SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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 verify_parent_transid(struct extent_io_tree *io_tree,
 209                                  struct extent_buffer *eb, u64 parent_transid)
 210 {
 211         int ret;
 212
 213         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 214                 return 0;
 215
 216         lock_extent(io_tree, eb->start, eb->start + eb->len - 1, GFP_NOFS);
 217         if (extent_buffer_uptodate(io_tree, eb) &&
 218             btrfs_header_generation(eb) == parent_transid) {
 219                 ret = 0;
 220                 goto out;
 221         }
 222         printk("parent transid verify failed on %llu wanted %llu found %llu\n",
 223                (unsigned long long)eb->start,
 224                (unsigned long long)parent_transid,
 225                (unsigned long long)btrfs_header_generation(eb));
 226         ret = 1;
 227 out:
 228         clear_extent_buffer_uptodate(io_tree, eb);
 229         unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
 230                       GFP_NOFS);
 231         return ret;
 232
 233 }
 234
 235 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 236                                           struct extent_buffer *eb,
 237                                           u64 start, u64 parent_transid)
 238 {
 239         struct extent_io_tree *io_tree;
 240         int ret;
 241         int num_copies = 0;
 242         int mirror_num = 0;
 243
 244         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 245         while (1) {
 246                 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
 247                                                btree_get_extent, mirror_num);
 248                 if (!ret &&
 249                     !verify_parent_transid(io_tree, eb, parent_transid))
 250                         return ret;
 251
 252                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
 253                                               eb->start, eb->len);
 254                 if (num_copies == 1)
 255                         return ret;
 256
 257                 mirror_num++;
 258                 if (mirror_num > num_copies)
 259                         return ret;
 260         }
 261         return -EIO;
 262 }
 263
 264 int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 265 {
 266         struct extent_io_tree *tree;
 267         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
 268         u64 found_start;
 269         int found_level;
 270         unsigned long len;
 271         struct extent_buffer *eb;
 272         int ret;
 273
 274         tree = &BTRFS_I(page->mapping->host)->io_tree;
 275
 276         if (page->private == EXTENT_PAGE_PRIVATE)
 277                 goto out;
 278         if (!page->private)
 279                 goto out;
 280         len = page->private >> 2;
 281         if (len == 0) {
 282                 WARN_ON(1);
 283         }
 284         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 285         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
 286                                              btrfs_header_generation(eb));
 287         BUG_ON(ret);
 288         btrfs_clear_buffer_defrag(eb);
 289         found_start = btrfs_header_bytenr(eb);
 290         if (found_start != start) {
 291                 printk("warning: eb start incorrect %Lu buffer %Lu len %lu\n",
 292                        start, found_start, len);
 293                 WARN_ON(1);
 294                 goto err;
 295         }
 296         if (eb->first_page != page) {
 297                 printk("bad first page %lu %lu\n", eb->first_page->index,
 298                        page->index);
 299                 WARN_ON(1);
 300                 goto err;
 301         }
 302         if (!PageUptodate(page)) {
 303                 printk("csum not up to date page %lu\n", page->index);
 304                 WARN_ON(1);
 305                 goto err;
 306         }
 307         found_level = btrfs_header_level(eb);
 308         spin_lock(&root->fs_info->hash_lock);
 309         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
 310         spin_unlock(&root->fs_info->hash_lock);
 311         csum_tree_block(root, eb, 0);
 312 err:
 313         free_extent_buffer(eb);
 314 out:
 315         return 0;
 316 }
 317
 318 static int btree_writepage_io_hook(struct page *page, u64 start, u64 end)
 319 {
 320         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 321
 322         csum_dirty_buffer(root, page);
 323         return 0;
 324 }
 325
 326 int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 327                                struct extent_state *state)
 328 {
 329         struct extent_io_tree *tree;
 330         u64 found_start;
 331         int found_level;
 332         unsigned long len;
 333         struct extent_buffer *eb;
 334         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 335         int ret = 0;
 336
 337         tree = &BTRFS_I(page->mapping->host)->io_tree;
 338         if (page->private == EXTENT_PAGE_PRIVATE)
 339                 goto out;
 340         if (!page->private)
 341                 goto out;
 342         len = page->private >> 2;
 343         if (len == 0) {
 344                 WARN_ON(1);
 345         }
 346         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 347
 348         btrfs_clear_buffer_defrag(eb);
 349         found_start = btrfs_header_bytenr(eb);
 350         if (found_start != start) {
 351                 ret = -EIO;
 352                 goto err;
 353         }
 354         if (eb->first_page != page) {
 355                 printk("bad first page %lu %lu\n", eb->first_page->index,
 356                        page->index);
 357                 WARN_ON(1);
 358                 ret = -EIO;
 359                 goto err;
 360         }
 361         if (memcmp_extent_buffer(eb, root->fs_info->fsid,
 362                                  (unsigned long)btrfs_header_fsid(eb),
 363                                  BTRFS_FSID_SIZE)) {
 364                 printk("bad fsid on block %Lu\n", eb->start);
 365                 ret = -EIO;
 366                 goto err;
 367         }
 368         found_level = btrfs_header_level(eb);
 369
 370         ret = csum_tree_block(root, eb, 1);
 371         if (ret)
 372                 ret = -EIO;
 373
 374         end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
 375         end = eb->start + end - 1;
 376         release_extent_buffer_tail_pages(eb);
 377 err:
 378         free_extent_buffer(eb);
 379 out:
 380         return ret;
 381 }
 382
 383 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
 384 static void end_workqueue_bio(struct bio *bio, int err)
 385 #else
 386 static int end_workqueue_bio(struct bio *bio,
 387                                    unsigned int bytes_done, int err)
 388 #endif
 389 {
 390         struct end_io_wq *end_io_wq = bio->bi_private;
 391         struct btrfs_fs_info *fs_info;
 392         unsigned long flags;
 393
 394 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
 395         if (bio->bi_size)
 396                 return 1;
 397 #endif
 398
 399         fs_info = end_io_wq->info;
 400         spin_lock_irqsave(&fs_info->end_io_work_lock, flags);
 401         end_io_wq->error = err;
 402         list_add_tail(&end_io_wq->list, &fs_info->end_io_work_list);
 403         spin_unlock_irqrestore(&fs_info->end_io_work_lock, flags);
 404         queue_work(end_io_workqueue, &fs_info->end_io_work);
 405
 406 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
