git.oblomov.eu Git - 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 "compat.h"
  30 #include "crc32c.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 "ref-cache.h"
  40 #include "tree-log.h"
  41
  42 static struct extent_io_ops btree_extent_io_ops;
  43 static void end_workqueue_fn(struct btrfs_work *work);
  44
  45 /*
  46  * end_io_wq structs are used to do processing in task context when an IO is
  47  * complete.  This is used during reads to verify checksums, and it is used
  48  * by writes to insert metadata for new file extents after IO is complete.
  49  */
  50 struct end_io_wq {
  51         struct bio *bio;
  52         bio_end_io_t *end_io;
  53         void *private;
  54         struct btrfs_fs_info *info;
  55         int error;
  56         int metadata;
  57         struct list_head list;
  58         struct btrfs_work work;
  59 };
  60
  61 /*
  62  * async submit bios are used to offload expensive checksumming
  63  * onto the worker threads.  They checksum file and metadata bios
  64  * just before they are sent down the IO stack.
  65  */
  66 struct async_submit_bio {
  67         struct inode *inode;
  68         struct bio *bio;
  69         struct list_head list;
  70         extent_submit_bio_hook_t *submit_bio_start;
  71         extent_submit_bio_hook_t *submit_bio_done;
  72         int rw;
  73         int mirror_num;
  74         unsigned long bio_flags;
  75         struct btrfs_work work;
  76 };
  77
  78 /* These are used to set the lockdep class on the extent buffer locks.
  79  * The class is set by the readpage_end_io_hook after the buffer has
  80  * passed csum validation but before the pages are unlocked.
  81  *
  82  * The lockdep class is also set by btrfs_init_new_buffer on freshly
  83  * allocated blocks.
  84  *
  85  * The class is based on the level in the tree block, which allows lockdep
  86  * to know that lower nodes nest inside the locks of higher nodes.
  87  *
  88  * We also add a check to make sure the highest level of the tree is
  89  * the same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this
  90  * code needs update as well.
  91  */
  92 #ifdef CONFIG_DEBUG_LOCK_ALLOC
  93 # if BTRFS_MAX_LEVEL != 8
  94 #  error
  95 # endif
  96 static struct lock_class_key btrfs_eb_class[BTRFS_MAX_LEVEL + 1];
  97 static const char *btrfs_eb_name[BTRFS_MAX_LEVEL + 1] = {
  98         /* leaf */
  99         "btrfs-extent-00",
 100         "btrfs-extent-01",
 101         "btrfs-extent-02",
 102         "btrfs-extent-03",
 103         "btrfs-extent-04",
 104         "btrfs-extent-05",
 105         "btrfs-extent-06",
 106         "btrfs-extent-07",
 107         /* highest possible level */
 108         "btrfs-extent-08",
 109 };
 110 #endif
 111
 112 /*
 113  * extents on the btree inode are pretty simple, there's one extent
 114  * that covers the entire device
 115  */
 116 static struct extent_map *btree_get_extent(struct inode *inode,
 117                 struct page *page, size_t page_offset, u64 start, u64 len,
 118                 int create)
 119 {
 120         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 121         struct extent_map *em;
 122         int ret;
 123
 124         spin_lock(&em_tree->lock);
 125         em = lookup_extent_mapping(em_tree, start, len);
 126         if (em) {
 127                 em->bdev =
 128                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 129                 spin_unlock(&em_tree->lock);
 130                 goto out;
 131         }
 132         spin_unlock(&em_tree->lock);
 133
 134         em = alloc_extent_map(GFP_NOFS);
 135         if (!em) {
 136                 em = ERR_PTR(-ENOMEM);
 137                 goto out;
 138         }
 139         em->start = 0;
 140         em->len = (u64)-1;
 141         em->block_len = (u64)-1;
 142         em->block_start = 0;
 143         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 144
 145         spin_lock(&em_tree->lock);
 146         ret = add_extent_mapping(em_tree, em);
 147         if (ret == -EEXIST) {
 148                 u64 failed_start = em->start;
 149                 u64 failed_len = em->len;
 150
 151                 free_extent_map(em);
 152                 em = lookup_extent_mapping(em_tree, start, len);
 153                 if (em) {
 154                         ret = 0;
 155                 } else {
 156                         em = lookup_extent_mapping(em_tree, failed_start,
 157                                                    failed_len);
 158                         ret = -EIO;
 159                 }
 160         } else if (ret) {
 161                 free_extent_map(em);
 162                 em = NULL;
 163         }
 164         spin_unlock(&em_tree->lock);
 165
 166         if (ret)
 167                 em = ERR_PTR(ret);
 168 out:
 169         return em;
 170 }
 171
 172 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
 173 {
 174         return btrfs_crc32c(seed, data, len);
 175 }
 176
 177 void btrfs_csum_final(u32 crc, char *result)
 178 {
 179         *(__le32 *)result = ~cpu_to_le32(crc);
 180 }
 181
 182 /*
 183  * compute the csum for a btree block, and either verify it or write it
 184  * into the csum field of the block.
 185  */
 186 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 187                            int verify)
 188 {
 189         u16 csum_size =
 190                 btrfs_super_csum_size(&root->fs_info->super_copy);
 191         char *result = NULL;
 192         unsigned long len;
 193         unsigned long cur_len;
 194         unsigned long offset = BTRFS_CSUM_SIZE;
 195         char *map_token = NULL;
 196         char *kaddr;
 197         unsigned long map_start;
 198         unsigned long map_len;
 199         int err;
 200         u32 crc = ~(u32)0;
 201         unsigned long inline_result;
 202
 203         len = buf->len - offset;
 204         while (len > 0) {
 205                 err = map_private_extent_buffer(buf, offset, 32,
 206                                         &map_token, &kaddr,
 207                                         &map_start, &map_len, KM_USER0);
 208                 if (err)
 209                         return 1;
 210                 cur_len = min(len, map_len - (offset - map_start));
 211                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
 212                                       crc, cur_len);
 213                 len -= cur_len;
 214                 offset += cur_len;
 215                 unmap_extent_buffer(buf, map_token, KM_USER0);
 216         }
 217         if (csum_size > sizeof(inline_result)) {
 218                 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
 219                 if (!result)
 220                         return 1;
 221         } else {
 222                 result = (char *)&inline_result;
 223         }
 224
 225         btrfs_csum_final(crc, result);
 226
 227         if (verify) {
 228                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
 229                         u32 val;
 230                         u32 found = 0;
 231                         memcpy(&found, result, csum_size);
 232
 233                         read_extent_buffer(buf, &val, 0, csum_size);
 234                         printk(KERN_INFO "btrfs: %s checksum verify failed "
 235                                "on %llu wanted %X found %X level %d\n",
 236                                root->fs_info->sb->s_id,
 237                                buf->start, val, found, btrfs_header_level(buf));
 238                         if (result != (char *)&inline_result)
 239                                 kfree(result);
 240                         return 1;
 241                 }
 242         } else {
 243                 write_extent_buffer(buf, result, 0, csum_size);
 244         }
 245         if (result != (char *)&inline_result)
 246                 kfree(result);
 247         return 0;
 248 }
 249
 250 /*
 251  * we can't consider a given block up to date unless the transid of the
 252  * block matches the transid in the parent node's pointer.  This is how we
 253  * detect blocks that either didn't get written at all or got written
 254  * in the wrong place.
 255  */
 256 static int verify_parent_transid(struct extent_io_tree *io_tree,
 257                                  struct extent_buffer *eb, u64 parent_transid)
 258 {
 259         int ret;
 260
 261         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 262                 return 0;
 263
 264         lock_extent(io_tree, eb->start, eb->start + eb->len - 1, GFP_NOFS);
 265         if (extent_buffer_uptodate(io_tree, eb) &&
 266             btrfs_header_generation(eb) == parent_transid) {
 267                 ret = 0;
 268                 goto out;
 269         }
 270         printk("parent transid verify failed on %llu wanted %llu found %llu\n",
 271                (unsigned long long)eb->start,
 272                (unsigned long long)parent_transid,
 273                (unsigned long long)btrfs_header_generation(eb));
 274         ret = 1;
 275         clear_extent_buffer_uptodate(io_tree, eb);
 276 out:
 277         unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
 278                       GFP_NOFS);
 279         return ret;
 280 }
 281
 282 /*
 283  * helper to read a given tree block, doing retries as required when
 284  * the checksums don't match and we have alternate mirrors to try.
