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 #include "free-space-cache.h"
  42
  43 static struct extent_io_ops btree_extent_io_ops;
  44 static void end_workqueue_fn(struct btrfs_work *work);
  45
  46 /*
  47  * end_io_wq structs are used to do processing in task context when an IO is
  48  * complete.  This is used during reads to verify checksums, and it is used
  49  * by writes to insert metadata for new file extents after IO is complete.
  50  */
  51 struct end_io_wq {
  52         struct bio *bio;
  53         bio_end_io_t *end_io;
  54         void *private;
  55         struct btrfs_fs_info *info;
  56         int error;
  57         int metadata;
  58         struct list_head list;
  59         struct btrfs_work work;
  60 };
  61
  62 /*
  63  * async submit bios are used to offload expensive checksumming
  64  * onto the worker threads.  They checksum file and metadata bios
  65  * just before they are sent down the IO stack.
  66  */
  67 struct async_submit_bio {
  68         struct inode *inode;
  69         struct bio *bio;
  70         struct list_head list;
  71         extent_submit_bio_hook_t *submit_bio_start;
  72         extent_submit_bio_hook_t *submit_bio_done;
  73         int rw;
  74         int mirror_num;
  75         unsigned long bio_flags;
  76         struct btrfs_work work;
  77 };
  78
  79 /* These are used to set the lockdep class on the extent buffer locks.
  80  * The class is set by the readpage_end_io_hook after the buffer has
  81  * passed csum validation but before the pages are unlocked.
  82  *
  83  * The lockdep class is also set by btrfs_init_new_buffer on freshly
  84  * allocated blocks.
  85  *
  86  * The class is based on the level in the tree block, which allows lockdep
  87  * to know that lower nodes nest inside the locks of higher nodes.
  88  *
  89  * We also add a check to make sure the highest level of the tree is
  90  * the same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this
  91  * code needs update as well.
  92  */
  93 #ifdef CONFIG_DEBUG_LOCK_ALLOC
  94 # if BTRFS_MAX_LEVEL != 8
  95 #  error
  96 # endif
  97 static struct lock_class_key btrfs_eb_class[BTRFS_MAX_LEVEL + 1];
  98 static const char *btrfs_eb_name[BTRFS_MAX_LEVEL + 1] = {
  99         /* leaf */
 100         "btrfs-extent-00",
 101         "btrfs-extent-01",
 102         "btrfs-extent-02",
 103         "btrfs-extent-03",
 104         "btrfs-extent-04",
 105         "btrfs-extent-05",
 106         "btrfs-extent-06",
 107         "btrfs-extent-07",
 108         /* highest possible level */
 109         "btrfs-extent-08",
 110 };
 111 #endif
 112
 113 /*
 114  * extents on the btree inode are pretty simple, there's one extent
 115  * that covers the entire device
 116  */
 117 static struct extent_map *btree_get_extent(struct inode *inode,
 118                 struct page *page, size_t page_offset, u64 start, u64 len,
 119                 int create)
 120 {
 121         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 122         struct extent_map *em;
 123         int ret;
 124
 125         spin_lock(&em_tree->lock);
 126         em = lookup_extent_mapping(em_tree, start, len);
 127         if (em) {
 128                 em->bdev =
 129                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 130                 spin_unlock(&em_tree->lock);
 131                 goto out;
 132         }
 133         spin_unlock(&em_tree->lock);
 134
 135         em = alloc_extent_map(GFP_NOFS);
 136         if (!em) {
 137                 em = ERR_PTR(-ENOMEM);
 138                 goto out;
 139         }
 140         em->start = 0;
 141         em->len = (u64)-1;
 142         em->block_len = (u64)-1;
 143         em->block_start = 0;
 144         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 145
 146         spin_lock(&em_tree->lock);
 147         ret = add_extent_mapping(em_tree, em);
 148         if (ret == -EEXIST) {
 149                 u64 failed_start = em->start;
 150                 u64 failed_len = em->len;
 151
 152                 free_extent_map(em);
 153                 em = lookup_extent_mapping(em_tree, start, len);
 154                 if (em) {
 155                         ret = 0;
 156                 } else {
 157                         em = lookup_extent_mapping(em_tree, failed_start,
 158                                                    failed_len);
 159                         ret = -EIO;
 160                 }
 161         } else if (ret) {
 162                 free_extent_map(em);
 163                 em = NULL;
 164         }
 165         spin_unlock(&em_tree->lock);
 166
 167         if (ret)
 168                 em = ERR_PTR(ret);
 169 out:
 170         return em;
 171 }
 172
 173 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
 174 {
 175         return btrfs_crc32c(seed, data, len);
 176 }
 177
 178 void btrfs_csum_final(u32 crc, char *result)
 179 {
 180         *(__le32 *)result = ~cpu_to_le32(crc);
 181 }
 182
 183 /*
 184  * compute the csum for a btree block, and either verify it or write it
 185  * into the csum field of the block.
 186  */
 187 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 188                            int verify)
 189 {
 190         u16 csum_size =
 191                 btrfs_super_csum_size(&root->fs_info->super_copy);
 192         char *result = NULL;
 193         unsigned long len;
 194         unsigned long cur_len;
 195         unsigned long offset = BTRFS_CSUM_SIZE;
 196         char *map_token = NULL;
 197         char *kaddr;
 198         unsigned long map_start;
 199         unsigned long map_len;
 200         int err;
 201         u32 crc = ~(u32)0;
 202         unsigned long inline_result;
 203
 204         len = buf->len - offset;
 205         while (len > 0) {
 206                 err = map_private_extent_buffer(buf, offset, 32,
 207                                         &map_token, &kaddr,
 208                                         &map_start, &map_len, KM_USER0);
 209                 if (err)
 210                         return 1;
 211                 cur_len = min(len, map_len - (offset - map_start));
 212                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
 213                                       crc, cur_len);
 214                 len -= cur_len;
 215                 offset += cur_len;
 216                 unmap_extent_buffer(buf, map_token, KM_USER0);
 217         }
 218         if (csum_size > sizeof(inline_result)) {
 219                 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
 220                 if (!result)
 221                         return 1;
 222         } else {
 223                 result = (char *)&inline_result;
 224         }
 225
 226         btrfs_csum_final(crc, result);
 227
 228         if (verify) {
 229                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
 230                         u32 val;
 231                         u32 found = 0;
 232                         memcpy(&found, result, csum_size);
 233
 234                         read_extent_buffer(buf, &val, 0, csum_size);
 235                         printk(KERN_INFO "btrfs: %s checksum verify failed "
 236                                "on %llu wanted %X found %X level %d\n",
 237                                root->fs_info->sb->s_id,
 238                                buf->start, val, found, btrfs_header_level(buf));
 239                         if (result != (char *)&inline_result)
 240                                 kfree(result);
 241                         return 1;
 242                 }
 243         } else {
 244                 write_extent_buffer(buf, result, 0, csum_size);
 245         }
 246         if (result != (char *)&inline_result)
 247                 kfree(result);
 248         return 0;
 249 }
 250
 251 /*
 252  * we can't consider a given block up to date unless the transid of the
 253  * block matches the transid in the parent node's pointer.  This is how we
 254  * detect blocks that either didn't get written at all or got written
 255  * in the wrong place.
 256  */
 257 static int verify_parent_transid(struct extent_io_tree *io_tree,
 258                                  struct extent_buffer *eb, u64 parent_transid)
 259 {
 260         int ret;
 261
 262         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 263                 return 0;
 264
 265         lock_extent(io_tree, eb->start, eb->start + eb->len - 1, GFP_NOFS);
 266         if (extent_buffer_uptodate(io_tree, eb) &&
 267             btrfs_header_generation(eb) == parent_transid) {
 268                 ret = 0;
 269                 goto out;
 270         }
 271         printk("parent transid verify failed on %llu wanted %llu found %llu\n",
 272                (unsigned long long)eb->start,
 273                (unsigned long long)parent_transid,
 274                (unsigned long long)btrfs_header_generation(eb));
 275         ret = 1;
 276         clear_extent_buffer_uptodate(io_tree, eb);
 277 out:
 278         unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
 279                       GFP_NOFS);
 280         return ret;
 281 }
 282
 283 /*
 284  * helper to read a given tree block, doing retries as required when
 285  * the checksums don't match and we have alternate mirrors to try.
