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         spin_unlock(&em_tree->lock);
  82         if (em)
  83                 goto out;
  84
  85         em = alloc_extent_map(GFP_NOFS);
  86         if (!em) {
  87                 em = ERR_PTR(-ENOMEM);
  88                 goto out;
  89         }
  90         em->start = 0;
  91         em->len = (u64)-1;
  92         em->block_start = 0;
  93         em->bdev = inode->i_sb->s_bdev;
  94
  95         spin_lock(&em_tree->lock);
  96         ret = add_extent_mapping(em_tree, em);
  97         if (ret == -EEXIST) {
  98                 u64 failed_start = em->start;
  99                 u64 failed_len = em->len;
 100
 101                 printk("failed to insert %Lu %Lu -> %Lu into tree\n",
 102                        em->start, em->len, em->block_start);
 103                 free_extent_map(em);
 104                 em = lookup_extent_mapping(em_tree, start, len);
 105                 if (em) {
 106                         printk("after failing, found %Lu %Lu %Lu\n",
 107                                em->start, em->len, em->block_start);
 108                         ret = 0;
 109                 } else {
 110                         em = lookup_extent_mapping(em_tree, failed_start,
 111                                                    failed_len);
 112                         if (em) {
 113                                 printk("double failure lookup gives us "
 114                                        "%Lu %Lu -> %Lu\n", em->start,
 115                                        em->len, em->block_start);
 116                                 free_extent_map(em);
 117                         }
 118                         ret = -EIO;
 119                 }
 120         } else if (ret) {
 121                 free_extent_map(em);
 122                 em = NULL;
 123         }
 124         spin_unlock(&em_tree->lock);
 125
 126         if (ret)
 127                 em = ERR_PTR(ret);
 128 out:
 129         return em;
 130 }
 131
 132 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
 133 {
 134         return btrfs_crc32c(seed, data, len);
 135 }
 136
 137 void btrfs_csum_final(u32 crc, char *result)
 138 {
 139         *(__le32 *)result = ~cpu_to_le32(crc);
 140 }
 141
 142 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 143                            int verify)
 144 {
 145         char result[BTRFS_CRC32_SIZE];
 146         unsigned long len;
 147         unsigned long cur_len;
 148         unsigned long offset = BTRFS_CSUM_SIZE;
 149         char *map_token = NULL;
 150         char *kaddr;
 151         unsigned long map_start;
 152         unsigned long map_len;
 153         int err;
 154         u32 crc = ~(u32)0;
 155
 156         len = buf->len - offset;
 157         while(len > 0) {
 158                 err = map_private_extent_buffer(buf, offset, 32,
 159                                         &map_token, &kaddr,
 160                                         &map_start, &map_len, KM_USER0);
 161                 if (err) {
 162                         printk("failed to map extent buffer! %lu\n",
 163                                offset);
 164                         return 1;
 165                 }
 166                 cur_len = min(len, map_len - (offset - map_start));
 167                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
 168                                       crc, cur_len);
 169                 len -= cur_len;
 170                 offset += cur_len;
 171                 unmap_extent_buffer(buf, map_token, KM_USER0);
 172         }
 173         btrfs_csum_final(crc, result);
 174
 175         if (verify) {
 176                 int from_this_trans = 0;
 177
 178                 if (root->fs_info->running_transaction &&
 179                     btrfs_header_generation(buf) ==
 180                     root->fs_info->running_transaction->transid)
 181                         from_this_trans = 1;
 182
 183                 /* FIXME, this is not good */
 184                 if (memcmp_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE)) {
 185                         u32 val;
 186                         u32 found = 0;
 187                         memcpy(&found, result, BTRFS_CRC32_SIZE);
 188
 189                         read_extent_buffer(buf, &val, 0, BTRFS_CRC32_SIZE);
 190                         printk("btrfs: %s checksum verify failed on %llu "
 191                                "wanted %X found %X from_this_trans %d "
 192                                "level %d\n",
 193                                root->fs_info->sb->s_id,
 194                                buf->start, val, found, from_this_trans,
 195                                btrfs_header_level(buf));
 196                         return 1;
 197                 }
 198         } else {
 199                 write_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE);
 200         }
 201         return 0;
 202 }
 203
 204 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 205                                           struct extent_buffer *eb,
 206                                           u64 start)
 207 {
 208         struct extent_io_tree *io_tree;
 209         int ret;
 210         int num_copies = 0;
 211         int mirror_num = 0;
 212
 213         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 214         while (1) {
 215                 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
 216                                                btree_get_extent, mirror_num);