 285  */
 286 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 287                                           struct extent_buffer *eb,
 288                                           u64 start, u64 parent_transid)
 289 {
 290         struct extent_io_tree *io_tree;
 291         int ret;
 292         int num_copies = 0;
 293         int mirror_num = 0;
 294
 295         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 296         while (1) {
 297                 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
 298                                                btree_get_extent, mirror_num);
 299                 if (!ret &&
 300                     !verify_parent_transid(io_tree, eb, parent_transid))
 301                         return ret;
 302
 303                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
 304                                               eb->start, eb->len);
 305                 if (num_copies == 1)
 306                         return ret;
 307
 308                 mirror_num++;
 309                 if (mirror_num > num_copies)
 310                         return ret;
 311         }
 312         return -EIO;
 313 }
 314
 315 /*
 316  * checksum a dirty tree block before IO.  This has extra checks to make sure
 317  * we only fill in the checksum field in the first page of a multi-page block
 318  */
 319
 320 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 321 {
 322         struct extent_io_tree *tree;
 323         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
 324         u64 found_start;
 325         int found_level;
 326         unsigned long len;
 327         struct extent_buffer *eb;
 328         int ret;
 329
 330         tree = &BTRFS_I(page->mapping->host)->io_tree;
 331
 332         if (page->private == EXTENT_PAGE_PRIVATE)
 333                 goto out;
 334         if (!page->private)
 335                 goto out;
 336         len = page->private >> 2;
 337         WARN_ON(len == 0);
 338
 339         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 340         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
 341                                              btrfs_header_generation(eb));
 342         BUG_ON(ret);
 343         found_start = btrfs_header_bytenr(eb);
 344         if (found_start != start) {
 345                 WARN_ON(1);
 346                 goto err;
 347         }
 348         if (eb->first_page != page) {
 349                 WARN_ON(1);
 350                 goto err;
 351         }
 352         if (!PageUptodate(page)) {
 353                 WARN_ON(1);
 354                 goto err;
 355         }
 356         found_level = btrfs_header_level(eb);
 357
 358         csum_tree_block(root, eb, 0);
 359 err:
 360         free_extent_buffer(eb);
 361 out:
 362         return 0;
 363 }
 364
 365 static int check_tree_block_fsid(struct btrfs_root *root,
 366                                  struct extent_buffer *eb)
 367 {
 368         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
 369         u8 fsid[BTRFS_UUID_SIZE];
 370         int ret = 1;
 371
 372         read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
 373                            BTRFS_FSID_SIZE);
 374         while (fs_devices) {
 375                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
 376                         ret = 0;
 377                         break;
 378                 }
 379                 fs_devices = fs_devices->seed;
 380         }
 381         return ret;
 382 }
 383
 384 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 385 void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level)
 386 {
 387         lockdep_set_class_and_name(&eb->lock,
 388                            &btrfs_eb_class[level],
 389                            btrfs_eb_name[level]);
 390 }
 391 #endif
 392
 393 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 394                                struct extent_state *state)
 395 {
 396         struct extent_io_tree *tree;
 397         u64 found_start;
 398         int found_level;
 399         unsigned long len;
 400         struct extent_buffer *eb;
 401         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 402         int ret = 0;
 403
 404         tree = &BTRFS_I(page->mapping->host)->io_tree;
 405         if (page->private == EXTENT_PAGE_PRIVATE)
 406                 goto out;
 407         if (!page->private)
 408                 goto out;
 409
 410         len = page->private >> 2;
 411         WARN_ON(len == 0);
 412
 413         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 414
 415         found_start = btrfs_header_bytenr(eb);
 416         if (found_start != start) {
 417                 printk(KERN_INFO "btrfs bad tree block start %llu %llu\n",
 418                        (unsigned long long)found_start,
 419                        (unsigned long long)eb->start);
 420                 ret = -EIO;
 421                 goto err;
 422         }
 423         if (eb->first_page != page) {
 424                 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
 425                        eb->first_page->index, page->index);
 426                 WARN_ON(1);
 427                 ret = -EIO;
 428                 goto err;
 429         }
 430         if (check_tree_block_fsid(root, eb)) {
 431                 printk(KERN_INFO "btrfs bad fsid on block %llu\n",
 432                        (unsigned long long)eb->start);
 433                 ret = -EIO;
 434                 goto err;
 435         }
 436         found_level = btrfs_header_level(eb);
 437
 438         btrfs_set_buffer_lockdep_class(eb, found_level);
 439
 440         ret = csum_tree_block(root, eb, 1);
 441         if (ret)
 442                 ret = -EIO;
 443
 444         end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
 445         end = eb->start + end - 1;
 446 err:
 447         free_extent_buffer(eb);
 448 out:
 449         return ret;
 450 }
 451
 452 static void end_workqueue_bio(struct bio *bio, int err)
 453 {
 454         struct end_io_wq *end_io_wq = bio->bi_private;
 455         struct btrfs_fs_info *fs_info;
 456
 457         fs_info = end_io_wq->info;
 458         end_io_wq->error = err;
 459         end_io_wq->work.func = end_workqueue_fn;
 460         end_io_wq->work.flags = 0;
 461
 462         if (bio->bi_rw & (1 << BIO_RW)) {
 463                 if (end_io_wq->metadata)
 464                         btrfs_queue_worker(&fs_info->endio_meta_write_workers,
 465                                            &end_io_wq->work);
 466                 else
 467                         btrfs_queue_worker(&fs_info->endio_write_workers,
 468                                            &end_io_wq->work);
 469         } else {
 470                 if (end_io_wq->metadata)
 471                         btrfs_queue_worker(&fs_info->endio_meta_workers,
 472                                            &end_io_wq->work);
 473                 else
 474                         btrfs_queue_worker(&fs_info->endio_workers,
 475                                            &end_io_wq->work);
 476         }
 477 }
 478
 479 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 480                         int metadata)
 481 {
 482         struct end_io_wq *end_io_wq;
 483         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
 484         if (!end_io_wq)
 485                 return -ENOMEM;
 486
 487         end_io_wq->private = bio->bi_private;
 488         end_io_wq->end_io = bio->bi_end_io;
 489         end_io_wq->info = info;
 490         end_io_wq->error = 0;
 491         end_io_wq->bio = bio;
 492         end_io_wq->metadata = metadata;
 493
 494         bio->bi_private = end_io_wq;
 495         bio->bi_end_io = end_workqueue_bio;
 496         return 0;
 497 }
 498
 499 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
 500 {
 501         unsigned long limit = min_t(unsigned long,
 502                                     info->workers.max_workers,
 503                                     info->fs_devices->open_devices);
 504         return 256 * limit;
 505 }
 506
 507 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
 508 {
 509         return atomic_read(&info->nr_async_bios) >
 510                 btrfs_async_submit_limit(info);
 511 }
 512
 513 static void run_one_async_start(struct btrfs_work *work)
 514 {
 515         struct btrfs_fs_info *fs_info;
 516         struct async_submit_bio *async;
 517
 518         async = container_of(work, struct  async_submit_bio, work);
 519         fs_info = BTRFS_I(async->inode)->root->fs_info;
 520         async->submit_bio_start(async->inode, async->rw, async->bio,
 521                                async->mirror_num, async->bio_flags);
 522 }
 523
 524 static void run_one_async_done(struct btrfs_work *work)
 525 {
 526         struct btrfs_fs_info *fs_info;
 527         struct async_submit_bio *async;
 528         int limit;
 529
 530         async = container_of(work, struct  async_submit_bio, work);
 531         fs_info = BTRFS_I(async->inode)->root->fs_info;
 532
 533         limit = btrfs_async_submit_limit(fs_info);
 534         limit = limit * 2 / 3;
 535
 536         atomic_dec(&fs_info->nr_async_submits);
 537
 538         if (atomic_read(&fs_info->nr_async_submits) < limit &&
 539             waitqueue_active(&fs_info->async_submit_wait))
 540                 wake_up(&fs_info->async_submit_wait);
 541
 542         async->submit_bio_done(async->inode, async->rw, async->bio,
 543                                async->mirror_num, async->bio_flags);
 544 }
 545
 546 static void run_one_async_free(struct btrfs_work *work)
 547 {
 548         struct async_submit_bio *async;
 549
 550         async = container_of(work, struct  async_submit_bio, work);
 551         kfree(async);
 552 }
 553
 554 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 555                         int rw, struct bio *bio, int mirror_num,
 556                         unsigned long bio_flags,
 557                         extent_submit_bio_hook_t *submit_bio_start,
 558                         extent_submit_bio_hook_t *submit_bio_done)
 559 {
 560         struct async_submit_bio *async;
 561
 562         async = kmalloc(sizeof(*async), GFP_NOFS);
 563         if (!async)
 564                 return -ENOMEM;
 565
 566         async->inode = inode;
 567         async->rw = rw;
 568         async->bio = bio;
 569         async->mirror_num = mirror_num;
 570         async->submit_bio_start = submit_bio_start;
 571         async->submit_bio_done = submit_bio_done;
 572
 573         async->work.func = run_one_async_start;
 574         async->work.ordered_func = run_one_async_done;
 575         async->work.ordered_free = run_one_async_free;
 576
 577         async->work.flags = 0;
 578         async->bio_flags = bio_flags;
 579
 580         atomic_inc(&fs_info->nr_async_submits);
 581         btrfs_queue_worker(&fs_info->workers, &async->work);
 582 #if 0
 583         int limit = btrfs_async_submit_limit(fs_info);
 584         if (atomic_read(&fs_info->nr_async_submits) > limit) {
 585                 wait_event_timeout(fs_info->async_submit_wait,
 586                            (atomic_read(&fs_info->nr_async_submits) < limit),
 587                            HZ/10);
 588
 589                 wait_event_timeout(fs_info->async_submit_wait,
 590                            (atomic_read(&fs_info->nr_async_bios) < limit),
 591                            HZ/10);
 592         }
 593 #endif
 594         while (atomic_read(&fs_info->async_submit_draining) &&
 595               atomic_read(&fs_info->nr_async_submits)) {
 596                 wait_event(fs_info->async_submit_wait,
 597                            (atomic_read(&fs_info->nr_async_submits) == 0));
 598         }
 599
 600         return 0;
 601 }
 602
 603 static int btree_csum_one_bio(struct bio *bio)
 604 {
 605         struct bio_vec *bvec = bio->bi_io_vec;
 606         int bio_index = 0;
 607         struct btrfs_root *root;
 608
 609         WARN_ON(bio->bi_vcnt <= 0);
 610         while (bio_index < bio->bi_vcnt) {
 611                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 612                 csum_dirty_buffer(root, bvec->bv_page);
 613                 bio_index++;
 614                 bvec++;
 615         }
 616         return 0;
 617 }
 618
 619 static int __btree_submit_bio_start(struct inode *inode, int rw,
 620                                     struct bio *bio, int mirror_num,
 621                                     unsigned long bio_flags)
 622 {
 623         /*
 624          * when we're called for a write, we're already in the async
 625          * submission context.  Just jump into btrfs_map_bio
 626          */
 627         btree_csum_one_bio(bio);
 628         return 0;
 629 }
 630
 631 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
 632                                  int mirror_num, unsigned long bio_flags)
 633 {
 634         /*
 635          * when we're called for a write, we're already in the async
 636          * submission context.  Just jump into btrfs_map_bio
 637          */
 638         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
 639 }
 640
 641 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 642                                  int mirror_num, unsigned long bio_flags)
 643 {
 644         int ret;
 645
 646         ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
 647                                           bio, 1);
 648         BUG_ON(ret);
 649
 650         if (!(rw & (1 << BIO_RW))) {
 651                 /*
 652                  * called for a read, do the setup so that checksum validation
 653                  * can happen in the async kernel threads
 654                  */
 655                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 656                                      mirror_num, 0);
 657         }
 658         /*
 659          * kthread helpers are used to submit writes so that checksumming