 286  */
 287 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 288                                           struct extent_buffer *eb,
 289                                           u64 start, u64 parent_transid)
 290 {
 291         struct extent_io_tree *io_tree;
 292         int ret;
 293         int num_copies = 0;
 294         int mirror_num = 0;
 295
 296         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 297         while (1) {
 298                 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
 299                                                btree_get_extent, mirror_num);
 300                 if (!ret &&
 301                     !verify_parent_transid(io_tree, eb, parent_transid))
 302                         return ret;
 303
 304                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
 305                                               eb->start, eb->len);
 306                 if (num_copies == 1)
 307                         return ret;
 308
 309                 mirror_num++;
 310                 if (mirror_num > num_copies)
 311                         return ret;
 312         }
 313         return -EIO;
 314 }
 315
 316 /*
 317  * checksum a dirty tree block before IO.  This has extra checks to make sure
 318  * we only fill in the checksum field in the first page of a multi-page block
 319  */
 320
 321 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 322 {
 323         struct extent_io_tree *tree;
 324         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
 325         u64 found_start;
 326         int found_level;
 327         unsigned long len;
 328         struct extent_buffer *eb;
 329         int ret;
 330
 331         tree = &BTRFS_I(page->mapping->host)->io_tree;
 332
 333         if (page->private == EXTENT_PAGE_PRIVATE)
 334                 goto out;
 335         if (!page->private)
 336                 goto out;
 337         len = page->private >> 2;
 338         WARN_ON(len == 0);
 339
 340         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 341         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
 342                                              btrfs_header_generation(eb));
 343         BUG_ON(ret);
 344         found_start = btrfs_header_bytenr(eb);
 345         if (found_start != start) {
 346                 WARN_ON(1);
 347                 goto err;
 348         }
 349         if (eb->first_page != page) {
 350                 WARN_ON(1);
 351                 goto err;
 352         }
 353         if (!PageUptodate(page)) {
 354                 WARN_ON(1);
 355                 goto err;
 356         }
 357         found_level = btrfs_header_level(eb);
 358
 359         csum_tree_block(root, eb, 0);
 360 err:
 361         free_extent_buffer(eb);
 362 out:
 363         return 0;
 364 }
 365
 366 static int check_tree_block_fsid(struct btrfs_root *root,
 367                                  struct extent_buffer *eb)
 368 {
 369         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
 370         u8 fsid[BTRFS_UUID_SIZE];
 371         int ret = 1;
 372
 373         read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
 374                            BTRFS_FSID_SIZE);
 375         while (fs_devices) {
 376                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
 377                         ret = 0;
 378                         break;
 379                 }
 380                 fs_devices = fs_devices->seed;
 381         }
 382         return ret;
 383 }
 384
 385 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 386 void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level)
 387 {
 388         lockdep_set_class_and_name(&eb->lock,
 389                            &btrfs_eb_class[level],
 390                            btrfs_eb_name[level]);
 391 }
 392 #endif
 393
 394 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 395                                struct extent_state *state)
 396 {
 397         struct extent_io_tree *tree;
 398         u64 found_start;
 399         int found_level;
 400         unsigned long len;
 401         struct extent_buffer *eb;
 402         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 403         int ret = 0;
 404
 405         tree = &BTRFS_I(page->mapping->host)->io_tree;
 406         if (page->private == EXTENT_PAGE_PRIVATE)
 407                 goto out;
 408         if (!page->private)
 409                 goto out;
 410
 411         len = page->private >> 2;
 412         WARN_ON(len == 0);
 413
 414         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 415
 416         found_start = btrfs_header_bytenr(eb);
 417         if (found_start != start) {
 418                 printk(KERN_INFO "btrfs bad tree block start %llu %llu\n",
 419                        (unsigned long long)found_start,
 420                        (unsigned long long)eb->start);
 421                 ret = -EIO;
 422                 goto err;
 423         }
 424         if (eb->first_page != page) {
 425                 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
 426                        eb->first_page->index, page->index);
 427                 WARN_ON(1);
 428                 ret = -EIO;
 429                 goto err;
 430         }
 431         if (check_tree_block_fsid(root, eb)) {
 432                 printk(KERN_INFO "btrfs bad fsid on block %llu\n",
 433                        (unsigned long long)eb->start);
 434                 ret = -EIO;
 435                 goto err;
 436         }
 437         found_level = btrfs_header_level(eb);
 438
 439         btrfs_set_buffer_lockdep_class(eb, found_level);
 440
 441         ret = csum_tree_block(root, eb, 1);
 442         if (ret)
 443                 ret = -EIO;
 444
 445         end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
 446         end = eb->start + end - 1;
 447 err:
 448         free_extent_buffer(eb);
 449 out:
 450         return ret;
 451 }
 452
 453 static void end_workqueue_bio(struct bio *bio, int err)
 454 {
 455         struct end_io_wq *end_io_wq = bio->bi_private;
 456         struct btrfs_fs_info *fs_info;
 457
 458         fs_info = end_io_wq->info;
 459         end_io_wq->error = err;
 460         end_io_wq->work.func = end_workqueue_fn;
 461         end_io_wq->work.flags = 0;
 462
 463         if (bio->bi_rw & (1 << BIO_RW)) {
 464                 if (end_io_wq->metadata)
 465                         btrfs_queue_worker(&fs_info->endio_meta_write_workers,
 466                                            &end_io_wq->work);
 467                 else
 468                         btrfs_queue_worker(&fs_info->endio_write_workers,
 469                                            &end_io_wq->work);
 470         } else {
 471                 if (end_io_wq->metadata)
 472                         btrfs_queue_worker(&fs_info->endio_meta_workers,
 473                                            &end_io_wq->work);
 474                 else
 475                         btrfs_queue_worker(&fs_info->endio_workers,
 476                                            &end_io_wq->work);
 477         }
 478 }
 479
 480 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 481                         int metadata)
 482 {
 483         struct end_io_wq *end_io_wq;
 484         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
 485         if (!end_io_wq)
 486                 return -ENOMEM;
 487
 488         end_io_wq->private = bio->bi_private;
 489         end_io_wq->end_io = bio->bi_end_io;
 490         end_io_wq->info = info;
 491         end_io_wq->error = 0;
 492         end_io_wq->bio = bio;
 493         end_io_wq->metadata = metadata;
 494
 495         bio->bi_private = end_io_wq;
 496         bio->bi_end_io = end_workqueue_bio;
 497         return 0;
 498 }
 499
 500 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
 501 {
 502         unsigned long limit = min_t(unsigned long,
 503                                     info->workers.max_workers,
 504                                     info->fs_devices->open_devices);
 505         return 256 * limit;
 506 }
 507
 508 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
 509 {
 510         return atomic_read(&info->nr_async_bios) >
 511                 btrfs_async_submit_limit(info);
 512 }
 513
 514 static void run_one_async_start(struct btrfs_work *work)
 515 {
 516         struct btrfs_fs_info *fs_info;
 517         struct async_submit_bio *async;
 518
 519         async = container_of(work, struct  async_submit_bio, work);
 520         fs_info = BTRFS_I(async->inode)->root->fs_info;
 521         async->submit_bio_start(async->inode, async->rw, async->bio,
 522                                async->mirror_num, async->bio_flags);
 523 }
 524
 525 static void run_one_async_done(struct btrfs_work *work)
 526 {
 527         struct btrfs_fs_info *fs_info;
 528         struct async_submit_bio *async;
 529         int limit;
 530
 531         async = container_of(work, struct  async_submit_bio, work);
 532         fs_info = BTRFS_I(async->inode)->root->fs_info;
 533
 534         limit = btrfs_async_submit_limit(fs_info);
 535         limit = limit * 2 / 3;
 536
 537         atomic_dec(&fs_info->nr_async_submits);
 538
 539         if (atomic_read(&fs_info->nr_async_submits) < limit &&
 540             waitqueue_active(&fs_info->async_submit_wait))
 541                 wake_up(&fs_info->async_submit_wait);
 542
 543         async->submit_bio_done(async->inode, async->rw, async->bio,
 544                                async->mirror_num, async->bio_flags);
 545 }
 546
 547 static void run_one_async_free(struct btrfs_work *work)
 548 {
 549         struct async_submit_bio *async;
 550
 551         async = container_of(work, struct  async_submit_bio, work);
 552         kfree(async);
 553 }
 554
 555 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 556                         int rw, struct bio *bio, int mirror_num,
 557                         unsigned long bio_flags,
 558                         extent_submit_bio_hook_t *submit_bio_start,
 559                         extent_submit_bio_hook_t *submit_bio_done)
 560 {
 561         struct async_submit_bio *async;
 562
 563         async = kmalloc(sizeof(*async), GFP_NOFS);
 564         if (!async)
 565                 return -ENOMEM;
 566
 567         async->inode = inode;
 568         async->rw = rw;
 569         async->bio = bio;
 570         async->mirror_num = mirror_num;
 571         async->submit_bio_start = submit_bio_start;
 572         async->submit_bio_done = submit_bio_done;
 573
 574         async->work.func = run_one_async_start;
 575         async->work.ordered_func = run_one_async_done;
 576         async->work.ordered_free = run_one_async_free;
 577
 578         async->work.flags = 0;
 579         async->bio_flags = bio_flags;
 580
 581         atomic_inc(&fs_info->nr_async_submits);
 582
 583         if (rw & (1 << BIO_RW_SYNCIO))
 584                 btrfs_set_work_high_prio(&async->work);
 585
 586         btrfs_queue_worker(&fs_info->workers, &async->work);
 587 #if 0
 588         int limit = btrfs_async_submit_limit(fs_info);
 589         if (atomic_read(&fs_info->nr_async_submits) > limit) {
 590                 wait_event_timeout(fs_info->async_submit_wait,
 591                            (atomic_read(&fs_info->nr_async_submits) < limit),
 592                            HZ/10);
 593
 594                 wait_event_timeout(fs_info->async_submit_wait,
 595                            (atomic_read(&fs_info->nr_async_bios) < limit),
 596                            HZ/10);
 597         }
 598 #endif
 599         while (atomic_read(&fs_info->async_submit_draining) &&
 600               atomic_read(&fs_info->nr_async_submits)) {
 601                 wait_event(fs_info->async_submit_wait,
 602                            (atomic_read(&fs_info->nr_async_submits) == 0));
 603         }
 604
 605         return 0;
 606 }
 607
 608 static int btree_csum_one_bio(struct bio *bio)
 609 {
 610         struct bio_vec *bvec = bio->bi_io_vec;
 611         int bio_index = 0;
 612         struct btrfs_root *root;
 613
 614         WARN_ON(bio->bi_vcnt <= 0);
 615         while (bio_index < bio->bi_vcnt) {
 616                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 617                 csum_dirty_buffer(root, bvec->bv_page);
 618                 bio_index++;
 619                 bvec++;
 620         }
 621         return 0;
 622 }
 623
 624 static int __btree_submit_bio_start(struct inode *inode, int rw,
 625                                     struct bio *bio, int mirror_num,
 626                                     unsigned long bio_flags)
 627 {
 628         /*
 629          * when we're called for a write, we're already in the async
 630          * submission context.  Just jump into btrfs_map_bio
 631          */
 632         btree_csum_one_bio(bio);
 633         return 0;
 634 }
 635
