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 <asm/unaligned.h>
  33 #include "compat.h"
  34 #include "ctree.h"
  35 #include "disk-io.h"
  36 #include "transaction.h"
  37 #include "btrfs_inode.h"
  38 #include "volumes.h"
  39 #include "print-tree.h"
  40 #include "async-thread.h"
  41 #include "locking.h"
  42 #include "tree-log.h"
  43 #include "free-space-cache.h"
  44
  45 static struct extent_io_ops btree_extent_io_ops;
  46 static void end_workqueue_fn(struct btrfs_work *work);
  47 static void free_fs_root(struct btrfs_root *root);
  48 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
  49                                     int read_only);
  50 static int btrfs_destroy_ordered_operations(struct btrfs_root *root);
  51 static int btrfs_destroy_ordered_extents(struct btrfs_root *root);
  52 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
  53                                       struct btrfs_root *root);
  54 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t);
  55 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
  56 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
  57                                         struct extent_io_tree *dirty_pages,
  58                                         int mark);
  59 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
  60                                        struct extent_io_tree *pinned_extents);
  61 static int btrfs_cleanup_transaction(struct btrfs_root *root);
  62
  63 /*
  64  * end_io_wq structs are used to do processing in task context when an IO is
  65  * complete.  This is used during reads to verify checksums, and it is used
  66  * by writes to insert metadata for new file extents after IO is complete.
  67  */
  68 struct end_io_wq {
  69         struct bio *bio;
  70         bio_end_io_t *end_io;
  71         void *private;
  72         struct btrfs_fs_info *info;
  73         int error;
  74         int metadata;
  75         struct list_head list;
  76         struct btrfs_work work;
  77 };
  78
  79 /*
  80  * async submit bios are used to offload expensive checksumming
  81  * onto the worker threads.  They checksum file and metadata bios
  82  * just before they are sent down the IO stack.
  83  */
  84 struct async_submit_bio {
  85         struct inode *inode;
  86         struct bio *bio;
  87         struct list_head list;
  88         extent_submit_bio_hook_t *submit_bio_start;
  89         extent_submit_bio_hook_t *submit_bio_done;
  90         int rw;
  91         int mirror_num;
  92         unsigned long bio_flags;
  93         /*
  94          * bio_offset is optional, can be used if the pages in the bio
  95          * can't tell us where in the file the bio should go
  96          */
  97         u64 bio_offset;
  98         struct btrfs_work work;
  99 };
 100
 101 /* These are used to set the lockdep class on the extent buffer locks.
 102  * The class is set by the readpage_end_io_hook after the buffer has
 103  * passed csum validation but before the pages are unlocked.
 104  *
 105  * The lockdep class is also set by btrfs_init_new_buffer on freshly
 106  * allocated blocks.
 107  *
 108  * The class is based on the level in the tree block, which allows lockdep
 109  * to know that lower nodes nest inside the locks of higher nodes.
 110  *
 111  * We also add a check to make sure the highest level of the tree is
 112  * the same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this
 113  * code needs update as well.
 114  */
 115 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 116 # if BTRFS_MAX_LEVEL != 8
 117 #  error
 118 # endif
 119 static struct lock_class_key btrfs_eb_class[BTRFS_MAX_LEVEL + 1];
 120 static const char *btrfs_eb_name[BTRFS_MAX_LEVEL + 1] = {
 121         /* leaf */
 122         "btrfs-extent-00",
 123         "btrfs-extent-01",
 124         "btrfs-extent-02",
 125         "btrfs-extent-03",
 126         "btrfs-extent-04",
 127         "btrfs-extent-05",
 128         "btrfs-extent-06",
 129         "btrfs-extent-07",
 130         /* highest possible level */
 131         "btrfs-extent-08",
 132 };
 133 #endif
 134
 135 /*
 136  * extents on the btree inode are pretty simple, there's one extent
 137  * that covers the entire device
 138  */
 139 static struct extent_map *btree_get_extent(struct inode *inode,
 140                 struct page *page, size_t page_offset, u64 start, u64 len,
 141                 int create)
 142 {
 143         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 144         struct extent_map *em;
 145         int ret;
 146
 147         read_lock(&em_tree->lock);
 148         em = lookup_extent_mapping(em_tree, start, len);
 149         if (em) {
 150                 em->bdev =
 151                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 152                 read_unlock(&em_tree->lock);
 153                 goto out;
 154         }
 155         read_unlock(&em_tree->lock);
 156
 157         em = alloc_extent_map(GFP_NOFS);
 158         if (!em) {
 159                 em = ERR_PTR(-ENOMEM);
 160                 goto out;
 161         }
 162         em->start = 0;
 163         em->len = (u64)-1;
 164         em->block_len = (u64)-1;
 165         em->block_start = 0;
 166         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 167
 168         write_lock(&em_tree->lock);
 169         ret = add_extent_mapping(em_tree, em);
 170         if (ret == -EEXIST) {
 171                 u64 failed_start = em->start;
 172                 u64 failed_len = em->len;
 173
 174                 free_extent_map(em);
 175                 em = lookup_extent_mapping(em_tree, start, len);
 176                 if (em) {
 177                         ret = 0;
 178                 } else {
 179                         em = lookup_extent_mapping(em_tree, failed_start,
 180                                                    failed_len);
 181                         ret = -EIO;
 182                 }
 183         } else if (ret) {
 184                 free_extent_map(em);
 185                 em = NULL;
 186         }
 187         write_unlock(&em_tree->lock);
 188
 189         if (ret)
 190                 em = ERR_PTR(ret);
 191 out:
 192         return em;
 193 }
 194
 195 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
 196 {
 197         return crc32c(seed, data, len);
 198 }
 199
 200 void btrfs_csum_final(u32 crc, char *result)
 201 {
 202         put_unaligned_le32(~crc, result);
 203 }
 204
 205 /*
 206  * compute the csum for a btree block, and either verify it or write it
 207  * into the csum field of the block.
 208  */
 209 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 210                            int verify)
 211 {
 212         u16 csum_size =
 213                 btrfs_super_csum_size(&root->fs_info->super_copy);
 214         char *result = NULL;
 215         unsigned long len;
 216         unsigned long cur_len;
 217         unsigned long offset = BTRFS_CSUM_SIZE;
 218         char *map_token = NULL;
 219         char *kaddr;
 220         unsigned long map_start;
 221         unsigned long map_len;
 222         int err;
 223         u32 crc = ~(u32)0;
 224         unsigned long inline_result;
 225
 226         len = buf->len - offset;
 227         while (len > 0) {
 228                 err = map_private_extent_buffer(buf, offset, 32,
 229                                         &map_token, &kaddr,
 230                                         &map_start, &map_len, KM_USER0);
 231                 if (err)
 232                         return 1;
 233                 cur_len = min(len, map_len - (offset - map_start));
 234                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
 235                                       crc, cur_len);
 236                 len -= cur_len;
 237                 offset += cur_len;
 238                 unmap_extent_buffer(buf, map_token, KM_USER0);
 239         }
 240         if (csum_size > sizeof(inline_result)) {
 241                 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
 242                 if (!result)
 243                         return 1;
 244         } else {
 245                 result = (char *)&inline_result;
 246         }
 247
 248         btrfs_csum_final(crc, result);
 249
 250         if (verify) {
 251                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
 252                         u32 val;
 253                         u32 found = 0;
 254                         memcpy(&found, result, csum_size);
 255
 256                         read_extent_buffer(buf, &val, 0, csum_size);
 257                         if (printk_ratelimit()) {
 258                                 printk(KERN_INFO "btrfs: %s checksum verify "
 259                                        "failed on %llu wanted %X found %X "
 260                                        "level %d\n",
 261                                        root->fs_info->sb->s_id,
 262                                        (unsigned long long)buf->start, val, found,
 263                                        btrfs_header_level(buf));
 264                         }
 265                         if (result != (char *)&inline_result)
 266                                 kfree(result);
 267                         return 1;
 268                 }
 269         } else {
 270                 write_extent_buffer(buf, result, 0, csum_size);
 271         }
 272         if (result != (char *)&inline_result)
 273                 kfree(result);
 274         return 0;
 275 }
 276
 277 /*
 278  * we can't consider a given block up to date unless the transid of the
 279  * block matches the transid in the parent node's pointer.  This is how we
 280  * detect blocks that either didn't get written at all or got written
 281  * in the wrong place.
 282  */
 283 static int verify_parent_transid(struct extent_io_tree *io_tree,
 284                                  struct extent_buffer *eb, u64 parent_transid)
 285 {
 286         struct extent_state *cached_state = NULL;
 287         int ret;
 288
 289         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 290                 return 0;
 291
 292         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
 293                          0, &cached_state, GFP_NOFS);
 294         if (extent_buffer_uptodate(io_tree, eb, cached_state) &&
 295             btrfs_header_generation(eb) == parent_transid) {
 296                 ret = 0;
 297                 goto out;
 298         }
 299         if (printk_ratelimit()) {
 300                 printk("parent transid verify failed on %llu wanted %llu "
 301                        "found %llu\n",
 302                        (unsigned long long)eb->start,
 303                        (unsigned long long)parent_transid,
 304                        (unsigned long long)btrfs_header_generation(eb));
 305         }
 306         ret = 1;
 307         clear_extent_buffer_uptodate(io_tree, eb, &cached_state);
 308 out:
 309         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
 310                              &cached_state, GFP_NOFS);
 311         return ret;
 312 }
 313
 314 /*
 315  * helper to read a given tree block, doing retries as required when
 316  * the checksums don't match and we have alternate mirrors to try.
 317  */
 318 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 319                                           struct extent_buffer *eb,
 320                                           u64 start, u64 parent_transid)
 321 {
 322         struct extent_io_tree *io_tree;
 323         int ret;
 324         int num_copies = 0;
 325         int mirror_num = 0;
 326
 327         clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 328         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 329         while (1) {
 330                 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
 331                                                btree_get_extent, mirror_num);
 332                 if (!ret &&
 333                     !verify_parent_transid(io_tree, eb, parent_transid))
 334                         return ret;
 335
 336                 /*
 337                  * This buffer's crc is fine, but its contents are corrupted, so
 338                  * there is no reason to read the other copies, they won't be
 339                  * any less wrong.
 340                  */
 341                 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
 342                         return ret;
 343
 344                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
 345                                               eb->start, eb->len);
 346                 if (num_copies == 1)
 347                         return ret;
 348
 349                 mirror_num++;
 350                 if (mirror_num > num_copies)
 351                         return ret;
 352         }
 353         return -EIO;
 354 }
 355
 356 /*
 357  * checksum a dirty tree block before IO.  This has extra checks to make sure
 358  * we only fill in the checksum field in the first page of a multi-page block
 359  */
 360
 361 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 362 {
 363         struct extent_io_tree *tree;
 364         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
 365         u64 found_start;
 366         unsigned long len;
 367         struct extent_buffer *eb;
 368         int ret;
 369
 370         tree = &BTRFS_I(page->mapping->host)->io_tree;
 371
 372         if (page->private == EXTENT_PAGE_PRIVATE) {
 373                 WARN_ON(1);
 374                 goto out;
 375         }
 376         if (!page->private) {
 377                 WARN_ON(1);
 378                 goto out;
 379         }
 380         len = page->private >> 2;
 381         WARN_ON(len == 0);
 382
 383         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 384         if (eb == NULL) {
 385                 WARN_ON(1);
 386                 goto out;
 387         }
 388         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
 389                                              btrfs_header_generation(eb));
 390         BUG_ON(ret);
 391         WARN_ON(!btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN));
 392
 393         found_start = btrfs_header_bytenr(eb);
 394         if (found_start != start) {
 395                 WARN_ON(1);
 396                 goto err;
 397         }
 398         if (eb->first_page != page) {
 399                 WARN_ON(1);
 400                 goto err;
 401         }
 402         if (!PageUptodate(page)) {
 403                 WARN_ON(1);
 404                 goto err;
 405         }
 406         csum_tree_block(root, eb, 0);
 407 err:
 408         free_extent_buffer(eb);
 409 out:
 410         return 0;
 411 }
 412
 413 static int check_tree_block_fsid(struct btrfs_root *root,
 414                                  struct extent_buffer *eb)
 415 {
 416         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
 417         u8 fsid[BTRFS_UUID_SIZE];
 418         int ret = 1;
 419
 420         read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
 421                            BTRFS_FSID_SIZE);
 422         while (fs_devices) {
 423                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
 424                         ret = 0;
 425                         break;
 426                 }
 427                 fs_devices = fs_devices->seed;
 428         }
 429         return ret;
 430 }
 431
 432 #define CORRUPT(reason, eb, root, slot)                         \
 433         printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
 434                "root=%llu, slot=%d\n", reason,                  \
 435                (unsigned long long)btrfs_header_bytenr(eb),     \
 436                (unsigned long long)root->objectid, slot)
 437
 438 static noinline int check_leaf(struct btrfs_root *root,
 439                                struct extent_buffer *leaf)
 440 {
 441         struct btrfs_key key;
 442         struct btrfs_key leaf_key;
 443         u32 nritems = btrfs_header_nritems(leaf);
 444         int slot;
 445
 446         if (nritems == 0)
 447                 return 0;
 448
 449         /* Check the 0 item */
 450         if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
 451             BTRFS_LEAF_DATA_SIZE(root)) {
 452                 CORRUPT("invalid item offset size pair", leaf, root, 0);
 453                 return -EIO;
 454         }
 455
 456         /*
 457          * Check to make sure each items keys are in the correct order and their
 458          * offsets make sense.  We only have to loop through nritems-1 because
 459          * we check the current slot against the next slot, which verifies the
 460          * next slot's offset+size makes sense and that the current's slot
 461          * offset is correct.