 660          * can happen in parallel across all CPUs
 661          */
 662         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 663                                    inode, rw, bio, mirror_num, 0,
 664                                    __btree_submit_bio_start,
 665                                    __btree_submit_bio_done);
 666 }
 667
 668 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 669 {
 670         struct extent_io_tree *tree;
 671         tree = &BTRFS_I(page->mapping->host)->io_tree;
 672
 673         if (current->flags & PF_MEMALLOC) {
 674                 redirty_page_for_writepage(wbc, page);
 675                 unlock_page(page);
 676                 return 0;
 677         }
 678         return extent_write_full_page(tree, page, btree_get_extent, wbc);
 679 }
 680
 681 static int btree_writepages(struct address_space *mapping,
 682                             struct writeback_control *wbc)
 683 {
 684         struct extent_io_tree *tree;
 685         tree = &BTRFS_I(mapping->host)->io_tree;
 686         if (wbc->sync_mode == WB_SYNC_NONE) {
 687                 u64 num_dirty;
 688                 u64 start = 0;
 689                 unsigned long thresh = 32 * 1024 * 1024;
 690
 691                 if (wbc->for_kupdate)
 692                         return 0;
 693
 694                 num_dirty = count_range_bits(tree, &start, (u64)-1,
 695                                              thresh, EXTENT_DIRTY);
 696                 if (num_dirty < thresh)
 697                         return 0;
 698         }
 699         return extent_writepages(tree, mapping, btree_get_extent, wbc);
 700 }
 701
 702 static int btree_readpage(struct file *file, struct page *page)
 703 {
 704         struct extent_io_tree *tree;
 705         tree = &BTRFS_I(page->mapping->host)->io_tree;
 706         return extent_read_full_page(tree, page, btree_get_extent);
 707 }
 708
 709 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 710 {
 711         struct extent_io_tree *tree;
 712         struct extent_map_tree *map;
 713         int ret;
 714
 715         if (PageWriteback(page) || PageDirty(page))
 716                 return 0;
 717
 718         tree = &BTRFS_I(page->mapping->host)->io_tree;
 719         map = &BTRFS_I(page->mapping->host)->extent_tree;
 720
 721         ret = try_release_extent_state(map, tree, page, gfp_flags);
 722         if (!ret)
 723                 return 0;
 724
 725         ret = try_release_extent_buffer(tree, page);
 726         if (ret == 1) {
 727                 ClearPagePrivate(page);
 728                 set_page_private(page, 0);
 729                 page_cache_release(page);
 730         }
 731
 732         return ret;
 733 }
 734
 735 static void btree_invalidatepage(struct page *page, unsigned long offset)
 736 {
 737         struct extent_io_tree *tree;
 738         tree = &BTRFS_I(page->mapping->host)->io_tree;
 739         extent_invalidatepage(tree, page, offset);
 740         btree_releasepage(page, GFP_NOFS);
 741         if (PagePrivate(page)) {
 742                 printk(KERN_WARNING "btrfs warning page private not zero "
 743                        "on page %llu\n", (unsigned long long)page_offset(page));
 744                 ClearPagePrivate(page);
 745                 set_page_private(page, 0);
 746                 page_cache_release(page);
 747         }
 748 }
 749
 750 #if 0
 751 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 752 {
 753         struct buffer_head *bh;
 754         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 755         struct buffer_head *head;
 756         if (!page_has_buffers(page)) {
 757                 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
 758                                         (1 << BH_Dirty)|(1 << BH_Uptodate));
 759         }
 760         head = page_buffers(page);
 761         bh = head;
 762         do {
 763                 if (buffer_dirty(bh))
 764                         csum_tree_block(root, bh, 0);
 765                 bh = bh->b_this_page;
 766         } while (bh != head);
 767         return block_write_full_page(page, btree_get_block, wbc);
 768 }
 769 #endif
 770
 771 static struct address_space_operations btree_aops = {
 772         .readpage       = btree_readpage,
 773         .writepage      = btree_writepage,
 774         .writepages     = btree_writepages,
 775         .releasepage    = btree_releasepage,
 776         .invalidatepage = btree_invalidatepage,
 777         .sync_page      = block_sync_page,
 778 };
 779
 780 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
 781                          u64 parent_transid)
 782 {
 783         struct extent_buffer *buf = NULL;
 784         struct inode *btree_inode = root->fs_info->btree_inode;
 785         int ret = 0;
 786
 787         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 788         if (!buf)
 789                 return 0;
 790         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
 791                                  buf, 0, 0, btree_get_extent, 0);
 792         free_extent_buffer(buf);
 793         return ret;
 794 }
 795
 796 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
 797                                             u64 bytenr, u32 blocksize)
 798 {
 799         struct inode *btree_inode = root->fs_info->btree_inode;
 800         struct extent_buffer *eb;
 801         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 802                                 bytenr, blocksize, GFP_NOFS);
 803         return eb;
 804 }
 805
 806 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
 807                                                  u64 bytenr, u32 blocksize)
 808 {
 809         struct inode *btree_inode = root->fs_info->btree_inode;
 810         struct extent_buffer *eb;
 811
 812         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 813                                  bytenr, blocksize, NULL, GFP_NOFS);
 814         return eb;
 815 }
 816
 817
 818 int btrfs_write_tree_block(struct extent_buffer *buf)
 819 {
 820         return btrfs_fdatawrite_range(buf->first_page->mapping, buf->start,
 821                                       buf->start + buf->len - 1, WB_SYNC_ALL);
 822 }
 823
 824 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
 825 {
 826         return btrfs_wait_on_page_writeback_range(buf->first_page->mapping,
 827                                   buf->start, buf->start + buf->len - 1);
 828 }
 829
 830 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
 831                                       u32 blocksize, u64 parent_transid)
 832 {
 833         struct extent_buffer *buf = NULL;
 834         struct inode *btree_inode = root->fs_info->btree_inode;
 835         struct extent_io_tree *io_tree;
 836         int ret;
 837
 838         io_tree = &BTRFS_I(btree_inode)->io_tree;
 839
 840         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 841         if (!buf)
 842                 return NULL;
 843
 844         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
 845
 846         if (ret == 0)
 847                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
 848         else
 849                 WARN_ON(1);
 850         return buf;
 851
 852 }
 853
 854 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 855                      struct extent_buffer *buf)
 856 {
 857         struct inode *btree_inode = root->fs_info->btree_inode;
 858         if (btrfs_header_generation(buf) ==
 859             root->fs_info->running_transaction->transid) {
 860                 btrfs_assert_tree_locked(buf);
 861
 862                 /* ugh, clear_extent_buffer_dirty can be expensive */
 863                 btrfs_set_lock_blocking(buf);
 864
 865                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
 866                                           buf);
 867         }
 868         return 0;
 869 }
 870
 871 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
 872                         u32 stripesize, struct btrfs_root *root,
 873                         struct btrfs_fs_info *fs_info,
 874                         u64 objectid)
 875 {
 876         root->node = NULL;
 877         root->commit_root = NULL;
 878         root->ref_tree = NULL;
 879         root->sectorsize = sectorsize;
 880         root->nodesize = nodesize;
 881         root->leafsize = leafsize;
 882         root->stripesize = stripesize;
 883         root->ref_cows = 0;
 884         root->track_dirty = 0;
 885
 886         root->fs_info = fs_info;
 887         root->objectid = objectid;
 888         root->last_trans = 0;
 889         root->highest_inode = 0;
 890         root->last_inode_alloc = 0;
 891         root->name = NULL;
 892         root->in_sysfs = 0;
 893
 894         INIT_LIST_HEAD(&root->dirty_list);
 895         INIT_LIST_HEAD(&root->orphan_list);
 896         INIT_LIST_HEAD(&root->dead_list);
 897         spin_lock_init(&root->node_lock);
 898         spin_lock_init(&root->list_lock);
 899         mutex_init(&root->objectid_mutex);
 900         mutex_init(&root->log_mutex);
 901         init_waitqueue_head(&root->log_writer_wait);
 902         init_waitqueue_head(&root->log_commit_wait[0]);
 903         init_waitqueue_head(&root->log_commit_wait[1]);
 904         atomic_set(&root->log_commit[0], 0);
 905         atomic_set(&root->log_commit[1], 0);
 906         atomic_set(&root->log_writers, 0);
 907         root->log_batch = 0;
 908         root->log_transid = 0;
 909         extent_io_tree_init(&root->dirty_log_pages,
 910                              fs_info->btree_inode->i_mapping, GFP_NOFS);
 911
 912         btrfs_leaf_ref_tree_init(&root->ref_tree_struct);
 913         root->ref_tree = &root->ref_tree_struct;
 914
 915         memset(&root->root_key, 0, sizeof(root->root_key));
 916         memset(&root->root_item, 0, sizeof(root->root_item));
 917         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
 918         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
 919         root->defrag_trans_start = fs_info->generation;
 920         init_completion(&root->kobj_unregister);
 921         root->defrag_running = 0;
 922         root->defrag_level = 0;
 923         root->root_key.objectid = objectid;
 924         root->anon_super.s_root = NULL;
 925         root->anon_super.s_dev = 0;
 926         INIT_LIST_HEAD(&root->anon_super.s_list);
 927         INIT_LIST_HEAD(&root->anon_super.s_instances);
 928         init_rwsem(&root->anon_super.s_umount);
 929
 930         return 0;
 931 }
 932
 933 static int find_and_setup_root(struct btrfs_root *tree_root,
 934                                struct btrfs_fs_info *fs_info,
 935                                u64 objectid,
 936                                struct btrfs_root *root)
 937 {
 938         int ret;
 939         u32 blocksize;
 940         u64 generation;
 941
 942         __setup_root(tree_root->nodesize, tree_root->leafsize,
 943                      tree_root->sectorsize, tree_root->stripesize,
 944                      root, fs_info, objectid);
 945         ret = btrfs_find_last_root(tree_root, objectid,
 946                                    &root->root_item, &root->root_key);
 947         BUG_ON(ret);
 948
 949         generation = btrfs_root_generation(&root->root_item);
 950         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 951         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 952                                      blocksize, generation);
 953         BUG_ON(!root->node);
 954         return 0;
 955 }
 956
 957 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
 958                              struct btrfs_fs_info *fs_info)
 959 {
 960         struct extent_buffer *eb;
 961         struct btrfs_root *log_root_tree = fs_info->log_root_tree;
 962         u64 start = 0;
 963         u64 end = 0;
 964         int ret;
 965
 966         if (!log_root_tree)
 967                 return 0;
 968
 969         while (1) {
 970                 ret = find_first_extent_bit(&log_root_tree->dirty_log_pages,
 971                                     0, &start, &end, EXTENT_DIRTY);
 972                 if (ret)
 973                         break;
 974
 975                 clear_extent_dirty(&log_root_tree->dirty_log_pages,
 976                                    start, end, GFP_NOFS);
 977         }
 978         eb = fs_info->log_root_tree->node;
 979
 980         WARN_ON(btrfs_header_level(eb) != 0);
 981         WARN_ON(btrfs_header_nritems(eb) != 0);
 982
 983         ret = btrfs_free_reserved_extent(fs_info->tree_root,
 984                                 eb->start, eb->len);
 985         BUG_ON(ret);
 986
 987         free_extent_buffer(eb);
 988         kfree(fs_info->log_root_tree);
 989         fs_info->log_root_tree = NULL;
 990         return 0;
 991 }
 992
 993 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
 994                                          struct btrfs_fs_info *fs_info)
 995 {
 996         struct btrfs_root *root;
 997         struct btrfs_root *tree_root = fs_info->tree_root;
 998         struct extent_buffer *leaf;
 999
1000         root = kzalloc(sizeof(*root), GFP_NOFS);
1001         if (!root)
1002                 return ERR_PTR(-ENOMEM);
1003
1004         __setup_root(tree_root->nodesize, tree_root->leafsize,
1005                      tree_root->sectorsize, tree_root->stripesize,
1006                      root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1007
1008         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1009         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1010         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1011         /*
1012          * log trees do not get reference counted because they go away
1013          * before a real commit is actually done.  They do store pointers
1014          * to file data extents, and those reference counts still get
1015          * updated (along with back refs to the log tree).
