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 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
  48                                     int read_only);
  49 static int btrfs_destroy_ordered_operations(struct btrfs_root *root);
  50 static int btrfs_destroy_ordered_extents(struct btrfs_root *root);
  51 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
  52                                       struct btrfs_root *root);
  53 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t);
  54 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
  55 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
  56                                         struct extent_io_tree *dirty_pages,
  57                                         int mark);
  58 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
  59                                        struct extent_io_tree *pinned_extents);
  60 static int btrfs_cleanup_transaction(struct btrfs_root *root);
  61
  62 /*
  63  * end_io_wq structs are used to do processing in task context when an IO is
  64  * complete.  This is used during reads to verify checksums, and it is used
  65  * by writes to insert metadata for new file extents after IO is complete.
  66  */
  67 struct end_io_wq {
  68         struct bio *bio;
  69         bio_end_io_t *end_io;
  70         void *private;
  71         struct btrfs_fs_info *info;
  72         int error;
  73         int metadata;
  74         struct list_head list;
  75         struct btrfs_work work;
  76 };
  77
  78 /*
  79  * async submit bios are used to offload expensive checksumming
  80  * onto the worker threads.  They checksum file and metadata bios
  81  * just before they are sent down the IO stack.
  82  */
  83 struct async_submit_bio {
  84         struct inode *inode;
  85         struct bio *bio;
  86         struct list_head list;
  87         extent_submit_bio_hook_t *submit_bio_start;
  88         extent_submit_bio_hook_t *submit_bio_done;
  89         int rw;
  90         int mirror_num;
  91         unsigned long bio_flags;
  92         /*
  93          * bio_offset is optional, can be used if the pages in the bio
  94          * can't tell us where in the file the bio should go
  95          */
  96         u64 bio_offset;
  97         struct btrfs_work work;
  98 };
  99
 100 /* These are used to set the lockdep class on the extent buffer locks.
 101  * The class is set by the readpage_end_io_hook after the buffer has
 102  * passed csum validation but before the pages are unlocked.
 103  *
 104  * The lockdep class is also set by btrfs_init_new_buffer on freshly
 105  * allocated blocks.
 106  *
 107  * The class is based on the level in the tree block, which allows lockdep
 108  * to know that lower nodes nest inside the locks of higher nodes.
 109  *
 110  * We also add a check to make sure the highest level of the tree is
 111  * the same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this
 112  * code needs update as well.
 113  */
 114 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 115 # if BTRFS_MAX_LEVEL != 8
 116 #  error
 117 # endif
 118 static struct lock_class_key btrfs_eb_class[BTRFS_MAX_LEVEL + 1];
 119 static const char *btrfs_eb_name[BTRFS_MAX_LEVEL + 1] = {
 120         /* leaf */
 121         "btrfs-extent-00",
 122         "btrfs-extent-01",
 123         "btrfs-extent-02",
 124         "btrfs-extent-03",
 125         "btrfs-extent-04",
 126         "btrfs-extent-05",
 127         "btrfs-extent-06",
 128         "btrfs-extent-07",
 129         /* highest possible level */
 130         "btrfs-extent-08",
 131 };
 132 #endif
 133
 134 /*
 135  * extents on the btree inode are pretty simple, there's one extent
 136  * that covers the entire device
 137  */
 138 static struct extent_map *btree_get_extent(struct inode *inode,
 139                 struct page *page, size_t page_offset, u64 start, u64 len,
 140                 int create)
 141 {
 142         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 143         struct extent_map *em;
 144         int ret;
 145
 146         read_lock(&em_tree->lock);
 147         em = lookup_extent_mapping(em_tree, start, len);
 148         if (em) {
 149                 em->bdev =
 150                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 151                 read_unlock(&em_tree->lock);
 152                 goto out;
 153         }
 154         read_unlock(&em_tree->lock);
 155
 156         em = alloc_extent_map(GFP_NOFS);
 157         if (!em) {
 158                 em = ERR_PTR(-ENOMEM);
 159                 goto out;
 160         }
 161         em->start = 0;
 162         em->len = (u64)-1;
 163         em->block_len = (u64)-1;
 164         em->block_start = 0;
 165         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 166
 167         write_lock(&em_tree->lock);
 168         ret = add_extent_mapping(em_tree, em);
 169         if (ret == -EEXIST) {
 170                 u64 failed_start = em->start;
 171                 u64 failed_len = em->len;
 172
 173                 free_extent_map(em);
 174                 em = lookup_extent_mapping(em_tree, start, len);
 175                 if (em) {
 176                         ret = 0;
 177                 } else {
 178                         em = lookup_extent_mapping(em_tree, failed_start,
 179                                                    failed_len);
 180                         ret = -EIO;
 181                 }
 182         } else if (ret) {
 183                 free_extent_map(em);
 184                 em = NULL;
 185         }
 186         write_unlock(&em_tree->lock);
 187
 188         if (ret)
 189                 em = ERR_PTR(ret);
 190 out:
 191         return em;
 192 }
 193
 194 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
 195 {
 196         return crc32c(seed, data, len);
 197 }
 198
 199 void btrfs_csum_final(u32 crc, char *result)
 200 {
 201         *(__le32 *)result = ~cpu_to_le32(crc);
 202 }
 203
 204 /*
 205  * compute the csum for a btree block, and either verify it or write it
 206  * into the csum field of the block.
 207  */
 208 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 209                            int verify)
 210 {
 211         u16 csum_size =
 212                 btrfs_super_csum_size(&root->fs_info->super_copy);
 213         char *result = NULL;
 214         unsigned long len;
 215         unsigned long cur_len;
 216         unsigned long offset = BTRFS_CSUM_SIZE;
 217         char *map_token = NULL;
 218         char *kaddr;
 219         unsigned long map_start;
 220         unsigned long map_len;
 221         int err;
 222         u32 crc = ~(u32)0;
 223         unsigned long inline_result;
 224
 225         len = buf->len - offset;
 226         while (len > 0) {
 227                 err = map_private_extent_buffer(buf, offset, 32,
 228                                         &map_token, &kaddr,
 229                                         &map_start, &map_len, KM_USER0);
 230                 if (err)
 231                         return 1;
 232                 cur_len = min(len, map_len - (offset - map_start));
 233                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
 234                                       crc, cur_len);
 235                 len -= cur_len;
 236                 offset += cur_len;
 237                 unmap_extent_buffer(buf, map_token, KM_USER0);
 238         }
 239         if (csum_size > sizeof(inline_result)) {
 240                 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
 241                 if (!result)
 242                         return 1;
 243         } else {
 244                 result = (char *)&inline_result;
 245         }
 246
 247         btrfs_csum_final(crc, result);
 248
 249         if (verify) {
 250                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
 251                         u32 val;
 252                         u32 found = 0;
 253                         memcpy(&found, result, csum_size);
 254
 255                         read_extent_buffer(buf, &val, 0, csum_size);
 256                         if (printk_ratelimit()) {
 257                                 printk(KERN_INFO "btrfs: %s checksum verify "
 258                                        "failed on %llu wanted %X found %X "
 259                                        "level %d\n",
 260                                        root->fs_info->sb->s_id,
 261                                        (unsigned long long)buf->start, val, found,
 262                                        btrfs_header_level(buf));
 263                         }
 264                         if (result != (char *)&inline_result)
 265                                 kfree(result);
 266                         return 1;
 267                 }
 268         } else {
 269                 write_extent_buffer(buf, result, 0, csum_size);
 270         }
 271         if (result != (char *)&inline_result)
 272                 kfree(result);
 273         return 0;
 274 }
 275
 276 /*
 277  * we can't consider a given block up to date unless the transid of the
 278  * block matches the transid in the parent node's pointer.  This is how we
 279  * detect blocks that either didn't get written at all or got written
 280  * in the wrong place.
 281  */
 282 static int verify_parent_transid(struct extent_io_tree *io_tree,
 283                                  struct extent_buffer *eb, u64 parent_transid)
 284 {
 285         struct extent_state *cached_state = NULL;
 286         int ret;
 287
 288         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 289                 return 0;
 290
 291         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
 292                          0, &cached_state, GFP_NOFS);
 293         if (extent_buffer_uptodate(io_tree, eb, cached_state) &&
 294             btrfs_header_generation(eb) == parent_transid) {
 295                 ret = 0;
 296                 goto out;
 297         }
 298         if (printk_ratelimit()) {
 299                 printk("parent transid verify failed on %llu wanted %llu "
 300                        "found %llu\n",
 301                        (unsigned long long)eb->start,
 302                        (unsigned long long)parent_transid,
 303                        (unsigned long long)btrfs_header_generation(eb));
 304         }
 305         ret = 1;
 306         clear_extent_buffer_uptodate(io_tree, eb, &cached_state);
 307 out:
 308         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
 309                              &cached_state, GFP_NOFS);
 310         return ret;
 311 }
 312
 313 /*
 314  * helper to read a given tree block, doing retries as required when
 315  * the checksums don't match and we have alternate mirrors to try.