 462          */
 463         for (slot = 0; slot < nritems - 1; slot++) {
 464                 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
 465                 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
 466
 467                 /* Make sure the keys are in the right order */
 468                 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
 469                         CORRUPT("bad key order", leaf, root, slot);
 470                         return -EIO;
 471                 }
 472
 473                 /*
 474                  * Make sure the offset and ends are right, remember that the
 475                  * item data starts at the end of the leaf and grows towards the
 476                  * front.
 477                  */
 478                 if (btrfs_item_offset_nr(leaf, slot) !=
 479                         btrfs_item_end_nr(leaf, slot + 1)) {
 480                         CORRUPT("slot offset bad", leaf, root, slot);
 481                         return -EIO;
 482                 }
 483
 484                 /*
 485                  * Check to make sure that we don't point outside of the leaf,
 486                  * just incase all the items are consistent to eachother, but
 487                  * all point outside of the leaf.
 488                  */
 489                 if (btrfs_item_end_nr(leaf, slot) >
 490                     BTRFS_LEAF_DATA_SIZE(root)) {
 491                         CORRUPT("slot end outside of leaf", leaf, root, slot);
 492                         return -EIO;
 493                 }
 494         }
 495
 496         return 0;
 497 }
 498
 499 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 500 void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level)
 501 {
 502         lockdep_set_class_and_name(&eb->lock,
 503                            &btrfs_eb_class[level],
 504                            btrfs_eb_name[level]);
 505 }
 506 #endif
 507
 508 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 509                                struct extent_state *state)
 510 {
 511         struct extent_io_tree *tree;
 512         u64 found_start;
 513         int found_level;
 514         unsigned long len;
 515         struct extent_buffer *eb;
 516         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 517         int ret = 0;
 518
 519         tree = &BTRFS_I(page->mapping->host)->io_tree;
 520         if (page->private == EXTENT_PAGE_PRIVATE)
 521                 goto out;
 522         if (!page->private)
 523                 goto out;
 524
 525         len = page->private >> 2;
 526         WARN_ON(len == 0);
 527
 528         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 529         if (eb == NULL) {
 530                 ret = -EIO;
 531                 goto out;
 532         }
 533
 534         found_start = btrfs_header_bytenr(eb);
 535         if (found_start != start) {
 536                 if (printk_ratelimit()) {
 537                         printk(KERN_INFO "btrfs bad tree block start "
 538                                "%llu %llu\n",
 539                                (unsigned long long)found_start,
 540                                (unsigned long long)eb->start);
 541                 }
 542                 ret = -EIO;
 543                 goto err;
 544         }
 545         if (eb->first_page != page) {
 546                 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
 547                        eb->first_page->index, page->index);
 548                 WARN_ON(1);
 549                 ret = -EIO;
 550                 goto err;
 551         }
 552         if (check_tree_block_fsid(root, eb)) {
 553                 if (printk_ratelimit()) {
 554                         printk(KERN_INFO "btrfs bad fsid on block %llu\n",
 555                                (unsigned long long)eb->start);
 556                 }
 557                 ret = -EIO;
 558                 goto err;
 559         }
 560         found_level = btrfs_header_level(eb);
 561
 562         btrfs_set_buffer_lockdep_class(eb, found_level);
 563
 564         ret = csum_tree_block(root, eb, 1);
 565         if (ret) {
 566                 ret = -EIO;
 567                 goto err;
 568         }
 569
 570         /*
 571          * If this is a leaf block and it is corrupt, set the corrupt bit so
 572          * that we don't try and read the other copies of this block, just
 573          * return -EIO.
 574          */
 575         if (found_level == 0 && check_leaf(root, eb)) {
 576                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 577                 ret = -EIO;
 578         }
 579
 580         end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
 581         end = eb->start + end - 1;
 582 err:
 583         free_extent_buffer(eb);
 584 out:
 585         return ret;
 586 }
 587
 588 static void end_workqueue_bio(struct bio *bio, int err)
 589 {
 590         struct end_io_wq *end_io_wq = bio->bi_private;
 591         struct btrfs_fs_info *fs_info;
 592
 593         fs_info = end_io_wq->info;
 594         end_io_wq->error = err;
 595         end_io_wq->work.func = end_workqueue_fn;
 596         end_io_wq->work.flags = 0;
 597
 598         if (bio->bi_rw & REQ_WRITE) {
 599                 if (end_io_wq->metadata == 1)
 600                         btrfs_queue_worker(&fs_info->endio_meta_write_workers,
 601                                            &end_io_wq->work);
 602                 else if (end_io_wq->metadata == 2)
 603                         btrfs_queue_worker(&fs_info->endio_freespace_worker,
 604                                            &end_io_wq->work);
 605                 else
 606                         btrfs_queue_worker(&fs_info->endio_write_workers,
 607                                            &end_io_wq->work);
 608         } else {
 609                 if (end_io_wq->metadata)
 610                         btrfs_queue_worker(&fs_info->endio_meta_workers,
 611                                            &end_io_wq->work);
 612                 else
 613                         btrfs_queue_worker(&fs_info->endio_workers,
 614                                            &end_io_wq->work);
 615         }
 616 }
 617
 618 /*
 619  * For the metadata arg you want
 620  *
 621  * 0 - if data
 622  * 1 - if normal metadta
 623  * 2 - if writing to the free space cache area
 624  */
 625 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 626                         int metadata)
 627 {
 628         struct end_io_wq *end_io_wq;
 629         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
 630         if (!end_io_wq)
 631                 return -ENOMEM;
 632
 633         end_io_wq->private = bio->bi_private;
 634         end_io_wq->end_io = bio->bi_end_io;
 635         end_io_wq->info = info;
 636         end_io_wq->error = 0;
 637         end_io_wq->bio = bio;
 638         end_io_wq->metadata = metadata;
 639
 640         bio->bi_private = end_io_wq;
 641         bio->bi_end_io = end_workqueue_bio;
 642         return 0;
 643 }
 644
 645 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
 646 {
 647         unsigned long limit = min_t(unsigned long,
 648                                     info->workers.max_workers,
 649                                     info->fs_devices->open_devices);
 650         return 256 * limit;
 651 }
 652
 653 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
 654 {
 655         return atomic_read(&info->nr_async_bios) >
 656                 btrfs_async_submit_limit(info);
 657 }
 658
 659 static void run_one_async_start(struct btrfs_work *work)
 660 {
 661         struct async_submit_bio *async;
 662
 663         async = container_of(work, struct  async_submit_bio, work);
 664         async->submit_bio_start(async->inode, async->rw, async->bio,
 665                                async->mirror_num, async->bio_flags,
 666                                async->bio_offset);
 667 }
 668
 669 static void run_one_async_done(struct btrfs_work *work)
 670 {
 671         struct btrfs_fs_info *fs_info;
 672         struct async_submit_bio *async;
 673         int limit;
 674
 675         async = container_of(work, struct  async_submit_bio, work);
 676         fs_info = BTRFS_I(async->inode)->root->fs_info;
 677
 678         limit = btrfs_async_submit_limit(fs_info);
 679         limit = limit * 2 / 3;
 680
 681         atomic_dec(&fs_info->nr_async_submits);
 682
 683         if (atomic_read(&fs_info->nr_async_submits) < limit &&
 684             waitqueue_active(&fs_info->async_submit_wait))
 685                 wake_up(&fs_info->async_submit_wait);
 686
 687         async->submit_bio_done(async->inode, async->rw, async->bio,
 688                                async->mirror_num, async->bio_flags,
 689                                async->bio_offset);
 690 }
 691
 692 static void run_one_async_free(struct btrfs_work *work)
 693 {
 694         struct async_submit_bio *async;
 695
 696         async = container_of(work, struct  async_submit_bio, work);
 697         kfree(async);
 698 }
 699
 700 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 701                         int rw, struct bio *bio, int mirror_num,
 702                         unsigned long bio_flags,
 703                         u64 bio_offset,
 704                         extent_submit_bio_hook_t *submit_bio_start,
 705                         extent_submit_bio_hook_t *submit_bio_done)
 706 {
 707         struct async_submit_bio *async;
 708
 709         async = kmalloc(sizeof(*async), GFP_NOFS);
 710         if (!async)
 711                 return -ENOMEM;
 712
 713         async->inode = inode;
 714         async->rw = rw;
 715         async->bio = bio;
 716         async->mirror_num = mirror_num;
 717         async->submit_bio_start = submit_bio_start;
 718         async->submit_bio_done = submit_bio_done;
 719
 720         async->work.func = run_one_async_start;
 721         async->work.ordered_func = run_one_async_done;
 722         async->work.ordered_free = run_one_async_free;
 723
 724         async->work.flags = 0;
 725         async->bio_flags = bio_flags;
 726         async->bio_offset = bio_offset;
 727
 728         atomic_inc(&fs_info->nr_async_submits);
 729
 730         if (rw & REQ_SYNC)
 731                 btrfs_set_work_high_prio(&async->work);
 732
 733         btrfs_queue_worker(&fs_info->workers, &async->work);
 734
 735         while (atomic_read(&fs_info->async_submit_draining) &&
 736               atomic_read(&fs_info->nr_async_submits)) {
 737                 wait_event(fs_info->async_submit_wait,
 738                            (atomic_read(&fs_info->nr_async_submits) == 0));
 739         }
 740
 741         return 0;
 742 }
 743
 744 static int btree_csum_one_bio(struct bio *bio)
 745 {
 746         struct bio_vec *bvec = bio->bi_io_vec;
 747         int bio_index = 0;
 748         struct btrfs_root *root;
 749
 750         WARN_ON(bio->bi_vcnt <= 0);
 751         while (bio_index < bio->bi_vcnt) {
 752                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 753                 csum_dirty_buffer(root, bvec->bv_page);
 754                 bio_index++;
 755                 bvec++;
 756         }
 757         return 0;
 758 }
 759
 760 static int __btree_submit_bio_start(struct inode *inode, int rw,
 761                                     struct bio *bio, int mirror_num,
 762                                     unsigned long bio_flags,
 763                                     u64 bio_offset)
 764 {
 765         /*
 766          * when we're called for a write, we're already in the async
 767          * submission context.  Just jump into btrfs_map_bio
 768          */
 769         btree_csum_one_bio(bio);
 770         return 0;
 771 }
 772
 773 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
 774                                  int mirror_num, unsigned long bio_flags,
 775                                  u64 bio_offset)
 776 {
 777         /*
 778          * when we're called for a write, we're already in the async
 779          * submission context.  Just jump into btrfs_map_bio
 780          */
 781         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
 782 }
 783
 784 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 785                                  int mirror_num, unsigned long bio_flags,
 786                                  u64 bio_offset)
 787 {
 788         int ret;
 789
 790         ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
 791                                           bio, 1);
 792         BUG_ON(ret);
 793
 794         if (!(rw & REQ_WRITE)) {
 795                 /*
 796                  * called for a read, do the setup so that checksum validation
 797                  * can happen in the async kernel threads
 798                  */
 799                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 800                                      mirror_num, 0);
 801         }
 802
 803         /*
 804          * kthread helpers are used to submit writes so that checksumming
 805          * can happen in parallel across all CPUs
 806          */
 807         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 808                                    inode, rw, bio, mirror_num, 0,
 809                                    bio_offset,
 810                                    __btree_submit_bio_start,
 811                                    __btree_submit_bio_done);
 812 }
 813
 814 #ifdef CONFIG_MIGRATION
 815 static int btree_migratepage(struct address_space *mapping,
 816                         struct page *newpage, struct page *page)
 817 {
 818         /*
 819          * we can't safely write a btree page from here,
 820          * we haven't done the locking hook
 821          */
 822         if (PageDirty(page))
 823                 return -EAGAIN;
 824         /*
 825          * Buffers may be managed in a filesystem specific way.