 636 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
 637                                  int mirror_num, unsigned long bio_flags)
 638 {
 639         /*
 640          * when we're called for a write, we're already in the async
 641          * submission context.  Just jump into btrfs_map_bio
 642          */
 643         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
 644 }
 645
 646 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 647                                  int mirror_num, unsigned long bio_flags)
 648 {
 649         int ret;
 650
 651         ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
 652                                           bio, 1);
 653         BUG_ON(ret);
 654
 655         if (!(rw & (1 << BIO_RW))) {
 656                 /*
 657                  * called for a read, do the setup so that checksum validation
 658                  * can happen in the async kernel threads
 659                  */
 660                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 661                                      mirror_num, 0);
 662         }
 663
 664         /*
 665          * kthread helpers are used to submit writes so that checksumming
 666          * can happen in parallel across all CPUs
 667          */
 668         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 669                                    inode, rw, bio, mirror_num, 0,
 670                                    __btree_submit_bio_start,
 671                                    __btree_submit_bio_done);
 672 }
 673
 674 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 675 {
 676         struct extent_io_tree *tree;
 677         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 678         struct extent_buffer *eb;
 679         int was_dirty;
 680
 681         tree = &BTRFS_I(page->mapping->host)->io_tree;
 682         if (!(current->flags & PF_MEMALLOC)) {
 683                 return extent_write_full_page(tree, page,
 684                                               btree_get_extent, wbc);
 685         }
 686
 687         redirty_page_for_writepage(wbc, page);
 688         eb = btrfs_find_tree_block(root, page_offset(page),
 689                                       PAGE_CACHE_SIZE);
 690         WARN_ON(!eb);
 691
 692         was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
 693         if (!was_dirty) {
 694                 spin_lock(&root->fs_info->delalloc_lock);
 695                 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
 696                 spin_unlock(&root->fs_info->delalloc_lock);
 697         }
 698         free_extent_buffer(eb);
 699
 700         unlock_page(page);
 701         return 0;
 702 }
 703
 704 static int btree_writepages(struct address_space *mapping,
 705                             struct writeback_control *wbc)
 706 {
 707         struct extent_io_tree *tree;
 708         tree = &BTRFS_I(mapping->host)->io_tree;
 709         if (wbc->sync_mode == WB_SYNC_NONE) {
 710                 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
 711                 u64 num_dirty;
 712                 unsigned long thresh = 32 * 1024 * 1024;
 713
 714                 if (wbc->for_kupdate)
 715                         return 0;
 716
 717                 /* this is a bit racy, but that's ok */
 718                 num_dirty = root->fs_info->dirty_metadata_bytes;
 719                 if (num_dirty < thresh)
 720                         return 0;
 721         }
 722         return extent_writepages(tree, mapping, btree_get_extent, wbc);
 723 }
 724
 725 static int btree_readpage(struct file *file, struct page *page)
 726 {
 727         struct extent_io_tree *tree;
 728         tree = &BTRFS_I(page->mapping->host)->io_tree;
 729         return extent_read_full_page(tree, page, btree_get_extent);
 730 }
 731
 732 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 733 {
 734         struct extent_io_tree *tree;
 735         struct extent_map_tree *map;
 736         int ret;
 737
 738         if (PageWriteback(page) || PageDirty(page))
 739                 return 0;
 740
 741         tree = &BTRFS_I(page->mapping->host)->io_tree;
 742         map = &BTRFS_I(page->mapping->host)->extent_tree;
 743
 744         ret = try_release_extent_state(map, tree, page, gfp_flags);
 745         if (!ret)
 746                 return 0;
 747
 748         ret = try_release_extent_buffer(tree, page);
 749         if (ret == 1) {
 750                 ClearPagePrivate(page);
 751                 set_page_private(page, 0);
 752                 page_cache_release(page);
 753         }
 754
 755         return ret;
 756 }
 757
 758 static void btree_invalidatepage(struct page *page, unsigned long offset)
 759 {
 760         struct extent_io_tree *tree;
 761         tree = &BTRFS_I(page->mapping->host)->io_tree;
 762         extent_invalidatepage(tree, page, offset);
 763         btree_releasepage(page, GFP_NOFS);
 764         if (PagePrivate(page)) {
 765                 printk(KERN_WARNING "btrfs warning page private not zero "
 766                        "on page %llu\n", (unsigned long long)page_offset(page));
 767                 ClearPagePrivate(page);
 768                 set_page_private(page, 0);
 769                 page_cache_release(page);
 770         }
 771 }
 772
 773 #if 0
 774 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 775 {
 776         struct buffer_head *bh;
 777         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 778         struct buffer_head *head;
 779         if (!page_has_buffers(page)) {
 780                 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
 781                                         (1 << BH_Dirty)|(1 << BH_Uptodate));
 782         }
 783         head = page_buffers(page);
 784         bh = head;
 785         do {
 786                 if (buffer_dirty(bh))
 787                         csum_tree_block(root, bh, 0);
 788                 bh = bh->b_this_page;
 789         } while (bh != head);
 790         return block_write_full_page(page, btree_get_block, wbc);
 791 }
 792 #endif
 793
 794 static struct address_space_operations btree_aops = {
 795         .readpage       = btree_readpage,
 796         .writepage      = btree_writepage,
 797         .writepages     = btree_writepages,
 798         .releasepage    = btree_releasepage,
 799         .invalidatepage = btree_invalidatepage,
 800         .sync_page      = block_sync_page,
 801 };
 802
 803 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
 804                          u64 parent_transid)
 805 {
 806         struct extent_buffer *buf = NULL;
 807         struct inode *btree_inode = root->fs_info->btree_inode;
 808         int ret = 0;
 809
 810         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 811         if (!buf)
 812                 return 0;
 813         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
 814                                  buf, 0, 0, btree_get_extent, 0);
 815         free_extent_buffer(buf);
 816         return ret;
 817 }
 818
 819 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
 820                                             u64 bytenr, u32 blocksize)
 821 {
 822         struct inode *btree_inode = root->fs_info->btree_inode;
 823         struct extent_buffer *eb;
 824         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 825                                 bytenr, blocksize, GFP_NOFS);
 826         return eb;
 827 }
 828
 829 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
 830                                                  u64 bytenr, u32 blocksize)
 831 {
 832         struct inode *btree_inode = root->fs_info->btree_inode;
 833         struct extent_buffer *eb;
 834
 835         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 836                                  bytenr, blocksize, NULL, GFP_NOFS);
 837         return eb;
 838 }
 839
 840
 841 int btrfs_write_tree_block(struct extent_buffer *buf)
 842 {
 843         return btrfs_fdatawrite_range(buf->first_page->mapping, buf->start,
 844                                       buf->start + buf->len - 1, WB_SYNC_ALL);
 845 }
 846
 847 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
 848 {
 849         return btrfs_wait_on_page_writeback_range(buf->first_page->mapping,
 850                                   buf->start, buf->start + buf->len - 1);
 851 }
 852
 853 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
 854                                       u32 blocksize, u64 parent_transid)
 855 {
 856         struct extent_buffer *buf = NULL;
 857         struct inode *btree_inode = root->fs_info->btree_inode;
 858         struct extent_io_tree *io_tree;
 859         int ret;
 860
 861         io_tree = &BTRFS_I(btree_inode)->io_tree;
 862
 863         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 864         if (!buf)
 865                 return NULL;
 866
 867         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
 868
 869         if (ret == 0)
 870                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
 871         else
 872                 WARN_ON(1);
 873         return buf;
 874
 875 }
 876
 877 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 878                      struct extent_buffer *buf)
 879 {
 880         struct inode *btree_inode = root->fs_info->btree_inode;
 881         if (btrfs_header_generation(buf) ==
 882             root->fs_info->running_transaction->transid) {
 883                 btrfs_assert_tree_locked(buf);
 884
 885                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
 886                         spin_lock(&root->fs_info->delalloc_lock);
 887                         if (root->fs_info->dirty_metadata_bytes >= buf->len)
 888                                 root->fs_info->dirty_metadata_bytes -= buf->len;
 889                         else
 890                                 WARN_ON(1);
 891                         spin_unlock(&root->fs_info->delalloc_lock);
 892                 }
 893
 894                 /* ugh, clear_extent_buffer_dirty needs to lock the page */
 895                 btrfs_set_lock_blocking(buf);
 896                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
 897                                           buf);
 898         }
 899         return 0;
 900 }
 901
 902 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
 903                         u32 stripesize, struct btrfs_root *root,
 904                         struct btrfs_fs_info *fs_info,
 905                         u64 objectid)
 906 {
 907         root->node = NULL;
 908         root->commit_root = NULL;
 909         root->ref_tree = NULL;
 910         root->sectorsize = sectorsize;
 911         root->nodesize = nodesize;
 912         root->leafsize = leafsize;
 913         root->stripesize = stripesize;
 914         root->ref_cows = 0;
 915         root->track_dirty = 0;
 916
 917         root->fs_info = fs_info;
 918         root->objectid = objectid;
 919         root->last_trans = 0;
 920         root->highest_inode = 0;
 921         root->last_inode_alloc = 0;
 922         root->name = NULL;
 923         root->in_sysfs = 0;
 924
 925         INIT_LIST_HEAD(&root->dirty_list);
 926         INIT_LIST_HEAD(&root->orphan_list);
 927         INIT_LIST_HEAD(&root->dead_list);
 928         spin_lock_init(&root->node_lock);
 929         spin_lock_init(&root->list_lock);
 930         mutex_init(&root->objectid_mutex);
 931         mutex_init(&root->log_mutex);
 932         init_waitqueue_head(&root->log_writer_wait);
 933         init_waitqueue_head(&root->log_commit_wait[0]);
 934         init_waitqueue_head(&root->log_commit_wait[1]);
 935         atomic_set(&root->log_commit[0], 0);
 936         atomic_set(&root->log_commit[1], 0);
 937         atomic_set(&root->log_writers, 0);
 938         root->log_batch = 0;
 939         root->log_transid = 0;
 940         extent_io_tree_init(&root->dirty_log_pages,
 941                              fs_info->btree_inode->i_mapping, GFP_NOFS);
 942
 943         btrfs_leaf_ref_tree_init(&root->ref_tree_struct);
 944         root->ref_tree = &root->ref_tree_struct;
 945
 946         memset(&root->root_key, 0, sizeof(root->root_key));
 947         memset(&root->root_item, 0, sizeof(root->root_item));
 948         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
 949         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
 950         root->defrag_trans_start = fs_info->generation;
 951         init_completion(&root->kobj_unregister);
 952         root->defrag_running = 0;
 953         root->defrag_level = 0;
 954         root->root_key.objectid = objectid;
 955         root->anon_super.s_root = NULL;
 956         root->anon_super.s_dev = 0;
 957         INIT_LIST_HEAD(&root->anon_super.s_list);
 958         INIT_LIST_HEAD(&root->anon_super.s_instances);
 959         init_rwsem(&root->anon_super.s_umount);
 960
