git.oblomov.eu Git - 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         atomic_inc(&fs_info->nr_async_submits);
 457         spin_unlock(&fs_info->async_submit_work_lock);
 458
 459         queue_work(async_submit_workqueue, &fs_info->async_submit_work);
 460         return 0;
 461 }
 462
 463 static int __btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 464                                  int mirror_num)
 465 {
 466         struct btrfs_root *root = BTRFS_I(inode)->root;
 467         u64 offset;
 468         int ret;
 469
 470         offset = bio->bi_sector << 9;
 471
 472         if (rw & (1 << BIO_RW)) {
 473                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num);
 474         }
 475
 476         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 1);
 477         BUG_ON(ret);
 478
 479         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num);
 480 }
 481
 482 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 483                                  int mirror_num)
 484 {
 485         if (!(rw & (1 << BIO_RW))) {
 486                 return __btree_submit_bio_hook(inode, rw, bio, mirror_num);
 487         }
 488         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 489                                    inode, rw, bio, mirror_num,
 490                                    __btree_submit_bio_hook);
 491 }
 492
 493 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 494 {
 495         struct extent_io_tree *tree;
 496         tree = &BTRFS_I(page->mapping->host)->io_tree;
 497         return extent_write_full_page(tree, page, btree_get_extent, wbc);
 498 }
 499
 500 static int btree_writepages(struct address_space *mapping,
 501                             struct writeback_control *wbc)
 502 {
 503         struct extent_io_tree *tree;
 504         tree = &BTRFS_I(mapping->host)->io_tree;
 505         if (wbc->sync_mode == WB_SYNC_NONE) {
 506                 u64 num_dirty;
 507                 u64 start = 0;
 508                 unsigned long thresh = 96 * 1024 * 1024;
 509
 510                 if (wbc->for_kupdate)
 511                         return 0;
 512
 513                 if (current_is_pdflush()) {
 514                         thresh = 96 * 1024 * 1024;
 515                 } else {
 516                         thresh = 8 * 1024 * 1024;
 517                 }
 518                 num_dirty = count_range_bits(tree, &start, (u64)-1,
 519                                              thresh, EXTENT_DIRTY);
 520                 if (num_dirty < thresh) {
 521                         return 0;
 522                 }
 523         }
 524         return extent_writepages(tree, mapping, btree_get_extent, wbc);
 525 }
 526
 527 int btree_readpage(struct file *file, struct page *page)
 528 {
 529         struct extent_io_tree *tree;
 530         tree = &BTRFS_I(page->mapping->host)->io_tree;
 531         return extent_read_full_page(tree, page, btree_get_extent);
 532 }
 533
 534 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 535 {
 536         struct extent_io_tree *tree;
 537         struct extent_map_tree *map;
 538         int ret;
 539
 540         if (page_count(page) > 3) {
 541                 /* once for page->private, once for the caller, once
 542                  * once for the page cache
 543                  */
 544                 return 0;
 545         }
 546         tree = &BTRFS_I(page->mapping->host)->io_tree;
 547         map = &BTRFS_I(page->mapping->host)->extent_tree;
 548         ret = try_release_extent_state(map, tree, page, gfp_flags);
 549         if (ret == 1) {
 550                 invalidate_extent_lru(tree, page_offset(page), PAGE_CACHE_SIZE);
 551                 ClearPagePrivate(page);
 552                 set_page_private(page, 0);
 553                 page_cache_release(page);
 554         }
 555         return ret;
 556 }
 557
 558 static void btree_invalidatepage(struct page *page, unsigned long offset)
 559 {
 560         struct extent_io_tree *tree;
 561         tree = &BTRFS_I(page->mapping->host)->io_tree;
 562         extent_invalidatepage(tree, page, offset);
 563         btree_releasepage(page, GFP_NOFS);
 564         if (PagePrivate(page)) {
 565                 invalidate_extent_lru(tree, page_offset(page), PAGE_CACHE_SIZE);
 566                 ClearPagePrivate(page);
 567                 set_page_private(page, 0);
 568                 page_cache_release(page);
 569         }
 570 }
 571
 572 #if 0
 573 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 574 {
 575         struct buffer_head *bh;
 576         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 577         struct buffer_head *head;
 578         if (!page_has_buffers(page)) {
 579                 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
 580                                         (1 << BH_Dirty)|(1 << BH_Uptodate));
 581         }
 582         head = page_buffers(page);
 583         bh = head;
 584         do {
 585                 if (buffer_dirty(bh))
 586                         csum_tree_block(root, bh, 0);
 587                 bh = bh->b_this_page;
 588         } while (bh != head);
 589         return block_write_full_page(page, btree_get_block, wbc);
 590 }
 591 #endif
 592
 593 static struct address_space_operations btree_aops = {
 594         .readpage       = btree_readpage,
 595         .writepage      = btree_writepage,
 596         .writepages     = btree_writepages,
 597         .releasepage    = btree_releasepage,
 598         .invalidatepage = btree_invalidatepage,
 599         .sync_page      = block_sync_page,
 600 };
 601
 602 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
 603                          u64 parent_transid)
 604 {
 605         struct extent_buffer *buf = NULL;
 606         struct inode *btree_inode = root->fs_info->btree_inode;
 607         int ret = 0;
 608
 609         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 610         if (!buf)
 611                 return 0;
 612         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
 613                                  buf, 0, 0, btree_get_extent, 0);
 614         free_extent_buffer(buf);
 615         return ret;
 616 }
 617
 618 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
 619                                             u64 bytenr, u32 blocksize)
 620 {
 621         struct inode *btree_inode = root->fs_info->btree_inode;
 622         struct extent_buffer *eb;
 623         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 624                                 bytenr, blocksize, GFP_NOFS);
 625         return eb;
 626 }
 627
 628 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
 629                                                  u64 bytenr, u32 blocksize)
