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> // for block_sync_page
  26 #include <linux/workqueue.h>
  27 #include "crc32c.h"
  28 #include "ctree.h"
  29 #include "disk-io.h"
  30 #include "transaction.h"
  31 #include "btrfs_inode.h"
  32 #include "volumes.h"
  33 #include "print-tree.h"
  34
  35 #if 0
  36 static int check_tree_block(struct btrfs_root *root, struct extent_buffer *buf)
  37 {
  38         if (extent_buffer_blocknr(buf) != btrfs_header_blocknr(buf)) {
  39                 printk(KERN_CRIT "buf blocknr(buf) is %llu, header is %llu\n",
  40                        (unsigned long long)extent_buffer_blocknr(buf),
  41                        (unsigned long long)btrfs_header_blocknr(buf));
  42                 return 1;
  43         }
  44         return 0;
  45 }
  46 #endif
  47
  48 static struct extent_io_ops btree_extent_io_ops;
  49 static struct workqueue_struct *end_io_workqueue;
  50 static struct workqueue_struct *async_submit_workqueue;
  51
  52 struct end_io_wq {
  53         struct bio *bio;
  54         bio_end_io_t *end_io;
  55         void *private;
  56         struct btrfs_fs_info *info;
  57         int error;
  58         int metadata;
  59         struct list_head list;
  60 };
  61
  62 struct async_submit_bio {
  63         struct inode *inode;
  64         struct bio *bio;
  65         struct list_head list;
  66         extent_submit_bio_hook_t *submit_bio_hook;
  67         int rw;
  68         int mirror_num;
  69 };
  70
  71 struct extent_map *btree_get_extent(struct inode *inode, struct page *page,
  72                                     size_t page_offset, u64 start, u64 len,
  73                                     int create)
  74 {
  75         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
  76         struct extent_map *em;
  77         int ret;
  78
  79         spin_lock(&em_tree->lock);
  80         em = lookup_extent_mapping(em_tree, start, len);
  81         spin_unlock(&em_tree->lock);
  82         if (em)
  83                 goto out;
  84
  85         em = alloc_extent_map(GFP_NOFS);
  86         if (!em) {
  87                 em = ERR_PTR(-ENOMEM);
  88                 goto out;
  89         }
  90         em->start = 0;
  91         em->len = i_size_read(inode);
  92         em->block_start = 0;
  93         em->bdev = inode->i_sb->s_bdev;
  94
  95         spin_lock(&em_tree->lock);
  96         ret = add_extent_mapping(em_tree, em);
  97         if (ret == -EEXIST) {
  98                 free_extent_map(em);
  99                 em = lookup_extent_mapping(em_tree, start, len);
 100                 if (em)
 101                         ret = 0;
 102                 else
 103                         ret = -EIO;
 104         } else if (ret) {
 105                 free_extent_map(em);
 106                 em = NULL;
 107         }
 108         spin_unlock(&em_tree->lock);
 109
 110         if (ret)
 111                 em = ERR_PTR(ret);
 112 out:
 113         return em;
 114 }
 115
 116 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
 117 {
 118         return btrfs_crc32c(seed, data, len);
 119 }
 120
 121 void btrfs_csum_final(u32 crc, char *result)
 122 {
 123         *(__le32 *)result = ~cpu_to_le32(crc);
 124 }
 125
 126 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 127                            int verify)
 128 {
 129         char result[BTRFS_CRC32_SIZE];
 130         unsigned long len;
 131         unsigned long cur_len;
 132         unsigned long offset = BTRFS_CSUM_SIZE;
 133         char *map_token = NULL;
 134         char *kaddr;
 135         unsigned long map_start;
 136         unsigned long map_len;
 137         int err;
 138         u32 crc = ~(u32)0;
 139
 140         len = buf->len - offset;
 141         while(len > 0) {
 142                 err = map_private_extent_buffer(buf, offset, 32,
 143                                         &map_token, &kaddr,
 144                                         &map_start, &map_len, KM_USER0);
 145                 if (err) {
 146                         printk("failed to map extent buffer! %lu\n",
 147                                offset);
 148                         return 1;
 149                 }
 150                 cur_len = min(len, map_len - (offset - map_start));
 151                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
 152                                       crc, cur_len);
 153                 len -= cur_len;
 154                 offset += cur_len;
 155                 unmap_extent_buffer(buf, map_token, KM_USER0);
 156         }
 157         btrfs_csum_final(crc, result);
 158
 159         if (verify) {
 160                 int from_this_trans = 0;
 161
 162                 if (root->fs_info->running_transaction &&
 163                     btrfs_header_generation(buf) ==
 164                     root->fs_info->running_transaction->transid)
 165                         from_this_trans = 1;
 166
 167                 /* FIXME, this is not good */
 168                 if (memcmp_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE)) {
 169                         u32 val;
 170                         u32 found = 0;
 171                         memcpy(&found, result, BTRFS_CRC32_SIZE);
 172
 173                         read_extent_buffer(buf, &val, 0, BTRFS_CRC32_SIZE);
 174                         printk("btrfs: %s checksum verify failed on %llu "
 175                                "wanted %X found %X from_this_trans %d "
 176                                "level %d\n",
 177                                root->fs_info->sb->s_id,
 178                                buf->start, val, found, from_this_trans,
 179                                btrfs_header_level(buf));
 180                         return 1;
 181                 }
 182         } else {
 183                 write_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE);
 184         }
 185         return 0;
 186 }
 187
 188 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 189                                           struct extent_buffer *eb,
 190                                           u64 start)
 191 {
 192         struct extent_io_tree *io_tree;
 193         int ret;
 194         int num_copies = 0;
 195         int mirror_num = 0;
 196
 197         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 198         while (1) {
 199                 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
 200                                                btree_get_extent, mirror_num);
 201                 if (!ret) {
 202                         if (mirror_num)
 203 printk("good read %Lu mirror %d total %d\n", eb->start, mirror_num, num_copies);
 204                         return ret;
 205                 }
 206                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
 207                                               eb->start, eb->len);
 208 printk("failed to read %Lu mirror %d total %d\n", eb->start, mirror_num, num_copies);
 209                 if (num_copies == 1) {
 210 printk("reading %Lu failed only one copy\n", eb->start);
 211                         return ret;
 212                 }
 213                 mirror_num++;
 214                 if (mirror_num > num_copies) {
 215 printk("bailing at mirror %d of %d\n", mirror_num, num_copies);
 216                         return ret;
 217                 }
 218         }
 219 printk("read extent buffer page last\n");
 220         return -EIO;
 221 }
 222
 223 int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 224 {
 225         struct extent_io_tree *tree;
 226         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
 227         u64 found_start;
 228         int found_level;
 229         unsigned long len;
 230         struct extent_buffer *eb;
 231         int ret;
 232
 233         tree = &BTRFS_I(page->mapping->host)->io_tree;
 234
 235         if (page->private == EXTENT_PAGE_PRIVATE)
 236                 goto out;
 237         if (!page->private)
 238                 goto out;