 961         return 0;
 962 }
 963
 964 static int find_and_setup_root(struct btrfs_root *tree_root,
 965                                struct btrfs_fs_info *fs_info,
 966                                u64 objectid,
 967                                struct btrfs_root *root)
 968 {
 969         int ret;
 970         u32 blocksize;
 971         u64 generation;
 972
 973         __setup_root(tree_root->nodesize, tree_root->leafsize,
 974                      tree_root->sectorsize, tree_root->stripesize,
 975                      root, fs_info, objectid);
 976         ret = btrfs_find_last_root(tree_root, objectid,
 977                                    &root->root_item, &root->root_key);
 978         BUG_ON(ret);
 979
 980         generation = btrfs_root_generation(&root->root_item);
 981         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 982         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 983                                      blocksize, generation);
 984         BUG_ON(!root->node);
 985         return 0;
 986 }
 987
 988 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
 989                              struct btrfs_fs_info *fs_info)
 990 {
 991         struct extent_buffer *eb;
 992         struct btrfs_root *log_root_tree = fs_info->log_root_tree;
 993         u64 start = 0;
 994         u64 end = 0;
 995         int ret;
 996
 997         if (!log_root_tree)
 998                 return 0;
 999
1000         while (1) {
1001                 ret = find_first_extent_bit(&log_root_tree->dirty_log_pages,
1002                                     0, &start, &end, EXTENT_DIRTY);
1003                 if (ret)
1004                         break;
1005
1006                 clear_extent_dirty(&log_root_tree->dirty_log_pages,
1007                                    start, end, GFP_NOFS);
1008         }
1009         eb = fs_info->log_root_tree->node;
1010
1011         WARN_ON(btrfs_header_level(eb) != 0);
1012         WARN_ON(btrfs_header_nritems(eb) != 0);
1013
1014         ret = btrfs_free_reserved_extent(fs_info->tree_root,
1015                                 eb->start, eb->len);
1016         BUG_ON(ret);
1017
1018         free_extent_buffer(eb);
1019         kfree(fs_info->log_root_tree);
1020         fs_info->log_root_tree = NULL;
1021         return 0;
1022 }
1023
1024 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1025                                          struct btrfs_fs_info *fs_info)
1026 {
1027         struct btrfs_root *root;
1028         struct btrfs_root *tree_root = fs_info->tree_root;
1029         struct extent_buffer *leaf;
1030
1031         root = kzalloc(sizeof(*root), GFP_NOFS);
1032         if (!root)
1033                 return ERR_PTR(-ENOMEM);
1034
1035         __setup_root(tree_root->nodesize, tree_root->leafsize,
1036                      tree_root->sectorsize, tree_root->stripesize,
1037                      root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1038
1039         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1040         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1041         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1042         /*
1043          * log trees do not get reference counted because they go away
1044          * before a real commit is actually done.  They do store pointers
1045          * to file data extents, and those reference counts still get
1046          * updated (along with back refs to the log tree).
1047          */
1048         root->ref_cows = 0;
1049
1050         leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1051                                       0, BTRFS_TREE_LOG_OBJECTID,
1052                                       trans->transid, 0, 0, 0);
1053         if (IS_ERR(leaf)) {
1054                 kfree(root);
1055                 return ERR_CAST(leaf);
1056         }
1057
1058         root->node = leaf;
1059         btrfs_set_header_nritems(root->node, 0);
1060         btrfs_set_header_level(root->node, 0);
1061         btrfs_set_header_bytenr(root->node, root->node->start);
1062         btrfs_set_header_generation(root->node, trans->transid);
1063         btrfs_set_header_owner(root->node, BTRFS_TREE_LOG_OBJECTID);
1064
1065         write_extent_buffer(root->node, root->fs_info->fsid,
1066                             (unsigned long)btrfs_header_fsid(root->node),
1067                             BTRFS_FSID_SIZE);
1068         btrfs_mark_buffer_dirty(root->node);
1069         btrfs_tree_unlock(root->node);
1070         return root;
1071 }
1072
1073 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1074                              struct btrfs_fs_info *fs_info)
1075 {
1076         struct btrfs_root *log_root;
1077
1078         log_root = alloc_log_tree(trans, fs_info);
1079         if (IS_ERR(log_root))
1080                 return PTR_ERR(log_root);
1081         WARN_ON(fs_info->log_root_tree);
1082         fs_info->log_root_tree = log_root;
1083         return 0;
1084 }
1085
1086 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1087                        struct btrfs_root *root)
1088 {
1089         struct btrfs_root *log_root;
1090         struct btrfs_inode_item *inode_item;
1091
1092         log_root = alloc_log_tree(trans, root->fs_info);
1093         if (IS_ERR(log_root))
1094                 return PTR_ERR(log_root);
1095
1096         log_root->last_trans = trans->transid;
1097         log_root->root_key.offset = root->root_key.objectid;
1098
1099         inode_item = &log_root->root_item.inode;
1100         inode_item->generation = cpu_to_le64(1);
1101         inode_item->size = cpu_to_le64(3);
1102         inode_item->nlink = cpu_to_le32(1);
1103         inode_item->nbytes = cpu_to_le64(root->leafsize);
1104         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1105
1106         btrfs_set_root_bytenr(&log_root->root_item, log_root->node->start);
1107         btrfs_set_root_generation(&log_root->root_item, trans->transid);
1108
1109         WARN_ON(root->log_root);
1110         root->log_root = log_root;
1111         root->log_transid = 0;
1112         return 0;
1113 }
1114
1115 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1116                                                struct btrfs_key *location)
1117 {
1118         struct btrfs_root *root;
1119         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1120         struct btrfs_path *path;
1121         struct extent_buffer *l;
1122         u64 highest_inode;
1123         u64 generation;
1124         u32 blocksize;
1125         int ret = 0;
1126
1127         root = kzalloc(sizeof(*root), GFP_NOFS);
1128         if (!root)
1129                 return ERR_PTR(-ENOMEM);
1130         if (location->offset == (u64)-1) {
1131                 ret = find_and_setup_root(tree_root, fs_info,
1132                                           location->objectid, root);
1133                 if (ret) {
1134                         kfree(root);
1135                         return ERR_PTR(ret);
1136                 }
1137                 goto insert;
1138         }
1139
1140         __setup_root(tree_root->nodesize, tree_root->leafsize,
1141                      tree_root->sectorsize, tree_root->stripesize,
1142                      root, fs_info, location->objectid);
1143
1144         path = btrfs_alloc_path();
1145         BUG_ON(!path);
1146         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1147         if (ret != 0) {
1148                 if (ret > 0)
1149                         ret = -ENOENT;
1150                 goto out;
1151         }
1152         l = path->nodes[0];
1153         read_extent_buffer(l, &root->root_item,
1154                btrfs_item_ptr_offset(l, path->slots[0]),
1155                sizeof(root->root_item));
1156         memcpy(&root->root_key, location, sizeof(*location));
1157         ret = 0;
1158 out:
1159         btrfs_release_path(root, path);
1160         btrfs_free_path(path);
1161         if (ret) {
1162                 kfree(root);
1163                 return ERR_PTR(ret);
1164         }
1165         generation = btrfs_root_generation(&root->root_item);
1166         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1167         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1168                                      blocksize, generation);
1169         BUG_ON(!root->node);
1170 insert:
1171         if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1172                 root->ref_cows = 1;
1173                 ret = btrfs_find_highest_inode(root, &highest_inode);
1174                 if (ret == 0) {
1175                         root->highest_inode = highest_inode;
1176                         root->last_inode_alloc = highest_inode;
1177                 }
1178         }
1179         return root;
1180 }
1181
1182 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1183                                         u64 root_objectid)
1184 {
1185         struct btrfs_root *root;
1186
1187         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1188                 return fs_info->tree_root;
1189         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1190                 return fs_info->extent_root;
1191
1192         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1193                                  (unsigned long)root_objectid);
1194         return root;
1195 }
1196
1197 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1198                                               struct btrfs_key *location)
1199 {
1200         struct btrfs_root *root;
1201         int ret;
1202
1203         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1204                 return fs_info->tree_root;
1205         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1206                 return fs_info->extent_root;
1207         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1208                 return fs_info->chunk_root;
1209         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1210                 return fs_info->dev_root;
1211         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1212                 return fs_info->csum_root;
1213
1214         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1215                                  (unsigned long)location->objectid);
1216         if (root)
1217                 return root;
1218
1219         root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1220         if (IS_ERR(root))
1221                 return root;
1222
1223         set_anon_super(&root->anon_super, NULL);
1224
1225         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1226                                 (unsigned long)root->root_key.objectid,
1227                                 root);
1228         if (ret) {
1229                 free_extent_buffer(root->node);
1230                 kfree(root);
1231                 return ERR_PTR(ret);
1232         }
1233         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
1234                 ret = btrfs_find_dead_roots(fs_info->tree_root,
1235                                             root->root_key.objectid, root);
1236                 BUG_ON(ret);
1237                 btrfs_orphan_cleanup(root);
1238         }
1239         return root;
1240 }
1241
1242 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1243                                       struct btrfs_key *location,
1244                                       const char *name, int namelen)
1245 {
1246         struct btrfs_root *root;
1247         int ret;
1248
1249         root = btrfs_read_fs_root_no_name(fs_info, location);
1250         if (!root)
1251                 return NULL;
1252
1253         if (root->in_sysfs)
1254                 return root;
1255
1256         ret = btrfs_set_root_name(root, name, namelen);
1257         if (ret) {
1258                 free_extent_buffer(root->node);
1259                 kfree(root);
1260                 return ERR_PTR(ret);
1261         }
1262 #if 0
1263         ret = btrfs_sysfs_add_root(root);
1264         if (ret) {
1265                 free_extent_buffer(root->node);
1266                 kfree(root->name);
1267                 kfree(root);
1268                 return ERR_PTR(ret);
1269         }
1270 #endif
1271         root->in_sysfs = 1;
1272         return root;
1273 }
1274
1275 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1276 {
1277         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1278         int ret = 0;
1279         struct btrfs_device *device;
1280         struct backing_dev_info *bdi;
1281 #if 0
1282         if ((bdi_bits & (1 << BDI_write_congested)) &&
1283             btrfs_congested_async(info, 0))
1284                 return 1;
1285 #endif
1286         list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1287                 if (!device->bdev)
1288                         continue;
1289                 bdi = blk_get_backing_dev_info(device->bdev);
