fs/btrfs/disk-io.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/fs.h>
  20 #include <linux/blkdev.h>
  21 #include <linux/scatterlist.h>
  22 #include <linux/swap.h>
  23 #include <linux/radix-tree.h>
  24 #include <linux/writeback.h>
  25 #include <linux/buffer_head.h>
  26 #include <linux/workqueue.h>
  27 #include <linux/kthread.h>
  28 #include <linux/freezer.h>
  29 #include <linux/crc32c.h>
  30 #include <linux/slab.h>
  31 #include <linux/migrate.h>
  32 #include "compat.h"
  33 #include "ctree.h"
  34 #include "disk-io.h"
  35 #include "transaction.h"
  36 #include "btrfs_inode.h"
  37 #include "volumes.h"
  38 #include "print-tree.h"
  39 #include "async-thread.h"
  40 #include "locking.h"
  41 #include "tree-log.h"
  42 #include "free-space-cache.h"
  43
  44 static struct extent_io_ops btree_extent_io_ops;
  45 static void end_workqueue_fn(struct btrfs_work *work);
  46 static void free_fs_root(struct btrfs_root *root);
  47
  48 /*
  49  * end_io_wq structs are used to do processing in task context when an IO is
  50  * complete.  This is used during reads to verify checksums, and it is used
  51  * by writes to insert metadata for new file extents after IO is complete.
  52  */
  53 struct end_io_wq {
  54         struct bio *bio;
  55         bio_end_io_t *end_io;
  56         void *private;
  57         struct btrfs_fs_info *info;
  58         int error;
  59         int metadata;
  60         struct list_head list;
  61         struct btrfs_work work;
  62 };
  63
  64 /*
  65  * async submit bios are used to offload expensive checksumming
  66  * onto the worker threads.  They checksum file and metadata bios
  67  * just before they are sent down the IO stack.
  68  */
  69 struct async_submit_bio {
  70         struct inode *inode;
  71         struct bio *bio;
  72         struct list_head list;
  73         extent_submit_bio_hook_t *submit_bio_start;
  74         extent_submit_bio_hook_t *submit_bio_done;
  75         int rw;
  76         int mirror_num;
  77         unsigned long bio_flags;
  78         /*
  79          * bio_offset is optional, can be used if the pages in the bio
  80          * can't tell us where in the file the bio should go
  81          */
  82         u64 bio_offset;
  83         struct btrfs_work work;
  84 };
  85
  86 /* These are used to set the lockdep class on the extent buffer locks.
  87  * The class is set by the readpage_end_io_hook after the buffer has
  88  * passed csum validation but before the pages are unlocked.
  89  *
  90  * The lockdep class is also set by btrfs_init_new_buffer on freshly
  91  * allocated blocks.
  92  *
  93  * The class is based on the level in the tree block, which allows lockdep
  94  * to know that lower nodes nest inside the locks of higher nodes.
  95  *
  96  * We also add a check to make sure the highest level of the tree is
  97  * the same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this
  98  * code needs update as well.
  99  */
 100 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 101 # if BTRFS_MAX_LEVEL != 8
 102 #  error
 103 # endif
 104 static struct lock_class_key btrfs_eb_class[BTRFS_MAX_LEVEL + 1];
 105 static const char *btrfs_eb_name[BTRFS_MAX_LEVEL + 1] = {
 106         /* leaf */
 107         "btrfs-extent-00",
 108         "btrfs-extent-01",
 109         "btrfs-extent-02",
 110         "btrfs-extent-03",
 111         "btrfs-extent-04",
 112         "btrfs-extent-05",
 113         "btrfs-extent-06",
 114         "btrfs-extent-07",
 115         /* highest possible level */
 116         "btrfs-extent-08",
 117 };
 118 #endif
 119
 120 /*
 121  * extents on the btree inode are pretty simple, there's one extent
 122  * that covers the entire device
 123  */
 124 static struct extent_map *btree_get_extent(struct inode *inode,
 125                 struct page *page, size_t page_offset, u64 start, u64 len,
 126                 int create)
 127 {
 128         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 129         struct extent_map *em;
 130         int ret;
 131
 132         read_lock(&em_tree->lock);
 133         em = lookup_extent_mapping(em_tree, start, len);
 134         if (em) {
 135                 em->bdev =
 136                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 137                 read_unlock(&em_tree->lock);
 138                 goto out;
 139         }
 140         read_unlock(&em_tree->lock);
 141
 142         em = alloc_extent_map(GFP_NOFS);
 143         if (!em) {
 144                 em = ERR_PTR(-ENOMEM);
 145                 goto out;
 146         }
 147         em->start = 0;
 148         em->len = (u64)-1;
 149         em->block_len = (u64)-1;
 150         em->block_start = 0;
 151         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 152
 153         write_lock(&em_tree->lock);
 154         ret = add_extent_mapping(em_tree, em);
 155         if (ret == -EEXIST) {
 156                 u64 failed_start = em->start;
 157                 u64 failed_len = em->len;
 158
 159                 free_extent_map(em);
 160                 em = lookup_extent_mapping(em_tree, start, len);
 161                 if (em) {
 162                         ret = 0;
 163                 } else {
 164                         em = lookup_extent_mapping(em_tree, failed_start,
 165                                                    failed_len);
 166                         ret = -EIO;
 167                 }
 168         } else if (ret) {
 169                 free_extent_map(em);
 170                 em = NULL;
 171         }
 172         write_unlock(&em_tree->lock);
 173
 174         if (ret)
 175                 em = ERR_PTR(ret);
 176 out:
 177         return em;
 178 }
 179
 180 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
 181 {
 182         return crc32c(seed, data, len);
 183 }
 184
 185 void btrfs_csum_final(u32 crc, char *result)
 186 {
 187         *(__le32 *)result = ~cpu_to_le32(crc);
 188 }
 189
 190 /*
 191  * compute the csum for a btree block, and either verify it or write it
 192  * into the csum field of the block.
 193  */
 194 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 195                            int verify)
 196 {
 197         u16 csum_size =
 198                 btrfs_super_csum_size(&root->fs_info->super_copy);
 199         char *result = NULL;
 200         unsigned long len;
 201         unsigned long cur_len;
 202         unsigned long offset = BTRFS_CSUM_SIZE;
 203         char *map_token = NULL;
 204         char *kaddr;
 205         unsigned long map_start;
 206         unsigned long map_len;
 207         int err;
 208         u32 crc = ~(u32)0;
 209         unsigned long inline_result;
 210
 211         len = buf->len - offset;
 212         while (len > 0) {
 213                 err = map_private_extent_buffer(buf, offset, 32,
 214                                         &map_token, &kaddr,
 215                                         &map_start, &map_len, KM_USER0);
 216                 if (err)
 217                         return 1;
 218                 cur_len = min(len, map_len - (offset - map_start));
 219                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
 220                                       crc, cur_len);
 221                 len -= cur_len;
 222                 offset += cur_len;
 223                 unmap_extent_buffer(buf, map_token, KM_USER0);
 224         }
 225         if (csum_size > sizeof(inline_result)) {
 226                 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
 227                 if (!result)
 228                         return 1;
 229         } else {
 230                 result = (char *)&inline_result;
 231         }
 232
 233         btrfs_csum_final(crc, result);
 234
 235         if (verify) {
 236                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
 237                         u32 val;
 238                         u32 found = 0;
 239                         memcpy(&found, result, csum_size);
 240
 241                         read_extent_buffer(buf, &val, 0, csum_size);
 242                         if (printk_ratelimit()) {
 243                                 printk(KERN_INFO "btrfs: %s checksum verify "
 244                                        "failed on %llu wanted %X found %X "
 245                                        "level %d\n",
 246                                        root->fs_info->sb->s_id,
 247                                        (unsigned long long)buf->start, val, found,
 248                                        btrfs_header_level(buf));
 249                         }
 250                         if (result != (char *)&inline_result)
 251                                 kfree(result);
 252                         return 1;
 253                 }
 254         } else {
 255                 write_extent_buffer(buf, result, 0, csum_size);
 256         }
 257         if (result != (char *)&inline_result)
 258                 kfree(result);
 259         return 0;
 260 }
 261
 262 /*
 263  * we can't consider a given block up to date unless the transid of the
 264  * block matches the transid in the parent node's pointer.  This is how we
 265  * detect blocks that either didn't get written at all or got written
 266  * in the wrong place.
 267  */
 268 static int verify_parent_transid(struct extent_io_tree *io_tree,
 269                                  struct extent_buffer *eb, u64 parent_transid)
 270 {
 271         struct extent_state *cached_state = NULL;
 272         int ret;
 273
 274         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 275                 return 0;
 276
 277         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
 278                          0, &cached_state, GFP_NOFS);
 279         if (extent_buffer_uptodate(io_tree, eb, cached_state) &&
 280             btrfs_header_generation(eb) == parent_transid) {
 281                 ret = 0;
 282                 goto out;
 283         }
 284         if (printk_ratelimit()) {
 285                 printk("parent transid verify failed on %llu wanted %llu "
 286                        "found %llu\n",
 287                        (unsigned long long)eb->start,
 288                        (unsigned long long)parent_transid,
 289                        (unsigned long long)btrfs_header_generation(eb));
 290         }
 291         ret = 1;
 292         clear_extent_buffer_uptodate(io_tree, eb, &cached_state);
 293 out:
 294         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
 295                              &cached_state, GFP_NOFS);
 296         return ret;
 297 }
 298
 299 /*
 300  * helper to read a given tree block, doing retries as required when
 301  * the checksums don't match and we have alternate mirrors to try.
