 info = BTRFS_I(page->mapping->host)->root->fs_info;
1422         }
1423         /* are we fully contained in this bio? */
1424         if (buf_len <= length)
1425                 return 1;
1426
1427         ret = extent_range_uptodate(io_tree, start + length,
1428                                     start + buf_len - 1);
1429         return ret;
1430 }
1431
1432 /*
1433  * called by the kthread helper functions to finally call the bio end_io
1434  * functions.  This is where read checksum verification actually happens
1435  */
1436 static void end_workqueue_fn(struct btrfs_work *work)
1437 {
1438         struct bio *bio;
1439         struct end_io_wq *end_io_wq;
1440         struct btrfs_fs_info *fs_info;
1441         int error;
1442
1443         end_io_wq = container_of(work, struct end_io_wq, work);
1444         bio = end_io_wq->bio;
1445         fs_info = end_io_wq->info;
1446
1447         /* metadata bio reads are special because the whole tree block must
1448          * be checksummed at once.  This makes sure the entire block is in
1449          * ram and up to date before trying to verify things.  For
1450          * blocksize <= pagesize, it is basically a noop
1451          */
1452         if (!(bio->bi_rw & (1 << BIO_RW)) && end_io_wq->metadata &&
1453             !bio_ready_for_csum(bio)) {
1454                 btrfs_queue_worker(&fs_info->endio_meta_workers,
1455                                    &end_io_wq->work);
1456                 return;
1457         }
1458         error = end_io_wq->error;
1459         bio->bi_private = end_io_wq->private;
1460         bio->bi_end_io = end_io_wq->end_io;
1461         kfree(end_io_wq);
1462         bio_endio(bio, error);
1463 }
1464
1465 static int cleaner_kthread(void *arg)
1466 {
1467         struct btrfs_root *root = arg;
1468
1469         do {
1470                 smp_mb();
1471                 if (root->fs_info->closing)
1472                         break;
1473
1474                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1475
1476                 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1477                     mutex_trylock(&root->fs_info->cleaner_mutex)) {
1478                         btrfs_clean_old_snapshots(root);
1479                         mutex_unlock(&root->fs_info->cleaner_mutex);
1480                 }
1481
1482                 if (freezing(current)) {
1483                         refrigerator();
1484                 } else {
1485                         smp_mb();
1486                         if (root->fs_info->closing)
1487                                 break;
1488                         set_current_state(TASK_INTERRUPTIBLE);
1489                         schedule();
1490                         __set_current_state(TASK_RUNNING);
1491                 }
1492         } while (!kthread_should_stop());
1493         return 0;
1494 }
1495
1496 static int transaction_kthread(void *arg)
1497 {
1498         struct btrfs_root *root = arg;
1499         struct btrfs_trans_handle *trans;
1500         struct btrfs_transaction *cur;
1501         unsigned long now;
1502         unsigned long delay;
1503         int ret;
1504
1505         do {
1506                 smp_mb();
1507                 if (root->fs_info->closing)
1508                         break;
1509
1510                 delay = HZ * 30;
1511                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1512                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1513
1514                 mutex_lock(&root->fs_info->trans_mutex);
1515                 cur = root->fs_info->running_transaction;
1516                 if (!cur) {
1517                         mutex_unlock(&root->fs_info->trans_mutex);
1518                         goto sleep;
1519                 }
1520
1521                 now = get_seconds();
1522                 if (now < cur->start_time || now - cur->start_time < 30) {
1523                         mutex_unlock(&root->fs_info->trans_mutex);
1524                         delay = HZ * 5;
1525                         goto sleep;
1526                 }
1527                 mutex_unlock(&root->fs_info->trans_mutex);
1528                 trans = btrfs_start_transaction(root, 1);
1529                 ret = btrfs_commit_transaction(trans, root);
1530
1531 sleep:
1532                 wake_up_process(root->fs_info->cleaner_kthread);
1533                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1534
1535                 if (freezing(current)) {
1536                         refrigerator();
1537                 } else {
1538                         if (root->fs_info->closing)
1539                                 break;
1540                         set_current_state(TASK_INTERRUPTIBLE);
1541                         schedule_timeout(delay);
1542                         __set_current_state(TASK_RUNNING);
1543                 }
1544         } while (!kthread_should_stop());
1545         return 0;
1546 }
1547
1548 struct btrfs_root *open_ctree(struct super_block *sb,
1549                               struct btrfs_fs_devices *fs_devices,
1550                               char *options)
1551 {
1552         u32 sectorsize;
1553         u32 nodesize;
1554         u32 leafsize;
1555         u32 blocksize;
1556         u32 stripesize;
1557         u64 generation;
1558         u64 features;
1559         struct btrfs_key location;
1560         struct buffer_head *bh;
1561         struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1562                                                  GFP_NOFS);
1563         struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1564                                                  GFP_NOFS);
1565         struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
1566                                                GFP_NOFS);
1567         struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1568                                                 GFP_NOFS);
1569         struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1570                                                 GFP_NOFS);