1290                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1291                         ret = 1;
1292                         break;
1293                 }
1294         }
1295         return ret;
1296 }
1297
1298 /*
1299  * this unplugs every device on the box, and it is only used when page
1300  * is null
1301  */
1302 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1303 {
1304         struct btrfs_device *device;
1305         struct btrfs_fs_info *info;
1306
1307         info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1308         list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1309                 if (!device->bdev)
1310                         continue;
1311
1312                 bdi = blk_get_backing_dev_info(device->bdev);
1313                 if (bdi->unplug_io_fn)
1314                         bdi->unplug_io_fn(bdi, page);
1315         }
1316 }
1317
1318 static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1319 {
1320         struct inode *inode;
1321         struct extent_map_tree *em_tree;
1322         struct extent_map *em;
1323         struct address_space *mapping;
1324         u64 offset;
1325
1326         /* the generic O_DIRECT read code does this */
1327         if (1 || !page) {
1328                 __unplug_io_fn(bdi, page);
1329                 return;
1330         }
1331
1332         /*
1333          * page->mapping may change at any time.  Get a consistent copy
1334          * and use that for everything below
1335          */
1336         smp_mb();
1337         mapping = page->mapping;
1338         if (!mapping)
1339                 return;
1340
1341         inode = mapping->host;
1342
1343         /*
1344          * don't do the expensive searching for a small number of
1345          * devices
1346          */
1347         if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1348                 __unplug_io_fn(bdi, page);
1349                 return;
1350         }
1351
1352         offset = page_offset(page);
1353
1354         em_tree = &BTRFS_I(inode)->extent_tree;
1355         spin_lock(&em_tree->lock);
1356         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1357         spin_unlock(&em_tree->lock);
1358         if (!em) {
1359                 __unplug_io_fn(bdi, page);
1360                 return;
1361         }
1362
1363         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1364                 free_extent_map(em);
1365                 __unplug_io_fn(bdi, page);
1366                 return;
1367         }
1368         offset = offset - em->start;
1369         btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1370                           em->block_start + offset, page);
1371         free_extent_map(em);
1372 }
1373
1374 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1375 {
1376         bdi_init(bdi);
1377         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1378         bdi->state              = 0;
1379         bdi->capabilities       = default_backing_dev_info.capabilities;
1380         bdi->unplug_io_fn       = btrfs_unplug_io_fn;
1381         bdi->unplug_io_data     = info;
1382         bdi->congested_fn       = btrfs_congested_fn;
1383         bdi->congested_data     = info;
1384         return 0;
1385 }
1386
1387 static int bio_ready_for_csum(struct bio *bio)
1388 {
1389         u64 length = 0;
1390         u64 buf_len = 0;
1391         u64 start = 0;
1392         struct page *page;
1393         struct extent_io_tree *io_tree = NULL;
1394         struct btrfs_fs_info *info = NULL;
1395         struct bio_vec *bvec;
1396         int i;
1397         int ret;
1398
1399         bio_for_each_segment(bvec, bio, i) {
1400                 page = bvec->bv_page;
1401                 if (page->private == EXTENT_PAGE_PRIVATE) {
1402                         length += bvec->bv_len;
1403                         continue;
1404                 }
1405                 if (!page->private) {
1406                         length += bvec->bv_len;
1407                         continue;
1408                 }
1409                 length = bvec->bv_len;
1410                 buf_len = page->private >> 2;
1411                 start = page_offset(page) + bvec->bv_offset;
1412                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1413                 info = BTRFS_I(page->mapping->host)->root->fs_info;
1414         }
1415         /* are we fully contained in this bio? */
1416         if (buf_len <= length)
1417                 return 1;
1418
1419         ret = extent_range_uptodate(io_tree, start + length,
1420                                     start + buf_len - 1);
1421         return ret;
1422 }
1423
1424 /*
1425  * called by the kthread helper functions to finally call the bio end_io
1426  * functions.  This is where read checksum verification actually happens
1427  */
1428 static void end_workqueue_fn(struct btrfs_work *work)
1429 {
1430         struct bio *bio;
1431         struct end_io_wq *end_io_wq;
1432         struct btrfs_fs_info *fs_info;
1433         int error;
1434
1435         end_io_wq = container_of(work, struct end_io_wq, work);
1436         bio = end_io_wq->bio;
1437         fs_info = end_io_wq->info;
1438
1439         /* metadata bio reads are special because the whole tree block must
1440          * be checksummed at once.  This makes sure the entire block is in
1441          * ram and up to date before trying to verify things.  For
1442          * blocksize <= pagesize, it is basically a noop
1443          */
1444         if (!(bio->bi_rw & (1 << BIO_RW)) && end_io_wq->metadata &&
1445             !bio_ready_for_csum(bio)) {
1446                 btrfs_queue_worker(&fs_info->endio_meta_workers,
1447                                    &end_io_wq->work);
1448                 return;
1449         }
1450         error = end_io_wq->error;
1451         bio->bi_private = end_io_wq->private;
1452         bio->bi_end_io = end_io_wq->end_io;
1453         kfree(end_io_wq);
1454         bio_endio(bio, error);
1455 }
1456
1457 static int cleaner_kthread(void *arg)
1458 {
1459         struct btrfs_root *root = arg;
1460
1461         do {
1462                 smp_mb();
1463                 if (root->fs_info->closing)
1464                         break;
1465
1466                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1467                 mutex_lock(&root->fs_info->cleaner_mutex);
1468                 btrfs_clean_old_snapshots(root);
1469                 mutex_unlock(&root->fs_info->cleaner_mutex);
1470
1471                 if (freezing(current)) {
1472                         refrigerator();
1473                 } else {
1474                         smp_mb();
1475                         if (root->fs_info->closing)
1476                                 break;
1477                         set_current_state(TASK_INTERRUPTIBLE);
1478                         schedule();
1479                         __set_current_state(TASK_RUNNING);
1480                 }
1481         } while (!kthread_should_stop());
1482         return 0;
1483 }
1484
1485 static int transaction_kthread(void *arg)
1486 {
1487         struct btrfs_root *root = arg;
1488         struct btrfs_trans_handle *trans;
1489         struct btrfs_transaction *cur;
1490         unsigned long now;
1491         unsigned long delay;
1492         int ret;
1493
1494         do {
1495                 smp_mb();
1496                 if (root->fs_info->closing)
1497                         break;
1498
1499                 delay = HZ * 30;
1500                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1501                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1502
1503                 mutex_lock(&root->fs_info->trans_mutex);
1504                 cur = root->fs_info->running_transaction;
1505                 if (!cur) {
1506                         mutex_unlock(&root->fs_info->trans_mutex);
1507                         goto sleep;
1508                 }
1509
1510                 now = get_seconds();
1511                 if (now < cur->start_time || now - cur->start_time < 30) {
1512                         mutex_unlock(&root->fs_info->trans_mutex);
1513                         delay = HZ * 5;
1514                         goto sleep;
1515                 }
1516                 mutex_unlock(&root->fs_info->trans_mutex);
1517                 trans = btrfs_start_transaction(root, 1);
1518                 ret = btrfs_commit_transaction(trans, root);
1519
1520 sleep:
1521                 wake_up_process(root->fs_info->cleaner_kthread);
1522                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1523
1524                 if (freezing(current)) {
1525                         refrigerator();
1526                 } else {
1527                         if (root->fs_info->closing)
1528                                 break;
1529                         set_current_state(TASK_INTERRUPTIBLE);
1530                         schedule_timeout(delay);
1531                         __set_current_state(TASK_RUNNING);
1532                 }
1533         } while (!kthread_should_stop());
1534         return 0;
1535 }
1536
1537 struct btrfs_root *open_ctree(struct super_block *sb,
1538                               struct btrfs_fs_devices *fs_devices,
1539                               char *options)
1540 {
1541         u32 sectorsize;
1542         u32 nodesize;
1543         u32 leafsize;
1544         u32 blocksize;
1545         u32 stripesize;
1546         u64 generation;
1547         u64 features;
1548         struct btrfs_key location;
1549         struct buffer_head *bh;
1550         struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1551                                                  GFP_NOFS);
1552         struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1553                                                  GFP_NOFS);
1554         struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
1555                                                GFP_NOFS);
1556         struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1557                                                 GFP_NOFS);
1558         struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1559                                                 GFP_NOFS);
1560         struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1561                                               GFP_NOFS);
1562         struct btrfs_root *log_tree_root;
1563
1564         int ret;
1565         int err = -EINVAL;
1566
1567         struct btrfs_super_block *disk_super;
1568
1569         if (!extent_root || !tree_root || !fs_info ||
1570             !chunk_root || !dev_root || !csum_root) {
1571                 err = -ENOMEM;
1572                 goto fail;
1573         }
1574         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1575         INIT_LIST_HEAD(&fs_info->trans_list);
1576         INIT_LIST_HEAD(&fs_info->dead_roots);
1577         INIT_LIST_HEAD(&fs_info->hashers);
1578         INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1579         INIT_LIST_HEAD(&fs_info->ordered_operations);
1580         spin_lock_init(&fs_info->delalloc_lock);
1581         spin_lock_init(&fs_info->new_trans_lock);
1582         spin_lock_init(&fs_info->ref_cache_lock);
1583
1584         init_completion(&fs_info->kobj_unregister);
1585         fs_info->tree_root = tree_root;
1586         fs_info->extent_root = extent_root;
1587         fs_info->csum_root = csum_root;
1588         fs_info->chunk_root = chunk_root;
1589         fs_info->dev_root = dev_root;
1590         fs_info->fs_devices = fs_devices;
1591         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1592         INIT_LIST_HEAD(&fs_info->space_info);
1593         btrfs_mapping_init(&fs_info->mapping_tree);
1594         atomic_set(&fs_info->nr_async_submits, 0);
1595         atomic_set(&fs_info->async_delalloc_pages, 0);
1596         atomic_set(&fs_info->async_submit_draining, 0);
1597         atomic_set(&fs_info->nr_async_bios, 0);
1598         atomic_set(&fs_info->throttles, 0);
1599         atomic_set(&fs_info->throttle_gen, 0);
1600         fs_info->sb = sb;
1601         fs_info->max_extent = (u64)-1;
1602         fs_info->max_inline = 8192 * 1024;
1603         setup_bdi(fs_info, &fs_info->bdi);
1604         fs_info->btree_inode = new_inode(sb);
1605         fs_info->btree_inode->i_ino = 1;
1606         fs_info->btree_inode->i_nlink = 1;
1607         fs_info->metadata_ratio = 8;
1608
1609         fs_info->thread_pool_size = min_t(unsigned long,
1610                                           num_online_cpus() + 2, 8);
1611
1612         INIT_LIST_HEAD(&fs_info->ordered_extents);
1613         spin_lock_init(&fs_info->ordered_extent_lock);
1614
1615         sb->s_blocksize = 4096;
1616         sb->s_blocksize_bits = blksize_bits(4096);
1617
1618         /*
1619          * we set the i_size on the btree inode to the max possible int.