 826          * We must have no buffers or drop them.
 827          */
 828         if (page_has_private(page) &&
 829             !try_to_release_page(page, GFP_KERNEL))
 830                 return -EAGAIN;
 831         return migrate_page(mapping, newpage, page);
 832 }
 833 #endif
 834
 835 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 836 {
 837         struct extent_io_tree *tree;
 838         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 839         struct extent_buffer *eb;
 840         int was_dirty;
 841
 842         tree = &BTRFS_I(page->mapping->host)->io_tree;
 843         if (!(current->flags & PF_MEMALLOC)) {
 844                 return extent_write_full_page(tree, page,
 845                                               btree_get_extent, wbc);
 846         }
 847
 848         redirty_page_for_writepage(wbc, page);
 849         eb = btrfs_find_tree_block(root, page_offset(page), PAGE_CACHE_SIZE);
 850         WARN_ON(!eb);
 851
 852         was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
 853         if (!was_dirty) {
 854                 spin_lock(&root->fs_info->delalloc_lock);
 855                 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
 856                 spin_unlock(&root->fs_info->delalloc_lock);
 857         }
 858         free_extent_buffer(eb);
 859
 860         unlock_page(page);
 861         return 0;
 862 }
 863
 864 static int btree_writepages(struct address_space *mapping,
 865                             struct writeback_control *wbc)
 866 {
 867         struct extent_io_tree *tree;
 868         tree = &BTRFS_I(mapping->host)->io_tree;
 869         if (wbc->sync_mode == WB_SYNC_NONE) {
 870                 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
 871                 u64 num_dirty;
 872                 unsigned long thresh = 32 * 1024 * 1024;
 873
 874                 if (wbc->for_kupdate)
 875                         return 0;
 876
 877                 /* this is a bit racy, but that's ok */
 878                 num_dirty = root->fs_info->dirty_metadata_bytes;
 879                 if (num_dirty < thresh)
 880                         return 0;
 881         }
 882         return extent_writepages(tree, mapping, btree_get_extent, wbc);
 883 }
 884
 885 static int btree_readpage(struct file *file, struct page *page)
 886 {
 887         struct extent_io_tree *tree;
 888         tree = &BTRFS_I(page->mapping->host)->io_tree;
 889         return extent_read_full_page(tree, page, btree_get_extent);
 890 }
 891
 892 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 893 {
 894         struct extent_io_tree *tree;
 895         struct extent_map_tree *map;
 896         int ret;
 897
 898         if (PageWriteback(page) || PageDirty(page))
 899                 return 0;
 900
 901         tree = &BTRFS_I(page->mapping->host)->io_tree;
 902         map = &BTRFS_I(page->mapping->host)->extent_tree;
 903
 904         ret = try_release_extent_state(map, tree, page, gfp_flags);
 905         if (!ret)
 906                 return 0;
 907
 908         ret = try_release_extent_buffer(tree, page);
 909         if (ret == 1) {
 910                 ClearPagePrivate(page);
 911                 set_page_private(page, 0);
 912                 page_cache_release(page);
 913         }
 914
 915         return ret;
 916 }
 917
 918 static void btree_invalidatepage(struct page *page, unsigned long offset)
 919 {
 920         struct extent_io_tree *tree;
 921         tree = &BTRFS_I(page->mapping->host)->io_tree;
 922         extent_invalidatepage(tree, page, offset);
 923         btree_releasepage(page, GFP_NOFS);
 924         if (PagePrivate(page)) {
 925                 printk(KERN_WARNING "btrfs warning page private not zero "
 926                        "on page %llu\n", (unsigned long long)page_offset(page));
 927                 ClearPagePrivate(page);
 928                 set_page_private(page, 0);
 929                 page_cache_release(page);
 930         }
 931 }
 932
 933 static const struct address_space_operations btree_aops = {
 934         .readpage       = btree_readpage,
 935         .writepage      = btree_writepage,
 936         .writepages     = btree_writepages,
 937         .releasepage    = btree_releasepage,
 938         .invalidatepage = btree_invalidatepage,
 939 #ifdef CONFIG_MIGRATION
 940         .migratepage    = btree_migratepage,
 941 #endif
 942 };
 943
 944 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
 945                          u64 parent_transid)
 946 {
 947         struct extent_buffer *buf = NULL;
 948         struct inode *btree_inode = root->fs_info->btree_inode;
 949         int ret = 0;
 950
 951         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 952         if (!buf)
 953                 return 0;
 954         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
 955                                  buf, 0, 0, btree_get_extent, 0);
 956         free_extent_buffer(buf);
 957         return ret;
 958 }
 959
 960 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
 961                                             u64 bytenr, u32 blocksize)
 962 {
 963         struct inode *btree_inode = root->fs_info->btree_inode;
 964         struct extent_buffer *eb;
 965         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 966                                 bytenr, blocksize, GFP_NOFS);
 967         return eb;
 968 }
 969
 970 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
 971                                                  u64 bytenr, u32 blocksize)
 972 {
 973         struct inode *btree_inode = root->fs_info->btree_inode;
 974         struct extent_buffer *eb;
 975
 976         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 977                                  bytenr, blocksize, NULL, GFP_NOFS);
 978         return eb;
 979 }
 980
 981
 982 int btrfs_write_tree_block(struct extent_buffer *buf)
 983 {
 984         return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
 985                                         buf->start + buf->len - 1);
 986 }
 987
 988 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
 989 {
 990         return filemap_fdatawait_range(buf->first_page->mapping,
 991                                        buf->start, buf->start + buf->len - 1);
 992 }
 993
 994 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
 995                                       u32 blocksize, u64 parent_transid)
 996 {
 997         struct extent_buffer *buf = NULL;
 998         int ret;
 999
1000         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1001         if (!buf)
1002                 return NULL;
1003
1004         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1005
1006         if (ret == 0)
1007                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
1008         return buf;
1009
1010 }
1011
1012 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1013                      struct extent_buffer *buf)
1014 {
1015         struct inode *btree_inode = root->fs_info->btree_inode;
1016         if (btrfs_header_generation(buf) ==
1017             root->fs_info->running_transaction->transid) {
1018                 btrfs_assert_tree_locked(buf);
1019
1020                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1021                         spin_lock(&root->fs_info->delalloc_lock);
1022                         if (root->fs_info->dirty_metadata_bytes >= buf->len)
1023                                 root->fs_info->dirty_metadata_bytes -= buf->len;
1024                         else
1025                                 WARN_ON(1);
1026                         spin_unlock(&root->fs_info->delalloc_lock);
1027                 }
1028
1029                 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1030                 btrfs_set_lock_blocking(buf);
1031                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
1032                                           buf);
1033         }
1034         return 0;
1035 }
1036
1037 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1038                         u32 stripesize, struct btrfs_root *root,
1039                         struct btrfs_fs_info *fs_info,
1040                         u64 objectid)
1041 {
1042         root->node = NULL;
1043         root->commit_root = NULL;
1044         root->sectorsize = sectorsize;
1045         root->nodesize = nodesize;
1046         root->leafsize = leafsize;
1047         root->stripesize = stripesize;
1048         root->ref_cows = 0;
1049         root->track_dirty = 0;
1050         root->in_radix = 0;
1051         root->orphan_item_inserted = 0;
1052         root->orphan_cleanup_state = 0;
1053
1054         root->fs_info = fs_info;
1055         root->objectid = objectid;
1056         root->last_trans = 0;
1057         root->highest_objectid = 0;
1058         root->name = NULL;
1059         root->in_sysfs = 0;
1060         root->inode_tree = RB_ROOT;
1061         root->block_rsv = NULL;
1062         root->orphan_block_rsv = NULL;
1063
1064         INIT_LIST_HEAD(&root->dirty_list);
1065         INIT_LIST_HEAD(&root->orphan_list);
1066         INIT_LIST_HEAD(&root->root_list);
1067         spin_lock_init(&root->node_lock);
1068         spin_lock_init(&root->orphan_lock);
1069         spin_lock_init(&root->inode_lock);
1070         spin_lock_init(&root->accounting_lock);
1071         mutex_init(&root->objectid_mutex);
1072         mutex_init(&root->log_mutex);
1073         init_waitqueue_head(&root->log_writer_wait);
1074         init_waitqueue_head(&root->log_commit_wait[0]);
1075         init_waitqueue_head(&root->log_commit_wait[1]);
1076         atomic_set(&root->log_commit[0], 0);
1077         atomic_set(&root->log_commit[1], 0);
1078         atomic_set(&root->log_writers, 0);
1079         root->log_batch = 0;
1080         root->log_transid = 0;
1081         root->last_log_commit = 0;
1082         extent_io_tree_init(&root->dirty_log_pages,
1083                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1084
1085         memset(&root->root_key, 0, sizeof(root->root_key));
1086         memset(&root->root_item, 0, sizeof(root->root_item));
1087         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1088         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1089         root->defrag_trans_start = fs_info->generation;
1090         init_completion(&root->kobj_unregister);
1091         root->defrag_running = 0;
1092         root->root_key.objectid = objectid;
1093         root->anon_super.s_root = NULL;
1094         root->anon_super.s_dev = 0;
1095         INIT_LIST_HEAD(&root->anon_super.s_list);
1096         INIT_LIST_HEAD(&root->anon_super.s_instances);
1097         init_rwsem(&root->anon_super.s_umount);
1098
1099         return 0;
1100 }
1101
1102 static int find_and_setup_root(struct btrfs_root *tree_root,
1103                                struct btrfs_fs_info *fs_info,
1104                                u64 objectid,
1105                                struct btrfs_root *root)
1106 {
1107         int ret;
1108         u32 blocksize;
1109         u64 generation;
1110
1111         __setup_root(tree_root->nodesize, tree_root->leafsize,
1112                      tree_root->sectorsize, tree_root->stripesize,
1113                      root, fs_info, objectid);
1114         ret = btrfs_find_last_root(tree_root, objectid,
1115                                    &root->root_item, &root->root_key);
1116         if (ret > 0)
1117                 return -ENOENT;
1118         BUG_ON(ret);
1119
1120         generation = btrfs_root_generation(&root->root_item);
1121         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1122         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1123                                      blocksize, generation);
1124         if (!root->node || !btrfs_buffer_uptodate(root->node, generation)) {
1125                 free_extent_buffer(root->node);
1126                 return -EIO;
1127         }
1128         root->commit_root = btrfs_root_node(root);
1129         return 0;
1130 }
1131
1132 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1133                                          struct btrfs_fs_info *fs_info)
1134 {
1135         struct btrfs_root *root;
1136         struct btrfs_root *tree_root = fs_info->tree_root;
1137         struct extent_buffer *leaf;
1138
1139         root = kzalloc(sizeof(*root), GFP_NOFS);
1140         if (!root)
1141                 return ERR_PTR(-ENOMEM);
1142
1143         __setup_root(tree_root->nodesize, tree_root->leafsize,
1144                      tree_root->sectorsize, tree_root->stripesize,
1145                      root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1146
1147         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1148         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1149         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1150         /*
1151          * log trees do not get reference counted because they go away
1152          * before a real commit is actually done.  They do store pointers
1153          * to file data extents, and those reference counts still get
1154          * updated (along with back refs to the log tree).