 316  */
 317 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 318                                           struct extent_buffer *eb,
 319                                           u64 start, u64 parent_transid)
 320 {
 321         struct extent_io_tree *io_tree;
 322         int ret;
 323         int num_copies = 0;
 324         int mirror_num = 0;
 325
 326         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 327         while (1) {
 328                 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
 329                                                btree_get_extent, mirror_num);
 330                 if (!ret &&
 331                     !verify_parent_transid(io_tree, eb, parent_transid))
 332                         return ret;
 333
 334                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
 335                                               eb->start, eb->len);
 336                 if (num_copies == 1)
 337                         return ret;
 338
 339                 mirror_num++;
 340                 if (mirror_num > num_copies)
 341                         return ret;
 342         }
 343         return -EIO;
 344 }
 345
 346 /*
 347  * checksum a dirty tree block before IO.  This has extra checks to make sure
 348  * we only fill in the checksum field in the first page of a multi-page block
 349  */
 350
 351 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 352 {
 353         struct extent_io_tree *tree;
 354         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
 355         u64 found_start;
 356         unsigned long len;
 357         struct extent_buffer *eb;
 358         int ret;
 359
 360         tree = &BTRFS_I(page->mapping->host)->io_tree;
 361
 362         if (page->private == EXTENT_PAGE_PRIVATE) {
 363                 WARN_ON(1);
 364                 goto out;
 365         }
 366         if (!page->private) {
 367                 WARN_ON(1);
 368                 goto out;
 369         }
 370         len = page->private >> 2;
 371         WARN_ON(len == 0);
 372
 373         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 374         if (eb == NULL) {
 375                 WARN_ON(1);
 376                 goto out;
 377         }
 378         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
 379                                              btrfs_header_generation(eb));
 380         BUG_ON(ret);
 381         WARN_ON(!btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN));
 382
 383         found_start = btrfs_header_bytenr(eb);
 384         if (found_start != start) {
 385                 WARN_ON(1);
 386                 goto err;
 387         }
 388         if (eb->first_page != page) {
 389                 WARN_ON(1);
 390                 goto err;
 391         }
 392         if (!PageUptodate(page)) {
 393                 WARN_ON(1);
 394                 goto err;
 395         }
 396         csum_tree_block(root, eb, 0);
 397 err:
 398         free_extent_buffer(eb);
 399 out:
 400         return 0;
 401 }
 402
 403 static int check_tree_block_fsid(struct btrfs_root *root,
 404                                  struct extent_buffer *eb)
 405 {
 406         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
 407         u8 fsid[BTRFS_UUID_SIZE];
 408         int ret = 1;
 409
 410         read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
 411                            BTRFS_FSID_SIZE);
 412         while (fs_devices) {
 413                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
 414                         ret = 0;
 415                         break;
 416                 }
 417                 fs_devices = fs_devices->seed;
 418         }
 419         return ret;
 420 }
 421
 422 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 423 void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level)
 424 {
 425         lockdep_set_class_and_name(&eb->lock,
 426                            &btrfs_eb_class[level],
 427                            btrfs_eb_name[level]);
 428 }
 429 #endif
 430
 431 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 432                                struct extent_state *state)
 433 {
 434         struct extent_io_tree *tree;
 435         u64 found_start;
 436         int found_level;
 437         unsigned long len;
 438         struct extent_buffer *eb;
 439         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 440         int ret = 0;
 441
 442         tree = &BTRFS_I(page->mapping->host)->io_tree;
 443         if (page->private == EXTENT_PAGE_PRIVATE)
 444                 goto out;
 445         if (!page->private)
 446                 goto out;
 447
 448         len = page->private >> 2;
 449         WARN_ON(len == 0);
 450
 451         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 452         if (eb == NULL) {
 453                 ret = -EIO;
 454                 goto out;
 455         }
 456
 457         found_start = btrfs_header_bytenr(eb);
 458         if (found_start != start) {
 459                 if (printk_ratelimit()) {
 460                         printk(KERN_INFO "btrfs bad tree block start "
 461                                "%llu %llu\n",
 462                                (unsigned long long)found_start,
 463                                (unsigned long long)eb->start);
 464                 }
 465                 ret = -EIO;
 466                 goto err;
 467         }
 468         if (eb->first_page != page) {
 469                 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
 470                        eb->first_page->index, page->index);
 471                 WARN_ON(1);
 472                 ret = -EIO;
 473                 goto err;
 474         }
 475         if (check_tree_block_fsid(root, eb)) {
 476                 if (printk_ratelimit()) {
 477                         printk(KERN_INFO "btrfs bad fsid on block %llu\n",
 478                                (unsigned long long)eb->start);
 479                 }
 480                 ret = -EIO;
 481                 goto err;
 482         }
 483         found_level = btrfs_header_level(eb);
 484
 485         btrfs_set_buffer_lockdep_class(eb, found_level);
 486
 487         ret = csum_tree_block(root, eb, 1);
 488         if (ret)
 489                 ret = -EIO;
 490
 491         end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
 492         end = eb->start + end - 1;
 493 err:
 494         free_extent_buffer(eb);
 495 out:
 496         return ret;
 497 }
 498
 499 static void end_workqueue_bio(struct bio *bio, int err)
 500 {
 501         struct end_io_wq *end_io_wq = bio->bi_private;
 502         struct btrfs_fs_info *fs_info;
 503
 504         fs_info = end_io_wq->info;
 505         end_io_wq->error = err;
 506         end_io_wq->work.func = end_workqueue_fn;
 507         end_io_wq->work.flags = 0;
 508
 509         if (bio->bi_rw & REQ_WRITE) {
 510                 if (end_io_wq->metadata == 1)
 511                         btrfs_queue_worker(&fs_info->endio_meta_write_workers,
 512                                            &end_io_wq->work);
 513                 else if (end_io_wq->metadata == 2)
 514                         btrfs_queue_worker(&fs_info->endio_freespace_worker,
 515                                            &end_io_wq->work);
 516                 else
 517                         btrfs_queue_worker(&fs_info->endio_write_workers,
 518                                            &end_io_wq->work);
 519         } else {
 520                 if (end_io_wq->metadata)
 521                         btrfs_queue_worker(&fs_info->endio_meta_workers,
 522                                            &end_io_wq->work);
 523                 else
 524                         btrfs_queue_worker(&fs_info->endio_workers,
 525                                            &end_io_wq->work);
 526         }
 527 }
 528
 529 /*
 530  * For the metadata arg you want
 531  *
 532  * 0 - if data
 533  * 1 - if normal metadta
 534  * 2 - if writing to the free space cache area
 535  */
 536 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 537                         int metadata)
 538 {
 539         struct end_io_wq *end_io_wq;
 540         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
 541         if (!end_io_wq)
 542                 return -ENOMEM;
 543
 544         end_io_wq->private = bio->bi_private;
 545         end_io_wq->end_io = bio->bi_end_io;
 546         end_io_wq->info = info;
 547         end_io_wq->error = 0;
 548         end_io_wq->bio = bio;
 549         end_io_wq->metadata = metadata;
 550
 551         bio->bi_private = end_io_wq;
 552         bio->bi_end_io = end_workqueue_bio;
 553         return 0;
 554 }
 555
 556 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
 557 {
 558         unsigned long limit = min_t(unsigned long,
 559                                     info->workers.max_workers,
 560                                     info->fs_devices->open_devices);
 561         return 256 * limit;
 562 }
 563
 564 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
 565 {
 566         return atomic_read(&info->nr_async_bios) >
 567                 btrfs_async_submit_limit(info);
 568 }
 569
 570 static void run_one_async_start(struct btrfs_work *work)
 571 {
 572         struct async_submit_bio *async;
 573
 574         async = container_of(work, struct  async_submit_bio, work);
 575         async->submit_bio_start(async->inode, async->rw, async->bio,
 576                                async->mirror_num, async->bio_flags,
 577                                async->bio_offset);
 578 }
 579
 580 static void run_one_async_done(struct btrfs_work *work)
 581 {
 582         struct btrfs_fs_info *fs_info;
 583         struct async_submit_bio *async;
 584         int limit;
 585
 586         async = container_of(work, struct  async_submit_bio, work);
 587         fs_info = BTRFS_I(async->inode)->root->fs_info;
 588
 589         limit = btrfs_async_submit_limit(fs_info);
 590         limit = limit * 2 / 3;
 591
 592         atomic_dec(&fs_info->nr_async_submits);
 593
 594         if (atomic_read(&fs_info->nr_async_submits) < limit &&
 595             waitqueue_active(&fs_info->async_submit_wait))
 596                 wake_up(&fs_info->async_submit_wait);
 597
 598         async->submit_bio_done(async->inode, async->rw, async->bio,
 599                                async->mirror_num, async->bio_flags,
 600                                async->bio_offset);
 601 }
 602
 603 static void run_one_async_free(struct btrfs_work *work)
 604 {
 605         struct async_submit_bio *async;
 606
 607         async = container_of(work, struct  async_submit_bio, work);
 608         kfree(async);
 609 }
 610
 611 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 612                         int rw, struct bio *bio, int mirror_num,
 613                         unsigned long bio_flags,
 614                         u64 bio_offset,
 615                         extent_submit_bio_hook_t *submit_bio_start,
 616                         extent_submit_bio_hook_t *submit_bio_done)
 617 {
 618         struct async_submit_bio *async;
 619
 620         async = kmalloc(sizeof(*async), GFP_NOFS);
 621         if (!async)
 622                 return -ENOMEM;
 623
 624         async->inode = inode;
 625         async->rw = rw;
 626         async->bio = bio;
 627         async->mirror_num = mirror_num;
 628         async->submit_bio_start = submit_bio_start;
 629         async->submit_bio_done = submit_bio_done;
 630
 631         async->work.func = run_one_async_start;
 632         async->work.ordered_func = run_one_async_done;
 633         async->work.ordered_free = run_one_async_free;
 634
 635         async->work.flags = 0;
 636         async->bio_flags = bio_flags;
 637         async->bio_offset = bio_offset;
 638
 639         atomic_inc(&fs_info->nr_async_submits);
 640
 641         if (rw & REQ_SYNC)
 642                 btrfs_set_work_high_prio(&async->work);
 643
 644         btrfs_queue_worker(&fs_info->workers, &async->work);
 645
 646         while (atomic_read(&fs_info->async_submit_draining) &&
 647               atomic_read(&fs_info->nr_async_submits)) {
 648                 wait_event(fs_info->async_submit_wait,
 649                            (atomic_read(&fs_info->nr_async_submits) == 0));
 650         }
 651
 652         return 0;
 653 }
 654
 655 static int btree_csum_one_bio(struct bio *bio)
 656 {
 657         struct bio_vec *bvec = bio->bi_io_vec;
 658         int bio_index = 0;
 659         struct btrfs_root *root;
 660
 661         WARN_ON(bio->bi_vcnt <= 0);
 662         while (bio_index < bio->bi_vcnt) {
 663                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 664                 csum_dirty_buffer(root, bvec->bv_page);
 665                 bio_index++;
 666                 bvec++;
 667         }
 668         return 0;
 669 }
 670
 671 static int __btree_submit_bio_start(struct inode *inode, int rw,
 672                                     struct bio *bio, int mirror_num,
 673                                     unsigned long bio_flags,
 674                                     u64 bio_offset)
 675 {
 676         /*
 677          * when we're called for a write, we're already in the async
 678          * submission context.  Just jump into btrfs_map_bio
 679          */
 680         btree_csum_one_bio(bio);
 681         return 0;
 682 }
 683
 684 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
 685                                  int mirror_num, unsigned long bio_flags,
 686                                  u64 bio_offset)
 687 {
 688         /*
 689          * when we're called for a write, we're already in the async
 690          * submission context.  Just jump into btrfs_map_bio
 691          */
 692         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
 693 }
 694
 695 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 696                                  int mirror_num, unsigned long bio_flags,
 697                                  u64 bio_offset)
 698 {
 699         int ret;
 700
 701         ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
 702                                           bio, 1);
 703         BUG_ON(ret);
 704
 705         if (!(rw & REQ_WRITE)) {
 706                 /*
 707                  * called for a read, do the setup so that checksum validation
 708                  * can happen in the async kernel threads
 709                  */
 710                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 711                                      mirror_num, 0);
 712         }
 713
 714         /*
 715          * kthread helpers are used to submit writes so that checksumming
 716          * can happen in parallel across all CPUs
 717          */
 718         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 719                                    inode, rw, bio, mirror_num, 0,
 720                                    bio_offset,
 721                                    __btree_submit_bio_start,
 722                                    __btree_submit_bio_done);
 723 }
 724
 725 #ifdef CONFIG_MIGRATION
 726 static int btree_migratepage(struct address_space *mapping,
 727                         struct page *newpage, struct page *page)
 728 {
 729         /*
 730          * we can't safely write a btree page from here,
 731          * we haven't done the locking hook
 732          */
 733         if (PageDirty(page))
 734                 return -EAGAIN;
 735         /*
 736          * Buffers may be managed in a filesystem specific way.
 737          * We must have no buffers or drop them.