 407         return 0;
 408 #endif
 409 }
 410
 411 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 412                         int metadata)
 413 {
 414         struct end_io_wq *end_io_wq;
 415         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
 416         if (!end_io_wq)
 417                 return -ENOMEM;
 418
 419         end_io_wq->private = bio->bi_private;
 420         end_io_wq->end_io = bio->bi_end_io;
 421         end_io_wq->info = info;
 422         end_io_wq->error = 0;
 423         end_io_wq->bio = bio;
 424         end_io_wq->metadata = metadata;
 425
 426         bio->bi_private = end_io_wq;
 427         bio->bi_end_io = end_workqueue_bio;
 428         return 0;
 429 }
 430
 431 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 432                         int rw, struct bio *bio, int mirror_num,
 433                         extent_submit_bio_hook_t *submit_bio_hook)
 434 {
 435         struct async_submit_bio *async;
 436
 437         /*
 438          * inline writerback should stay inline, only hop to the async
 439          * queue if we're pdflush
 440          */
 441         if (!current_is_pdflush())
 442                 return submit_bio_hook(inode, rw, bio, mirror_num);
 443
 444         async = kmalloc(sizeof(*async), GFP_NOFS);
 445         if (!async)
 446                 return -ENOMEM;
 447
 448         async->inode = inode;
 449         async->rw = rw;
 450         async->bio = bio;
 451         async->mirror_num = mirror_num;
 452         async->submit_bio_hook = submit_bio_hook;
 453
 454         spin_lock(&fs_info->async_submit_work_lock);
 455         list_add_tail(&async->list, &fs_info->async_submit_work_list);
 456         spin_unlock(&fs_info->async_submit_work_lock);
 457
 458         queue_work(async_submit_workqueue, &fs_info->async_submit_work);
 459         return 0;
 460 }
 461
 462 static int __btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 463                                  int mirror_num)
 464 {
 465         struct btrfs_root *root = BTRFS_I(inode)->root;
 466         u64 offset;
 467         int ret;
 468
 469         offset = bio->bi_sector << 9;
 470
 471         if (rw & (1 << BIO_RW)) {
 472                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num);
 473         }
 474
 475         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 1);
 476         BUG_ON(ret);
 477
 478         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num);
 479 }
 480
 481 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 482                                  int mirror_num)
 483 {
 484         if (!(rw & (1 << BIO_RW))) {
 485                 return __btree_submit_bio_hook(inode, rw, bio, mirror_num);
 486         }
 487         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 488                                    inode, rw, bio, mirror_num,
 489                                    __btree_submit_bio_hook);
 490 }
 491
 492 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 493 {
 494         struct extent_io_tree *tree;
 495         tree = &BTRFS_I(page->mapping->host)->io_tree;
 496         return extent_write_full_page(tree, page, btree_get_extent, wbc);
 497 }
 498
 499 static int btree_writepages(struct address_space *mapping,
 500                             struct writeback_control *wbc)
 501 {
 502         struct extent_io_tree *tree;
 503         tree = &BTRFS_I(mapping->host)->io_tree;
 504         if (wbc->sync_mode == WB_SYNC_NONE) {
 505                 u64 num_dirty;
 506                 u64 start = 0;
 507                 unsigned long thresh = 96 * 1024 * 1024;
 508
 509                 if (wbc->for_kupdate)
 510                         return 0;
 511
 512                 if (current_is_pdflush()) {
 513                         thresh = 96 * 1024 * 1024;
 514                 } else {
 515                         thresh = 8 * 1024 * 1024;
 516                 }
 517                 num_dirty = count_range_bits(tree, &start, (u64)-1,
 518                                              thresh, EXTENT_DIRTY);
 519                 if (num_dirty < thresh) {
 520                         return 0;
 521                 }
 522         }
 523         return extent_writepages(tree, mapping, btree_get_extent, wbc);
 524 }
 525
 526 int btree_readpage(struct file *file, struct page *page)
 527 {
 528         struct extent_io_tree *tree;
 529         tree = &BTRFS_I(page->mapping->host)->io_tree;
 530         return extent_read_full_page(tree, page, btree_get_extent);
 531 }
 532
 533 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 534 {
 535         struct extent_io_tree *tree;
 536         struct extent_map_tree *map;
 537         int ret;
 538
 539         if (page_count(page) > 3) {
 540                 /* once for page->private, once for the caller, once
 541                  * once for the page cache
 542                  */
 543                 return 0;
 544         }
 545         tree = &BTRFS_I(page->mapping->host)->io_tree;
 546         map = &BTRFS_I(page->mapping->host)->extent_tree;
 547         ret = try_release_extent_state(map, tree, page, gfp_flags);
 548         if (ret == 1) {
 549                 invalidate_extent_lru(tree, page_offset(page), PAGE_CACHE_SIZE);
 550                 ClearPagePrivate(page);
 551                 set_page_private(page, 0);
 552                 page_cache_release(page);
 553         }
 554         return ret;
 555 }
 556
 557 static void btree_invalidatepage(struct page *page, unsigned long offset)
 558 {
 559         struct extent_io_tree *tree;
 560         tree = &BTRFS_I(page->mapping->host)->io_tree;
 561         extent_invalidatepage(tree, page, offset);
 562         btree_releasepage(page, GFP_NOFS);
 563         if (PagePrivate(page)) {
 564                 invalidate_extent_lru(tree, page_offset(page), PAGE_CACHE_SIZE);
 565                 ClearPagePrivate(page);
 566                 set_page_private(page, 0);
 567                 page_cache_release(page);
 568         }
 569 }
 570
 571 #if 0
 572 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 573 {
 574         struct buffer_head *bh;
 575         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 576         struct buffer_head *head;
 577         if (!page_has_buffers(page)) {
 578                 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
 579                                         (1 << BH_Dirty)|(1 << BH_Uptodate));
 580         }
 581         head = page_buffers(page);
 582         bh = head;
 583         do {
 584                 if (buffer_dirty(bh))
 585                         csum_tree_block(root, bh, 0);
 586                 bh = bh->b_this_page;
 587         } while (bh != head);
 588         return block_write_full_page(page, btree_get_block, wbc);
 589 }
 590 #endif
 591
 592 static struct address_space_operations btree_aops = {
 593         .readpage       = btree_readpage,
 594         .writepage      = btree_writepage,
 595         .writepages     = btree_writepages,
 596         .releasepage    = btree_releasepage,
 597         .invalidatepage = btree_invalidatepage,
 598         .sync_page      = block_sync_page,
 599 };
 600
 601 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
 602                          u64 parent_transid)
 603 {
 604         struct extent_buffer *buf = NULL;
 605         struct inode *btree_inode = root->fs_info->btree_inode;
 606         int ret = 0;
 607
 608         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 609         if (!buf)
 610                 return 0;
 611         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