1155          */
1156         root->ref_cows = 0;
1157
1158         leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1159                                       BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
1160         if (IS_ERR(leaf)) {
1161                 kfree(root);
1162                 return ERR_CAST(leaf);
1163         }
1164
1165         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1166         btrfs_set_header_bytenr(leaf, leaf->start);
1167         btrfs_set_header_generation(leaf, trans->transid);
1168         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1169         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1170         root->node = leaf;
1171
1172         write_extent_buffer(root->node, root->fs_info->fsid,
1173                             (unsigned long)btrfs_header_fsid(root->node),
1174                             BTRFS_FSID_SIZE);
1175         btrfs_mark_buffer_dirty(root->node);
1176         btrfs_tree_unlock(root->node);
1177         return root;
1178 }
1179
1180 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1181                              struct btrfs_fs_info *fs_info)
1182 {
1183         struct btrfs_root *log_root;
1184
1185         log_root = alloc_log_tree(trans, fs_info);
1186         if (IS_ERR(log_root))
1187                 return PTR_ERR(log_root);
1188         WARN_ON(fs_info->log_root_tree);
1189         fs_info->log_root_tree = log_root;
1190         return 0;
1191 }
1192
1193 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1194                        struct btrfs_root *root)
1195 {
1196         struct btrfs_root *log_root;
1197         struct btrfs_inode_item *inode_item;
1198
1199         log_root = alloc_log_tree(trans, root->fs_info);
1200         if (IS_ERR(log_root))
1201                 return PTR_ERR(log_root);
1202
1203         log_root->last_trans = trans->transid;
1204         log_root->root_key.offset = root->root_key.objectid;
1205
1206         inode_item = &log_root->root_item.inode;
1207         inode_item->generation = cpu_to_le64(1);
1208         inode_item->size = cpu_to_le64(3);
1209         inode_item->nlink = cpu_to_le32(1);
1210         inode_item->nbytes = cpu_to_le64(root->leafsize);
1211         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1212
1213         btrfs_set_root_node(&log_root->root_item, log_root->node);
1214
1215         WARN_ON(root->log_root);
1216         root->log_root = log_root;
1217         root->log_transid = 0;
1218         root->last_log_commit = 0;
1219         return 0;
1220 }
1221
1222 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1223                                                struct btrfs_key *location)
1224 {
1225         struct btrfs_root *root;
1226         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1227         struct btrfs_path *path;
1228         struct extent_buffer *l;
1229         u64 generation;
1230         u32 blocksize;
1231         int ret = 0;
1232
1233         root = kzalloc(sizeof(*root), GFP_NOFS);
1234         if (!root)
1235                 return ERR_PTR(-ENOMEM);
1236         if (location->offset == (u64)-1) {
1237                 ret = find_and_setup_root(tree_root, fs_info,
1238                                           location->objectid, root);
1239                 if (ret) {
1240                         kfree(root);
1241                         return ERR_PTR(ret);
1242                 }
1243                 goto out;
1244         }
1245
1246         __setup_root(tree_root->nodesize, tree_root->leafsize,
1247                      tree_root->sectorsize, tree_root->stripesize,
1248                      root, fs_info, location->objectid);
1249
1250         path = btrfs_alloc_path();
1251         if (!path) {
1252                 kfree(root);
1253                 return ERR_PTR(-ENOMEM);
1254         }
1255         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1256         if (ret == 0) {
1257                 l = path->nodes[0];
1258                 read_extent_buffer(l, &root->root_item,
1259                                 btrfs_item_ptr_offset(l, path->slots[0]),
1260                                 sizeof(root->root_item));
1261                 memcpy(&root->root_key, location, sizeof(*location));
1262         }
1263         btrfs_free_path(path);
1264         if (ret) {
1265                 kfree(root);
1266                 if (ret > 0)
1267                         ret = -ENOENT;
1268                 return ERR_PTR(ret);
1269         }
1270
1271         generation = btrfs_root_generation(&root->root_item);
1272         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1273         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1274                                      blocksize, generation);
1275         root->commit_root = btrfs_root_node(root);
1276         BUG_ON(!root->node);
1277 out:
1278         if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1279                 root->ref_cows = 1;
1280                 btrfs_check_and_init_root_item(&root->root_item);
1281         }
1282
1283         return root;
1284 }
1285
1286 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1287                                         u64 root_objectid)
1288 {
1289         struct btrfs_root *root;
1290
1291         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1292                 return fs_info->tree_root;
1293         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1294                 return fs_info->extent_root;
1295
1296         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1297                                  (unsigned long)root_objectid);
1298         return root;
1299 }
1300
1301 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1302                                               struct btrfs_key *location)
1303 {
1304         struct btrfs_root *root;
1305         int ret;
1306
1307         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1308                 return fs_info->tree_root;
1309         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1310                 return fs_info->extent_root;
1311         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1312                 return fs_info->chunk_root;
1313         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1314                 return fs_info->dev_root;
1315         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1316                 return fs_info->csum_root;
1317 again:
1318         spin_lock(&fs_info->fs_roots_radix_lock);
1319         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1320                                  (unsigned long)location->objectid);
1321         spin_unlock(&fs_info->fs_roots_radix_lock);
1322         if (root)
1323                 return root;
1324
1325         root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1326         if (IS_ERR(root))
1327                 return root;
1328
1329         set_anon_super(&root->anon_super, NULL);
1330
1331         if (btrfs_root_refs(&root->root_item) == 0) {
1332                 ret = -ENOENT;
1333                 goto fail;
1334         }
1335
1336         ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1337         if (ret < 0)
1338                 goto fail;
1339         if (ret == 0)
1340                 root->orphan_item_inserted = 1;
1341
1342         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1343         if (ret)
1344                 goto fail;
1345
1346         spin_lock(&fs_info->fs_roots_radix_lock);
1347         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1348                                 (unsigned long)root->root_key.objectid,
1349                                 root);
1350         if (ret == 0)
1351                 root->in_radix = 1;
1352
1353         spin_unlock(&fs_info->fs_roots_radix_lock);
1354         radix_tree_preload_end();
1355         if (ret) {
1356                 if (ret == -EEXIST) {
1357                         free_fs_root(root);
1358                         goto again;
1359                 }
1360                 goto fail;
1361         }
1362
1363         ret = btrfs_find_dead_roots(fs_info->tree_root,
1364                                     root->root_key.objectid);
1365         WARN_ON(ret);
1366         return root;
1367 fail:
1368         free_fs_root(root);
1369         return ERR_PTR(ret);
1370 }
1371
1372 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1373                                       struct btrfs_key *location,
1374                                       const char *name, int namelen)
1375 {
1376         return btrfs_read_fs_root_no_name(fs_info, location);
1377 #if 0
1378         struct btrfs_root *root;
1379         int ret;
1380
1381         root = btrfs_read_fs_root_no_name(fs_info, location);
1382         if (!root)
1383                 return NULL;
1384
1385         if (root->in_sysfs)
1386                 return root;
1387
1388         ret = btrfs_set_root_name(root, name, namelen);
1389         if (ret) {
1390                 free_extent_buffer(root->node);
1391                 kfree(root);
1392                 return ERR_PTR(ret);
1393         }
1394
1395         ret = btrfs_sysfs_add_root(root);
1396         if (ret) {
1397                 free_extent_buffer(root->node);
1398                 kfree(root->name);
1399                 kfree(root);
1400                 return ERR_PTR(ret);
1401         }
1402         root->in_sysfs = 1;
1403         return root;
1404 #endif
1405 }
1406
1407 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1408 {
1409         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1410         int ret = 0;
1411         struct btrfs_device *device;
1412         struct backing_dev_info *bdi;
1413
1414         list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1415                 if (!device->bdev)
1416                         continue;
1417                 bdi = blk_get_backing_dev_info(device->bdev);
1418                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1419                         ret = 1;
1420                         break;
1421                 }
1422         }
1423         return ret;
1424 }
1425
1426 /*
1427  * If this fails, caller must call bdi_destroy() to get rid of the
1428  * bdi again.
1429  */
1430 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1431 {
1432         int err;
1433
1434         bdi->capabilities = BDI_CAP_MAP_COPY;
1435         err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1436         if (err)
1437                 return err;
1438
1439         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1440         bdi->congested_fn       = btrfs_congested_fn;
1441         bdi->congested_data     = info;
1442         return 0;
1443 }
1444
1445 static int bio_ready_for_csum(struct bio *bio)
1446 {
1447         u64 length = 0;
1448         u64 buf_len = 0;
1449         u64 start = 0;
1450         struct page *page;
1451         struct extent_io_tree *io_tree = NULL;
1452         struct bio_vec *bvec;
1453         int i;
1454         int ret;
1455
1456         bio_for_each_segment(bvec, bio, i) {
1457                 page = bvec->bv_page;
1458                 if (page->private == EXTENT_PAGE_PRIVATE) {
1459                         length += bvec->bv_len;
1460                         continue;
1461                 }
1462                 if (!page->private) {
1463                         length += bvec->bv_len;
1464                         continue;
1465                 }
1466                 length = bvec->bv_len;
1467                 buf_len = page->private >> 2;
1468                 start = page_offset(page) + bvec->bv_offset;
1469                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1470         }
1471         /* are we fully contained in this bio? */
1472         if (buf_len <= length)
1473                 return 1;
1474
1475         ret = extent_range_uptodate(io_tree, start + length,
1476                                     start + buf_len - 1);
1477         return ret;
1478 }
1479
1480 /*
1481  * called by the kthread helper functions to finally call the bio end_io
1482  * functions.  This is where read checksum verification actually happens
1483  */
1484 static void end_workqueue_fn(struct btrfs_work *work)
1485 {
1486         struct bio *bio;
1487         struct end_io_wq *end_io_wq;
1488         struct btrfs_fs_info *fs_info;
1489         int error;
1490
1491         end_io_wq = container_of(work, struct end_io_wq, work);
1492         bio = end_io_wq->bio;
1493         fs_info = end_io_wq->info;
1494
1495         /* metadata bio reads are special because the whole tree block must
1496          * be checksummed at once.  This makes sure the entire block is in
1497          * ram and up to date before trying to verify things.  For
1498          * blocksize <= pagesize, it is basically a noop
1499          */
1500         if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata &&
1501             !bio_ready_for_csum(bio)) {
1502                 btrfs_queue_worker(&fs_info->endio_meta_workers,
1503                                    &end_io_wq->work);
1504                 return;
1505         }
1506         error = end_io_wq->error;
1507         bio->bi_private = end_io_wq->private;
1508         bio->bi_end_io = end_io_wq->end_io;
1509         kfree(end_io_wq);
1510         bio_endio(bio, error);
1511 }
1512
1513 static int cleaner_kthread(void *arg)
1514 {
1515         struct btrfs_root *root = arg;
1516
1517         do {
1518                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1519
1520                 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1521                     mutex_trylock(&root->fs_info->cleaner_mutex)) {
1522                         btrfs_run_delayed_iputs(root);
1523                         btrfs_clean_old_snapshots(root);
1524                         mutex_unlock(&root->fs_info->cleaner_mutex);
1525                 }
1526
1527                 if (freezing(current)) {
1528                         refrigerator();
1529                 } else {
1530                         set_current_state(TASK_INTERRUPTIBLE);
1531                         if (!kthread_should_stop())
1532                                 schedule();
1533                         __set_current_state(TASK_RUNNING);
1534                 }
1535         } while (!kthread_should_stop());
1536         return 0;
1537 }
1538
1539 static int transaction_kthread(void *arg)
1540 {
1541         struct btrfs_root *root = arg;
1542         struct btrfs_trans_handle *trans;
1543         struct btrfs_transaction *cur;
1544         u64 transid;
1545         unsigned long now;
1546         unsigned long delay;
1547         int ret;
1548
1549         do {
1550                 delay = HZ * 30;
1551                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1552                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1553