1571         struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1572                                               GFP_NOFS);
1573         struct btrfs_root *log_tree_root;
1574
1575         int ret;
1576         int err = -EINVAL;
1577
1578         struct btrfs_super_block *disk_super;
1579
1580         if (!extent_root || !tree_root || !fs_info ||
1581             !chunk_root || !dev_root || !csum_root) {
1582                 err = -ENOMEM;
1583                 goto fail;
1584         }
1585
1586         ret = init_srcu_struct(&fs_info->subvol_srcu);
1587         if (ret) {
1588                 err = ret;
1589                 goto fail;
1590         }
1591
1592         ret = setup_bdi(fs_info, &fs_info->bdi);
1593         if (ret) {
1594                 err = ret;
1595                 goto fail_srcu;
1596         }
1597
1598         fs_info->btree_inode = new_inode(sb);
1599         if (!fs_info->btree_inode) {
1600                 err = -ENOMEM;
1601                 goto fail_bdi;
1602         }
1603
1604         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1605         INIT_LIST_HEAD(&fs_info->trans_list);
1606         INIT_LIST_HEAD(&fs_info->dead_roots);
1607         INIT_LIST_HEAD(&fs_info->hashers);
1608         INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1609         INIT_LIST_HEAD(&fs_info->ordered_operations);
1610         INIT_LIST_HEAD(&fs_info->caching_block_groups);
1611         spin_lock_init(&fs_info->delalloc_lock);
1612         spin_lock_init(&fs_info->new_trans_lock);
1613         spin_lock_init(&fs_info->ref_cache_lock);
1614         spin_lock_init(&fs_info->fs_roots_radix_lock);
1615
1616         init_completion(&fs_info->kobj_unregister);
1617         fs_info->tree_root = tree_root;
1618         fs_info->extent_root = extent_root;
1619         fs_info->csum_root = csum_root;
1620         fs_info->chunk_root = chunk_root;
1621         fs_info->dev_root = dev_root;
1622         fs_info->fs_devices = fs_devices;
1623         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1624         INIT_LIST_HEAD(&fs_info->space_info);
1625         btrfs_mapping_init(&fs_info->mapping_tree);
1626         atomic_set(&fs_info->nr_async_submits, 0);
1627         atomic_set(&fs_info->async_delalloc_pages, 0);
1628         atomic_set(&fs_info->async_submit_draining, 0);
1629         atomic_set(&fs_info->nr_async_bios, 0);
1630         fs_info->sb = sb;
1631         fs_info->max_extent = (u64)-1;
1632         fs_info->max_inline = 8192 * 1024;
1633         fs_info->metadata_ratio = 0;
1634
1635         fs_info->thread_pool_size = min_t(unsigned long,
1636                                           num_online_cpus() + 2, 8);
1637
1638         INIT_LIST_HEAD(&fs_info->ordered_extents);
1639         spin_lock_init(&fs_info->ordered_extent_lock);
1640
1641         sb->s_blocksize = 4096;
1642         sb->s_blocksize_bits = blksize_bits(4096);
1643         sb->s_bdi = &fs_info->bdi;
1644
1645         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1646         fs_info->btree_inode->i_nlink = 1;
1647         /*
1648          * we set the i_size on the btree inode to the max possible int.
1649          * the real end of the address space is determined by all of
1650          * the devices in the system
1651          */
1652         fs_info->btree_inode->i_size = OFFSET_MAX;
1653         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1654         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1655
1656         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
1657         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1658                              fs_info->btree_inode->i_mapping,
1659                              GFP_NOFS);
1660         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1661                              GFP_NOFS);
1662
1663         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1664
1665         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1666         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1667                sizeof(struct btrfs_key));
1668         BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
1669         insert_inode_hash(fs_info->btree_inode);
1670
1671         spin_lock_init(&fs_info->block_group_cache_lock);
1672         fs_info->block_group_cache_tree.rb_node = NULL;
1673
1674         extent_io_tree_init(&fs_info->freed_extents[0],
1675                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1676         extent_io_tree_init(&fs_info->freed_extents[1],
1677                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1678         fs_info->pinned_extents = &fs_info->freed_extents[0];
1679         fs_info->do_barriers = 1;
1680
1681
1682         mutex_init(&fs_info->trans_mutex);
1683         mutex_init(&fs_info->ordered_operations_mutex);
1684         mutex_init(&fs_info->tree_log_mutex);
1685         mutex_init(&fs_info->chunk_mutex);
1686         mutex_init(&fs_info->transaction_kthread_mutex);
1687         mutex_init(&fs_info->cleaner_mutex);
1688         mutex_init(&fs_info->volume_mutex);
1689         init_rwsem(&fs_info->extent_commit_sem);
1690         init_rwsem(&fs_info->subvol_sem);
1691
1692         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1693         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1694
1695         init_waitqueue_head(&fs_info->transaction_throttle);
1696         init_waitqueue_head(&fs_info->transaction_wait);
1697         init_waitqueue_head(&fs_info->async_submit_wait);
1698
1699         __setup_root(4096, 4096, 4096, 4096, tree_root,
1700                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1701
1702
1703         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1704         if (!bh)
1705                 goto fail_iput;
1706
1707         memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1708         memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1709                sizeof(fs_info->super_for_commit));