 302  */
 303 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 304                                           struct extent_buffer *eb,
 305                                           u64 start, u64 parent_transid)
 306 {
 307         struct extent_io_tree *io_tree;
 308         int ret;
 309         int num_copies = 0;
 310         int mirror_num = 0;
 311
 312         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 313         while (1) {
 314                 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
 315                                                btree_get_extent, mirror_num);
 316                 if (!ret &&
 317                     !verify_parent_transid(io_tree, eb, parent_transid))
 318                         return ret;
 319
 320                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
 321                                               eb->start, eb->len);
 322                 if (num_copies == 1)
 323                         return ret;
 324
 325                 mirror_num++;
 326                 if (mirror_num > num_copies)
 327                         return ret;
 328         }
 329         return -EIO;
 330 }
 331
 332 /*
 333  * checksum a dirty tree block before IO.  This has extra checks to make sure
 334  * we only fill in the checksum field in the first page of a multi-page block
 335  */
 336
 337 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 338 {
 339         struct extent_io_tree *tree;
 340         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
 341         u64 found_start;
 342         unsigned long len;
 343         struct extent_buffer *eb;
 344         int ret;
 345
 346         tree = &BTRFS_I(page->mapping->host)->io_tree;
 347
 348         if (page->private == EXTENT_PAGE_PRIVATE)
 349                 goto out;
 350         if (!page->private)
 351                 goto out;
 352         len = page->private >> 2;
 353         WARN_ON(len == 0);
 354
 355         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 356         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
 357                                              btrfs_header_generation(eb));
 358         BUG_ON(ret);
 359         WARN_ON(!btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN));
 360
 361         found_start = btrfs_header_bytenr(eb);
 362         if (found_start != start) {
 363                 WARN_ON(1);
 364                 goto err;
 365         }
 366         if (eb->first_page != page) {
 367                 WARN_ON(1);
 368                 goto err;
 369         }
 370         if (!PageUptodate(page)) {
 371                 WARN_ON(1);
 372                 goto err;
 373         }
 374         csum_tree_block(root, eb, 0);
 375 err:
 376         free_extent_buffer(eb);
 377 out:
 378         return 0;
 379 }
 380
 381 static int check_tree_block_fsid(struct btrfs_root *root,
 382                                  struct extent_buffer *eb)
 383 {
 384         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
 385         u8 fsid[BTRFS_UUID_SIZE];
 386         int ret = 1;
 387
 388         read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
 389                            BTRFS_FSID_SIZE);
 390         while (fs_devices) {
 391                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
 392                         ret = 0;
 393                         break;
 394                 }
 395                 fs_devices = fs_devices->seed;
 396         }
 397         return ret;
 398 }
 399
 400 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 401 void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level)
 402 {
 403         lockdep_set_class_and_name(&eb->lock,
 404                            &btrfs_eb_class[level],
 405                            btrfs_eb_name[level]);
 406 }
 407 #endif
 408
 409 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 410                                struct extent_state *state)
 411 {
 412         struct extent_io_tree *tree;
 413         u64 found_start;
 414         int found_level;
 415         unsigned long len;
 416         struct extent_buffer *eb;
 417         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 418         int ret = 0;
 419
 420         tree = &BTRFS_I(page->mapping->host)->io_tree;
 421         if (page->private == EXTENT_PAGE_PRIVATE)
 422                 goto out;
 423         if (!page->private)
 424                 goto out;
 425
 426         len = page->private >> 2;
 427         WARN_ON(len == 0);
 428
 429         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 430
 431         found_start = btrfs_header_bytenr(eb);
 432         if (found_start != start) {
 433                 if (printk_ratelimit()) {
 434                         printk(KERN_INFO "btrfs bad tree block start "
 435                                "%llu %llu\n",
 436                                (unsigned long long)found_start,
 437                                (unsigned long long)eb->start);
 438                 }
 439                 ret = -EIO;
 440                 goto err;
 441         }
 442         if (eb->first_page != page) {
 443                 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
 444                        eb->first_page->index, page->index);
 445                 WARN_ON(1);
 446                 ret = -EIO;
 447                 goto err;
 448         }
 449         if (check_tree_block_fsid(root, eb)) {
 450                 if (printk_ratelimit()) {
 451                         printk(KERN_INFO "btrfs bad fsid on block %llu\n",
 452                                (unsigned long long)eb->start);
 453                 }
 454                 ret = -EIO;
 455                 goto err;
 456         }
 457         found_level = btrfs_header_level(eb);
 458
 459         btrfs_set_buffer_lockdep_class(eb, found_level);
 460
 461         ret = csum_tree_block(root, eb, 1);
 462         if (ret)
 463                 ret = -EIO;
 464
 465         end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
 466         end = eb->start + end - 1;
 467 err:
 468         free_extent_buffer(eb);
 469 out:
 470         return ret;
 471 }
 472
 473 static void end_workqueue_bio(struct bio *bio, int err)
 474 {
 475         struct end_io_wq *end_io_wq = bio->bi_private;
 476         struct btrfs_fs_info *fs_info;
 477
 478         fs_info = end_io_wq->info;
 479         end_io_wq->error = err;
 480         end_io_wq->work.func = end_workqueue_fn;
 481         end_io_wq->work.flags = 0;
 482
 483         if (bio->bi_rw & REQ_WRITE) {
 484                 if (end_io_wq->metadata == 1)
 485                         btrfs_queue_worker(&fs_info->endio_meta_write_workers,
 486                                            &end_io_wq->work);
 487                 else if (end_io_wq->metadata == 2)
 488                         btrfs_queue_worker(&fs_info->endio_freespace_worker,
 489                                            &end_io_wq->work);
 490                 else
 491                         btrfs_queue_worker(&fs_info->endio_write_workers,
 492                                            &end_io_wq->work);
 493         } else {
 494                 if (end_io_wq->metadata)
 495                         btrfs_queue_worker(&fs_info->endio_meta_workers,
 496                                            &end_io_wq->work);
 497                 else
 498                         btrfs_queue_worker(&fs_info->endio_workers,
 499                                            &end_io_wq->work);
 500         }
 501 }
 502
 503 /*
 504  * For the metadata arg you want
 505  *
 506  * 0 - if data
 507  * 1 - if normal metadta
 508  * 2 - if writing to the free space cache area
 509  */
 510 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 511                         int metadata)
 512 {
 513         struct end_io_wq *end_io_wq;
 514         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
 515         if (!end_io_wq)
 516                 return -ENOMEM;
 517
 518         end_io_wq->private = bio->bi_private;
 519         end_io_wq->end_io = bio->bi_end_io;
 520         end_io_wq->info = info;
 521         end_io_wq->error = 0;
 522         end_io_wq->bio = bio;
 523         end_io_wq->metadata = metadata;
 524
 525         bio->bi_private = end_io_wq;
 526         bio->bi_end_io = end_workqueue_bio;
 527         return 0;
 528 }
 529
 530 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
 531 {
 532         unsigned long limit = min_t(unsigned long,
 533                                     info->workers.max_workers,
 534                                     info->fs_devices->open_devices);
 535         return 256 * limit;
 536 }
 537
 538 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
 539 {
 540         return atomic_read(&info->nr_async_bios) >
 541                 btrfs_async_submit_limit(info);
 542 }
 543
 544 static void run_one_async_start(struct btrfs_work *work)
 545 {
 546         struct async_submit_bio *async;
 547
 548         async = container_of(work, struct  async_submit_bio, work);
 549         async->submit_bio_start(async->inode, async->rw, async->bio,
 550                                async->mirror_num, async->bio_flags,
 551                                async->bio_offset);
 552 }
 553
 554 static void run_one_async_done(struct btrfs_work *work)
 555 {
 556         struct btrfs_fs_info *fs_info;
 557         struct async_submit_bio *async;
 558         int limit;
 559
 560         async = container_of(work, struct  async_submit_bio, work);
 561         fs_info = BTRFS_I(async->inode)->root->fs_info;
 562
 563         limit = btrfs_async_submit_limit(fs_info);
 564         limit = limit * 2 / 3;
 565
 566         atomic_dec(&fs_info->nr_async_submits);
 567
 568         if (atomic_read(&fs_info->nr_async_submits) < limit &&
 569             waitqueue_active(&fs_info->async_submit_wait))
 570                 wake_up(&fs_info->async_submit_wait);
 571
 572         async->submit_bio_done(async->inode, async->rw, async->bio,
 573                                async->mirror_num, async->bio_flags,
 574                                async->bio_offset);
 575 }
 576
 577 static void run_one_async_free(struct btrfs_work *work)
 578 {
 579         struct async_submit_bio *async;
 580
 581         async = container_of(work, struct  async_submit_bio, work);
 582         kfree(async);
 583 }
 584
 585 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 586                         int rw, struct bio *bio, int mirror_num,
 587                         unsigned long bio_flags,
 588                         u64 bio_offset,
 589                         extent_submit_bio_hook_t *submit_bio_start,
 590                         extent_submit_bio_hook_t *submit_bio_done)
 591 {
 592         struct async_submit_bio *async;
 593
 594         async = kmalloc(sizeof(*async), GFP_NOFS);
 595         if (!async)
 596                 return -ENOMEM;
 597
 598         async->inode = inode;
 599         async->rw = rw;
 600         async->bio = bio;
 601         async->mirror_num = mirror_num;
 602         async->submit_bio_start = submit_bio_start;
 603         async->submit_bio_done = submit_bio_done;
 604
 605         async->work.func = run_one_async_start;
 606         async->work.ordered_func = run_one_async_done;
 607         async->work.ordered_free = run_one_async_free;
 608
 609         async->work.flags = 0;
 610         async->bio_flags = bio_flags;
 611         async->bio_offset = bio_offset;
 612
 613         atomic_inc(&fs_info->nr_async_submits);
 614
 615         if (rw & REQ_SYNC)
 616                 btrfs_set_work_high_prio(&async->work);
 617
 618         btrfs_queue_worker(&fs_info->workers, &async->work);
 619
 620         while (atomic_read(&fs_info->async_submit_draining) &&
 621               atomic_read(&fs_info->nr_async_submits)) {
 622                 wait_event(fs_info->async_submit_wait,
 623                            (atomic_read(&fs_info->nr_async_submits) == 0));
 624         }
 625
 626         return 0;
 627 }
 628
 629 static int btree_csum_one_bio(struct bio *bio)
 630 {
 631         struct bio_vec *bvec = bio->bi_io_vec;
 632         int bio_index = 0;
 633         struct btrfs_root *root;
 634
 635         WARN_ON(bio->bi_vcnt <= 0);
 636         while (bio_index < bio->bi_vcnt) {
 637                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 638                 csum_dirty_buffer(root, bvec->bv_page);
 639                 bio_index++;
 640                 bvec++;
 641         }
 642         return 0;
 643 }
 644
 645 static int __btree_submit_bio_start(struct inode *inode, int rw,
 646                                     struct bio *bio, int mirror_num,
 647                                     unsigned long bio_flags,
 648                                     u64 bio_offset)
 649 {
 650         /*
 651          * when we're called for a write, we're already in the async
 652          * submission context.  Just jump into btrfs_map_bio
 653          */
 654         btree_csum_one_bio(bio);
 655         return 0;
 656 }
 657
 658 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
 659                                  int mirror_num, unsigned long bio_flags,
 660                                  u64 bio_offset)
 661 {
 662         /*
 663          * when we're called for a write, we're already in the async
 664          * submission context.  Just jump into btrfs_map_bio
 665          */
 666         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
 667 }
 668
 669 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 670                                  int mirror_num, unsigned long bio_flags,
 671                                  u64 bio_offset)
 672 {
 673         int ret;
 674
 675         ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
 676                                           bio, 1);
 677         BUG_ON(ret);
 678
 679         if (!(rw & REQ_WRITE)) {
 680                 /*
 681                  * called for a read, do the setup so that checksum validation
 682                  * can happen in the async kernel threads
 683                  */
 684                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 685                                      mirror_num, 0);
 686         }
 687
 688         /*
 689          * kthread helpers are used to submit writes so that checksumming
 690          * can happen in parallel across all CPUs
 691          */
 692         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 693                                    inode, rw, bio, mirror_num, 0,
 694                                    bio_offset,
 695                                    __btree_submit_bio_start,
 696                                    __btree_submit_bio_done);
 697 }
 698
 699 static int btree_migratepage(struct address_space *mapping,
 700                         struct page *newpage, struct page *page)
 701 {
 702         /*
 703          * we can't safely write a btree page from here,
 704          * we haven't done the locking hook
 705          */
 706         if (PageDirty(page))
 707                 return -EAGAIN;
 708         /*
 709          * Buffers may be managed in a filesystem specific way.