 217                 if (!ret) {
 218                         if (mirror_num)
 219 printk("good read %Lu mirror %d total %d\n", eb->start, mirror_num, num_copies);
 220                         return ret;
 221                 }
 222                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
 223                                               eb->start, eb->len);
 224 printk("failed to read %Lu mirror %d total %d\n", eb->start, mirror_num, num_copies);
 225                 if (num_copies == 1) {
 226 printk("reading %Lu failed only one copy\n", eb->start);
 227                         return ret;
 228                 }
 229                 mirror_num++;
 230                 if (mirror_num > num_copies) {
 231 printk("bailing at mirror %d of %d\n", mirror_num, num_copies);
 232                         return ret;
 233                 }
 234         }
 235 printk("read extent buffer page last\n");
 236         return -EIO;
 237 }
 238
 239 int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 240 {
 241         struct extent_io_tree *tree;
 242         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
 243         u64 found_start;
 244         int found_level;
 245         unsigned long len;
 246         struct extent_buffer *eb;
 247         int ret;
 248
 249         tree = &BTRFS_I(page->mapping->host)->io_tree;
 250
 251         if (page->private == EXTENT_PAGE_PRIVATE)
 252                 goto out;
 253         if (!page->private)
 254                 goto out;
 255         len = page->private >> 2;
 256         if (len == 0) {
 257                 WARN_ON(1);
 258         }
 259         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 260         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE);
 261         BUG_ON(ret);
 262         btrfs_clear_buffer_defrag(eb);
 263         found_start = btrfs_header_bytenr(eb);
 264         if (found_start != start) {
 265                 printk("warning: eb start incorrect %Lu buffer %Lu len %lu\n",
 266                        start, found_start, len);
 267                 WARN_ON(1);
 268                 goto err;
 269         }
 270         if (eb->first_page != page) {
 271                 printk("bad first page %lu %lu\n", eb->first_page->index,
 272                        page->index);
 273                 WARN_ON(1);
 274                 goto err;
 275         }
 276         if (!PageUptodate(page)) {
 277                 printk("csum not up to date page %lu\n", page->index);
 278                 WARN_ON(1);
 279                 goto err;
 280         }
 281         found_level = btrfs_header_level(eb);
 282         spin_lock(&root->fs_info->hash_lock);
 283         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
 284         spin_unlock(&root->fs_info->hash_lock);
 285         csum_tree_block(root, eb, 0);
 286 err:
 287         free_extent_buffer(eb);
 288 out:
 289         return 0;
 290 }
 291
 292 static int btree_writepage_io_hook(struct page *page, u64 start, u64 end)
 293 {
 294         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 295
 296         csum_dirty_buffer(root, page);
 297         return 0;
 298 }
 299
 300 int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 301                                struct extent_state *state)
 302 {
 303         struct extent_io_tree *tree;
 304         u64 found_start;
 305         int found_level;
 306         unsigned long len;
 307         struct extent_buffer *eb;
 308         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 309         int ret = 0;
 310
 311         tree = &BTRFS_I(page->mapping->host)->io_tree;
 312         if (page->private == EXTENT_PAGE_PRIVATE)
 313                 goto out;
 314         if (!page->private)
 315                 goto out;
 316         len = page->private >> 2;
 317         if (len == 0) {
 318                 WARN_ON(1);
 319         }
 320         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 321
 322         btrfs_clear_buffer_defrag(eb);
 323         found_start = btrfs_header_bytenr(eb);
 324         if (found_start != start) {
 325 printk("bad start on %Lu found %Lu\n", eb->start, found_start);
 326                 ret = -EIO;
 327                 goto err;
 328         }
 329         if (eb->first_page != page) {
 330                 printk("bad first page %lu %lu\n", eb->first_page->index,
 331                        page->index);
 332                 WARN_ON(1);
 333                 ret = -EIO;
 334                 goto err;
 335         }
 336         found_level = btrfs_header_level(eb);
 337
 338         ret = csum_tree_block(root, eb, 1);
 339         if (ret)
 340                 ret = -EIO;
 341
 342         end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
 343         end = eb->start + end - 1;
 344         release_extent_buffer_tail_pages(eb);
 345 err:
 346         free_extent_buffer(eb);
 347 out:
 348         return ret;
 349 }
 350
 351 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
 352 static void end_workqueue_bio(struct bio *bio, int err)
 353 #else
 354 static int end_workqueue_bio(struct bio *bio,
 355                                    unsigned int bytes_done, int err)
 356 #endif
 357 {
 358         struct end_io_wq *end_io_wq = bio->bi_private;
 359         struct btrfs_fs_info *fs_info;
 360         unsigned long flags;
 361
 362 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
 363         if (bio->bi_size)
 364                 return 1;
 365 #endif
 366
 367         fs_info = end_io_wq->info;