1016          */
1017         root->ref_cows = 0;
1018
1019         leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1020                                       0, BTRFS_TREE_LOG_OBJECTID,
1021                                       trans->transid, 0, 0, 0);
1022         if (IS_ERR(leaf)) {
1023                 kfree(root);
1024                 return ERR_CAST(leaf);
1025         }
1026
1027         root->node = leaf;
1028         btrfs_set_header_nritems(root->node, 0);
1029         btrfs_set_header_level(root->node, 0);
1030         btrfs_set_header_bytenr(root->node, root->node->start);
1031         btrfs_set_header_generation(root->node, trans->transid);
1032         btrfs_set_header_owner(root->node, BTRFS_TREE_LOG_OBJECTID);
1033
1034         write_extent_buffer(root->node, root->fs_info->fsid,
1035                             (unsigned long)btrfs_header_fsid(root->node),
1036                             BTRFS_FSID_SIZE);
1037         btrfs_mark_buffer_dirty(root->node);
1038         btrfs_tree_unlock(root->node);
1039         return root;
1040 }
1041
1042 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1043                              struct btrfs_fs_info *fs_info)
1044 {
1045         struct btrfs_root *log_root;
1046
1047         log_root = alloc_log_tree(trans, fs_info);
1048         if (IS_ERR(log_root))
1049                 return PTR_ERR(log_root);
1050         WARN_ON(fs_info->log_root_tree);
1051         fs_info->log_root_tree = log_root;
1052         return 0;
1053 }
1054
1055 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1056                        struct btrfs_root *root)
1057 {
1058         struct btrfs_root *log_root;
1059         struct btrfs_inode_item *inode_item;
1060
1061         log_root = alloc_log_tree(trans, root->fs_info);
1062         if (IS_ERR(log_root))
1063                 return PTR_ERR(log_root);
1064
1065         log_root->last_trans = trans->transid;
1066         log_root->root_key.offset = root->root_key.objectid;
1067
1068         inode_item = &log_root->root_item.inode;
1069         inode_item->generation = cpu_to_le64(1);
1070         inode_item->size = cpu_to_le64(3);
1071         inode_item->nlink = cpu_to_le32(1);
1072         inode_item->nbytes = cpu_to_le64(root->leafsize);
1073         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1074
1075         btrfs_set_root_bytenr(&log_root->root_item, log_root->node->start);
1076         btrfs_set_root_generation(&log_root->root_item, trans->transid);
1077
1078         WARN_ON(root->log_root);
1079         root->log_root = log_root;
1080         root->log_transid = 0;
1081         return 0;
1082 }
1083
1084 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1085                                                struct btrfs_key *location)
1086 {
1087         struct btrfs_root *root;
1088         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1089         struct btrfs_path *path;
1090         struct extent_buffer *l;
1091         u64 highest_inode;
1092         u64 generation;
1093         u32 blocksize;
1094         int ret = 0;
1095
1096         root = kzalloc(sizeof(*root), GFP_NOFS);
1097         if (!root)
1098                 return ERR_PTR(-ENOMEM);
1099         if (location->offset == (u64)-1) {
1100                 ret = find_and_setup_root(tree_root, fs_info,
1101                                           location->objectid, root);
1102                 if (ret) {
1103                         kfree(root);
1104                         return ERR_PTR(ret);
1105                 }
1106                 goto insert;
1107         }
1108
1109         __setup_root(tree_root->nodesize, tree_root->leafsize,
1110                      tree_root->sectorsize, tree_root->stripesize,
1111                      root, fs_info, location->objectid);
1112
1113         path = btrfs_alloc_path();
1114         BUG_ON(!path);
1115         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1116         if (ret != 0) {
1117                 if (ret > 0)
1118                         ret = -ENOENT;
1119                 goto out;
1120         }
1121         l = path->nodes[0];
1122         read_extent_buffer(l, &root->root_item,
1123                btrfs_item_ptr_offset(l, path->slots[0]),
1124                sizeof(root->root_item));
1125         memcpy(&root->root_key, location, sizeof(*location));
1126         ret = 0;
1127 out:
1128         btrfs_release_path(root, path);
1129         btrfs_free_path(path);
1130         if (ret) {
1131                 kfree(root);
1132                 return ERR_PTR(ret);
1133         }
1134         generation = btrfs_root_generation(&root->root_item);
1135         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1136         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1137                                      blocksize, generation);
1138         BUG_ON(!root->node);
1139 insert:
1140         if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1141                 root->ref_cows = 1;
1142                 ret = btrfs_find_highest_inode(root, &highest_inode);
1143                 if (ret == 0) {
1144                         root->highest_inode = highest_inode;
1145                         root->last_inode_alloc = highest_inode;
1146                 }
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
1183         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1184                                  (unsigned long)location->objectid);
1185         if (root)
1186                 return root;
1187
1188         root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1189         if (IS_ERR(root))
1190                 return root;
1191
1192         set_anon_super(&root->anon_super, NULL);
1193
1194         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1195                                 (unsigned long)root->root_key.objectid,
1196                                 root);
1197         if (ret) {
1198                 free_extent_buffer(root->node);
1199                 kfree(root);
1200                 return ERR_PTR(ret);
1201         }
1202         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
1203                 ret = btrfs_find_dead_roots(fs_info->tree_root,
1204                                             root->root_key.objectid, root);
1205                 BUG_ON(ret);
1206                 btrfs_orphan_cleanup(root);
1207         }
1208         return root;
1209 }
1210
1211 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1212                                       struct btrfs_key *location,
1213                                       const char *name, int namelen)
1214 {
1215         struct btrfs_root *root;
1216         int ret;
1217
1218         root = btrfs_read_fs_root_no_name(fs_info, location);
1219         if (!root)
1220                 return NULL;
1221
1222         if (root->in_sysfs)
1223                 return root;
1224
1225         ret = btrfs_set_root_name(root, name, namelen);
1226         if (ret) {
1227                 free_extent_buffer(root->node);
1228                 kfree(root);
1229                 return ERR_PTR(ret);
1230         }
1231 #if 0
1232         ret = btrfs_sysfs_add_root(root);
1233         if (ret) {
1234                 free_extent_buffer(root->node);
1235                 kfree(root->name);
1236                 kfree(root);
1237                 return ERR_PTR(ret);
1238         }
1239 #endif
1240         root->in_sysfs = 1;
1241         return root;
1242 }
1243
1244 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1245 {
1246         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1247         int ret = 0;
1248         struct btrfs_device *device;
1249         struct backing_dev_info *bdi;
1250 #if 0
1251         if ((bdi_bits & (1 << BDI_write_congested)) &&
1252             btrfs_congested_async(info, 0))
1253                 return 1;
1254 #endif
1255         list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1256                 if (!device->bdev)
1257                         continue;
1258                 bdi = blk_get_backing_dev_info(device->bdev);
1259                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1260                         ret = 1;
1261                         break;
1262                 }
1263         }
1264         return ret;
1265 }
1266
1267 /*
1268  * this unplugs every device on the box, and it is only used when page
1269  * is null
1270  */
1271 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1272 {
1273         struct btrfs_device *device;
1274         struct btrfs_fs_info *info;
1275
1276         info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1277         list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1278                 if (!device->bdev)
1279                         continue;
1280
1281                 bdi = blk_get_backing_dev_info(device->bdev);
1282                 if (bdi->unplug_io_fn)
1283                         bdi->unplug_io_fn(bdi, page);
1284         }
1285 }
1286
1287 static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1288 {
1289         struct inode *inode;
1290         struct extent_map_tree *em_tree;
1291         struct extent_map *em;
1292         struct address_space *mapping;
1293         u64 offset;
1294
1295         /* the generic O_DIRECT read code does this */
1296         if (1 || !page) {
1297                 __unplug_io_fn(bdi, page);
1298                 return;
1299         }
1300
1301         /*
1302          * page->mapping may change at any time.  Get a consistent copy
1303          * and use that for everything below
1304          */
1305         smp_mb();
1306         mapping = page->mapping;
1307         if (!mapping)
1308                 return;
1309
1310         inode = mapping->host;
1311
1312         /*
1313          * don't do the expensive searching for a small number of
1314          * devices
1315          */
1316         if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1317                 __unplug_io_fn(bdi, page);
1318                 return;
1319         }
1320
1321         offset = page_offset(page);
1322
1323         em_tree = &BTRFS_I(inode)->extent_tree;
1324         spin_lock(&em_tree->lock);
1325         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1326         spin_unlock(&em_tree->lock);
1327         if (!em) {
1328                 __unplug_io_fn(bdi, page);
1329                 return;
1330         }
1331
1332         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1333                 free_extent_map(em);
1334                 __unplug_io_fn(bdi, page);
1335                 return;
1336         }