1554                 spin_lock(&root->fs_info->new_trans_lock);
1555                 cur = root->fs_info->running_transaction;
1556                 if (!cur) {
1557                         spin_unlock(&root->fs_info->new_trans_lock);
1558                         goto sleep;
1559                 }
1560
1561                 now = get_seconds();
1562                 if (!cur->blocked &&
1563                     (now < cur->start_time || now - cur->start_time < 30)) {
1564                         spin_unlock(&root->fs_info->new_trans_lock);
1565                         delay = HZ * 5;
1566                         goto sleep;
1567                 }
1568                 transid = cur->transid;
1569                 spin_unlock(&root->fs_info->new_trans_lock);
1570
1571                 trans = btrfs_join_transaction(root, 1);
1572                 BUG_ON(IS_ERR(trans));
1573                 if (transid == trans->transid) {
1574                         ret = btrfs_commit_transaction(trans, root);
1575                         BUG_ON(ret);
1576                 } else {
1577                         btrfs_end_transaction(trans, root);
1578                 }
1579 sleep:
1580                 wake_up_process(root->fs_info->cleaner_kthread);
1581                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1582
1583                 if (freezing(current)) {
1584                         refrigerator();
1585                 } else {
1586                         set_current_state(TASK_INTERRUPTIBLE);
1587                         if (!kthread_should_stop() &&
1588                             !btrfs_transaction_blocked(root->fs_info))
1589                                 schedule_timeout(delay);
1590                         __set_current_state(TASK_RUNNING);
1591                 }
1592         } while (!kthread_should_stop());
1593         return 0;
1594 }
1595
1596 struct btrfs_root *open_ctree(struct super_block *sb,
1597                               struct btrfs_fs_devices *fs_devices,
1598                               char *options)
1599 {
1600         u32 sectorsize;
1601         u32 nodesize;
1602         u32 leafsize;
1603         u32 blocksize;
1604         u32 stripesize;
1605         u64 generation;
1606         u64 features;
1607         struct btrfs_key location;
1608         struct buffer_head *bh;
1609         struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1610                                                  GFP_NOFS);
1611         struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1612                                                  GFP_NOFS);
1613         struct btrfs_root *tree_root = btrfs_sb(sb);
1614         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1615         struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1616                                                 GFP_NOFS);
1617         struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1618                                               GFP_NOFS);
1619         struct btrfs_root *log_tree_root;
1620
1621         int ret;
1622         int err = -EINVAL;
1623
1624         struct btrfs_super_block *disk_super;
1625
1626         if (!extent_root || !tree_root || !fs_info ||
1627             !chunk_root || !dev_root || !csum_root) {
1628                 err = -ENOMEM;
1629                 goto fail;
1630         }
1631
1632         ret = init_srcu_struct(&fs_info->subvol_srcu);
1633         if (ret) {
1634                 err = ret;
1635                 goto fail;
1636         }
1637
1638         ret = setup_bdi(fs_info, &fs_info->bdi);
1639         if (ret) {
1640                 err = ret;
1641                 goto fail_srcu;
1642         }
1643
1644         fs_info->btree_inode = new_inode(sb);
1645         if (!fs_info->btree_inode) {
1646                 err = -ENOMEM;
1647                 goto fail_bdi;
1648         }
1649
1650         fs_info->btree_inode->i_mapping->flags &= ~__GFP_FS;
1651
1652         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1653         INIT_LIST_HEAD(&fs_info->trans_list);
1654         INIT_LIST_HEAD(&fs_info->dead_roots);
1655         INIT_LIST_HEAD(&fs_info->delayed_iputs);
1656         INIT_LIST_HEAD(&fs_info->hashers);
1657         INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1658         INIT_LIST_HEAD(&fs_info->ordered_operations);
1659         INIT_LIST_HEAD(&fs_info->caching_block_groups);
1660         spin_lock_init(&fs_info->delalloc_lock);
1661         spin_lock_init(&fs_info->new_trans_lock);
1662         spin_lock_init(&fs_info->ref_cache_lock);
1663         spin_lock_init(&fs_info->fs_roots_radix_lock);
1664         spin_lock_init(&fs_info->delayed_iput_lock);
1665
1666         init_completion(&fs_info->kobj_unregister);
1667         fs_info->tree_root = tree_root;
1668         fs_info->extent_root = extent_root;
1669         fs_info->csum_root = csum_root;
1670         fs_info->chunk_root = chunk_root;
1671         fs_info->dev_root = dev_root;
1672         fs_info->fs_devices = fs_devices;
1673         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1674         INIT_LIST_HEAD(&fs_info->space_info);
1675         btrfs_mapping_init(&fs_info->mapping_tree);
1676         btrfs_init_block_rsv(&fs_info->global_block_rsv);
1677         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1678         btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1679         btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1680         btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1681         INIT_LIST_HEAD(&fs_info->durable_block_rsv_list);
1682         mutex_init(&fs_info->durable_block_rsv_mutex);
1683         atomic_set(&fs_info->nr_async_submits, 0);
1684         atomic_set(&fs_info->async_delalloc_pages, 0);
1685         atomic_set(&fs_info->async_submit_draining, 0);
1686         atomic_set(&fs_info->nr_async_bios, 0);
1687         fs_info->sb = sb;
1688         fs_info->max_inline = 8192 * 1024;
1689         fs_info->metadata_ratio = 0;
1690
1691         fs_info->thread_pool_size = min_t(unsigned long,
1692                                           num_online_cpus() + 2, 8);
1693
1694         INIT_LIST_HEAD(&fs_info->ordered_extents);
1695         spin_lock_init(&fs_info->ordered_extent_lock);
1696
1697         sb->s_blocksize = 4096;
1698         sb->s_blocksize_bits = blksize_bits(4096);
1699         sb->s_bdi = &fs_info->bdi;
1700
1701         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1702         fs_info->btree_inode->i_nlink = 1;
1703         /*
1704          * we set the i_size on the btree inode to the max possible int.
1705          * the real end of the address space is determined by all of
1706          * the devices in the system
1707          */
1708         fs_info->btree_inode->i_size = OFFSET_MAX;
1709         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1710         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1711
1712         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
1713         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1714                              fs_info->btree_inode->i_mapping,
1715                              GFP_NOFS);
1716         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1717                              GFP_NOFS);
1718
1719         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1720
1721         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1722         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1723                sizeof(struct btrfs_key));
1724         BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
1725         insert_inode_hash(fs_info->btree_inode);
1726
1727         spin_lock_init(&fs_info->block_group_cache_lock);
1728         fs_info->block_group_cache_tree = RB_ROOT;
1729
1730         extent_io_tree_init(&fs_info->freed_extents[0],
1731                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1732         extent_io_tree_init(&fs_info->freed_extents[1],
1733                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1734         fs_info->pinned_extents = &fs_info->freed_extents[0];
1735         fs_info->do_barriers = 1;
1736
1737
1738         mutex_init(&fs_info->trans_mutex);
1739         mutex_init(&fs_info->ordered_operations_mutex);
1740         mutex_init(&fs_info->tree_log_mutex);
1741         mutex_init(&fs_info->chunk_mutex);
1742         mutex_init(&fs_info->transaction_kthread_mutex);
1743         mutex_init(&fs_info->cleaner_mutex);
1744         mutex_init(&fs_info->volume_mutex);
1745         init_rwsem(&fs_info->extent_commit_sem);
1746         init_rwsem(&fs_info->cleanup_work_sem);
1747         init_rwsem(&fs_info->subvol_sem);
1748
1749         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1750         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1751
1752         init_waitqueue_head(&fs_info->transaction_throttle);
1753         init_waitqueue_head(&fs_info->transaction_wait);
1754         init_waitqueue_head(&fs_info->transaction_blocked_wait);
1755         init_waitqueue_head(&fs_info->async_submit_wait);
1756
1757         __setup_root(4096, 4096, 4096, 4096, tree_root,
1758                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1759
1760         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1761         if (!bh) {
1762                 err = -EINVAL;
1763                 goto fail_iput;
1764         }
1765
1766         memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1767         memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1768                sizeof(fs_info->super_for_commit));
1769         brelse(bh);
1770
1771         memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1772
1773         disk_super = &fs_info->super_copy;
1774         if (!btrfs_super_root(disk_super))
1775                 goto fail_iput;
1776
1777         /* check FS state, whether FS is broken. */
1778         fs_info->fs_state |= btrfs_super_flags(disk_super);
1779
1780         btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
1781
1782         /*
1783          * In the long term, we'll store the compression type in the super
1784          * block, and it'll be used for per file compression control.
1785          */
1786         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
1787
1788         ret = btrfs_parse_options(tree_root, options);
1789         if (ret) {
1790                 err = ret;
1791                 goto fail_iput;
1792         }
1793
1794         features = btrfs_super_incompat_flags(disk_super) &
1795                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1796         if (features) {
1797                 printk(KERN_ERR "BTRFS: couldn't mount because of "
1798                        "unsupported optional features (%Lx).\n",
1799                        (unsigned long long)features);
1800                 err = -EINVAL;
1801                 goto fail_iput;
1802         }
1803
1804         features = btrfs_super_incompat_flags(disk_super);
1805         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
1806         if (tree_root->fs_info->compress_type & BTRFS_COMPRESS_LZO)
1807                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
1808         btrfs_set_super_incompat_flags(disk_super, features);
1809
1810         features = btrfs_super_compat_ro_flags(disk_super) &
1811                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1812         if (!(sb->s_flags & MS_RDONLY) && features) {
1813                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1814                        "unsupported option features (%Lx).\n",
1815                        (unsigned long long)features);
1816                 err = -EINVAL;
1817                 goto fail_iput;
1818         }
1819
1820         btrfs_init_workers(&fs_info->generic_worker,
1821                            "genwork", 1, NULL);
1822
1823         btrfs_init_workers(&fs_info->workers, "worker",
1824                            fs_info->thread_pool_size,
1825                            &fs_info->generic_worker);
1826
1827         btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1828                            fs_info->thread_pool_size,
1829                            &fs_info->generic_worker);
1830
1831         btrfs_init_workers(&fs_info->submit_workers, "submit",
1832                            min_t(u64, fs_devices->num_devices,
1833                            fs_info->thread_pool_size),
1834                            &fs_info->generic_worker);
1835
1836         /* a higher idle thresh on the submit workers makes it much more
1837          * likely that bios will be send down in a sane order to the
1838          * devices
1839          */
1840         fs_info->submit_workers.idle_thresh = 64;
1841
1842         fs_info->workers.idle_thresh = 16;
1843         fs_info->workers.ordered = 1;
1844
1845         fs_info->delalloc_workers.idle_thresh = 2;
1846         fs_info->delalloc_workers.ordered = 1;
1847
1848         btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
1849                            &fs_info->generic_worker);
1850         btrfs_init_workers(&fs_info->endio_workers, "endio",
1851                            fs_info->thread_pool_size,
1852                            &fs_info->generic_worker);
1853         btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1854                            fs_info->thread_pool_size,
1855                            &fs_info->generic_worker);
1856         btrfs_init_workers(&fs_info->endio_meta_write_workers,
1857                            "endio-meta-write", fs_info->thread_pool_size,
1858                            &fs_info->generic_worker);
1859         btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1860                            fs_info->thread_pool_size,
1861                            &fs_info->generic_worker);
1862         btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
1863                            1, &fs_info->generic_worker);
1864
1865         /*
1866          * endios are largely parallel and should have a very
1867          * low idle thresh
1868          */
1869         fs_info->endio_workers.idle_thresh = 4;
1870         fs_info->endio_meta_workers.idle_thresh = 4;
1871
1872         fs_info->endio_write_workers.idle_thresh = 2;
1873         fs_info->endio_meta_write_workers.idle_thresh = 2;
1874
1875         btrfs_start_workers(&fs_info->workers, 1);
1876         btrfs_start_workers(&fs_info->generic_worker, 1);
1877         btrfs_start_workers(&fs_info->submit_workers, 1);
1878         btrfs_start_workers(&fs_info->delalloc_workers, 1);
1879         btrfs_start_workers(&fs_info->fixup_workers, 1);
1880         btrfs_start_workers(&fs_info->endio_workers, 1);