1710         brelse(bh);
1711
1712         memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1713
1714         disk_super = &fs_info->super_copy;
1715         if (!btrfs_super_root(disk_super))
1716                 goto fail_iput;
1717
1718         ret = btrfs_parse_options(tree_root, options);
1719         if (ret) {
1720                 err = ret;
1721                 goto fail_iput;
1722         }
1723
1724         features = btrfs_super_incompat_flags(disk_super) &
1725                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1726         if (features) {
1727                 printk(KERN_ERR "BTRFS: couldn't mount because of "
1728                        "unsupported optional features (%Lx).\n",
1729                        (unsigned long long)features);
1730                 err = -EINVAL;
1731                 goto fail_iput;
1732         }
1733
1734         features = btrfs_super_incompat_flags(disk_super);
1735         if (!(features & BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF)) {
1736                 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
1737                 btrfs_set_super_incompat_flags(disk_super, features);
1738         }
1739
1740         features = btrfs_super_compat_ro_flags(disk_super) &
1741                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1742         if (!(sb->s_flags & MS_RDONLY) && features) {
1743                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1744                        "unsupported option features (%Lx).\n",
1745                        (unsigned long long)features);
1746                 err = -EINVAL;
1747                 goto fail_iput;
1748         }
1749
1750         btrfs_init_workers(&fs_info->generic_worker,
1751                            "genwork", 1, NULL);
1752
1753         btrfs_init_workers(&fs_info->workers, "worker",
1754                            fs_info->thread_pool_size,
1755                            &fs_info->generic_worker);
1756
1757         btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1758                            fs_info->thread_pool_size,
1759                            &fs_info->generic_worker);
1760
1761         btrfs_init_workers(&fs_info->submit_workers, "submit",
1762                            min_t(u64, fs_devices->num_devices,
1763                            fs_info->thread_pool_size),
1764                            &fs_info->generic_worker);
1765         btrfs_init_workers(&fs_info->enospc_workers, "enospc",
1766                            fs_info->thread_pool_size,
1767                            &fs_info->generic_worker);
1768
1769         /* a higher idle thresh on the submit workers makes it much more
1770          * likely that bios will be send down in a sane order to the
1771          * devices
1772          */
1773         fs_info->submit_workers.idle_thresh = 64;
1774
1775         fs_info->workers.idle_thresh = 16;
1776         fs_info->workers.ordered = 1;
1777
1778         fs_info->delalloc_workers.idle_thresh = 2;
1779         fs_info->delalloc_workers.ordered = 1;
1780
1781         btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
1782                            &fs_info->generic_worker);
1783         btrfs_init_workers(&fs_info->endio_workers, "endio",
1784                            fs_info->thread_pool_size,
1785                            &fs_info->generic_worker);
1786         btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1787                            fs_info->thread_pool_size,
1788                            &fs_info->generic_worker);
1789         btrfs_init_workers(&fs_info->endio_meta_write_workers,
1790                            "endio-meta-write", fs_info->thread_pool_size,
1791                            &fs_info->generic_worker);
1792         btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1793                            fs_info->thread_pool_size,
1794                            &fs_info->generic_worker);
1795
1796         /*
1797          * endios are largely parallel and should have a very
1798          * low idle thresh
1799          */
1800         fs_info->endio_workers.idle_thresh = 4;
1801         fs_info->endio_meta_workers.idle_thresh = 4;
1802
1803         fs_info->endio_write_workers.idle_thresh = 2;
1804         fs_info->endio_meta_write_workers.idle_thresh = 2;
1805
1806         btrfs_start_workers(&fs_info->workers, 1);
1807         btrfs_start_workers(&fs_info->generic_worker, 1);
1808         btrfs_start_workers(&fs_info->submit_workers, 1);
1809         btrfs_start_workers(&fs_info->delalloc_workers, 1);
1810         btrfs_start_workers(&fs_info->fixup_workers, 1);
1811         btrfs_start_workers(&fs_info->endio_workers, 1);
1812         btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1813         btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1814         btrfs_start_workers(&fs_info->endio_write_workers, 1);
1815         btrfs_start_workers(&fs_info->enospc_workers, 1);
1816
1817         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1818         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1819                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1820
1821         nodesize = btrfs_super_nodesize(disk_super);
1822         leafsize = btrfs_super_leafsize(disk_super);
1823         sectorsize = btrfs_super_sectorsize(disk_super);
1824         stripesize = btrfs_super_stripesize(disk_super);
1825         tree_root->nodesize = nodesize;
1826         tree_root->leafsize = leafsize;
1827         tree_root->sectorsize = sectorsize;
1828         tree_root->stripesize = stripesize;
1829
1830         sb->s_blocksize = sectorsize;
1831         sb->s_blocksize_bits = blksize_bits(sectorsize);
1832
1833         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1834                     sizeof(disk_super->magic))) {
1835                 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
1836                 goto fail_sb_buffer;
