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         if (em) {
  82                 em->bdev =
  83                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
  84                 spin_unlock(&em_tree->lock);
  85                 goto out;
  86         }
  87         spin_unlock(&em_tree->lock);
  88
  89         em = alloc_extent_map(GFP_NOFS);
  90         if (!em) {
  91                 em = ERR_PTR(-ENOMEM);
  92                 goto out;
  93         }
  94         em->start = 0;
  95         em->len = (u64)-1;
  96         em->block_start = 0;
  97         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
  98
  99         spin_lock(&em_tree->lock);
 100         ret = add_extent_mapping(em_tree, em);
 101         if (ret == -EEXIST) {
 102                 u64 failed_start = em->start;
 103                 u64 failed_len = em->len;
 104
 105                 printk("failed to insert %Lu %Lu -> %Lu into tree\n",
 106                        em->start, em->len, em->block_start);
 107                 free_extent_map(em);
 108                 em = lookup_extent_mapping(em_tree, start, len);
 109                 if (em) {
 110                         printk("after failing, found %Lu %Lu %Lu\n",
 111                                em->start, em->len, em->block_start);
 112                         ret = 0;
 113                 } else {
 114                         em = lookup_extent_mapping(em_tree, failed_start,
 115                                                    failed_len);
 116                         if (em) {
 117                                 printk("double failure lookup gives us "
 118                                        "%Lu %Lu -> %Lu\n", em->start,
 119                                        em->len, em->block_start);
 120                                 free_extent_map(em);
 121                         }
 122                         ret = -EIO;
 123                 }
 124         } else if (ret) {
 125                 free_extent_map(em);
 126                 em = NULL;
 127         }
 128         spin_unlock(&em_tree->lock);
 129
 130         if (ret)
 131                 em = ERR_PTR(ret);
 132 out:
 133         return em;
 134 }
 135
 136 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
 137 {
 138         return btrfs_crc32c(seed, data, len);
 139 }
 140
 141 void btrfs_csum_final(u32 crc, char *result)
 142 {
 143         *(__le32 *)result = ~cpu_to_le32(crc);
 144 }
 145
 146 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 147                            int verify)
 148 {
 149         char result[BTRFS_CRC32_SIZE];
 150         unsigned long len;
 151         unsigned long cur_len;
 152         unsigned long offset = BTRFS_CSUM_SIZE;
 153         char *map_token = NULL;
 154         char *kaddr;
 155         unsigned long map_start;
 156         unsigned long map_len;
 157         int err;
 158         u32 crc = ~(u32)0;
 159
 160         len = buf->len - offset;
 161         while(len > 0) {
 162                 err = map_private_extent_buffer(buf, offset, 32,
 163                                         &map_token, &kaddr,
 164                                         &map_start, &map_len, KM_USER0);
 165                 if (err) {
 166                         printk("failed to map extent buffer! %lu\n",
 167                                offset);
 168                         return 1;
 169                 }
 170                 cur_len = min(len, map_len - (offset - map_start));
 171                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
 172                                       crc, cur_len);
 173                 len -= cur_len;
 174                 offset += cur_len;
 175                 unmap_extent_buffer(buf, map_token, KM_USER0);
 176         }
 177         btrfs_csum_final(crc, result);
 178
 179         if (verify) {
 180                 int from_this_trans = 0;
 181
 182                 if (root->fs_info->running_transaction &&
 183                     btrfs_header_generation(buf) ==
 184                     root->fs_info->running_transaction->transid)
 185                         from_this_trans = 1;
 186
 187                 /* FIXME, this is not good */
 188                 if (memcmp_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE)) {
 189                         u32 val;
 190                         u32 found = 0;
 191                         memcpy(&found, result, BTRFS_CRC32_SIZE);
 192
 193                         read_extent_buffer(buf, &val, 0, BTRFS_CRC32_SIZE);
 194                         printk("btrfs: %s checksum verify failed on %llu "
 195                                "wanted %X found %X from_this_trans %d "
 196                                "level %d\n",
 197                                root->fs_info->sb->s_id,
 198                                buf->start, val, found, from_this_trans,
 199                                btrfs_header_level(buf));
 200                         return 1;
 201                 }
 202         } else {
 203                 write_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE);
 204         }
 205         return 0;
 206 }
 207
 208 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 209                                           struct extent_buffer *eb,
 210                                           u64 start, u64 parent_transid)
 211 {
 212         struct extent_io_tree *io_tree;
 213         int ret;
 214         int num_copies = 0;
 215         int mirror_num = 0;
 216
 217         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 218         while (1) {
 219                 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
 220                                                btree_get_extent, mirror_num);
 221                 if (!ret)
 222                         return ret;
 223
 224                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
 225                                               eb->start, eb->len);
 226                 if (num_copies == 1)
 227                         return ret;
 228
 229                 mirror_num++;
 230                 if (mirror_num > num_copies)
 231                         return ret;
 232         }
 233         return -EIO;
 234 }
 235
 236 int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 237 {
 238         struct extent_io_tree *tree;
 239         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
 240         u64 found_start;
 241         int found_level;
 242         unsigned long len;
 243         struct extent_buffer *eb;
 244         int ret;
 245
 246         tree = &BTRFS_I(page->mapping->host)->io_tree;
 247
 248         if (page->private == EXTENT_PAGE_PRIVATE)
 249                 goto out;
 250         if (!page->private)
 251                 goto out;
 252         len = page->private >> 2;
 253         if (len == 0) {
 254                 WARN_ON(1);
 255         }
 256         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 257         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
 258                                              btrfs_header_generation(eb));
 259         BUG_ON(ret);
 260         btrfs_clear_buffer_defrag(eb);
 261         found_start = btrfs_header_bytenr(eb);
 262         if (found_start != start) {
 263                 printk("warning: eb start incorrect %Lu buffer %Lu len %lu\n",
 264                        start, found_start, len);
 265                 WARN_ON(1);
 266                 goto err;
 267         }
 268         if (eb->first_page != page) {
 269                 printk("bad first page %lu %lu\n", eb->first_page->index,
 270                        page->index);
 271                 WARN_ON(1);
 272                 goto err;
 273         }
 274         if (!PageUptodate(page)) {
 275                 printk("csum not up to date page %lu\n", page->index);
 276                 WARN_ON(1);
 277                 goto err;
 278         }
 279         found_level = btrfs_header_level(eb);
 280         spin_lock(&root->fs_info->hash_lock);
 281         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
 282         spin_unlock(&root->fs_info->hash_lock);
 283         csum_tree_block(root, eb, 0);
 284 err:
 285         free_extent_buffer(eb);
 286 out:
 287         return 0;
 288 }
 289
 290 static int btree_writepage_io_hook(struct page *page, u64 start, u64 end)
 291 {
 292         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 293