 630 {
 631         struct inode *btree_inode = root->fs_info->btree_inode;
 632         struct extent_buffer *eb;
 633
 634         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 635                                  bytenr, blocksize, NULL, GFP_NOFS);
 636         return eb;
 637 }
 638
 639
 640 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
 641                                       u32 blocksize, u64 parent_transid)
 642 {
 643         struct extent_buffer *buf = NULL;
 644         struct inode *btree_inode = root->fs_info->btree_inode;
 645         struct extent_io_tree *io_tree;
 646         int ret;
 647
 648         io_tree = &BTRFS_I(btree_inode)->io_tree;
 649
 650         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 651         if (!buf)
 652                 return NULL;
 653
 654         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
 655
 656         if (ret == 0) {
 657                 buf->flags |= EXTENT_UPTODATE;
 658         }
 659         return buf;
 660
 661 }
 662
 663 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 664                      struct extent_buffer *buf)
 665 {
 666         struct inode *btree_inode = root->fs_info->btree_inode;
 667         if (btrfs_header_generation(buf) ==
 668             root->fs_info->running_transaction->transid)
 669                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
 670                                           buf);
 671         return 0;
 672 }
 673
 674 int wait_on_tree_block_writeback(struct btrfs_root *root,
 675                                  struct extent_buffer *buf)
 676 {
 677         struct inode *btree_inode = root->fs_info->btree_inode;
 678         wait_on_extent_buffer_writeback(&BTRFS_I(btree_inode)->io_tree,
 679                                         buf);
 680         return 0;
 681 }
 682
 683 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
 684                         u32 stripesize, struct btrfs_root *root,
 685                         struct btrfs_fs_info *fs_info,
 686                         u64 objectid)
 687 {
 688         root->node = NULL;
 689         root->inode = NULL;
 690         root->commit_root = NULL;
 691         root->sectorsize = sectorsize;
 692         root->nodesize = nodesize;
 693         root->leafsize = leafsize;
 694         root->stripesize = stripesize;
 695         root->ref_cows = 0;
 696         root->track_dirty = 0;
 697
 698         root->fs_info = fs_info;
 699         root->objectid = objectid;
 700         root->last_trans = 0;
 701         root->highest_inode = 0;
 702         root->last_inode_alloc = 0;
 703         root->name = NULL;
 704         root->in_sysfs = 0;
 705
 706         INIT_LIST_HEAD(&root->dirty_list);
 707         memset(&root->root_key, 0, sizeof(root->root_key));
 708         memset(&root->root_item, 0, sizeof(root->root_item));
 709         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
 710         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
 711         init_completion(&root->kobj_unregister);
 712         root->defrag_running = 0;
 713         root->defrag_level = 0;
 714         root->root_key.objectid = objectid;
 715         return 0;
 716 }
 717
 718 static int find_and_setup_root(struct btrfs_root *tree_root,
 719                                struct btrfs_fs_info *fs_info,
 720                                u64 objectid,
 721                                struct btrfs_root *root)
 722 {
 723         int ret;
 724         u32 blocksize;
 725
 726         __setup_root(tree_root->nodesize, tree_root->leafsize,
 727                      tree_root->sectorsize, tree_root->stripesize,
 728                      root, fs_info, objectid);
 729         ret = btrfs_find_last_root(tree_root, objectid,
 730                                    &root->root_item, &root->root_key);
 731         BUG_ON(ret);
 732
 733         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 734         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 735                                      blocksize, 0);
 736         BUG_ON(!root->node);
 737         return 0;
 738 }
 739
 740 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_fs_info *fs_info,
 741                                                struct btrfs_key *location)
 742 {
 743         struct btrfs_root *root;
 744         struct btrfs_root *tree_root = fs_info->tree_root;
 745         struct btrfs_path *path;
 746         struct extent_buffer *l;
 747         u64 highest_inode;
 748         u32 blocksize;
 749         int ret = 0;
 750
 751         root = kzalloc(sizeof(*root), GFP_NOFS);
 752         if (!root)
 753                 return ERR_PTR(-ENOMEM);
 754         if (location->offset == (u64)-1) {
 755                 ret = find_and_setup_root(tree_root, fs_info,
 756                                           location->objectid, root);
 757                 if (ret) {
 758                         kfree(root);
 759                         return ERR_PTR(ret);
 760                 }
 761                 goto insert;
 762         }
 763
 764         __setup_root(tree_root->nodesize, tree_root->leafsize,
 765                      tree_root->sectorsize, tree_root->stripesize,
 766                      root, fs_info, location->objectid);
 767
 768         path = btrfs_alloc_path();
 769         BUG_ON(!path);
 770         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
 771         if (ret != 0) {
 772                 if (ret > 0)
 773                         ret = -ENOENT;
 774                 goto out;
 775         }
 776         l = path->nodes[0];
 777         read_extent_buffer(l, &root->root_item,
 778                btrfs_item_ptr_offset(l, path->slots[0]),
 779                sizeof(root->root_item));
 780         memcpy(&root->root_key, location, sizeof(*location));
 781         ret = 0;
 782 out:
 783         btrfs_release_path(root, path);
 784         btrfs_free_path(path);
 785         if (ret) {
 786                 kfree(root);
 787                 return ERR_PTR(ret);
 788         }
 789         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 790         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 791                                      blocksize, 0);
 792         BUG_ON(!root->node);
 793 insert:
 794         root->ref_cows = 1;
 795         ret = btrfs_find_highest_inode(root, &highest_inode);
 796         if (ret == 0) {
 797                 root->highest_inode = highest_inode;
 798                 root->last_inode_alloc = highest_inode;
 799         }
 800         return root;
 801 }
 802
 803 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
 804                                         u64 root_objectid)
 805 {