 612                                  buf, 0, 0, btree_get_extent, 0);
 613         free_extent_buffer(buf);
 614         return ret;
 615 }
 616
 617 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
 618                                             u64 bytenr, u32 blocksize)
 619 {
 620         struct inode *btree_inode = root->fs_info->btree_inode;
 621         struct extent_buffer *eb;
 622         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 623                                 bytenr, blocksize, GFP_NOFS);
 624         return eb;
 625 }
 626
 627 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
 628                                                  u64 bytenr, u32 blocksize)
 629 {
 630         struct inode *btree_inode = root->fs_info->btree_inode;
 631         struct extent_buffer *eb;
 632
 633         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 634                                  bytenr, blocksize, NULL, GFP_NOFS);
 635         return eb;
 636 }
 637
 638
 639 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
 640                                       u32 blocksize, u64 parent_transid)
 641 {
 642         struct extent_buffer *buf = NULL;
 643         struct inode *btree_inode = root->fs_info->btree_inode;
 644         struct extent_io_tree *io_tree;
 645         int ret;
 646
 647         io_tree = &BTRFS_I(btree_inode)->io_tree;
 648
 649         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 650         if (!buf)
 651                 return NULL;
 652
 653         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
 654
 655         if (ret == 0) {
 656                 buf->flags |= EXTENT_UPTODATE;
 657         }
 658         return buf;
 659
 660 }
 661
 662 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 663                      struct extent_buffer *buf)
 664 {
 665         struct inode *btree_inode = root->fs_info->btree_inode;
 666         if (btrfs_header_generation(buf) ==
 667             root->fs_info->running_transaction->transid)
 668                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
 669                                           buf);
 670         return 0;
 671 }
 672
 673 int wait_on_tree_block_writeback(struct btrfs_root *root,
 674                                  struct extent_buffer *buf)
 675 {
 676         struct inode *btree_inode = root->fs_info->btree_inode;
 677         wait_on_extent_buffer_writeback(&BTRFS_I(btree_inode)->io_tree,
 678                                         buf);
 679         return 0;
 680 }
 681
 682 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
 683                         u32 stripesize, struct btrfs_root *root,
 684                         struct btrfs_fs_info *fs_info,
 685                         u64 objectid)
 686 {
 687         root->node = NULL;
 688         root->inode = NULL;
 689         root->commit_root = NULL;
 690         root->sectorsize = sectorsize;
 691         root->nodesize = nodesize;
 692         root->leafsize = leafsize;
 693         root->stripesize = stripesize;
 694         root->ref_cows = 0;
 695         root->track_dirty = 0;
 696
 697         root->fs_info = fs_info;
 698         root->objectid = objectid;
 699         root->last_trans = 0;
 700         root->highest_inode = 0;
 701         root->last_inode_alloc = 0;
 702         root->name = NULL;
 703         root->in_sysfs = 0;
 704
 705         INIT_LIST_HEAD(&root->dirty_list);
 706         memset(&root->root_key, 0, sizeof(root->root_key));
 707         memset(&root->root_item, 0, sizeof(root->root_item));
 708         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
 709         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
 710         init_completion(&root->kobj_unregister);
 711         root->defrag_running = 0;
 712         root->defrag_level = 0;
 713         root->root_key.objectid = objectid;
 714         return 0;
 715 }
 716
 717 static int find_and_setup_root(struct btrfs_root *tree_root,
 718                                struct btrfs_fs_info *fs_info,
 719                                u64 objectid,
 720                                struct btrfs_root *root)
 721 {
 722         int ret;
 723         u32 blocksize;
 724
 725         __setup_root(tree_root->nodesize, tree_root->leafsize,
 726                      tree_root->sectorsize, tree_root->stripesize,
 727                      root, fs_info, objectid);
 728         ret = btrfs_find_last_root(tree_root, objectid,
 729                                    &root->root_item, &root->root_key);
 730         BUG_ON(ret);
 731
 732         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 733         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 734                                      blocksize, 0);
 735         BUG_ON(!root->node);
 736         return 0;
 737 }
 738
 739 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_fs_info *fs_info,
 740                                                struct btrfs_key *location)
 741 {
 742         struct btrfs_root *root;
 743         struct btrfs_root *tree_root = fs_info->tree_root;
 744         struct btrfs_path *path;
 745         struct extent_buffer *l;
 746         u64 highest_inode;
 747         u32 blocksize;
 748         int ret = 0;
 749
 750         root = kzalloc(sizeof(*root), GFP_NOFS);
 751         if (!root)
 752                 return ERR_PTR(-ENOMEM);
 753         if (location->offset == (u64)-1) {
 754                 ret = find_and_setup_root(tree_root, fs_info,
 755                                           location->objectid, root);
 756                 if (ret) {
 757                         kfree(root);
 758                         return ERR_PTR(ret);
 759                 }
 760                 goto insert;
 761         }
 762
 763         __setup_root(tree_root->nodesize, tree_root->leafsize,
 764                      tree_root->sectorsize, tree_root->stripesize,
 765                      root, fs_info, location->objectid);
 766
 767         path = btrfs_alloc_path();
 768         BUG_ON(!path);
 769         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
 770         if (ret != 0) {
 771                 if (ret > 0)
 772                         ret = -ENOENT;
 773                 goto out;
 774         }
 775         l = path->nodes[0];
 776         read_extent_buffer(l, &root->root_item,
 777                btrfs_item_ptr_offset(l, path->slots[0]),
 778                sizeof(root->root_item));
 779         memcpy(&root->root_key, location, sizeof(*location));
 780         ret = 0;
 781 out:
 782         btrfs_release_path(root, path);
 783         btrfs_free_path(path);
 784         if (ret) {
 785                 kfree(root);
 786                 return ERR_PTR(ret);
 787         }
 788         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 789         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 790                                      blocksize, 0);
 791         BUG_ON(!root->node);
 792 insert:
 793         root->ref_cows = 1;
 794         ret = btrfs_find_highest_inode(root, &highest_inode);
 795         if (ret == 0) {
 796                 root->highest_inode = highest_inode;
 797                 root->last_inode_alloc = highest_inode;
 798         }
 799         return root;
 800 }
 801
 802 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
 803                                         u64 root_objectid)
 804 {
 805         struct btrfs_root *root;
 806
 807         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
 808                 return fs_info->tree_root;
 809         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