1837         }
1838
1839         mutex_lock(&fs_info->chunk_mutex);
1840         ret = btrfs_read_sys_array(tree_root);
1841         mutex_unlock(&fs_info->chunk_mutex);
1842         if (ret) {
1843                 printk(KERN_WARNING "btrfs: failed to read the system "
1844                        "array on %s\n", sb->s_id);
1845                 goto fail_sb_buffer;
1846         }
1847
1848         blocksize = btrfs_level_size(tree_root,
1849                                      btrfs_super_chunk_root_level(disk_super));
1850         generation = btrfs_super_chunk_root_generation(disk_super);
1851
1852         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1853                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1854
1855         chunk_root->node = read_tree_block(chunk_root,
1856                                            btrfs_super_chunk_root(disk_super),
1857                                            blocksize, generation);
1858         BUG_ON(!chunk_root->node);
1859         if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
1860                 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
1861                        sb->s_id);
1862                 goto fail_chunk_root;
1863         }
1864         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
1865         chunk_root->commit_root = btrfs_root_node(chunk_root);
1866
1867         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1868            (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1869            BTRFS_UUID_SIZE);
1870
1871         mutex_lock(&fs_info->chunk_mutex);
1872         ret = btrfs_read_chunk_tree(chunk_root);
1873         mutex_unlock(&fs_info->chunk_mutex);
1874         if (ret) {
1875                 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
1876                        sb->s_id);
1877                 goto fail_chunk_root;
1878         }
1879
1880         btrfs_close_extra_devices(fs_devices);
1881
1882         blocksize = btrfs_level_size(tree_root,
1883                                      btrfs_super_root_level(disk_super));
1884         generation = btrfs_super_generation(disk_super);
1885
1886         tree_root->node = read_tree_block(tree_root,
1887                                           btrfs_super_root(disk_super),
1888                                           blocksize, generation);
1889         if (!tree_root->node)
1890                 goto fail_chunk_root;
1891         if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
1892                 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
1893                        sb->s_id);
1894                 goto fail_tree_root;
1895         }
1896         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
1897         tree_root->commit_root = btrfs_root_node(tree_root);
1898
1899         ret = find_and_setup_root(tree_root, fs_info,
1900                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1901         if (ret)
1902                 goto fail_tree_root;
1903         extent_root->track_dirty = 1;
1904
1905         ret = find_and_setup_root(tree_root, fs_info,
1906                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
1907         if (ret)
1908                 goto fail_extent_root;
1909         dev_root->track_dirty = 1;
1910
1911         ret = find_and_setup_root(tree_root, fs_info,
1912                                   BTRFS_CSUM_TREE_OBJECTID, csum_root);
1913         if (ret)
1914                 goto fail_dev_root;
1915
1916         csum_root->track_dirty = 1;
1917
1918         btrfs_read_block_groups(extent_root);
1919
1920         fs_info->generation = generation;
1921         fs_info->last_trans_committed = generation;
1922         fs_info->data_alloc_profile = (u64)-1;
1923         fs_info->metadata_alloc_profile = (u64)-1;
1924         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1925         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1926                                                "btrfs-cleaner");
1927         if (IS_ERR(fs_info->cleaner_kthread))
1928                 goto fail_csum_root;
1929
1930         fs_info->transaction_kthread = kthread_run(transaction_kthread,
1931                                                    tree_root,
1932                                                    "btrfs-transaction");
1933         if (IS_ERR(fs_info->transaction_kthread))
1934                 goto fail_cleaner;
1935
1936         if (!btrfs_test_opt(tree_root, SSD) &&
1937             !btrfs_test_opt(tree_root, NOSSD) &&
1938             !fs_info->fs_devices->rotating) {
1939                 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
1940                        "mode\n");
1941                 btrfs_set_opt(fs_info->mount_opt, SSD);
1942         }
1943
1944         if (btrfs_super_log_root(disk_super) != 0) {
1945                 u64 bytenr = btrfs_super_log_root(disk_super);
1946
1947                 if (fs_devices->rw_devices == 0) {
1948                         printk(KERN_WARNING "Btrfs log replay required "
1949                                "on RO media\n");
1950                         err = -EIO;
1951                         goto fail_trans_kthread;
1952                 }
1953                 blocksize =
1954                      btrfs_level_size(tree_root,
1955                                       btrfs_super_log_root_level(disk_super));
1956
1957                 log_tree_root = kzalloc(sizeof(struct btrfs_root),
1958                                                       GFP_NOFS);
1959
1960                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1961                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1962
1963                 log_tree_root->node = read_tree_block(tree_root, bytenr,
1964                                                       blocksize,
1965                                                       generation + 1);
1966                 ret = btrfs_recover_log_trees(log_tree_root);
1967                 BUG_ON(ret);