 294         csum_dirty_buffer(root, page);
 295         return 0;
 296 }
 297
 298 int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 299                                struct extent_state *state)
 300 {
 301         struct extent_io_tree *tree;
 302         u64 found_start;
 303         int found_level;
 304         unsigned long len;
 305         struct extent_buffer *eb;
 306         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 307         int ret = 0;
 308
 309         tree = &BTRFS_I(page->mapping->host)->io_tree;
 310         if (page->private == EXTENT_PAGE_PRIVATE)
 311                 goto out;
 312         if (!page->private)
 313                 goto out;
 314         len = page->private >> 2;
 315         if (len == 0) {
 316                 WARN_ON(1);
 317         }
 318         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 319
 320         btrfs_clear_buffer_defrag(eb);
 321         found_start = btrfs_header_bytenr(eb);
 322         if (found_start != start) {
 323                 ret = -EIO;
 324                 goto err;
 325         }
 326         if (eb->first_page != page) {
 327                 printk("bad first page %lu %lu\n", eb->first_page->index,
 328                        page->index);
 329                 WARN_ON(1);
 330                 ret = -EIO;
 331                 goto err;
 332         }
 333         found_level = btrfs_header_level(eb);
 334
 335         ret = csum_tree_block(root, eb, 1);
 336         if (ret)
 337                 ret = -EIO;
 338
 339         end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
 340         end = eb->start + end - 1;
 341         release_extent_buffer_tail_pages(eb);
 342 err:
 343         free_extent_buffer(eb);
 344 out:
 345         return ret;
 346 }
 347
 348 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
 349 static void end_workqueue_bio(struct bio *bio, int err)
 350 #else
 351 static int end_workqueue_bio(struct bio *bio,
 352                                    unsigned int bytes_done, int err)
 353 #endif
 354 {
 355         struct end_io_wq *end_io_wq = bio->bi_private;
 356         struct btrfs_fs_info *fs_info;
 357         unsigned long flags;
 358
 359 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
 360         if (bio->bi_size)
 361                 return 1;
 362 #endif
 363
 364         fs_info = end_io_wq->info;
 365         spin_lock_irqsave(&fs_info->end_io_work_lock, flags);
 366         end_io_wq->error = err;
 367         list_add_tail(&end_io_wq->list, &fs_info->end_io_work_list);
 368         spin_unlock_irqrestore(&fs_info->end_io_work_lock, flags);
 369         queue_work(end_io_workqueue, &fs_info->end_io_work);
 370
 371 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
 372         return 0;
 373 #endif
 374 }
 375
 376 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 377                         int metadata)
 378 {
 379         struct end_io_wq *end_io_wq;
 380         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
 381         if (!end_io_wq)
 382                 return -ENOMEM;
 383
 384         end_io_wq->private = bio->bi_private;
 385         end_io_wq->end_io = bio->bi_end_io;
 386         end_io_wq->info = info;
 387         end_io_wq->error = 0;
 388         end_io_wq->bio = bio;
 389         end_io_wq->metadata = metadata;
 390
 391         bio->bi_private = end_io_wq;
 392         bio->bi_end_io = end_workqueue_bio;
 393         return 0;
 394 }
 395
 396 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 397                         int rw, struct bio *bio, int mirror_num,
 398                         extent_submit_bio_hook_t *submit_bio_hook)
 399 {
 400         struct async_submit_bio *async;
 401
 402         /*
 403          * inline writerback should stay inline, only hop to the async
 404          * queue if we're pdflush
 405          */
 406         if (!current_is_pdflush())
 407                 return submit_bio_hook(inode, rw, bio, mirror_num);
 408
 409         async = kmalloc(sizeof(*async), GFP_NOFS);
 410         if (!async)
 411                 return -ENOMEM;
 412
 413         async->inode = inode;
 414         async->rw = rw;
 415         async->bio = bio;
 416         async->mirror_num = mirror_num;
 417         async->submit_bio_hook = submit_bio_hook;
 418
 419         spin_lock(&fs_info->async_submit_work_lock);
 420         list_add_tail(&async->list, &fs_info->async_submit_work_list);
 421         spin_unlock(&fs_info->async_submit_work_lock);
 422
 423         queue_work(async_submit_workqueue, &fs_info->async_submit_work);
 424         return 0;
 425 }
 426
 427 static int __btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 428                                  int mirror_num)
 429 {
 430         struct btrfs_root *root = BTRFS_I(inode)->root;
 431         u64 offset;
 432         int ret;
 433
 434         offset = bio->bi_sector << 9;
 435
 436         if (rw & (1 << BIO_RW)) {
 437                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num);
 438         }
 439
 440         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 1);
 441         BUG_ON(ret);
 442
 443         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num);
 444 }
 445
 446 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 447                                  int mirror_num)
 448 {
 449         if (!(rw & (1 << BIO_RW))) {
 450                 return __btree_submit_bio_hook(inode, rw, bio, mirror_num);
 451         }
 452         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 453                                    inode, rw, bio, mirror_num,
 454                                    __btree_submit_bio_hook);
 455 }
 456
 457 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 458 {
 459         struct extent_io_tree *tree;
 460         tree = &BTRFS_I(page->mapping->host)->io_tree;
 461         return extent_write_full_page(tree, page, btree_get_extent, wbc);
 462 }
 463
 464 static int btree_writepages(struct address_space *mapping,
 465                             struct writeback_control *wbc)
 466 {
 467         struct extent_io_tree *tree;
 468         tree = &BTRFS_I(mapping->host)->io_tree;
 469         if (wbc->sync_mode == WB_SYNC_NONE) {
 470                 u64 num_dirty;
 471                 u64 start = 0;
 472                 unsigned long thresh = 96 * 1024 * 1024;
 473
 474                 if (wbc->for_kupdate)
 475                         return 0;
 476
 477                 if (current_is_pdflush()) {
 478                         thresh = 96 * 1024 * 1024;
 479                 } else {
 480                         thresh = 8 * 1024 * 1024;
 481                 }
 482                 num_dirty = count_range_bits(tree, &start, (u64)-1,
 483                                              thresh, EXTENT_DIRTY);
 484                 if (num_dirty < thresh) {
 485                         return 0;
 486                 }
 487         }
 488         return extent_writepages(tree, mapping, btree_get_extent, wbc);
 489 }
 490
 491 int btree_readpage(struct file *file, struct page *page)
 492 {
 493         struct extent_io_tree *tree;
 494         tree = &BTRFS_I(page->mapping->host)->io_tree;
 495         return extent_read_full_page(tree, page, btree_get_extent);
 496 }
 497
 498 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 499 {
 500         struct extent_io_tree *tree;
 501         struct extent_map_tree *map;
 502         int ret;
 503
 504         if (page_count(page) > 3) {
 505                 /* once for page->private, once for the caller, once
 506                  * once for the page cache
 507                  */
 508                 return 0;
 509         }
 510         tree = &BTRFS_I(page->mapping->host)->io_tree;
 511         map = &BTRFS_I(page->mapping->host)->extent_tree;
 512         ret = try_release_extent_state(map, tree, page, gfp_flags);
 513         if (ret == 1) {
 514                 invalidate_extent_lru(tree, page_offset(page), PAGE_CACHE_SIZE);
 515                 ClearPagePrivate(page);
 516                 set_page_private(page, 0);
 517                 page_cache_release(page);
 518         }
 519         return ret;
 520 }
 521
 522 static void btree_invalidatepage(struct page *page, unsigned long offset)
 523 {
 524         struct extent_io_tree *tree;
 525         tree = &BTRFS_I(page->mapping->host)->io_tree;