 806         struct btrfs_root *root;
 807
 808         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
 809                 return fs_info->tree_root;
 810         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
 811                 return fs_info->extent_root;
 812
 813         root = radix_tree_lookup(&fs_info->fs_roots_radix,
 814                                  (unsigned long)root_objectid);
 815         return root;
 816 }
 817
 818 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
 819                                               struct btrfs_key *location)
 820 {
 821         struct btrfs_root *root;
 822         int ret;
 823
 824         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
 825                 return fs_info->tree_root;
 826         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
 827                 return fs_info->extent_root;
 828         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
 829                 return fs_info->chunk_root;
 830         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
 831                 return fs_info->dev_root;
 832
 833         root = radix_tree_lookup(&fs_info->fs_roots_radix,
 834                                  (unsigned long)location->objectid);
 835         if (root)
 836                 return root;
 837
 838         root = btrfs_read_fs_root_no_radix(fs_info, location);
 839         if (IS_ERR(root))
 840                 return root;
 841         ret = radix_tree_insert(&fs_info->fs_roots_radix,
 842                                 (unsigned long)root->root_key.objectid,
 843                                 root);
 844         if (ret) {
 845                 free_extent_buffer(root->node);
 846                 kfree(root);
 847                 return ERR_PTR(ret);
 848         }
 849         ret = btrfs_find_dead_roots(fs_info->tree_root,
 850                                     root->root_key.objectid, root);
 851         BUG_ON(ret);
 852
 853         return root;
 854 }
 855
 856 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
 857                                       struct btrfs_key *location,
 858                                       const char *name, int namelen)
 859 {
 860         struct btrfs_root *root;
 861         int ret;
 862
 863         root = btrfs_read_fs_root_no_name(fs_info, location);
 864         if (!root)
 865                 return NULL;
 866
 867         if (root->in_sysfs)
 868                 return root;
 869
 870         ret = btrfs_set_root_name(root, name, namelen);
 871         if (ret) {
 872                 free_extent_buffer(root->node);
 873                 kfree(root);
 874                 return ERR_PTR(ret);
 875         }
 876
 877         ret = btrfs_sysfs_add_root(root);
 878         if (ret) {
 879                 free_extent_buffer(root->node);
 880                 kfree(root->name);
 881                 kfree(root);
 882                 return ERR_PTR(ret);
 883         }
 884         root->in_sysfs = 1;
 885         return root;
 886 }
 887 #if 0
 888 static int add_hasher(struct btrfs_fs_info *info, char *type) {
 889         struct btrfs_hasher *hasher;
 890
 891         hasher = kmalloc(sizeof(*hasher), GFP_NOFS);
 892         if (!hasher)
 893                 return -ENOMEM;
 894         hasher->hash_tfm = crypto_alloc_hash(type, 0, CRYPTO_ALG_ASYNC);
 895         if (!hasher->hash_tfm) {
 896                 kfree(hasher);
 897                 return -EINVAL;
 898         }
 899         spin_lock(&info->hash_lock);
 900         list_add(&hasher->list, &info->hashers);
 901         spin_unlock(&info->hash_lock);
 902         return 0;
 903 }
 904 #endif
 905
 906 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
 907 {
 908         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
 909         int ret = 0;
 910         int limit = 256 * info->fs_devices->open_devices;
 911         struct list_head *cur;
 912         struct btrfs_device *device;
 913         struct backing_dev_info *bdi;
 914
 915         if ((bdi_bits & (1 << BDI_write_congested)) &&
 916             atomic_read(&info->nr_async_submits) > limit) {
 917                 return 1;
 918         }
 919
 920         list_for_each(cur, &info->fs_devices->devices) {
 921                 device = list_entry(cur, struct btrfs_device, dev_list);
 922                 if (!device->bdev)
 923                         continue;
 924                 bdi = blk_get_backing_dev_info(device->bdev);
 925                 if (bdi && bdi_congested(bdi, bdi_bits)) {
 926                         ret = 1;
 927                         break;
 928                 }
 929         }
 930         return ret;
 931 }
 932
 933 /*
 934  * this unplugs every device on the box, and it is only used when page
 935  * is null
 936  */
 937 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
 938 {
 939         struct list_head *cur;
 940         struct btrfs_device *device;
 941         struct btrfs_fs_info *info;
 942
 943         info = (struct btrfs_fs_info *)bdi->unplug_io_data;
 944         list_for_each(cur, &info->fs_devices->devices) {
 945                 device = list_entry(cur, struct btrfs_device, dev_list);
 946                 bdi = blk_get_backing_dev_info(device->bdev);
 947                 if (bdi->unplug_io_fn) {
 948                         bdi->unplug_io_fn(bdi, page);
 949                 }
 950         }
 951 }
 952
 953 void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
 954 {
 955         struct inode *inode;
 956         struct extent_map_tree *em_tree;
 957         struct extent_map *em;
 958         struct address_space *mapping;
 959         u64 offset;
 960
 961         /* the generic O_DIRECT read code does this */
 962         if (!page) {
 963                 __unplug_io_fn(bdi, page);
 964                 return;
 965         }
 966
 967         /*
 968          * page->mapping may change at any time.  Get a consistent copy
 969          * and use that for everything below
 970          */
 971         smp_mb();
 972         mapping = page->mapping;
 973         if (!mapping)
 974                 return;
 975
 976         inode = mapping->host;
 977         offset = page_offset(page);
 978
 979         em_tree = &BTRFS_I(inode)->extent_tree;
 980         spin_lock(&em_tree->lock);
 981         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
 982         spin_unlock(&em_tree->lock);
 983         if (!em)
 984                 return;
 985
 986         offset = offset - em->start;
 987         btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
 988                           em->block_start + offset, page);
 989         free_extent_map(em);
 990 }
 991
 992 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
 993 {