 368         spin_lock_irqsave(&fs_info->end_io_work_lock, flags);
 369         end_io_wq->error = err;
 370         list_add_tail(&end_io_wq->list, &fs_info->end_io_work_list);
 371         spin_unlock_irqrestore(&fs_info->end_io_work_lock, flags);
 372         queue_work(end_io_workqueue, &fs_info->end_io_work);
 373
 374 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
 375         return 0;
 376 #endif
 377 }
 378
 379 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 380                         int metadata)
 381 {
 382         struct end_io_wq *end_io_wq;
 383         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
 384         if (!end_io_wq)
 385                 return -ENOMEM;
 386
 387         end_io_wq->private = bio->bi_private;
 388         end_io_wq->end_io = bio->bi_end_io;
 389         end_io_wq->info = info;
 390         end_io_wq->error = 0;
 391         end_io_wq->bio = bio;
 392         end_io_wq->metadata = metadata;
 393
 394         bio->bi_private = end_io_wq;
 395         bio->bi_end_io = end_workqueue_bio;
 396         return 0;
 397 }
 398
 399 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 400                         int rw, struct bio *bio, int mirror_num,
 401                         extent_submit_bio_hook_t *submit_bio_hook)
 402 {
 403         struct async_submit_bio *async;
 404
 405         /*
 406          * inline writerback should stay inline, only hop to the async
 407          * queue if we're pdflush
 408          */
 409         if (!current_is_pdflush())
 410                 return submit_bio_hook(inode, rw, bio, mirror_num);
 411
 412         async = kmalloc(sizeof(*async), GFP_NOFS);
 413         if (!async)
 414                 return -ENOMEM;
 415
 416         async->inode = inode;
 417         async->rw = rw;
 418         async->bio = bio;
 419         async->mirror_num = mirror_num;
 420         async->submit_bio_hook = submit_bio_hook;
 421
 422         spin_lock(&fs_info->async_submit_work_lock);
 423         list_add_tail(&async->list, &fs_info->async_submit_work_list);
 424         spin_unlock(&fs_info->async_submit_work_lock);
 425
 426         queue_work(async_submit_workqueue, &fs_info->async_submit_work);
 427         return 0;
 428 }
 429
 430 static int __btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 431                                  int mirror_num)
 432 {
 433         struct btrfs_root *root = BTRFS_I(inode)->root;
 434         u64 offset;
 435         int ret;
 436
 437         offset = bio->bi_sector << 9;
 438
 439         if (rw & (1 << BIO_RW)) {
 440                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num);
 441         }
 442
 443         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 1);
 444         BUG_ON(ret);
 445
 446         if (offset == BTRFS_SUPER_INFO_OFFSET) {
 447                 bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
 448                 submit_bio(rw, bio);
 449                 return 0;
 450         }
 451         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num);
 452 }
 453
 454 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 455                                  int mirror_num)
 456 {
 457         if (!(rw & (1 << BIO_RW))) {
 458                 return __btree_submit_bio_hook(inode, rw, bio, mirror_num);
 459         }
 460         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 461                                    inode, rw, bio, mirror_num,
 462                                    __btree_submit_bio_hook);
 463 }
 464
 465 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 466 {
 467         struct extent_io_tree *tree;
 468         tree = &BTRFS_I(page->mapping->host)->io_tree;
 469         return extent_write_full_page(tree, page, btree_get_extent, wbc);
 470 }
 471
 472 static int btree_writepages(struct address_space *mapping,
 473                             struct writeback_control *wbc)
 474 {
 475         struct extent_io_tree *tree;
 476         tree = &BTRFS_I(mapping->host)->io_tree;
 477         if (wbc->sync_mode == WB_SYNC_NONE) {
 478                 u64 num_dirty;
 479                 u64 start = 0;
 480                 unsigned long thresh = 96 * 1024 * 1024;
 481
 482                 if (wbc->for_kupdate)
 483                         return 0;
 484
 485                 if (current_is_pdflush()) {
 486                         thresh = 96 * 1024 * 1024;
 487                 } else {
 488                         thresh = 8 * 1024 * 1024;
 489                 }
 490                 num_dirty = count_range_bits(tree, &start, (u64)-1,
 491                                              thresh, EXTENT_DIRTY);
 492                 if (num_dirty < thresh) {
 493                         return 0;
 494                 }
 495         }
 496         return extent_writepages(tree, mapping, btree_get_extent, wbc);
 497 }
 498
 499 int btree_readpage(struct file *file, struct page *page)
 500 {
 501         struct extent_io_tree *tree;
 502         tree = &BTRFS_I(page->mapping->host)->io_tree;
 503         return extent_read_full_page(tree, page, btree_get_extent);
 504 }
 505
 506 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 507 {
 508         struct extent_io_tree *tree;
 509         struct extent_map_tree *map;
 510         int ret;
 511
 512         if (page_count(page) > 3) {
 513                 /* once for page->private, once for the caller, once
 514                  * once for the page cache
 515                  */
 516                 return 0;
 517         }