 710          * We must have no buffers or drop them.
 711          */
 712         if (page_has_private(page) &&
 713             !try_to_release_page(page, GFP_KERNEL))
 714                 return -EAGAIN;
 715
 716         return migrate_page(mapping, newpage, page);
 717 }
 718
 719 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 720 {
 721         struct extent_io_tree *tree;
 722         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 723         struct extent_buffer *eb;
 724         int was_dirty;
 725
 726         tree = &BTRFS_I(page->mapping->host)->io_tree;
 727         if (!(current->flags & PF_MEMALLOC)) {
 728                 return extent_write_full_page(tree, page,
 729                                               btree_get_extent, wbc);
 730         }
 731
 732         redirty_page_for_writepage(wbc, page);
 733         eb = btrfs_find_tree_block(root, page_offset(page), PAGE_CACHE_SIZE);
 734         WARN_ON(!eb);
 735
 736         was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
 737         if (!was_dirty) {
 738                 spin_lock(&root->fs_info->delalloc_lock);
 739                 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
 740                 spin_unlock(&root->fs_info->delalloc_lock);
 741         }
 742         free_extent_buffer(eb);
 743
 744         unlock_page(page);
 745         return 0;
 746 }
 747
 748 static int btree_writepages(struct address_space *mapping,
 749                             struct writeback_control *wbc)
 750 {
 751         struct extent_io_tree *tree;
 752         tree = &BTRFS_I(mapping->host)->io_tree;
 753         if (wbc->sync_mode == WB_SYNC_NONE) {
 754                 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
 755                 u64 num_dirty;
 756                 unsigned long thresh = 32 * 1024 * 1024;
 757
 758                 if (wbc->for_kupdate)
 759                         return 0;
 760
 761                 /* this is a bit racy, but that's ok */
 762                 num_dirty = root->fs_info->dirty_metadata_bytes;
 763                 if (num_dirty < thresh)
 764                         return 0;
 765         }
 766         return extent_writepages(tree, mapping, btree_get_extent, wbc);
 767 }
 768
 769 static int btree_readpage(struct file *file, struct page *page)
 770 {
 771         struct extent_io_tree *tree;
 772         tree = &BTRFS_I(page->mapping->host)->io_tree;
 773         return extent_read_full_page(tree, page, btree_get_extent);
 774 }
 775
 776 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 777 {
 778         struct extent_io_tree *tree;
 779         struct extent_map_tree *map;
 780         int ret;
 781
 782         if (PageWriteback(page) || PageDirty(page))
 783                 return 0;
 784
 785         tree = &BTRFS_I(page->mapping->host)->io_tree;
 786         map = &BTRFS_I(page->mapping->host)->extent_tree;
 787
 788         ret = try_release_extent_state(map, tree, page, gfp_flags);
 789         if (!ret)
 790                 return 0;
 791
 792         ret = try_release_extent_buffer(tree, page);
 793         if (ret == 1) {
 794                 ClearPagePrivate(page);
 795                 set_page_private(page, 0);
 796                 page_cache_release(page);
 797         }
 798
 799         return ret;
 800 }
 801
 802 static void btree_invalidatepage(struct page *page, unsigned long offset)
 803 {
 804         struct extent_io_tree *tree;
 805         tree = &BTRFS_I(page->mapping->host)->io_tree;
 806         extent_invalidatepage(tree, page, offset);
 807         btree_releasepage(page, GFP_NOFS);
 808         if (PagePrivate(page)) {
 809                 printk(KERN_WARNING "btrfs warning page private not zero "
 810                        "on page %llu\n", (unsigned long long)page_offset(page));
 811                 ClearPagePrivate(page);
 812                 set_page_private(page, 0);
 813                 page_cache_release(page);
 814         }
 815 }
 816
 817 static const struct address_space_operations btree_aops = {
 818         .readpage       = btree_readpage,
 819         .writepage      = btree_writepage,
 820         .writepages     = btree_writepages,
 821         .releasepage    = btree_releasepage,
 822         .invalidatepage = btree_invalidatepage,
 823         .sync_page      = block_sync_page,
 824         .migratepage    = btree_migratepage,
 825 };
 826
 827 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
 828                          u64 parent_transid)
 829 {
 830         struct extent_buffer *buf = NULL;
 831         struct inode *btree_inode = root->fs_info->btree_inode;
 832         int ret = 0;
 833
 834         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 835         if (!buf)
 836                 return 0;
 837         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
 838                                  buf, 0, 0, btree_get_extent, 0);
 839         free_extent_buffer(buf);
 840         return ret;
 841 }
 842
 843 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
 844                                             u64 bytenr, u32 blocksize)
 845 {
 846         struct inode *btree_inode = root->fs_info->btree_inode;
 847         struct extent_buffer *eb;
 848         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 849                                 bytenr, blocksize, GFP_NOFS);
 850         return eb;
 851 }
 852
 853 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
 854                                                  u64 bytenr, u32 blocksize)
 855 {
 856         struct inode *btree_inode = root->fs_info->btree_inode;
 857         struct extent_buffer *eb;
 858
 859         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 860                                  bytenr, blocksize, NULL, GFP_NOFS);
 861         return eb;
 862 }
 863
 864
 865 int btrfs_write_tree_block(struct extent_buffer *buf)
 866 {
 867         return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
 868                                         buf->start + buf->len - 1);
 869 }
 870
 871 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
 872 {
 873         return filemap_fdatawait_range(buf->first_page->mapping,
 874                                        buf->start, buf->start + buf->len - 1);
 875 }
 876
 877 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
 878                                       u32 blocksize, u64 parent_transid)
 879 {
 880         struct extent_buffer *buf = NULL;
 881         int ret;
 882
 883         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 884         if (!buf)
 885                 return NULL;
 886
 887         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
 888
 889         if (ret == 0)
 890                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
 891         return buf;
 892
 893 }
 894
 895 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 896                      struct extent_buffer *buf)
 897 {
 898         struct inode *btree_inode = root->fs_info->btree_inode;
 899         if (btrfs_header_generation(buf) ==
 900             root->fs_info->running_transaction->transid) {
 901                 btrfs_assert_tree_locked(buf);
 902
 903                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
 904                         spin_lock(&root->fs_info->delalloc_lock);
 905                         if (root->fs_info->dirty_metadata_bytes >= buf->len)
 906                                 root->fs_info->dirty_metadata_bytes -= buf->len;
 907                         else
 908                                 WARN_ON(1);
 909                         spin_unlock(&root->fs_info->delalloc_lock);
 910                 }
 911
 912                 /* ugh, clear_extent_buffer_dirty needs to lock the page */
 913                 btrfs_set_lock_blocking(buf);
 914                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
 915                                           buf);
 916         }
 917         return 0;
 918 }
 919
 920 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
 921                         u32 stripesize, struct btrfs_root *root,
 922                         struct btrfs_fs_info *fs_info,
 923                         u64 objectid)
 924 {
 925         root->node = NULL;
 926         root->commit_root = NULL;
 927         root->sectorsize = sectorsize;
 928         root->nodesize = nodesize;
 929         root->leafsize = leafsize;
 930         root->stripesize = stripesize;
 931         root->ref_cows = 0;
 932         root->track_dirty = 0;
 933         root->in_radix = 0;
 934         root->orphan_item_inserted = 0;
 935         root->orphan_cleanup_state = 0;
 936
 937         root->fs_info = fs_info;
 938         root->objectid = objectid;
 939         root->last_trans = 0;
 940         root->highest_objectid = 0;
 941         root->name = NULL;
 942         root->in_sysfs = 0;
 943         root->inode_tree = RB_ROOT;
 944         root->block_rsv = NULL;
 945         root->orphan_block_rsv = NULL;
 946
 947         INIT_LIST_HEAD(&root->dirty_list);
 948         INIT_LIST_HEAD(&root->orphan_list);
 949         INIT_LIST_HEAD(&root->root_list);
 950         spin_lock_init(&root->node_lock);
 951         spin_lock_init(&root->orphan_lock);
 952         spin_lock_init(&root->inode_lock);
 953         spin_lock_init(&root->accounting_lock);
 954         mutex_init(&root->objectid_mutex);
 955         mutex_init(&root->log_mutex);
 956         init_waitqueue_head(&root->log_writer_wait);
 957         init_waitqueue_head(&root->log_commit_wait[0]);
 958         init_waitqueue_head(&root->log_commit_wait[1]);
 959         atomic_set(&root->log_commit[0], 0);
 960         atomic_set(&root->log_commit[1], 0);
 961         atomic_set(&root->log_writers, 0);
 962         root->log_batch = 0;
 963         root->log_transid = 0;
 964         root->last_log_commit = 0;
 965         extent_io_tree_init(&root->dirty_log_pages,
 966                              fs_info->btree_inode->i_mapping, GFP_NOFS);
 967
 968         memset(&root->root_key, 0, sizeof(root->root_key));
 969         memset(&root->root_item, 0, sizeof(root->root_item));
 970         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
 971         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
 972         root->defrag_trans_start = fs_info->generation;
 973         init_completion(&root->kobj_unregister);
 974         root->defrag_running = 0;
 975         root->root_key.objectid = objectid;
 976         root->anon_super.s_root = NULL;
 977         root->anon_super.s_dev = 0;
 978         INIT_LIST_HEAD(&root->anon_super.s_list);
 979         INIT_LIST_HEAD(&root->anon_super.s_instances);
 980         init_rwsem(&root->anon_super.s_umount);
 981
 982         return 0;
 983 }
 984
 985 static int find_and_setup_root(struct btrfs_root *tree_root,
 986                                struct btrfs_fs_info *fs_info,
 987                                u64 objectid,
 988                                struct btrfs_root *root)
 989 {
 990         int ret;
 991         u32 blocksize;
 992         u64 generation;
 993
 994         __setup_root(tree_root->nodesize, tree_root->leafsize,
 995                      tree_root->sectorsize, tree_root->stripesize,
 996                      root, fs_info, objectid);
 997         ret = btrfs_find_last_root(tree_root, objectid,
 998                                    &root->root_item, &root->root_key);
 999         if (ret > 0)
1000                 return -ENOENT;
1001         BUG_ON(ret);
1002
1003         generation = btrfs_root_generation(&root->root_item);
1004         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1005         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1006                                      blocksize, generation);
1007         BUG_ON(!root->node);
1008         root->commit_root = btrfs_root_node(root);
1009         return 0;
1010 }
1011
1012 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1013                                          struct btrfs_fs_info *fs_info)
1014 {
1015         struct btrfs_root *root;
1016         struct btrfs_root *tree_root = fs_info->tree_root;
1017         struct extent_buffer *leaf;
1018
1019         root = kzalloc(sizeof(*root), GFP_NOFS);
1020         if (!root)
1021                 return ERR_PTR(-ENOMEM);
1022
1023         __setup_root(tree_root->nodesize, tree_root->leafsize,
1024                      tree_root->sectorsize, tree_root->stripesize,
1025                      root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1026
1027         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1028         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1029         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1030         /*
1031          * log trees do not get reference counted because they go away
1032          * before a real commit is actually done.  They do store pointers
1033          * to file data extents, and those reference counts still get
1034          * updated (along with back refs to the log tree).