1968
1969                 if (sb->s_flags & MS_RDONLY) {
1970                         ret =  btrfs_commit_super(tree_root);
1971                         BUG_ON(ret);
1972                 }
1973         }
1974
1975         ret = btrfs_find_orphan_roots(tree_root);
1976         BUG_ON(ret);
1977
1978         if (!(sb->s_flags & MS_RDONLY)) {
1979                 ret = btrfs_recover_relocation(tree_root);
1980                 BUG_ON(ret);
1981         }
1982
1983         location.objectid = BTRFS_FS_TREE_OBJECTID;
1984         location.type = BTRFS_ROOT_ITEM_KEY;
1985         location.offset = (u64)-1;
1986
1987         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
1988         if (!fs_info->fs_root)
1989                 goto fail_trans_kthread;
1990
1991         return tree_root;
1992
1993 fail_trans_kthread:
1994         kthread_stop(fs_info->transaction_kthread);
1995 fail_cleaner:
1996         kthread_stop(fs_info->cleaner_kthread);
1997
1998         /*
1999          * make sure we're done with the btree inode before we stop our
2000          * kthreads
2001          */
2002         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2003         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2004
2005 fail_csum_root:
2006         free_extent_buffer(csum_root->node);
2007         free_extent_buffer(csum_root->commit_root);
2008 fail_dev_root:
2009         free_extent_buffer(dev_root->node);
2010         free_extent_buffer(dev_root->commit_root);
2011 fail_extent_root:
2012         free_extent_buffer(extent_root->node);
2013         free_extent_buffer(extent_root->commit_root);
2014 fail_tree_root:
2015         free_extent_buffer(tree_root->node);
2016         free_extent_buffer(tree_root->commit_root);
2017 fail_chunk_root:
2018         free_extent_buffer(chunk_root->node);
2019         free_extent_buffer(chunk_root->commit_root);
2020 fail_sb_buffer:
2021         btrfs_stop_workers(&fs_info->generic_worker);
2022         btrfs_stop_workers(&fs_info->fixup_workers);
2023         btrfs_stop_workers(&fs_info->delalloc_workers);
2024         btrfs_stop_workers(&fs_info->workers);
2025         btrfs_stop_workers(&fs_info->endio_workers);
2026         btrfs_stop_workers(&fs_info->endio_meta_workers);
2027         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2028         btrfs_stop_workers(&fs_info->endio_write_workers);
2029         btrfs_stop_workers(&fs_info->submit_workers);
2030         btrfs_stop_workers(&fs_info->enospc_workers);
2031 fail_iput:
2032         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2033         iput(fs_info->btree_inode);
2034
2035         btrfs_close_devices(fs_info->fs_devices);
2036         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2037 fail_bdi:
2038         bdi_destroy(&fs_info->bdi);
2039 fail_srcu:
2040         cleanup_srcu_struct(&fs_info->subvol_srcu);
2041 fail:
2042         kfree(extent_root);
2043         kfree(tree_root);
2044         kfree(fs_info);
2045         kfree(chunk_root);
2046         kfree(dev_root);
2047         kfree(csum_root);
2048         return ERR_PTR(err);
2049 }
2050
2051 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2052 {
2053         char b[BDEVNAME_SIZE];
2054
2055         if (uptodate) {
2056                 set_buffer_uptodate(bh);
2057         } else {
2058                 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
2059                         printk(KERN_WARNING "lost page write due to "
2060                                         "I/O error on %s\n",
2061                                        bdevname(bh->b_bdev, b));
2062                 }
2063                 /* note, we dont' set_buffer_write_io_error because we have
2064                  * our own ways of dealing with the IO errors
2065                  */
2066                 clear_buffer_uptodate(bh);
2067         }
2068         unlock_buffer(bh);
2069         put_bh(bh);
2070 }
2071
2072 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2073 {
2074         struct buffer_head *bh;
2075         struct buffer_head *latest = NULL;
2076         struct btrfs_super_block *super;
2077         int i;
2078         u64 transid = 0;
2079         u64 bytenr;
2080
2081         /* we would like to check all the supers, but that would make
2082          * a btrfs mount succeed after a mkfs from a different FS.
2083          * So, we need to add a special mount option to scan for
2084          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2085          */
2086         for (i = 0; i < 1; i++) {
2087                 bytenr = btrfs_sb_offset(i);
2088                 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2089                         break;
2090                 bh = __bread(bdev, bytenr / 4096, 4096);
2091                 if (!bh)
2092                         continue;
2093
2094                 super = (struct btrfs_super_block *)bh->b_data;
2095                 if (btrfs_super_bytenr(super) != bytenr ||
2096                     strncmp((char *)(&super->magic), BTRFS_MAGIC,
2097                             sizeof(super->magic))) {
2098                         brelse(bh);
2099                         continue;
2100                 }
2101
2102                 if (!latest || btrfs_super_generation(super) > transid) {
2103                         brelse(latest);
2104                         latest = bh;
2105                         transid = btrfs_super_generation(super);
2106                 } else {
2107                         brelse(bh);
2108                 }
2109         }
2110         return latest;
2111 }
2112
2113 /*
2114  * this should be called twice, once with wait == 0 and
2115  * once with wait == 1.  When wait == 0 is done, all the buffer heads
2116  * we write are pinned.
2117  *
2118  * They are released when wait == 1 is done.
2119  * max_mirrors must be the same for both runs, and it indicates how
2120  * many supers on this one device should be written.
2121  *
2122  * max_mirrors == 0 means to write them all.
