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