 738          */
 739         if (page_has_private(page) &&
 740             !try_to_release_page(page, GFP_KERNEL))
 741                 return -EAGAIN;
 742         return migrate_page(mapping, newpage, page);
 743 }
 744 #endif
 745
 746 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 747 {
 748         struct extent_io_tree *tree;
 749         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 750         struct extent_buffer *eb;
 751         int was_dirty;
 752
 753         tree = &BTRFS_I(page->mapping->host)->io_tree;
 754         if (!(current->flags & PF_MEMALLOC)) {
 755                 return extent_write_full_page(tree, page,
 756                                               btree_get_extent, wbc);
 757         }
 758
 759         redirty_page_for_writepage(wbc, page);
 760         eb = btrfs_find_tree_block(root, page_offset(page), PAGE_CACHE_SIZE);
 761         WARN_ON(!eb);
 762
 763         was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
 764         if (!was_dirty) {
 765                 spin_lock(&root->fs_info->delalloc_lock);
 766                 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
 767                 spin_unlock(&root->fs_info->delalloc_lock);
 768         }
 769         free_extent_buffer(eb);
 770
 771         unlock_page(page);
 772         return 0;
 773 }
 774
 775 static int btree_writepages(struct address_space *mapping,
 776                             struct writeback_control *wbc)
 777 {
 778         struct extent_io_tree *tree;
 779         tree = &BTRFS_I(mapping->host)->io_tree;
 780         if (wbc->sync_mode == WB_SYNC_NONE) {
 781                 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
 782                 u64 num_dirty;
 783                 unsigned long thresh = 32 * 1024 * 1024;
 784
 785                 if (wbc->for_kupdate)
 786                         return 0;
 787
 788                 /* this is a bit racy, but that's ok */
 789                 num_dirty = root->fs_info->dirty_metadata_bytes;
 790                 if (num_dirty < thresh)
 791                         return 0;
 792         }
 793         return extent_writepages(tree, mapping, btree_get_extent, wbc);
 794 }
 795
 796 static int btree_readpage(struct file *file, struct page *page)
 797 {
 798         struct extent_io_tree *tree;
 799         tree = &BTRFS_I(page->mapping->host)->io_tree;
 800         return extent_read_full_page(tree, page, btree_get_extent);
 801 }
 802
 803 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 804 {
 805         struct extent_io_tree *tree;
 806         struct extent_map_tree *map;
 807         int ret;
 808
 809         if (PageWriteback(page) || PageDirty(page))
 810                 return 0;
 811
 812         tree = &BTRFS_I(page->mapping->host)->io_tree;
 813         map = &BTRFS_I(page->mapping->host)->extent_tree;
 814
 815         ret = try_release_extent_state(map, tree, page, gfp_flags);
 816         if (!ret)
 817                 return 0;
 818
 819         ret = try_release_extent_buffer(tree, page);
 820         if (ret == 1) {
 821                 ClearPagePrivate(page);
 822                 set_page_private(page, 0);
 823                 page_cache_release(page);
 824         }
 825
 826         return ret;
 827 }
 828
 829 static void btree_invalidatepage(struct page *page, unsigned long offset)
 830 {
 831         struct extent_io_tree *tree;
 832         tree = &BTRFS_I(page->mapping->host)->io_tree;
 833         extent_invalidatepage(tree, page, offset);
 834         btree_releasepage(page, GFP_NOFS);
 835         if (PagePrivate(page)) {
 836                 printk(KERN_WARNING "btrfs warning page private not zero "
 837                        "on page %llu\n", (unsigned long long)page_offset(page));
 838                 ClearPagePrivate(page);
 839                 set_page_private(page, 0);
 840                 page_cache_release(page);
 841         }
 842 }
 843
 844 static const struct address_space_operations btree_aops = {
 845         .readpage       = btree_readpage,
 846         .writepage      = btree_writepage,
 847         .writepages     = btree_writepages,
 848         .releasepage    = btree_releasepage,
 849         .invalidatepage = btree_invalidatepage,
 850         .sync_page      = block_sync_page,
 851 #ifdef CONFIG_MIGRATION
 852         .migratepage    = btree_migratepage,
 853 #endif
 854 };
 855
 856 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
 857                          u64 parent_transid)
 858 {
 859         struct extent_buffer *buf = NULL;
 860         struct inode *btree_inode = root->fs_info->btree_inode;
 861         int ret = 0;
 862
 863         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 864         if (!buf)
 865                 return 0;
 866         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
 867                                  buf, 0, 0, btree_get_extent, 0);
 868         free_extent_buffer(buf);
 869         return ret;
 870 }
 871
 872 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
 873                                             u64 bytenr, u32 blocksize)
 874 {
 875         struct inode *btree_inode = root->fs_info->btree_inode;
 876         struct extent_buffer *eb;
 877         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 878                                 bytenr, blocksize, GFP_NOFS);
 879         return eb;
 880 }
 881
 882 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
 883                                                  u64 bytenr, u32 blocksize)
 884 {
 885         struct inode *btree_inode = root->fs_info->btree_inode;
 886         struct extent_buffer *eb;
 887
 888         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 889                                  bytenr, blocksize, NULL, GFP_NOFS);
 890         return eb;
 891 }
 892
 893
 894 int btrfs_write_tree_block(struct extent_buffer *buf)
 895 {
 896         return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
 897                                         buf->start + buf->len - 1);
 898 }
 899
 900 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
 901 {
 902         return filemap_fdatawait_range(buf->first_page->mapping,
 903                                        buf->start, buf->start + buf->len - 1);
 904 }
 905
 906 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
 907                                       u32 blocksize, u64 parent_transid)
 908 {
 909         struct extent_buffer *buf = NULL;
 910         int ret;
 911
 912         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 913         if (!buf)
 914                 return NULL;
 915
 916         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
 917
 918         if (ret == 0)
 919                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
 920         return buf;
 921
 922 }
 923
 924 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 925                      struct extent_buffer *buf)
 926 {
 927         struct inode *btree_inode = root->fs_info->btree_inode;
 928         if (btrfs_header_generation(buf) ==
 929             root->fs_info->running_transaction->transid) {
 930                 btrfs_assert_tree_locked(buf);
 931
 932                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
 933                         spin_lock(&root->fs_info->delalloc_lock);
 934                         if (root->fs_info->dirty_metadata_bytes >= buf->len)
 935                                 root->fs_info->dirty_metadata_bytes -= buf->len;
 936                         else
 937                                 WARN_ON(1);
 938                         spin_unlock(&root->fs_info->delalloc_lock);
 939                 }
 940
 941                 /* ugh, clear_extent_buffer_dirty needs to lock the page */
 942                 btrfs_set_lock_blocking(buf);
 943                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
 944                                           buf);
 945         }
 946         return 0;
 947 }
 948
 949 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
 950                         u32 stripesize, struct btrfs_root *root,
 951                         struct btrfs_fs_info *fs_info,
 952                         u64 objectid)
 953 {
 954         root->node = NULL;
 955         root->commit_root = NULL;
 956         root->sectorsize = sectorsize;
 957         root->nodesize = nodesize;
 958         root->leafsize = leafsize;
 959         root->stripesize = stripesize;
 960         root->ref_cows = 0;
 961         root->track_dirty = 0;
 962         root->in_radix = 0;
 963         root->orphan_item_inserted = 0;
 964         root->orphan_cleanup_state = 0;
 965
 966         root->fs_info = fs_info;
 967         root->objectid = objectid;
 968         root->last_trans = 0;
 969         root->highest_objectid = 0;
 970         root->name = NULL;
 971         root->in_sysfs = 0;
 972         root->inode_tree = RB_ROOT;
 973         root->block_rsv = NULL;
 974         root->orphan_block_rsv = NULL;
 975
 976         INIT_LIST_HEAD(&root->dirty_list);
 977         INIT_LIST_HEAD(&root->orphan_list);
 978         INIT_LIST_HEAD(&root->root_list);
 979         spin_lock_init(&root->node_lock);
 980         spin_lock_init(&root->orphan_lock);
 981         spin_lock_init(&root->inode_lock);
 982         spin_lock_init(&root->accounting_lock);
 983         mutex_init(&root->objectid_mutex);
 984         mutex_init(&root->log_mutex);
 985         init_waitqueue_head(&root->log_writer_wait);
 986         init_waitqueue_head(&root->log_commit_wait[0]);
 987         init_waitqueue_head(&root->log_commit_wait[1]);
 988         atomic_set(&root->log_commit[0], 0);
 989         atomic_set(&root->log_commit[1], 0);
 990         atomic_set(&root->log_writers, 0);
 991         root->log_batch = 0;
 992         root->log_transid = 0;
 993         root->last_log_commit = 0;
 994         extent_io_tree_init(&root->dirty_log_pages,
 995                              fs_info->btree_inode->i_mapping, GFP_NOFS);
 996
 997         memset(&root->root_key, 0, sizeof(root->root_key));
 998         memset(&root->root_item, 0, sizeof(root->root_item));
 999         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1000         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1001         root->defrag_trans_start = fs_info->generation;
1002         init_completion(&root->kobj_unregister);
1003         root->defrag_running = 0;
1004         root->root_key.objectid = objectid;
1005         root->anon_super.s_root = NULL;
1006         root->anon_super.s_dev = 0;
1007         INIT_LIST_HEAD(&root->anon_super.s_list);
1008         INIT_LIST_HEAD(&root->anon_super.s_instances);
1009         init_rwsem(&root->anon_super.s_umount);
1010
1011         return 0;
1012 }
1013
1014 static int find_and_setup_root(struct btrfs_root *tree_root,
1015                                struct btrfs_fs_info *fs_info,
1016                                u64 objectid,
1017                                struct btrfs_root *root)
1018 {
1019         int ret;
1020         u32 blocksize;
1021         u64 generation;
1022
1023         __setup_root(tree_root->nodesize, tree_root->leafsize,
1024                      tree_root->sectorsize, tree_root->stripesize,
1025                      root, fs_info, objectid);
1026         ret = btrfs_find_last_root(tree_root, objectid,
1027                                    &root->root_item, &root->root_key);
1028         if (ret > 0)
1029                 return -ENOENT;
1030         BUG_ON(ret);
1031
1032         generation = btrfs_root_generation(&root->root_item);
1033         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1034         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1035                                      blocksize, generation);
1036         if (!root->node || !btrfs_buffer_uptodate(root->node, generation)) {
1037                 free_extent_buffer(root->node);
1038                 return -EIO;
1039         }
1040         root->commit_root = btrfs_root_node(root);
1041         return 0;
1042 }
1043
1044 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1045                                          struct btrfs_fs_info *fs_info)
1046 {
1047         struct btrfs_root *root;
1048         struct btrfs_root *tree_root = fs_info->tree_root;
1049         struct extent_buffer *leaf;
1050
1051         root = kzalloc(sizeof(*root), GFP_NOFS);
1052         if (!root)
1053                 return ERR_PTR(-ENOMEM);
1054
1055         __setup_root(tree_root->nodesize, tree_root->leafsize,
1056                      tree_root->sectorsize, tree_root->stripesize,
1057                      root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1058
1059         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1060         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1061         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1062         /*
1063          * log trees do not get reference counted because they go away
1064          * before a real commit is actually done.  They do store pointers
1065          * to file data extents, and those reference counts still get
1066          * updated (along with back refs to the log tree).