 526         extent_invalidatepage(tree, page, offset);
 527         btree_releasepage(page, GFP_NOFS);
 528         if (PagePrivate(page)) {
 529                 invalidate_extent_lru(tree, page_offset(page), PAGE_CACHE_SIZE);
 530                 ClearPagePrivate(page);
 531                 set_page_private(page, 0);
 532                 page_cache_release(page);
 533         }
 534 }
 535
 536 #if 0
 537 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 538 {
 539         struct buffer_head *bh;
 540         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 541         struct buffer_head *head;
 542         if (!page_has_buffers(page)) {
 543                 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
 544                                         (1 << BH_Dirty)|(1 << BH_Uptodate));
 545         }
 546         head = page_buffers(page);
 547         bh = head;
 548         do {
 549                 if (buffer_dirty(bh))
 550                         csum_tree_block(root, bh, 0);
 551                 bh = bh->b_this_page;
 552         } while (bh != head);
 553         return block_write_full_page(page, btree_get_block, wbc);
 554 }
 555 #endif
 556
 557 static struct address_space_operations btree_aops = {
 558         .readpage       = btree_readpage,
 559         .writepage      = btree_writepage,
 560         .writepages     = btree_writepages,
 561         .releasepage    = btree_releasepage,
 562         .invalidatepage = btree_invalidatepage,
 563         .sync_page      = block_sync_page,
 564 };
 565
 566 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
 567                          u64 parent_transid)
 568 {
 569         struct extent_buffer *buf = NULL;
 570         struct inode *btree_inode = root->fs_info->btree_inode;
 571         int ret = 0;
 572
 573         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 574         if (!buf)
 575                 return 0;
 576         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
 577                                  buf, 0, 0, btree_get_extent, 0);
 578         free_extent_buffer(buf);
 579         return ret;
 580 }
 581
 582 static int close_all_devices(struct btrfs_fs_info *fs_info)
 583 {
 584         struct list_head *list;
 585         struct list_head *next;
 586         struct btrfs_device *device;
 587
 588         list = &fs_info->fs_devices->devices;
 589         list_for_each(next, list) {
 590                 device = list_entry(next, struct btrfs_device, dev_list);
 591                 close_bdev_excl(device->bdev);
 592                 device->bdev = NULL;
 593         }
 594         return 0;
 595 }
 596
 597 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
 598                                             u64 bytenr, u32 blocksize)
 599 {
 600         struct inode *btree_inode = root->fs_info->btree_inode;
 601         struct extent_buffer *eb;
 602         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 603                                 bytenr, blocksize, GFP_NOFS);
 604         return eb;
 605 }
 606
 607 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
 608                                                  u64 bytenr, u32 blocksize)
 609 {
 610         struct inode *btree_inode = root->fs_info->btree_inode;
 611         struct extent_buffer *eb;
 612
 613         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 614                                  bytenr, blocksize, NULL, GFP_NOFS);
 615         return eb;
 616 }
 617
 618
 619 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
 620                                       u32 blocksize, u64 parent_transid)
 621 {
 622         struct extent_buffer *buf = NULL;
 623         struct inode *btree_inode = root->fs_info->btree_inode;
 624         struct extent_io_tree *io_tree;
 625         int ret;
 626
 627         io_tree = &BTRFS_I(btree_inode)->io_tree;
 628
 629         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 630         if (!buf)
 631                 return NULL;
 632
 633         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
 634
 635         if (ret == 0) {
 636                 buf->flags |= EXTENT_UPTODATE;
 637         }
 638         return buf;
 639
 640 }
 641
 642 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 643                      struct extent_buffer *buf)
 644 {
 645         struct inode *btree_inode = root->fs_info->btree_inode;
 646         if (btrfs_header_generation(buf) ==
 647             root->fs_info->running_transaction->transid)
 648                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
 649                                           buf);
 650         return 0;
 651 }
 652
 653 int wait_on_tree_block_writeback(struct btrfs_root *root,
 654                                  struct extent_buffer *buf)
 655 {
 656         struct inode *btree_inode = root->fs_info->btree_inode;
 657         wait_on_extent_buffer_writeback(&BTRFS_I(btree_inode)->io_tree,
 658                                         buf);
 659         return 0;
 660 }
 661
 662 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
 663                         u32 stripesize, struct btrfs_root *root,
 664                         struct btrfs_fs_info *fs_info,
 665                         u64 objectid)
 666 {
 667         root->node = NULL;
 668         root->inode = NULL;
 669         root->commit_root = NULL;
 670         root->sectorsize = sectorsize;
 671         root->nodesize = nodesize;
 672         root->leafsize = leafsize;
 673         root->stripesize = stripesize;
 674         root->ref_cows = 0;
 675         root->track_dirty = 0;
 676
 677         root->fs_info = fs_info;
 678         root->objectid = objectid;
 679         root->last_trans = 0;
 680         root->highest_inode = 0;
 681         root->last_inode_alloc = 0;
 682         root->name = NULL;
 683         root->in_sysfs = 0;
 684
 685         INIT_LIST_HEAD(&root->dirty_list);
 686         memset(&root->root_key, 0, sizeof(root->root_key));
 687         memset(&root->root_item, 0, sizeof(root->root_item));
 688         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
 689         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
 690         init_completion(&root->kobj_unregister);
 691         root->defrag_running = 0;
 692         root->defrag_level = 0;
 693         root->root_key.objectid = objectid;
 694         return 0;
 695 }
 696
 697 static int find_and_setup_root(struct btrfs_root *tree_root,
 698                                struct btrfs_fs_info *fs_info,
 699                                u64 objectid,
 700                                struct btrfs_root *root)
 701 {
 702         int ret;
 703         u32 blocksize;
 704
 705         __setup_root(tree_root->nodesize, tree_root->leafsize,
 706                      tree_root->sectorsize, tree_root->stripesize,
 707                      root, fs_info, objectid);
 708         ret = btrfs_find_last_root(tree_root, objectid,
 709                                    &root->root_item, &root->root_key);
 710         BUG_ON(ret);
 711
 712         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 713         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 714                                      blocksize, 0);
 715         BUG_ON(!root->node);
 716         return 0;
 717 }
 718
 719 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_fs_info *fs_info,
 720                                                struct btrfs_key *location)
 721 {
 722         struct btrfs_root *root;
 723         struct btrfs_root *tree_root = fs_info->tree_root;
 724         struct btrfs_path *path;
 725         struct extent_buffer *l;
 726         u64 highest_inode;
 727         u32 blocksize;
 728         int ret = 0;
 729
 730         root = kzalloc(sizeof(*root), GFP_NOFS);
 731         if (!root)
 732                 return ERR_PTR(-ENOMEM);
 733         if (location->offset == (u64)-1) {
 734                 ret = find_and_setup_root(tree_root, fs_info,
 735                                           location->objectid, root);
 736                 if (ret) {
 737                         kfree(root);
 738                         return ERR_PTR(ret);
 739                 }
 740                 goto insert;
 741         }
 742
 743         __setup_root(tree_root->nodesize, tree_root->leafsize,
 744                      tree_root->sectorsize, tree_root->stripesize,