 239         len = page->private >> 2;
 240         if (len == 0) {
 241                 WARN_ON(1);
 242         }
 243         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 244         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE);
 245         BUG_ON(ret);
 246         btrfs_clear_buffer_defrag(eb);
 247         found_start = btrfs_header_bytenr(eb);
 248         if (found_start != start) {
 249                 printk("warning: eb start incorrect %Lu buffer %Lu len %lu\n",
 250                        start, found_start, len);
 251                 WARN_ON(1);
 252                 goto err;
 253         }
 254         if (eb->first_page != page) {
 255                 printk("bad first page %lu %lu\n", eb->first_page->index,
 256                        page->index);
 257                 WARN_ON(1);
 258                 goto err;
 259         }
 260         if (!PageUptodate(page)) {
 261                 printk("csum not up to date page %lu\n", page->index);
 262                 WARN_ON(1);
 263                 goto err;
 264         }
 265         found_level = btrfs_header_level(eb);
 266         spin_lock(&root->fs_info->hash_lock);
 267         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
 268         spin_unlock(&root->fs_info->hash_lock);
 269         csum_tree_block(root, eb, 0);
 270 err:
 271         free_extent_buffer(eb);
 272 out:
 273         return 0;
 274 }
 275
 276 static int btree_writepage_io_hook(struct page *page, u64 start, u64 end)
 277 {
 278         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 279
 280         csum_dirty_buffer(root, page);
 281         return 0;
 282 }
 283
 284 int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 285                                struct extent_state *state)
 286 {
 287         struct extent_io_tree *tree;
 288         u64 found_start;
 289         int found_level;
 290         unsigned long len;
 291         struct extent_buffer *eb;
 292         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 293         int ret = 0;
 294
 295         tree = &BTRFS_I(page->mapping->host)->io_tree;
 296         if (page->private == EXTENT_PAGE_PRIVATE)
 297                 goto out;
 298         if (!page->private)
 299                 goto out;
 300         len = page->private >> 2;
 301         if (len == 0) {
 302                 WARN_ON(1);
 303         }
 304         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 305
 306         btrfs_clear_buffer_defrag(eb);
 307         found_start = btrfs_header_bytenr(eb);
 308         if (found_start != start) {
 309 printk("bad start on %Lu found %Lu\n", eb->start, found_start);
 310                 ret = -EIO;
 311                 goto err;
 312         }
 313         if (eb->first_page != page) {
 314                 printk("bad first page %lu %lu\n", eb->first_page->index,
 315                        page->index);
 316                 WARN_ON(1);
 317                 ret = -EIO;
 318                 goto err;
 319         }
 320         found_level = btrfs_header_level(eb);
 321
 322         ret = csum_tree_block(root, eb, 1);
 323         if (ret)
 324                 ret = -EIO;
 325
 326         end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
 327         end = eb->start + end - 1;
 328         release_extent_buffer_tail_pages(eb);
 329 err:
 330         free_extent_buffer(eb);
 331 out:
 332         return ret;
 333 }
 334
 335 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
 336 static void end_workqueue_bio(struct bio *bio, int err)
 337 #else
 338 static int end_workqueue_bio(struct bio *bio,
 339                                    unsigned int bytes_done, int err)
 340 #endif
 341 {
 342         struct end_io_wq *end_io_wq = bio->bi_private;
 343         struct btrfs_fs_info *fs_info;
 344         unsigned long flags;
 345
 346 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
 347         if (bio->bi_size)
 348                 return 1;
 349 #endif
 350
 351         fs_info = end_io_wq->info;
 352         spin_lock_irqsave(&fs_info->end_io_work_lock, flags);
 353         end_io_wq->error = err;
 354         list_add_tail(&end_io_wq->list, &fs_info->end_io_work_list);
 355         spin_unlock_irqrestore(&fs_info->end_io_work_lock, flags);
 356         queue_work(end_io_workqueue, &fs_info->end_io_work);
 357
 358 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
 359         return 0;
 360 #endif
 361 }
 362
 363 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 364                         int metadata)
 365 {
 366         struct end_io_wq *end_io_wq;
 367         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
 368         if (!end_io_wq)
 369                 return -ENOMEM;
 370
 371         end_io_wq->private = bio->bi_private;
 372         end_io_wq->end_io = bio->bi_end_io;
 373         end_io_wq->info = info;
 374         end_io_wq->error = 0;
 375         end_io_wq->bio = bio;
 376         end_io_wq->metadata = metadata;
 377
 378         bio->bi_private = end_io_wq;
 379         bio->bi_end_io = end_workqueue_bio;
 380         return 0;
 381 }
 382
 383 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 384                         int rw, struct bio *bio, int mirror_num,
 385                         extent_submit_bio_hook_t *submit_bio_hook)
 386 {
 387         struct async_submit_bio *async;
 388
 389         /*
 390          * inline writerback should stay inline, only hop to the async
 391          * queue if we're pdflush
 392          */
 393         if (!current_is_pdflush())
 394                 return submit_bio_hook(inode, rw, bio, mirror_num);
 395
 396         async = kmalloc(sizeof(*async), GFP_NOFS);
 397         if (!async)
 398                 return -ENOMEM;
 399
 400         async->inode = inode;
 401         async->rw = rw;
 402         async->bio = bio;
 403         async->mirror_num = mirror_num;
 404         async->submit_bio_hook = submit_bio_hook;
 405
 406         spin_lock(&fs_info->async_submit_work_lock);
 407         list_add_tail(&async->list, &fs_info->async_submit_work_list);
 408         spin_unlock(&fs_info->async_submit_work_lock);
 409
 410         queue_work(async_submit_workqueue, &fs_info->async_submit_work);
 411         return 0;
 412 }
 413
 414 static int __btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 415                                  int mirror_num)
 416 {
 417         struct btrfs_root *root = BTRFS_I(inode)->root;
 418         u64 offset;
 419         int ret;
 420
 421         offset = bio->bi_sector << 9;
 422
 423         if (rw & (1 << BIO_RW)) {
 424                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num);
 425         }
 426
 427         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 1);
 428         BUG_ON(ret);
 429
 430         if (offset == BTRFS_SUPER_INFO_OFFSET) {
 431                 bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
 432                 submit_bio(rw, bio);
 433                 return 0;
 434         }
 435         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num);
 436 }
 437
 438 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 439                                  int mirror_num)
 440 {
 441         if (!(rw & (1 << BIO_RW))) {
 442                 return __btree_submit_bio_hook(inode, rw, bio, mirror_num);
 443         }
 444         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 445                                    inode, rw, bio, mirror_num,
 446                                    __btree_submit_bio_hook);
 447 }
 448
 449 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 450 {
 451         struct extent_io_tree *tree;
 452         tree = &BTRFS_I(page->mapping->host)->io_tree;
 453         return extent_write_full_page(tree, page, btree_get_extent, wbc);
 454 }
 455
 456 static int btree_writepages(struct address_space *mapping,
 457                             struct writeback_control *wbc)
 458 {
 459         struct extent_io_tree *tree;
 460         tree = &BTRFS_I(mapping->host)->io_tree;