1067          */
1068         root->ref_cows = 0;
1069
1070         leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1071                                       BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
1072         if (IS_ERR(leaf)) {
1073                 kfree(root);
1074                 return ERR_CAST(leaf);
1075         }
1076
1077         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1078         btrfs_set_header_bytenr(leaf, leaf->start);
1079         btrfs_set_header_generation(leaf, trans->transid);
1080         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1081         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1082         root->node = leaf;
1083
1084         write_extent_buffer(root->node, root->fs_info->fsid,
1085                             (unsigned long)btrfs_header_fsid(root->node),
1086                             BTRFS_FSID_SIZE);
1087         btrfs_mark_buffer_dirty(root->node);
1088         btrfs_tree_unlock(root->node);
1089         return root;
1090 }
1091
1092 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1093                              struct btrfs_fs_info *fs_info)
1094 {
1095         struct btrfs_root *log_root;
1096
1097         log_root = alloc_log_tree(trans, fs_info);
1098         if (IS_ERR(log_root))
1099                 return PTR_ERR(log_root);
1100         WARN_ON(fs_info->log_root_tree);
1101         fs_info->log_root_tree = log_root;
1102         return 0;
1103 }
1104
1105 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1106                        struct btrfs_root *root)
1107 {
1108         struct btrfs_root *log_root;
1109         struct btrfs_inode_item *inode_item;
1110
1111         log_root = alloc_log_tree(trans, root->fs_info);
1112         if (IS_ERR(log_root))
1113                 return PTR_ERR(log_root);
1114
1115         log_root->last_trans = trans->transid;
1116         log_root->root_key.offset = root->root_key.objectid;
1117
1118         inode_item = &log_root->root_item.inode;
1119         inode_item->generation = cpu_to_le64(1);
1120         inode_item->size = cpu_to_le64(3);
1121         inode_item->nlink = cpu_to_le32(1);
1122         inode_item->nbytes = cpu_to_le64(root->leafsize);
1123         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1124
1125         btrfs_set_root_node(&log_root->root_item, log_root->node);
1126
1127         WARN_ON(root->log_root);
1128         root->log_root = log_root;
1129         root->log_transid = 0;
1130         root->last_log_commit = 0;
1131         return 0;
1132 }
1133
1134 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1135                                                struct btrfs_key *location)
1136 {
1137         struct btrfs_root *root;
1138         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1139         struct btrfs_path *path;
1140         struct extent_buffer *l;
1141         u64 generation;
1142         u32 blocksize;
1143         int ret = 0;
1144
1145         root = kzalloc(sizeof(*root), GFP_NOFS);
1146         if (!root)
1147                 return ERR_PTR(-ENOMEM);
1148         if (location->offset == (u64)-1) {
1149                 ret = find_and_setup_root(tree_root, fs_info,
1150                                           location->objectid, root);
1151                 if (ret) {
1152                         kfree(root);
1153                         return ERR_PTR(ret);
1154                 }
1155                 goto out;
1156         }
1157
1158         __setup_root(tree_root->nodesize, tree_root->leafsize,
1159                      tree_root->sectorsize, tree_root->stripesize,
1160                      root, fs_info, location->objectid);
1161
1162         path = btrfs_alloc_path();
1163         BUG_ON(!path);
1164         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1165         if (ret == 0) {
1166                 l = path->nodes[0];
1167                 read_extent_buffer(l, &root->root_item,
1168                                 btrfs_item_ptr_offset(l, path->slots[0]),
1169                                 sizeof(root->root_item));
1170                 memcpy(&root->root_key, location, sizeof(*location));
1171         }
1172         btrfs_free_path(path);
1173         if (ret) {
1174                 kfree(root);
1175                 if (ret > 0)
1176                         ret = -ENOENT;
1177                 return ERR_PTR(ret);
1178         }
1179
1180         generation = btrfs_root_generation(&root->root_item);
1181         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1182         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1183                                      blocksize, generation);
1184         root->commit_root = btrfs_root_node(root);
1185         BUG_ON(!root->node);
1186 out:
1187         if (location->objectid != BTRFS_TREE_LOG_OBJECTID)
1188                 root->ref_cows = 1;
1189
1190         return root;
1191 }
1192
1193 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1194                                         u64 root_objectid)
1195 {
1196         struct btrfs_root *root;
1197
1198         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1199                 return fs_info->tree_root;
1200         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1201                 return fs_info->extent_root;
1202
1203         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1204                                  (unsigned long)root_objectid);
1205         return root;
1206 }
1207
1208 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1209                                               struct btrfs_key *location)
1210 {
1211         struct btrfs_root *root;
1212         int ret;
1213
1214         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1215                 return fs_info->tree_root;
1216         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1217                 return fs_info->extent_root;
1218         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1219                 return fs_info->chunk_root;
1220         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1221                 return fs_info->dev_root;
1222         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1223                 return fs_info->csum_root;
1224 again:
1225         spin_lock(&fs_info->fs_roots_radix_lock);
1226         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1227                                  (unsigned long)location->objectid);
1228         spin_unlock(&fs_info->fs_roots_radix_lock);
1229         if (root)
1230                 return root;
1231
1232         root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1233         if (IS_ERR(root))
1234                 return root;
1235
1236         set_anon_super(&root->anon_super, NULL);
1237
1238         if (btrfs_root_refs(&root->root_item) == 0) {
1239                 ret = -ENOENT;
1240                 goto fail;
1241         }
1242
1243         ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1244         if (ret < 0)
1245                 goto fail;
1246         if (ret == 0)
1247                 root->orphan_item_inserted = 1;
1248
1249         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1250         if (ret)
1251                 goto fail;
1252
1253         spin_lock(&fs_info->fs_roots_radix_lock);
1254         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1255                                 (unsigned long)root->root_key.objectid,
1256                                 root);
1257         if (ret == 0)
1258                 root->in_radix = 1;
1259
1260         spin_unlock(&fs_info->fs_roots_radix_lock);
1261         radix_tree_preload_end();
1262         if (ret) {
1263                 if (ret == -EEXIST) {
1264                         free_fs_root(root);
1265                         goto again;
1266                 }
1267                 goto fail;
1268         }
1269
1270         ret = btrfs_find_dead_roots(fs_info->tree_root,
1271                                     root->root_key.objectid);
1272         WARN_ON(ret);
1273         return root;
1274 fail:
1275         free_fs_root(root);
1276         return ERR_PTR(ret);
1277 }
1278
1279 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1280                                       struct btrfs_key *location,
1281                                       const char *name, int namelen)
1282 {
1283         return btrfs_read_fs_root_no_name(fs_info, location);
1284 #if 0
1285         struct btrfs_root *root;
1286         int ret;
1287
1288         root = btrfs_read_fs_root_no_name(fs_info, location);
1289         if (!root)
1290                 return NULL;
1291
1292         if (root->in_sysfs)
1293                 return root;
1294
1295         ret = btrfs_set_root_name(root, name, namelen);
1296         if (ret) {
1297                 free_extent_buffer(root->node);
1298                 kfree(root);
1299                 return ERR_PTR(ret);
1300         }
1301
1302         ret = btrfs_sysfs_add_root(root);
1303         if (ret) {
1304                 free_extent_buffer(root->node);
1305                 kfree(root->name);
1306                 kfree(root);
1307                 return ERR_PTR(ret);
1308         }
1309         root->in_sysfs = 1;
1310         return root;
1311 #endif
1312 }
1313
1314 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1315 {
1316         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1317         int ret = 0;
1318         struct btrfs_device *device;
1319         struct backing_dev_info *bdi;
1320
1321         list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1322                 if (!device->bdev)
1323                         continue;
1324                 bdi = blk_get_backing_dev_info(device->bdev);
1325                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1326                         ret = 1;
1327                         break;
1328                 }
1329         }
1330         return ret;
1331 }
1332
1333 /*
1334  * this unplugs every device on the box, and it is only used when page
1335  * is null
1336  */
1337 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1338 {
1339         struct btrfs_device *device;
1340         struct btrfs_fs_info *info;
1341
1342         info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1343         list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1344                 if (!device->bdev)
1345                         continue;
1346
1347                 bdi = blk_get_backing_dev_info(device->bdev);
1348                 if (bdi->unplug_io_fn)
1349                         bdi->unplug_io_fn(bdi, page);
1350         }
1351 }
1352
1353 static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1354 {
1355         struct inode *inode;
1356         struct extent_map_tree *em_tree;
1357         struct extent_map *em;
1358         struct address_space *mapping;
1359         u64 offset;
1360
1361         /* the generic O_DIRECT read code does this */
1362         if (1 || !page) {
1363                 __unplug_io_fn(bdi, page);
1364                 return;
1365         }
1366
1367         /*
1368          * page->mapping may change at any time.  Get a consistent copy
1369          * and use that for everything below
1370          */
1371         smp_mb();
1372         mapping = page->mapping;
1373         if (!mapping)
1374                 return;
1375
1376         inode = mapping->host;
1377
1378         /*
1379          * don't do the expensive searching for a small number of
1380          * devices
1381          */
1382         if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1383                 __unplug_io_fn(bdi, page);
1384                 return;
1385         }
1386
1387         offset = page_offset(page);
1388
1389         em_tree = &BTRFS_I(inode)->extent_tree;
1390         read_lock(&em_tree->lock);
1391         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1392         read_unlock(&em_tree->lock);
1393         if (!em) {
1394                 __unplug_io_fn(bdi, page);
1395                 return;
1396         }
1397
1398         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1399                 free_extent_map(em);
1400                 __unplug_io_fn(bdi, page);
1401                 return;
1402         }
1403         offset = offset - em->start;
1404         btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1405                           em->block_start + offset, page);
1406         free_extent_map(em);
1407 }
1408
1409 /*
1410  * If this fails, caller must call bdi_destroy() to get rid of the
1411  * bdi again.
1412  */
1413 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1414 {
1415         int err;
1416
1417         bdi->capabilities = BDI_CAP_MAP_COPY;
1418         err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1419         if (err)
1420                 return err;
1421
1422         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1423         bdi->unplug_io_fn       = btrfs_unplug_io_fn;
1424         bdi->unplug_io_data     = info;
1425         bdi->congested_fn       = btrfs_congested_fn;
1426         bdi->congested_data     = info;
1427         return 0;
1428 }
1429
1430 static int bio_ready_for_csum(struct bio *bio)
1431 {
1432         u64 length = 0;
1433         u64 buf_len = 0;
1434         u64 start = 0;
1435         struct page *page;
1436         struct extent_io_tree *io_tree = NULL;
1437         struct bio_vec *bvec;
1438         int i;
1439         int ret;
1440
1441         bio_for_each_segment(bvec, bio, i) {
1442                 page = bvec->bv_page;
1443                 if (page->private == EXTENT_PAGE_PRIVATE) {
1444                         length += bvec->bv_len;
1445                         continue;
1446                 }
1447                 if (!page->private) {
1448                         length += bvec->bv_len;
1449                         continue;
1450                 }
1451                 length = bvec->bv_len;
1452                 buf_len = page->private >> 2;
1453                 start = page_offset(page) + bvec->bv_offset;
1454                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1455         }
1456         /* are we fully contained in this bio? */
1457         if (buf_len <= length)
1458                 return 1;
1459
1460         ret = extent_range_uptodate(io_tree, start + length,
1461                                     start + buf_len - 1);
1462         return ret;
1463 }
1464
1465 /*
1466  * called by the kthread helper functions to finally call the bio end_io
1467  * functions.  This is where read checksum verification actually happens
1468  */
1469 static void end_workqueue_fn(struct btrfs_work *work)
1470 {
1471         struct bio *bio;
1472         struct end_io_wq *end_io_wq;
1473         struct btrfs_fs_info *fs_info;
1474         int error;
1475
1476         end_io_wq = container_of(work, struct end_io_wq, work);
1477         bio = end_io_wq->bio;
1478         fs_info = end_io_wq->info;
1479
1480         /* metadata bio reads are special because the whole tree block must
1481          * be checksummed at once.  This makes sure the entire block is in
1482          * ram and up to date before trying to verify things.  For
1483          * blocksize <= pagesize, it is basically a noop
1484          */
1485         if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata &&
1486             !bio_ready_for_csum(bio)) {