2123  */
2124 static int write_dev_supers(struct btrfs_device *device,
2125                             struct btrfs_super_block *sb,
2126                             int do_barriers, int wait, int max_mirrors)
2127 {
2128         struct buffer_head *bh;
2129         int i;
2130         int ret;
2131         int errors = 0;
2132         u32 crc;
2133         u64 bytenr;
2134         int last_barrier = 0;
2135
2136         if (max_mirrors == 0)
2137                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2138
2139         /* make sure only the last submit_bh does a barrier */
2140         if (do_barriers) {
2141                 for (i = 0; i < max_mirrors; i++) {
2142                         bytenr = btrfs_sb_offset(i);
2143                         if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2144                             device->total_bytes)
2145                                 break;
2146                         last_barrier = i;
2147                 }
2148         }
2149
2150         for (i = 0; i < max_mirrors; i++) {
2151                 bytenr = btrfs_sb_offset(i);
2152                 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2153                         break;
2154
2155                 if (wait) {
2156                         bh = __find_get_block(device->bdev, bytenr / 4096,
2157                                               BTRFS_SUPER_INFO_SIZE);
2158                         BUG_ON(!bh);
2159                         wait_on_buffer(bh);
2160                         if (!buffer_uptodate(bh))
2161                                 errors++;
2162
2163                         /* drop our reference */
2164                         brelse(bh);
2165
2166                         /* drop the reference from the wait == 0 run */
2167                         brelse(bh);
2168                         continue;
2169                 } else {
2170                         btrfs_set_super_bytenr(sb, bytenr);
2171
2172                         crc = ~(u32)0;
2173                         crc = btrfs_csum_data(NULL, (char *)sb +
2174                                               BTRFS_CSUM_SIZE, crc,
2175                                               BTRFS_SUPER_INFO_SIZE -
2176                                               BTRFS_CSUM_SIZE);
2177                         btrfs_csum_final(crc, sb->csum);
2178
2179                         /*
2180                          * one reference for us, and we leave it for the
2181                          * caller
2182                          */
2183                         bh = __getblk(device->bdev, bytenr / 4096,
2184                                       BTRFS_SUPER_INFO_SIZE);
2185                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2186
2187                         /* one reference for submit_bh */
2188                         get_bh(bh);
2189
2190                         set_buffer_uptodate(bh);
2191                         lock_buffer(bh);
2192                         bh->b_end_io = btrfs_end_buffer_write_sync;
2193                 }
2194
2195                 if (i == last_barrier && do_barriers && device->barriers) {
2196                         ret = submit_bh(WRITE_BARRIER, bh);
2197                         if (ret == -EOPNOTSUPP) {
2198                                 printk("btrfs: disabling barriers on dev %s\n",
2199                                        device->name);
2200                                 set_buffer_uptodate(bh);
2201                                 device->barriers = 0;
2202                                 /* one reference for submit_bh */
2203                                 get_bh(bh);
2204                                 lock_buffer(bh);
2205                                 ret = submit_bh(WRITE_SYNC, bh);
2206                         }
2207                 } else {
2208                         ret = submit_bh(WRITE_SYNC, bh);
2209                 }
2210
2211                 if (ret)
2212                         errors++;
2213         }
2214         return errors < i ? 0 : -1;
2215 }
2216
2217 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2218 {
2219         struct list_head *head;
2220         struct btrfs_device *dev;
2221         struct btrfs_super_block *sb;
2222         struct btrfs_dev_item *dev_item;
2223         int ret;
2224         int do_barriers;
2225         int max_errors;
2226         int total_errors = 0;
2227         u64 flags;
2228
2229         max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2230         do_barriers = !btrfs_test_opt(root, NOBARRIER);
2231
2232         sb = &root->fs_info->super_for_commit;
2233         dev_item = &sb->dev_item;
2234
2235         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2236         head = &root->fs_info->fs_devices->devices;
2237         list_for_each_entry(dev, head, dev_list) {
2238                 if (!dev->bdev) {
2239                         total_errors++;
2240                         continue;
2241                 }
2242                 if (!dev->in_fs_metadata || !dev->writeable)
2243                         continue;
2244
2245                 btrfs_set_stack_device_generation(dev_item, 0);
2246                 btrfs_set_stack_device_type(dev_item, dev->type);
2247                 btrfs_set_stack_device_id(dev_item, dev->devid);
2248                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2249                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2250                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2251                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2252                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2253                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2254                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2255
2256                 flags = btrfs_super_flags(sb);
2257                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2258
2259                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2260                 if (ret)
2261                         total_errors++;
2262         }
2263         if (total_errors > max_errors) {
2264                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2265                        total_errors);