 810                 return fs_info->extent_root;
 811
 812         root = radix_tree_lookup(&fs_info->fs_roots_radix,
 813                                  (unsigned long)root_objectid);
 814         return root;
 815 }
 816
 817 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
 818                                               struct btrfs_key *location)
 819 {
 820         struct btrfs_root *root;
 821         int ret;
 822
 823         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
 824                 return fs_info->tree_root;
 825         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
 826                 return fs_info->extent_root;
 827         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
 828                 return fs_info->chunk_root;
 829         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
 830                 return fs_info->dev_root;
 831
 832         root = radix_tree_lookup(&fs_info->fs_roots_radix,
 833                                  (unsigned long)location->objectid);
 834         if (root)
 835                 return root;
 836
 837         root = btrfs_read_fs_root_no_radix(fs_info, location);
 838         if (IS_ERR(root))
 839                 return root;
 840         ret = radix_tree_insert(&fs_info->fs_roots_radix,
 841                                 (unsigned long)root->root_key.objectid,
 842                                 root);
 843         if (ret) {
 844                 free_extent_buffer(root->node);
 845                 kfree(root);
 846                 return ERR_PTR(ret);
 847         }
 848         ret = btrfs_find_dead_roots(fs_info->tree_root,
 849                                     root->root_key.objectid, root);
 850         BUG_ON(ret);
 851
 852         return root;
 853 }
 854
 855 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
 856                                       struct btrfs_key *location,
 857                                       const char *name, int namelen)
 858 {
 859         struct btrfs_root *root;
 860         int ret;
 861
 862         root = btrfs_read_fs_root_no_name(fs_info, location);
 863         if (!root)
 864                 return NULL;
 865
 866         if (root->in_sysfs)
 867                 return root;
 868
 869         ret = btrfs_set_root_name(root, name, namelen);
 870         if (ret) {
 871                 free_extent_buffer(root->node);
 872                 kfree(root);
 873                 return ERR_PTR(ret);
 874         }
 875
 876         ret = btrfs_sysfs_add_root(root);
 877         if (ret) {
 878                 free_extent_buffer(root->node);
 879                 kfree(root->name);
 880                 kfree(root);
 881                 return ERR_PTR(ret);
 882         }
 883         root->in_sysfs = 1;
 884         return root;
 885 }
 886 #if 0
 887 static int add_hasher(struct btrfs_fs_info *info, char *type) {
 888         struct btrfs_hasher *hasher;
 889
 890         hasher = kmalloc(sizeof(*hasher), GFP_NOFS);
 891         if (!hasher)
 892                 return -ENOMEM;
 893         hasher->hash_tfm = crypto_alloc_hash(type, 0, CRYPTO_ALG_ASYNC);
 894         if (!hasher->hash_tfm) {
 895                 kfree(hasher);
 896                 return -EINVAL;
 897         }
 898         spin_lock(&info->hash_lock);
 899         list_add(&hasher->list, &info->hashers);
 900         spin_unlock(&info->hash_lock);
 901         return 0;
 902 }
 903 #endif
 904
 905 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
 906 {
 907         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
 908         int ret = 0;
 909         struct list_head *cur;
 910         struct btrfs_device *device;
 911         struct backing_dev_info *bdi;
 912
 913         list_for_each(cur, &info->fs_devices->devices) {
 914                 device = list_entry(cur, struct btrfs_device, dev_list);
 915                 if (!device->bdev)
 916                         continue;
 917                 bdi = blk_get_backing_dev_info(device->bdev);
 918                 if (bdi && bdi_congested(bdi, bdi_bits)) {
 919                         ret = 1;
 920                         break;
 921                 }
 922         }
 923         return ret;
 924 }
 925
 926 /*
 927  * this unplugs every device on the box, and it is only used when page
 928  * is null
 929  */
 930 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
 931 {
 932         struct list_head *cur;
 933         struct btrfs_device *device;
 934         struct btrfs_fs_info *info;
 935
 936         info = (struct btrfs_fs_info *)bdi->unplug_io_data;
 937         list_for_each(cur, &info->fs_devices->devices) {
 938                 device = list_entry(cur, struct btrfs_device, dev_list);
 939                 bdi = blk_get_backing_dev_info(device->bdev);
 940                 if (bdi->unplug_io_fn) {
 941                         bdi->unplug_io_fn(bdi, page);
 942                 }
 943         }
 944 }
 945
 946 void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
 947 {
 948         struct inode *inode;
 949         struct extent_map_tree *em_tree;
 950         struct extent_map *em;
 951         struct address_space *mapping;
 952         u64 offset;
 953
 954         /* the generic O_DIRECT read code does this */
 955         if (!page) {
 956                 __unplug_io_fn(bdi, page);
 957                 return;
 958         }
 959
 960         /*
 961          * page->mapping may change at any time.  Get a consistent copy
 962          * and use that for everything below
 963          */
 964         smp_mb();
 965         mapping = page->mapping;
 966         if (!mapping)
 967                 return;
 968
 969         inode = mapping->host;
 970         offset = page_offset(page);
 971
 972         em_tree = &BTRFS_I(inode)->extent_tree;
 973         spin_lock(&em_tree->lock);
 974         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
 975         spin_unlock(&em_tree->lock);
 976         if (!em)
 977                 return;
 978
 979         offset = offset - em->start;
 980         btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
 981                           em->block_start + offset, page);
 982         free_extent_map(em);
 983 }
 984
 985 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
 986 {
 987 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,23)
 988         bdi_init(bdi);
 989 #endif
 990         bdi->ra_pages   = default_backing_dev_info.ra_pages;
 991         bdi->state              = 0;
 992         bdi->capabilities       = default_backing_dev_info.capabilities;
 993         bdi->unplug_io_fn       = btrfs_unplug_io_fn;
 994         bdi->unplug_io_data     = info;
 995         bdi->congested_fn       = btrfs_congested_fn;
 996         bdi->congested_data     = info;
 997         return 0;
 998 }
 999
1000 static int bio_ready_for_csum(struct bio *bio)
1001 {
1002         u64 length = 0;
1003         u64 buf_len = 0;
1004         u64 start = 0;
1005         struct page *page;
1006         struct extent_io_tree *io_tree = NULL;
1007         struct btrfs_fs_info *info = NULL;
1008         struct bio_vec *bvec;
1009         int i;
1010         int ret;
1011
1012         bio_for_each_segment(bvec, bio, i) {
1013                 page = bvec->bv_page;
1014                 if (page->private == EXTENT_PAGE_PRIVATE) {
1015                         length += bvec->bv_len;
1016                         continue;
1017                 }
1018                 if (!page->private) {
1019                         length += bvec->bv_len;
1020                         continue;
1021                 }