 461         if (wbc->sync_mode == WB_SYNC_NONE) {
 462                 u64 num_dirty;
 463                 u64 start = 0;
 464                 unsigned long thresh = 96 * 1024 * 1024;
 465
 466                 if (wbc->for_kupdate)
 467                         return 0;
 468
 469                 if (current_is_pdflush()) {
 470                         thresh = 96 * 1024 * 1024;
 471                 } else {
 472                         thresh = 8 * 1024 * 1024;
 473                 }
 474                 num_dirty = count_range_bits(tree, &start, (u64)-1,
 475                                              thresh, EXTENT_DIRTY);
 476                 if (num_dirty < thresh) {
 477                         return 0;
 478                 }
 479         }
 480         return extent_writepages(tree, mapping, btree_get_extent, wbc);
 481 }
 482
 483 int btree_readpage(struct file *file, struct page *page)
 484 {
 485         struct extent_io_tree *tree;
 486         tree = &BTRFS_I(page->mapping->host)->io_tree;
 487         return extent_read_full_page(tree, page, btree_get_extent);
 488 }
 489
 490 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 491 {
 492         struct extent_io_tree *tree;
 493         struct extent_map_tree *map;
 494         int ret;
 495
 496         if (page_count(page) > 3) {
 497                 /* once for page->private, once for the caller, once
 498                  * once for the page cache
 499                  */
 500                 return 0;
 501         }
 502         tree = &BTRFS_I(page->mapping->host)->io_tree;
 503         map = &BTRFS_I(page->mapping->host)->extent_tree;
 504         ret = try_release_extent_state(map, tree, page, gfp_flags);
 505         if (ret == 1) {
 506                 invalidate_extent_lru(tree, page_offset(page), PAGE_CACHE_SIZE);
 507                 ClearPagePrivate(page);
 508                 set_page_private(page, 0);
 509                 page_cache_release(page);
 510         }
 511         return ret;
 512 }
 513
 514 static void btree_invalidatepage(struct page *page, unsigned long offset)
 515 {
 516         struct extent_io_tree *tree;
 517         tree = &BTRFS_I(page->mapping->host)->io_tree;
 518         extent_invalidatepage(tree, page, offset);
 519         btree_releasepage(page, GFP_NOFS);
 520 }
 521
 522 #if 0
 523 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 524 {
 525         struct buffer_head *bh;
 526         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 527         struct buffer_head *head;
 528         if (!page_has_buffers(page)) {
 529                 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
 530                                         (1 << BH_Dirty)|(1 << BH_Uptodate));
 531         }
 532         head = page_buffers(page);
 533         bh = head;
 534         do {
 535                 if (buffer_dirty(bh))
 536                         csum_tree_block(root, bh, 0);
 537                 bh = bh->b_this_page;
 538         } while (bh != head);
 539         return block_write_full_page(page, btree_get_block, wbc);
 540 }
 541 #endif
 542
 543 static struct address_space_operations btree_aops = {
 544         .readpage       = btree_readpage,
 545         .writepage      = btree_writepage,
 546         .writepages     = btree_writepages,
 547         .releasepage    = btree_releasepage,
 548         .invalidatepage = btree_invalidatepage,
 549         .sync_page      = block_sync_page,
 550 };
 551
 552 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize)
 553 {
 554         struct extent_buffer *buf = NULL;
 555         struct inode *btree_inode = root->fs_info->btree_inode;
 556         int ret = 0;
 557
 558         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 559         if (!buf)
 560                 return 0;
 561         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
 562                                  buf, 0, 0, btree_get_extent, 0);
 563         free_extent_buffer(buf);
 564         return ret;
 565 }
 566
 567 static int close_all_devices(struct btrfs_fs_info *fs_info)
 568 {
 569         struct list_head *list;
 570         struct list_head *next;
 571         struct btrfs_device *device;
 572
 573         list = &fs_info->fs_devices->devices;
 574         list_for_each(next, list) {
 575                 device = list_entry(next, struct btrfs_device, dev_list);
 576                 if (device->bdev && device->bdev != fs_info->sb->s_bdev)
 577                         close_bdev_excl(device->bdev);
 578                 device->bdev = NULL;
 579         }
 580         return 0;
 581 }
 582
 583 int btrfs_verify_block_csum(struct btrfs_root *root,
 584                             struct extent_buffer *buf)
 585 {
 586         return btrfs_buffer_uptodate(buf);
 587 }
 588
 589 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
 590                                             u64 bytenr, u32 blocksize)
 591 {
 592         struct inode *btree_inode = root->fs_info->btree_inode;
 593         struct extent_buffer *eb;
 594         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 595                                 bytenr, blocksize, GFP_NOFS);
 596         return eb;
 597 }
 598
 599 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
 600                                                  u64 bytenr, u32 blocksize)
 601 {
 602         struct inode *btree_inode = root->fs_info->btree_inode;
 603         struct extent_buffer *eb;
 604
 605         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 606                                  bytenr, blocksize, NULL, GFP_NOFS);
 607         return eb;
 608 }
 609
 610
 611 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
 612                                       u32 blocksize)
 613 {
 614         struct extent_buffer *buf = NULL;
 615         struct inode *btree_inode = root->fs_info->btree_inode;
 616         struct extent_io_tree *io_tree;
 617         int ret;
 618
 619         io_tree = &BTRFS_I(btree_inode)->io_tree;
 620
 621         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 622         if (!buf)
 623                 return NULL;
 624
 625         ret = btree_read_extent_buffer_pages(root, buf, 0);
 626
 627         if (ret == 0) {
 628                 buf->flags |= EXTENT_UPTODATE;
 629         }
 630         return buf;
 631
 632 }
 633
 634 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 635                      struct extent_buffer *buf)
 636 {
 637         struct inode *btree_inode = root->fs_info->btree_inode;
 638         if (btrfs_header_generation(buf) ==
 639             root->fs_info->running_transaction->transid)
 640                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
 641                                           buf);
 642         return 0;
 643 }
 644
 645 int wait_on_tree_block_writeback(struct btrfs_root *root,
 646                                  struct extent_buffer *buf)
 647 {
 648         struct inode *btree_inode = root->fs_info->btree_inode;
 649         wait_on_extent_buffer_writeback(&BTRFS_I(btree_inode)->io_tree,
 650                                         buf);
 651         return 0;
 652 }
 653
 654 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
 655                         u32 stripesize, struct btrfs_root *root,
 656                         struct btrfs_fs_info *fs_info,
 657                         u64 objectid)
 658 {
 659         root->node = NULL;
 660         root->inode = NULL;
 661         root->commit_root = NULL;
 662         root->sectorsize = sectorsize;
 663         root->nodesize = nodesize;
 664         root->leafsize = leafsize;
 665         root->stripesize = stripesize;
 666         root->ref_cows = 0;
 667         root->track_dirty = 0;
 668
 669         root->fs_info = fs_info;
 670         root->objectid = objectid;
 671         root->last_trans = 0;
 672         root->highest_inode = 0;
 673         root->last_inode_alloc = 0;
 674         root->name = NULL;
 675         root->in_sysfs = 0;
 676