2266                 BUG();
2267         }
2268
2269         total_errors = 0;
2270         list_for_each_entry(dev, head, dev_list) {
2271                 if (!dev->bdev)
2272                         continue;
2273                 if (!dev->in_fs_metadata || !dev->writeable)
2274                         continue;
2275
2276                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2277                 if (ret)
2278                         total_errors++;
2279         }
2280         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2281         if (total_errors > max_errors) {
2282                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2283                        total_errors);
2284                 BUG();
2285         }
2286         return 0;
2287 }
2288
2289 int write_ctree_super(struct btrfs_trans_handle *trans,
2290                       struct btrfs_root *root, int max_mirrors)
2291 {
2292         int ret;
2293
2294         ret = write_all_supers(root, max_mirrors);
2295         return ret;
2296 }
2297
2298 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2299 {
2300         spin_lock(&fs_info->fs_roots_radix_lock);
2301         radix_tree_delete(&fs_info->fs_roots_radix,
2302                           (unsigned long)root->root_key.objectid);
2303         spin_unlock(&fs_info->fs_roots_radix_lock);
2304
2305         if (btrfs_root_refs(&root->root_item) == 0)
2306                 synchronize_srcu(&fs_info->subvol_srcu);
2307
2308         free_fs_root(root);
2309         return 0;
2310 }
2311
2312 static void free_fs_root(struct btrfs_root *root)
2313 {
2314         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2315         if (root->anon_super.s_dev) {
2316                 down_write(&root->anon_super.s_umount);
2317                 kill_anon_super(&root->anon_super);
2318         }
2319         free_extent_buffer(root->node);
2320         free_extent_buffer(root->commit_root);
2321         kfree(root->name);
2322         kfree(root);
2323 }
2324
2325 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2326 {
2327         int ret;
2328         struct btrfs_root *gang[8];
2329         int i;
2330
2331         while (!list_empty(&fs_info->dead_roots)) {
2332                 gang[0] = list_entry(fs_info->dead_roots.next,
2333                                      struct btrfs_root, root_list);
2334                 list_del(&gang[0]->root_list);
2335
2336                 if (gang[0]->in_radix) {
2337                         btrfs_free_fs_root(fs_info, gang[0]);
2338                 } else {
2339                         free_extent_buffer(gang[0]->node);
2340                         free_extent_buffer(gang[0]->commit_root);
2341                         kfree(gang[0]);
2342                 }
2343         }
2344
2345         while (1) {
2346                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2347                                              (void **)gang, 0,
2348                                              ARRAY_SIZE(gang));
2349                 if (!ret)
2350                         break;
2351                 for (i = 0; i < ret; i++)
2352                         btrfs_free_fs_root(fs_info, gang[i]);
2353         }
2354         return 0;
2355 }
2356
2357 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2358 {
2359         u64 root_objectid = 0;
2360         struct btrfs_root *gang[8];
2361         int i;
2362         int ret;
2363
2364         while (1) {
2365                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2366                                              (void **)gang, root_objectid,
2367                                              ARRAY_SIZE(gang));
2368                 if (!ret)
2369                         break;
2370
2371                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2372                 for (i = 0; i < ret; i++) {
2373                         root_objectid = gang[i]->root_key.objectid;
2374                         btrfs_orphan_cleanup(gang[i]);
2375                 }
2376                 root_objectid++;
2377         }
2378         return 0;
2379 }
2380
2381 int btrfs_commit_super(struct btrfs_root *root)
2382 {
2383         struct btrfs_trans_handle *trans;
2384         int ret;
2385
2386         mutex_lock(&root->fs_info->cleaner_mutex);
2387         btrfs_clean_old_snapshots(root);
2388         mutex_unlock(&root->fs_info->cleaner_mutex);
2389         trans = btrfs_start_transaction(root, 1);
2390         ret = btrfs_commit_transaction(trans, root);
2391         BUG_ON(ret);
2392         /* run commit again to drop the original snapshot */
2393         trans = btrfs_start_transaction(root, 1);
2394         btrfs_commit_transaction(trans, root);
2395         ret = btrfs_write_and_wait_transaction(NULL, root);
2396         BUG_ON(ret);
2397
2398         ret = write_ctree_super(NULL, root, 0);
2399         return ret;
2400 }
2401
2402 int close_ctree(struct btrfs_root *root)
2403 {
2404         struct btrfs_fs_info *fs_info = root->fs_info;
2405         int ret;
2406
2407         fs_info->closing = 1;
2408         smp_mb();
2409
2410         kthread_stop(root->fs_info->transaction_kthread);
2411         kthread_stop(root->fs_info->cleaner_kthread);
2412
2413         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2414                 ret =  btrfs_commit_super(root);
2415                 if (ret)
2416                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2417         }
2418
2419         fs_info->closing = 2;
2420         smp_mb();
2421
2422         if (fs_info->delalloc_bytes) {
2423                 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2424                        (unsigned long long)fs_info->delalloc_bytes);
2425         }
2426         if (fs_info->total_ref_cache_size) {
2427                 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2428                        (unsigned long long)fs_info->total_ref_cache_size);
2429         }
2430
2431         free_extent_buffer(fs_info->extent_root->node);
2432         free_extent_buffer(fs_info->extent_root->commit_root);
2433         free_extent_buffer(fs_info->tree_root->node);
2434         free_extent_buffer(fs_info->tree_root->commit_root);