1022                 length = bvec->bv_len;
1023                 buf_len = page->private >> 2;
1024                 start = page_offset(page) + bvec->bv_offset;
1025                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1026                 info = BTRFS_I(page->mapping->host)->root->fs_info;
1027         }
1028         /* are we fully contained in this bio? */
1029         if (buf_len <= length)
1030                 return 1;
1031
1032         ret = extent_range_uptodate(io_tree, start + length,
1033                                     start + buf_len - 1);
1034         if (ret == 1)
1035                 return ret;
1036         return ret;
1037 }
1038
1039 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1040 static void btrfs_end_io_csum(void *p)
1041 #else
1042 static void btrfs_end_io_csum(struct work_struct *work)
1043 #endif
1044 {
1045 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1046         struct btrfs_fs_info *fs_info = p;
1047 #else
1048         struct btrfs_fs_info *fs_info = container_of(work,
1049                                                      struct btrfs_fs_info,
1050                                                      end_io_work);
1051 #endif
1052         unsigned long flags;
1053         struct end_io_wq *end_io_wq;
1054         struct bio *bio;
1055         struct list_head *next;
1056         int error;
1057         int was_empty;
1058
1059         while(1) {
1060                 spin_lock_irqsave(&fs_info->end_io_work_lock, flags);
1061                 if (list_empty(&fs_info->end_io_work_list)) {
1062                         spin_unlock_irqrestore(&fs_info->end_io_work_lock,
1063                                                flags);
1064                         return;
1065                 }
1066                 next = fs_info->end_io_work_list.next;
1067                 list_del(next);
1068                 spin_unlock_irqrestore(&fs_info->end_io_work_lock, flags);
1069
1070                 end_io_wq = list_entry(next, struct end_io_wq, list);
1071
1072                 bio = end_io_wq->bio;
1073                 if (end_io_wq->metadata && !bio_ready_for_csum(bio)) {
1074                         spin_lock_irqsave(&fs_info->end_io_work_lock, flags);
1075                         was_empty = list_empty(&fs_info->end_io_work_list);
1076                         list_add_tail(&end_io_wq->list,
1077                                       &fs_info->end_io_work_list);
1078                         spin_unlock_irqrestore(&fs_info->end_io_work_lock,
1079                                                flags);
1080                         if (was_empty)
1081                                 return;
1082                         continue;
1083                 }
1084                 error = end_io_wq->error;
1085                 bio->bi_private = end_io_wq->private;
1086                 bio->bi_end_io = end_io_wq->end_io;
1087                 kfree(end_io_wq);
1088 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1089                 bio_endio(bio, bio->bi_size, error);
1090 #else
1091                 bio_endio(bio, error);
1092 #endif
1093         }
1094 }
1095
1096 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1097 static void btrfs_async_submit_work(void *p)
1098 #else
1099 static void btrfs_async_submit_work(struct work_struct *work)
1100 #endif
1101 {
1102 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1103         struct btrfs_fs_info *fs_info = p;
1104 #else
1105         struct btrfs_fs_info *fs_info = container_of(work,
1106                                                      struct btrfs_fs_info,
1107                                                      async_submit_work);
1108 #endif
1109         struct async_submit_bio *async;
1110         struct list_head *next;
1111
1112         while(1) {
1113                 spin_lock(&fs_info->async_submit_work_lock);
1114                 if (list_empty(&fs_info->async_submit_work_list)) {
1115                         spin_unlock(&fs_info->async_submit_work_lock);
1116                         return;
1117                 }
1118                 next = fs_info->async_submit_work_list.next;
1119                 list_del(next);
1120                 spin_unlock(&fs_info->async_submit_work_lock);
1121
1122                 async = list_entry(next, struct async_submit_bio, list);
1123                 async->submit_bio_hook(async->inode, async->rw, async->bio,
1124                                        async->mirror_num);
1125                 kfree(async);
1126         }
1127 }
1128
1129 struct btrfs_root *open_ctree(struct super_block *sb,
1130                               struct btrfs_fs_devices *fs_devices,
1131                               char *options)
1132 {
1133         u32 sectorsize;
1134         u32 nodesize;
1135         u32 leafsize;
1136         u32 blocksize;
1137         u32 stripesize;
1138         struct buffer_head *bh;
1139         struct btrfs_root *extent_root = kmalloc(sizeof(struct btrfs_root),
1140                                                  GFP_NOFS);
1141         struct btrfs_root *tree_root = kmalloc(sizeof(struct btrfs_root),
1142                                                GFP_NOFS);
1143         struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1144                                                 GFP_NOFS);
1145         struct btrfs_root *chunk_root = kmalloc(sizeof(struct btrfs_root),
1146                                                 GFP_NOFS);
1147         struct btrfs_root *dev_root = kmalloc(sizeof(struct btrfs_root),
1148                                               GFP_NOFS);
1149         int ret;
1150         int err = -EINVAL;
1151         struct btrfs_super_block *disk_super;
1152
1153         if (!extent_root || !tree_root || !fs_info) {
1154                 err = -ENOMEM;
1155                 goto fail;
1156         }
1157         end_io_workqueue = create_workqueue("btrfs-end-io");
1158         BUG_ON(!end_io_workqueue);
1159         async_submit_workqueue = create_workqueue("btrfs-async-submit");
1160
1161         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1162         INIT_LIST_HEAD(&fs_info->trans_list);
1163         INIT_LIST_HEAD(&fs_info->dead_roots);
1164         INIT_LIST_HEAD(&fs_info->hashers);
1165         INIT_LIST_HEAD(&fs_info->end_io_work_list);
1166         INIT_LIST_HEAD(&fs_info->async_submit_work_list);
1167         spin_lock_init(&fs_info->hash_lock);
1168         spin_lock_init(&fs_info->end_io_work_lock);
1169         spin_lock_init(&fs_info->async_submit_work_lock);
1170         spin_lock_init(&fs_info->delalloc_lock);
1171         spin_lock_init(&fs_info->new_trans_lock);
1172
1173         init_completion(&fs_info->kobj_unregister);
1174         fs_info->tree_root = tree_root;
1175         fs_info->extent_root = extent_root;
1176         fs_info->chunk_root = chunk_root;
1177         fs_info->dev_root = dev_root;
1178         fs_info->fs_devices = fs_devices;
1179         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1180         INIT_LIST_HEAD(&fs_info->space_info);
1181         btrfs_mapping_init(&fs_info->mapping_tree);
1182         fs_info->sb = sb;
1183         fs_info->max_extent = (u64)-1;
1184         fs_info->max_inline = 8192 * 1024;
1185         setup_bdi(fs_info, &fs_info->bdi);
1186         fs_info->btree_inode = new_inode(sb);
1187         fs_info->btree_inode->i_ino = 1;
1188         fs_info->btree_inode->i_nlink = 1;
1189
1190         sb->s_blocksize = 4096;
1191         sb->s_blocksize_bits = blksize_bits(4096);
1192
1193         /*
1194          * we set the i_size on the btree inode to the max possible int.