 745                      root, fs_info, location->objectid);
 746
 747         path = btrfs_alloc_path();
 748         BUG_ON(!path);
 749         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
 750         if (ret != 0) {
 751                 if (ret > 0)
 752                         ret = -ENOENT;
 753                 goto out;
 754         }
 755         l = path->nodes[0];
 756         read_extent_buffer(l, &root->root_item,
 757                btrfs_item_ptr_offset(l, path->slots[0]),
 758                sizeof(root->root_item));
 759         memcpy(&root->root_key, location, sizeof(*location));
 760         ret = 0;
 761 out:
 762         btrfs_release_path(root, path);
 763         btrfs_free_path(path);
 764         if (ret) {
 765                 kfree(root);
 766                 return ERR_PTR(ret);
 767         }
 768         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 769         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 770                                      blocksize, 0);
 771         BUG_ON(!root->node);
 772 insert:
 773         root->ref_cows = 1;
 774         ret = btrfs_find_highest_inode(root, &highest_inode);
 775         if (ret == 0) {
 776                 root->highest_inode = highest_inode;
 777                 root->last_inode_alloc = highest_inode;
 778         }
 779         return root;
 780 }
 781
 782 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
 783                                         u64 root_objectid)
 784 {
 785         struct btrfs_root *root;
 786
 787         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
 788                 return fs_info->tree_root;
 789         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
 790                 return fs_info->extent_root;
 791
 792         root = radix_tree_lookup(&fs_info->fs_roots_radix,
 793                                  (unsigned long)root_objectid);
 794         return root;
 795 }
 796
 797 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
 798                                               struct btrfs_key *location)
 799 {
 800         struct btrfs_root *root;
 801         int ret;
 802
 803         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
 804                 return fs_info->tree_root;
 805         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
 806                 return fs_info->extent_root;
 807         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
 808                 return fs_info->chunk_root;
 809         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
 810                 return fs_info->dev_root;
 811
 812         root = radix_tree_lookup(&fs_info->fs_roots_radix,
 813                                  (unsigned long)location->objectid);
 814         if (root)
 815                 return root;
 816
 817         root = btrfs_read_fs_root_no_radix(fs_info, location);
 818         if (IS_ERR(root))
 819                 return root;
 820         ret = radix_tree_insert(&fs_info->fs_roots_radix,
 821                                 (unsigned long)root->root_key.objectid,
 822                                 root);
 823         if (ret) {
 824                 free_extent_buffer(root->node);
 825                 kfree(root);
 826                 return ERR_PTR(ret);
 827         }
 828         ret = btrfs_find_dead_roots(fs_info->tree_root,
 829                                     root->root_key.objectid, root);
 830         BUG_ON(ret);
 831
 832         return root;
 833 }
 834
 835 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
 836                                       struct btrfs_key *location,
 837                                       const char *name, int namelen)
 838 {
 839         struct btrfs_root *root;
 840         int ret;
 841
 842         root = btrfs_read_fs_root_no_name(fs_info, location);
 843         if (!root)
 844                 return NULL;
 845
 846         if (root->in_sysfs)
 847                 return root;
 848
 849         ret = btrfs_set_root_name(root, name, namelen);
 850         if (ret) {
 851                 free_extent_buffer(root->node);
 852                 kfree(root);
 853                 return ERR_PTR(ret);
 854         }
 855
 856         ret = btrfs_sysfs_add_root(root);
 857         if (ret) {
 858                 free_extent_buffer(root->node);
 859                 kfree(root->name);
 860                 kfree(root);
 861                 return ERR_PTR(ret);
 862         }
 863         root->in_sysfs = 1;
 864         return root;
 865 }
 866 #if 0
 867 static int add_hasher(struct btrfs_fs_info *info, char *type) {
 868         struct btrfs_hasher *hasher;
 869
 870         hasher = kmalloc(sizeof(*hasher), GFP_NOFS);
 871         if (!hasher)
 872                 return -ENOMEM;
 873         hasher->hash_tfm = crypto_alloc_hash(type, 0, CRYPTO_ALG_ASYNC);
 874         if (!hasher->hash_tfm) {
 875                 kfree(hasher);
 876                 return -EINVAL;
 877         }
 878         spin_lock(&info->hash_lock);
 879         list_add(&hasher->list, &info->hashers);
 880         spin_unlock(&info->hash_lock);
 881         return 0;
 882 }
 883 #endif
 884
 885 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
 886 {
 887         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
 888         int ret = 0;
 889         struct list_head *cur;
 890         struct btrfs_device *device;
 891         struct backing_dev_info *bdi;
 892
 893         list_for_each(cur, &info->fs_devices->devices) {
 894                 device = list_entry(cur, struct btrfs_device, dev_list);
 895                 bdi = blk_get_backing_dev_info(device->bdev);
 896                 if (bdi && bdi_congested(bdi, bdi_bits)) {
 897                         ret = 1;
 898                         break;
 899                 }
 900         }
 901         return ret;
 902 }
 903
 904 /*
 905  * this unplugs every device on the box, and it is only used when page
 906  * is null
 907  */
 908 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
 909 {
 910         struct list_head *cur;
 911         struct btrfs_device *device;
 912         struct btrfs_fs_info *info;
 913
 914         info = (struct btrfs_fs_info *)bdi->unplug_io_data;
 915         list_for_each(cur, &info->fs_devices->devices) {
 916                 device = list_entry(cur, struct btrfs_device, dev_list);
 917                 bdi = blk_get_backing_dev_info(device->bdev);
 918                 if (bdi->unplug_io_fn) {
 919                         bdi->unplug_io_fn(bdi, page);
 920                 }
 921         }
 922 }
 923
 924 void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
 925 {
 926         struct inode *inode;
 927         struct extent_map_tree *em_tree;
 928         struct extent_map *em;
 929         struct address_space *mapping;
 930         u64 offset;
 931
 932         /* the generic O_DIRECT read code does this */
 933         if (!page) {
 934                 __unplug_io_fn(bdi, page);
 935                 return;
 936         }
 937
 938         /*
 939          * page->mapping may change at any time.  Get a consistent copy
 940          * and use that for everything below
 941          */
 942         smp_mb();
 943         mapping = page->mapping;
 944         if (!mapping)
 945                 return;
 946
 947         inode = mapping->host;
 948         offset = page_offset(page);
 949
 950         em_tree = &BTRFS_I(inode)->extent_tree;
 951         spin_lock(&em_tree->lock);
 952         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
 953         spin_unlock(&em_tree->lock);
 954         if (!em)
 955                 return;
 956
 957         offset = offset - em->start;
 958         btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
 959                           em->block_start + offset, page);
 960         free_extent_map(em);
 961 }
 962
 963 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
 964 {
 965 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,23)
 966         bdi_init(bdi);
 967 #endif
 968         bdi->ra_pages   = default_backing_dev_info.ra_pages;
 969         bdi->state              = 0;
 970         bdi->capabilities       = default_backing_dev_info.capabilities;
 971         bdi->unplug_io_fn       = btrfs_unplug_io_fn;
 972         bdi->unplug_io_data     = info;
 973         bdi->congested_fn       = btrfs_congested_fn;
 974         bdi->congested_data     = info;