1035          */
1036         root->ref_cows = 0;
1037
1038         leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1039                                       BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
1040         if (IS_ERR(leaf)) {
1041                 kfree(root);
1042                 return ERR_CAST(leaf);
1043         }
1044
1045         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1046         btrfs_set_header_bytenr(leaf, leaf->start);
1047         btrfs_set_header_generation(leaf, trans->transid);
1048         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1049         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1050         root->node = leaf;
1051
1052         write_extent_buffer(root->node, root->fs_info->fsid,
1053                             (unsigned long)btrfs_header_fsid(root->node),
1054                             BTRFS_FSID_SIZE);
1055         btrfs_mark_buffer_dirty(root->node);
1056         btrfs_tree_unlock(root->node);
1057         return root;
1058 }
1059
1060 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1061                              struct btrfs_fs_info *fs_info)
1062 {
1063         struct btrfs_root *log_root;
1064
1065         log_root = alloc_log_tree(trans, fs_info);
1066         if (IS_ERR(log_root))
1067                 return PTR_ERR(log_root);
1068         WARN_ON(fs_info->log_root_tree);
1069         fs_info->log_root_tree = log_root;
1070         return 0;
1071 }
1072
1073 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1074                        struct btrfs_root *root)
1075 {
1076         struct btrfs_root *log_root;
1077         struct btrfs_inode_item *inode_item;
1078
1079         log_root = alloc_log_tree(trans, root->fs_info);
1080         if (IS_ERR(log_root))
1081                 return PTR_ERR(log_root);
1082
1083         log_root->last_trans = trans->transid;
1084         log_root->root_key.offset = root->root_key.objectid;
1085
1086         inode_item = &log_root->root_item.inode;
1087         inode_item->generation = cpu_to_le64(1);
1088         inode_item->size = cpu_to_le64(3);
1089         inode_item->nlink = cpu_to_le32(1);
1090         inode_item->nbytes = cpu_to_le64(root->leafsize);
1091         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1092
1093         btrfs_set_root_node(&log_root->root_item, log_root->node);
1094
1095         WARN_ON(root->log_root);
1096         root->log_root = log_root;
1097         root->log_transid = 0;
1098         root->last_log_commit = 0;
1099         return 0;
1100 }
1101
1102 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1103                                                struct btrfs_key *location)
1104 {
1105         struct btrfs_root *root;
1106         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1107         struct btrfs_path *path;
1108         struct extent_buffer *l;
1109         u64 generation;
1110         u32 blocksize;
1111         int ret = 0;
1112
1113         root = kzalloc(sizeof(*root), GFP_NOFS);
1114         if (!root)
1115                 return ERR_PTR(-ENOMEM);
1116         if (location->offset == (u64)-1) {
1117                 ret = find_and_setup_root(tree_root, fs_info,
1118                                           location->objectid, root);
1119                 if (ret) {
1120                         kfree(root);
1121                         return ERR_PTR(ret);
1122                 }
1123                 goto out;
1124         }
1125
1126         __setup_root(tree_root->nodesize, tree_root->leafsize,
1127                      tree_root->sectorsize, tree_root->stripesize,
1128                      root, fs_info, location->objectid);
1129
1130         path = btrfs_alloc_path();
1131         BUG_ON(!path);
1132         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1133         if (ret == 0) {
1134                 l = path->nodes[0];
1135                 read_extent_buffer(l, &root->root_item,
1136                                 btrfs_item_ptr_offset(l, path->slots[0]),
1137                                 sizeof(root->root_item));
1138                 memcpy(&root->root_key, location, sizeof(*location));
1139         }
1140         btrfs_free_path(path);
1141         if (ret) {
1142                 if (ret > 0)
1143                         ret = -ENOENT;
1144                 return ERR_PTR(ret);
1145         }
1146
1147         generation = btrfs_root_generation(&root->root_item);
1148         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1149         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1150                                      blocksize, generation);
1151         root->commit_root = btrfs_root_node(root);
1152         BUG_ON(!root->node);
1153 out:
1154         if (location->objectid != BTRFS_TREE_LOG_OBJECTID)
1155                 root->ref_cows = 1;
1156
1157         return root;
1158 }
1159
1160 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1161                                         u64 root_objectid)
1162 {
1163         struct btrfs_root *root;
1164
1165         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1166                 return fs_info->tree_root;
1167         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1168                 return fs_info->extent_root;
1169
1170         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1171                                  (unsigned long)root_objectid);
1172         return root;
1173 }
1174
1175 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1176                                               struct btrfs_key *location)
1177 {
1178         struct btrfs_root *root;
1179         int ret;
1180
1181         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1182                 return fs_info->tree_root;
1183         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1184                 return fs_info->extent_root;
1185         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1186                 return fs_info->chunk_root;
1187         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1188                 return fs_info->dev_root;
1189         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1190                 return fs_info->csum_root;
1191 again:
1192         spin_lock(&fs_info->fs_roots_radix_lock);
1193         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1194                                  (unsigned long)location->objectid);
1195         spin_unlock(&fs_info->fs_roots_radix_lock);
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         if (btrfs_root_refs(&root->root_item) == 0) {
1206                 ret = -ENOENT;
1207                 goto fail;
1208         }
1209
1210         ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1211         if (ret < 0)
1212                 goto fail;
1213         if (ret == 0)
1214                 root->orphan_item_inserted = 1;
1215
1216         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1217         if (ret)
1218                 goto fail;
1219
1220         spin_lock(&fs_info->fs_roots_radix_lock);
1221         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1222                                 (unsigned long)root->root_key.objectid,
1223                                 root);
1224         if (ret == 0)
1225                 root->in_radix = 1;
1226
1227         spin_unlock(&fs_info->fs_roots_radix_lock);
1228         radix_tree_preload_end();
1229         if (ret) {
1230                 if (ret == -EEXIST) {
1231                         free_fs_root(root);
1232                         goto again;
1233                 }
1234                 goto fail;
1235         }
1236
1237         ret = btrfs_find_dead_roots(fs_info->tree_root,
1238                                     root->root_key.objectid);
1239         WARN_ON(ret);
1240         return root;
1241 fail:
1242         free_fs_root(root);
1243         return ERR_PTR(ret);
1244 }
1245
1246 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1247                                       struct btrfs_key *location,
1248                                       const char *name, int namelen)
1249 {
1250         return btrfs_read_fs_root_no_name(fs_info, location);
1251 #if 0
1252         struct btrfs_root *root;
1253         int ret;
1254
1255         root = btrfs_read_fs_root_no_name(fs_info, location);
1256         if (!root)
1257                 return NULL;
1258
1259         if (root->in_sysfs)
1260                 return root;
1261
1262         ret = btrfs_set_root_name(root, name, namelen);
1263         if (ret) {
1264                 free_extent_buffer(root->node);
1265                 kfree(root);
1266                 return ERR_PTR(ret);
1267         }
1268
1269         ret = btrfs_sysfs_add_root(root);
1270         if (ret) {
1271                 free_extent_buffer(root->node);
1272                 kfree(root->name);
1273                 kfree(root);
1274                 return ERR_PTR(ret);
1275         }
1276         root->in_sysfs = 1;
1277         return root;
1278 #endif
1279 }
1280
1281 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1282 {
1283         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1284         int ret = 0;
1285         struct btrfs_device *device;
1286         struct backing_dev_info *bdi;
1287
1288         list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1289                 if (!device->bdev)
1290                         continue;
1291                 bdi = blk_get_backing_dev_info(device->bdev);
1292                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1293                         ret = 1;
1294                         break;
1295                 }
1296         }
1297         return ret;
1298 }
1299
1300 /*
1301  * this unplugs every device on the box, and it is only used when page
1302  * is null
1303  */
1304 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1305 {
1306         struct btrfs_device *device;
1307         struct btrfs_fs_info *info;
1308
1309         info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1310         list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1311                 if (!device->bdev)
1312                         continue;
1313
1314                 bdi = blk_get_backing_dev_info(device->bdev);
1315                 if (bdi->unplug_io_fn)
1316                         bdi->unplug_io_fn(bdi, page);
1317         }
1318 }
1319
1320 static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1321 {
1322         struct inode *inode;
1323         struct extent_map_tree *em_tree;
1324         struct extent_map *em;
1325         struct address_space *mapping;
1326         u64 offset;
1327
1328         /* the generic O_DIRECT read code does this */
1329         if (1 || !page) {
1330                 __unplug_io_fn(bdi, page);
1331                 return;
1332         }
1333
1334         /*
1335          * page->mapping may change at any time.  Get a consistent copy
1336          * and use that for everything below
1337          */
1338         smp_mb();
1339         mapping = page->mapping;
1340         if (!mapping)
1341                 return;
1342
1343         inode = mapping->host;
1344
1345         /*
1346          * don't do the expensive searching for a small number of
1347          * devices
1348          */
1349         if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1350                 __unplug_io_fn(bdi, page);
1351                 return;
1352         }
1353
1354         offset = page_offset(page);
1355
1356         em_tree = &BTRFS_I(inode)->extent_tree;
1357         read_lock(&em_tree->lock);
1358         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1359         read_unlock(&em_tree->lock);
1360         if (!em) {
1361                 __unplug_io_fn(bdi, page);
1362                 return;
1363         }
1364
1365         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1366                 free_extent_map(em);
1367                 __unplug_io_fn(bdi, page);
1368                 return;
1369         }
1370         offset = offset - em->start;
1371         btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1372                           em->block_start + offset, page);
1373         free_extent_map(em);
1374 }
1375
1376 /*
1377  * If this fails, caller must call bdi_destroy() to get rid of the
1378  * bdi again.