 994 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,23)
 995         bdi_init(bdi);
 996 #endif
 997         bdi->ra_pages   = default_backing_dev_info.ra_pages;
 998         bdi->state              = 0;
 999         bdi->capabilities       = default_backing_dev_info.capabilities;
1000         bdi->unplug_io_fn       = btrfs_unplug_io_fn;
1001         bdi->unplug_io_data     = info;
1002         bdi->congested_fn       = btrfs_congested_fn;
1003         bdi->congested_data     = info;
1004         return 0;
1005 }
1006
1007 static int bio_ready_for_csum(struct bio *bio)
1008 {
1009         u64 length = 0;
1010         u64 buf_len = 0;
1011         u64 start = 0;
1012         struct page *page;
1013         struct extent_io_tree *io_tree = NULL;
1014         struct btrfs_fs_info *info = NULL;
1015         struct bio_vec *bvec;
1016         int i;
1017         int ret;
1018
1019         bio_for_each_segment(bvec, bio, i) {
1020                 page = bvec->bv_page;
1021                 if (page->private == EXTENT_PAGE_PRIVATE) {
1022                         length += bvec->bv_len;
1023                         continue;
1024                 }
1025                 if (!page->private) {
1026                         length += bvec->bv_len;
1027                         continue;
1028                 }
1029                 length = bvec->bv_len;
1030                 buf_len = page->private >> 2;
1031                 start = page_offset(page) + bvec->bv_offset;
1032                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1033                 info = BTRFS_I(page->mapping->host)->root->fs_info;
1034         }
1035         /* are we fully contained in this bio? */
1036         if (buf_len <= length)
1037                 return 1;
1038
1039         ret = extent_range_uptodate(io_tree, start + length,
1040                                     start + buf_len - 1);
1041         if (ret == 1)
1042                 return ret;
1043         return ret;
1044 }
1045
1046 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1047 static void btrfs_end_io_csum(void *p)
1048 #else
1049 static void btrfs_end_io_csum(struct work_struct *work)
1050 #endif
1051 {
1052 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1053         struct btrfs_fs_info *fs_info = p;
1054 #else
1055         struct btrfs_fs_info *fs_info = container_of(work,
1056                                                      struct btrfs_fs_info,
1057                                                      end_io_work);
1058 #endif
1059         unsigned long flags;
1060         struct end_io_wq *end_io_wq;
1061         struct bio *bio;
1062         struct list_head *next;
1063         int error;
1064         int was_empty;
1065
1066         while(1) {
1067                 spin_lock_irqsave(&fs_info->end_io_work_lock, flags);
1068                 if (list_empty(&fs_info->end_io_work_list)) {
1069                         spin_unlock_irqrestore(&fs_info->end_io_work_lock,
1070                                                flags);
1071                         return;
1072                 }
1073                 next = fs_info->end_io_work_list.next;
1074                 list_del(next);
1075                 spin_unlock_irqrestore(&fs_info->end_io_work_lock, flags);
1076
1077                 end_io_wq = list_entry(next, struct end_io_wq, list);
1078
1079                 bio = end_io_wq->bio;
1080                 if (end_io_wq->metadata && !bio_ready_for_csum(bio)) {
1081                         spin_lock_irqsave(&fs_info->end_io_work_lock, flags);
1082                         was_empty = list_empty(&fs_info->end_io_work_list);
1083                         list_add_tail(&end_io_wq->list,
1084                                       &fs_info->end_io_work_list);
1085                         spin_unlock_irqrestore(&fs_info->end_io_work_lock,
1086                                                flags);
1087                         if (was_empty)
1088                                 return;
1089                         continue;
1090                 }
1091                 error = end_io_wq->error;
1092                 bio->bi_private = end_io_wq->private;
1093                 bio->bi_end_io = end_io_wq->end_io;
1094                 kfree(end_io_wq);
1095 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1096                 bio_endio(bio, bio->bi_size, error);
1097 #else
1098                 bio_endio(bio, error);
1099 #endif
1100         }
1101 }
1102
1103 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1104 static void btrfs_async_submit_work(void *p)
1105 #else
1106 static void btrfs_async_submit_work(struct work_struct *work)
1107 #endif
1108 {
1109 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1110         struct btrfs_fs_info *fs_info = p;
1111 #else
1112         struct btrfs_fs_info *fs_info = container_of(work,
1113                                                      struct btrfs_fs_info,
1114                                                      async_submit_work);
1115 #endif
1116         struct async_submit_bio *async;
1117         struct list_head *next;
1118
1119         while(1) {
1120                 spin_lock(&fs_info->async_submit_work_lock);
1121                 if (list_empty(&fs_info->async_submit_work_list)) {
1122                         spin_unlock(&fs_info->async_submit_work_lock);
1123                         return;
1124                 }
1125                 next = fs_info->async_submit_work_list.next;
1126                 list_del(next);
1127                 atomic_dec(&fs_info->nr_async_submits);
1128                 spin_unlock(&fs_info->async_submit_work_lock);
1129
1130                 async = list_entry(next, struct async_submit_bio, list);
1131                 async->submit_bio_hook(async->inode, async->rw, async->bio,
1132                                        async->mirror_num);
1133                 kfree(async);
1134         }
1135 }
1136
1137 struct btrfs_root *open_ctree(struct super_block *sb,
1138                               struct btrfs_fs_devices *fs_devices,
1139                               char *options)
1140 {
1141         u32 sectorsize;
1142         u32 nodesize;
1143         u32 leafsize;
1144         u32 blocksize;
1145         u32 stripesize;
1146         struct buffer_head *bh;
1147         struct btrfs_root *extent_root = kmalloc(sizeof(struct btrfs_root),
1148                                                  GFP_NOFS);
1149         struct btrfs_root *tree_root = kmalloc(sizeof(struct btrfs_root),
1150                                                GFP_NOFS);
1151         struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1152                                                 GFP_NOFS);
1153         struct btrfs_root *chunk_root = kmalloc(sizeof(struct btrfs_root),