1620          * the real end of the address space is determined by all of
1621          * the devices in the system
1622          */
1623         fs_info->btree_inode->i_size = OFFSET_MAX;
1624         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1625         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1626
1627         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1628                              fs_info->btree_inode->i_mapping,
1629                              GFP_NOFS);
1630         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1631                              GFP_NOFS);
1632
1633         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1634
1635         spin_lock_init(&fs_info->block_group_cache_lock);
1636         fs_info->block_group_cache_tree.rb_node = NULL;
1637
1638         extent_io_tree_init(&fs_info->pinned_extents,
1639                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1640         fs_info->do_barriers = 1;
1641
1642         INIT_LIST_HEAD(&fs_info->dead_reloc_roots);
1643         btrfs_leaf_ref_tree_init(&fs_info->reloc_ref_tree);
1644         btrfs_leaf_ref_tree_init(&fs_info->shared_ref_tree);
1645
1646         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1647         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1648                sizeof(struct btrfs_key));
1649         insert_inode_hash(fs_info->btree_inode);
1650
1651         mutex_init(&fs_info->trans_mutex);
1652         mutex_init(&fs_info->ordered_operations_mutex);
1653         mutex_init(&fs_info->tree_log_mutex);
1654         mutex_init(&fs_info->drop_mutex);
1655         mutex_init(&fs_info->chunk_mutex);
1656         mutex_init(&fs_info->transaction_kthread_mutex);
1657         mutex_init(&fs_info->cleaner_mutex);
1658         mutex_init(&fs_info->volume_mutex);
1659         mutex_init(&fs_info->tree_reloc_mutex);
1660
1661         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1662         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1663
1664         init_waitqueue_head(&fs_info->transaction_throttle);
1665         init_waitqueue_head(&fs_info->transaction_wait);
1666         init_waitqueue_head(&fs_info->async_submit_wait);
1667
1668         __setup_root(4096, 4096, 4096, 4096, tree_root,
1669                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1670
1671
1672         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1673         if (!bh)
1674                 goto fail_iput;
1675
1676         memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1677         memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1678                sizeof(fs_info->super_for_commit));
1679         brelse(bh);
1680
1681         memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1682
1683         disk_super = &fs_info->super_copy;
1684         if (!btrfs_super_root(disk_super))
1685                 goto fail_iput;
1686
1687         ret = btrfs_parse_options(tree_root, options);
1688         if (ret) {
1689                 err = ret;
1690                 goto fail_iput;
1691         }
1692
1693         features = btrfs_super_incompat_flags(disk_super) &
1694                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1695         if (features) {
1696                 printk(KERN_ERR "BTRFS: couldn't mount because of "
1697                        "unsupported optional features (%Lx).\n",
1698                        features);
1699                 err = -EINVAL;
1700                 goto fail_iput;
1701         }
1702
1703         features = btrfs_super_compat_ro_flags(disk_super) &
1704                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1705         if (!(sb->s_flags & MS_RDONLY) && features) {
1706                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1707                        "unsupported option features (%Lx).\n",
1708                        features);
1709                 err = -EINVAL;
1710                 goto fail_iput;
1711         }
1712
1713         /*
1714          * we need to start all the end_io workers up front because the
1715          * queue work function gets called at interrupt time, and so it
1716          * cannot dynamically grow.
1717          */
1718         btrfs_init_workers(&fs_info->workers, "worker",
1719                            fs_info->thread_pool_size);
1720
1721         btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1722                            fs_info->thread_pool_size);
1723
1724         btrfs_init_workers(&fs_info->submit_workers, "submit",
1725                            min_t(u64, fs_devices->num_devices,
1726                            fs_info->thread_pool_size));
1727
1728         /* a higher idle thresh on the submit workers makes it much more
1729          * likely that bios will be send down in a sane order to the
1730          * devices
1731          */
1732         fs_info->submit_workers.idle_thresh = 64;
1733
1734         fs_info->workers.idle_thresh = 16;
1735         fs_info->workers.ordered = 1;
1736
1737         fs_info->delalloc_workers.idle_thresh = 2;
1738         fs_info->delalloc_workers.ordered = 1;
1739
1740         btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1);
1741         btrfs_init_workers(&fs_info->endio_workers, "endio",
1742                            fs_info->thread_pool_size);
1743         btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1744                            fs_info->thread_pool_size);
1745         btrfs_init_workers(&fs_info->endio_meta_write_workers,
1746                            "endio-meta-write", fs_info->thread_pool_size);
1747         btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1748                            fs_info->thread_pool_size);
1749
1750         /*
1751          * endios are largely parallel and should have a very
1752          * low idle thresh
1753          */
1754         fs_info->endio_workers.idle_thresh = 4;
1755         fs_info->endio_meta_workers.idle_thresh = 4;
1756
1757         fs_info->endio_write_workers.idle_thresh = 64;
1758         fs_info->endio_meta_write_workers.idle_thresh = 64;
1759
1760         btrfs_start_workers(&fs_info->workers, 1);
1761         btrfs_start_workers(&fs_info->submit_workers, 1);
1762         btrfs_start_workers(&fs_info->delalloc_workers, 1);
1763         btrfs_start_workers(&fs_info->fixup_workers, 1);
1764         btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
1765         btrfs_start_workers(&fs_info->endio_meta_workers,
1766                             fs_info->thread_pool_size);
1767         btrfs_start_workers(&fs_info->endio_meta_write_workers,
1768                             fs_info->thread_pool_size);
1769         btrfs_start_workers(&fs_info->endio_write_workers,
1770                             fs_info->thread_pool_size);
1771
1772         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1773         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1774                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1775
1776         nodesize = btrfs_super_nodesize(disk_super);
1777         leafsize = btrfs_super_leafsize(disk_super);
1778         sectorsize = btrfs_super_sectorsize(disk_super);
1779         stripesize = btrfs_super_stripesize(disk_super);
1780         tree_root->nodesize = nodesize;
1781         tree_root->leafsize = leafsize;
1782         tree_root->sectorsize = sectorsize;
1783         tree_root->stripesize = stripesize;
1784
1785         sb->s_blocksize = sectorsize;
1786         sb->s_blocksize_bits = blksize_bits(sectorsize);
1787
1788         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1789                     sizeof(disk_super->magic))) {
1790                 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
1791                 goto fail_sb_buffer;
1792         }
1793
1794         mutex_lock(&fs_info->chunk_mutex);
1795         ret = btrfs_read_sys_array(tree_root);
1796         mutex_unlock(&fs_info->chunk_mutex);
1797         if (ret) {
1798                 printk(KERN_WARNING "btrfs: failed to read the system "
1799                        "array on %s\n", sb->s_id);
1800                 goto fail_sys_array;
1801         }
1802
1803         blocksize = btrfs_level_size(tree_root,
1804                                      btrfs_super_chunk_root_level(disk_super));
1805         generation = btrfs_super_chunk_root_generation(disk_super);
1806
1807         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1808                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1809
1810         chunk_root->node = read_tree_block(chunk_root,
1811                                            btrfs_super_chunk_root(disk_super),
1812                                            blocksize, generation);
1813         BUG_ON(!chunk_root->node);
1814
1815         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1816            (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1817            BTRFS_UUID_SIZE);
1818
1819         mutex_lock(&fs_info->chunk_mutex);
1820         ret = btrfs_read_chunk_tree(chunk_root);
1821         mutex_unlock(&fs_info->chunk_mutex);
1822         if (ret) {
1823                 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
1824                        sb->s_id);
1825                 goto fail_chunk_root;
1826         }
1827
1828         btrfs_close_extra_devices(fs_devices);
1829
1830         blocksize = btrfs_level_size(tree_root,
1831                                      btrfs_super_root_level(disk_super));
1832         generation = btrfs_super_generation(disk_super);
1833
1834         tree_root->node = read_tree_block(tree_root,
1835                                           btrfs_super_root(disk_super),
1836                                           blocksize, generation);
1837         if (!tree_root->node)
1838                 goto fail_chunk_root;
1839
1840
1841         ret = find_and_setup_root(tree_root, fs_info,
1842                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1843         if (ret)
1844                 goto fail_tree_root;
1845         extent_root->track_dirty = 1;
1846
1847         ret = find_and_setup_root(tree_root, fs_info,
1848                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
1849         dev_root->track_dirty = 1;
1850         if (ret)
1851                 goto fail_extent_root;
1852
1853         ret = find_and_setup_root(tree_root, fs_info,
1854                                   BTRFS_CSUM_TREE_OBJECTID, csum_root);
1855         if (ret)
1856                 goto fail_extent_root;
1857
1858         csum_root->track_dirty = 1;
1859
1860         btrfs_read_block_groups(extent_root);
1861
1862         fs_info->generation = generation;
1863         fs_info->last_trans_committed = generation;
1864         fs_info->data_alloc_profile = (u64)-1;
1865         fs_info->metadata_alloc_profile = (u64)-1;
1866         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1867         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1868                                                "btrfs-cleaner");
1869         if (IS_ERR(fs_info->cleaner_kthread))
1870                 goto fail_csum_root;
1871
1872         fs_info->transaction_kthread = kthread_run(transaction_kthread,
1873                                                    tree_root,
1874                                                    "btrfs-transaction");
1875         if (IS_ERR(fs_info->transaction_kthread))
1876                 goto fail_cleaner;
1877
1878         if (btrfs_super_log_root(disk_super) != 0) {
1879                 u64 bytenr = btrfs_super_log_root(disk_super);
1880
1881                 if (fs_devices->rw_devices == 0) {
1882                         printk(KERN_WARNING "Btrfs log replay required "
1883                                "on RO media\n");
1884                         err = -EIO;