1379  */
1380 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1381 {
1382         int err;
1383
1384         bdi->capabilities = BDI_CAP_MAP_COPY;
1385         err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1386         if (err)
1387                 return err;
1388
1389         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1390         bdi->unplug_io_fn       = btrfs_unplug_io_fn;
1391         bdi->unplug_io_data     = info;
1392         bdi->congested_fn       = btrfs_congested_fn;
1393         bdi->congested_data     = info;
1394         return 0;
1395 }
1396
1397 static int bio_ready_for_csum(struct bio *bio)
1398 {
1399         u64 length = 0;
1400         u64 buf_len = 0;
1401         u64 start = 0;
1402         struct page *page;
1403         struct extent_io_tree *io_tree = NULL;
1404         struct bio_vec *bvec;
1405         int i;
1406         int ret;
1407
1408         bio_for_each_segment(bvec, bio, i) {
1409                 page = bvec->bv_page;
1410                 if (page->private == EXTENT_PAGE_PRIVATE) {
1411                         length += bvec->bv_len;
1412                         continue;
1413                 }
1414                 if (!page->private) {
1415                         length += bvec->bv_len;
1416                         continue;
1417                 }
1418                 length = bvec->bv_len;
1419                 buf_len = page->private >> 2;
1420                 start = page_offset(page) + bvec->bv_offset;
1421                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1422         }
1423         /* are we fully contained in this bio? */
1424         if (buf_len <= length)
1425                 return 1;
1426
1427         ret = extent_range_uptodate(io_tree, start + length,
1428                                     start + buf_len - 1);
1429         return ret;
1430 }
1431
1432 /*
1433  * called by the kthread helper functions to finally call the bio end_io
1434  * functions.  This is where read checksum verification actually happens
1435  */
1436 static void end_workqueue_fn(struct btrfs_work *work)
1437 {
1438         struct bio *bio;
1439         struct end_io_wq *end_io_wq;
1440         struct btrfs_fs_info *fs_info;
1441         int error;
1442
1443         end_io_wq = container_of(work, struct end_io_wq, work);
1444         bio = end_io_wq->bio;
1445         fs_info = end_io_wq->info;
1446
1447         /* metadata bio reads are special because the whole tree block must
1448          * be checksummed at once.  This makes sure the entire block is in
1449          * ram and up to date before trying to verify things.  For
1450          * blocksize <= pagesize, it is basically a noop
1451          */
1452         if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata &&
1453             !bio_ready_for_csum(bio)) {
1454                 btrfs_queue_worker(&fs_info->endio_meta_workers,
1455                                    &end_io_wq->work);
1456                 return;
1457         }
1458         error = end_io_wq->error;
1459         bio->bi_private = end_io_wq->private;
1460         bio->bi_end_io = end_io_wq->end_io;
1461         kfree(end_io_wq);
1462         bio_endio(bio, error);
1463 }
1464
1465 static int cleaner_kthread(void *arg)
1466 {
1467         struct btrfs_root *root = arg;
1468
1469         do {
1470                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1471
1472                 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1473                     mutex_trylock(&root->fs_info->cleaner_mutex)) {
1474                         btrfs_run_delayed_iputs(root);
1475                         btrfs_clean_old_snapshots(root);
1476                         mutex_unlock(&root->fs_info->cleaner_mutex);
1477                 }
1478
1479                 if (freezing(current)) {
1480                         refrigerator();
1481                 } else {
1482                         set_current_state(TASK_INTERRUPTIBLE);
1483                         if (!kthread_should_stop())
1484                                 schedule();
1485                         __set_current_state(TASK_RUNNING);
1486                 }
1487         } while (!kthread_should_stop());
1488         return 0;
1489 }
1490
1491 static int transaction_kthread(void *arg)
1492 {
1493         struct btrfs_root *root = arg;
1494         struct btrfs_trans_handle *trans;
1495         struct btrfs_transaction *cur;
1496         u64 transid;
1497         unsigned long now;
1498         unsigned long delay;
1499         int ret;
1500
1501         do {
1502                 delay = HZ * 30;
1503                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1504                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1505
1506                 spin_lock(&root->fs_info->new_trans_lock);
1507                 cur = root->fs_info->running_transaction;
1508                 if (!cur) {
1509                         spin_unlock(&root->fs_info->new_trans_lock);
1510                         goto sleep;
1511                 }
1512
1513                 now = get_seconds();
1514                 if (!cur->blocked &&
1515                     (now < cur->start_time || now - cur->start_time < 30)) {
1516                         spin_unlock(&root->fs_info->new_trans_lock);
1517                         delay = HZ * 5;
1518                         goto sleep;
1519                 }
1520                 transid = cur->transid;
1521                 spin_unlock(&root->fs_info->new_trans_lock);
1522
1523                 trans = btrfs_join_transaction(root, 1);
1524                 if (transid == trans->transid) {
1525                         ret = btrfs_commit_transaction(trans, root);
1526                         BUG_ON(ret);
1527                 } else {
1528                         btrfs_end_transaction(trans, root);
1529                 }
1530 sleep:
1531                 wake_up_process(root->fs_info->cleaner_kthread);
1532                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1533
1534                 if (freezing(current)) {
1535                         refrigerator();
1536                 } else {
1537                         set_current_state(TASK_INTERRUPTIBLE);
1538                         if (!kthread_should_stop() &&
1539                             !btrfs_transaction_blocked(root->fs_info))
1540                                 schedule_timeout(delay);
1541                         __set_current_state(TASK_RUNNING);
1542                 }
1543         } while (!kthread_should_stop());
1544         return 0;
1545 }
1546
1547 struct btrfs_root *open_ctree(struct super_block *sb,
1548                               struct btrfs_fs_devices *fs_devices,
1549                               char *options)
1550 {
1551         u32 sectorsize;
1552         u32 nodesize;
1553         u32 leafsize;
1554         u32 blocksize;
1555         u32 stripesize;
1556         u64 generation;
1557         u64 features;
1558         struct btrfs_key location;
1559         struct buffer_head *bh;
1560         struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1561                                                  GFP_NOFS);
1562         struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1563                                                  GFP_NOFS);
1564         struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
1565                                                GFP_NOFS);
1566         struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1567                                                 GFP_NOFS);
1568         struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1569                                                 GFP_NOFS);
1570         struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1571                                               GFP_NOFS);
1572         struct btrfs_root *log_tree_root;
1573
1574         int ret;
1575         int err = -EINVAL;
1576
1577         struct btrfs_super_block *disk_super;
1578
1579         if (!extent_root || !tree_root || !fs_info ||
1580             !chunk_root || !dev_root || !csum_root) {
1581                 err = -ENOMEM;
1582                 goto fail;
1583         }
1584
1585         ret = init_srcu_struct(&fs_info->subvol_srcu);
1586         if (ret) {
1587                 err = ret;
1588                 goto fail;
1589         }
1590
1591         ret = setup_bdi(fs_info, &fs_info->bdi);
1592         if (ret) {
1593                 err = ret;
1594                 goto fail_srcu;
1595         }
1596
1597         fs_info->btree_inode = new_inode(sb);
1598         if (!fs_info->btree_inode) {
1599                 err = -ENOMEM;
1600                 goto fail_bdi;
1601         }
1602
1603         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1604         INIT_LIST_HEAD(&fs_info->trans_list);
1605         INIT_LIST_HEAD(&fs_info->dead_roots);
1606         INIT_LIST_HEAD(&fs_info->delayed_iputs);
1607         INIT_LIST_HEAD(&fs_info->hashers);
1608         INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1609         INIT_LIST_HEAD(&fs_info->ordered_operations);
1610         INIT_LIST_HEAD(&fs_info->caching_block_groups);
1611         spin_lock_init(&fs_info->delalloc_lock);
1612         spin_lock_init(&fs_info->new_trans_lock);
1613         spin_lock_init(&fs_info->ref_cache_lock);
1614         spin_lock_init(&fs_info->fs_roots_radix_lock);
1615         spin_lock_init(&fs_info->delayed_iput_lock);
1616
1617         init_completion(&fs_info->kobj_unregister);
1618         fs_info->tree_root = tree_root;
1619         fs_info->extent_root = extent_root;
1620         fs_info->csum_root = csum_root;
1621         fs_info->chunk_root = chunk_root;
1622         fs_info->dev_root = dev_root;
1623         fs_info->fs_devices = fs_devices;
1624         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1625         INIT_LIST_HEAD(&fs_info->space_info);
1626         btrfs_mapping_init(&fs_info->mapping_tree);
1627         btrfs_init_block_rsv(&fs_info->global_block_rsv);
1628         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1629         btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1630         btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1631         btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1632         INIT_LIST_HEAD(&fs_info->durable_block_rsv_list);
1633         mutex_init(&fs_info->durable_block_rsv_mutex);
1634         atomic_set(&fs_info->nr_async_submits, 0);
1635         atomic_set(&fs_info->async_delalloc_pages, 0);
1636         atomic_set(&fs_info->async_submit_draining, 0);
1637         atomic_set(&fs_info->nr_async_bios, 0);
1638         fs_info->sb = sb;
1639         fs_info->max_inline = 8192 * 1024;
1640         fs_info->metadata_ratio = 0;
1641
1642         fs_info->thread_pool_size = min_t(unsigned long,
1643                                           num_online_cpus() + 2, 8);
1644
1645         INIT_LIST_HEAD(&fs_info->ordered_extents);
1646         spin_lock_init(&fs_info->ordered_extent_lock);
1647
1648         sb->s_blocksize = 4096;
1649         sb->s_blocksize_bits = blksize_bits(4096);
1650         sb->s_bdi = &fs_info->bdi;
1651
1652         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1653         fs_info->btree_inode->i_nlink = 1;
1654         /*
1655          * we set the i_size on the btree inode to the max possible int.