1881         btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1882         btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1883         btrfs_start_workers(&fs_info->endio_write_workers, 1);
1884         btrfs_start_workers(&fs_info->endio_freespace_worker, 1);
1885
1886         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1887         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1888                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1889
1890         nodesize = btrfs_super_nodesize(disk_super);
1891         leafsize = btrfs_super_leafsize(disk_super);
1892         sectorsize = btrfs_super_sectorsize(disk_super);
1893         stripesize = btrfs_super_stripesize(disk_super);
1894         tree_root->nodesize = nodesize;
1895         tree_root->leafsize = leafsize;
1896         tree_root->sectorsize = sectorsize;
1897         tree_root->stripesize = stripesize;
1898
1899         sb->s_blocksize = sectorsize;
1900         sb->s_blocksize_bits = blksize_bits(sectorsize);
1901
1902         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1903                     sizeof(disk_super->magic))) {
1904                 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
1905                 goto fail_sb_buffer;
1906         }
1907
1908         mutex_lock(&fs_info->chunk_mutex);
1909         ret = btrfs_read_sys_array(tree_root);
1910         mutex_unlock(&fs_info->chunk_mutex);
1911         if (ret) {
1912                 printk(KERN_WARNING "btrfs: failed to read the system "
1913                        "array on %s\n", sb->s_id);
1914                 goto fail_sb_buffer;
1915         }
1916
1917         blocksize = btrfs_level_size(tree_root,
1918                                      btrfs_super_chunk_root_level(disk_super));
1919         generation = btrfs_super_chunk_root_generation(disk_super);
1920
1921         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1922                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1923
1924         chunk_root->node = read_tree_block(chunk_root,
1925                                            btrfs_super_chunk_root(disk_super),
1926                                            blocksize, generation);
1927         BUG_ON(!chunk_root->node);
1928         if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
1929                 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
1930                        sb->s_id);
1931                 goto fail_chunk_root;
1932         }
1933         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
1934         chunk_root->commit_root = btrfs_root_node(chunk_root);
1935
1936         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1937            (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1938            BTRFS_UUID_SIZE);
1939
1940         mutex_lock(&fs_info->chunk_mutex);
1941         ret = btrfs_read_chunk_tree(chunk_root);
1942         mutex_unlock(&fs_info->chunk_mutex);
1943         if (ret) {
1944                 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
1945                        sb->s_id);
1946                 goto fail_chunk_root;
1947         }
1948
1949         btrfs_close_extra_devices(fs_devices);
1950
1951         blocksize = btrfs_level_size(tree_root,
1952                                      btrfs_super_root_level(disk_super));
1953         generation = btrfs_super_generation(disk_super);
1954
1955         tree_root->node = read_tree_block(tree_root,
1956                                           btrfs_super_root(disk_super),
1957                                           blocksize, generation);
1958         if (!tree_root->node)
1959                 goto fail_chunk_root;
1960         if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
1961                 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
1962                        sb->s_id);
1963                 goto fail_tree_root;
1964         }
1965         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
1966         tree_root->commit_root = btrfs_root_node(tree_root);
1967
1968         ret = find_and_setup_root(tree_root, fs_info,
1969                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1970         if (ret)
1971                 goto fail_tree_root;
1972         extent_root->track_dirty = 1;
1973
1974         ret = find_and_setup_root(tree_root, fs_info,
1975                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
1976         if (ret)
1977                 goto fail_extent_root;
1978         dev_root->track_dirty = 1;
1979
1980         ret = find_and_setup_root(tree_root, fs_info,
1981                                   BTRFS_CSUM_TREE_OBJECTID, csum_root);
1982         if (ret)
1983                 goto fail_dev_root;
1984
1985         csum_root->track_dirty = 1;
1986
1987         fs_info->generation = generation;
1988         fs_info->last_trans_committed = generation;
1989         fs_info->data_alloc_profile = (u64)-1;
1990         fs_info->metadata_alloc_profile = (u64)-1;
1991         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1992
1993         ret = btrfs_init_space_info(fs_info);
1994         if (ret) {
1995                 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
1996                 goto fail_block_groups;
1997         }
1998
1999         ret = btrfs_read_block_groups(extent_root);
2000         if (ret) {
2001                 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2002                 goto fail_block_groups;
2003         }
2004
2005         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2006                                                "btrfs-cleaner");
2007         if (IS_ERR(fs_info->cleaner_kthread))
2008                 goto fail_block_groups;
2009
2010         fs_info->transaction_kthread = kthread_run(transaction_kthread,
2011                                                    tree_root,
2012                                                    "btrfs-transaction");
2013         if (IS_ERR(fs_info->transaction_kthread))
2014                 goto fail_cleaner;
2015
2016         if (!btrfs_test_opt(tree_root, SSD) &&
2017             !btrfs_test_opt(tree_root, NOSSD) &&
2018             !fs_info->fs_devices->rotating) {
2019                 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2020                        "mode\n");
2021                 btrfs_set_opt(fs_info->mount_opt, SSD);
2022         }
2023
2024         /* do not make disk changes in broken FS */
2025         if (btrfs_super_log_root(disk_super) != 0 &&
2026             !(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) {
2027                 u64 bytenr = btrfs_super_log_root(disk_super);
2028
2029                 if (fs_devices->rw_devices == 0) {
2030                         printk(KERN_WARNING "Btrfs log replay required "
2031                                "on RO media\n");
2032                         err = -EIO;
2033                         goto fail_trans_kthread;
2034                 }
2035                 blocksize =
2036                      btrfs_level_size(tree_root,
2037                                       btrfs_super_log_root_level(disk_super));
2038
2039                 log_tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
2040                 if (!log_tree_root) {
2041                         err = -ENOMEM;
2042                         goto fail_trans_kthread;
2043                 }
2044
2045                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2046                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2047
2048                 log_tree_root->node = read_tree_block(tree_root, bytenr,
2049                                                       blocksize,
2050                                                       generation + 1);
2051                 ret = btrfs_recover_log_trees(log_tree_root);
2052                 BUG_ON(ret);
2053
2054                 if (sb->s_flags & MS_RDONLY) {
2055                         ret =  btrfs_commit_super(tree_root);
2056                         BUG_ON(ret);
2057                 }
2058         }
2059
2060         ret = btrfs_find_orphan_roots(tree_root);
2061         BUG_ON(ret);
2062
2063         if (!(sb->s_flags & MS_RDONLY)) {
2064                 ret = btrfs_cleanup_fs_roots(fs_info);
2065                 BUG_ON(ret);
2066
2067                 ret = btrfs_recover_relocation(tree_root);
2068                 if (ret < 0) {
2069                         printk(KERN_WARNING
2070                                "btrfs: failed to recover relocation\n");
2071                         err = -EINVAL;
2072                         goto fail_trans_kthread;
2073                 }
2074         }
2075
2076         location.objectid = BTRFS_FS_TREE_OBJECTID;
2077         location.type = BTRFS_ROOT_ITEM_KEY;
2078         location.offset = (u64)-1;
2079
2080         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2081         if (!fs_info->fs_root)
2082                 goto fail_trans_kthread;
2083         if (IS_ERR(fs_info->fs_root)) {
2084                 err = PTR_ERR(fs_info->fs_root);
2085                 goto fail_trans_kthread;
2086         }
2087
2088         if (!(sb->s_flags & MS_RDONLY)) {
2089                 down_read(&fs_info->cleanup_work_sem);
2090                 err = btrfs_orphan_cleanup(fs_info->fs_root);
2091                 if (!err)
2092                         err = btrfs_orphan_cleanup(fs_info->tree_root);
2093                 up_read(&fs_info->cleanup_work_sem);
2094                 if (err) {
2095                         close_ctree(tree_root);
2096                         return ERR_PTR(err);
2097                 }
2098         }
2099
2100         return tree_root;
2101
2102 fail_trans_kthread:
2103         kthread_stop(fs_info->transaction_kthread);
2104 fail_cleaner:
2105         kthread_stop(fs_info->cleaner_kthread);
2106
2107         /*
2108          * make sure we're done with the btree inode before we stop our
2109          * kthreads
2110          */
2111         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2112         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2113
2114 fail_block_groups:
2115         btrfs_free_block_groups(fs_info);
2116         free_extent_buffer(csum_root->node);
2117         free_extent_buffer(csum_root->commit_root);
2118 fail_dev_root:
2119         free_extent_buffer(dev_root->node);
2120         free_extent_buffer(dev_root->commit_root);
2121 fail_extent_root:
2122         free_extent_buffer(extent_root->node);
2123         free_extent_buffer(extent_root->commit_root);
2124 fail_tree_root:
2125         free_extent_buffer(tree_root->node);
2126         free_extent_buffer(tree_root->commit_root);
2127 fail_chunk_root:
2128         free_extent_buffer(chunk_root->node);
2129         free_extent_buffer(chunk_root->commit_root);
2130 fail_sb_buffer:
2131         btrfs_stop_workers(&fs_info->generic_worker);
2132         btrfs_stop_workers(&fs_info->fixup_workers);
2133         btrfs_stop_workers(&fs_info->delalloc_workers);
2134         btrfs_stop_workers(&fs_info->workers);
2135         btrfs_stop_workers(&fs_info->endio_workers);
2136         btrfs_stop_workers(&fs_info->endio_meta_workers);
2137         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2138         btrfs_stop_workers(&fs_info->endio_write_workers);
2139         btrfs_stop_workers(&fs_info->endio_freespace_worker);
2140         btrfs_stop_workers(&fs_info->submit_workers);
2141 fail_iput:
2142         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2143         iput(fs_info->btree_inode);
2144
2145         btrfs_close_devices(fs_info->fs_devices);
2146         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2147 fail_bdi:
2148         bdi_destroy(&fs_info->bdi);
2149 fail_srcu:
2150         cleanup_srcu_struct(&fs_info->subvol_srcu);
2151 fail:
2152         kfree(extent_root);
2153         kfree(tree_root);
2154         kfree(fs_info);
2155         kfree(chunk_root);
2156         kfree(dev_root);
2157         kfree(csum_root);
2158         return ERR_PTR(err);
2159 }
2160
2161 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2162 {
2163         char b[BDEVNAME_SIZE];
2164
2165         if (uptodate) {
2166                 set_buffer_uptodate(bh);
2167         } else {
2168                 if (printk_ratelimit()) {
2169                         printk(KERN_WARNING "lost page write due to "
2170                                         "I/O error on %s\n",
2171                                        bdevname(bh->b_bdev, b));
2172                 }
2173                 /* note, we dont' set_buffer_write_io_error because we have
2174                  * our own ways of dealing with the IO errors
2175                  */
2176                 clear_buffer_uptodate(bh);
2177         }
2178         unlock_buffer(bh);
2179         put_bh(bh);
2180 }
2181
2182 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2183 {
2184         struct buffer_head *bh;
2185         struct buffer_head *latest = NULL;
2186         struct btrfs_super_block *super;
2187         int i;
2188         u64 transid = 0;
2189         u64 bytenr;
2190
2191         /* we would like to check all the supers, but that would make
2192          * a btrfs mount succeed after a mkfs from a different FS.
2193          * So, we need to add a special mount option to scan for
2194          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2195          */
2196         for (i = 0; i < 1; i++) {
2197                 bytenr = btrfs_sb_offset(i);
2198                 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2199                         break;
2200                 bh = __bread(bdev, bytenr / 4096, 4096);
2201                 if (!bh)
2202                         continue;
2203
2204                 super = (struct btrfs_super_block *)bh->b_data;
2205                 if (btrfs_super_bytenr(super) != bytenr ||
2206                     strncmp((char *)(&super->magic), BTRFS_MAGIC,
2207                             sizeof(super->magic))) {
2208                         brelse(bh);
2209                         continue;
2210                 }
2211
2212                 if (!latest || btrfs_super_generation(super) > transid) {
2213                         brelse(latest);
2214                         latest = bh;
2215                         transid = btrfs_super_generation(super);
2216                 } else {
2217                         brelse(bh);
2218                 }
2219         }
2220         return latest;
2221 }
2222
2223 /*
2224  * this should be called twice, once with wait == 0 and
2225  * once with wait == 1.  When wait == 0 is done, all the buffer heads
2226  * we write are pinned.
2227  *
2228  * They are released when wait == 1 is done.
2229  * max_mirrors must be the same for both runs, and it indicates how
2230  * many supers on this one device should be written.
2231  *
2232  * max_mirrors == 0 means to write them all.