 975         return 0;
 976 }
 977
 978 static int bio_ready_for_csum(struct bio *bio)
 979 {
 980         u64 length = 0;
 981         u64 buf_len = 0;
 982         u64 start = 0;
 983         struct page *page;
 984         struct extent_io_tree *io_tree = NULL;
 985         struct btrfs_fs_info *info = NULL;
 986         struct bio_vec *bvec;
 987         int i;
 988         int ret;
 989
 990         bio_for_each_segment(bvec, bio, i) {
 991                 page = bvec->bv_page;
 992                 if (page->private == EXTENT_PAGE_PRIVATE) {
 993                         length += bvec->bv_len;
 994                         continue;
 995                 }
 996                 if (!page->private) {
 997                         length += bvec->bv_len;
 998                         continue;
 999                 }
1000                 length = bvec->bv_len;
1001                 buf_len = page->private >> 2;
1002                 start = page_offset(page) + bvec->bv_offset;
1003                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1004                 info = BTRFS_I(page->mapping->host)->root->fs_info;
1005         }
1006         /* are we fully contained in this bio? */
1007         if (buf_len <= length)
1008                 return 1;
1009
1010         ret = extent_range_uptodate(io_tree, start + length,
1011                                     start + buf_len - 1);
1012         if (ret == 1)
1013                 return ret;
1014         return ret;
1015 }
1016
1017 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1018 static void btrfs_end_io_csum(void *p)
1019 #else
1020 static void btrfs_end_io_csum(struct work_struct *work)
1021 #endif
1022 {
1023 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1024         struct btrfs_fs_info *fs_info = p;
1025 #else
1026         struct btrfs_fs_info *fs_info = container_of(work,
1027                                                      struct btrfs_fs_info,
1028                                                      end_io_work);
1029 #endif
1030         unsigned long flags;
1031         struct end_io_wq *end_io_wq;
1032         struct bio *bio;
1033         struct list_head *next;
1034         int error;
1035         int was_empty;
1036
1037         while(1) {
1038                 spin_lock_irqsave(&fs_info->end_io_work_lock, flags);
1039                 if (list_empty(&fs_info->end_io_work_list)) {
1040                         spin_unlock_irqrestore(&fs_info->end_io_work_lock,
1041                                                flags);
1042                         return;
1043                 }
1044                 next = fs_info->end_io_work_list.next;
1045                 list_del(next);
1046                 spin_unlock_irqrestore(&fs_info->end_io_work_lock, flags);
1047
1048                 end_io_wq = list_entry(next, struct end_io_wq, list);
1049
1050                 bio = end_io_wq->bio;
1051                 if (end_io_wq->metadata && !bio_ready_for_csum(bio)) {
1052                         spin_lock_irqsave(&fs_info->end_io_work_lock, flags);
1053                         was_empty = list_empty(&fs_info->end_io_work_list);
1054                         list_add_tail(&end_io_wq->list,
1055                                       &fs_info->end_io_work_list);
1056                         spin_unlock_irqrestore(&fs_info->end_io_work_lock,
1057                                                flags);
1058                         if (was_empty)
1059                                 return;
1060                         continue;
1061                 }
1062                 error = end_io_wq->error;
1063                 bio->bi_private = end_io_wq->private;
1064                 bio->bi_end_io = end_io_wq->end_io;
1065                 kfree(end_io_wq);
1066 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1067                 bio_endio(bio, bio->bi_size, error);
1068 #else
1069                 bio_endio(bio, error);
1070 #endif
1071         }
1072 }
1073
1074 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1075 static void btrfs_async_submit_work(void *p)
1076 #else
1077 static void btrfs_async_submit_work(struct work_struct *work)
1078 #endif
1079 {
1080 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1081         struct btrfs_fs_info *fs_info = p;
1082 #else
1083         struct btrfs_fs_info *fs_info = container_of(work,
1084                                                      struct btrfs_fs_info,
1085                                                      async_submit_work);
1086 #endif
1087         struct async_submit_bio *async;
1088         struct list_head *next;
1089
1090         while(1) {
1091                 spin_lock(&fs_info->async_submit_work_lock);
1092                 if (list_empty(&fs_info->async_submit_work_list)) {
1093                         spin_unlock(&fs_info->async_submit_work_lock);
1094                         return;
1095                 }
1096                 next = fs_info->async_submit_work_list.next;
1097                 list_del(next);
1098                 spin_unlock(&fs_info->async_submit_work_lock);
1099
1100                 async = list_entry(next, struct async_submit_bio, list);
1101                 async->submit_bio_hook(async->inode, async->rw, async->bio,
1102                                        async->mirror_num);
1103                 kfree(async);
1104         }
1105 }
1106
1107 struct btrfs_root *open_ctree(struct super_block *sb,
1108                               struct btrfs_fs_devices *fs_devices)
1109 {
1110         u32 sectorsize;
1111         u32 nodesize;
1112         u32 leafsize;
1113         u32 blocksize;
1114         u32 stripesize;
1115         struct buffer_head *bh;
1116         struct btrfs_root *extent_root = kmalloc(sizeof(struct btrfs_root),
1117                                                  GFP_NOFS);
1118         struct btrfs_root *tree_root = kmalloc(sizeof(struct btrfs_root),
1119                                                GFP_NOFS);
1120         struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1121                                                 GFP_NOFS);
1122         struct btrfs_root *chunk_root = kmalloc(sizeof(struct btrfs_root),
1123                                                 GFP_NOFS);
1124         struct btrfs_root *dev_root = kmalloc(sizeof(struct btrfs_root),
1125                                               GFP_NOFS);
1126         int ret;
1127         int err = -EINVAL;
1128         struct btrfs_super_block *disk_super;
1129
1130         if (!extent_root || !tree_root || !fs_info) {
1131                 err = -ENOMEM;
1132                 goto fail;
1133         }
1134         end_io_workqueue = create_workqueue("btrfs-end-io");
1135         BUG_ON(!end_io_workqueue);
1136         async_submit_workqueue = create_workqueue("btrfs-async-submit");
1137
1138         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1139         INIT_LIST_HEAD(&fs_info->trans_list);
1140         INIT_LIST_HEAD(&fs_info->dead_roots);
1141         INIT_LIST_HEAD(&fs_info->hashers);
1142         INIT_LIST_HEAD(&fs_info->end_io_work_list);
1143         INIT_LIST_HEAD(&fs_info->async_submit_work_list);
1144         spin_lock_init(&fs_info->hash_lock);
1145         spin_lock_init(&fs_info->end_io_work_lock);
1146         spin_lock_init(&fs_info->async_submit_work_lock);
1147         spin_lock_init(&fs_info->delalloc_lock);
1148         spin_lock_init(&fs_info->new_trans_lock);
1149
1150         init_completion(&fs_info->kobj_unregister);
1151         fs_info->tree_root = tree_root;
1152         fs_info->extent_root = extent_root;
1153         fs_info->chunk_root = chunk_root;
1154         fs_info->dev_root = dev_root;
1155         fs_info->fs_devices = fs_devices;
1156         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1157         INIT_LIST_HEAD(&fs_info->space_info);
1158         btrfs_mapping_init(&fs_info->mapping_tree);
1159         fs_info->sb = sb;
1160         fs_info->max_extent = (u64)-1;
1161         fs_info->max_inline = 8192 * 1024;
1162         setup_bdi(fs_info, &fs_info->bdi);
1163         fs_info->btree_inode = new_inode(sb);
1164         fs_info->btree_inode->i_ino = 1;
1165         fs_info->btree_inode->i_nlink = 1;
1166
1167         sb->s_blocksize = 4096;
1168         sb->s_blocksize_bits = blksize_bits(4096);
1169
1170         /*
1171          * we set the i_size on the btree inode to the max possible int.