1487                 btrfs_queue_worker(&fs_info->endio_meta_workers,
1488                                    &end_io_wq->work);
1489                 return;
1490         }
1491         error = end_io_wq->error;
1492         bio->bi_private = end_io_wq->private;
1493         bio->bi_end_io = end_io_wq->end_io;
1494         kfree(end_io_wq);
1495         bio_endio(bio, error);
1496 }
1497
1498 static int cleaner_kthread(void *arg)
1499 {
1500         struct btrfs_root *root = arg;
1501
1502         do {
1503                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1504
1505                 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1506                     mutex_trylock(&root->fs_info->cleaner_mutex)) {
1507                         btrfs_run_delayed_iputs(root);
1508                         btrfs_clean_old_snapshots(root);
1509                         mutex_unlock(&root->fs_info->cleaner_mutex);
1510                 }
1511
1512                 if (freezing(current)) {
1513                         refrigerator();
1514                 } else {
1515                         set_current_state(TASK_INTERRUPTIBLE);
1516                         if (!kthread_should_stop())
1517                                 schedule();
1518                         __set_current_state(TASK_RUNNING);
1519                 }
1520         } while (!kthread_should_stop());
1521         return 0;
1522 }
1523
1524 static int transaction_kthread(void *arg)
1525 {
1526         struct btrfs_root *root = arg;
1527         struct btrfs_trans_handle *trans;
1528         struct btrfs_transaction *cur;
1529         u64 transid;
1530         unsigned long now;
1531         unsigned long delay;
1532         int ret;
1533
1534         do {
1535                 delay = HZ * 30;
1536                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1537                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1538
1539                 spin_lock(&root->fs_info->new_trans_lock);
1540                 cur = root->fs_info->running_transaction;
1541                 if (!cur) {
1542                         spin_unlock(&root->fs_info->new_trans_lock);
1543                         goto sleep;
1544                 }
1545
1546                 now = get_seconds();
1547                 if (!cur->blocked &&
1548                     (now < cur->start_time || now - cur->start_time < 30)) {
1549                         spin_unlock(&root->fs_info->new_trans_lock);
1550                         delay = HZ * 5;
1551                         goto sleep;
1552                 }
1553                 transid = cur->transid;
1554                 spin_unlock(&root->fs_info->new_trans_lock);
1555
1556                 trans = btrfs_join_transaction(root, 1);
1557                 BUG_ON(IS_ERR(trans));
1558                 if (transid == trans->transid) {
1559                         ret = btrfs_commit_transaction(trans, root);
1560                         BUG_ON(ret);
1561                 } else {
1562                         btrfs_end_transaction(trans, root);
1563                 }
1564 sleep:
1565                 wake_up_process(root->fs_info->cleaner_kthread);
1566                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1567
1568                 if (freezing(current)) {
1569                         refrigerator();
1570                 } else {
1571                         set_current_state(TASK_INTERRUPTIBLE);
1572                         if (!kthread_should_stop() &&
1573                             !btrfs_transaction_blocked(root->fs_info))
1574                                 schedule_timeout(delay);
1575                         __set_current_state(TASK_RUNNING);
1576                 }
1577         } while (!kthread_should_stop());
1578         return 0;
1579 }
1580
1581 struct btrfs_root *open_ctree(struct super_block *sb,
1582                               struct btrfs_fs_devices *fs_devices,
1583                               char *options)
1584 {
1585         u32 sectorsize;
1586         u32 nodesize;
1587         u32 leafsize;
1588         u32 blocksize;
1589         u32 stripesize;
1590         u64 generation;
1591         u64 features;
1592         struct btrfs_key location;
1593         struct buffer_head *bh;
1594         struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1595                                                  GFP_NOFS);
1596         struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1597                                                  GFP_NOFS);
1598         struct btrfs_root *tree_root = btrfs_sb(sb);
1599         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1600         struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1601                                                 GFP_NOFS);
1602         struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1603                                               GFP_NOFS);
1604         struct btrfs_root *log_tree_root;
1605
1606         int ret;
1607         int err = -EINVAL;
1608
1609         struct btrfs_super_block *disk_super;
1610
1611         if (!extent_root || !tree_root || !fs_info ||
1612             !chunk_root || !dev_root || !csum_root) {
1613                 err = -ENOMEM;
1614                 goto fail;
1615         }
1616
1617         ret = init_srcu_struct(&fs_info->subvol_srcu);
1618         if (ret) {
1619                 err = ret;
1620                 goto fail;
1621         }
1622
1623         ret = setup_bdi(fs_info, &fs_info->bdi);
1624         if (ret) {
1625                 err = ret;
1626                 goto fail_srcu;
1627         }
1628
1629         fs_info->btree_inode = new_inode(sb);
1630         if (!fs_info->btree_inode) {
1631                 err = -ENOMEM;
1632                 goto fail_bdi;
1633         }
1634
1635         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1636         INIT_LIST_HEAD(&fs_info->trans_list);
1637         INIT_LIST_HEAD(&fs_info->dead_roots);
1638         INIT_LIST_HEAD(&fs_info->delayed_iputs);
1639         INIT_LIST_HEAD(&fs_info->hashers);
1640         INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1641         INIT_LIST_HEAD(&fs_info->ordered_operations);
1642         INIT_LIST_HEAD(&fs_info->caching_block_groups);
1643         spin_lock_init(&fs_info->delalloc_lock);
1644         spin_lock_init(&fs_info->new_trans_lock);
1645         spin_lock_init(&fs_info->ref_cache_lock);
1646         spin_lock_init(&fs_info->fs_roots_radix_lock);
1647         spin_lock_init(&fs_info->delayed_iput_lock);
1648
1649         init_completion(&fs_info->kobj_unregister);
1650         fs_info->tree_root = tree_root;
1651         fs_info->extent_root = extent_root;
1652         fs_info->csum_root = csum_root;
1653         fs_info->chunk_root = chunk_root;
1654         fs_info->dev_root = dev_root;
1655         fs_info->fs_devices = fs_devices;
1656         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1657         INIT_LIST_HEAD(&fs_info->space_info);
1658         btrfs_mapping_init(&fs_info->mapping_tree);
1659         btrfs_init_block_rsv(&fs_info->global_block_rsv);
1660         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1661         btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1662         btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1663         btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1664         INIT_LIST_HEAD(&fs_info->durable_block_rsv_list);
1665         mutex_init(&fs_info->durable_block_rsv_mutex);
1666         atomic_set(&fs_info->nr_async_submits, 0);
1667         atomic_set(&fs_info->async_delalloc_pages, 0);
1668         atomic_set(&fs_info->async_submit_draining, 0);
1669         atomic_set(&fs_info->nr_async_bios, 0);
1670         fs_info->sb = sb;
1671         fs_info->max_inline = 8192 * 1024;
1672         fs_info->metadata_ratio = 0;
1673
1674         fs_info->thread_pool_size = min_t(unsigned long,
1675                                           num_online_cpus() + 2, 8);
1676
1677         INIT_LIST_HEAD(&fs_info->ordered_extents);
1678         spin_lock_init(&fs_info->ordered_extent_lock);
1679
1680         sb->s_blocksize = 4096;
1681         sb->s_blocksize_bits = blksize_bits(4096);
1682         sb->s_bdi = &fs_info->bdi;
1683
1684         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1685         fs_info->btree_inode->i_nlink = 1;
1686         /*
1687          * we set the i_size on the btree inode to the max possible int.
1688          * the real end of the address space is determined by all of
1689          * the devices in the system
1690          */
1691         fs_info->btree_inode->i_size = OFFSET_MAX;
1692         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1693         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1694
1695         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
1696         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1697                              fs_info->btree_inode->i_mapping,
1698                              GFP_NOFS);
1699         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1700                              GFP_NOFS);
1701
1702         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1703
1704         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1705         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1706                sizeof(struct btrfs_key));
1707         BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
1708         insert_inode_hash(fs_info->btree_inode);
1709
1710         spin_lock_init(&fs_info->block_group_cache_lock);
1711         fs_info->block_group_cache_tree = RB_ROOT;
1712
1713         extent_io_tree_init(&fs_info->freed_extents[0],
1714                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1715         extent_io_tree_init(&fs_info->freed_extents[1],
1716                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1717         fs_info->pinned_extents = &fs_info->freed_extents[0];
1718         fs_info->do_barriers = 1;
1719
1720
1721         mutex_init(&fs_info->trans_mutex);
1722         mutex_init(&fs_info->ordered_operations_mutex);
1723         mutex_init(&fs_info->tree_log_mutex);
1724         mutex_init(&fs_info->chunk_mutex);
1725         mutex_init(&fs_info->transaction_kthread_mutex);
1726         mutex_init(&fs_info->cleaner_mutex);
1727         mutex_init(&fs_info->volume_mutex);
1728         init_rwsem(&fs_info->extent_commit_sem);
1729         init_rwsem(&fs_info->cleanup_work_sem);
1730         init_rwsem(&fs_info->subvol_sem);
1731
1732         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1733         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1734
1735         init_waitqueue_head(&fs_info->transaction_throttle);
1736         init_waitqueue_head(&fs_info->transaction_wait);
1737         init_waitqueue_head(&fs_info->transaction_blocked_wait);
1738         init_waitqueue_head(&fs_info->async_submit_wait);
1739
1740         __setup_root(4096, 4096, 4096, 4096, tree_root,
1741                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1742
1743         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1744         if (!bh) {
1745                 err = -EINVAL;
1746                 goto fail_iput;
1747         }
1748
1749         memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1750         memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1751                sizeof(fs_info->super_for_commit));
1752         brelse(bh);
1753
1754         memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1755
1756         disk_super = &fs_info->super_copy;
1757         if (!btrfs_super_root(disk_super))
1758                 goto fail_iput;
1759
1760         /* check FS state, whether FS is broken. */
1761         fs_info->fs_state |= btrfs_super_flags(disk_super);
1762
1763         btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
1764
1765         ret = btrfs_parse_options(tree_root, options);
1766         if (ret) {
1767                 err = ret;
1768                 goto fail_iput;
1769         }
1770
1771         features = btrfs_super_incompat_flags(disk_super) &
1772                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1773         if (features) {
1774                 printk(KERN_ERR "BTRFS: couldn't mount because of "
1775                        "unsupported optional features (%Lx).\n",
1776                        (unsigned long long)features);
1777                 err = -EINVAL;
1778                 goto fail_iput;
1779         }
1780
1781         features = btrfs_super_incompat_flags(disk_super);
1782         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
1783         if (tree_root->fs_info->compress_type & BTRFS_COMPRESS_LZO)
1784                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
1785         btrfs_set_super_incompat_flags(disk_super, features);
1786
1787         features = btrfs_super_compat_ro_flags(disk_super) &
1788                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1789         if (!(sb->s_flags & MS_RDONLY) && features) {
1790                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1791                        "unsupported option features (%Lx).\n",
1792                        (unsigned long long)features);
1793                 err = -EINVAL;
1794                 goto fail_iput;
1795         }
1796
1797         btrfs_init_workers(&fs_info->generic_worker,
1798                            "genwork", 1, NULL);
1799
1800         btrfs_init_workers(&fs_info->workers, "worker",
1801                            fs_info->thread_pool_size,
1802                            &fs_info->generic_worker);
1803
1804         btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1805                            fs_info->thread_pool_size,
1806                            &fs_info->generic_worker);
1807
1808         btrfs_init_workers(&fs_info->submit_workers, "submit",
1809                            min_t(u64, fs_devices->num_devices,
1810                            fs_info->thread_pool_size),
1811                            &fs_info->generic_worker);
1812
1813         /* a higher idle thresh on the submit workers makes it much more
1814          * likely that bios will be send down in a sane order to the
1815          * devices
1816          */
1817         fs_info->submit_workers.idle_thresh = 64;
1818
1819         fs_info->workers.idle_thresh = 16;
1820         fs_info->workers.ordered = 1;
1821
1822         fs_info->delalloc_workers.idle_thresh = 2;
1823         fs_info->delalloc_workers.ordered = 1;
1824
1825         btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
1826                            &fs_info->generic_worker);
1827         btrfs_init_workers(&fs_info->endio_workers, "endio",
1828                            fs_info->thread_pool_size,
1829                            &fs_info->generic_worker);
1830         btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1831                            fs_info->thread_pool_size,
1832                            &fs_info->generic_worker);
1833         btrfs_init_workers(&fs_info->endio_meta_write_workers,
1834                            "endio-meta-write", fs_info->thread_pool_size,
1835                            &fs_info->generic_worker);
1836         btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1837                            fs_info->thread_pool_size,
