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