1337         offset = offset - em->start;
1338         btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1339                           em->block_start + offset, page);
1340         free_extent_map(em);
1341 }
1342
1343 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1344 {
1345         bdi_init(bdi);
1346         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1347         bdi->state              = 0;
1348         bdi->capabilities       = default_backing_dev_info.capabilities;
1349         bdi->unplug_io_fn       = btrfs_unplug_io_fn;
1350         bdi->unplug_io_data     = info;
1351         bdi->congested_fn       = btrfs_congested_fn;
1352         bdi->congested_data     = info;
1353         return 0;
1354 }
1355
1356 static int bio_ready_for_csum(struct bio *bio)
1357 {
1358         u64 length = 0;
1359         u64 buf_len = 0;
1360         u64 start = 0;
1361         struct page *page;
1362         struct extent_io_tree *io_tree = NULL;
1363         struct btrfs_fs_info *info = NULL;
1364         struct bio_vec *bvec;
1365         int i;
1366         int ret;
1367
1368         bio_for_each_segment(bvec, bio, i) {
1369                 page = bvec->bv_page;
1370                 if (page->private == EXTENT_PAGE_PRIVATE) {
1371                         length += bvec->bv_len;
1372                         continue;
1373                 }
1374                 if (!page->private) {
1375                         length += bvec->bv_len;
1376                         continue;
1377                 }
1378                 length = bvec->bv_len;
1379                 buf_len = page->private >> 2;
1380                 start = page_offset(page) + bvec->bv_offset;
1381                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1382                 info = BTRFS_I(page->mapping->host)->root->fs_info;
1383         }
1384         /* are we fully contained in this bio? */
1385         if (buf_len <= length)
1386                 return 1;
1387
1388         ret = extent_range_uptodate(io_tree, start + length,
1389                                     start + buf_len - 1);
1390         if (ret == 1)
1391                 return ret;
1392         return ret;
1393 }
1394
1395 /*
1396  * called by the kthread helper functions to finally call the bio end_io
1397  * functions.  This is where read checksum verification actually happens
1398  */
1399 static void end_workqueue_fn(struct btrfs_work *work)
1400 {
1401         struct bio *bio;
1402         struct end_io_wq *end_io_wq;
1403         struct btrfs_fs_info *fs_info;
1404         int error;
1405
1406         end_io_wq = container_of(work, struct end_io_wq, work);
1407         bio = end_io_wq->bio;
1408         fs_info = end_io_wq->info;
1409
1410         /* metadata bio reads are special because the whole tree block must
1411          * be checksummed at once.  This makes sure the entire block is in
1412          * ram and up to date before trying to verify things.  For
1413          * blocksize <= pagesize, it is basically a noop
1414          */
1415         if (!(bio->bi_rw & (1 << BIO_RW)) && end_io_wq->metadata &&
1416             !bio_ready_for_csum(bio)) {
1417                 btrfs_queue_worker(&fs_info->endio_meta_workers,
1418                                    &end_io_wq->work);
1419                 return;
1420         }
1421         error = end_io_wq->error;
1422         bio->bi_private = end_io_wq->private;
1423         bio->bi_end_io = end_io_wq->end_io;
1424         kfree(end_io_wq);
1425         bio_endio(bio, error);
1426 }
1427
1428 static int cleaner_kthread(void *arg)
1429 {
1430         struct btrfs_root *root = arg;
1431
1432         do {
1433                 smp_mb();
1434                 if (root->fs_info->closing)
1435                         break;
1436
1437                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1438                 mutex_lock(&root->fs_info->cleaner_mutex);
1439                 btrfs_clean_old_snapshots(root);
1440                 mutex_unlock(&root->fs_info->cleaner_mutex);
1441
1442                 if (freezing(current)) {
1443                         refrigerator();
1444                 } else {
1445                         smp_mb();
1446                         if (root->fs_info->closing)
1447                                 break;
1448                         set_current_state(TASK_INTERRUPTIBLE);
1449                         schedule();
1450                         __set_current_state(TASK_RUNNING);
1451                 }
1452         } while (!kthread_should_stop());
1453         return 0;
1454 }
1455
1456 static int transaction_kthread(void *arg)
1457 {
1458         struct btrfs_root *root = arg;
1459         struct btrfs_trans_handle *trans;
1460         struct btrfs_transaction *cur;
1461         unsigned long now;
1462         unsigned long delay;
1463         int ret;
1464
1465         do {
1466                 smp_mb();
1467                 if (root->fs_info->closing)
1468                         break;
1469
1470                 delay = HZ * 30;
1471                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1472                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1473
1474                 if (root->fs_info->total_ref_cache_size > 20 * 1024 * 1024) {
1475                         printk(KERN_INFO "btrfs: total reference cache "
1476                                "size %llu\n",
1477                                root->fs_info->total_ref_cache_size);
1478                 }
1479
1480                 mutex_lock(&root->fs_info->trans_mutex);
1481                 cur = root->fs_info->running_transaction;
1482                 if (!cur) {
1483                         mutex_unlock(&root->fs_info->trans_mutex);
1484                         goto sleep;
1485                 }
1486
1487                 now = get_seconds();
1488                 if (now < cur->start_time || now - cur->start_time < 30) {
1489                         mutex_unlock(&root->fs_info->trans_mutex);
1490                         delay = HZ * 5;
1491                         goto sleep;
1492                 }
1493                 mutex_unlock(&root->fs_info->trans_mutex);
1494                 trans = btrfs_start_transaction(root, 1);
1495                 ret = btrfs_commit_transaction(trans, root);
1496 sleep:
1497                 wake_up_process(root->fs_info->cleaner_kthread);
1498                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1499
1500                 if (freezing(current)) {
1501                         refrigerator();
1502                 } else {
1503                         if (root->fs_info->closing)
1504                                 break;
1505                         set_current_state(TASK_INTERRUPTIBLE);
1506                         schedule_timeout(delay);
1507                         __set_current_state(TASK_RUNNING);
1508                 }
1509         } while (!kthread_should_stop());
1510         return 0;
1511 }
1512
1513 struct btrfs_root *open_ctree(struct super_block *sb,
1514                               struct btrfs_fs_devices *fs_devices,
1515                               char *options)
1516 {
1517         u32 sectorsize;
1518         u32 nodesize;
1519         u32 leafsize;
1520         u32 blocksize;
1521         u32 stripesize;
1522         u64 generation;
1523         u64 features;
1524         struct btrfs_key location;
1525         struct buffer_head *bh;
1526         struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1527                                                  GFP_NOFS);
1528         struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1529                                                  GFP_NOFS);
1530         struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
1531                                                GFP_NOFS);
1532         struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1533                                                 GFP_NOFS);
1534         struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1535                                                 GFP_NOFS);
1536         struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1537                                               GFP_NOFS);
1538         struct btrfs_root *log_tree_root;
1539
1540         int ret;
1541         int err = -EINVAL;
1542
1543         struct btrfs_super_block *disk_super;
1544
1545         if (!extent_root || !tree_root || !fs_info ||
1546             !chunk_root || !dev_root || !csum_root) {
1547                 err = -ENOMEM;
1548                 goto fail;
1549         }
1550         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1551         INIT_LIST_HEAD(&fs_info->trans_list);
1552         INIT_LIST_HEAD(&fs_info->dead_roots);
1553         INIT_LIST_HEAD(&fs_info->hashers);
1554         INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1555         spin_lock_init(&fs_info->delalloc_lock);
1556         spin_lock_init(&fs_info->new_trans_lock);
1557         spin_lock_init(&fs_info->ref_cache_lock);
1558
1559         init_completion(&fs_info->kobj_unregister);
1560         fs_info->tree_root = tree_root;
1561         fs_info->extent_root = extent_root;
1562         fs_info->csum_root = csum_root;
1563         fs_info->chunk_root = chunk_root;
1564         fs_info->dev_root = dev_root;
1565         fs_info->fs_devices = fs_devices;
1566         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1567         INIT_LIST_HEAD(&fs_info->space_info);
1568         btrfs_mapping_init(&fs_info->mapping_tree);
1569         atomic_set(&fs_info->nr_async_submits, 0);
1570         atomic_set(&fs_info->async_delalloc_pages, 0);
1571         atomic_set(&fs_info->async_submit_draining, 0);
1572         atomic_set(&fs_info->nr_async_bios, 0);
1573         atomic_set(&fs_info->throttles, 0);
1574         atomic_set(&fs_info->throttle_gen, 0);
1575         fs_info->sb = sb;
1576         fs_info->max_extent = (u64)-1;
1577         fs_info->max_inline = 8192 * 1024;
1578         setup_bdi(fs_info, &fs_info->bdi);
1579         fs_info->btree_inode = new_inode(sb);
1580         fs_info->btree_inode->i_ino = 1;
1581         fs_info->btree_inode->i_nlink = 1;
1582
1583         fs_info->thread_pool_size = min_t(unsigned long,
1584                                           num_online_cpus() + 2, 8);
1585
1586         INIT_LIST_HEAD(&fs_info->ordered_extents);
1587         spin_lock_init(&fs_info->ordered_extent_lock);
1588
1589         sb->s_blocksize = 4096;
1590         sb->s_blocksize_bits = blksize_bits(4096);
1591
1592         /*
1593          * we set the i_size on the btree inode to the max possible int.