1154                                                 GFP_NOFS);
1155         struct btrfs_root *dev_root = kmalloc(sizeof(struct btrfs_root),
1156                                               GFP_NOFS);
1157         int ret;
1158         int err = -EINVAL;
1159         struct btrfs_super_block *disk_super;
1160
1161         if (!extent_root || !tree_root || !fs_info) {
1162                 err = -ENOMEM;
1163                 goto fail;
1164         }
1165         end_io_workqueue = create_workqueue("btrfs-end-io");
1166         BUG_ON(!end_io_workqueue);
1167         async_submit_workqueue = create_workqueue("btrfs-async-submit");
1168
1169         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1170         INIT_LIST_HEAD(&fs_info->trans_list);
1171         INIT_LIST_HEAD(&fs_info->dead_roots);
1172         INIT_LIST_HEAD(&fs_info->hashers);
1173         INIT_LIST_HEAD(&fs_info->end_io_work_list);
1174         INIT_LIST_HEAD(&fs_info->async_submit_work_list);
1175         spin_lock_init(&fs_info->hash_lock);
1176         spin_lock_init(&fs_info->end_io_work_lock);
1177         spin_lock_init(&fs_info->async_submit_work_lock);
1178         spin_lock_init(&fs_info->delalloc_lock);
1179         spin_lock_init(&fs_info->new_trans_lock);
1180
1181         init_completion(&fs_info->kobj_unregister);
1182         fs_info->tree_root = tree_root;
1183         fs_info->extent_root = extent_root;
1184         fs_info->chunk_root = chunk_root;
1185         fs_info->dev_root = dev_root;
1186         fs_info->fs_devices = fs_devices;
1187         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1188         INIT_LIST_HEAD(&fs_info->space_info);
1189         btrfs_mapping_init(&fs_info->mapping_tree);
1190         atomic_set(&fs_info->nr_async_submits, 0);
1191         fs_info->sb = sb;
1192         fs_info->max_extent = (u64)-1;
1193         fs_info->max_inline = 8192 * 1024;
1194         setup_bdi(fs_info, &fs_info->bdi);
1195         fs_info->btree_inode = new_inode(sb);
1196         fs_info->btree_inode->i_ino = 1;
1197         fs_info->btree_inode->i_nlink = 1;
1198
1199         sb->s_blocksize = 4096;
1200         sb->s_blocksize_bits = blksize_bits(4096);
1201
1202         /*
1203          * we set the i_size on the btree inode to the max possible int.
1204          * the real end of the address space is determined by all of
1205          * the devices in the system
1206          */
1207         fs_info->btree_inode->i_size = OFFSET_MAX;
1208         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1209         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1210
1211         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1212                              fs_info->btree_inode->i_mapping,
1213                              GFP_NOFS);
1214         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1215                              GFP_NOFS);
1216
1217         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1218
1219         extent_io_tree_init(&fs_info->free_space_cache,
1220                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1221         extent_io_tree_init(&fs_info->block_group_cache,
1222                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1223         extent_io_tree_init(&fs_info->pinned_extents,
1224                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1225         extent_io_tree_init(&fs_info->pending_del,
1226                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1227         extent_io_tree_init(&fs_info->extent_ins,
1228                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1229         fs_info->do_barriers = 1;
1230
1231 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1232         INIT_WORK(&fs_info->end_io_work, btrfs_end_io_csum, fs_info);
1233         INIT_WORK(&fs_info->async_submit_work, btrfs_async_submit_work,
1234                   fs_info);
1235         INIT_WORK(&fs_info->trans_work, btrfs_transaction_cleaner, fs_info);
1236 #else
1237         INIT_WORK(&fs_info->end_io_work, btrfs_end_io_csum);
1238         INIT_WORK(&fs_info->async_submit_work, btrfs_async_submit_work);
1239         INIT_DELAYED_WORK(&fs_info->trans_work, btrfs_transaction_cleaner);
1240 #endif
1241         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1242         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1243                sizeof(struct btrfs_key));
1244         insert_inode_hash(fs_info->btree_inode);
1245         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1246
1247         mutex_init(&fs_info->trans_mutex);
1248         mutex_init(&fs_info->fs_mutex);
1249
1250 #if 0
1251         ret = add_hasher(fs_info, "crc32c");
1252         if (ret) {
1253                 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
1254                 err = -ENOMEM;
1255                 goto fail_iput;
1256         }
1257 #endif
1258         __setup_root(4096, 4096, 4096, 4096, tree_root,
1259                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1260
1261
1262         bh = __bread(fs_devices->latest_bdev,
1263                      BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
1264         if (!bh)
1265                 goto fail_iput;
1266
1267         memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1268         brelse(bh);
1269
1270         memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1271
1272         disk_super = &fs_info->super_copy;
1273         if (!btrfs_super_root(disk_super))
1274                 goto fail_sb_buffer;
1275
1276         btrfs_parse_options(options, tree_root, NULL);
1277
1278         if (btrfs_super_num_devices(disk_super) > fs_devices->open_devices) {
1279                 printk("Btrfs: wanted %llu devices, but found %llu\n",
1280                        (unsigned long long)btrfs_super_num_devices(disk_super),
1281                        (unsigned long long)fs_devices->open_devices);
1282                 if (btrfs_test_opt(tree_root, DEGRADED))
1283                         printk("continuing in degraded mode\n");
1284                 else {
1285                         goto fail_sb_buffer;
1286                 }
1287         }
1288
1289         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1290
1291         nodesize = btrfs_super_nodesize(disk_super);
1292         leafsize = btrfs_super_leafsize(disk_super);
1293         sectorsize = btrfs_super_sectorsize(disk_super);
1294         stripesize = btrfs_super_stripesize(disk_super);
1295         tree_root->nodesize = nodesize;
1296         tree_root->leafsize = leafsize;
1297         tree_root->sectorsize = sectorsize;
1298         tree_root->stripesize = stripesize;
1299
1300         sb->s_blocksize = sectorsize;