 677         INIT_LIST_HEAD(&root->dirty_list);
 678         memset(&root->root_key, 0, sizeof(root->root_key));
 679         memset(&root->root_item, 0, sizeof(root->root_item));
 680         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
 681         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
 682         init_completion(&root->kobj_unregister);
 683         root->defrag_running = 0;
 684         root->defrag_level = 0;
 685         root->root_key.objectid = objectid;
 686         return 0;
 687 }
 688
 689 static int find_and_setup_root(struct btrfs_root *tree_root,
 690                                struct btrfs_fs_info *fs_info,
 691                                u64 objectid,
 692                                struct btrfs_root *root)
 693 {
 694         int ret;
 695         u32 blocksize;
 696
 697         __setup_root(tree_root->nodesize, tree_root->leafsize,
 698                      tree_root->sectorsize, tree_root->stripesize,
 699                      root, fs_info, objectid);
 700         ret = btrfs_find_last_root(tree_root, objectid,
 701                                    &root->root_item, &root->root_key);
 702         BUG_ON(ret);
 703
 704         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 705         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 706                                      blocksize);
 707         BUG_ON(!root->node);
 708         return 0;
 709 }
 710
 711 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_fs_info *fs_info,
 712                                                struct btrfs_key *location)
 713 {
 714         struct btrfs_root *root;
 715         struct btrfs_root *tree_root = fs_info->tree_root;
 716         struct btrfs_path *path;
 717         struct extent_buffer *l;
 718         u64 highest_inode;
 719         u32 blocksize;
 720         int ret = 0;
 721
 722         root = kzalloc(sizeof(*root), GFP_NOFS);
 723         if (!root)
 724                 return ERR_PTR(-ENOMEM);
 725         if (location->offset == (u64)-1) {
 726                 ret = find_and_setup_root(tree_root, fs_info,
 727                                           location->objectid, root);
 728                 if (ret) {
 729                         kfree(root);
 730                         return ERR_PTR(ret);
 731                 }
 732                 goto insert;
 733         }
 734
 735         __setup_root(tree_root->nodesize, tree_root->leafsize,
 736                      tree_root->sectorsize, tree_root->stripesize,
 737                      root, fs_info, location->objectid);
 738
 739         path = btrfs_alloc_path();
 740         BUG_ON(!path);
 741         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
 742         if (ret != 0) {
 743                 if (ret > 0)
 744                         ret = -ENOENT;
 745                 goto out;
 746         }
 747         l = path->nodes[0];
 748         read_extent_buffer(l, &root->root_item,
 749                btrfs_item_ptr_offset(l, path->slots[0]),
 750                sizeof(root->root_item));
 751         memcpy(&root->root_key, location, sizeof(*location));
 752         ret = 0;
 753 out:
 754         btrfs_release_path(root, path);
 755         btrfs_free_path(path);
 756         if (ret) {
 757                 kfree(root);
 758                 return ERR_PTR(ret);
 759         }
 760         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 761         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 762                                      blocksize);
 763         BUG_ON(!root->node);
 764 insert:
 765         root->ref_cows = 1;
 766         ret = btrfs_find_highest_inode(root, &highest_inode);
 767         if (ret == 0) {
 768                 root->highest_inode = highest_inode;
 769                 root->last_inode_alloc = highest_inode;
 770         }
 771         return root;
 772 }
 773
 774 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
 775                                         u64 root_objectid)
 776 {
 777         struct btrfs_root *root;
 778
 779         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
 780                 return fs_info->tree_root;
 781         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
 782                 return fs_info->extent_root;
 783
 784         root = radix_tree_lookup(&fs_info->fs_roots_radix,
 785                                  (unsigned long)root_objectid);
 786         return root;
 787 }
 788
 789 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
 790                                               struct btrfs_key *location)
 791 {
 792         struct btrfs_root *root;
 793         int ret;
 794
 795         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
 796                 return fs_info->tree_root;
 797         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
 798                 return fs_info->extent_root;
 799
 800         root = radix_tree_lookup(&fs_info->fs_roots_radix,
 801                                  (unsigned long)location->objectid);
 802         if (root)
 803                 return root;
 804
 805         root = btrfs_read_fs_root_no_radix(fs_info, location);
 806         if (IS_ERR(root))
 807                 return root;
 808         ret = radix_tree_insert(&fs_info->fs_roots_radix,
 809                                 (unsigned long)root->root_key.objectid,
 810                                 root);
 811         if (ret) {
 812                 free_extent_buffer(root->node);
 813                 kfree(root);
 814                 return ERR_PTR(ret);
 815         }
 816         ret = btrfs_find_dead_roots(fs_info->tree_root,
 817                                     root->root_key.objectid, root);
 818         BUG_ON(ret);
 819
 820         return root;
 821 }
 822
 823 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
 824                                       struct btrfs_key *location,
 825                                       const char *name, int namelen)
 826 {
 827         struct btrfs_root *root;
 828         int ret;
 829
 830         root = btrfs_read_fs_root_no_name(fs_info, location);
 831         if (!root)
 832                 return NULL;
 833
 834         if (root->in_sysfs)
 835                 return root;
 836
 837         ret = btrfs_set_root_name(root, name, namelen);
 838         if (ret) {
 839                 free_extent_buffer(root->node);
 840                 kfree(root);
 841                 return ERR_PTR(ret);
 842         }
 843
 844         ret = btrfs_sysfs_add_root(root);
 845         if (ret) {
 846                 free_extent_buffer(root->node);
 847                 kfree(root->name);
 848                 kfree(root);
 849                 return ERR_PTR(ret);
 850         }
 851         root->in_sysfs = 1;
 852         return root;
 853 }
 854 #if 0
 855 static int add_hasher(struct btrfs_fs_info *info, char *type) {
 856         struct btrfs_hasher *hasher;
 857
 858         hasher = kmalloc(sizeof(*hasher), GFP_NOFS);
 859         if (!hasher)
 860                 return -ENOMEM;
 861         hasher->hash_tfm = crypto_alloc_hash(type, 0, CRYPTO_ALG_ASYNC);
 862         if (!hasher->hash_tfm) {
 863                 kfree(hasher);
 864                 return -EINVAL;
 865         }
 866         spin_lock(&info->hash_lock);
 867         list_add(&hasher->list, &info->hashers);
 868         spin_unlock(&info->hash_lock);
 869         return 0;
 870 }
 871 #endif
 872
 873 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
 874 {
 875         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
 876         int ret = 0;
 877         struct list_head *cur;
 878         struct btrfs_device *device;
 879         struct backing_dev_info *bdi;
 880
 881         list_for_each(cur, &info->fs_devices->devices) {
 882                 device = list_entry(cur, struct btrfs_device, dev_list);
 883                 bdi = blk_get_backing_dev_info(device->bdev);
 884                 if (bdi && bdi_congested(bdi, bdi_bits)) {