1172          * the real end of the address space is determined by all of
1173          * the devices in the system
1174          */
1175         fs_info->btree_inode->i_size = OFFSET_MAX;
1176         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1177         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1178
1179         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1180                              fs_info->btree_inode->i_mapping,
1181                              GFP_NOFS);
1182         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1183                              GFP_NOFS);
1184
1185         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1186
1187         extent_io_tree_init(&fs_info->free_space_cache,
1188                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1189         extent_io_tree_init(&fs_info->block_group_cache,
1190                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1191         extent_io_tree_init(&fs_info->pinned_extents,
1192                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1193         extent_io_tree_init(&fs_info->pending_del,
1194                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1195         extent_io_tree_init(&fs_info->extent_ins,
1196                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1197         fs_info->do_barriers = 1;
1198
1199 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1200         INIT_WORK(&fs_info->end_io_work, btrfs_end_io_csum, fs_info);
1201         INIT_WORK(&fs_info->async_submit_work, btrfs_async_submit_work,
1202                   fs_info);
1203         INIT_WORK(&fs_info->trans_work, btrfs_transaction_cleaner, fs_info);
1204 #else
1205         INIT_WORK(&fs_info->end_io_work, btrfs_end_io_csum);
1206         INIT_WORK(&fs_info->async_submit_work, btrfs_async_submit_work);
1207         INIT_DELAYED_WORK(&fs_info->trans_work, btrfs_transaction_cleaner);
1208 #endif
1209         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1210         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1211                sizeof(struct btrfs_key));
1212         insert_inode_hash(fs_info->btree_inode);
1213         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1214
1215         mutex_init(&fs_info->trans_mutex);
1216         mutex_init(&fs_info->fs_mutex);
1217
1218 #if 0
1219         ret = add_hasher(fs_info, "crc32c");
1220         if (ret) {
1221                 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
1222                 err = -ENOMEM;
1223                 goto fail_iput;
1224         }
1225 #endif
1226         __setup_root(4096, 4096, 4096, 4096, tree_root,
1227                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1228
1229
1230         bh = __bread(fs_devices->latest_bdev,
1231                      BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
1232         if (!bh)
1233                 goto fail_iput;
1234
1235         memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1236         brelse(bh);
1237
1238         memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1239
1240         disk_super = &fs_info->super_copy;
1241         if (!btrfs_super_root(disk_super))
1242                 goto fail_sb_buffer;
1243
1244         if (btrfs_super_num_devices(disk_super) != fs_devices->num_devices) {
1245                 printk("Btrfs: wanted %llu devices, but found %llu\n",
1246                        (unsigned long long)btrfs_super_num_devices(disk_super),
1247                        (unsigned long long)fs_devices->num_devices);
1248                 goto fail_sb_buffer;
1249         }
1250         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1251
1252         nodesize = btrfs_super_nodesize(disk_super);
1253         leafsize = btrfs_super_leafsize(disk_super);
1254         sectorsize = btrfs_super_sectorsize(disk_super);
1255         stripesize = btrfs_super_stripesize(disk_super);
1256         tree_root->nodesize = nodesize;
1257         tree_root->leafsize = leafsize;
1258         tree_root->sectorsize = sectorsize;
1259         tree_root->stripesize = stripesize;
1260
1261         sb->s_blocksize = sectorsize;
1262         sb->s_blocksize_bits = blksize_bits(sectorsize);
1263
1264         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1265                     sizeof(disk_super->magic))) {
1266                 printk("btrfs: valid FS not found on %s\n", sb->s_id);
1267                 goto fail_sb_buffer;
1268         }
1269
1270         mutex_lock(&fs_info->fs_mutex);
1271
1272         ret = btrfs_read_sys_array(tree_root);
1273         if (ret) {
1274                 printk("btrfs: failed to read the system array on %s\n",
1275                        sb->s_id);
1276                 goto fail_sys_array;
1277         }
1278
1279         blocksize = btrfs_level_size(tree_root,
1280                                      btrfs_super_chunk_root_level(disk_super));
1281
1282         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1283                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1284
1285         chunk_root->node = read_tree_block(chunk_root,
1286                                            btrfs_super_chunk_root(disk_super),
1287                                            blocksize, 0);
1288         BUG_ON(!chunk_root->node);
1289
1290         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1291                  (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1292                  BTRFS_UUID_SIZE);
1293
1294         ret = btrfs_read_chunk_tree(chunk_root);
1295         BUG_ON(ret);
1296
1297         blocksize = btrfs_level_size(tree_root,
1298                                      btrfs_super_root_level(disk_super));
1299
1300
1301         tree_root->node = read_tree_block(tree_root,
1302                                           btrfs_super_root(disk_super),
1303                                           blocksize, 0);
1304         if (!tree_root->node)
1305                 goto fail_sb_buffer;
1306
1307
1308         ret = find_and_setup_root(tree_root, fs_info,
1309                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1310         if (ret)
1311                 goto fail_tree_root;
1312         extent_root->track_dirty = 1;
1313
1314         ret = find_and_setup_root(tree_root, fs_info,
1315                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
1316         dev_root->track_dirty = 1;
1317
1318         if (ret)
1319                 goto fail_extent_root;
1320
1321         btrfs_read_block_groups(extent_root);
1322
1323         fs_info->generation = btrfs_super_generation(disk_super) + 1;
1324         fs_info->data_alloc_profile = (u64)-1;
1325         fs_info->metadata_alloc_profile = (u64)-1;
1326         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1327
1328         mutex_unlock(&fs_info->fs_mutex);
1329         return tree_root;
1330
1331 fail_extent_root:
1332         free_extent_buffer(extent_root->node);
1333 fail_tree_root:
1334         free_extent_buffer(tree_root->node);
1335 fail_sys_array:
1336         mutex_unlock(&fs_info->fs_mutex);
1337 fail_sb_buffer:
1338         extent_io_tree_empty_lru(&BTRFS_I(fs_info->btree_inode)->io_tree);
1339 fail_iput:
1340         iput(fs_info->btree_inode);
1341 fail:
1342         close_all_devices(fs_info);
1343         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1344
1345         kfree(extent_root);
1346         kfree(tree_root);
1347 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,23)
1348         bdi_destroy(&fs_info->bdi);
1349 #endif
1350         kfree(fs_info);
1351         return ERR_PTR(err);
1352 }
1353
1354 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1355 {
1356         char b[BDEVNAME_SIZE];
1357
1358         if (uptodate) {
1359                 set_buffer_uptodate(bh);
1360         } else {