1301         sb->s_blocksize_bits = blksize_bits(sectorsize);
1302
1303         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1304                     sizeof(disk_super->magic))) {
1305                 printk("btrfs: valid FS not found on %s\n", sb->s_id);
1306                 goto fail_sb_buffer;
1307         }
1308
1309         mutex_lock(&fs_info->fs_mutex);
1310
1311         ret = btrfs_read_sys_array(tree_root);
1312         if (ret) {
1313                 printk("btrfs: failed to read the system array on %s\n",
1314                        sb->s_id);
1315                 goto fail_sys_array;
1316         }
1317
1318         blocksize = btrfs_level_size(tree_root,
1319                                      btrfs_super_chunk_root_level(disk_super));
1320
1321         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1322                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1323
1324         chunk_root->node = read_tree_block(chunk_root,
1325                                            btrfs_super_chunk_root(disk_super),
1326                                            blocksize, 0);
1327         BUG_ON(!chunk_root->node);
1328
1329         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1330                  (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1331                  BTRFS_UUID_SIZE);
1332
1333         ret = btrfs_read_chunk_tree(chunk_root);
1334         BUG_ON(ret);
1335
1336         btrfs_close_extra_devices(fs_devices);
1337
1338         blocksize = btrfs_level_size(tree_root,
1339                                      btrfs_super_root_level(disk_super));
1340
1341
1342         tree_root->node = read_tree_block(tree_root,
1343                                           btrfs_super_root(disk_super),
1344                                           blocksize, 0);
1345         if (!tree_root->node)
1346                 goto fail_sb_buffer;
1347
1348
1349         ret = find_and_setup_root(tree_root, fs_info,
1350                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1351         if (ret)
1352                 goto fail_tree_root;
1353         extent_root->track_dirty = 1;
1354
1355         ret = find_and_setup_root(tree_root, fs_info,
1356                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
1357         dev_root->track_dirty = 1;
1358
1359         if (ret)
1360                 goto fail_extent_root;
1361
1362         btrfs_read_block_groups(extent_root);
1363
1364         fs_info->generation = btrfs_super_generation(disk_super) + 1;
1365         fs_info->data_alloc_profile = (u64)-1;
1366         fs_info->metadata_alloc_profile = (u64)-1;
1367         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1368
1369         mutex_unlock(&fs_info->fs_mutex);
1370         return tree_root;
1371
1372 fail_extent_root:
1373         free_extent_buffer(extent_root->node);
1374 fail_tree_root:
1375         free_extent_buffer(tree_root->node);
1376 fail_sys_array:
1377         mutex_unlock(&fs_info->fs_mutex);
1378 fail_sb_buffer:
1379         extent_io_tree_empty_lru(&BTRFS_I(fs_info->btree_inode)->io_tree);
1380 fail_iput:
1381         iput(fs_info->btree_inode);
1382 fail:
1383         btrfs_close_devices(fs_info->fs_devices);
1384         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1385
1386         kfree(extent_root);
1387         kfree(tree_root);
1388 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,23)
1389         bdi_destroy(&fs_info->bdi);
1390 #endif
1391         kfree(fs_info);
1392         return ERR_PTR(err);
1393 }
1394
1395 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1396 {
1397         char b[BDEVNAME_SIZE];
1398
1399         if (uptodate) {
1400                 set_buffer_uptodate(bh);
1401         } else {
1402                 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1403                         printk(KERN_WARNING "lost page write due to "
1404                                         "I/O error on %s\n",
1405                                        bdevname(bh->b_bdev, b));
1406                 }
1407                 /* note, we dont' set_buffer_write_io_error because we have
1408                  * our own ways of dealing with the IO errors
1409                  */
1410                 clear_buffer_uptodate(bh);
1411         }
1412         unlock_buffer(bh);
1413         put_bh(bh);
1414 }
1415
1416 int write_all_supers(struct btrfs_root *root)
1417 {
1418         struct list_head *cur;
1419         struct list_head *head = &root->fs_info->fs_devices->devices;
1420         struct btrfs_device *dev;
1421         struct btrfs_super_block *sb;
1422         struct btrfs_dev_item *dev_item;
1423         struct buffer_head *bh;
1424         int ret;
1425         int do_barriers;
1426         int max_errors;
1427         int total_errors = 0;
1428         u32 crc;
1429         u64 flags;
1430
1431         max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1432         do_barriers = !btrfs_test_opt(root, NOBARRIER);
1433
1434         sb = &root->fs_info->super_for_commit;
1435         dev_item = &sb->dev_item;
1436         list_for_each(cur, head) {
1437                 dev = list_entry(cur, struct btrfs_device, dev_list);
1438                 if (!dev->bdev) {
1439                         total_errors++;
1440                         continue;
1441                 }
1442                 if (!dev->in_fs_metadata)
1443                         continue;
1444
1445                 btrfs_set_stack_device_type(dev_item, dev->type);
1446                 btrfs_set_stack_device_id(dev_item, dev->devid);
1447                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
1448                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
1449                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
1450                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
1451                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
1452                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
1453                 flags = btrfs_super_flags(sb);
1454                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
1455
1456
1457                 crc = ~(u32)0;
1458                 crc = btrfs_csum_data(root, (char *)sb + BTRFS_CSUM_SIZE, crc,
1459                                       BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
1460                 btrfs_csum_final(crc, sb->csum);
1461
1462                 bh = __getblk(dev->bdev, BTRFS_SUPER_INFO_OFFSET / 4096,
1463                               BTRFS_SUPER_INFO_SIZE);
1464
1465                 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
1466                 dev->pending_io = bh;
1467
1468                 get_bh(bh);