1838                            &fs_info->generic_worker);
1839         btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
1840                            1, &fs_info->generic_worker);
1841
1842         /*
1843          * endios are largely parallel and should have a very
1844          * low idle thresh
1845          */
1846         fs_info->endio_workers.idle_thresh = 4;
1847         fs_info->endio_meta_workers.idle_thresh = 4;
1848
1849         fs_info->endio_write_workers.idle_thresh = 2;
1850         fs_info->endio_meta_write_workers.idle_thresh = 2;
1851
1852         btrfs_start_workers(&fs_info->workers, 1);
1853         btrfs_start_workers(&fs_info->generic_worker, 1);
1854         btrfs_start_workers(&fs_info->submit_workers, 1);
1855         btrfs_start_workers(&fs_info->delalloc_workers, 1);
1856         btrfs_start_workers(&fs_info->fixup_workers, 1);
1857         btrfs_start_workers(&fs_info->endio_workers, 1);
1858         btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1859         btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1860         btrfs_start_workers(&fs_info->endio_write_workers, 1);
1861         btrfs_start_workers(&fs_info->endio_freespace_worker, 1);
1862
1863         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1864         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1865                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1866
1867         nodesize = btrfs_super_nodesize(disk_super);
1868         leafsize = btrfs_super_leafsize(disk_super);
1869         sectorsize = btrfs_super_sectorsize(disk_super);
1870         stripesize = btrfs_super_stripesize(disk_super);
1871         tree_root->nodesize = nodesize;
1872         tree_root->leafsize = leafsize;
1873         tree_root->sectorsize = sectorsize;
1874         tree_root->stripesize = stripesize;
1875
1876         sb->s_blocksize = sectorsize;
1877         sb->s_blocksize_bits = blksize_bits(sectorsize);
1878
1879         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1880                     sizeof(disk_super->magic))) {
1881                 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
1882                 goto fail_sb_buffer;
1883         }
1884
1885         mutex_lock(&fs_info->chunk_mutex);
1886         ret = btrfs_read_sys_array(tree_root);
1887         mutex_unlock(&fs_info->chunk_mutex);
1888         if (ret) {
1889                 printk(KERN_WARNING "btrfs: failed to read the system "
1890                        "array on %s\n", sb->s_id);
1891                 goto fail_sb_buffer;
1892         }
1893
1894         blocksize = btrfs_level_size(tree_root,
1895                                      btrfs_super_chunk_root_level(disk_super));
1896         generation = btrfs_super_chunk_root_generation(disk_super);
1897
1898         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1899                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1900
1901         chunk_root->node = read_tree_block(chunk_root,
1902                                            btrfs_super_chunk_root(disk_super),
1903                                            blocksize, generation);
1904         BUG_ON(!chunk_root->node);
1905         if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
1906                 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
1907                        sb->s_id);
1908                 goto fail_chunk_root;
1909         }
1910         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
1911         chunk_root->commit_root = btrfs_root_node(chunk_root);
1912
1913         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1914            (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1915            BTRFS_UUID_SIZE);
1916
1917         mutex_lock(&fs_info->chunk_mutex);
1918         ret = btrfs_read_chunk_tree(chunk_root);
1919         mutex_unlock(&fs_info->chunk_mutex);
1920         if (ret) {
1921                 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
1922                        sb->s_id);
1923                 goto fail_chunk_root;
1924         }
1925
1926         btrfs_close_extra_devices(fs_devices);
1927
1928         blocksize = btrfs_level_size(tree_root,
1929                                      btrfs_super_root_level(disk_super));
1930         generation = btrfs_super_generation(disk_super);
1931
1932         tree_root->node = read_tree_block(tree_root,
1933                                           btrfs_super_root(disk_super),
1934                                           blocksize, generation);
1935         if (!tree_root->node)
1936                 goto fail_chunk_root;
1937         if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
1938                 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
1939                        sb->s_id);
1940                 goto fail_tree_root;
1941         }
1942         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
1943         tree_root->commit_root = btrfs_root_node(tree_root);
1944
1945         ret = find_and_setup_root(tree_root, fs_info,
1946                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1947         if (ret)
1948                 goto fail_tree_root;
1949         extent_root->track_dirty = 1;
1950
1951         ret = find_and_setup_root(tree_root, fs_info,
1952                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
1953         if (ret)
1954                 goto fail_extent_root;
1955         dev_root->track_dirty = 1;
1956
1957         ret = find_and_setup_root(tree_root, fs_info,
1958                                   BTRFS_CSUM_TREE_OBJECTID, csum_root);
1959         if (ret)
1960                 goto fail_dev_root;
1961
1962         csum_root->track_dirty = 1;
1963
1964         fs_info->generation = generation;
1965         fs_info->last_trans_committed = generation;
1966         fs_info->data_alloc_profile = (u64)-1;
1967         fs_info->metadata_alloc_profile = (u64)-1;
1968         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1969
1970         ret = btrfs_read_block_groups(extent_root);
1971         if (ret) {
1972                 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
1973                 goto fail_block_groups;
1974         }
1975
1976         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1977                                                "btrfs-cleaner");
1978         if (IS_ERR(fs_info->cleaner_kthread))
1979                 goto fail_block_groups;
1980
1981         fs_info->transaction_kthread = kthread_run(transaction_kthread,
1982                                                    tree_root,
1983                                                    "btrfs-transaction");
1984         if (IS_ERR(fs_info->transaction_kthread))
1985                 goto fail_cleaner;
1986
1987         if (!btrfs_test_opt(tree_root, SSD) &&
1988             !btrfs_test_opt(tree_root, NOSSD) &&
1989             !fs_info->fs_devices->rotating) {
1990                 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
1991                        "mode\n");
1992                 btrfs_set_opt(fs_info->mount_opt, SSD);
1993         }
1994
1995         /* do not make disk changes in broken FS */
1996         if (btrfs_super_log_root(disk_super) != 0 &&
1997             !(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) {
1998                 u64 bytenr = btrfs_super_log_root(disk_super);
1999
2000                 if (fs_devices->rw_devices == 0) {
2001                         printk(KERN_WARNING "Btrfs log replay required "
2002                                "on RO media\n");
2003                         err = -EIO;
2004                         goto fail_trans_kthread;
2005                 }
2006                 blocksize =
2007                      btrfs_level_size(tree_root,
2008                                       btrfs_super_log_root_level(disk_super));
2009
2010                 log_tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
2011                 if (!log_tree_root) {
2012                         err = -ENOMEM;
2013                         goto fail_trans_kthread;
2014                 }
2015
2016                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2017                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2018
2019                 log_tree_root->node = read_tree_block(tree_root, bytenr,
2020                                                       blocksize,
2021                                                       generation + 1);
2022                 ret = btrfs_recover_log_trees(log_tree_root);
2023                 BUG_ON(ret);
2024
2025                 if (sb->s_flags & MS_RDONLY) {
2026                         ret =  btrfs_commit_super(tree_root);
2027                         BUG_ON(ret);
2028                 }
2029         }
2030
2031         ret = btrfs_find_orphan_roots(tree_root);
2032         BUG_ON(ret);
2033
2034         if (!(sb->s_flags & MS_RDONLY)) {
2035                 ret = btrfs_cleanup_fs_roots(fs_info);
2036                 BUG_ON(ret);
2037
2038                 ret = btrfs_recover_relocation(tree_root);
2039                 if (ret < 0) {
2040                         printk(KERN_WARNING
2041                                "btrfs: failed to recover relocation\n");
2042                         err = -EINVAL;
2043                         goto fail_trans_kthread;
2044                 }
2045         }
2046
2047         location.objectid = BTRFS_FS_TREE_OBJECTID;
2048         location.type = BTRFS_ROOT_ITEM_KEY;
2049         location.offset = (u64)-1;
2050
2051         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2052         if (!fs_info->fs_root)
2053                 goto fail_trans_kthread;
2054         if (IS_ERR(fs_info->fs_root)) {
2055                 err = PTR_ERR(fs_info->fs_root);
2056                 goto fail_trans_kthread;
2057         }
2058
2059         if (!(sb->s_flags & MS_RDONLY)) {
2060                 down_read(&fs_info->cleanup_work_sem);
2061                 btrfs_orphan_cleanup(fs_info->fs_root);
2062                 btrfs_orphan_cleanup(fs_info->tree_root);
2063                 up_read(&fs_info->cleanup_work_sem);
2064         }
2065
2066         return tree_root;
2067
2068 fail_trans_kthread:
2069         kthread_stop(fs_info->transaction_kthread);
2070 fail_cleaner:
2071         kthread_stop(fs_info->cleaner_kthread);
2072
2073         /*
2074          * make sure we're done with the btree inode before we stop our
2075          * kthreads
2076          */
2077         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2078         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2079
2080 fail_block_groups:
2081         btrfs_free_block_groups(fs_info);
2082         free_extent_buffer(csum_root->node);
2083         free_extent_buffer(csum_root->commit_root);
2084 fail_dev_root:
2085         free_extent_buffer(dev_root->node);
2086         free_extent_buffer(dev_root->commit_root);
2087 fail_extent_root:
2088         free_extent_buffer(extent_root->node);
2089         free_extent_buffer(extent_root->commit_root);
2090 fail_tree_root:
2091         free_extent_buffer(tree_root->node);
2092         free_extent_buffer(tree_root->commit_root);
2093 fail_chunk_root:
2094         free_extent_buffer(chunk_root->node);
2095         free_extent_buffer(chunk_root->commit_root);
2096 fail_sb_buffer:
2097         btrfs_stop_workers(&fs_info->generic_worker);
2098         btrfs_stop_workers(&fs_info->fixup_workers);
2099         btrfs_stop_workers(&fs_info->delalloc_workers);
2100         btrfs_stop_workers(&fs_info->workers);
2101         btrfs_stop_workers(&fs_info->endio_workers);
2102         btrfs_stop_workers(&fs_info->endio_meta_workers);
2103         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2104         btrfs_stop_workers(&fs_info->endio_write_workers);
2105         btrfs_stop_workers(&fs_info->endio_freespace_worker);
2106         btrfs_stop_workers(&fs_info->submit_workers);
2107 fail_iput:
2108         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2109         iput(fs_info->btree_inode);
2110
2111         btrfs_close_devices(fs_info->fs_devices);
2112         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2113 fail_bdi:
2114         bdi_destroy(&fs_info->bdi);
2115 fail_srcu:
2116         cleanup_srcu_struct(&fs_info->subvol_srcu);
2117 fail:
2118         kfree(extent_root);
2119         kfree(tree_root);
2120         kfree(fs_info);
2121         kfree(chunk_root);
2122         kfree(dev_root);
2123         kfree(csum_root);
2124         return ERR_PTR(err);
2125 }
2126
2127 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2128 {
2129         char b[BDEVNAME_SIZE];
2130
2131         if (uptodate) {
2132                 set_buffer_uptodate(bh);
2133         } else {
2134                 if (printk_ratelimit()) {
2135                         printk(KERN_WARNING "lost page write due to "
2136                                         "I/O error on %s\n",
2137                                        bdevname(bh->b_bdev, b));
2138                 }
2139                 /* note, we dont' set_buffer_write_io_error because we have
2140                  * our own ways of dealing with the IO errors
2141                  */
2142                 clear_buffer_uptodate(bh);
2143         }
2144         unlock_buffer(bh);
2145         put_bh(bh);
2146 }
2147
2148 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2149 {
2150         struct buffer_head *bh;
2151         struct buffer_head *latest = NULL;
2152         struct btrfs_super_block *super;
2153         int i;
2154         u64 transid = 0;
2155         u64 bytenr;
2156
2157         /* we would like to check all the supers, but that would make
2158          * a btrfs mount succeed after a mkfs from a different FS.
2159          * So, we need to add a special mount option to scan for
2160          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2161          */
2162         for (i = 0; i < 1; i++) {
2163                 bytenr = btrfs_sb_offset(i);
2164                 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2165                         break;
2166                 bh = __bread(bdev, bytenr / 4096, 4096);
2167                 if (!bh)
2168                         continue;
2169
2170                 super = (struct btrfs_super_block *)bh->b_data;
2171                 if (btrfs_super_bytenr(super) != bytenr ||
2172                     strncmp((char *)(&super->magic), BTRFS_MAGIC,
2173                             sizeof(super->magic))) {
2174                         brelse(bh);
2175                         continue;
2176                 }
2177
2178                 if (!latest || btrfs_super_generation(super) > transid) {
2179                         brelse(latest);
2180                         latest = bh;
2181                         transid = btrfs_super_generation(super);
2182                 } else {
2183                         brelse(bh);
2184                 }
2185         }
2186         return latest;
2187 }
2188
2189 /*
2190  * this should be called twice, once with wait == 0 and
2191  * once with wait == 1.  When wait == 0 is done, all the buffer heads
2192  * we write are pinned.
2193  *
2194  * They are released when wait == 1 is done.
2195  * max_mirrors must be the same for both runs, and it indicates how
2196  * many supers on this one device should be written.
2197  *
2198  * max_mirrors == 0 means to write them all.