1656          * the real end of the address space is determined by all of
1657          * the devices in the system
1658          */
1659         fs_info->btree_inode->i_size = OFFSET_MAX;
1660         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1661         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1662
1663         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
1664         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1665                              fs_info->btree_inode->i_mapping,
1666                              GFP_NOFS);
1667         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1668                              GFP_NOFS);
1669
1670         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1671
1672         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1673         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1674                sizeof(struct btrfs_key));
1675         BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
1676         insert_inode_hash(fs_info->btree_inode);
1677
1678         spin_lock_init(&fs_info->block_group_cache_lock);
1679         fs_info->block_group_cache_tree = RB_ROOT;
1680
1681         extent_io_tree_init(&fs_info->freed_extents[0],
1682                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1683         extent_io_tree_init(&fs_info->freed_extents[1],
1684                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1685         fs_info->pinned_extents = &fs_info->freed_extents[0];
1686         fs_info->do_barriers = 1;
1687
1688
1689         mutex_init(&fs_info->trans_mutex);
1690         mutex_init(&fs_info->ordered_operations_mutex);
1691         mutex_init(&fs_info->tree_log_mutex);
1692         mutex_init(&fs_info->chunk_mutex);
1693         mutex_init(&fs_info->transaction_kthread_mutex);
1694         mutex_init(&fs_info->cleaner_mutex);
1695         mutex_init(&fs_info->volume_mutex);
1696         init_rwsem(&fs_info->extent_commit_sem);
1697         init_rwsem(&fs_info->cleanup_work_sem);
1698         init_rwsem(&fs_info->subvol_sem);
1699
1700         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1701         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1702
1703         init_waitqueue_head(&fs_info->transaction_throttle);
1704         init_waitqueue_head(&fs_info->transaction_wait);
1705         init_waitqueue_head(&fs_info->transaction_blocked_wait);
1706         init_waitqueue_head(&fs_info->async_submit_wait);
1707
1708         __setup_root(4096, 4096, 4096, 4096, tree_root,
1709                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1710
1711         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1712         if (!bh)
1713                 goto fail_iput;
1714
1715         memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1716         memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1717                sizeof(fs_info->super_for_commit));
1718         brelse(bh);
1719
1720         memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1721
1722         disk_super = &fs_info->super_copy;
1723         if (!btrfs_super_root(disk_super))
1724                 goto fail_iput;
1725
1726         ret = btrfs_parse_options(tree_root, options);
1727         if (ret) {
1728                 err = ret;
1729                 goto fail_iput;
1730         }
1731
1732         features = btrfs_super_incompat_flags(disk_super) &
1733                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1734         if (features) {
1735                 printk(KERN_ERR "BTRFS: couldn't mount because of "
1736                        "unsupported optional features (%Lx).\n",
1737                        (unsigned long long)features);
1738                 err = -EINVAL;
1739                 goto fail_iput;
1740         }
1741
1742         features = btrfs_super_incompat_flags(disk_super);
1743         if (!(features & BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF)) {
1744                 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
1745                 btrfs_set_super_incompat_flags(disk_super, features);
1746         }
1747
1748         features = btrfs_super_compat_ro_flags(disk_super) &
1749                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1750         if (!(sb->s_flags & MS_RDONLY) && features) {
1751                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1752                        "unsupported option features (%Lx).\n",
1753                        (unsigned long long)features);
1754                 err = -EINVAL;
1755                 goto fail_iput;
1756         }
1757
1758         btrfs_init_workers(&fs_info->generic_worker,
1759                            "genwork", 1, NULL);
1760
1761         btrfs_init_workers(&fs_info->workers, "worker",
1762                            fs_info->thread_pool_size,
1763                            &fs_info->generic_worker);
1764
1765         btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1766                            fs_info->thread_pool_size,
1767                            &fs_info->generic_worker);
1768
1769         btrfs_init_workers(&fs_info->submit_workers, "submit",
1770                            min_t(u64, fs_devices->num_devices,
1771                            fs_info->thread_pool_size),
1772                            &fs_info->generic_worker);
1773
1774         /* a higher idle thresh on the submit workers makes it much more
1775          * likely that bios will be send down in a sane order to the
1776          * devices
1777          */
1778         fs_info->submit_workers.idle_thresh = 64;
1779
1780         fs_info->workers.idle_thresh = 16;
1781         fs_info->workers.ordered = 1;
1782
1783         fs_info->delalloc_workers.idle_thresh = 2;
1784         fs_info->delalloc_workers.ordered = 1;
1785
1786         btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
1787                            &fs_info->generic_worker);
1788         btrfs_init_workers(&fs_info->endio_workers, "endio",
1789                            fs_info->thread_pool_size,
1790                            &fs_info->generic_worker);
1791         btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1792                            fs_info->thread_pool_size,
1793                            &fs_info->generic_worker);
1794         btrfs_init_workers(&fs_info->endio_meta_write_workers,
1795                            "endio-meta-write", fs_info->thread_pool_size,
1796                            &fs_info->generic_worker);
1797         btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1798                            fs_info->thread_pool_size,
1799                            &fs_info->generic_worker);
1800         btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
1801                            1, &fs_info->generic_worker);
1802
1803         /*
1804          * endios are largely parallel and should have a very
1805          * low idle thresh
1806          */
1807         fs_info->endio_workers.idle_thresh = 4;
1808         fs_info->endio_meta_workers.idle_thresh = 4;
1809
1810         fs_info->endio_write_workers.idle_thresh = 2;
1811         fs_info->endio_meta_write_workers.idle_thresh = 2;
1812
1813         btrfs_start_workers(&fs_info->workers, 1);
1814         btrfs_start_workers(&fs_info->generic_worker, 1);
1815         btrfs_start_workers(&fs_info->submit_workers, 1);
1816         btrfs_start_workers(&fs_info->delalloc_workers, 1);
1817         btrfs_start_workers(&fs_info->fixup_workers, 1);
1818         btrfs_start_workers(&fs_info->endio_workers, 1);
1819         btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1820         btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1821         btrfs_start_workers(&fs_info->endio_write_workers, 1);
1822         btrfs_start_workers(&fs_info->endio_freespace_worker, 1);
1823
1824         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1825         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1826                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1827
1828         nodesize = btrfs_super_nodesize(disk_super);
1829         leafsize = btrfs_super_leafsize(disk_super);
1830         sectorsize = btrfs_super_sectorsize(disk_super);
1831         stripesize = btrfs_super_stripesize(disk_super);
1832         tree_root->nodesize = nodesize;
1833         tree_root->leafsize = leafsize;
1834         tree_root->sectorsize = sectorsize;
1835         tree_root->stripesize = stripesize;
1836
1837         sb->s_blocksize = sectorsize;
1838         sb->s_blocksize_bits = blksize_bits(sectorsize);
1839
1840         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1841                     sizeof(disk_super->magic))) {
1842                 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
1843                 goto fail_sb_buffer;
1844         }
1845
1846         mutex_lock(&fs_info->chunk_mutex);
1847         ret = btrfs_read_sys_array(tree_root);
1848         mutex_unlock(&fs_info->chunk_mutex);
1849         if (ret) {
1850                 printk(KERN_WARNING "btrfs: failed to read the system "
1851                        "array on %s\n", sb->s_id);
1852                 goto fail_sb_buffer;
1853         }
1854
1855         blocksize = btrfs_level_size(tree_root,
1856                                      btrfs_super_chunk_root_level(disk_super));
1857         generation = btrfs_super_chunk_root_generation(disk_super);
1858
1859         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1860                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1861
1862         chunk_root->node = read_tree_block(chunk_root,
1863                                            btrfs_super_chunk_root(disk_super),
1864                                            blocksize, generation);
1865         BUG_ON(!chunk_root->node);
1866         if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
1867                 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
1868                        sb->s_id);
1869                 goto fail_chunk_root;
1870         }
1871         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
1872         chunk_root->commit_root = btrfs_root_node(chunk_root);
1873
1874         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1875            (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1876            BTRFS_UUID_SIZE);
1877
1878         mutex_lock(&fs_info->chunk_mutex);
1879         ret = btrfs_read_chunk_tree(chunk_root);
1880         mutex_unlock(&fs_info->chunk_mutex);
1881         if (ret) {
1882                 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
1883                        sb->s_id);
1884                 goto fail_chunk_root;
1885         }
1886
1887         btrfs_close_extra_devices(fs_devices);
1888
1889         blocksize = btrfs_level_size(tree_root,
1890                                      btrfs_super_root_level(disk_super));
1891         generation = btrfs_super_generation(disk_super);
1892
1893         tree_root->node = read_tree_block(tree_root,
1894                                           btrfs_super_root(disk_super),
1895                                           blocksize, generation);
1896         if (!tree_root->node)
1897                 goto fail_chunk_root;
1898         if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
1899                 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
1900                        sb->s_id);
1901                 goto fail_tree_root;
1902         }
1903         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
1904         tree_root->commit_root = btrfs_root_node(tree_root);
1905
1906         ret = find_and_setup_root(tree_root, fs_info,
1907                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1908         if (ret)
1909                 goto fail_tree_root;
1910         extent_root->track_dirty = 1;
1911
1912         ret = find_and_setup_root(tree_root, fs_info,
1913                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
1914         if (ret)
1915                 goto fail_extent_root;
1916         dev_root->track_dirty = 1;
1917
1918         ret = find_and_setup_root(tree_root, fs_info,
1919                                   BTRFS_CSUM_TREE_OBJECTID, csum_root);
1920         if (ret)
1921                 goto fail_dev_root;
1922
1923         csum_root->track_dirty = 1;
1924
1925         fs_info->generation = generation;
1926         fs_info->last_trans_committed = generation;
1927         fs_info->data_alloc_profile = (u64)-1;
1928         fs_info->metadata_alloc_profile = (u64)-1;
1929         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1930
1931         ret = btrfs_read_block_groups(extent_root);
1932         if (ret) {
1933                 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
1934                 goto fail_block_groups;
1935         }
1936
1937         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1938                                                "btrfs-cleaner");
1939         if (IS_ERR(fs_info->cleaner_kthread))
1940                 goto fail_block_groups;
1941
1942         fs_info->transaction_kthread = kthread_run(transaction_kthread,
1943                                                    tree_root,
1944                                                    "btrfs-transaction");
1945         if (IS_ERR(fs_info->transaction_kthread))
1946                 goto fail_cleaner;
1947
1948         if (!btrfs_test_opt(tree_root, SSD) &&
1949             !btrfs_test_opt(tree_root, NOSSD) &&
1950             !fs_info->fs_devices->rotating) {
1951                 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