1469                 set_buffer_uptodate(bh);
1470                 lock_buffer(bh);
1471                 bh->b_end_io = btrfs_end_buffer_write_sync;
1472
1473                 if (do_barriers && dev->barriers) {
1474                         ret = submit_bh(WRITE_BARRIER, bh);
1475                         if (ret == -EOPNOTSUPP) {
1476                                 printk("btrfs: disabling barriers on dev %s\n",
1477                                        dev->name);
1478                                 set_buffer_uptodate(bh);
1479                                 dev->barriers = 0;
1480                                 get_bh(bh);
1481                                 lock_buffer(bh);
1482                                 ret = submit_bh(WRITE, bh);
1483                         }
1484                 } else {
1485                         ret = submit_bh(WRITE, bh);
1486                 }
1487                 if (ret)
1488                         total_errors++;
1489         }
1490         if (total_errors > max_errors) {
1491                 printk("btrfs: %d errors while writing supers\n", total_errors);
1492                 BUG();
1493         }
1494         total_errors = 0;
1495
1496         list_for_each(cur, head) {
1497                 dev = list_entry(cur, struct btrfs_device, dev_list);
1498                 if (!dev->bdev)
1499                         continue;
1500                 if (!dev->in_fs_metadata)
1501                         continue;
1502
1503                 BUG_ON(!dev->pending_io);
1504                 bh = dev->pending_io;
1505                 wait_on_buffer(bh);
1506                 if (!buffer_uptodate(dev->pending_io)) {
1507                         if (do_barriers && dev->barriers) {
1508                                 printk("btrfs: disabling barriers on dev %s\n",
1509                                        dev->name);
1510                                 set_buffer_uptodate(bh);
1511                                 get_bh(bh);
1512                                 lock_buffer(bh);
1513                                 dev->barriers = 0;
1514                                 ret = submit_bh(WRITE, bh);
1515                                 BUG_ON(ret);
1516                                 wait_on_buffer(bh);
1517                                 if (!buffer_uptodate(bh))
1518                                         total_errors++;
1519                         } else {
1520                                 total_errors++;
1521                         }
1522
1523                 }
1524                 dev->pending_io = NULL;
1525                 brelse(bh);
1526         }
1527         if (total_errors > max_errors) {
1528                 printk("btrfs: %d errors while writing supers\n", total_errors);
1529                 BUG();
1530         }
1531         return 0;
1532 }
1533
1534 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
1535                       *root)
1536 {
1537         int ret;
1538
1539         ret = write_all_supers(root);
1540         return ret;
1541 }
1542
1543 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
1544 {
1545         radix_tree_delete(&fs_info->fs_roots_radix,
1546                           (unsigned long)root->root_key.objectid);
1547         if (root->in_sysfs)
1548                 btrfs_sysfs_del_root(root);
1549         if (root->inode)
1550                 iput(root->inode);
1551         if (root->node)
1552                 free_extent_buffer(root->node);
1553         if (root->commit_root)
1554                 free_extent_buffer(root->commit_root);
1555         if (root->name)
1556                 kfree(root->name);
1557         kfree(root);
1558         return 0;
1559 }
1560
1561 static int del_fs_roots(struct btrfs_fs_info *fs_info)
1562 {
1563         int ret;
1564         struct btrfs_root *gang[8];
1565         int i;
1566
1567         while(1) {
1568                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1569                                              (void **)gang, 0,
1570                                              ARRAY_SIZE(gang));
1571                 if (!ret)
1572                         break;
1573                 for (i = 0; i < ret; i++)
1574                         btrfs_free_fs_root(fs_info, gang[i]);
1575         }
1576         return 0;
1577 }
1578
1579 int close_ctree(struct btrfs_root *root)
1580 {
1581         int ret;
1582         struct btrfs_trans_handle *trans;
1583         struct btrfs_fs_info *fs_info = root->fs_info;
1584
1585         fs_info->closing = 1;
1586         btrfs_transaction_flush_work(root);
1587         mutex_lock(&fs_info->fs_mutex);
1588         btrfs_defrag_dirty_roots(root->fs_info);
1589         trans = btrfs_start_transaction(root, 1);
1590         ret = btrfs_commit_transaction(trans, root);
1591         /* run commit again to  drop the original snapshot */
1592         trans = btrfs_start_transaction(root, 1);
1593         btrfs_commit_transaction(trans, root);
1594         ret = btrfs_write_and_wait_transaction(NULL, root);
1595         BUG_ON(ret);
1596
1597         write_ctree_super(NULL, root);
1598         mutex_unlock(&fs_info->fs_mutex);
1599
1600         btrfs_transaction_flush_work(root);
1601
1602         if (fs_info->delalloc_bytes) {
1603                 printk("btrfs: at unmount delalloc count %Lu\n",
1604                        fs_info->delalloc_bytes);
1605         }
1606         if (fs_info->extent_root->node)
1607                 free_extent_buffer(fs_info->extent_root->node);
1608
1609         if (fs_info->tree_root->node)
1610                 free_extent_buffer(fs_info->tree_root->node);
1611
1612         if (root->fs_info->chunk_root->node);
1613                 free_extent_buffer(root->fs_info->chunk_root->node);
1614
1615         if (root->fs_info->dev_root->node);
1616                 free_extent_buffer(root->fs_info->dev_root->node);
1617
1618         btrfs_free_block_groups(root->fs_info);
1619         del_fs_roots(fs_info);
1620
1621         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1622
1623         extent_io_tree_empty_lru(&fs_info->free_space_cache);
1624         extent_io_tree_empty_lru(&fs_info->block_group_cache);
1625         extent_io_tree_empty_lru(&fs_info->pinned_extents);
1626         extent_io_tree_empty_lru(&fs_info->pending_del);
1627         extent_io_tree_empty_lru(&fs_info->extent_ins);
1628         extent_io_tree_empty_lru(&BTRFS_I(fs_info->btree_inode)->io_tree);
1629
1630         flush_workqueue(async_submit_workqueue);
1631         flush_workqueue(end_io_workqueue);
1632
1633         truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
1634
1635         flush_workqueue(async_submit_workqueue);
1636         destroy_workqueue(async_submit_workqueue);
1637