2199  */
2200 static int write_dev_supers(struct btrfs_device *device,
2201                             struct btrfs_super_block *sb,
2202                             int do_barriers, int wait, int max_mirrors)
2203 {
2204         struct buffer_head *bh;
2205         int i;
2206         int ret;
2207         int errors = 0;
2208         u32 crc;
2209         u64 bytenr;
2210         int last_barrier = 0;
2211
2212         if (max_mirrors == 0)
2213                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2214
2215         /* make sure only the last submit_bh does a barrier */
2216         if (do_barriers) {
2217                 for (i = 0; i < max_mirrors; i++) {
2218                         bytenr = btrfs_sb_offset(i);
2219                         if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2220                             device->total_bytes)
2221                                 break;
2222                         last_barrier = i;
2223                 }
2224         }
2225
2226         for (i = 0; i < max_mirrors; i++) {
2227                 bytenr = btrfs_sb_offset(i);
2228                 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2229                         break;
2230
2231                 if (wait) {
2232                         bh = __find_get_block(device->bdev, bytenr / 4096,
2233                                               BTRFS_SUPER_INFO_SIZE);
2234                         BUG_ON(!bh);
2235                         wait_on_buffer(bh);
2236                         if (!buffer_uptodate(bh))
2237                                 errors++;
2238
2239                         /* drop our reference */
2240                         brelse(bh);
2241
2242                         /* drop the reference from the wait == 0 run */
2243                         brelse(bh);
2244                         continue;
2245                 } else {
2246                         btrfs_set_super_bytenr(sb, bytenr);
2247
2248                         crc = ~(u32)0;
2249                         crc = btrfs_csum_data(NULL, (char *)sb +
2250                                               BTRFS_CSUM_SIZE, crc,
2251                                               BTRFS_SUPER_INFO_SIZE -
2252                                               BTRFS_CSUM_SIZE);
2253                         btrfs_csum_final(crc, sb->csum);
2254
2255                         /*
2256                          * one reference for us, and we leave it for the
2257                          * caller
2258                          */
2259                         bh = __getblk(device->bdev, bytenr / 4096,
2260                                       BTRFS_SUPER_INFO_SIZE);
2261                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2262
2263                         /* one reference for submit_bh */
2264                         get_bh(bh);
2265
2266                         set_buffer_uptodate(bh);
2267                         lock_buffer(bh);
2268                         bh->b_end_io = btrfs_end_buffer_write_sync;
2269                 }
2270
2271                 if (i == last_barrier && do_barriers)
2272                         ret = submit_bh(WRITE_FLUSH_FUA, bh);
2273                 else
2274                         ret = submit_bh(WRITE_SYNC, bh);
2275
2276                 if (ret)
2277                         errors++;
2278         }
2279         return errors < i ? 0 : -1;
2280 }
2281
2282 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2283 {
2284         struct list_head *head;
2285         struct btrfs_device *dev;
2286         struct btrfs_super_block *sb;
2287         struct btrfs_dev_item *dev_item;
2288         int ret;
2289         int do_barriers;
2290         int max_errors;
2291         int total_errors = 0;
2292         u64 flags;
2293
2294         max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2295         do_barriers = !btrfs_test_opt(root, NOBARRIER);
2296
2297         sb = &root->fs_info->super_for_commit;
2298         dev_item = &sb->dev_item;
2299
2300         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2301         head = &root->fs_info->fs_devices->devices;
2302         list_for_each_entry(dev, head, dev_list) {
2303                 if (!dev->bdev) {
2304                         total_errors++;
2305                         continue;
2306                 }
2307                 if (!dev->in_fs_metadata || !dev->writeable)
2308                         continue;
2309
2310                 btrfs_set_stack_device_generation(dev_item, 0);
2311                 btrfs_set_stack_device_type(dev_item, dev->type);
2312                 btrfs_set_stack_device_id(dev_item, dev->devid);
2313                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2314                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2315                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2316                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2317                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2318                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2319                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2320
2321                 flags = btrfs_super_flags(sb);
2322                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2323
2324                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2325                 if (ret)
2326                         total_errors++;
2327         }
2328         if (total_errors > max_errors) {
2329                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2330                        total_errors);
2331                 BUG();
2332         }
2333
2334         total_errors = 0;
2335         list_for_each_entry(dev, head, dev_list) {
2336                 if (!dev->bdev)
2337                         continue;
2338                 if (!dev->in_fs_metadata || !dev->writeable)
2339                         continue;
2340
2341                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2342                 if (ret)
2343                         total_errors++;
2344         }
2345         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2346         if (total_errors > max_errors) {
2347                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2348                        total_errors);
2349                 BUG();
2350         }
2351         return 0;
2352 }
2353
2354 int write_ctree_super(struct btrfs_trans_handle *trans,
2355                       struct btrfs_root *root, int max_mirrors)
2356 {
2357         int ret;
2358
2359         ret = write_all_supers(root, max_mirrors);
2360         return ret;
2361 }
2362
2363 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2364 {
2365         spin_lock(&fs_info->fs_roots_radix_lock);
2366         radix_tree_delete(&fs_info->fs_roots_radix,
2367                           (unsigned long)root->root_key.objectid);
2368         spin_unlock(&fs_info->fs_roots_radix_lock);
2369
2370         if (btrfs_root_refs(&root->root_item) == 0)
2371                 synchronize_srcu(&fs_info->subvol_srcu);
2372
2373         free_fs_root(root);
2374         return 0;
2375 }
2376
2377 static void free_fs_root(struct btrfs_root *root)
2378 {
2379         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2380         if (root->anon_super.s_dev) {
2381                 down_write(&root->anon_super.s_umount);
2382                 kill_anon_super(&root->anon_super);
2383         }
2384         free_extent_buffer(root->node);
2385         free_extent_buffer(root->commit_root);
2386         kfree(root->name);
2387         kfree(root);
2388 }
2389
2390 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2391 {
2392         int ret;
2393         struct btrfs_root *gang[8];
2394         int i;
2395
2396         while (!list_empty(&fs_info->dead_roots)) {
2397                 gang[0] = list_entry(fs_info->dead_roots.next,
2398                                      struct btrfs_root, root_list);
2399                 list_del(&gang[0]->root_list);
2400
2401                 if (gang[0]->in_radix) {
2402                         btrfs_free_fs_root(fs_info, gang[0]);
2403                 } else {
2404                         free_extent_buffer(gang[0]->node);
2405                         free_extent_buffer(gang[0]->commit_root);
2406                         kfree(gang[0]);
2407                 }
2408         }
2409
2410         while (1) {
2411                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2412                                              (void **)gang, 0,
2413                                              ARRAY_SIZE(gang));
2414                 if (!ret)
2415                         break;
2416                 for (i = 0; i < ret; i++)
2417                         btrfs_free_fs_root(fs_info, gang[i]);
2418         }
2419         return 0;
2420 }
2421
2422 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2423 {
2424         u64 root_objectid = 0;
2425         struct btrfs_root *gang[8];
2426         int i;
2427         int ret;
2428
2429         while (1) {
2430                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2431                                              (void **)gang, root_objectid,
2432                                              ARRAY_SIZE(gang));
2433                 if (!ret)
2434                         break;
2435
2436                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2437                 for (i = 0; i < ret; i++) {
2438                         root_objectid = gang[i]->root_key.objectid;
2439                         btrfs_orphan_cleanup(gang[i]);
2440                 }
2441                 root_objectid++;
2442         }
2443         return 0;
2444 }
2445
2446 int btrfs_commit_super(struct btrfs_root *root)
2447 {
2448         struct btrfs_trans_handle *trans;
2449         int ret;
2450
2451         mutex_lock(&root->fs_info->cleaner_mutex);
2452         btrfs_run_delayed_iputs(root);
2453         btrfs_clean_old_snapshots(root);
2454         mutex_unlock(&root->fs_info->cleaner_mutex);
2455
2456         /* wait until ongoing cleanup work done */
2457         down_write(&root->fs_info->cleanup_work_sem);
2458         up_write(&root->fs_info->cleanup_work_sem);
2459
2460         trans = btrfs_join_transaction(root, 1);
2461         if (IS_ERR(trans))
2462                 return PTR_ERR(trans);
2463         ret = btrfs_commit_transaction(trans, root);
2464         BUG_ON(ret);
2465         /* run commit again to drop the original snapshot */
2466         trans = btrfs_join_transaction(root, 1);
2467         if (IS_ERR(trans))
2468                 return PTR_ERR(trans);
2469         btrfs_commit_transaction(trans, root);
2470         ret = btrfs_write_and_wait_transaction(NULL, root);
2471         BUG_ON(ret);
2472
2473         ret = write_ctree_super(NULL, root, 0);
2474         return ret;
2475 }
2476
2477 int close_ctree(struct btrfs_root *root)
2478 {
2479         struct btrfs_fs_info *fs_info = root->fs_info;
2480         int ret;
2481
2482         fs_info->closing = 1;
2483         smp_mb();
2484
2485         btrfs_put_block_group_cache(fs_info);
2486
2487         /*
2488          * Here come 2 situations when btrfs is broken to flip readonly:
2489          *
2490          * 1. when btrfs flips readonly somewhere else before
2491          * btrfs_commit_super, sb->s_flags has MS_RDONLY flag,
2492          * and btrfs will skip to write sb directly to keep
2493          * ERROR state on disk.
2494          *
2495          * 2. when btrfs flips readonly just in btrfs_commit_super,
2496          * and in such case, btrfs cannot write sb via btrfs_commit_super,
2497          * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag,
2498          * btrfs will cleanup all FS resources first and write sb then.
2499          */
2500         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2501                 ret = btrfs_commit_super(root);
2502                 if (ret)
2503                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2504         }
2505
2506         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
2507                 ret = btrfs_error_commit_super(root);
2508                 if (ret)
2509                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2510         }
2511
2512         kthread_stop(root->fs_info->transaction_kthread);
2513         kthread_stop(root->fs_info->cleaner_kthread);
2514
2515         fs_info->closing = 2;
2516         smp_mb();
2517
2518         if (fs_info->delalloc_bytes) {
2519                 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2520                        (unsigned long long)fs_info->delalloc_bytes);
2521         }
2522         if (fs_info->total_ref_cache_size) {
2523                 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2524                        (unsigned long long)fs_info->total_ref_cache_size);
2525         }
2526
2527         free_extent_buffer(fs_info->extent_root->node);
2528         free_extent_buffer(fs_info->extent_root->commit_root);
2529         free_extent_buffer(fs_info->tree_root->node);
2530         free_extent_buffer(fs_info->tree_root->commit_root);
2531         free_extent_buffer(root->fs_info->chunk_root->node);
2532         free_extent_buffer(root->fs_info->chunk_root->commit_root);
2533         free_extent_buffer(root->fs_info->dev_root->node);
2534         free_extent_buffer(root->fs_info->dev_root->commit_root);
2535         free_extent_buffer(root->fs_info->csum_root->node);
2536         free_extent_buffer(root->fs_info->csum_root->commit_root);
2537
2538         btrfs_free_block_groups(root->fs_info);
2539
2540         del_fs_roots(fs_info);
2541
2542         iput(fs_info->btree_inode);
2543
2544         btrfs_stop_workers(&fs_info->generic_worker);
2545         btrfs_stop_workers(&fs_info->fixup_workers);
2546         btrfs_stop_workers(&fs_info->delalloc_workers);
2547         btrfs_stop_workers(&fs_info->workers);
2548         btrfs_stop_workers(&fs_info->endio_workers);
2549         btrfs_stop_workers(&fs_info->endio_meta_workers);
2550         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2551         btrfs_stop_workers(&fs_info->endio_write_workers);
2552         btrfs_stop_workers(&fs_info->endio_freespace_worker);
2553         btrfs_stop_workers(&fs_info->submit_workers);
2554
2555         btrfs_close_devices(fs_info->fs_devices);
2556         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2557
2558         bdi_destroy(&fs_info->bdi);
2559         cleanup_srcu_struct(&fs_info->subvol_srcu);
2560
2561         kfree(fs_info->extent_root);
2562         kfree(fs_info->tree_root);
2563         kfree(fs_info->chunk_root);
2564         kfree(fs_info->dev_root);
2565         kfree(fs_info->csum_root);
2566         kfree(fs_info);
2567
2568         return 0;
2569 }
2570
2571 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2572 {
2573         int ret;
2574         struct inode *btree_inode = buf->first_page->mapping->host;
2575
2576         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
2577                                      NULL);
2578         if (!ret)
2579                 return ret;
2580
2581         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2582                                     parent_transid);
2583         return !ret;
2584 }
2585
2586 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2587 {
2588         struct inode *btree_inode = buf->first_page->mapping->host;
2589         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2590                                           buf);
2591 }
2592
2593 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2594 {
2595         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2596         u64 transid = btrfs_header_generation(buf);
2597         struct inode *btree_inode = root->fs_info->btree_inode;
2598         int was_dirty;
2599
2600         btrfs_assert_tree_locked(buf);
2601         if (transid != root->fs_info->generation) {
2602                 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2603                        "found %llu running %llu\n",
2604                         (unsigned long long)buf->start,
2605                         (unsigned long long)transid,