1195          * the real end of the address space is determined by all of
1196          * the devices in the system
1197          */
1198         fs_info->btree_inode->i_size = OFFSET_MAX;
1199         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1200         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1201
1202         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1203                              fs_info->btree_inode->i_mapping,
1204                              GFP_NOFS);
1205         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1206                              GFP_NOFS);
1207
1208         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1209
1210         extent_io_tree_init(&fs_info->free_space_cache,
1211                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1212         extent_io_tree_init(&fs_info->block_group_cache,
1213                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1214         extent_io_tree_init(&fs_info->pinned_extents,
1215                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1216         extent_io_tree_init(&fs_info->pending_del,
1217                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1218         extent_io_tree_init(&fs_info->extent_ins,
1219                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1220         fs_info->do_barriers = 1;
1221
1222 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1223         INIT_WORK(&fs_info->end_io_work, btrfs_end_io_csum, fs_info);
1224         INIT_WORK(&fs_info->async_submit_work, btrfs_async_submit_work,
1225                   fs_info);
1226         INIT_WORK(&fs_info->trans_work, btrfs_transaction_cleaner, fs_info);
1227 #else
1228         INIT_WORK(&fs_info->end_io_work, btrfs_end_io_csum);
1229         INIT_WORK(&fs_info->async_submit_work, btrfs_async_submit_work);
1230         INIT_DELAYED_WORK(&fs_info->trans_work, btrfs_transaction_cleaner);
1231 #endif
1232         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1233         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1234                sizeof(struct btrfs_key));
1235         insert_inode_hash(fs_info->btree_inode);
1236         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1237
1238         mutex_init(&fs_info->trans_mutex);
1239         mutex_init(&fs_info->fs_mutex);
1240
1241 #if 0
1242         ret = add_hasher(fs_info, "crc32c");
1243         if (ret) {
1244                 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
1245                 err = -ENOMEM;
1246                 goto fail_iput;
1247         }
1248 #endif
1249         __setup_root(4096, 4096, 4096, 4096, tree_root,
1250                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1251
1252
1253         bh = __bread(fs_devices->latest_bdev,
1254                      BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
1255         if (!bh)
1256                 goto fail_iput;
1257
1258         memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1259         brelse(bh);
1260
1261         memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1262
1263         disk_super = &fs_info->super_copy;
1264         if (!btrfs_super_root(disk_super))
1265                 goto fail_sb_buffer;
1266
1267         btrfs_parse_options(options, tree_root, NULL);
1268
1269         if (btrfs_super_num_devices(disk_super) > fs_devices->num_devices) {
1270                 printk("Btrfs: wanted %llu devices, but found %llu\n",
1271                        (unsigned long long)btrfs_super_num_devices(disk_super),
1272                        (unsigned long long)fs_devices->num_devices);
1273                 if (btrfs_test_opt(tree_root, DEGRADED))
1274                         printk("continuing in degraded mode\n");
1275                 else {
1276                         goto fail_sb_buffer;
1277                 }
1278         }
1279
1280         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1281
1282         nodesize = btrfs_super_nodesize(disk_super);
1283         leafsize = btrfs_super_leafsize(disk_super);
1284         sectorsize = btrfs_super_sectorsize(disk_super);
1285         stripesize = btrfs_super_stripesize(disk_super);
1286         tree_root->nodesize = nodesize;
1287         tree_root->leafsize = leafsize;
1288         tree_root->sectorsize = sectorsize;
1289         tree_root->stripesize = stripesize;
1290
1291         sb->s_blocksize = sectorsize;
1292         sb->s_blocksize_bits = blksize_bits(sectorsize);
1293
1294         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1295                     sizeof(disk_super->magic))) {
1296                 printk("btrfs: valid FS not found on %s\n", sb->s_id);
1297                 goto fail_sb_buffer;
1298         }
1299
1300         mutex_lock(&fs_info->fs_mutex);
1301
1302         ret = btrfs_read_sys_array(tree_root);
1303         if (ret) {
1304                 printk("btrfs: failed to read the system array on %s\n",
1305                        sb->s_id);
1306                 goto fail_sys_array;
1307         }
1308
1309         blocksize = btrfs_level_size(tree_root,
1310                                      btrfs_super_chunk_root_level(disk_super));
1311
1312         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1313                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1314
1315         chunk_root->node = read_tree_block(chunk_root,
1316                                            btrfs_super_chunk_root(disk_super),
1317                                            blocksize, 0);
1318         BUG_ON(!chunk_root->node);
1319
1320         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1321                  (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1322                  BTRFS_UUID_SIZE);
1323
1324         ret = btrfs_read_chunk_tree(chunk_root);
1325         BUG_ON(ret);
1326
1327         btrfs_close_extra_devices(fs_devices);
1328
1329         blocksize = btrfs_level_size(tree_root,
1330                                      btrfs_super_root_level(disk_super));
1331
1332
1333         tree_root->node = read_tree_block(tree_root,
1334                                           btrfs_super_root(disk_super),
1335                                           blocksize, 0);
1336         if (!tree_root->node)
1337                 goto fail_sb_buffer;
1338
1339
1340         ret = find_and_setup_root(tree_root, fs_info,
1341                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1342         if (ret)
1343                 goto fail_tree_root;
1344         extent_root->track_dirty = 1;
1345
1346         ret = find_and_setup_root(tree_root, fs_info,
1347                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
1348         dev_root->track_dirty = 1;
1349
1350         if (ret)
1351                 goto fail_extent_root;
1352
1353         btrfs_read_block_groups(extent_root);
1354
1355         fs_info->generation = btrfs_super_generation(disk_super) + 1;
1356         fs_info->data_alloc_profile = (u64)-1;
1357         fs_info->metadata_alloc_profile = (u64)-1;
1358         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1359
1360         mutex_unlock(&fs_info->fs_mutex);
1361         return tree_root;
1362
1363 fail_extent_root:
1364         free_extent_buffer(extent_root->node);
1365 fail_tree_root:
1366         free_extent_buffer(tree_root->node);
1367 fail_sys_array:
1368         mutex_unlock(&fs_info->fs_mutex);
1369 fail_sb_buffer:
1370         extent_io_tree_empty_lru(&BTRFS_I(fs_info->btree_inode)->io_tree);
1371 fail_iput:
1372         iput(fs_info->btree_inode);
1373 fail:
1374         btrfs_close_devices(fs_info->fs_devices);
1375         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1376
1377         kfree(extent_root);
1378         kfree(tree_root);
1379 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,23)
1380         bdi_destroy(&fs_info->bdi);
1381 #endif
1382         kfree(fs_info);
1383         return ERR_PTR(err);
1384 }
1385
1386 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1387 {
1388         char b[BDEVNAME_SIZE];
1389
1390         if (uptodate) {
1391                 set_buffer_uptodate(bh);
1392         } else {