2233  */
2234 static int write_dev_supers(struct btrfs_device *device,
2235                             struct btrfs_super_block *sb,
2236                             int do_barriers, int wait, int max_mirrors)
2237 {
2238         struct buffer_head *bh;
2239         int i;
2240         int ret;
2241         int errors = 0;
2242         u32 crc;
2243         u64 bytenr;
2244         int last_barrier = 0;
2245
2246         if (max_mirrors == 0)
2247                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2248
2249         /* make sure only the last submit_bh does a barrier */
2250         if (do_barriers) {
2251                 for (i = 0; i < max_mirrors; i++) {
2252                         bytenr = btrfs_sb_offset(i);
2253                         if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2254                             device->total_bytes)
2255                                 break;
2256                         last_barrier = i;
2257                 }
2258         }
2259
2260         for (i = 0; i < max_mirrors; i++) {
2261                 bytenr = btrfs_sb_offset(i);
2262                 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2263                         break;
2264
2265                 if (wait) {
2266                         bh = __find_get_block(device->bdev, bytenr / 4096,
2267                                               BTRFS_SUPER_INFO_SIZE);
2268                         BUG_ON(!bh);
2269                         wait_on_buffer(bh);
2270                         if (!buffer_uptodate(bh))
2271                                 errors++;
2272
2273                         /* drop our reference */
2274                         brelse(bh);
2275
2276                         /* drop the reference from the wait == 0 run */
2277                         brelse(bh);
2278                         continue;
2279                 } else {
2280                         btrfs_set_super_bytenr(sb, bytenr);
2281
2282                         crc = ~(u32)0;
2283                         crc = btrfs_csum_data(NULL, (char *)sb +
2284                                               BTRFS_CSUM_SIZE, crc,
2285                                               BTRFS_SUPER_INFO_SIZE -
2286                                               BTRFS_CSUM_SIZE);
2287                         btrfs_csum_final(crc, sb->csum);
2288
2289                         /*
2290                          * one reference for us, and we leave it for the
2291                          * caller
2292                          */
2293                         bh = __getblk(device->bdev, bytenr / 4096,
2294                                       BTRFS_SUPER_INFO_SIZE);
2295                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2296
2297                         /* one reference for submit_bh */
2298                         get_bh(bh);
2299
2300                         set_buffer_uptodate(bh);
2301                         lock_buffer(bh);
2302                         bh->b_end_io = btrfs_end_buffer_write_sync;
2303                 }
2304
2305                 if (i == last_barrier && do_barriers)
2306                         ret = submit_bh(WRITE_FLUSH_FUA, bh);
2307                 else
2308                         ret = submit_bh(WRITE_SYNC, bh);
2309
2310                 if (ret)
2311                         errors++;
2312         }
2313         return errors < i ? 0 : -1;
2314 }
2315
2316 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2317 {
2318         struct list_head *head;
2319         struct btrfs_device *dev;
2320         struct btrfs_super_block *sb;
2321         struct btrfs_dev_item *dev_item;
2322         int ret;
2323         int do_barriers;
2324         int max_errors;
2325         int total_errors = 0;
2326         u64 flags;
2327
2328         max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2329         do_barriers = !btrfs_test_opt(root, NOBARRIER);
2330
2331         sb = &root->fs_info->super_for_commit;
2332         dev_item = &sb->dev_item;
2333
2334         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2335         head = &root->fs_info->fs_devices->devices;
2336         list_for_each_entry(dev, head, dev_list) {
2337                 if (!dev->bdev) {
2338                         total_errors++;
2339                         continue;
2340                 }
2341                 if (!dev->in_fs_metadata || !dev->writeable)
2342                         continue;
2343
2344                 btrfs_set_stack_device_generation(dev_item, 0);
2345                 btrfs_set_stack_device_type(dev_item, dev->type);
2346                 btrfs_set_stack_device_id(dev_item, dev->devid);
2347                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2348                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2349                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2350                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2351                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2352                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2353                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2354
2355                 flags = btrfs_super_flags(sb);
2356                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2357
2358                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2359                 if (ret)
2360                         total_errors++;
2361         }
2362         if (total_errors > max_errors) {
2363                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2364                        total_errors);
2365                 BUG();
2366         }
2367
2368         total_errors = 0;
2369         list_for_each_entry(dev, head, dev_list) {
2370                 if (!dev->bdev)
2371                         continue;
2372                 if (!dev->in_fs_metadata || !dev->writeable)
2373                         continue;
2374
2375                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2376                 if (ret)
2377                         total_errors++;
2378         }
2379         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2380         if (total_errors > max_errors) {
2381                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2382                        total_errors);
2383                 BUG();
2384         }
2385         return 0;
2386 }
2387
2388 int write_ctree_super(struct btrfs_trans_handle *trans,
2389                       struct btrfs_root *root, int max_mirrors)
2390 {
2391         int ret;
2392
2393         ret = write_all_supers(root, max_mirrors);
2394         return ret;
2395 }
2396
2397 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2398 {
2399         spin_lock(&fs_info->fs_roots_radix_lock);
2400         radix_tree_delete(&fs_info->fs_roots_radix,
2401                           (unsigned long)root->root_key.objectid);
2402         spin_unlock(&fs_info->fs_roots_radix_lock);
2403
2404         if (btrfs_root_refs(&root->root_item) == 0)
2405                 synchronize_srcu(&fs_info->subvol_srcu);
2406
2407         free_fs_root(root);
2408         return 0;
2409 }
2410
2411 static void free_fs_root(struct btrfs_root *root)
2412 {
2413         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2414         if (root->anon_super.s_dev) {
2415                 down_write(&root->anon_super.s_umount);
2416                 kill_anon_super(&root->anon_super);
2417         }
2418         free_extent_buffer(root->node);
2419         free_extent_buffer(root->commit_root);
2420         kfree(root->name);
2421         kfree(root);
2422 }
2423
2424 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2425 {
2426         int ret;
2427         struct btrfs_root *gang[8];
2428         int i;
2429
2430         while (!list_empty(&fs_info->dead_roots)) {
2431                 gang[0] = list_entry(fs_info->dead_roots.next,
2432                                      struct btrfs_root, root_list);
2433                 list_del(&gang[0]->root_list);
2434
2435                 if (gang[0]->in_radix) {
2436                         btrfs_free_fs_root(fs_info, gang[0]);
2437                 } else {
2438                         free_extent_buffer(gang[0]->node);
2439                         free_extent_buffer(gang[0]->commit_root);
2440                         kfree(gang[0]);
2441                 }
2442         }
2443
2444         while (1) {
2445                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2446                                              (void **)gang, 0,
2447                                              ARRAY_SIZE(gang));
2448                 if (!ret)
2449                         break;
2450                 for (i = 0; i < ret; i++)
2451                         btrfs_free_fs_root(fs_info, gang[i]);
2452         }
2453         return 0;
2454 }
2455
2456 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2457 {
2458         u64 root_objectid = 0;
2459         struct btrfs_root *gang[8];
2460         int i;
2461         int ret;
2462
2463         while (1) {
2464                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2465                                              (void **)gang, root_objectid,
2466                                              ARRAY_SIZE(gang));
2467                 if (!ret)
2468                         break;
2469
2470                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2471                 for (i = 0; i < ret; i++) {
2472                         int err;
2473
2474                         root_objectid = gang[i]->root_key.objectid;
2475                         err = btrfs_orphan_cleanup(gang[i]);
2476                         if (err)
2477                                 return err;
2478                 }
2479                 root_objectid++;
2480         }
2481         return 0;
2482 }
2483
2484 int btrfs_commit_super(struct btrfs_root *root)
2485 {
2486         struct btrfs_trans_handle *trans;
2487         int ret;
2488
2489         mutex_lock(&root->fs_info->cleaner_mutex);
2490         btrfs_run_delayed_iputs(root);
2491         btrfs_clean_old_snapshots(root);
2492         mutex_unlock(&root->fs_info->cleaner_mutex);
2493
2494         /* wait until ongoing cleanup work done */
2495         down_write(&root->fs_info->cleanup_work_sem);
2496         up_write(&root->fs_info->cleanup_work_sem);
2497
2498         trans = btrfs_join_transaction(root, 1);
2499         if (IS_ERR(trans))
2500                 return PTR_ERR(trans);
2501         ret = btrfs_commit_transaction(trans, root);
2502         BUG_ON(ret);
2503         /* run commit again to drop the original snapshot */
2504         trans = btrfs_join_transaction(root, 1);
2505         if (IS_ERR(trans))
2506                 return PTR_ERR(trans);
2507         btrfs_commit_transaction(trans, root);
2508         ret = btrfs_write_and_wait_transaction(NULL, root);
2509         BUG_ON(ret);
2510
2511         ret = write_ctree_super(NULL, root, 0);
2512         return ret;
2513 }
2514
2515 int close_ctree(struct btrfs_root *root)
2516 {
2517         struct btrfs_fs_info *fs_info = root->fs_info;
2518         int ret;
2519
2520         fs_info->closing = 1;
2521         smp_mb();
2522
2523         btrfs_put_block_group_cache(fs_info);
2524
2525         /*
2526          * Here come 2 situations when btrfs is broken to flip readonly:
2527          *
2528          * 1. when btrfs flips readonly somewhere else before
2529          * btrfs_commit_super, sb->s_flags has MS_RDONLY flag,
2530          * and btrfs will skip to write sb directly to keep
2531          * ERROR state on disk.
2532          *
2533          * 2. when btrfs flips readonly just in btrfs_commit_super,
2534          * and in such case, btrfs cannot write sb via btrfs_commit_super,
2535          * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag,
2536          * btrfs will cleanup all FS resources first and write sb then.
2537          */
2538         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2539                 ret = btrfs_commit_super(root);
2540                 if (ret)
2541                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2542         }
2543
2544         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
2545                 ret = btrfs_error_commit_super(root);
2546                 if (ret)
2547                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2548         }
2549
2550         kthread_stop(root->fs_info->transaction_kthread);
2551         kthread_stop(root->fs_info->cleaner_kthread);
2552
2553         fs_info->closing = 2;
2554         smp_mb();
2555
2556         if (fs_info->delalloc_bytes) {
2557                 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2558                        (unsigned long long)fs_info->delalloc_bytes);
2559         }
2560         if (fs_info->total_ref_cache_size) {
2561                 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2562                        (unsigned long long)fs_info->total_ref_cache_size);
2563         }
2564
2565         free_extent_buffer(fs_info->extent_root->node);
2566         free_extent_buffer(fs_info->extent_root->commit_root);
2567         free_extent_buffer(fs_info->tree_root->node);
2568         free_extent_buffer(fs_info->tree_root->commit_root);
2569         free_extent_buffer(root->fs_info->chunk_root->node);
2570         free_extent_buffer(root->fs_info->chunk_root->commit_root);
2571         free_extent_buffer(root->fs_info->dev_root->node);
2572         free_extent_buffer(root->fs_info->dev_root->commit_root);
2573         free_extent_buffer(root->fs_info->csum_root->node);
2574         free_extent_buffer(root->fs_info->csum_root->commit_root);
2575
2576         btrfs_free_block_groups(root->fs_info);
2577
2578         del_fs_roots(fs_info);
2579
2580         iput(fs_info->btree_inode);
2581
2582         btrfs_stop_workers(&fs_info->generic_worker);
2583         btrfs_stop_workers(&fs_info->fixup_workers);
2584         btrfs_stop_workers(&fs_info->delalloc_workers);
2585         btrfs_stop_workers(&fs_info->workers);
2586         btrfs_stop_workers(&fs_info->endio_workers);
2587         btrfs_stop_workers(&fs_info->endio_meta_workers);
2588         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2589         btrfs_stop_workers(&fs_info->endio_write_workers);
2590         btrfs_stop_workers(&fs_info->endio_freespace_worker);
2591         btrfs_stop_workers(&fs_info->submit_workers);
2592
2593         btrfs_close_devices(fs_info->fs_devices);
2594         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2595
2596         bdi_destroy(&fs_info->bdi);
2597         cleanup_srcu_struct(&fs_info->subvol_srcu);
2598
2599         kfree(fs_info->extent_root);
2600         kfree(fs_info->tree_root);
2601         kfree(fs_info->chunk_root);
2602         kfree(fs_info->dev_root);
2603         kfree(fs_info->csum_root);
2604         kfree(fs_info);
2605
2606         return 0;
2607 }
2608
2609 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2610 {
2611         int ret;
2612         struct inode *btree_inode = buf->first_page->mapping->host;
2613
2614         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
2615                                      NULL);
2616         if (!ret)
2617                 return ret;
2618
2619         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2620                                     parent_transid);
2621         return !ret;
2622 }
2623
2624 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2625 {
2626         struct inode *btree_inode = buf->first_page->mapping->host;
2627         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2628                                           buf);
2629 }
2630
2631 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2632 {
2633         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2634         u64 transid = btrfs_header_generation(buf);
2635         struct inode *btree_inode = root->fs_info->btree_inode;
2636         int was_dirty;
2637
2638         btrfs_assert_tree_locked(buf);
2639         if (transid != root->fs_info->generation) {
2640                 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2641                        "found %llu running %llu\n",
2642                         (unsigned long long)buf->start,
2643                         (unsigned long long)transid,
2644                         (unsigned long long)root->fs_info->generation);