1952                        "mode\n");
1953                 btrfs_set_opt(fs_info->mount_opt, SSD);
1954         }
1955
1956         if (btrfs_super_log_root(disk_super) != 0) {
1957                 u64 bytenr = btrfs_super_log_root(disk_super);
1958
1959                 if (fs_devices->rw_devices == 0) {
1960                         printk(KERN_WARNING "Btrfs log replay required "
1961                                "on RO media\n");
1962                         err = -EIO;
1963                         goto fail_trans_kthread;
1964                 }
1965                 blocksize =
1966                      btrfs_level_size(tree_root,
1967                                       btrfs_super_log_root_level(disk_super));
1968
1969                 log_tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
1970                 if (!log_tree_root) {
1971                         err = -ENOMEM;
1972                         goto fail_trans_kthread;
1973                 }
1974
1975                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1976                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1977
1978                 log_tree_root->node = read_tree_block(tree_root, bytenr,
1979                                                       blocksize,
1980                                                       generation + 1);
1981                 ret = btrfs_recover_log_trees(log_tree_root);
1982                 BUG_ON(ret);
1983
1984                 if (sb->s_flags & MS_RDONLY) {
1985                         ret =  btrfs_commit_super(tree_root);
1986                         BUG_ON(ret);
1987                 }
1988         }
1989
1990         ret = btrfs_find_orphan_roots(tree_root);
1991         BUG_ON(ret);
1992
1993         if (!(sb->s_flags & MS_RDONLY)) {
1994                 ret = btrfs_cleanup_fs_roots(fs_info);
1995                 BUG_ON(ret);
1996
1997                 ret = btrfs_recover_relocation(tree_root);
1998                 if (ret < 0) {
1999                         printk(KERN_WARNING
2000                                "btrfs: failed to recover relocation\n");
2001                         err = -EINVAL;
2002                         goto fail_trans_kthread;
2003                 }
2004         }
2005
2006         location.objectid = BTRFS_FS_TREE_OBJECTID;
2007         location.type = BTRFS_ROOT_ITEM_KEY;
2008         location.offset = (u64)-1;
2009
2010         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2011         if (!fs_info->fs_root)
2012                 goto fail_trans_kthread;
2013         if (IS_ERR(fs_info->fs_root)) {
2014                 err = PTR_ERR(fs_info->fs_root);
2015                 goto fail_trans_kthread;
2016         }
2017
2018         if (!(sb->s_flags & MS_RDONLY)) {
2019                 down_read(&fs_info->cleanup_work_sem);
2020                 btrfs_orphan_cleanup(fs_info->fs_root);
2021                 btrfs_orphan_cleanup(fs_info->tree_root);
2022                 up_read(&fs_info->cleanup_work_sem);
2023         }
2024
2025         return tree_root;
2026
2027 fail_trans_kthread:
2028         kthread_stop(fs_info->transaction_kthread);
2029 fail_cleaner:
2030         kthread_stop(fs_info->cleaner_kthread);
2031
2032         /*
2033          * make sure we're done with the btree inode before we stop our
2034          * kthreads
2035          */
2036         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2037         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2038
2039 fail_block_groups:
2040         btrfs_free_block_groups(fs_info);
2041         free_extent_buffer(csum_root->node);
2042         free_extent_buffer(csum_root->commit_root);
2043 fail_dev_root:
2044         free_extent_buffer(dev_root->node);
2045         free_extent_buffer(dev_root->commit_root);
2046 fail_extent_root:
2047         free_extent_buffer(extent_root->node);
2048         free_extent_buffer(extent_root->commit_root);
2049 fail_tree_root:
2050         free_extent_buffer(tree_root->node);
2051         free_extent_buffer(tree_root->commit_root);
2052 fail_chunk_root:
2053         free_extent_buffer(chunk_root->node);
2054         free_extent_buffer(chunk_root->commit_root);
2055 fail_sb_buffer:
2056         btrfs_stop_workers(&fs_info->generic_worker);
2057         btrfs_stop_workers(&fs_info->fixup_workers);
2058         btrfs_stop_workers(&fs_info->delalloc_workers);
2059         btrfs_stop_workers(&fs_info->workers);
2060         btrfs_stop_workers(&fs_info->endio_workers);
2061         btrfs_stop_workers(&fs_info->endio_meta_workers);
2062         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2063         btrfs_stop_workers(&fs_info->endio_write_workers);
2064         btrfs_stop_workers(&fs_info->endio_freespace_worker);
2065         btrfs_stop_workers(&fs_info->submit_workers);
2066 fail_iput:
2067         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2068         iput(fs_info->btree_inode);
2069
2070         btrfs_close_devices(fs_info->fs_devices);
2071         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2072 fail_bdi:
2073         bdi_destroy(&fs_info->bdi);
2074 fail_srcu:
2075         cleanup_srcu_struct(&fs_info->subvol_srcu);
2076 fail:
2077         kfree(extent_root);
2078         kfree(tree_root);
2079         kfree(fs_info);
2080         kfree(chunk_root);
2081         kfree(dev_root);
2082         kfree(csum_root);
2083         return ERR_PTR(err);
2084 }
2085
2086 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2087 {
2088         char b[BDEVNAME_SIZE];
2089
2090         if (uptodate) {
2091                 set_buffer_uptodate(bh);
2092         } else {
2093                 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
2094                         printk(KERN_WARNING "lost page write due to "
2095                                         "I/O error on %s\n",
2096                                        bdevname(bh->b_bdev, b));
2097                 }
2098                 /* note, we dont' set_buffer_write_io_error because we have
2099                  * our own ways of dealing with the IO errors
2100                  */
2101                 clear_buffer_uptodate(bh);
2102         }
2103         unlock_buffer(bh);
2104         put_bh(bh);
2105 }
2106
2107 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2108 {
2109         struct buffer_head *bh;
2110         struct buffer_head *latest = NULL;
2111         struct btrfs_super_block *super;
2112         int i;
2113         u64 transid = 0;
2114         u64 bytenr;
2115
2116         /* we would like to check all the supers, but that would make
2117          * a btrfs mount succeed after a mkfs from a different FS.
2118          * So, we need to add a special mount option to scan for
2119          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2120          */
2121         for (i = 0; i < 1; i++) {
2122                 bytenr = btrfs_sb_offset(i);
2123                 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2124                         break;
2125                 bh = __bread(bdev, bytenr / 4096, 4096);
2126                 if (!bh)
2127                         continue;
2128
2129                 super = (struct btrfs_super_block *)bh->b_data;
2130                 if (btrfs_super_bytenr(super) != bytenr ||
2131                     strncmp((char *)(&super->magic), BTRFS_MAGIC,
2132                             sizeof(super->magic))) {
2133                         brelse(bh);
2134                         continue;
2135                 }
2136
2137                 if (!latest || btrfs_super_generation(super) > transid) {
2138                         brelse(latest);
2139                         latest = bh;
2140                         transid = btrfs_super_generation(super);
2141                 } else {
2142                         brelse(bh);
2143                 }
2144         }
2145         return latest;
2146 }
2147
2148 /*
2149  * this should be called twice, once with wait == 0 and
2150  * once with wait == 1.  When wait == 0 is done, all the buffer heads
2151  * we write are pinned.
2152  *
2153  * They are released when wait == 1 is done.
2154  * max_mirrors must be the same for both runs, and it indicates how
2155  * many supers on this one device should be written.
2156  *
2157  * max_mirrors == 0 means to write them all.
2158  */
2159 static int write_dev_supers(struct btrfs_device *device,
2160                             struct btrfs_super_block *sb,
2161                             int do_barriers, int wait, int max_mirrors)
2162 {
2163         struct buffer_head *bh;
2164         int i;
2165         int ret;
2166         int errors = 0;
2167         u32 crc;
2168         u64 bytenr;
2169         int last_barrier = 0;
2170
2171         if (max_mirrors == 0)
2172                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2173
2174         /* make sure only the last submit_bh does a barrier */
2175         if (do_barriers) {
2176                 for (i = 0; i < max_mirrors; i++) {
2177                         bytenr = btrfs_sb_offset(i);
2178                         if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2179                             device->total_bytes)
2180                                 break;
2181                         last_barrier = i;
2182                 }
2183         }
2184
2185         for (i = 0; i < max_mirrors; i++) {
2186                 bytenr = btrfs_sb_offset(i);
2187                 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2188                         break;
2189
2190                 if (wait) {
2191                         bh = __find_get_block(device->bdev, bytenr / 4096,
2192                                               BTRFS_SUPER_INFO_SIZE);
2193                         BUG_ON(!bh);
2194                         wait_on_buffer(bh);
2195                         if (!buffer_uptodate(bh))
2196                                 errors++;
2197
2198                         /* drop our reference */
2199                         brelse(bh);
2200
2201                         /* drop the reference from the wait == 0 run */
2202                         brelse(bh);
2203                         continue;
2204                 } else {
2205                         btrfs_set_super_bytenr(sb, bytenr);
2206
2207                         crc = ~(u32)0;
2208                         crc = btrfs_csum_data(NULL, (char *)sb +
2209                                               BTRFS_CSUM_SIZE, crc,
2210                                               BTRFS_SUPER_INFO_SIZE -
2211                                               BTRFS_CSUM_SIZE);
2212                         btrfs_csum_final(crc, sb->csum);
2213
2214                         /*
2215                          * one reference for us, and we leave it for the
2216                          * caller
2217                          */
2218                         bh = __getblk(device->bdev, bytenr / 4096,
2219                                       BTRFS_SUPER_INFO_SIZE);
2220                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2221
2222                         /* one reference for submit_bh */
2223                         get_bh(bh);
2224
2225                         set_buffer_uptodate(bh);
2226                         lock_buffer(bh);
2227                         bh->b_end_io = btrfs_end_buffer_write_sync;
2228                 }
2229
2230                 if (i == last_barrier && do_barriers && device->barriers) {
2231                         ret = submit_bh(WRITE_BARRIER, bh);
2232                         if (ret == -EOPNOTSUPP) {
2233                                 printk("btrfs: disabling barriers on dev %s\n",
2234                                        device->name);
2235                                 set_buffer_uptodate(bh);
2236                                 device->barriers = 0;
2237                                 /* one reference for submit_bh */
2238                                 get_bh(bh);
2239                                 lock_buffer(bh);
2240                                 ret = submit_bh(WRITE_SYNC, bh);
2241                         }
2242                 } else {
2243                         ret = submit_bh(WRITE_SYNC, bh);
2244                 }
2245
2246                 if (ret)
2247                         errors++;
2248         }
2249         return errors < i ? 0 : -1;
2250 }
2251
2252 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2253 {
2254         struct list_head *head;
2255         struct btrfs_device *dev;
2256         struct btrfs_super_block *sb;
2257         struct btrfs_dev_item *dev_item;
2258         int ret;
2259         int do_barriers;
2260         int max_errors;
2261         int total_errors = 0;
2262         u64 flags;
2263
2264         max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2265         do_barriers = !btrfs_test_opt(root, NOBARRIER);
2266
2267         sb = &root->fs_info->super_for_commit;
2268         dev_item = &sb->dev_item;
2269
2270         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2271         head = &root->fs_info->fs_devices->devices;
2272         list_for_each_entry(dev, head, dev_list) {
2273                 if (!dev->bdev) {
2274                         total_errors++;
2275                         continue;
2276                 }
2277                 if (!dev->in_fs_metadata || !dev->writeable)
2278                         continue;
2279
2280                 btrfs_set_stack_device_generation(dev_item, 0);
2281                 btrfs_set_stack_device_type(dev_item, dev->type);
2282                 btrfs_set_stack_device_id(dev_item, dev->devid);