 885                         ret = 1;
 886                         break;
 887                 }
 888         }
 889         return ret;
 890 }
 891
 892 void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
 893 {
 894         struct list_head *cur;
 895         struct btrfs_device *device;
 896         struct btrfs_fs_info *info;
 897
 898         info = (struct btrfs_fs_info *)bdi->unplug_io_data;
 899         list_for_each(cur, &info->fs_devices->devices) {
 900                 device = list_entry(cur, struct btrfs_device, dev_list);
 901                 bdi = blk_get_backing_dev_info(device->bdev);
 902                 if (bdi->unplug_io_fn) {
 903                         bdi->unplug_io_fn(bdi, page);
 904                 }
 905         }
 906 }
 907
 908 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
 909 {
 910 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,23)
 911         bdi_init(bdi);
 912 #endif
 913         bdi->ra_pages   = default_backing_dev_info.ra_pages * 4;
 914         bdi->state              = 0;
 915         bdi->capabilities       = default_backing_dev_info.capabilities;
 916         bdi->unplug_io_fn       = btrfs_unplug_io_fn;
 917         bdi->unplug_io_data     = info;
 918         bdi->congested_fn       = btrfs_congested_fn;
 919         bdi->congested_data     = info;
 920         return 0;
 921 }
 922
 923 static int bio_ready_for_csum(struct bio *bio)
 924 {
 925         u64 length = 0;
 926         u64 buf_len = 0;
 927         u64 start = 0;
 928         struct page *page;
 929         struct extent_io_tree *io_tree = NULL;
 930         struct btrfs_fs_info *info = NULL;
 931         struct bio_vec *bvec;
 932         int i;
 933         int ret;
 934
 935         bio_for_each_segment(bvec, bio, i) {
 936                 page = bvec->bv_page;
 937                 if (page->private == EXTENT_PAGE_PRIVATE) {
 938                         length += bvec->bv_len;
 939                         continue;
 940                 }
 941                 if (!page->private) {
 942                         length += bvec->bv_len;
 943                         continue;
 944                 }
 945                 length = bvec->bv_len;
 946                 buf_len = page->private >> 2;
 947                 start = page_offset(page) + bvec->bv_offset;
 948                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
 949                 info = BTRFS_I(page->mapping->host)->root->fs_info;
 950         }
 951         /* are we fully contained in this bio? */
 952         if (buf_len <= length)
 953                 return 1;
 954
 955         ret = extent_range_uptodate(io_tree, start + length,
 956                                     start + buf_len - 1);
 957         if (ret == 1)
 958                 return ret;
 959         return ret;
 960 }
 961
 962 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
 963 static void btrfs_end_io_csum(void *p)
 964 #else
 965 static void btrfs_end_io_csum(struct work_struct *work)
 966 #endif
 967 {
 968 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
 969         struct btrfs_fs_info *fs_info = p;
 970 #else
 971         struct btrfs_fs_info *fs_info = container_of(work,
 972                                                      struct btrfs_fs_info,
 973                                                      end_io_work);
 974 #endif
 975         unsigned long flags;
 976         struct end_io_wq *end_io_wq;
 977         struct bio *bio;
 978         struct list_head *next;
 979         int error;
 980         int was_empty;
 981
 982         while(1) {
 983                 spin_lock_irqsave(&fs_info->end_io_work_lock, flags);
 984                 if (list_empty(&fs_info->end_io_work_list)) {
 985                         spin_unlock_irqrestore(&fs_info->end_io_work_lock,
 986                                                flags);
 987                         return;
 988                 }
 989                 next = fs_info->end_io_work_list.next;
 990                 list_del(next);
 991                 spin_unlock_irqrestore(&fs_info->end_io_work_lock, flags);
 992
 993                 end_io_wq = list_entry(next, struct end_io_wq, list);
 994
 995                 bio = end_io_wq->bio;
 996                 if (end_io_wq->metadata && !bio_ready_for_csum(bio)) {
 997                         spin_lock_irqsave(&fs_info->end_io_work_lock, flags);
 998                         was_empty = list_empty(&fs_info->end_io_work_list);
 999                         list_add_tail(&end_io_wq->list,
1000                                       &fs_info->end_io_work_list);
1001                         spin_unlock_irqrestore(&fs_info->end_io_work_lock,
1002                                                flags);
1003                         if (was_empty)
1004                                 return;
1005                         continue;
1006                 }
1007                 error = end_io_wq->error;
1008                 bio->bi_private = end_io_wq->private;
1009                 bio->bi_end_io = end_io_wq->end_io;
1010                 kfree(end_io_wq);
1011 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1012                 bio_endio(bio, bio->bi_size, error);
1013 #else
1014                 bio_endio(bio, error);
1015 #endif
1016         }
1017 }
1018
1019 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1020 static void btrfs_async_submit_work(void *p)
1021 #else
1022 static void btrfs_async_submit_work(struct work_struct *work)
1023 #endif
1024 {
1025 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1026         struct btrfs_fs_info *fs_info = p;
1027 #else
1028         struct btrfs_fs_info *fs_info = container_of(work,
1029                                                      struct btrfs_fs_info,
1030                                                      async_submit_work);
1031 #endif
1032         struct async_submit_bio *async;
1033         struct list_head *next;
1034
1035         while(1) {
1036                 spin_lock(&fs_info->async_submit_work_lock);
1037                 if (list_empty(&fs_info->async_submit_work_list)) {
1038                         spin_unlock(&fs_info->async_submit_work_lock);
1039                         return;
1040                 }
1041                 next = fs_info->async_submit_work_list.next;
1042                 list_del(next);
1043                 spin_unlock(&fs_info->async_submit_work_lock);
1044
1045                 async = list_entry(next, struct async_submit_bio, list);
1046                 async->submit_bio_hook(async->inode, async->rw, async->bio,
1047                                        async->mirror_num);
1048                 kfree(async);
1049         }
1050 }
1051
1052 struct btrfs_root *open_ctree(struct super_block *sb,
1053                               struct btrfs_fs_devices *fs_devices)
1054 {
1055         u32 sectorsize;
1056         u32 nodesize;
1057         u32 leafsize;
1058         u32 blocksize;
1059         u32 stripesize;
1060         struct btrfs_root *extent_root = kmalloc(sizeof(struct btrfs_root),
1061                                                  GFP_NOFS);
1062         struct btrfs_root *tree_root = kmalloc(sizeof(struct btrfs_root),
1063                                                GFP_NOFS);
1064         struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1065                                                 GFP_NOFS);
1066         struct btrfs_root *chunk_root = kmalloc(sizeof(struct btrfs_root),
1067                                                 GFP_NOFS);
1068         struct btrfs_root *dev_root = kmalloc(sizeof(struct btrfs_root),
1069                                               GFP_NOFS);
1070         int ret;
1071         int err = -EINVAL;
1072         struct btrfs_super_block *disk_super;
1073
1074         if (!extent_root || !tree_root || !fs_info) {
1075                 err = -ENOMEM;
1076                 goto fail;
1077         }
1078         end_io_workqueue = create_workqueue("btrfs-end-io");
1079         BUG_ON(!end_io_workqueue);