 518         tree = &BTRFS_I(page->mapping->host)->io_tree;
 519         map = &BTRFS_I(page->mapping->host)->extent_tree;
 520         ret = try_release_extent_state(map, tree, page, gfp_flags);
 521         if (ret == 1) {
 522                 invalidate_extent_lru(tree, page_offset(page), PAGE_CACHE_SIZE);
 523                 ClearPagePrivate(page);
 524                 set_page_private(page, 0);
 525                 page_cache_release(page);
 526         }
 527         return ret;
 528 }
 529
 530 static void btree_invalidatepage(struct page *page, unsigned long offset)
 531 {
 532         struct extent_io_tree *tree;
 533         tree = &BTRFS_I(page->mapping->host)->io_tree;
 534         extent_invalidatepage(tree, page, offset);
 535         btree_releasepage(page, GFP_NOFS);
 536 }
 537
 538 #if 0
 539 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 540 {
 541         struct buffer_head *bh;
 542         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 543         struct buffer_head *head;
 544         if (!page_has_buffers(page)) {
 545                 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
 546                                         (1 << BH_Dirty)|(1 << BH_Uptodate));
 547         }
 548         head = page_buffers(page);
 549         bh = head;
 550         do {
 551                 if (buffer_dirty(bh))
 552                         csum_tree_block(root, bh, 0);
 553                 bh = bh->b_this_page;
 554         } while (bh != head);
 555         return block_write_full_page(page, btree_get_block, wbc);
 556 }
 557 #endif
 558
 559 static struct address_space_operations btree_aops = {
 560         .readpage       = btree_readpage,
 561         .writepage      = btree_writepage,
 562         .writepages     = btree_writepages,
 563         .releasepage    = btree_releasepage,
 564         .invalidatepage = btree_invalidatepage,
 565         .sync_page      = block_sync_page,
 566 };
 567
 568 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize)
 569 {
 570         struct extent_buffer *buf = NULL;
 571         struct inode *btree_inode = root->fs_info->btree_inode;
 572         int ret = 0;
 573
 574         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 575         if (!buf)
 576                 return 0;
 577         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
 578                                  buf, 0, 0, btree_get_extent, 0);
 579         free_extent_buffer(buf);
 580         return ret;
 581 }
 582
 583 static int close_all_devices(struct btrfs_fs_info *fs_info)
 584 {
 585         struct list_head *list;
 586         struct list_head *next;
 587         struct btrfs_device *device;
 588
 589         list = &fs_info->fs_devices->devices;
 590         list_for_each(next, list) {
 591                 device = list_entry(next, struct btrfs_device, dev_list);
 592                 if (device->bdev && device->bdev != fs_info->sb->s_bdev)
 593                         close_bdev_excl(device->bdev);
 594                 device->bdev = NULL;
 595         }
 596         return 0;
 597 }
 598
 599 int btrfs_verify_block_csum(struct btrfs_root *root,
 600                             struct extent_buffer *buf)
 601 {
 602         return btrfs_buffer_uptodate(buf);
 603 }
 604
 605 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
 606                                             u64 bytenr, u32 blocksize)
 607 {
 608         struct inode *btree_inode = root->fs_info->btree_inode;
 609         struct extent_buffer *eb;
 610         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 611                                 bytenr, blocksize, GFP_NOFS);
 612         return eb;
 613 }
 614
 615 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
 616                                                  u64 bytenr, u32 blocksize)
 617 {
 618         struct inode *btree_inode = root->fs_info->btree_inode;
 619         struct extent_buffer *eb;
 620
 621         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 622                                  bytenr, blocksize, NULL, GFP_NOFS);
 623         return eb;
 624 }
 625
 626
 627 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
 628                                       u32 blocksize)
 629 {
 630         struct extent_buffer *buf = NULL;
 631         struct inode *btree_inode = root->fs_info->btree_inode;
 632         struct extent_io_tree *io_tree;
 633         int ret;
 634
 635         io_tree = &BTRFS_I(btree_inode)->io_tree;
 636
 637         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 638         if (!buf)
 639                 return NULL;
 640
 641         ret = btree_read_extent_buffer_pages(root, buf, 0);
 642
 643         if (ret == 0) {
 644                 buf->flags |= EXTENT_UPTODATE;
 645         }
 646         return buf;
 647
 648 }
 649
 650 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 651                      struct extent_buffer *buf)
 652 {
 653         struct inode *btree_inode = root->fs_info->btree_inode;
 654         if (btrfs_header_generation(buf) ==
 655             root->fs_info->running_transaction->transid)
 656                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
 657                                           buf);
 658         return 0;
 659 }
 660
 661 int wait_on_tree_block_writeback(struct btrfs_root *root,
 662                                  struct extent_buffer *buf)
 663 {
 664         struct inode *btree_inode = root->fs_info->btree_inode;
 665         wait_on_extent_buffer_writeback(&BTRFS_I(btree_inode)->io_tree,