1361                 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1362                         printk(KERN_WARNING "lost page write due to "
1363                                         "I/O error on %s\n",
1364                                        bdevname(bh->b_bdev, b));
1365                 }
1366                 set_buffer_write_io_error(bh);
1367                 clear_buffer_uptodate(bh);
1368         }
1369         unlock_buffer(bh);
1370         put_bh(bh);
1371 }
1372
1373 int write_all_supers(struct btrfs_root *root)
1374 {
1375         struct list_head *cur;
1376         struct list_head *head = &root->fs_info->fs_devices->devices;
1377         struct btrfs_device *dev;
1378         struct btrfs_super_block *sb;
1379         struct btrfs_dev_item *dev_item;
1380         struct buffer_head *bh;
1381         int ret;
1382         int do_barriers;
1383         int max_errors;
1384         int total_errors = 0;
1385         u32 crc;
1386         u64 flags;
1387
1388         max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1389         do_barriers = !btrfs_test_opt(root, NOBARRIER);
1390
1391         sb = &root->fs_info->super_for_commit;
1392         dev_item = &sb->dev_item;
1393         list_for_each(cur, head) {
1394                 dev = list_entry(cur, struct btrfs_device, dev_list);
1395                 btrfs_set_stack_device_type(dev_item, dev->type);
1396                 btrfs_set_stack_device_id(dev_item, dev->devid);
1397                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
1398                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
1399                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
1400                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
1401                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
1402                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
1403                 flags = btrfs_super_flags(sb);
1404                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
1405
1406
1407                 crc = ~(u32)0;
1408                 crc = btrfs_csum_data(root, (char *)sb + BTRFS_CSUM_SIZE, crc,
1409                                       BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
1410                 btrfs_csum_final(crc, sb->csum);
1411
1412                 bh = __getblk(dev->bdev, BTRFS_SUPER_INFO_OFFSET / 4096,
1413                               BTRFS_SUPER_INFO_SIZE);
1414
1415                 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
1416                 dev->pending_io = bh;
1417
1418                 get_bh(bh);
1419                 set_buffer_uptodate(bh);
1420                 lock_buffer(bh);
1421                 bh->b_end_io = btrfs_end_buffer_write_sync;
1422
1423                 if (do_barriers && dev->barriers) {
1424                         ret = submit_bh(WRITE_BARRIER, bh);
1425                         if (ret == -EOPNOTSUPP) {
1426                                 printk("btrfs: disabling barriers on dev %s\n",
1427                                        dev->name);
1428                                 set_buffer_uptodate(bh);
1429                                 dev->barriers = 0;
1430                                 get_bh(bh);
1431                                 lock_buffer(bh);
1432                                 ret = submit_bh(WRITE, bh);
1433                         }
1434                 } else {
1435                         ret = submit_bh(WRITE, bh);
1436                 }
1437                 if (ret)
1438                         total_errors++;
1439         }
1440         if (total_errors > max_errors) {
1441                 printk("btrfs: %d errors while writing supers\n", total_errors);
1442                 BUG();
1443         }
1444         total_errors = 0;
1445
1446         list_for_each(cur, head) {
1447                 dev = list_entry(cur, struct btrfs_device, dev_list);
1448                 BUG_ON(!dev->pending_io);
1449                 bh = dev->pending_io;
1450                 wait_on_buffer(bh);
1451                 if (!buffer_uptodate(dev->pending_io)) {
1452                         if (do_barriers && dev->barriers) {
1453                                 printk("btrfs: disabling barriers on dev %s\n",
1454                                        dev->name);
1455                                 set_buffer_uptodate(bh);
1456                                 get_bh(bh);
1457                                 lock_buffer(bh);
1458                                 dev->barriers = 0;
1459                                 ret = submit_bh(WRITE, bh);
1460                                 BUG_ON(ret);
1461                                 wait_on_buffer(bh);
1462                                 BUG_ON(!buffer_uptodate(bh));
1463                         } else {
1464                                 total_errors++;
1465                         }
1466
1467                 }
1468                 dev->pending_io = NULL;
1469                 brelse(bh);
1470         }
1471         if (total_errors > max_errors) {
1472                 printk("btrfs: %d errors while writing supers\n", total_errors);
1473                 BUG();
1474         }
1475         return 0;
1476 }
1477
1478 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
1479                       *root)
1480 {
1481         int ret;
1482
1483         ret = write_all_supers(root);
1484         return ret;
1485 }
1486
1487 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
1488 {
1489         radix_tree_delete(&fs_info->fs_roots_radix,
1490                           (unsigned long)root->root_key.objectid);
1491         if (root->in_sysfs)
1492                 btrfs_sysfs_del_root(root);
1493         if (root->inode)
1494                 iput(root->inode);
1495         if (root->node)
1496                 free_extent_buffer(root->node);
1497         if (root->commit_root)
1498                 free_extent_buffer(root->commit_root);
1499         if (root->name)
1500                 kfree(root->name);
1501         kfree(root);
1502         return 0;
1503 }
1504
1505 static int del_fs_roots(struct btrfs_fs_info *fs_info)
1506 {
1507         int ret;
1508         struct btrfs_root *gang[8];
1509         int i;
1510
1511         while(1) {
1512                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1513                                              (void **)gang, 0,
1514                                              ARRAY_SIZE(gang));
1515                 if (!ret)
1516                         break;
1517                 for (i = 0; i < ret; i++)
1518                         btrfs_free_fs_root(fs_info, gang[i]);
1519         }
1520         return 0;
1521 }
1522
1523 int close_ctree(struct btrfs_root *root)
1524 {
1525         int ret;
1526         struct btrfs_trans_handle *trans;
1527         struct btrfs_fs_info *fs_info = root->fs_info;
1528
1529         fs_info->closing = 1;
1530         btrfs_transaction_flush_work(root);
1531         mutex_lock(&fs_info->fs_mutex);
1532         btrfs_defrag_dirty_roots(root->fs_info);
1533         trans = btrfs_start_transaction(root, 1);
1534         ret = btrfs_commit_transaction(trans, root);
1535         /* run commit again to  drop the original snapshot */
1536         trans = btrfs_start_transaction(root, 1);
1537         btrfs_commit_transaction(trans, root);
1538         ret = btrfs_write_and_wait_transaction(NULL, root);
1539         BUG_ON(ret);
1540
1541         write_ctree_super(NULL, root);
1542         mutex_unlock(&fs_info->fs_mutex);
1543
1544         btrfs_transaction_flush_work(root);
1545
1546         if (fs_info->delalloc_bytes) {
1547                 printk("btrfs: at unmount delalloc count %Lu\n",
1548                        fs_info->delalloc_bytes);
1549         }
1550         if (fs_info->extent_root->node)
1551                 free_extent_buffer(fs_info->extent_root->node);