1080         async_submit_workqueue = create_workqueue("btrfs-async-submit");
1081
1082         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1083         INIT_LIST_HEAD(&fs_info->trans_list);
1084         INIT_LIST_HEAD(&fs_info->dead_roots);
1085         INIT_LIST_HEAD(&fs_info->hashers);
1086         INIT_LIST_HEAD(&fs_info->end_io_work_list);
1087         INIT_LIST_HEAD(&fs_info->async_submit_work_list);
1088         spin_lock_init(&fs_info->hash_lock);
1089         spin_lock_init(&fs_info->end_io_work_lock);
1090         spin_lock_init(&fs_info->async_submit_work_lock);
1091         spin_lock_init(&fs_info->delalloc_lock);
1092         spin_lock_init(&fs_info->new_trans_lock);
1093
1094         init_completion(&fs_info->kobj_unregister);
1095         sb_set_blocksize(sb, BTRFS_SUPER_INFO_SIZE);
1096         fs_info->tree_root = tree_root;
1097         fs_info->extent_root = extent_root;
1098         fs_info->chunk_root = chunk_root;
1099         fs_info->dev_root = dev_root;
1100         fs_info->fs_devices = fs_devices;
1101         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1102         INIT_LIST_HEAD(&fs_info->space_info);
1103         btrfs_mapping_init(&fs_info->mapping_tree);
1104         fs_info->sb = sb;
1105         fs_info->max_extent = (u64)-1;
1106         fs_info->max_inline = 8192 * 1024;
1107         setup_bdi(fs_info, &fs_info->bdi);
1108         fs_info->btree_inode = new_inode(sb);
1109         fs_info->btree_inode->i_ino = 1;
1110         fs_info->btree_inode->i_nlink = 1;
1111         fs_info->btree_inode->i_size = sb->s_bdev->bd_inode->i_size;
1112         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1113         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1114
1115         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1116                              fs_info->btree_inode->i_mapping,
1117                              GFP_NOFS);
1118         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1119                              GFP_NOFS);
1120
1121         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1122
1123         extent_io_tree_init(&fs_info->free_space_cache,
1124                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1125         extent_io_tree_init(&fs_info->block_group_cache,
1126                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1127         extent_io_tree_init(&fs_info->pinned_extents,
1128                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1129         extent_io_tree_init(&fs_info->pending_del,
1130                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1131         extent_io_tree_init(&fs_info->extent_ins,
1132                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1133         fs_info->do_barriers = 1;
1134
1135 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1136         INIT_WORK(&fs_info->end_io_work, btrfs_end_io_csum, fs_info);
1137         INIT_WORK(&fs_info->async_submit_work, btrfs_async_submit_work,
1138                   fs_info);
1139         INIT_WORK(&fs_info->trans_work, btrfs_transaction_cleaner, fs_info);
1140 #else
1141         INIT_WORK(&fs_info->end_io_work, btrfs_end_io_csum);
1142         INIT_WORK(&fs_info->async_submit_work, btrfs_async_submit_work);
1143         INIT_DELAYED_WORK(&fs_info->trans_work, btrfs_transaction_cleaner);
1144 #endif
1145         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1146         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1147                sizeof(struct btrfs_key));
1148         insert_inode_hash(fs_info->btree_inode);
1149         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1150
1151         mutex_init(&fs_info->trans_mutex);
1152         mutex_init(&fs_info->fs_mutex);
1153
1154 #if 0
1155         ret = add_hasher(fs_info, "crc32c");
1156         if (ret) {
1157                 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
1158                 err = -ENOMEM;
1159                 goto fail_iput;
1160         }
1161 #endif
1162         __setup_root(4096, 4096, 4096, 4096, tree_root,
1163                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1164
1165         fs_info->sb_buffer = read_tree_block(tree_root,
1166                                              BTRFS_SUPER_INFO_OFFSET,
1167                                              4096);
1168
1169         if (!fs_info->sb_buffer)
1170                 goto fail_iput;
1171
1172         read_extent_buffer(fs_info->sb_buffer, &fs_info->super_copy, 0,
1173                            sizeof(fs_info->super_copy));
1174
1175         read_extent_buffer(fs_info->sb_buffer, fs_info->fsid,
1176                            (unsigned long)btrfs_super_fsid(fs_info->sb_buffer),
1177                            BTRFS_FSID_SIZE);
1178
1179         disk_super = &fs_info->super_copy;
1180         if (!btrfs_super_root(disk_super))
1181                 goto fail_sb_buffer;
1182
1183         if (btrfs_super_num_devices(disk_super) != fs_devices->num_devices) {
1184                 printk("Btrfs: wanted %llu devices, but found %llu\n",
1185                        (unsigned long long)btrfs_super_num_devices(disk_super),
1186                        (unsigned long long)fs_devices->num_devices);
1187                 goto fail_sb_buffer;
1188         }
1189         nodesize = btrfs_super_nodesize(disk_super);
1190         leafsize = btrfs_super_leafsize(disk_super);
1191         sectorsize = btrfs_super_sectorsize(disk_super);
1192         stripesize = btrfs_super_stripesize(disk_super);
1193         tree_root->nodesize = nodesize;
1194         tree_root->leafsize = leafsize;
1195         tree_root->sectorsize = sectorsize;
1196         tree_root->stripesize = stripesize;
1197         sb_set_blocksize(sb, sectorsize);
1198
1199         i_size_write(fs_info->btree_inode,
1200                      btrfs_super_total_bytes(disk_super));
1201
1202         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1203                     sizeof(disk_super->magic))) {
1204                 printk("btrfs: valid FS not found on %s\n", sb->s_id);
1205                 goto fail_sb_buffer;
1206         }
1207
1208         mutex_lock(&fs_info->fs_mutex);
1209
1210         ret = btrfs_read_sys_array(tree_root);
1211         BUG_ON(ret);
1212
1213         blocksize = btrfs_level_size(tree_root,
1214                                      btrfs_super_chunk_root_level(disk_super));
1215
1216         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1217                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1218
1219         chunk_root->node = read_tree_block(chunk_root,
1220                                            btrfs_super_chunk_root(disk_super),
1221                                            blocksize);
1222         BUG_ON(!chunk_root->node);
1223
1224         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1225                  (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1226                  BTRFS_UUID_SIZE);
1227
1228         ret = btrfs_read_chunk_tree(chunk_root);
1229         BUG_ON(ret);
1230
1231         blocksize = btrfs_level_size(tree_root,
1232                                      btrfs_super_root_level(disk_super));
1233
1234
1235         tree_root->node = read_tree_block(tree_root,
1236                                           btrfs_super_root(disk_super),
1237                                           blocksize);
1238         if (!tree_root->node)
1239                 goto fail_sb_buffer;
1240
1241
1242         ret = find_and_setup_root(tree_root, fs_info,
1243                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1244         if (ret)
1245                 goto fail_tree_root;
1246         extent_root->track_dirty = 1;
1247
1248         ret = find_and_setup_root(tree_root, fs_info,
1249                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
1250         dev_root->track_dirty = 1;
1251
1252         if (ret)
1253                 goto fail_extent_root;
1254
1255         btrfs_read_block_groups(extent_root);