1638         flush_workqueue(end_io_workqueue);
1639         destroy_workqueue(end_io_workqueue);
1640
1641         iput(fs_info->btree_inode);
1642 #if 0
1643         while(!list_empty(&fs_info->hashers)) {
1644                 struct btrfs_hasher *hasher;
1645                 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
1646                                     hashers);
1647                 list_del(&hasher->hashers);
1648                 crypto_free_hash(&fs_info->hash_tfm);
1649                 kfree(hasher);
1650         }
1651 #endif
1652         btrfs_close_devices(fs_info->fs_devices);
1653         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1654
1655 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,23)
1656         bdi_destroy(&fs_info->bdi);
1657 #endif
1658
1659         kfree(fs_info->extent_root);
1660         kfree(fs_info->tree_root);
1661         kfree(fs_info->chunk_root);
1662         kfree(fs_info->dev_root);
1663         return 0;
1664 }
1665
1666 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
1667 {
1668         int ret;
1669         struct inode *btree_inode = buf->first_page->mapping->host;
1670
1671         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
1672         if (!ret)
1673                 return ret;
1674
1675         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
1676                                     parent_transid);
1677         return !ret;
1678 }
1679
1680 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
1681 {
1682         struct inode *btree_inode = buf->first_page->mapping->host;
1683         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
1684                                           buf);
1685 }
1686
1687 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
1688 {
1689         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1690         u64 transid = btrfs_header_generation(buf);
1691         struct inode *btree_inode = root->fs_info->btree_inode;
1692
1693         if (transid != root->fs_info->generation) {
1694                 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
1695                         (unsigned long long)buf->start,
1696                         transid, root->fs_info->generation);
1697                 WARN_ON(1);
1698         }
1699         set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
1700 }
1701
1702 void btrfs_throttle(struct btrfs_root *root)
1703 {
1704         struct backing_dev_info *bdi;
1705
1706         bdi = &root->fs_info->bdi;
1707         if (root->fs_info->throttles && bdi_write_congested(bdi)) {
1708 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,18)
1709                 congestion_wait(WRITE, HZ/20);
1710 #else
1711                 blk_congestion_wait(WRITE, HZ/20);
1712 #endif
1713         }
1714 }
1715
1716 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
1717 {
1718         /*
1719          * looks as though older kernels can get into trouble with
1720          * this code, they end up stuck in balance_dirty_pages forever
1721          */
1722         struct extent_io_tree *tree;
1723         u64 num_dirty;
1724         u64 start = 0;
1725         unsigned long thresh = 16 * 1024 * 1024;
1726         tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
1727
1728         if (current_is_pdflush())
1729                 return;
1730
1731         num_dirty = count_range_bits(tree, &start, (u64)-1,
1732                                      thresh, EXTENT_DIRTY);
1733         if (num_dirty > thresh) {
1734                 balance_dirty_pages_ratelimited_nr(
1735                                    root->fs_info->btree_inode->i_mapping, 1);
1736         }
1737         return;
1738 }
1739
1740 void btrfs_set_buffer_defrag(struct extent_buffer *buf)
1741 {
1742         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1743         struct inode *btree_inode = root->fs_info->btree_inode;
1744         set_extent_bits(&BTRFS_I(btree_inode)->io_tree, buf->start,
1745                         buf->start + buf->len - 1, EXTENT_DEFRAG, GFP_NOFS);
1746 }
1747
1748 void btrfs_set_buffer_defrag_done(struct extent_buffer *buf)
1749 {
1750         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1751         struct inode *btree_inode = root->fs_info->btree_inode;
1752         set_extent_bits(&BTRFS_I(btree_inode)->io_tree, buf->start,
1753                         buf->start + buf->len - 1, EXTENT_DEFRAG_DONE,
1754                         GFP_NOFS);
1755 }
1756
1757 int btrfs_buffer_defrag(struct extent_buffer *buf)
1758 {
1759         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1760         struct inode *btree_inode = root->fs_info->btree_inode;
1761         return test_range_bit(&BTRFS_I(btree_inode)->io_tree,
1762                      buf->start, buf->start + buf->len - 1, EXTENT_DEFRAG, 0);
1763 }
1764
1765 int btrfs_buffer_defrag_done(struct extent_buffer *buf)
1766 {
1767         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1768         struct inode *btree_inode = root->fs_info->btree_inode;
1769         return test_range_bit(&BTRFS_I(btree_inode)->io_tree,
1770                      buf->start, buf->start + buf->len - 1,
1771                      EXTENT_DEFRAG_DONE, 0);
1772 }
1773
1774 int btrfs_clear_buffer_defrag_done(struct extent_buffer *buf)
1775 {
1776         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1777         struct inode *btree_inode = root->fs_info->btree_inode;
1778         return clear_extent_bits(&BTRFS_I(btree_inode)->io_tree,
1779                      buf->start, buf->start + buf->len - 1,
1780                      EXTENT_DEFRAG_DONE, GFP_NOFS);
1781 }
1782
1783 int btrfs_clear_buffer_defrag(struct extent_buffer *buf)
1784 {
1785         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1786         struct inode *btree_inode = root->fs_info->btree_inode;
1787         return clear_extent_bits(&BTRFS_I(btree_inode)->io_tree,
1788                      buf->start, buf->start + buf->len - 1,
1789                      EXTENT_DEFRAG, GFP_NOFS);
1790 }
1791
1792 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
1793 {
1794         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1795         int ret;
1796         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1797         if (ret == 0) {
1798                 buf->flags |= EXTENT_UPTODATE;
1799         }
1800         return ret;
1801 }
1802
1803 static struct extent_io_ops btree_extent_io_ops = {
1804         .writepage_io_hook = btree_writepage_io_hook,
1805         .readpage_end_io_hook = btree_readpage_end_io_hook,
1806         .submit_bio_hook = btree_submit_bio_hook,
1807         /* note we're sharing with inode.c for the merge bio hook */
1808         .merge_bio_hook = btrfs_merge_bio_hook,
1809 };