2283                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2284                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2285                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2286                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2287                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2288                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2289                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2290
2291                 flags = btrfs_super_flags(sb);
2292                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2293
2294                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2295                 if (ret)
2296                         total_errors++;
2297         }
2298         if (total_errors > max_errors) {
2299                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2300                        total_errors);
2301                 BUG();
2302         }
2303
2304         total_errors = 0;
2305         list_for_each_entry(dev, head, dev_list) {
2306                 if (!dev->bdev)
2307                         continue;
2308                 if (!dev->in_fs_metadata || !dev->writeable)
2309                         continue;
2310
2311                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2312                 if (ret)
2313                         total_errors++;
2314         }
2315         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2316         if (total_errors > max_errors) {
2317                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2318                        total_errors);
2319                 BUG();
2320         }
2321         return 0;
2322 }
2323
2324 int write_ctree_super(struct btrfs_trans_handle *trans,
2325                       struct btrfs_root *root, int max_mirrors)
2326 {
2327         int ret;
2328
2329         ret = write_all_supers(root, max_mirrors);
2330         return ret;
2331 }
2332
2333 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2334 {
2335         spin_lock(&fs_info->fs_roots_radix_lock);
2336         radix_tree_delete(&fs_info->fs_roots_radix,
2337                           (unsigned long)root->root_key.objectid);
2338         spin_unlock(&fs_info->fs_roots_radix_lock);
2339
2340         if (btrfs_root_refs(&root->root_item) == 0)
2341                 synchronize_srcu(&fs_info->subvol_srcu);
2342
2343         free_fs_root(root);
2344         return 0;
2345 }
2346
2347 static void free_fs_root(struct btrfs_root *root)
2348 {
2349         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2350         if (root->anon_super.s_dev) {
2351                 down_write(&root->anon_super.s_umount);
2352                 kill_anon_super(&root->anon_super);
2353         }
2354         free_extent_buffer(root->node);
2355         free_extent_buffer(root->commit_root);
2356         kfree(root->name);
2357         kfree(root);
2358 }
2359
2360 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2361 {
2362         int ret;
2363         struct btrfs_root *gang[8];
2364         int i;
2365
2366         while (!list_empty(&fs_info->dead_roots)) {
2367                 gang[0] = list_entry(fs_info->dead_roots.next,
2368                                      struct btrfs_root, root_list);
2369                 list_del(&gang[0]->root_list);
2370
2371                 if (gang[0]->in_radix) {
2372                         btrfs_free_fs_root(fs_info, gang[0]);
2373                 } else {
2374                         free_extent_buffer(gang[0]->node);
2375                         free_extent_buffer(gang[0]->commit_root);
2376                         kfree(gang[0]);
2377                 }
2378         }
2379
2380         while (1) {
2381                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2382                                              (void **)gang, 0,
2383                                              ARRAY_SIZE(gang));
2384                 if (!ret)
2385                         break;
2386                 for (i = 0; i < ret; i++)
2387                         btrfs_free_fs_root(fs_info, gang[i]);
2388         }
2389         return 0;
2390 }
2391
2392 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2393 {
2394         u64 root_objectid = 0;
2395         struct btrfs_root *gang[8];
2396         int i;
2397         int ret;
2398
2399         while (1) {
2400                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2401                                              (void **)gang, root_objectid,
2402                                              ARRAY_SIZE(gang));
2403                 if (!ret)
2404                         break;
2405
2406                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2407                 for (i = 0; i < ret; i++) {
2408                         root_objectid = gang[i]->root_key.objectid;
2409                         btrfs_orphan_cleanup(gang[i]);
2410                 }
2411                 root_objectid++;
2412         }
2413         return 0;
2414 }
2415
2416 int btrfs_commit_super(struct btrfs_root *root)
2417 {
2418         struct btrfs_trans_handle *trans;
2419         int ret;
2420
2421         mutex_lock(&root->fs_info->cleaner_mutex);
2422         btrfs_run_delayed_iputs(root);
2423         btrfs_clean_old_snapshots(root);
2424         mutex_unlock(&root->fs_info->cleaner_mutex);
2425
2426         /* wait until ongoing cleanup work done */
2427         down_write(&root->fs_info->cleanup_work_sem);
2428         up_write(&root->fs_info->cleanup_work_sem);
2429
2430         trans = btrfs_join_transaction(root, 1);
2431         ret = btrfs_commit_transaction(trans, root);
2432         BUG_ON(ret);
2433         /* run commit again to drop the original snapshot */
2434         trans = btrfs_join_transaction(root, 1);
2435         btrfs_commit_transaction(trans, root);
2436         ret = btrfs_write_and_wait_transaction(NULL, root);
2437         BUG_ON(ret);
2438
2439         ret = write_ctree_super(NULL, root, 0);
2440         return ret;
2441 }
2442
2443 int close_ctree(struct btrfs_root *root)
2444 {
2445         struct btrfs_fs_info *fs_info = root->fs_info;
2446         int ret;
2447
2448         fs_info->closing = 1;
2449         smp_mb();
2450
2451         btrfs_put_block_group_cache(fs_info);
2452         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2453                 ret =  btrfs_commit_super(root);
2454                 if (ret)
2455                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2456         }
2457
2458         kthread_stop(root->fs_info->transaction_kthread);
2459         kthread_stop(root->fs_info->cleaner_kthread);
2460
2461         fs_info->closing = 2;
2462         smp_mb();
2463
2464         if (fs_info->delalloc_bytes) {
2465                 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2466                        (unsigned long long)fs_info->delalloc_bytes);
2467         }
2468         if (fs_info->total_ref_cache_size) {
2469                 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2470                        (unsigned long long)fs_info->total_ref_cache_size);
2471         }
2472
2473         free_extent_buffer(fs_info->extent_root->node);
2474         free_extent_buffer(fs_info->extent_root->commit_root);
2475         free_extent_buffer(fs_info->tree_root->node);
2476         free_extent_buffer(fs_info->tree_root->commit_root);
2477         free_extent_buffer(root->fs_info->chunk_root->node);
2478         free_extent_buffer(root->fs_info->chunk_root->commit_root);
2479         free_extent_buffer(root->fs_info->dev_root->node);
2480         free_extent_buffer(root->fs_info->dev_root->commit_root);
2481         free_extent_buffer(root->fs_info->csum_root->node);
2482         free_extent_buffer(root->fs_info->csum_root->commit_root);
2483
2484         btrfs_free_block_groups(root->fs_info);
2485
2486         del_fs_roots(fs_info);
2487
2488         iput(fs_info->btree_inode);
2489
2490         btrfs_stop_workers(&fs_info->generic_worker);
2491         btrfs_stop_workers(&fs_info->fixup_workers);
2492         btrfs_stop_workers(&fs_info->delalloc_workers);
2493         btrfs_stop_workers(&fs_info->workers);
2494         btrfs_stop_workers(&fs_info->endio_workers);
2495         btrfs_stop_workers(&fs_info->endio_meta_workers);
2496         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2497         btrfs_stop_workers(&fs_info->endio_write_workers);
2498         btrfs_stop_workers(&fs_info->endio_freespace_worker);
2499         btrfs_stop_workers(&fs_info->submit_workers);
2500
2501         btrfs_close_devices(fs_info->fs_devices);
2502         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2503
2504         bdi_destroy(&fs_info->bdi);
2505         cleanup_srcu_struct(&fs_info->subvol_srcu);
2506
2507         kfree(fs_info->extent_root);
2508         kfree(fs_info->tree_root);
2509         kfree(fs_info->chunk_root);
2510         kfree(fs_info->dev_root);
2511         kfree(fs_info->csum_root);
2512         return 0;
2513 }
2514
2515 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2516 {
2517         int ret;
2518         struct inode *btree_inode = buf->first_page->mapping->host;
2519
2520         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
2521                                      NULL);
2522         if (!ret)
2523                 return ret;
2524
2525         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2526                                     parent_transid);
2527         return !ret;
2528 }
2529
2530 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2531 {
2532         struct inode *btree_inode = buf->first_page->mapping->host;
2533         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2534                                           buf);
2535 }
2536
2537 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2538 {
2539         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2540         u64 transid = btrfs_header_generation(buf);
2541         struct inode *btree_inode = root->fs_info->btree_inode;
2542         int was_dirty;
2543
2544         btrfs_assert_tree_locked(buf);
2545         if (transid != root->fs_info->generation) {
2546                 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2547                        "found %llu running %llu\n",
2548                         (unsigned long long)buf->start,
2549                         (unsigned long long)transid,
2550                         (unsigned long long)root->fs_info->generation);
2551                 WARN_ON(1);
2552         }
2553         was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2554                                             buf);
2555         if (!was_dirty) {
2556                 spin_lock(&root->fs_info->delalloc_lock);
2557                 root->fs_info->dirty_metadata_bytes += buf->len;
2558                 spin_unlock(&root->fs_info->delalloc_lock);
2559         }
2560 }
2561
2562 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2563 {
2564         /*
2565          * looks as though older kernels can get into trouble with
2566          * this code, they end up stuck in balance_dirty_pages forever
2567          */
2568         u64 num_dirty;
2569         unsigned long thresh = 32 * 1024 * 1024;
2570
2571         if (current->flags & PF_MEMALLOC)
2572                 return;
2573
2574         num_dirty = root->fs_info->dirty_metadata_bytes;
2575
2576         if (num_dirty > thresh) {
2577                 balance_dirty_pages_ratelimited_nr(
2578                                    root->fs_info->btree_inode->i_mapping, 1);
2579         }
2580         return;
2581 }
2582
2583 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2584 {
2585         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2586         int ret;
2587         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2588         if (ret == 0)
2589                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
2590         return ret;
2591 }
2592
2593 int btree_lock_page_hook(struct page *page)
2594 {
2595         struct inode *inode = page->mapping->host;
2596         struct btrfs_root *root = BTRFS_I(inode)->root;
2597         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2598         struct extent_buffer *eb;
2599         unsigned long len;
2600         u64 bytenr = page_offset(page);
2601
2602         if (page->private == EXTENT_PAGE_PRIVATE)
2603                 goto out;
2604
2605         len = page->private >> 2;
2606         eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2607         if (!eb)
2608                 goto out;
2609
2610         btrfs_tree_lock(eb);
2611         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2612
2613         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2614                 spin_lock(&root->fs_info->delalloc_lock);
2615                 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2616                         root->fs_info->dirty_metadata_bytes -= eb->len;
2617                 else
2618                         WARN_ON(1);
2619                 spin_unlock(&root->fs_info->delalloc_lock);
2620         }
2621
2622         btrfs_tree_unlock(eb);
2623         free_extent_buffer(eb);
2624 out:
2625         lock_page(page);
2626         return 0;
2627 }
2628
2629 static struct extent_io_ops btree_extent_io_ops = {
2630         .write_cache_pages_lock_hook = btree_lock_page_hook,
2631         .readpage_end_io_hook = btree_readpage_end_io_hook,
2632         .submit_bio_hook = btree_submit_bio_hook,
2633         /* note we're sharing with inode.c for the merge bio hook */
2634         .merge_bio_hook = btrfs_merge_bio_hook,
2635 };