1256
1257         fs_info->generation = btrfs_super_generation(disk_super) + 1;
1258         fs_info->data_alloc_profile = (u64)-1;
1259         fs_info->metadata_alloc_profile = (u64)-1;
1260         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1261
1262         mutex_unlock(&fs_info->fs_mutex);
1263         return tree_root;
1264
1265 fail_extent_root:
1266         free_extent_buffer(extent_root->node);
1267 fail_tree_root:
1268         mutex_unlock(&fs_info->fs_mutex);
1269         free_extent_buffer(tree_root->node);
1270 fail_sb_buffer:
1271         free_extent_buffer(fs_info->sb_buffer);
1272         extent_io_tree_empty_lru(&BTRFS_I(fs_info->btree_inode)->io_tree);
1273 fail_iput:
1274         iput(fs_info->btree_inode);
1275 fail:
1276         close_all_devices(fs_info);
1277         kfree(extent_root);
1278         kfree(tree_root);
1279 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,23)
1280         bdi_destroy(&fs_info->bdi);
1281 #endif
1282         kfree(fs_info);
1283         return ERR_PTR(err);
1284 }
1285
1286 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1287 {
1288         char b[BDEVNAME_SIZE];
1289
1290         if (uptodate) {
1291                 set_buffer_uptodate(bh);
1292         } else {
1293                 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1294                         printk(KERN_WARNING "lost page write due to "
1295                                         "I/O error on %s\n",
1296                                        bdevname(bh->b_bdev, b));
1297                 }
1298                 set_buffer_write_io_error(bh);
1299                 clear_buffer_uptodate(bh);
1300         }
1301         unlock_buffer(bh);
1302         put_bh(bh);
1303 }
1304
1305 int write_all_supers(struct btrfs_root *root)
1306 {
1307         struct list_head *cur;
1308         struct list_head *head = &root->fs_info->fs_devices->devices;
1309         struct btrfs_device *dev;
1310         struct extent_buffer *sb;
1311         struct btrfs_dev_item *dev_item;
1312         struct buffer_head *bh;
1313         int ret;
1314         int do_barriers;
1315
1316         do_barriers = !btrfs_test_opt(root, NOBARRIER);
1317
1318         sb = root->fs_info->sb_buffer;
1319         dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
1320                                                       dev_item);
1321         list_for_each(cur, head) {
1322                 dev = list_entry(cur, struct btrfs_device, dev_list);
1323                 btrfs_set_device_type(sb, dev_item, dev->type);
1324                 btrfs_set_device_id(sb, dev_item, dev->devid);
1325                 btrfs_set_device_total_bytes(sb, dev_item, dev->total_bytes);
1326                 btrfs_set_device_bytes_used(sb, dev_item, dev->bytes_used);
1327                 btrfs_set_device_io_align(sb, dev_item, dev->io_align);
1328                 btrfs_set_device_io_width(sb, dev_item, dev->io_width);
1329                 btrfs_set_device_sector_size(sb, dev_item, dev->sector_size);
1330                 write_extent_buffer(sb, dev->uuid,
1331                                     (unsigned long)btrfs_device_uuid(dev_item),
1332                                     BTRFS_UUID_SIZE);
1333
1334                 btrfs_set_header_flag(sb, BTRFS_HEADER_FLAG_WRITTEN);
1335                 csum_tree_block(root, sb, 0);
1336
1337                 bh = __getblk(dev->bdev, BTRFS_SUPER_INFO_OFFSET /
1338                               root->fs_info->sb->s_blocksize,
1339                               BTRFS_SUPER_INFO_SIZE);
1340
1341                 read_extent_buffer(sb, bh->b_data, 0, BTRFS_SUPER_INFO_SIZE);
1342                 dev->pending_io = bh;
1343
1344                 get_bh(bh);
1345                 set_buffer_uptodate(bh);
1346                 lock_buffer(bh);
1347                 bh->b_end_io = btrfs_end_buffer_write_sync;
1348
1349                 if (do_barriers && dev->barriers) {
1350                         ret = submit_bh(WRITE_BARRIER, bh);
1351                         if (ret == -EOPNOTSUPP) {
1352                                 printk("btrfs: disabling barriers on dev %s\n",
1353                                        dev->name);
1354                                 set_buffer_uptodate(bh);
1355                                 dev->barriers = 0;
1356                                 get_bh(bh);
1357                                 lock_buffer(bh);
1358                                 ret = submit_bh(WRITE, bh);
1359                         }
1360                 } else {
1361                         ret = submit_bh(WRITE, bh);
1362                 }
1363                 BUG_ON(ret);
1364         }
1365
1366         list_for_each(cur, head) {
1367                 dev = list_entry(cur, struct btrfs_device, dev_list);
1368                 BUG_ON(!dev->pending_io);
1369                 bh = dev->pending_io;
1370                 wait_on_buffer(bh);
1371                 if (!buffer_uptodate(dev->pending_io)) {
1372                         if (do_barriers && dev->barriers) {
1373                                 printk("btrfs: disabling barriers on dev %s\n",
1374                                        dev->name);
1375                                 set_buffer_uptodate(bh);
1376                                 get_bh(bh);
1377                                 lock_buffer(bh);
1378                                 dev->barriers = 0;
1379                                 ret = submit_bh(WRITE, bh);
1380                                 BUG_ON(ret);
1381                                 wait_on_buffer(bh);
1382                                 BUG_ON(!buffer_uptodate(bh));
1383                         } else {
1384                                 BUG();
1385                         }
1386
1387                 }
1388                 dev->pending_io = NULL;
1389                 brelse(bh);
1390         }
1391         return 0;
1392 }
1393
1394 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
1395                       *root)
1396 {
1397         int ret;
1398
1399         ret = write_all_supers(root);
1400 #if 0
1401         if (!btrfs_test_opt(root, NOBARRIER))
1402                 blkdev_issue_flush(sb->s_bdev, NULL);
1403         set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, super);
1404         ret = sync_page_range_nolock(btree_inode, btree_inode->i_mapping,
1405                                      super->start, super->len);
1406         if (!btrfs_test_opt(root, NOBARRIER))
1407                 blkdev_issue_flush(sb->s_bdev, NULL);
1408 #endif
1409         return ret;
1410 }
1411
1412 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
1413 {
1414         radix_tree_delete(&fs_info->fs_roots_radix,
1415                           (unsigned long)root->root_key.objectid);
1416         if (root->in_sysfs)
1417                 btrfs_sysfs_del_root(root);
1418         if (root->inode)
1419                 iput(root->inode);
1420         if (root->node)
1421                 free_extent_buffer(root->node);
1422         if (root->commit_root)
1423                 free_extent_buffer(root->commit_root);
1424         if (root->name)
1425                 kfree(root->name);
1426         kfree(root);
1427         return 0;
1428 }
1429
1430 static int del_fs_roots(struct btrfs_fs_info *fs_info)
1431 {
1432         int ret;
1433         struct btrfs_root *gang[8];
1434         int i;
1435
1436         while(1) {
1437                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1438                                              (void **)gang, 0,
1439                                              ARRAY_SIZE(gang));
1440                 if (!ret)
1441                         break;
1442                 for (i = 0; i < ret; i++)
1443                         btrfs_free_fs_root(fs_info, gang[i]);
1444         }
1445         return 0;
1446 }
1447
1448 int close_ctree(struct btrfs_root *root)
1449 {
1450         int ret;
1451         struct btrfs_trans_handle *trans;
1452         struct btrfs_fs_info *fs_info = root->fs_info;