1594          * the real end of the address space is determined by all of
1595          * the devices in the system
1596          */
1597         fs_info->btree_inode->i_size = OFFSET_MAX;
1598         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1599         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1600
1601         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1602                              fs_info->btree_inode->i_mapping,
1603                              GFP_NOFS);
1604         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1605                              GFP_NOFS);
1606
1607         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1608
1609         spin_lock_init(&fs_info->block_group_cache_lock);
1610         fs_info->block_group_cache_tree.rb_node = NULL;
1611
1612         extent_io_tree_init(&fs_info->pinned_extents,
1613                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1614         extent_io_tree_init(&fs_info->pending_del,
1615                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1616         extent_io_tree_init(&fs_info->extent_ins,
1617                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1618         fs_info->do_barriers = 1;
1619
1620         INIT_LIST_HEAD(&fs_info->dead_reloc_roots);
1621         btrfs_leaf_ref_tree_init(&fs_info->reloc_ref_tree);
1622         btrfs_leaf_ref_tree_init(&fs_info->shared_ref_tree);
1623
1624         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1625         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1626                sizeof(struct btrfs_key));
1627         insert_inode_hash(fs_info->btree_inode);
1628
1629         mutex_init(&fs_info->trans_mutex);
1630         mutex_init(&fs_info->tree_log_mutex);
1631         mutex_init(&fs_info->drop_mutex);
1632         mutex_init(&fs_info->extent_ins_mutex);
1633         mutex_init(&fs_info->pinned_mutex);
1634         mutex_init(&fs_info->chunk_mutex);
1635         mutex_init(&fs_info->transaction_kthread_mutex);
1636         mutex_init(&fs_info->cleaner_mutex);
1637         mutex_init(&fs_info->volume_mutex);
1638         mutex_init(&fs_info->tree_reloc_mutex);
1639         init_waitqueue_head(&fs_info->transaction_throttle);
1640         init_waitqueue_head(&fs_info->transaction_wait);
1641         init_waitqueue_head(&fs_info->async_submit_wait);
1642
1643         __setup_root(4096, 4096, 4096, 4096, tree_root,
1644                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1645
1646
1647         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1648         if (!bh)
1649                 goto fail_iput;
1650
1651         memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1652         memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1653                sizeof(fs_info->super_for_commit));
1654         brelse(bh);
1655
1656         memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1657
1658         disk_super = &fs_info->super_copy;
1659         if (!btrfs_super_root(disk_super))
1660                 goto fail_iput;
1661
1662         ret = btrfs_parse_options(tree_root, options);
1663         if (ret) {
1664                 err = ret;
1665                 goto fail_iput;
1666         }
1667
1668         features = btrfs_super_incompat_flags(disk_super) &
1669                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1670         if (features) {
1671                 printk(KERN_ERR "BTRFS: couldn't mount because of "
1672                        "unsupported optional features (%Lx).\n",
1673                        features);
1674                 err = -EINVAL;
1675                 goto fail_iput;
1676         }
1677
1678         features = btrfs_super_compat_ro_flags(disk_super) &
1679                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1680         if (!(sb->s_flags & MS_RDONLY) && features) {
1681                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1682                        "unsupported option features (%Lx).\n",
1683                        features);
1684                 err = -EINVAL;
1685                 goto fail_iput;
1686         }
1687
1688         /*
1689          * we need to start all the end_io workers up front because the
1690          * queue work function gets called at interrupt time, and so it
1691          * cannot dynamically grow.
1692          */
1693         btrfs_init_workers(&fs_info->workers, "worker",
1694                            fs_info->thread_pool_size);
1695
1696         btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1697                            fs_info->thread_pool_size);
1698
1699         btrfs_init_workers(&fs_info->submit_workers, "submit",
1700                            min_t(u64, fs_devices->num_devices,
1701                            fs_info->thread_pool_size));
1702
1703         /* a higher idle thresh on the submit workers makes it much more
1704          * likely that bios will be send down in a sane order to the
1705          * devices
1706          */
1707         fs_info->submit_workers.idle_thresh = 64;
1708
1709         fs_info->workers.idle_thresh = 16;
1710         fs_info->workers.ordered = 1;
1711
1712         fs_info->delalloc_workers.idle_thresh = 2;
1713         fs_info->delalloc_workers.ordered = 1;
1714
1715         btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1);
1716         btrfs_init_workers(&fs_info->endio_workers, "endio",
1717                            fs_info->thread_pool_size);
1718         btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1719                            fs_info->thread_pool_size);
1720         btrfs_init_workers(&fs_info->endio_meta_write_workers,
1721                            "endio-meta-write", fs_info->thread_pool_size);
1722         btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1723                            fs_info->thread_pool_size);
1724
1725         /*
1726          * endios are largely parallel and should have a very
1727          * low idle thresh
1728          */
1729         fs_info->endio_workers.idle_thresh = 4;
1730         fs_info->endio_meta_workers.idle_thresh = 4;
1731
1732         fs_info->endio_write_workers.idle_thresh = 64;
1733         fs_info->endio_meta_write_workers.idle_thresh = 64;
1734
1735         btrfs_start_workers(&fs_info->workers, 1);
1736         btrfs_start_workers(&fs_info->submit_workers, 1);
1737         btrfs_start_workers(&fs_info->delalloc_workers, 1);
1738         btrfs_start_workers(&fs_info->fixup_workers, 1);
1739         btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
1740         btrfs_start_workers(&fs_info->endio_meta_workers,
1741                             fs_info->thread_pool_size);
1742         btrfs_start_workers(&fs_info->endio_meta_write_workers,
1743                             fs_info->thread_pool_size);
1744         btrfs_start_workers(&fs_info->endio_write_workers,
1745                             fs_info->thread_pool_size);
1746
1747         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1748         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1749                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1750
1751         nodesize = btrfs_super_nodesize(disk_super);
1752         leafsize = btrfs_super_leafsize(disk_super);
1753         sectorsize = btrfs_super_sectorsize(disk_super);
1754         stripesize = btrfs_super_stripesize(disk_super);
1755         tree_root->nodesize = nodesize;
1756         tree_root->leafsize = leafsize;
1757         tree_root->sectorsize = sectorsize;
1758         tree_root->stripesize = stripesize;
1759
1760         sb->s_blocksize = sectorsize;
1761         sb->s_blocksize_bits = blksize_bits(sectorsize);
1762
1763         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1764                     sizeof(disk_super->magic))) {
1765                 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
1766                 goto fail_sb_buffer;
1767         }
1768
1769         mutex_lock(&fs_info->chunk_mutex);
1770         ret = btrfs_read_sys_array(tree_root);
1771         mutex_unlock(&fs_info->chunk_mutex);
1772         if (ret) {
1773                 printk(KERN_WARNING "btrfs: failed to read the system "
1774                        "array on %s\n", sb->s_id);
1775                 goto fail_sys_array;
1776         }
1777
1778         blocksize = btrfs_level_size(tree_root,
1779                                      btrfs_super_chunk_root_level(disk_super));
1780         generation = btrfs_super_chunk_root_generation(disk_super);
1781
1782         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1783                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1784
1785         chunk_root->node = read_tree_block(chunk_root,
1786                                            btrfs_super_chunk_root(disk_super),
1787                                            blocksize, generation);
1788         BUG_ON(!chunk_root->node);
1789
1790         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1791            (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1792            BTRFS_UUID_SIZE);
1793
1794         mutex_lock(&fs_info->chunk_mutex);
1795         ret = btrfs_read_chunk_tree(chunk_root);
1796         mutex_unlock(&fs_info->chunk_mutex);
1797         if (ret) {
1798                 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
1799                        sb->s_id);
1800                 goto fail_chunk_root;
1801         }
1802
1803         btrfs_close_extra_devices(fs_devices);
1804
1805         blocksize = btrfs_level_size(tree_root,
1806                                      btrfs_super_root_level(disk_super));
1807         generation = btrfs_super_generation(disk_super);
1808
1809         tree_root->node = read_tree_block(tree_root,
1810                                           btrfs_super_root(disk_super),
1811                                           blocksize, generation);
1812         if (!tree_root->node)
1813                 goto fail_chunk_root;
1814
1815
1816         ret = find_and_setup_root(tree_root, fs_info,
1817                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1818         if (ret)
1819                 goto fail_tree_root;
1820         extent_root->track_dirty = 1;
1821
1822         ret = find_and_setup_root(tree_root, fs_info,
1823                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
1824         dev_root->track_dirty = 1;
1825         if (ret)
1826                 goto fail_extent_root;
1827
1828         ret = find_and_setup_root(tree_root, fs_info,
1829                                   BTRFS_CSUM_TREE_OBJECTID, csum_root);
1830         if (ret)
1831                 goto fail_extent_root;
1832
1833         csum_root->track_dirty = 1;
1834
1835         btrfs_read_block_groups(extent_root);
1836
1837         fs_info->generation = generation;
1838         fs_info->last_trans_committed = generation;
1839         fs_info->data_alloc_profile = (u64)-1;
1840         fs_info->metadata_alloc_profile = (u64)-1;
1841         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1842         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1843                                                "btrfs-cleaner");
1844         if (IS_ERR(fs_info->cleaner_kthread))
1845                 goto fail_csum_root;
1846
1847         fs_info->transaction_kthread = kthread_run(transaction_kthread,
1848                                                    tree_root,
1849                                                    "btrfs-transaction");
1850         if (IS_ERR(fs_info->transaction_kthread))
1851                 goto fail_cleaner;