 666                                         buf);
 667         return 0;
 668 }
 669
 670 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
 671                         u32 stripesize, struct btrfs_root *root,
 672                         struct btrfs_fs_info *fs_info,
 673                         u64 objectid)
 674 {
 675         root->node = NULL;
 676         root->inode = NULL;
 677         root->commit_root = NULL;
 678         root->sectorsize = sectorsize;
 679         root->nodesize = nodesize;
 680         root->leafsize = leafsize;
 681         root->stripesize = stripesize;
 682         root->ref_cows = 0;
 683         root->track_dirty = 0;
 684
 685         root->fs_info = fs_info;
 686         root->objectid = objectid;
 687         root->last_trans = 0;
 688         root->highest_inode = 0;
 689         root->last_inode_alloc = 0;
 690         root->name = NULL;
 691         root->in_sysfs = 0;
 692
 693         INIT_LIST_HEAD(&root->dirty_list);
 694         memset(&root->root_key, 0, sizeof(root->root_key));
 695         memset(&root->root_item, 0, sizeof(root->root_item));
 696         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
 697         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
 698         init_completion(&root->kobj_unregister);
 699         root->defrag_running = 0;
 700         root->defrag_level = 0;
 701         root->root_key.objectid = objectid;
 702         return 0;
 703 }
 704
 705 static int find_and_setup_root(struct btrfs_root *tree_root,
 706                                struct btrfs_fs_info *fs_info,
 707                                u64 objectid,
 708                                struct btrfs_root *root)
 709 {
 710         int ret;
 711         u32 blocksize;
 712
 713         __setup_root(tree_root->nodesize, tree_root->leafsize,
 714                      tree_root->sectorsize, tree_root->stripesize,
 715                      root, fs_info, objectid);
 716         ret = btrfs_find_last_root(tree_root, objectid,
 717                                    &root->root_item, &root->root_key);
 718         BUG_ON(ret);
 719
 720         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 721         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 722                                      blocksize);
 723         BUG_ON(!root->node);
 724         return 0;
 725 }
 726
 727 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_fs_info *fs_info,
 728                                                struct btrfs_key *location)
 729 {
 730         struct btrfs_root *root;
 731         struct btrfs_root *tree_root = fs_info->tree_root;
 732         struct btrfs_path *path;
 733         struct extent_buffer *l;
 734         u64 highest_inode;
 735         u32 blocksize;
 736         int ret = 0;
 737
 738         root = kzalloc(sizeof(*root), GFP_NOFS);
 739         if (!root)
 740                 return ERR_PTR(-ENOMEM);
 741         if (location->offset == (u64)-1) {
 742                 ret = find_and_setup_root(tree_root, fs_info,
 743                                           location->objectid, root);
 744                 if (ret) {
 745                         kfree(root);
 746                         return ERR_PTR(ret);
 747                 }
 748                 goto insert;
 749         }
 750
 751         __setup_root(tree_root->nodesize, tree_root->leafsize,
 752                      tree_root->sectorsize, tree_root->stripesize,
 753                      root, fs_info, location->objectid);
 754
 755         path = btrfs_alloc_path();
 756         BUG_ON(!path);
 757         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
 758         if (ret != 0) {
 759                 if (ret > 0)
 760                         ret = -ENOENT;
 761                 goto out;
 762         }
 763         l = path->nodes[0];
 764         read_extent_buffer(l, &root->root_item,
 765                btrfs_item_ptr_offset(l, path->slots[0]),
 766                sizeof(root->root_item));
 767         memcpy(&root->root_key, location, sizeof(*location));
 768         ret = 0;
 769 out:
 770         btrfs_release_path(root, path);
 771         btrfs_free_path(path);
 772         if (ret) {
 773                 kfree(root);
 774                 return ERR_PTR(ret);
 775         }
 776         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 777         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 778                                      blocksize);
 779         BUG_ON(!root->node);
 780 insert:
 781         root->ref_cows = 1;
 782         ret = btrfs_find_highest_inode(root, &highest_inode);
 783         if (ret == 0) {
 784                 root->highest_inode = highest_inode;
 785                 root->last_inode_alloc = highest_inode;
 786         }
 787         return root;
 788 }
 789
 790 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
 791                                         u64 root_objectid)
 792 {
 793         struct btrfs_root *root;
 794
 795         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
 796                 return fs_info->tree_root;
 797         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
 798                 return fs_info->extent_root;
 799
 800         root = radix_tree_lookup(&fs_info->fs_roots_radix,
 801                                  (unsigned long)root_objectid);
 802         return root;
 803 }
 804
 805 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
 806                                               struct btrfs_key *location)