1393                 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1394                         printk(KERN_WARNING "lost page write due to "
1395                                         "I/O error on %s\n",
1396                                        bdevname(bh->b_bdev, b));
1397                 }
1398                 /* note, we dont' set_buffer_write_io_error because we have
1399                  * our own ways of dealing with the IO errors
1400                  */
1401                 clear_buffer_uptodate(bh);
1402         }
1403         unlock_buffer(bh);
1404         put_bh(bh);
1405 }
1406
1407 int write_all_supers(struct btrfs_root *root)
1408 {
1409         struct list_head *cur;
1410         struct list_head *head = &root->fs_info->fs_devices->devices;
1411         struct btrfs_device *dev;
1412         struct btrfs_super_block *sb;
1413         struct btrfs_dev_item *dev_item;
1414         struct buffer_head *bh;
1415         int ret;
1416         int do_barriers;
1417         int max_errors;
1418         int total_errors = 0;
1419         u32 crc;
1420         u64 flags;
1421
1422         max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1423         do_barriers = !btrfs_test_opt(root, NOBARRIER);
1424
1425         sb = &root->fs_info->super_for_commit;
1426         dev_item = &sb->dev_item;
1427         list_for_each(cur, head) {
1428                 dev = list_entry(cur, struct btrfs_device, dev_list);
1429                 if (!dev->bdev) {
1430                         total_errors++;
1431                         continue;
1432                 }
1433                 if (!dev->in_fs_metadata)
1434                         continue;
1435
1436                 btrfs_set_stack_device_type(dev_item, dev->type);
1437                 btrfs_set_stack_device_id(dev_item, dev->devid);
1438                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
1439                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
1440                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
1441                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
1442                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
1443                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
1444                 flags = btrfs_super_flags(sb);
1445                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
1446
1447
1448                 crc = ~(u32)0;
1449                 crc = btrfs_csum_data(root, (char *)sb + BTRFS_CSUM_SIZE, crc,
1450                                       BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
1451                 btrfs_csum_final(crc, sb->csum);
1452
1453                 bh = __getblk(dev->bdev, BTRFS_SUPER_INFO_OFFSET / 4096,
1454                               BTRFS_SUPER_INFO_SIZE);
1455
1456                 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
1457                 dev->pending_io = bh;
1458
1459                 get_bh(bh);
1460                 set_buffer_uptodate(bh);
1461                 lock_buffer(bh);
1462                 bh->b_end_io = btrfs_end_buffer_write_sync;
1463
1464                 if (do_barriers && dev->barriers) {
1465                         ret = submit_bh(WRITE_BARRIER, bh);
1466                         if (ret == -EOPNOTSUPP) {
1467                                 printk("btrfs: disabling barriers on dev %s\n",
1468                                        dev->name);
1469                                 set_buffer_uptodate(bh);
1470                                 dev->barriers = 0;
1471                                 get_bh(bh);
1472                                 lock_buffer(bh);
1473                                 ret = submit_bh(WRITE, bh);
1474                         }
1475                 } else {
1476                         ret = submit_bh(WRITE, bh);
1477                 }
1478                 if (ret)
1479                         total_errors++;
1480         }
1481         if (total_errors > max_errors) {
1482                 printk("btrfs: %d errors while writing supers\n", total_errors);
1483                 BUG();
1484         }
1485         total_errors = 0;
1486
1487         list_for_each(cur, head) {
1488                 dev = list_entry(cur, struct btrfs_device, dev_list);
1489                 if (!dev->bdev)
1490                         continue;
1491                 if (!dev->in_fs_metadata)
1492                         continue;
1493
1494                 BUG_ON(!dev->pending_io);
1495                 bh = dev->pending_io;
1496                 wait_on_buffer(bh);
1497                 if (!buffer_uptodate(dev->pending_io)) {
1498                         if (do_barriers && dev->barriers) {
1499                                 printk("btrfs: disabling barriers on dev %s\n",
1500                                        dev->name);
1501                                 set_buffer_uptodate(bh);
1502                                 get_bh(bh);
1503                                 lock_buffer(bh);
1504                                 dev->barriers = 0;
1505                                 ret = submit_bh(WRITE, bh);
1506                                 BUG_ON(ret);
1507                                 wait_on_buffer(bh);
1508                                 if (!buffer_uptodate(bh))
1509                                         total_errors++;
1510                         } else {
1511                                 total_errors++;
1512                         }
1513
1514                 }
1515                 dev->pending_io = NULL;
1516                 brelse(bh);
1517         }
1518         if (total_errors > max_errors) {
1519                 printk("btrfs: %d errors while writing supers\n", total_errors);
1520                 BUG();
1521         }
1522         return 0;
1523 }
1524
1525 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
1526                       *root)
1527 {
1528         int ret;
1529
1530         ret = write_all_supers(root);
1531         return ret;
1532 }
1533
1534 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
1535 {
1536         radix_tree_delete(&fs_info->fs_roots_radix,
1537                           (unsigned long)root->root_key.objectid);
1538         if (root->in_sysfs)
1539                 btrfs_sysfs_del_root(root);
1540         if (root->inode)
1541                 iput(root->inode);
1542         if (root->node)
1543                 free_extent_buffer(root->node);
1544         if (root->commit_root)
1545                 free_extent_buffer(root->commit_root);
1546         if (root->name)
1547                 kfree(root->name);
1548         kfree(root);
1549         return 0;
1550 }
1551
1552 static int del_fs_roots(struct btrfs_fs_info *fs_info)
1553 {
1554         int ret;
1555         struct btrfs_root *gang[8];
1556         int i;
1557
1558         while(1) {
1559                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1560                                              (void **)gang, 0,
1561                                              ARRAY_SIZE(gang));
1562                 if (!ret)
1563                         break;
1564                 for (i = 0; i < ret; i++)
1565                         btrfs_free_fs_root(fs_info, gang[i]);
1566         }
1567         return 0;
1568 }
1569
1570 int close_ctree(struct btrfs_root *root)
1571 {
1572         int ret;
1573         struct btrfs_trans_handle *trans;
1574         struct btrfs_fs_info *fs_info = root->fs_info;
1575
1576         fs_info->closing = 1;
1577         btrfs_transaction_flush_work(root);
1578         mutex_lock(&fs_info->fs_mutex);
1579         btrfs_defrag_dirty_roots(root->fs_info);
1580         trans = btrfs_start_transaction(root, 1);
1581         ret = btrfs_commit_transaction(trans, root);
1582         /* run commit again to  drop the original snapshot */
1583         trans = btrfs_start_transaction(root, 1);
1584         btrfs_commit_transaction(trans, root);
1585         ret = btrfs_write_and_wait_transaction(NULL, root);
1586         BUG_ON(ret);
1587
1588         write_ctree_super(NULL, root);
1589         mutex_unlock(&fs_info->fs_mutex);
1590
1591         btrfs_transaction_flush_work(root);
1592
1593         if (fs_info->delalloc_bytes) {
1594                 printk("btrfs: at unmount delalloc count %Lu\n",