1885                         goto fail_trans_kthread;
1886                 }
1887                 blocksize =
1888                      btrfs_level_size(tree_root,
1889                                       btrfs_super_log_root_level(disk_super));
1890
1891                 log_tree_root = kzalloc(sizeof(struct btrfs_root),
1892                                                       GFP_NOFS);
1893
1894                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1895                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1896
1897                 log_tree_root->node = read_tree_block(tree_root, bytenr,
1898                                                       blocksize,
1899                                                       generation + 1);
1900                 ret = btrfs_recover_log_trees(log_tree_root);
1901                 BUG_ON(ret);
1902
1903                 if (sb->s_flags & MS_RDONLY) {
1904                         ret =  btrfs_commit_super(tree_root);
1905                         BUG_ON(ret);
1906                 }
1907         }
1908
1909         if (!(sb->s_flags & MS_RDONLY)) {
1910                 ret = btrfs_cleanup_reloc_trees(tree_root);
1911                 BUG_ON(ret);
1912         }
1913
1914         location.objectid = BTRFS_FS_TREE_OBJECTID;
1915         location.type = BTRFS_ROOT_ITEM_KEY;
1916         location.offset = (u64)-1;
1917
1918         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
1919         if (!fs_info->fs_root)
1920                 goto fail_trans_kthread;
1921         return tree_root;
1922
1923 fail_trans_kthread:
1924         kthread_stop(fs_info->transaction_kthread);
1925 fail_cleaner:
1926         kthread_stop(fs_info->cleaner_kthread);
1927
1928         /*
1929          * make sure we're done with the btree inode before we stop our
1930          * kthreads
1931          */
1932         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1933         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
1934
1935 fail_csum_root:
1936         free_extent_buffer(csum_root->node);
1937 fail_extent_root:
1938         free_extent_buffer(extent_root->node);
1939 fail_tree_root:
1940         free_extent_buffer(tree_root->node);
1941 fail_chunk_root:
1942         free_extent_buffer(chunk_root->node);
1943 fail_sys_array:
1944         free_extent_buffer(dev_root->node);
1945 fail_sb_buffer:
1946         btrfs_stop_workers(&fs_info->fixup_workers);
1947         btrfs_stop_workers(&fs_info->delalloc_workers);
1948         btrfs_stop_workers(&fs_info->workers);
1949         btrfs_stop_workers(&fs_info->endio_workers);
1950         btrfs_stop_workers(&fs_info->endio_meta_workers);
1951         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
1952         btrfs_stop_workers(&fs_info->endio_write_workers);
1953         btrfs_stop_workers(&fs_info->submit_workers);
1954 fail_iput:
1955         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
1956         iput(fs_info->btree_inode);
1957
1958         btrfs_close_devices(fs_info->fs_devices);
1959         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1960         bdi_destroy(&fs_info->bdi);
1961
1962 fail:
1963         kfree(extent_root);
1964         kfree(tree_root);
1965         kfree(fs_info);
1966         kfree(chunk_root);
1967         kfree(dev_root);
1968         kfree(csum_root);
1969         return ERR_PTR(err);
1970 }
1971
1972 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1973 {
1974         char b[BDEVNAME_SIZE];
1975
1976         if (uptodate) {
1977                 set_buffer_uptodate(bh);
1978         } else {
1979                 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1980                         printk(KERN_WARNING "lost page write due to "
1981                                         "I/O error on %s\n",
1982                                        bdevname(bh->b_bdev, b));
1983                 }
1984                 /* note, we dont' set_buffer_write_io_error because we have
1985                  * our own ways of dealing with the IO errors
1986                  */
1987                 clear_buffer_uptodate(bh);
1988         }
1989         unlock_buffer(bh);
1990         put_bh(bh);
1991 }
1992
1993 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
1994 {
1995         struct buffer_head *bh;
1996         struct buffer_head *latest = NULL;
1997         struct btrfs_super_block *super;
1998         int i;
1999         u64 transid = 0;
2000         u64 bytenr;
2001
2002         /* we would like to check all the supers, but that would make
2003          * a btrfs mount succeed after a mkfs from a different FS.
2004          * So, we need to add a special mount option to scan for
2005          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2006          */
2007         for (i = 0; i < 1; i++) {
2008                 bytenr = btrfs_sb_offset(i);
2009                 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2010                         break;
2011                 bh = __bread(bdev, bytenr / 4096, 4096);
2012                 if (!bh)
2013                         continue;
2014
2015                 super = (struct btrfs_super_block *)bh->b_data;
2016                 if (btrfs_super_bytenr(super) != bytenr ||
2017                     strncmp((char *)(&super->magic), BTRFS_MAGIC,
2018                             sizeof(super->magic))) {
2019                         brelse(bh);
2020                         continue;
2021                 }
2022
2023                 if (!latest || btrfs_super_generation(super) > transid) {
2024                         brelse(latest);
2025                         latest = bh;
2026                         transid = btrfs_super_generation(super);
2027                 } else {
2028                         brelse(bh);
2029                 }
2030         }
2031         return latest;
2032 }
2033
2034 static int write_dev_supers(struct btrfs_device *device,
2035                             struct btrfs_super_block *sb,
2036                             int do_barriers, int wait, int max_mirrors)
2037 {
2038         struct buffer_head *bh;
2039         int i;
2040         int ret;
2041         int errors = 0;
2042         u32 crc;
2043         u64 bytenr;
2044         int last_barrier = 0;
2045
2046         if (max_mirrors == 0)
2047                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2048
2049         /* make sure only the last submit_bh does a barrier */
2050         if (do_barriers) {
2051                 for (i = 0; i < max_mirrors; i++) {
2052                         bytenr = btrfs_sb_offset(i);
2053                         if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2054                             device->total_bytes)
2055                                 break;
2056                         last_barrier = i;
2057                 }
2058         }
2059
2060         for (i = 0; i < max_mirrors; i++) {
2061                 bytenr = btrfs_sb_offset(i);
2062                 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2063                         break;
2064
2065                 if (wait) {
2066                         bh = __find_get_block(device->bdev, bytenr / 4096,
2067                                               BTRFS_SUPER_INFO_SIZE);
2068                         BUG_ON(!bh);
2069                         brelse(bh);
2070                         wait_on_buffer(bh);
2071                         if (buffer_uptodate(bh)) {
2072                                 brelse(bh);
2073                                 continue;
2074                         }
2075                 } else {
2076                         btrfs_set_super_bytenr(sb, bytenr);
2077
2078                         crc = ~(u32)0;
2079                         crc = btrfs_csum_data(NULL, (char *)sb +
2080                                               BTRFS_CSUM_SIZE, crc,
2081                                               BTRFS_SUPER_INFO_SIZE -
2082                                               BTRFS_CSUM_SIZE);
2083                         btrfs_csum_final(crc, sb->csum);
2084
2085                         bh = __getblk(device->bdev, bytenr / 4096,
2086                                       BTRFS_SUPER_INFO_SIZE);
2087                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2088
2089                         set_buffer_uptodate(bh);
2090                         get_bh(bh);
2091                         lock_buffer(bh);
2092                         bh->b_end_io = btrfs_end_buffer_write_sync;
2093                 }
2094
2095                 if (i == last_barrier && do_barriers && device->barriers) {
2096                         ret = submit_bh(WRITE_BARRIER, bh);
2097                         if (ret == -EOPNOTSUPP) {
2098                                 printk("btrfs: disabling barriers on dev %s\n",
2099                                        device->name);
2100                                 set_buffer_uptodate(bh);
2101                                 device->barriers = 0;
2102                                 get_bh(bh);
2103                                 lock_buffer(bh);
2104                                 ret = submit_bh(WRITE_SYNC, bh);
2105                         }
2106                 } else {
2107                         ret = submit_bh(WRITE_SYNC, bh);
2108                 }
2109
2110                 if (!ret && wait) {
2111                         wait_on_buffer(bh);
2112                         if (!buffer_uptodate(bh))
2113                                 errors++;
2114                 } else if (ret) {
2115                         errors++;
2116                 }
2117                 if (wait)
2118                         brelse(bh);
2119         }
2120         return errors < i ? 0 : -1;
2121 }
2122
2123 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2124 {
2125         struct list_head *head = &root->fs_info->fs_devices->devices;
2126         struct btrfs_device *dev;
2127         struct btrfs_super_block *sb;
2128         struct btrfs_dev_item *dev_item;
2129         int ret;
2130         int do_barriers;
2131         int max_errors;
2132         int total_errors = 0;
2133         u64 flags;
2134
2135         max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2136         do_barriers = !btrfs_test_opt(root, NOBARRIER);
2137
2138         sb = &root->fs_info->super_for_commit;
2139         dev_item = &sb->dev_item;
2140         list_for_each_entry(dev, head, dev_list) {
2141                 if (!dev->bdev) {
2142                         total_errors++;
2143                         continue;
2144                 }
2145                 if (!dev->in_fs_metadata || !dev->writeable)
2146                         continue;
2147
2148                 btrfs_set_stack_device_generation(dev_item, 0);
2149                 btrfs_set_stack_device_type(dev_item, dev->type);
2150                 btrfs_set_stack_device_id(dev_item, dev->devid);
2151                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2152                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2153                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2154                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2155                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2156                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2157                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2158
2159                 flags = btrfs_super_flags(sb);
2160                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2161
2162                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2163                 if (ret)
2164                         total_errors++;
2165         }
2166         if (total_errors > max_errors) {
2167                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2168                        total_errors);