 807 {
 808         struct btrfs_root *root;
 809         int ret;
 810
 811         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
 812                 return fs_info->tree_root;
 813         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
 814                 return fs_info->extent_root;
 815
 816         root = radix_tree_lookup(&fs_info->fs_roots_radix,
 817                                  (unsigned long)location->objectid);
 818         if (root)
 819                 return root;
 820
 821         root = btrfs_read_fs_root_no_radix(fs_info, location);
 822         if (IS_ERR(root))
 823                 return root;
 824         ret = radix_tree_insert(&fs_info->fs_roots_radix,
 825                                 (unsigned long)root->root_key.objectid,
 826                                 root);
 827         if (ret) {
 828                 free_extent_buffer(root->node);
 829                 kfree(root);
 830                 return ERR_PTR(ret);
 831         }
 832         ret = btrfs_find_dead_roots(fs_info->tree_root,
 833                                     root->root_key.objectid, root);
 834         BUG_ON(ret);
 835
 836         return root;
 837 }
 838
 839 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
 840                                       struct btrfs_key *location,
 841                                       const char *name, int namelen)
 842 {
 843         struct btrfs_root *root;
 844         int ret;
 845
 846         root = btrfs_read_fs_root_no_name(fs_info, location);
 847         if (!root)
 848                 return NULL;
 849
 850         if (root->in_sysfs)
 851                 return root;
 852
 853         ret = btrfs_set_root_name(root, name, namelen);
 854         if (ret) {
 855                 free_extent_buffer(root->node);
 856                 kfree(root);
 857                 return ERR_PTR(ret);
 858         }
 859
 860         ret = btrfs_sysfs_add_root(root);
 861         if (ret) {
 862                 free_extent_buffer(root->node);
 863                 kfree(root->name);
 864                 kfree(root);
 865                 return ERR_PTR(ret);
 866         }
 867         root->in_sysfs = 1;
 868         return root;
 869 }
 870 #if 0
 871 static int add_hasher(struct btrfs_fs_info *info, char *type) {
 872         struct btrfs_hasher *hasher;
 873
 874         hasher = kmalloc(sizeof(*hasher), GFP_NOFS);
 875         if (!hasher)
 876                 return -ENOMEM;
 877         hasher->hash_tfm = crypto_alloc_hash(type, 0, CRYPTO_ALG_ASYNC);
 878         if (!hasher->hash_tfm) {
 879                 kfree(hasher);
 880                 return -EINVAL;
 881         }
 882         spin_lock(&info->hash_lock);
 883         list_add(&hasher->list, &info->hashers);
 884         spin_unlock(&info->hash_lock);
 885         return 0;
 886 }
 887 #endif
 888
 889 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
 890 {
 891         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
 892         int ret = 0;
 893         struct list_head *cur;
 894         struct btrfs_device *device;
 895         struct backing_dev_info *bdi;
 896
 897         list_for_each(cur, &info->fs_devices->devices) {
 898                 device = list_entry(cur, struct btrfs_device, dev_list);
 899                 bdi = blk_get_backing_dev_info(device->bdev);
 900                 if (bdi && bdi_congested(bdi, bdi_bits)) {
 901                         ret = 1;
 902                         break;
 903                 }
 904         }
 905         return ret;
 906 }
 907
 908 void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
 909 {
 910         struct list_head *cur;
 911         struct btrfs_device *device;
 912         struct btrfs_fs_info *info;
 913
 914         info = (struct btrfs_fs_info *)bdi->unplug_io_data;
 915         list_for_each(cur, &info->fs_devices->devices) {
 916                 device = list_entry(cur, struct btrfs_device, dev_list);
 917                 bdi = blk_get_backing_dev_info(device->bdev);
 918                 if (bdi->unplug_io_fn) {
 919                         bdi->unplug_io_fn(bdi, page);
 920                 }
 921         }
 922 }
 923
 924 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
 925 {
 926 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,23)
 927         bdi_init(bdi);
 928 #endif
 929         bdi->ra_pages   = default_backing_dev_info.ra_pages * 4;
 930         bdi->state              = 0;
 931         bdi->capabilities       = default_backing_dev_info.capabilities;
 932         bdi->unplug_io_fn       = btrfs_unplug_io_fn;
 933         bdi->unplug_io_data     = info;
 934         bdi->congested_fn       = btrfs_congested_fn;
 935         bdi->congested_data     = info;
 936         return 0;
 937 }
 938
 939 static int bio_ready_for_csum(struct bio *bio)
 940 {
 941         u64 length = 0;
 942         u64 buf_len = 0;
 943         u64 start = 0;
 944         struct page *page;
 945         struct extent_io_tree *io_tree = NULL;
 946         struct btrfs_fs_info *info = NULL;
 947         struct bio_vec *bvec;
 948         int i;
 949         int ret;
 950
 951         bio_for_each_segment(bvec, bio, i) {
 952                 page = bvec->bv_page;
 953                 if (page->private == EXTENT_PAGE_PRIVATE) {
 954                         length += bvec->bv_len;
 955                         continue;
 956                 }
 957                 if (!page->private) {
 958                         length += bvec->bv_len;