1852
1853         if (btrfs_super_log_root(disk_super) != 0) {
1854                 u64 bytenr = btrfs_super_log_root(disk_super);
1855
1856                 if (fs_devices->rw_devices == 0) {
1857                         printk(KERN_WARNING "Btrfs log replay required "
1858                                "on RO media\n");
1859                         err = -EIO;
1860                         goto fail_trans_kthread;
1861                 }
1862                 blocksize =
1863                      btrfs_level_size(tree_root,
1864                                       btrfs_super_log_root_level(disk_super));
1865
1866                 log_tree_root = kzalloc(sizeof(struct btrfs_root),
1867                                                       GFP_NOFS);
1868
1869                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1870                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1871
1872                 log_tree_root->node = read_tree_block(tree_root, bytenr,
1873                                                       blocksize,
1874                                                       generation + 1);
1875                 ret = btrfs_recover_log_trees(log_tree_root);
1876                 BUG_ON(ret);
1877
1878                 if (sb->s_flags & MS_RDONLY) {
1879                         ret =  btrfs_commit_super(tree_root);
1880                         BUG_ON(ret);
1881                 }
1882         }
1883
1884         if (!(sb->s_flags & MS_RDONLY)) {
1885                 ret = btrfs_cleanup_reloc_trees(tree_root);
1886                 BUG_ON(ret);
1887         }
1888
1889         location.objectid = BTRFS_FS_TREE_OBJECTID;
1890         location.type = BTRFS_ROOT_ITEM_KEY;
1891         location.offset = (u64)-1;
1892
1893         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
1894         if (!fs_info->fs_root)
1895                 goto fail_trans_kthread;
1896         return tree_root;
1897
1898 fail_trans_kthread:
1899         kthread_stop(fs_info->transaction_kthread);
1900 fail_cleaner:
1901         kthread_stop(fs_info->cleaner_kthread);
1902
1903         /*
1904          * make sure we're done with the btree inode before we stop our
1905          * kthreads
1906          */
1907         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1908         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
1909
1910 fail_csum_root:
1911         free_extent_buffer(csum_root->node);
1912 fail_extent_root:
1913         free_extent_buffer(extent_root->node);
1914 fail_tree_root:
1915         free_extent_buffer(tree_root->node);
1916 fail_chunk_root:
1917         free_extent_buffer(chunk_root->node);
1918 fail_sys_array:
1919         free_extent_buffer(dev_root->node);
1920 fail_sb_buffer:
1921         btrfs_stop_workers(&fs_info->fixup_workers);
1922         btrfs_stop_workers(&fs_info->delalloc_workers);
1923         btrfs_stop_workers(&fs_info->workers);
1924         btrfs_stop_workers(&fs_info->endio_workers);
1925         btrfs_stop_workers(&fs_info->endio_meta_workers);
1926         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
1927         btrfs_stop_workers(&fs_info->endio_write_workers);
1928         btrfs_stop_workers(&fs_info->submit_workers);
1929 fail_iput:
1930         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
1931         iput(fs_info->btree_inode);
1932
1933         btrfs_close_devices(fs_info->fs_devices);
1934         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1935         bdi_destroy(&fs_info->bdi);
1936
1937 fail:
1938         kfree(extent_root);
1939         kfree(tree_root);
1940         kfree(fs_info);
1941         kfree(chunk_root);
1942         kfree(dev_root);
1943         kfree(csum_root);
1944         return ERR_PTR(err);
1945 }
1946
1947 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1948 {
1949         char b[BDEVNAME_SIZE];
1950
1951         if (uptodate) {
1952                 set_buffer_uptodate(bh);
1953         } else {
1954                 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1955                         printk(KERN_WARNING "lost page write due to "
1956                                         "I/O error on %s\n",
1957                                        bdevname(bh->b_bdev, b));
1958                 }
1959                 /* note, we dont' set_buffer_write_io_error because we have
1960                  * our own ways of dealing with the IO errors
1961                  */
1962                 clear_buffer_uptodate(bh);
1963         }
1964         unlock_buffer(bh);
1965         put_bh(bh);
1966 }
1967
1968 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
1969 {
1970         struct buffer_head *bh;
1971         struct buffer_head *latest = NULL;
1972         struct btrfs_super_block *super;
1973         int i;
1974         u64 transid = 0;
1975         u64 bytenr;
1976
1977         /* we would like to check all the supers, but that would make
1978          * a btrfs mount succeed after a mkfs from a different FS.
1979          * So, we need to add a special mount option to scan for
1980          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1981          */
1982         for (i = 0; i < 1; i++) {
1983                 bytenr = btrfs_sb_offset(i);
1984                 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
1985                         break;
1986                 bh = __bread(bdev, bytenr / 4096, 4096);
1987                 if (!bh)
1988                         continue;
1989
1990                 super = (struct btrfs_super_block *)bh->b_data;
1991                 if (btrfs_super_bytenr(super) != bytenr ||
1992                     strncmp((char *)(&super->magic), BTRFS_MAGIC,
1993                             sizeof(super->magic))) {
1994                         brelse(bh);
1995                         continue;
1996                 }
1997
1998                 if (!latest || btrfs_super_generation(super) > transid) {
1999                         brelse(latest);
2000                         latest = bh;
2001                         transid = btrfs_super_generation(super);
2002                 } else {
2003                         brelse(bh);
2004                 }
2005         }
2006         return latest;
2007 }
2008
2009 static int write_dev_supers(struct btrfs_device *device,
2010                             struct btrfs_super_block *sb,
2011                             int do_barriers, int wait, int max_mirrors)
2012 {
2013         struct buffer_head *bh;
2014         int i;
2015         int ret;
2016         int errors = 0;
2017         u32 crc;
2018         u64 bytenr;
2019         int last_barrier = 0;
2020
2021         if (max_mirrors == 0)
2022                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2023
2024         /* make sure only the last submit_bh does a barrier */
2025         if (do_barriers) {
2026                 for (i = 0; i < max_mirrors; i++) {
2027                         bytenr = btrfs_sb_offset(i);
2028                         if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2029                             device->total_bytes)
2030                                 break;
2031                         last_barrier = i;
2032                 }
2033         }
2034
2035         for (i = 0; i < max_mirrors; i++) {
2036                 bytenr = btrfs_sb_offset(i);
2037                 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2038                         break;
2039
2040                 if (wait) {
2041                         bh = __find_get_block(device->bdev, bytenr / 4096,
2042                                               BTRFS_SUPER_INFO_SIZE);
2043                         BUG_ON(!bh);
2044                         brelse(bh);
2045                         wait_on_buffer(bh);
2046                         if (buffer_uptodate(bh)) {
2047                                 brelse(bh);
2048                                 continue;
2049                         }
2050                 } else {
2051                         btrfs_set_super_bytenr(sb, bytenr);
2052
2053                         crc = ~(u32)0;
2054                         crc = btrfs_csum_data(NULL, (char *)sb +
2055                                               BTRFS_CSUM_SIZE, crc,
2056                                               BTRFS_SUPER_INFO_SIZE -
2057                                               BTRFS_CSUM_SIZE);
2058                         btrfs_csum_final(crc, sb->csum);
2059
2060                         bh = __getblk(device->bdev, bytenr / 4096,
2061                                       BTRFS_SUPER_INFO_SIZE);
2062                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2063
2064                         set_buffer_uptodate(bh);
2065                         get_bh(bh);
2066                         lock_buffer(bh);
2067                         bh->b_end_io = btrfs_end_buffer_write_sync;
2068                 }
2069
2070                 if (i == last_barrier && do_barriers && device->barriers) {
2071                         ret = submit_bh(WRITE_BARRIER, bh);
2072                         if (ret == -EOPNOTSUPP) {
2073                                 printk("btrfs: disabling barriers on dev %s\n",
2074                                        device->name);
2075                                 set_buffer_uptodate(bh);
2076                                 device->barriers = 0;
2077                                 get_bh(bh);
2078                                 lock_buffer(bh);
2079                                 ret = submit_bh(WRITE, bh);
2080                         }
2081                 } else {
2082                         ret = submit_bh(WRITE, bh);
2083                 }
2084
2085                 if (!ret && wait) {
2086                         wait_on_buffer(bh);
2087                         if (!buffer_uptodate(bh))
2088                                 errors++;
2089                 } else if (ret) {
2090                         errors++;
2091                 }
2092                 if (wait)
2093                         brelse(bh);
2094         }
2095         return errors < i ? 0 : -1;
2096 }
2097
2098 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2099 {
2100         struct list_head *head = &root->fs_info->fs_devices->devices;
2101         struct btrfs_device *dev;
2102         struct btrfs_super_block *sb;
2103         struct btrfs_dev_item *dev_item;
2104         int ret;
2105         int do_barriers;
2106         int max_errors;
2107         int total_errors = 0;
2108         u64 flags;
2109
2110         max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2111         do_barriers = !btrfs_test_opt(root, NOBARRIER);
2112
2113         sb = &root->fs_info->super_for_commit;
2114         dev_item = &sb->dev_item;
2115         list_for_each_entry(dev, head, dev_list) {
2116                 if (!dev->bdev) {
2117                         total_errors++;
2118                         continue;
2119                 }
2120                 if (!dev->in_fs_metadata || !dev->writeable)
2121                         continue;
2122
2123                 btrfs_set_stack_device_generation(dev_item, 0);
2124                 btrfs_set_stack_device_type(dev_item, dev->type);
2125                 btrfs_set_stack_device_id(dev_item, dev->devid);
2126                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2127                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2128                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2129                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2130                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2131                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2132                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2133