2435         free_extent_buffer(root->fs_info->chunk_root->node);
2436         free_extent_buffer(root->fs_info->chunk_root->commit_root);
2437         free_extent_buffer(root->fs_info->dev_root->node);
2438         free_extent_buffer(root->fs_info->dev_root->commit_root);
2439         free_extent_buffer(root->fs_info->csum_root->node);
2440         free_extent_buffer(root->fs_info->csum_root->commit_root);
2441
2442         btrfs_free_block_groups(root->fs_info);
2443
2444         del_fs_roots(fs_info);
2445
2446         iput(fs_info->btree_inode);
2447
2448         btrfs_stop_workers(&fs_info->generic_worker);
2449         btrfs_stop_workers(&fs_info->fixup_workers);
2450         btrfs_stop_workers(&fs_info->delalloc_workers);
2451         btrfs_stop_workers(&fs_info->workers);
2452         btrfs_stop_workers(&fs_info->endio_workers);
2453         btrfs_stop_workers(&fs_info->endio_meta_workers);
2454         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2455         btrfs_stop_workers(&fs_info->endio_write_workers);
2456         btrfs_stop_workers(&fs_info->submit_workers);
2457         btrfs_stop_workers(&fs_info->enospc_workers);
2458
2459         btrfs_close_devices(fs_info->fs_devices);
2460         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2461
2462         bdi_destroy(&fs_info->bdi);
2463         cleanup_srcu_struct(&fs_info->subvol_srcu);
2464
2465         kfree(fs_info->extent_root);
2466         kfree(fs_info->tree_root);
2467         kfree(fs_info->chunk_root);
2468         kfree(fs_info->dev_root);
2469         kfree(fs_info->csum_root);
2470         return 0;
2471 }
2472
2473 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2474 {
2475         int ret;
2476         struct inode *btree_inode = buf->first_page->mapping->host;
2477
2478         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
2479         if (!ret)
2480                 return ret;
2481
2482         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2483                                     parent_transid);
2484         return !ret;
2485 }
2486
2487 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2488 {
2489         struct inode *btree_inode = buf->first_page->mapping->host;
2490         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2491                                           buf);
2492 }
2493
2494 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2495 {
2496         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2497         u64 transid = btrfs_header_generation(buf);
2498         struct inode *btree_inode = root->fs_info->btree_inode;
2499         int was_dirty;
2500
2501         btrfs_assert_tree_locked(buf);
2502         if (transid != root->fs_info->generation) {
2503                 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2504                        "found %llu running %llu\n",
2505                         (unsigned long long)buf->start,
2506                         (unsigned long long)transid,
2507                         (unsigned long long)root->fs_info->generation);
2508                 WARN_ON(1);
2509         }
2510         was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2511                                             buf);
2512         if (!was_dirty) {
2513                 spin_lock(&root->fs_info->delalloc_lock);
2514                 root->fs_info->dirty_metadata_bytes += buf->len;
2515                 spin_unlock(&root->fs_info->delalloc_lock);
2516         }
2517 }
2518
2519 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2520 {
2521         /*
2522          * looks as though older kernels can get into trouble with
2523          * this code, they end up stuck in balance_dirty_pages forever
2524          */
2525         u64 num_dirty;
2526         unsigned long thresh = 32 * 1024 * 1024;
2527
2528         if (current->flags & PF_MEMALLOC)
2529                 return;
2530
2531         num_dirty = root->fs_info->dirty_metadata_bytes;
2532
2533         if (num_dirty > thresh) {
2534                 balance_dirty_pages_ratelimited_nr(
2535                                    root->fs_info->btree_inode->i_mapping, 1);
2536         }
2537         return;
2538 }
2539
2540 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2541 {
2542         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2543         int ret;
2544         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2545         if (ret == 0)
2546                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
2547         return ret;
2548 }
2549
2550 int btree_lock_page_hook(struct page *page)
2551 {
2552         struct inode *inode = page->mapping->host;
2553         struct btrfs_root *root = BTRFS_I(inode)->root;
2554         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2555         struct extent_buffer *eb;
2556         unsigned long len;
2557         u64 bytenr = page_offset(page);
2558
2559         if (page->private == EXTENT_PAGE_PRIVATE)
2560                 goto out;
2561
2562         len = page->private >> 2;
2563         eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2564         if (!eb)
2565                 goto out;
2566
2567         btrfs_tree_lock(eb);
2568         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2569
2570         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2571                 spin_lock(&root->fs_info->delalloc_lock);
2572                 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2573                         root->fs_info->dirty_metadata_bytes -= eb->len;
2574                 else
2575                         WARN_ON(1);
2576                 spin_unlock(&root->fs_info->delalloc_lock);
2577         }
2578
2579         btrfs_tree_unlock(eb);
2580         free_extent_buffer(eb);
2581 out:
2582         lock_page(page);
2583         return 0;
2584 }
2585
2586 static struct extent_io_ops btree_extent_io_ops = {
2587         .write_cache_pages_lock_hook = btree_lock_page_hook,
2588         .readpage_end_io_hook = btree_readpage_end_io_hook,
2589         .submit_bio_hook = btree_submit_bio_hook,
2590         /* note we're sharing with inode.c for the merge bio hook */
2591         .merge_bio_hook = btrfs_merge_bio_hook,
2592 };