 959                         continue;
 960                 }
 961                 length = bvec->bv_len;
 962                 buf_len = page->private >> 2;
 963                 start = page_offset(page) + bvec->bv_offset;
 964                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
 965                 info = BTRFS_I(page->mapping->host)->root->fs_info;
 966         }
 967         /* are we fully contained in this bio? */
 968         if (buf_len <= length)
 969                 return 1;
 970
 971         ret = extent_range_uptodate(io_tree, start + length,
 972                                     start + buf_len - 1);
 973         if (ret == 1)
 974                 return ret;
 975         return ret;
 976 }
 977
 978 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
 979 static void btrfs_end_io_csum(void *p)
 980 #else
 981 static void btrfs_end_io_csum(struct work_struct *work)
 982 #endif
 983 {
 984 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
 985         struct btrfs_fs_info *fs_info = p;
 986 #else
 987         struct btrfs_fs_info *fs_info = container_of(work,
 988                                                      struct btrfs_fs_info,
 989                                                      end_io_work);
 990 #endif
 991         unsigned long flags;
 992         struct end_io_wq *end_io_wq;
 993         struct bio *bio;
 994         struct list_head *next;
 995         int error;
 996         int was_empty;
 997
 998         while(1) {
 999                 spin_lock_irqsave(&fs_info->end_io_work_lock, flags);
1000                 if (list_empty(&fs_info->end_io_work_list)) {
1001                         spin_unlock_irqrestore(&fs_info->end_io_work_lock,
1002                                                flags);
1003                         return;
1004                 }
1005                 next = fs_info->end_io_work_list.next;
1006                 list_del(next);
1007                 spin_unlock_irqrestore(&fs_info->end_io_work_lock, flags);
1008
1009                 end_io_wq = list_entry(next, struct end_io_wq, list);
1010
1011                 bio = end_io_wq->bio;
1012                 if (end_io_wq->metadata && !bio_ready_for_csum(bio)) {
1013                         spin_lock_irqsave(&fs_info->end_io_work_lock, flags);
1014                         was_empty = list_empty(&fs_info->end_io_work_list);
1015                         list_add_tail(&end_io_wq->list,
1016                                       &fs_info->end_io_work_list);
1017                         spin_unlock_irqrestore(&fs_info->end_io_work_lock,
1018                                                flags);
1019                         if (was_empty)
1020                                 return;
1021                         continue;
1022                 }
1023                 error = end_io_wq->error;
1024                 bio->bi_private = end_io_wq->private;
1025                 bio->bi_end_io = end_io_wq->end_io;
1026                 kfree(end_io_wq);
1027 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1028                 bio_endio(bio, bio->bi_size, error);
1029 #else
1030                 bio_endio(bio, error);
1031 #endif
1032         }
1033 }
1034
1035 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1036 static void btrfs_async_submit_work(void *p)
1037 #else
1038 static void btrfs_async_submit_work(struct work_struct *work)
1039 #endif
1040 {
1041 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1042         struct btrfs_fs_info *fs_info = p;
1043 #else
1044         struct btrfs_fs_info *fs_info = container_of(work,
1045                                                      struct btrfs_fs_info,
1046                                                      async_submit_work);
1047 #endif
1048         struct async_submit_bio *async;
1049         struct list_head *next;
1050
1051         while(1) {
1052                 spin_lock(&fs_info->async_submit_work_lock);
1053                 if (list_empty(&fs_info->async_submit_work_list)) {
1054                         spin_unlock(&fs_info->async_submit_work_lock);
1055                         return;
1056                 }
1057                 next = fs_info->async_submit_work_list.next;
1058                 list_del(next);
1059                 spin_unlock(&fs_info->async_submit_work_lock);
1060
1061                 async = list_entry(next, struct async_submit_bio, list);
1062                 async->submit_bio_hook(async->inode, async->rw, async->bio,
1063                                        async->mirror_num);
1064                 kfree(async);
1065         }
1066 }
1067
1068 struct btrfs_root *open_ctree(struct super_block *sb,
1069                               struct btrfs_fs_devices *fs_devices)
1070 {
1071         u32 sectorsize;
1072         u32 nodesize;
1073         u32 leafsize;
1074         u32 blocksize;
1075         u32 stripesize;
1076         struct btrfs_root *extent_root = kmalloc(sizeof(struct btrfs_root),
1077                                                  GFP_NOFS);
1078         struct btrfs_root *tree_root = kmalloc(sizeof(struct btrfs_root),
1079                                                GFP_NOFS);
1080         struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1081                                                 GFP_NOFS);
1082         struct btrfs_root *chunk_root = kmalloc(sizeof(struct btrfs_root),
1083                                                 GFP_NOFS);
1084         struct btrfs_root *dev_root = kmalloc(sizeof(struct btrfs_root),