2645                 WARN_ON(1);
2646         }
2647         was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2648                                             buf);
2649         if (!was_dirty) {
2650                 spin_lock(&root->fs_info->delalloc_lock);
2651                 root->fs_info->dirty_metadata_bytes += buf->len;
2652                 spin_unlock(&root->fs_info->delalloc_lock);
2653         }
2654 }
2655
2656 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2657 {
2658         /*
2659          * looks as though older kernels can get into trouble with
2660          * this code, they end up stuck in balance_dirty_pages forever
2661          */
2662         u64 num_dirty;
2663         unsigned long thresh = 32 * 1024 * 1024;
2664
2665         if (current->flags & PF_MEMALLOC)
2666                 return;
2667
2668         num_dirty = root->fs_info->dirty_metadata_bytes;
2669
2670         if (num_dirty > thresh) {
2671                 balance_dirty_pages_ratelimited_nr(
2672                                    root->fs_info->btree_inode->i_mapping, 1);
2673         }
2674         return;
2675 }
2676
2677 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2678 {
2679         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2680         int ret;
2681         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2682         if (ret == 0)
2683                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
2684         return ret;
2685 }
2686
2687 int btree_lock_page_hook(struct page *page)
2688 {
2689         struct inode *inode = page->mapping->host;
2690         struct btrfs_root *root = BTRFS_I(inode)->root;
2691         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2692         struct extent_buffer *eb;
2693         unsigned long len;
2694         u64 bytenr = page_offset(page);
2695
2696         if (page->private == EXTENT_PAGE_PRIVATE)
2697                 goto out;
2698
2699         len = page->private >> 2;
2700         eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2701         if (!eb)
2702                 goto out;
2703
2704         btrfs_tree_lock(eb);
2705         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2706
2707         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2708                 spin_lock(&root->fs_info->delalloc_lock);
2709                 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2710                         root->fs_info->dirty_metadata_bytes -= eb->len;
2711                 else
2712                         WARN_ON(1);
2713                 spin_unlock(&root->fs_info->delalloc_lock);
2714         }
2715
2716         btrfs_tree_unlock(eb);
2717         free_extent_buffer(eb);
2718 out:
2719         lock_page(page);
2720         return 0;
2721 }
2722
2723 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
2724                               int read_only)
2725 {
2726         if (read_only)
2727                 return;
2728
2729         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
2730                 printk(KERN_WARNING "warning: mount fs with errors, "
2731                        "running btrfsck is recommended\n");
2732 }
2733
2734 int btrfs_error_commit_super(struct btrfs_root *root)
2735 {
2736         int ret;
2737
2738         mutex_lock(&root->fs_info->cleaner_mutex);
2739         btrfs_run_delayed_iputs(root);
2740         mutex_unlock(&root->fs_info->cleaner_mutex);
2741
2742         down_write(&root->fs_info->cleanup_work_sem);
2743         up_write(&root->fs_info->cleanup_work_sem);
2744
2745         /* cleanup FS via transaction */
2746         btrfs_cleanup_transaction(root);
2747
2748         ret = write_ctree_super(NULL, root, 0);
2749
2750         return ret;
2751 }
2752
2753 static int btrfs_destroy_ordered_operations(struct btrfs_root *root)
2754 {
2755         struct btrfs_inode *btrfs_inode;
2756         struct list_head splice;
2757
2758         INIT_LIST_HEAD(&splice);
2759
2760         mutex_lock(&root->fs_info->ordered_operations_mutex);
2761         spin_lock(&root->fs_info->ordered_extent_lock);
2762
2763         list_splice_init(&root->fs_info->ordered_operations, &splice);
2764         while (!list_empty(&splice)) {
2765                 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2766                                          ordered_operations);
2767
2768                 list_del_init(&btrfs_inode->ordered_operations);
2769
2770                 btrfs_invalidate_inodes(btrfs_inode->root);
2771         }
2772
2773         spin_unlock(&root->fs_info->ordered_extent_lock);
2774         mutex_unlock(&root->fs_info->ordered_operations_mutex);
2775
2776         return 0;
2777 }
2778
2779 static int btrfs_destroy_ordered_extents(struct btrfs_root *root)
2780 {
2781         struct list_head splice;
2782         struct btrfs_ordered_extent *ordered;
2783         struct inode *inode;
2784
2785         INIT_LIST_HEAD(&splice);
2786
2787         spin_lock(&root->fs_info->ordered_extent_lock);
2788
2789         list_splice_init(&root->fs_info->ordered_extents, &splice);
2790         while (!list_empty(&splice)) {
2791                 ordered = list_entry(splice.next, struct btrfs_ordered_extent,
2792                                      root_extent_list);
2793
2794                 list_del_init(&ordered->root_extent_list);
2795                 atomic_inc(&ordered->refs);
2796
2797                 /* the inode may be getting freed (in sys_unlink path). */
2798                 inode = igrab(ordered->inode);
2799
2800                 spin_unlock(&root->fs_info->ordered_extent_lock);
2801                 if (inode)
2802                         iput(inode);
2803
2804                 atomic_set(&ordered->refs, 1);
2805                 btrfs_put_ordered_extent(ordered);
2806
2807                 spin_lock(&root->fs_info->ordered_extent_lock);
2808         }
2809
2810         spin_unlock(&root->fs_info->ordered_extent_lock);
2811
2812         return 0;
2813 }
2814
2815 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
2816                                       struct btrfs_root *root)
2817 {
2818         struct rb_node *node;
2819         struct btrfs_delayed_ref_root *delayed_refs;
2820         struct btrfs_delayed_ref_node *ref;
2821         int ret = 0;
2822
2823         delayed_refs = &trans->delayed_refs;
2824
2825         spin_lock(&delayed_refs->lock);
2826         if (delayed_refs->num_entries == 0) {
2827                 spin_unlock(&delayed_refs->lock);
2828                 printk(KERN_INFO "delayed_refs has NO entry\n");
2829                 return ret;
2830         }
2831
2832         node = rb_first(&delayed_refs->root);
2833         while (node) {
2834                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2835                 node = rb_next(node);
2836
2837                 ref->in_tree = 0;
2838                 rb_erase(&ref->rb_node, &delayed_refs->root);
2839                 delayed_refs->num_entries--;
2840
2841                 atomic_set(&ref->refs, 1);
2842                 if (btrfs_delayed_ref_is_head(ref)) {
2843                         struct btrfs_delayed_ref_head *head;
2844
2845                         head = btrfs_delayed_node_to_head(ref);
2846                         mutex_lock(&head->mutex);
2847                         kfree(head->extent_op);
2848                         delayed_refs->num_heads--;
2849                         if (list_empty(&head->cluster))
2850                                 delayed_refs->num_heads_ready--;
2851                         list_del_init(&head->cluster);
2852                         mutex_unlock(&head->mutex);
2853                 }
2854
2855                 spin_unlock(&delayed_refs->lock);
2856                 btrfs_put_delayed_ref(ref);
2857
2858                 cond_resched();
2859                 spin_lock(&delayed_refs->lock);
2860         }
2861
2862         spin_unlock(&delayed_refs->lock);
2863
2864         return ret;
2865 }
2866
2867 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
2868 {
2869         struct btrfs_pending_snapshot *snapshot;
2870         struct list_head splice;
2871
2872         INIT_LIST_HEAD(&splice);
2873
2874         list_splice_init(&t->pending_snapshots, &splice);
2875
2876         while (!list_empty(&splice)) {
2877                 snapshot = list_entry(splice.next,
2878                                       struct btrfs_pending_snapshot,
2879                                       list);
2880
2881                 list_del_init(&snapshot->list);
2882
2883                 kfree(snapshot);
2884         }
2885
2886         return 0;
2887 }
2888
2889 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
2890 {
2891         struct btrfs_inode *btrfs_inode;
2892         struct list_head splice;
2893
2894         INIT_LIST_HEAD(&splice);
2895
2896         list_splice_init(&root->fs_info->delalloc_inodes, &splice);
2897
2898         spin_lock(&root->fs_info->delalloc_lock);
2899
2900         while (!list_empty(&splice)) {
2901                 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2902                                     delalloc_inodes);
2903
2904                 list_del_init(&btrfs_inode->delalloc_inodes);
2905
2906                 btrfs_invalidate_inodes(btrfs_inode->root);
2907         }
2908
2909         spin_unlock(&root->fs_info->delalloc_lock);
2910
2911         return 0;
2912 }
2913
2914 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
2915                                         struct extent_io_tree *dirty_pages,
2916                                         int mark)
2917 {
2918         int ret;
2919         struct page *page;
2920         struct inode *btree_inode = root->fs_info->btree_inode;
2921         struct extent_buffer *eb;
2922         u64 start = 0;
2923         u64 end;
2924         u64 offset;
2925         unsigned long index;
2926
2927         while (1) {
2928                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
2929                                             mark);
2930                 if (ret)
2931                         break;
2932
2933                 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
2934                 while (start <= end) {
2935                         index = start >> PAGE_CACHE_SHIFT;
2936                         start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
2937                         page = find_get_page(btree_inode->i_mapping, index);
2938                         if (!page)
2939                                 continue;
2940                         offset = page_offset(page);
2941
2942                         spin_lock(&dirty_pages->buffer_lock);
2943                         eb = radix_tree_lookup(
2944                              &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
2945                                                offset >> PAGE_CACHE_SHIFT);
2946                         spin_unlock(&dirty_pages->buffer_lock);
2947                         if (eb) {
2948                                 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
2949                                                          &eb->bflags);
2950                                 atomic_set(&eb->refs, 1);
2951                         }
2952                         if (PageWriteback(page))
2953                                 end_page_writeback(page);
2954
2955                         lock_page(page);
2956                         if (PageDirty(page)) {
2957                                 clear_page_dirty_for_io(page);
2958                                 spin_lock_irq(&page->mapping->tree_lock);
2959                                 radix_tree_tag_clear(&page->mapping->page_tree,
2960                                                         page_index(page),
2961                                                         PAGECACHE_TAG_DIRTY);
2962                                 spin_unlock_irq(&page->mapping->tree_lock);
2963                         }
2964
2965                         page->mapping->a_ops->invalidatepage(page, 0);
2966                         unlock_page(page);
2967                 }
2968         }
2969
2970         return ret;
2971 }
2972
2973 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
2974                                        struct extent_io_tree *pinned_extents)
2975 {
2976         struct extent_io_tree *unpin;
2977         u64 start;
2978         u64 end;
2979         int ret;
2980
2981         unpin = pinned_extents;
2982         while (1) {
2983                 ret = find_first_extent_bit(unpin, 0, &start, &end,
2984                                             EXTENT_DIRTY);
2985                 if (ret)
2986                         break;
2987
2988                 /* opt_discard */
2989                 if (btrfs_test_opt(root, DISCARD))
2990                         ret = btrfs_error_discard_extent(root, start,
2991                                                          end + 1 - start,
2992                                                          NULL);
2993
2994                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
2995                 btrfs_error_unpin_extent_range(root, start, end);
2996                 cond_resched();
2997         }
2998
2999         return 0;
3000 }
3001
3002 static int btrfs_cleanup_transaction(struct btrfs_root *root)
3003 {
3004         struct btrfs_transaction *t;
3005         LIST_HEAD(list);
3006
3007         WARN_ON(1);
3008
3009         mutex_lock(&root->fs_info->trans_mutex);
3010         mutex_lock(&root->fs_info->transaction_kthread_mutex);
3011
3012         list_splice_init(&root->fs_info->trans_list, &list);
3013         while (!list_empty(&list)) {
3014                 t = list_entry(list.next, struct btrfs_transaction, list);
3015                 if (!t)
3016                         break;
3017
3018                 btrfs_destroy_ordered_operations(root);
3019
3020                 btrfs_destroy_ordered_extents(root);
3021
3022                 btrfs_destroy_delayed_refs(t, root);
3023
3024                 btrfs_block_rsv_release(root,
3025                                         &root->fs_info->trans_block_rsv,
3026                                         t->dirty_pages.dirty_bytes);
3027
3028                 /* FIXME: cleanup wait for commit */
3029                 t->in_commit = 1;
3030                 t->blocked = 1;
3031                 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3032                         wake_up(&root->fs_info->transaction_blocked_wait);
3033
3034                 t->blocked = 0;
3035                 if (waitqueue_active(&root->fs_info->transaction_wait))
3036                         wake_up(&root->fs_info->transaction_wait);
3037                 mutex_unlock(&root->fs_info->trans_mutex);
3038
3039                 mutex_lock(&root->fs_info->trans_mutex);
3040                 t->commit_done = 1;
3041                 if (waitqueue_active(&t->commit_wait))
3042                         wake_up(&t->commit_wait);
3043                 mutex_unlock(&root->fs_info->trans_mutex);
3044
3045                 mutex_lock(&root->fs_info->trans_mutex);
3046
3047                 btrfs_destroy_pending_snapshots(t);
3048
3049                 btrfs_destroy_delalloc_inodes(root);
3050
3051                 spin_lock(&root->fs_info->new_trans_lock);
3052                 root->fs_info->running_transaction = NULL;
3053                 spin_unlock(&root->fs_info->new_trans_lock);
3054
3055                 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3056                                              EXTENT_DIRTY);
3057
3058                 btrfs_destroy_pinned_extent(root,
3059                                             root->fs_info->pinned_extents);
3060
3061                 atomic_set(&t->use_count, 0);
3062                 list_del_init(&t->list);
3063                 memset(t, 0, sizeof(*t));
3064                 kmem_cache_free(btrfs_transaction_cachep, t);
3065         }
3066
3067         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3068         mutex_unlock(&root->fs_info->trans_mutex);
3069
3070         return 0;
3071 }
3072
3073 static struct extent_io_ops btree_extent_io_ops = {
3074         .write_cache_pages_lock_hook = btree_lock_page_hook,
3075         .readpage_end_io_hook = btree_readpage_end_io_hook,
3076         .submit_bio_hook = btree_submit_bio_hook,
3077         /* note we're sharing with inode.c for the merge bio hook */
3078         .merge_bio_hook = btrfs_merge_bio_hook,
3079 };