2169                 BUG();
2170         }
2171
2172         total_errors = 0;
2173         list_for_each_entry(dev, head, dev_list) {
2174                 if (!dev->bdev)
2175                         continue;
2176                 if (!dev->in_fs_metadata || !dev->writeable)
2177                         continue;
2178
2179                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2180                 if (ret)
2181                         total_errors++;
2182         }
2183         if (total_errors > max_errors) {
2184                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2185                        total_errors);
2186                 BUG();
2187         }
2188         return 0;
2189 }
2190
2191 int write_ctree_super(struct btrfs_trans_handle *trans,
2192                       struct btrfs_root *root, int max_mirrors)
2193 {
2194         int ret;
2195
2196         ret = write_all_supers(root, max_mirrors);
2197         return ret;
2198 }
2199
2200 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2201 {
2202         radix_tree_delete(&fs_info->fs_roots_radix,
2203                           (unsigned long)root->root_key.objectid);
2204         if (root->anon_super.s_dev) {
2205                 down_write(&root->anon_super.s_umount);
2206                 kill_anon_super(&root->anon_super);
2207         }
2208         if (root->node)
2209                 free_extent_buffer(root->node);
2210         if (root->commit_root)
2211                 free_extent_buffer(root->commit_root);
2212         kfree(root->name);
2213         kfree(root);
2214         return 0;
2215 }
2216
2217 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2218 {
2219         int ret;
2220         struct btrfs_root *gang[8];
2221         int i;
2222
2223         while (1) {
2224                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2225                                              (void **)gang, 0,
2226                                              ARRAY_SIZE(gang));
2227                 if (!ret)
2228                         break;
2229                 for (i = 0; i < ret; i++)
2230                         btrfs_free_fs_root(fs_info, gang[i]);
2231         }
2232         return 0;
2233 }
2234
2235 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2236 {
2237         u64 root_objectid = 0;
2238         struct btrfs_root *gang[8];
2239         int i;
2240         int ret;
2241
2242         while (1) {
2243                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2244                                              (void **)gang, root_objectid,
2245                                              ARRAY_SIZE(gang));
2246                 if (!ret)
2247                         break;
2248                 for (i = 0; i < ret; i++) {
2249                         root_objectid = gang[i]->root_key.objectid;
2250                         ret = btrfs_find_dead_roots(fs_info->tree_root,
2251                                                     root_objectid, gang[i]);
2252                         BUG_ON(ret);
2253                         btrfs_orphan_cleanup(gang[i]);
2254                 }
2255                 root_objectid++;
2256         }
2257         return 0;
2258 }
2259
2260 int btrfs_commit_super(struct btrfs_root *root)
2261 {
2262         struct btrfs_trans_handle *trans;
2263         int ret;
2264
2265         mutex_lock(&root->fs_info->cleaner_mutex);
2266         btrfs_clean_old_snapshots(root);
2267         mutex_unlock(&root->fs_info->cleaner_mutex);
2268         trans = btrfs_start_transaction(root, 1);
2269         ret = btrfs_commit_transaction(trans, root);
2270         BUG_ON(ret);
2271         /* run commit again to drop the original snapshot */
2272         trans = btrfs_start_transaction(root, 1);
2273         btrfs_commit_transaction(trans, root);
2274         ret = btrfs_write_and_wait_transaction(NULL, root);
2275         BUG_ON(ret);
2276
2277         ret = write_ctree_super(NULL, root, 0);
2278         return ret;
2279 }
2280
2281 int close_ctree(struct btrfs_root *root)
2282 {
2283         struct btrfs_fs_info *fs_info = root->fs_info;
2284         int ret;
2285
2286         fs_info->closing = 1;
2287         smp_mb();
2288
2289         kthread_stop(root->fs_info->transaction_kthread);
2290         kthread_stop(root->fs_info->cleaner_kthread);
2291
2292         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2293                 ret =  btrfs_commit_super(root);
2294                 if (ret)
2295                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2296         }
2297
2298         if (fs_info->delalloc_bytes) {
2299                 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2300                        fs_info->delalloc_bytes);
2301         }
2302         if (fs_info->total_ref_cache_size) {
2303                 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2304                        (unsigned long long)fs_info->total_ref_cache_size);
2305         }
2306
2307         if (fs_info->extent_root->node)
2308                 free_extent_buffer(fs_info->extent_root->node);
2309
2310         if (fs_info->tree_root->node)
2311                 free_extent_buffer(fs_info->tree_root->node);
2312
2313         if (root->fs_info->chunk_root->node)
2314                 free_extent_buffer(root->fs_info->chunk_root->node);
2315
2316         if (root->fs_info->dev_root->node)
2317                 free_extent_buffer(root->fs_info->dev_root->node);
2318
2319         if (root->fs_info->csum_root->node)
2320                 free_extent_buffer(root->fs_info->csum_root->node);
2321
2322         btrfs_free_block_groups(root->fs_info);
2323
2324         del_fs_roots(fs_info);
2325
2326         iput(fs_info->btree_inode);
2327
2328         btrfs_stop_workers(&fs_info->fixup_workers);
2329         btrfs_stop_workers(&fs_info->delalloc_workers);
2330         btrfs_stop_workers(&fs_info->workers);
2331         btrfs_stop_workers(&fs_info->endio_workers);
2332         btrfs_stop_workers(&fs_info->endio_meta_workers);
2333         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2334         btrfs_stop_workers(&fs_info->endio_write_workers);
2335         btrfs_stop_workers(&fs_info->submit_workers);
2336
2337 #if 0
2338         while (!list_empty(&fs_info->hashers)) {
2339                 struct btrfs_hasher *hasher;
2340                 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
2341                                     hashers);
2342                 list_del(&hasher->hashers);
2343                 crypto_free_hash(&fs_info->hash_tfm);
2344                 kfree(hasher);
2345         }
2346 #endif
2347         btrfs_close_devices(fs_info->fs_devices);
2348         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2349
2350         bdi_destroy(&fs_info->bdi);
2351
2352         kfree(fs_info->extent_root);
2353         kfree(fs_info->tree_root);
2354         kfree(fs_info->chunk_root);
2355         kfree(fs_info->dev_root);
2356         kfree(fs_info->csum_root);
2357         return 0;
2358 }
2359
2360 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2361 {
2362         int ret;
2363         struct inode *btree_inode = buf->first_page->mapping->host;
2364
2365         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
2366         if (!ret)
2367                 return ret;
2368
2369         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2370                                     parent_transid);
2371         return !ret;
2372 }
2373
2374 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2375 {
2376         struct inode *btree_inode = buf->first_page->mapping->host;
2377         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2378                                           buf);
2379 }
2380
2381 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2382 {
2383         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2384         u64 transid = btrfs_header_generation(buf);
2385         struct inode *btree_inode = root->fs_info->btree_inode;
2386         int was_dirty;
2387
2388         btrfs_assert_tree_locked(buf);
2389         if (transid != root->fs_info->generation) {
2390                 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2391                        "found %llu running %llu\n",
2392                         (unsigned long long)buf->start,
2393                         (unsigned long long)transid,
2394                         (unsigned long long)root->fs_info->generation);
2395                 WARN_ON(1);
2396         }
2397         was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2398                                             buf);
2399         if (!was_dirty) {
2400                 spin_lock(&root->fs_info->delalloc_lock);
2401                 root->fs_info->dirty_metadata_bytes += buf->len;
2402                 spin_unlock(&root->fs_info->delalloc_lock);
2403         }
2404 }
2405
2406 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2407 {
2408         /*
2409          * looks as though older kernels can get into trouble with
2410          * this code, they end up stuck in balance_dirty_pages forever
2411          */
2412         struct extent_io_tree *tree;
2413         u64 num_dirty;
2414         u64 start = 0;
2415         unsigned long thresh = 32 * 1024 * 1024;
2416         tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
2417
2418         if (current->flags & PF_MEMALLOC)
2419                 return;
2420
2421         num_dirty = count_range_bits(tree, &start, (u64)-1,
2422                                      thresh, EXTENT_DIRTY);
2423         if (num_dirty > thresh) {
2424                 balance_dirty_pages_ratelimited_nr(
2425                                    root->fs_info->btree_inode->i_mapping, 1);
2426         }
2427         return;
2428 }
2429
2430 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2431 {
2432         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2433         int ret;
2434         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2435         if (ret == 0)
2436                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
2437         return ret;
2438 }
2439
2440 int btree_lock_page_hook(struct page *page)
2441 {
2442         struct inode *inode = page->mapping->host;
2443         struct btrfs_root *root = BTRFS_I(inode)->root;
2444         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2445         struct extent_buffer *eb;
2446         unsigned long len;
2447         u64 bytenr = page_offset(page);
2448
2449         if (page->private == EXTENT_PAGE_PRIVATE)
2450                 goto out;
2451
2452         len = page->private >> 2;
2453         eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2454         if (!eb)
2455                 goto out;
2456
2457         btrfs_tree_lock(eb);
2458         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2459
2460         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2461                 spin_lock(&root->fs_info->delalloc_lock);
2462                 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2463                         root->fs_info->dirty_metadata_bytes -= eb->len;
2464                 else
2465                         WARN_ON(1);
2466                 spin_unlock(&root->fs_info->delalloc_lock);
2467         }
2468
2469         btrfs_tree_unlock(eb);
2470         free_extent_buffer(eb);
2471 out:
2472         lock_page(page);
2473         return 0;
2474 }
2475
2476 static struct extent_io_ops btree_extent_io_ops = {
2477         .write_cache_pages_lock_hook = btree_lock_page_hook,
2478         .readpage_end_io_hook = btree_readpage_end_io_hook,
2479         .submit_bio_hook = btree_submit_bio_hook,
2480         /* note we're sharing with inode.c for the merge bio hook */
2481         .merge_bio_hook = btrfs_merge_bio_hook,
2482 };