2134                 flags = btrfs_super_flags(sb);
2135                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2136
2137                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2138                 if (ret)
2139                         total_errors++;
2140         }
2141         if (total_errors > max_errors) {
2142                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2143                        total_errors);
2144                 BUG();
2145         }
2146
2147         total_errors = 0;
2148         list_for_each_entry(dev, head, dev_list) {
2149                 if (!dev->bdev)
2150                         continue;
2151                 if (!dev->in_fs_metadata || !dev->writeable)
2152                         continue;
2153
2154                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2155                 if (ret)
2156                         total_errors++;
2157         }
2158         if (total_errors > max_errors) {
2159                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2160                        total_errors);
2161                 BUG();
2162         }
2163         return 0;
2164 }
2165
2166 int write_ctree_super(struct btrfs_trans_handle *trans,
2167                       struct btrfs_root *root, int max_mirrors)
2168 {
2169         int ret;
2170
2171         ret = write_all_supers(root, max_mirrors);
2172         return ret;
2173 }
2174
2175 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2176 {
2177         radix_tree_delete(&fs_info->fs_roots_radix,
2178                           (unsigned long)root->root_key.objectid);
2179         if (root->anon_super.s_dev) {
2180                 down_write(&root->anon_super.s_umount);
2181                 kill_anon_super(&root->anon_super);
2182         }
2183         if (root->node)
2184                 free_extent_buffer(root->node);
2185         if (root->commit_root)
2186                 free_extent_buffer(root->commit_root);
2187         kfree(root->name);
2188         kfree(root);
2189         return 0;
2190 }
2191
2192 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2193 {
2194         int ret;
2195         struct btrfs_root *gang[8];
2196         int i;
2197
2198         while (1) {
2199                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2200                                              (void **)gang, 0,
2201                                              ARRAY_SIZE(gang));
2202                 if (!ret)
2203                         break;
2204                 for (i = 0; i < ret; i++)
2205                         btrfs_free_fs_root(fs_info, gang[i]);
2206         }
2207         return 0;
2208 }
2209
2210 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2211 {
2212         u64 root_objectid = 0;
2213         struct btrfs_root *gang[8];
2214         int i;
2215         int ret;
2216
2217         while (1) {
2218                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2219                                              (void **)gang, root_objectid,
2220                                              ARRAY_SIZE(gang));
2221                 if (!ret)
2222                         break;
2223                 for (i = 0; i < ret; i++) {
2224                         root_objectid = gang[i]->root_key.objectid;
2225                         ret = btrfs_find_dead_roots(fs_info->tree_root,
2226                                                     root_objectid, gang[i]);
2227                         BUG_ON(ret);
2228                         btrfs_orphan_cleanup(gang[i]);
2229                 }
2230                 root_objectid++;
2231         }
2232         return 0;
2233 }
2234
2235 int btrfs_commit_super(struct btrfs_root *root)
2236 {
2237         struct btrfs_trans_handle *trans;
2238         int ret;
2239
2240         mutex_lock(&root->fs_info->cleaner_mutex);
2241         btrfs_clean_old_snapshots(root);
2242         mutex_unlock(&root->fs_info->cleaner_mutex);
2243         trans = btrfs_start_transaction(root, 1);
2244         ret = btrfs_commit_transaction(trans, root);
2245         BUG_ON(ret);
2246         /* run commit again to drop the original snapshot */
2247         trans = btrfs_start_transaction(root, 1);
2248         btrfs_commit_transaction(trans, root);
2249         ret = btrfs_write_and_wait_transaction(NULL, root);
2250         BUG_ON(ret);
2251
2252         ret = write_ctree_super(NULL, root, 0);
2253         return ret;
2254 }
2255
2256 int close_ctree(struct btrfs_root *root)
2257 {
2258         struct btrfs_fs_info *fs_info = root->fs_info;
2259         int ret;
2260
2261         fs_info->closing = 1;
2262         smp_mb();
2263
2264         kthread_stop(root->fs_info->transaction_kthread);
2265         kthread_stop(root->fs_info->cleaner_kthread);
2266
2267         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2268                 ret =  btrfs_commit_super(root);
2269                 if (ret)
2270                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2271         }
2272
2273         if (fs_info->delalloc_bytes) {
2274                 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2275                        fs_info->delalloc_bytes);
2276         }
2277         if (fs_info->total_ref_cache_size) {
2278                 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2279                        (unsigned long long)fs_info->total_ref_cache_size);
2280         }
2281
2282         if (fs_info->extent_root->node)
2283                 free_extent_buffer(fs_info->extent_root->node);
2284
2285         if (fs_info->tree_root->node)
2286                 free_extent_buffer(fs_info->tree_root->node);
2287
2288         if (root->fs_info->chunk_root->node)
2289                 free_extent_buffer(root->fs_info->chunk_root->node);
2290
2291         if (root->fs_info->dev_root->node)
2292                 free_extent_buffer(root->fs_info->dev_root->node);
2293
2294         if (root->fs_info->csum_root->node)
2295                 free_extent_buffer(root->fs_info->csum_root->node);
2296
2297         btrfs_free_block_groups(root->fs_info);
2298
2299         del_fs_roots(fs_info);
2300
2301         iput(fs_info->btree_inode);
2302
2303         btrfs_stop_workers(&fs_info->fixup_workers);
2304         btrfs_stop_workers(&fs_info->delalloc_workers);
2305         btrfs_stop_workers(&fs_info->workers);
2306         btrfs_stop_workers(&fs_info->endio_workers);
2307         btrfs_stop_workers(&fs_info->endio_meta_workers);
2308         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2309         btrfs_stop_workers(&fs_info->endio_write_workers);
2310         btrfs_stop_workers(&fs_info->submit_workers);
2311
2312 #if 0
2313         while (!list_empty(&fs_info->hashers)) {
2314                 struct btrfs_hasher *hasher;
2315                 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
2316                                     hashers);
2317                 list_del(&hasher->hashers);
2318                 crypto_free_hash(&fs_info->hash_tfm);
2319                 kfree(hasher);
2320         }
2321 #endif
2322         btrfs_close_devices(fs_info->fs_devices);
2323         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2324
2325         bdi_destroy(&fs_info->bdi);
2326
2327         kfree(fs_info->extent_root);
2328         kfree(fs_info->tree_root);
2329         kfree(fs_info->chunk_root);
2330         kfree(fs_info->dev_root);
2331         kfree(fs_info->csum_root);
2332         return 0;
2333 }
2334
2335 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2336 {
2337         int ret;
2338         struct inode *btree_inode = buf->first_page->mapping->host;
2339
2340         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
2341         if (!ret)
2342                 return ret;
2343
2344         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2345                                     parent_transid);
2346         return !ret;
2347 }
2348
2349 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2350 {
2351         struct inode *btree_inode = buf->first_page->mapping->host;
2352         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2353                                           buf);
2354 }
2355
2356 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2357 {
2358         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2359         u64 transid = btrfs_header_generation(buf);
2360         struct inode *btree_inode = root->fs_info->btree_inode;
2361
2362         btrfs_set_lock_blocking(buf);
2363
2364         btrfs_assert_tree_locked(buf);
2365         if (transid != root->fs_info->generation) {
2366                 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2367                        "found %llu running %llu\n",
2368                         (unsigned long long)buf->start,
2369                         (unsigned long long)transid,
2370                         (unsigned long long)root->fs_info->generation);
2371                 WARN_ON(1);
2372         }
2373         set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
2374 }
2375
2376 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2377 {
2378         /*
2379          * looks as though older kernels can get into trouble with
2380          * this code, they end up stuck in balance_dirty_pages forever
2381          */
2382         struct extent_io_tree *tree;
2383         u64 num_dirty;
2384         u64 start = 0;
2385         unsigned long thresh = 32 * 1024 * 1024;
2386         tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
2387
2388         if (current->flags & PF_MEMALLOC)
2389                 return;
2390
2391         num_dirty = count_range_bits(tree, &start, (u64)-1,
2392                                      thresh, EXTENT_DIRTY);
2393         if (num_dirty > thresh) {
2394                 balance_dirty_pages_ratelimited_nr(
2395                                    root->fs_info->btree_inode->i_mapping, 1);
2396         }
2397         return;
2398 }
2399
2400 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2401 {
2402         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2403         int ret;
2404         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2405         if (ret == 0)
2406                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
2407         return ret;
2408 }
2409
2410 int btree_lock_page_hook(struct page *page)
2411 {
2412         struct inode *inode = page->mapping->host;
2413         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2414         struct extent_buffer *eb;
2415         unsigned long len;
2416         u64 bytenr = page_offset(page);
2417
2418         if (page->private == EXTENT_PAGE_PRIVATE)
2419                 goto out;
2420
2421         len = page->private >> 2;
2422         eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2423         if (!eb)
2424                 goto out;
2425
2426         btrfs_tree_lock(eb);
2427         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2428         btrfs_tree_unlock(eb);
2429         free_extent_buffer(eb);
2430 out:
2431         lock_page(page);
2432         return 0;
2433 }
2434
2435 static struct extent_io_ops btree_extent_io_ops = {
2436         .write_cache_pages_lock_hook = btree_lock_page_hook,
2437         .readpage_end_io_hook = btree_readpage_end_io_hook,
2438         .submit_bio_hook = btree_submit_bio_hook,
2439         /* note we're sharing with inode.c for the merge bio hook */
2440         .merge_bio_hook = btrfs_merge_bio_hook,
2441 };