1453
1454         fs_info->closing = 1;
1455         btrfs_transaction_flush_work(root);
1456         mutex_lock(&fs_info->fs_mutex);
1457         btrfs_defrag_dirty_roots(root->fs_info);
1458         trans = btrfs_start_transaction(root, 1);
1459         ret = btrfs_commit_transaction(trans, root);
1460         /* run commit again to  drop the original snapshot */
1461         trans = btrfs_start_transaction(root, 1);
1462         btrfs_commit_transaction(trans, root);
1463         ret = btrfs_write_and_wait_transaction(NULL, root);
1464         BUG_ON(ret);
1465         write_ctree_super(NULL, root);
1466         mutex_unlock(&fs_info->fs_mutex);
1467
1468         if (fs_info->delalloc_bytes) {
1469                 printk("btrfs: at unmount delalloc count %Lu\n",
1470                        fs_info->delalloc_bytes);
1471         }
1472         if (fs_info->extent_root->node)
1473                 free_extent_buffer(fs_info->extent_root->node);
1474
1475         if (fs_info->tree_root->node)
1476                 free_extent_buffer(fs_info->tree_root->node);
1477
1478         if (root->fs_info->chunk_root->node);
1479                 free_extent_buffer(root->fs_info->chunk_root->node);
1480
1481         if (root->fs_info->dev_root->node);
1482                 free_extent_buffer(root->fs_info->dev_root->node);
1483
1484         free_extent_buffer(fs_info->sb_buffer);
1485
1486         btrfs_free_block_groups(root->fs_info);
1487         del_fs_roots(fs_info);
1488
1489         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1490
1491         extent_io_tree_empty_lru(&fs_info->free_space_cache);
1492         extent_io_tree_empty_lru(&fs_info->block_group_cache);
1493         extent_io_tree_empty_lru(&fs_info->pinned_extents);
1494         extent_io_tree_empty_lru(&fs_info->pending_del);
1495         extent_io_tree_empty_lru(&fs_info->extent_ins);
1496         extent_io_tree_empty_lru(&BTRFS_I(fs_info->btree_inode)->io_tree);
1497
1498         truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
1499         flush_workqueue(end_io_workqueue);
1500         destroy_workqueue(end_io_workqueue);
1501
1502         flush_workqueue(async_submit_workqueue);
1503         destroy_workqueue(async_submit_workqueue);
1504
1505         iput(fs_info->btree_inode);
1506 #if 0
1507         while(!list_empty(&fs_info->hashers)) {
1508                 struct btrfs_hasher *hasher;
1509                 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
1510                                     hashers);
1511                 list_del(&hasher->hashers);
1512                 crypto_free_hash(&fs_info->hash_tfm);
1513                 kfree(hasher);
1514         }
1515 #endif
1516         close_all_devices(fs_info);
1517         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1518
1519 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,23)
1520         bdi_destroy(&fs_info->bdi);
1521 #endif
1522
1523         kfree(fs_info->extent_root);
1524         kfree(fs_info->tree_root);
1525         kfree(fs_info->chunk_root);
1526         kfree(fs_info->dev_root);
1527         return 0;
1528 }
1529
1530 int btrfs_buffer_uptodate(struct extent_buffer *buf)
1531 {
1532         struct inode *btree_inode = buf->first_page->mapping->host;
1533         return extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
1534 }
1535
1536 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
1537 {
1538         struct inode *btree_inode = buf->first_page->mapping->host;
1539         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
1540                                           buf);
1541 }
1542
1543 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
1544 {
1545         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1546         u64 transid = btrfs_header_generation(buf);
1547         struct inode *btree_inode = root->fs_info->btree_inode;
1548
1549         if (transid != root->fs_info->generation) {
1550                 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
1551                         (unsigned long long)buf->start,
1552                         transid, root->fs_info->generation);
1553                 WARN_ON(1);
1554         }
1555         set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
1556 }
1557
1558 void btrfs_throttle(struct btrfs_root *root)
1559 {
1560         struct backing_dev_info *bdi;
1561
1562         bdi = root->fs_info->sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
1563         if (root->fs_info->throttles && bdi_write_congested(bdi)) {
1564 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,18)
1565                 congestion_wait(WRITE, HZ/20);
1566 #else
1567                 blk_congestion_wait(WRITE, HZ/20);
1568 #endif
1569         }
1570 }
1571
1572 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
1573 {
1574         balance_dirty_pages_ratelimited_nr(
1575                                    root->fs_info->btree_inode->i_mapping, 1);
1576 }
1577
1578 void btrfs_set_buffer_defrag(struct extent_buffer *buf)
1579 {
1580         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1581         struct inode *btree_inode = root->fs_info->btree_inode;
1582         set_extent_bits(&BTRFS_I(btree_inode)->io_tree, buf->start,
1583                         buf->start + buf->len - 1, EXTENT_DEFRAG, GFP_NOFS);
1584 }
1585
1586 void btrfs_set_buffer_defrag_done(struct extent_buffer *buf)
1587 {
1588         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1589         struct inode *btree_inode = root->fs_info->btree_inode;
1590         set_extent_bits(&BTRFS_I(btree_inode)->io_tree, buf->start,
1591                         buf->start + buf->len - 1, EXTENT_DEFRAG_DONE,
1592                         GFP_NOFS);
1593 }
1594
1595 int btrfs_buffer_defrag(struct extent_buffer *buf)
1596 {
1597         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1598         struct inode *btree_inode = root->fs_info->btree_inode;
1599         return test_range_bit(&BTRFS_I(btree_inode)->io_tree,
1600                      buf->start, buf->start + buf->len - 1, EXTENT_DEFRAG, 0);
1601 }
1602
1603 int btrfs_buffer_defrag_done(struct extent_buffer *buf)
1604 {
1605         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1606         struct inode *btree_inode = root->fs_info->btree_inode;
1607         return test_range_bit(&BTRFS_I(btree_inode)->io_tree,
1608                      buf->start, buf->start + buf->len - 1,
1609                      EXTENT_DEFRAG_DONE, 0);
1610 }
1611
1612 int btrfs_clear_buffer_defrag_done(struct extent_buffer *buf)
1613 {
1614         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1615         struct inode *btree_inode = root->fs_info->btree_inode;
1616         return clear_extent_bits(&BTRFS_I(btree_inode)->io_tree,
1617                      buf->start, buf->start + buf->len - 1,
1618                      EXTENT_DEFRAG_DONE, GFP_NOFS);
1619 }
1620
1621 int btrfs_clear_buffer_defrag(struct extent_buffer *buf)
1622 {
1623         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1624         struct inode *btree_inode = root->fs_info->btree_inode;
1625         return clear_extent_bits(&BTRFS_I(btree_inode)->io_tree,
1626                      buf->start, buf->start + buf->len - 1,
1627                      EXTENT_DEFRAG, GFP_NOFS);
1628 }
1629
1630 int btrfs_read_buffer(struct extent_buffer *buf)
1631 {
1632         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1633         int ret;
1634         ret = btree_read_extent_buffer_pages(root, buf, 0);
1635         if (ret == 0) {
1636                 buf->flags |= EXTENT_UPTODATE;
1637         }
1638         return ret;
1639 }
1640
1641 static struct extent_io_ops btree_extent_io_ops = {
1642         .writepage_io_hook = btree_writepage_io_hook,
1643         .readpage_end_io_hook = btree_readpage_end_io_hook,
1644         .submit_bio_hook = btree_submit_bio_hook,
1645         /* note we're sharing with inode.c for the merge bio hook */
1646         .merge_bio_hook = btrfs_merge_bio_hook,
1647 };