1552
1553         if (fs_info->tree_root->node)
1554                 free_extent_buffer(fs_info->tree_root->node);
1555
1556         if (root->fs_info->chunk_root->node);
1557                 free_extent_buffer(root->fs_info->chunk_root->node);
1558
1559         if (root->fs_info->dev_root->node);
1560                 free_extent_buffer(root->fs_info->dev_root->node);
1561
1562         btrfs_free_block_groups(root->fs_info);
1563         del_fs_roots(fs_info);
1564
1565         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1566
1567         extent_io_tree_empty_lru(&fs_info->free_space_cache);
1568         extent_io_tree_empty_lru(&fs_info->block_group_cache);
1569         extent_io_tree_empty_lru(&fs_info->pinned_extents);
1570         extent_io_tree_empty_lru(&fs_info->pending_del);
1571         extent_io_tree_empty_lru(&fs_info->extent_ins);
1572         extent_io_tree_empty_lru(&BTRFS_I(fs_info->btree_inode)->io_tree);
1573
1574         flush_workqueue(async_submit_workqueue);
1575         flush_workqueue(end_io_workqueue);
1576
1577         truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
1578
1579         flush_workqueue(async_submit_workqueue);
1580         destroy_workqueue(async_submit_workqueue);
1581
1582         flush_workqueue(end_io_workqueue);
1583         destroy_workqueue(end_io_workqueue);
1584
1585         iput(fs_info->btree_inode);
1586 #if 0
1587         while(!list_empty(&fs_info->hashers)) {
1588                 struct btrfs_hasher *hasher;
1589                 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
1590                                     hashers);
1591                 list_del(&hasher->hashers);
1592                 crypto_free_hash(&fs_info->hash_tfm);
1593                 kfree(hasher);
1594         }
1595 #endif
1596         close_all_devices(fs_info);
1597         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1598
1599 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,23)
1600         bdi_destroy(&fs_info->bdi);
1601 #endif
1602
1603         kfree(fs_info->extent_root);
1604         kfree(fs_info->tree_root);
1605         kfree(fs_info->chunk_root);
1606         kfree(fs_info->dev_root);
1607         return 0;
1608 }
1609
1610 int btrfs_buffer_uptodate(struct extent_buffer *buf)
1611 {
1612         struct inode *btree_inode = buf->first_page->mapping->host;
1613         return extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
1614 }
1615
1616 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
1617 {
1618         struct inode *btree_inode = buf->first_page->mapping->host;
1619         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
1620                                           buf);
1621 }
1622
1623 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
1624 {
1625         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1626         u64 transid = btrfs_header_generation(buf);
1627         struct inode *btree_inode = root->fs_info->btree_inode;
1628
1629         if (transid != root->fs_info->generation) {
1630                 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
1631                         (unsigned long long)buf->start,
1632                         transid, root->fs_info->generation);
1633                 WARN_ON(1);
1634         }
1635         set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
1636 }
1637
1638 void btrfs_throttle(struct btrfs_root *root)
1639 {
1640         struct backing_dev_info *bdi;
1641
1642         bdi = &root->fs_info->bdi;
1643         if (root->fs_info->throttles && bdi_write_congested(bdi)) {
1644 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,18)
1645                 congestion_wait(WRITE, HZ/20);
1646 #else
1647                 blk_congestion_wait(WRITE, HZ/20);
1648 #endif
1649         }
1650 }
1651
1652 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
1653 {
1654         /*
1655          * looks as though older kernels can get into trouble with
1656          * this code, they end up stuck in balance_dirty_pages forever
1657          */
1658 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,23)
1659         struct extent_io_tree *tree;
1660         u64 num_dirty;
1661         u64 start = 0;
1662         unsigned long thresh = 16 * 1024 * 1024;
1663         tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
1664
1665         if (current_is_pdflush())
1666                 return;
1667
1668         num_dirty = count_range_bits(tree, &start, (u64)-1,
1669                                      thresh, EXTENT_DIRTY);
1670         if (num_dirty > thresh) {
1671                 balance_dirty_pages_ratelimited_nr(
1672                                    root->fs_info->btree_inode->i_mapping, 1);
1673         }
1674 #else
1675         return;
1676 #endif
1677 }
1678
1679 void btrfs_set_buffer_defrag(struct extent_buffer *buf)
1680 {
1681         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1682         struct inode *btree_inode = root->fs_info->btree_inode;
1683         set_extent_bits(&BTRFS_I(btree_inode)->io_tree, buf->start,
1684                         buf->start + buf->len - 1, EXTENT_DEFRAG, GFP_NOFS);
1685 }
1686
1687 void btrfs_set_buffer_defrag_done(struct extent_buffer *buf)
1688 {
1689         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1690         struct inode *btree_inode = root->fs_info->btree_inode;
1691         set_extent_bits(&BTRFS_I(btree_inode)->io_tree, buf->start,
1692                         buf->start + buf->len - 1, EXTENT_DEFRAG_DONE,
1693                         GFP_NOFS);
1694 }
1695
1696 int btrfs_buffer_defrag(struct extent_buffer *buf)
1697 {
1698         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1699         struct inode *btree_inode = root->fs_info->btree_inode;
1700         return test_range_bit(&BTRFS_I(btree_inode)->io_tree,
1701                      buf->start, buf->start + buf->len - 1, EXTENT_DEFRAG, 0);
1702 }
1703
1704 int btrfs_buffer_defrag_done(struct extent_buffer *buf)
1705 {
1706         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1707         struct inode *btree_inode = root->fs_info->btree_inode;
1708         return test_range_bit(&BTRFS_I(btree_inode)->io_tree,
1709                      buf->start, buf->start + buf->len - 1,
1710                      EXTENT_DEFRAG_DONE, 0);
1711 }
1712
1713 int btrfs_clear_buffer_defrag_done(struct extent_buffer *buf)
1714 {
1715         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1716         struct inode *btree_inode = root->fs_info->btree_inode;
1717         return clear_extent_bits(&BTRFS_I(btree_inode)->io_tree,
1718                      buf->start, buf->start + buf->len - 1,
1719                      EXTENT_DEFRAG_DONE, GFP_NOFS);
1720 }
1721
1722 int btrfs_clear_buffer_defrag(struct extent_buffer *buf)
1723 {
1724         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1725         struct inode *btree_inode = root->fs_info->btree_inode;
1726         return clear_extent_bits(&BTRFS_I(btree_inode)->io_tree,
1727                      buf->start, buf->start + buf->len - 1,
1728                      EXTENT_DEFRAG, GFP_NOFS);
1729 }
1730
1731 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
1732 {
1733         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1734         int ret;
1735         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1736         if (ret == 0) {
1737                 buf->flags |= EXTENT_UPTODATE;
1738         }
1739         return ret;
1740 }
1741
1742 static struct extent_io_ops btree_extent_io_ops = {
1743         .writepage_io_hook = btree_writepage_io_hook,
1744         .readpage_end_io_hook = btree_readpage_end_io_hook,
1745         .submit_bio_hook = btree_submit_bio_hook,
1746         /* note we're sharing with inode.c for the merge bio hook */
1747         .merge_bio_hook = btrfs_merge_bio_hook,
1748 };