1595                        fs_info->delalloc_bytes);
1596         }
1597         if (fs_info->extent_root->node)
1598                 free_extent_buffer(fs_info->extent_root->node);
1599
1600         if (fs_info->tree_root->node)
1601                 free_extent_buffer(fs_info->tree_root->node);
1602
1603         if (root->fs_info->chunk_root->node);
1604                 free_extent_buffer(root->fs_info->chunk_root->node);
1605
1606         if (root->fs_info->dev_root->node);
1607                 free_extent_buffer(root->fs_info->dev_root->node);
1608
1609         btrfs_free_block_groups(root->fs_info);
1610         del_fs_roots(fs_info);
1611
1612         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1613
1614         extent_io_tree_empty_lru(&fs_info->free_space_cache);
1615         extent_io_tree_empty_lru(&fs_info->block_group_cache);
1616         extent_io_tree_empty_lru(&fs_info->pinned_extents);
1617         extent_io_tree_empty_lru(&fs_info->pending_del);
1618         extent_io_tree_empty_lru(&fs_info->extent_ins);
1619         extent_io_tree_empty_lru(&BTRFS_I(fs_info->btree_inode)->io_tree);
1620
1621         flush_workqueue(async_submit_workqueue);
1622         flush_workqueue(end_io_workqueue);
1623
1624         truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
1625
1626         flush_workqueue(async_submit_workqueue);
1627         destroy_workqueue(async_submit_workqueue);
1628
1629         flush_workqueue(end_io_workqueue);
1630         destroy_workqueue(end_io_workqueue);
1631
1632         iput(fs_info->btree_inode);
1633 #if 0
1634         while(!list_empty(&fs_info->hashers)) {
1635                 struct btrfs_hasher *hasher;
1636                 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
1637                                     hashers);
1638                 list_del(&hasher->hashers);
1639                 crypto_free_hash(&fs_info->hash_tfm);
1640                 kfree(hasher);
1641         }
1642 #endif
1643         btrfs_close_devices(fs_info->fs_devices);
1644         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1645
1646 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,23)
1647         bdi_destroy(&fs_info->bdi);
1648 #endif
1649
1650         kfree(fs_info->extent_root);
1651         kfree(fs_info->tree_root);
1652         kfree(fs_info->chunk_root);
1653         kfree(fs_info->dev_root);
1654         return 0;
1655 }
1656
1657 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
1658 {
1659         int ret;
1660         struct inode *btree_inode = buf->first_page->mapping->host;
1661
1662         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
1663         if (!ret)
1664                 return ret;
1665
1666         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
1667                                     parent_transid);
1668         return !ret;
1669 }
1670
1671 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
1672 {
1673         struct inode *btree_inode = buf->first_page->mapping->host;
1674         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
1675                                           buf);
1676 }
1677
1678 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
1679 {
1680         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1681         u64 transid = btrfs_header_generation(buf);
1682         struct inode *btree_inode = root->fs_info->btree_inode;
1683
1684         if (transid != root->fs_info->generation) {
1685                 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
1686                         (unsigned long long)buf->start,
1687                         transid, root->fs_info->generation);
1688                 WARN_ON(1);
1689         }
1690         set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
1691 }
1692
1693 void btrfs_throttle(struct btrfs_root *root)
1694 {
1695         struct backing_dev_info *bdi;
1696
1697         bdi = &root->fs_info->bdi;
1698         if (root->fs_info->throttles && bdi_write_congested(bdi)) {
1699 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,18)
1700                 congestion_wait(WRITE, HZ/20);
1701 #else
1702                 blk_congestion_wait(WRITE, HZ/20);
1703 #endif
1704         }
1705 }
1706
1707 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
1708 {
1709         /*
1710          * looks as though older kernels can get into trouble with
1711          * this code, they end up stuck in balance_dirty_pages forever
1712          */
1713 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,23)
1714         struct extent_io_tree *tree;
1715         u64 num_dirty;
1716         u64 start = 0;
1717         unsigned long thresh = 16 * 1024 * 1024;
1718         tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
1719
1720         if (current_is_pdflush())
1721                 return;
1722
1723         num_dirty = count_range_bits(tree, &start, (u64)-1,
1724                                      thresh, EXTENT_DIRTY);
1725         if (num_dirty > thresh) {
1726                 balance_dirty_pages_ratelimited_nr(
1727                                    root->fs_info->btree_inode->i_mapping, 1);
1728         }
1729 #else
1730         return;
1731 #endif
1732 }
1733
1734 void btrfs_set_buffer_defrag(struct extent_buffer *buf)
1735 {
1736         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1737         struct inode *btree_inode = root->fs_info->btree_inode;
1738         set_extent_bits(&BTRFS_I(btree_inode)->io_tree, buf->start,
1739                         buf->start + buf->len - 1, EXTENT_DEFRAG, GFP_NOFS);
1740 }
1741
1742 void btrfs_set_buffer_defrag_done(struct extent_buffer *buf)
1743 {
1744         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1745         struct inode *btree_inode = root->fs_info->btree_inode;
1746         set_extent_bits(&BTRFS_I(btree_inode)->io_tree, buf->start,
1747                         buf->start + buf->len - 1, EXTENT_DEFRAG_DONE,
1748                         GFP_NOFS);
1749 }
1750
1751 int btrfs_buffer_defrag(struct extent_buffer *buf)
1752 {
1753         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1754         struct inode *btree_inode = root->fs_info->btree_inode;
1755         return test_range_bit(&BTRFS_I(btree_inode)->io_tree,
1756                      buf->start, buf->start + buf->len - 1, EXTENT_DEFRAG, 0);
1757 }
1758
1759 int btrfs_buffer_defrag_done(struct extent_buffer *buf)
1760 {
1761         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1762         struct inode *btree_inode = root->fs_info->btree_inode;
1763         return test_range_bit(&BTRFS_I(btree_inode)->io_tree,
1764                      buf->start, buf->start + buf->len - 1,
1765                      EXTENT_DEFRAG_DONE, 0);
1766 }
1767
1768 int btrfs_clear_buffer_defrag_done(struct extent_buffer *buf)
1769 {
1770         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1771         struct inode *btree_inode = root->fs_info->btree_inode;
1772         return clear_extent_bits(&BTRFS_I(btree_inode)->io_tree,
1773                      buf->start, buf->start + buf->len - 1,
1774                      EXTENT_DEFRAG_DONE, GFP_NOFS);
1775 }
1776
1777 int btrfs_clear_buffer_defrag(struct extent_buffer *buf)
1778 {
1779         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1780         struct inode *btree_inode = root->fs_info->btree_inode;
1781         return clear_extent_bits(&BTRFS_I(btree_inode)->io_tree,
1782                      buf->start, buf->start + buf->len - 1,
1783                      EXTENT_DEFRAG, GFP_NOFS);
1784 }
1785
1786 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
1787 {
1788         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1789         int ret;
1790         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1791         if (ret == 0) {
1792                 buf->flags |= EXTENT_UPTODATE;
1793         }
1794         return ret;
1795 }
1796
1797 static struct extent_io_ops btree_extent_io_ops = {
1798         .writepage_io_hook = btree_writepage_io_hook,
1799         .readpage_end_io_hook = btree_readpage_end_io_hook,
1800         .submit_bio_hook = btree_submit_bio_hook,
1801         /* note we're sharing with inode.c for the merge bio hook */
1802         .merge_bio_hook = btrfs_merge_bio_hook,
1803 };