2606                         (unsigned long long)root->fs_info->generation);
2607                 WARN_ON(1);
2608         }
2609         was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2610                                             buf);
2611         if (!was_dirty) {
2612                 spin_lock(&root->fs_info->delalloc_lock);
2613                 root->fs_info->dirty_metadata_bytes += buf->len;
2614                 spin_unlock(&root->fs_info->delalloc_lock);
2615         }
2616 }
2617
2618 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2619 {
2620         /*
2621          * looks as though older kernels can get into trouble with
2622          * this code, they end up stuck in balance_dirty_pages forever
2623          */
2624         u64 num_dirty;
2625         unsigned long thresh = 32 * 1024 * 1024;
2626
2627         if (current->flags & PF_MEMALLOC)
2628                 return;
2629
2630         num_dirty = root->fs_info->dirty_metadata_bytes;
2631
2632         if (num_dirty > thresh) {
2633                 balance_dirty_pages_ratelimited_nr(
2634                                    root->fs_info->btree_inode->i_mapping, 1);
2635         }
2636         return;
2637 }
2638
2639 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2640 {
2641         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2642         int ret;
2643         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2644         if (ret == 0)
2645                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
2646         return ret;
2647 }
2648
2649 int btree_lock_page_hook(struct page *page)
2650 {
2651         struct inode *inode = page->mapping->host;
2652         struct btrfs_root *root = BTRFS_I(inode)->root;
2653         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2654         struct extent_buffer *eb;
2655         unsigned long len;
2656         u64 bytenr = page_offset(page);
2657
2658         if (page->private == EXTENT_PAGE_PRIVATE)
2659                 goto out;
2660
2661         len = page->private >> 2;
2662         eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2663         if (!eb)
2664                 goto out;
2665
2666         btrfs_tree_lock(eb);
2667         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2668
2669         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2670                 spin_lock(&root->fs_info->delalloc_lock);
2671                 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2672                         root->fs_info->dirty_metadata_bytes -= eb->len;
2673                 else
2674                         WARN_ON(1);
2675                 spin_unlock(&root->fs_info->delalloc_lock);
2676         }
2677
2678         btrfs_tree_unlock(eb);
2679         free_extent_buffer(eb);
2680 out:
2681         lock_page(page);
2682         return 0;
2683 }
2684
2685 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
2686                               int read_only)
2687 {
2688         if (read_only)
2689                 return;
2690
2691         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
2692                 printk(KERN_WARNING "warning: mount fs with errors, "
2693                        "running btrfsck is recommended\n");
2694 }
2695
2696 int btrfs_error_commit_super(struct btrfs_root *root)
2697 {
2698         int ret;
2699
2700         mutex_lock(&root->fs_info->cleaner_mutex);
2701         btrfs_run_delayed_iputs(root);
2702         mutex_unlock(&root->fs_info->cleaner_mutex);
2703
2704         down_write(&root->fs_info->cleanup_work_sem);
2705         up_write(&root->fs_info->cleanup_work_sem);
2706
2707         /* cleanup FS via transaction */
2708         btrfs_cleanup_transaction(root);
2709
2710         ret = write_ctree_super(NULL, root, 0);
2711
2712         return ret;
2713 }
2714
2715 static int btrfs_destroy_ordered_operations(struct btrfs_root *root)
2716 {
2717         struct btrfs_inode *btrfs_inode;
2718         struct list_head splice;
2719
2720         INIT_LIST_HEAD(&splice);
2721
2722         mutex_lock(&root->fs_info->ordered_operations_mutex);
2723         spin_lock(&root->fs_info->ordered_extent_lock);
2724
2725         list_splice_init(&root->fs_info->ordered_operations, &splice);
2726         while (!list_empty(&splice)) {
2727                 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2728                                          ordered_operations);
2729
2730                 list_del_init(&btrfs_inode->ordered_operations);
2731
2732                 btrfs_invalidate_inodes(btrfs_inode->root);
2733         }
2734
2735         spin_unlock(&root->fs_info->ordered_extent_lock);
2736         mutex_unlock(&root->fs_info->ordered_operations_mutex);
2737
2738         return 0;
2739 }
2740
2741 static int btrfs_destroy_ordered_extents(struct btrfs_root *root)
2742 {
2743         struct list_head splice;
2744         struct btrfs_ordered_extent *ordered;
2745         struct inode *inode;
2746
2747         INIT_LIST_HEAD(&splice);
2748
2749         spin_lock(&root->fs_info->ordered_extent_lock);
2750
2751         list_splice_init(&root->fs_info->ordered_extents, &splice);
2752         while (!list_empty(&splice)) {
2753                 ordered = list_entry(splice.next, struct btrfs_ordered_extent,
2754                                      root_extent_list);
2755
2756                 list_del_init(&ordered->root_extent_list);
2757                 atomic_inc(&ordered->refs);
2758
2759                 /* the inode may be getting freed (in sys_unlink path). */
2760                 inode = igrab(ordered->inode);
2761
2762                 spin_unlock(&root->fs_info->ordered_extent_lock);
2763                 if (inode)
2764                         iput(inode);
2765
2766                 atomic_set(&ordered->refs, 1);
2767                 btrfs_put_ordered_extent(ordered);
2768
2769                 spin_lock(&root->fs_info->ordered_extent_lock);
2770         }
2771
2772         spin_unlock(&root->fs_info->ordered_extent_lock);
2773
2774         return 0;
2775 }
2776
2777 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
2778                                       struct btrfs_root *root)
2779 {
2780         struct rb_node *node;
2781         struct btrfs_delayed_ref_root *delayed_refs;
2782         struct btrfs_delayed_ref_node *ref;
2783         int ret = 0;
2784
2785         delayed_refs = &trans->delayed_refs;
2786
2787         spin_lock(&delayed_refs->lock);
2788         if (delayed_refs->num_entries == 0) {
2789                 printk(KERN_INFO "delayed_refs has NO entry\n");
2790                 return ret;
2791         }
2792
2793         node = rb_first(&delayed_refs->root);
2794         while (node) {
2795                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2796                 node = rb_next(node);
2797
2798                 ref->in_tree = 0;
2799                 rb_erase(&ref->rb_node, &delayed_refs->root);
2800                 delayed_refs->num_entries--;
2801
2802                 atomic_set(&ref->refs, 1);
2803                 if (btrfs_delayed_ref_is_head(ref)) {
2804                         struct btrfs_delayed_ref_head *head;
2805
2806                         head = btrfs_delayed_node_to_head(ref);
2807                         mutex_lock(&head->mutex);
2808                         kfree(head->extent_op);
2809                         delayed_refs->num_heads--;
2810                         if (list_empty(&head->cluster))
2811                                 delayed_refs->num_heads_ready--;
2812                         list_del_init(&head->cluster);
2813                         mutex_unlock(&head->mutex);
2814                 }
2815
2816                 spin_unlock(&delayed_refs->lock);
2817                 btrfs_put_delayed_ref(ref);
2818
2819                 cond_resched();
2820                 spin_lock(&delayed_refs->lock);
2821         }
2822
2823         spin_unlock(&delayed_refs->lock);
2824
2825         return ret;
2826 }
2827
2828 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
2829 {
2830         struct btrfs_pending_snapshot *snapshot;
2831         struct list_head splice;
2832
2833         INIT_LIST_HEAD(&splice);
2834
2835         list_splice_init(&t->pending_snapshots, &splice);
2836
2837         while (!list_empty(&splice)) {
2838                 snapshot = list_entry(splice.next,
2839                                       struct btrfs_pending_snapshot,
2840                                       list);
2841
2842                 list_del_init(&snapshot->list);
2843
2844                 kfree(snapshot);
2845         }
2846
2847         return 0;
2848 }
2849
2850 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
2851 {
2852         struct btrfs_inode *btrfs_inode;
2853         struct list_head splice;
2854
2855         INIT_LIST_HEAD(&splice);
2856
2857         list_splice_init(&root->fs_info->delalloc_inodes, &splice);
2858
2859         spin_lock(&root->fs_info->delalloc_lock);
2860
2861         while (!list_empty(&splice)) {
2862                 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2863                                     delalloc_inodes);
2864
2865                 list_del_init(&btrfs_inode->delalloc_inodes);
2866
2867                 btrfs_invalidate_inodes(btrfs_inode->root);
2868         }
2869
2870         spin_unlock(&root->fs_info->delalloc_lock);
2871
2872         return 0;
2873 }
2874
2875 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
2876                                         struct extent_io_tree *dirty_pages,
2877                                         int mark)
2878 {
2879         int ret;
2880         struct page *page;
2881         struct inode *btree_inode = root->fs_info->btree_inode;
2882         struct extent_buffer *eb;
2883         u64 start = 0;
2884         u64 end;
2885         u64 offset;
2886         unsigned long index;
2887
2888         while (1) {
2889                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
2890                                             mark);
2891                 if (ret)
2892                         break;
2893
2894                 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
2895                 while (start <= end) {
2896                         index = start >> PAGE_CACHE_SHIFT;
2897                         start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
2898                         page = find_get_page(btree_inode->i_mapping, index);
2899                         if (!page)
2900                                 continue;
2901                         offset = page_offset(page);
2902
2903                         spin_lock(&dirty_pages->buffer_lock);
2904                         eb = radix_tree_lookup(
2905                              &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
2906                                                offset >> PAGE_CACHE_SHIFT);
2907                         spin_unlock(&dirty_pages->buffer_lock);
2908                         if (eb) {
2909                                 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
2910                                                          &eb->bflags);
2911                                 atomic_set(&eb->refs, 1);
2912                         }
2913                         if (PageWriteback(page))
2914                                 end_page_writeback(page);
2915
2916                         lock_page(page);
2917                         if (PageDirty(page)) {
2918                                 clear_page_dirty_for_io(page);
2919                                 spin_lock_irq(&page->mapping->tree_lock);
2920                                 radix_tree_tag_clear(&page->mapping->page_tree,
2921                                                         page_index(page),
2922                                                         PAGECACHE_TAG_DIRTY);
2923                                 spin_unlock_irq(&page->mapping->tree_lock);
2924                         }
2925
2926                         page->mapping->a_ops->invalidatepage(page, 0);
2927                         unlock_page(page);
2928                 }
2929         }
2930
2931         return ret;
2932 }
2933
2934 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
2935                                        struct extent_io_tree *pinned_extents)
2936 {
2937         struct extent_io_tree *unpin;
2938         u64 start;
2939         u64 end;
2940         int ret;
2941
2942         unpin = pinned_extents;
2943         while (1) {
2944                 ret = find_first_extent_bit(unpin, 0, &start, &end,
2945                                             EXTENT_DIRTY);
2946                 if (ret)
2947                         break;
2948
2949                 /* opt_discard */
2950                 ret = btrfs_error_discard_extent(root, start, end + 1 - start);
2951
2952                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
2953                 btrfs_error_unpin_extent_range(root, start, end);
2954                 cond_resched();
2955         }
2956
2957         return 0;
2958 }
2959
2960 static int btrfs_cleanup_transaction(struct btrfs_root *root)
2961 {
2962         struct btrfs_transaction *t;
2963         LIST_HEAD(list);
2964
2965         WARN_ON(1);
2966
2967         mutex_lock(&root->fs_info->trans_mutex);
2968         mutex_lock(&root->fs_info->transaction_kthread_mutex);
2969
2970         list_splice_init(&root->fs_info->trans_list, &list);
2971         while (!list_empty(&list)) {
2972                 t = list_entry(list.next, struct btrfs_transaction, list);
2973                 if (!t)
2974                         break;
2975
2976                 btrfs_destroy_ordered_operations(root);
2977
2978                 btrfs_destroy_ordered_extents(root);
2979
2980                 btrfs_destroy_delayed_refs(t, root);
2981
2982                 btrfs_block_rsv_release(root,
2983                                         &root->fs_info->trans_block_rsv,
2984                                         t->dirty_pages.dirty_bytes);
2985
2986                 /* FIXME: cleanup wait for commit */
2987                 t->in_commit = 1;
2988                 t->blocked = 1;
2989                 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
2990                         wake_up(&root->fs_info->transaction_blocked_wait);
2991
2992                 t->blocked = 0;
2993                 if (waitqueue_active(&root->fs_info->transaction_wait))
2994                         wake_up(&root->fs_info->transaction_wait);
2995                 mutex_unlock(&root->fs_info->trans_mutex);
2996
2997                 mutex_lock(&root->fs_info->trans_mutex);
2998                 t->commit_done = 1;
2999                 if (waitqueue_active(&t->commit_wait))
3000                         wake_up(&t->commit_wait);
3001                 mutex_unlock(&root->fs_info->trans_mutex);
3002
3003                 mutex_lock(&root->fs_info->trans_mutex);
3004
3005                 btrfs_destroy_pending_snapshots(t);
3006
3007                 btrfs_destroy_delalloc_inodes(root);
3008
3009                 spin_lock(&root->fs_info->new_trans_lock);
3010                 root->fs_info->running_transaction = NULL;
3011                 spin_unlock(&root->fs_info->new_trans_lock);
3012
3013                 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3014                                              EXTENT_DIRTY);
3015
3016                 btrfs_destroy_pinned_extent(root,
3017                                             root->fs_info->pinned_extents);
3018
3019                 t->use_count = 0;
3020                 list_del_init(&t->list);
3021                 memset(t, 0, sizeof(*t));
3022                 kmem_cache_free(btrfs_transaction_cachep, t);
3023         }
3024
3025         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3026         mutex_unlock(&root->fs_info->trans_mutex);
3027
3028         return 0;
3029 }
3030
3031 static struct extent_io_ops btree_extent_io_ops = {
3032         .write_cache_pages_lock_hook = btree_lock_page_hook,
3033         .readpage_end_io_hook = btree_readpage_end_io_hook,
3034         .submit_bio_hook = btree_submit_bio_hook,
3035         /* note we're sharing with inode.c for the merge bio hook */
3036         .merge_bio_hook = btrfs_merge_bio_hook,
3037 };