1085                                               GFP_NOFS);
1086         int ret;
1087         int err = -EINVAL;
1088         struct btrfs_super_block *disk_super;
1089
1090         if (!extent_root || !tree_root || !fs_info) {
1091                 err = -ENOMEM;
1092                 goto fail;
1093         }
1094         end_io_workqueue = create_workqueue("btrfs-end-io");
1095         BUG_ON(!end_io_workqueue);
1096         async_submit_workqueue = create_workqueue("btrfs-async-submit");
1097
1098         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1099         INIT_LIST_HEAD(&fs_info->trans_list);
1100         INIT_LIST_HEAD(&fs_info->dead_roots);
1101         INIT_LIST_HEAD(&fs_info->hashers);
1102         INIT_LIST_HEAD(&fs_info->end_io_work_list);
1103         INIT_LIST_HEAD(&fs_info->async_submit_work_list);
1104         spin_lock_init(&fs_info->hash_lock);
1105         spin_lock_init(&fs_info->end_io_work_lock);
1106         spin_lock_init(&fs_info->async_submit_work_lock);
1107         spin_lock_init(&fs_info->delalloc_lock);
1108         spin_lock_init(&fs_info->new_trans_lock);
1109
1110         init_completion(&fs_info->kobj_unregister);
1111         sb_set_blocksize(sb, BTRFS_SUPER_INFO_SIZE);
1112         fs_info->tree_root = tree_root;
1113         fs_info->extent_root = extent_root;
1114         fs_info->chunk_root = chunk_root;
1115         fs_info->dev_root = dev_root;
1116         fs_info->fs_devices = fs_devices;
1117         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1118         INIT_LIST_HEAD(&fs_info->space_info);
1119         btrfs_mapping_init(&fs_info->mapping_tree);
1120         fs_info->sb = sb;
1121         fs_info->max_extent = (u64)-1;
1122         fs_info->max_inline = 8192 * 1024;
1123         setup_bdi(fs_info, &fs_info->bdi);
1124         fs_info->btree_inode = new_inode(sb);
1125         fs_info->btree_inode->i_ino = 1;
1126         fs_info->btree_inode->i_nlink = 1;
1127
1128         /*
1129          * we set the i_size on the btree inode to the max possible int.
1130          * the real end of the address space is determined by all of
1131          * the devices in the system
1132          */
1133         fs_info->btree_inode->i_size = OFFSET_MAX;
1134         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1135         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1136
1137         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1138                              fs_info->btree_inode->i_mapping,
1139                              GFP_NOFS);
1140         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1141                              GFP_NOFS);
1142
1143         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1144
1145         extent_io_tree_init(&fs_info->free_space_cache,
1146                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1147         extent_io_tree_init(&fs_info->block_group_cache,
1148                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1149         extent_io_tree_init(&fs_info->pinned_extents,
1150                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1151         extent_io_tree_init(&fs_info->pending_del,
1152                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1153         extent_io_tree_init(&fs_info->extent_ins,
1154                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1155         fs_info->do_barriers = 1;
1156
1157 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1158         INIT_WORK(&fs_info->end_io_work, btrfs_end_io_csum, fs_info);
1159         INIT_WORK(&fs_info->async_submit_work, btrfs_async_submit_work,
1160                   fs_info);
1161         INIT_WORK(&fs_info->trans_work, btrfs_transaction_cleaner, fs_info);
1162 #else
1163         INIT_WORK(&fs_info->end_io_work, btrfs_end_io_csum);
1164         INIT_WORK(&fs_info->async_submit_work, btrfs_async_submit_work);
1165         INIT_DELAYED_WORK(&fs_info->trans_work, btrfs_transaction_cleaner);
1166 #endif
1167         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1168         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1169                sizeof(struct btrfs_key));
1170         insert_inode_hash(fs_info->btree_inode);
1171         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1172
1173         mutex_init(&fs_info->trans_mutex);
1174         mutex_init(&fs_info->fs_mutex);
1175
1176 #if 0
1177         ret = add_hasher(fs_info, "crc32c");
1178         if (ret) {
1179                 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
1180                 err = -ENOMEM;
1181                 goto fail_iput;
1182         }
1183 #endif
1184         __setup_root(4096, 4096, 4096, 4096, tree_root,
1185                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1186
1187         fs_info->sb_buffer = read_tree_block(tree_root,
1188                                              BTRFS_SUPER_INFO_OFFSET,
1189                                              4096);
1190
1191         if (!fs_info->sb_buffer)
1192                 goto fail_iput;
1193
1194         read_extent_buffer(fs_info->sb_buffer, &fs_info->super_copy, 0,
1195                            sizeof(fs_info->super_copy));
1196
1197         read_extent_buffer(fs_info->sb_buffer, fs_info->fsid,
1198                            (unsigned long)btrfs_super_fsid(fs_info->sb_buffer),
1199                            BTRFS_FSID_SIZE);
1200
1201         disk_super = &fs_info->super_copy;
1202         if (!btrfs_super_root(disk_super))
1203                 goto fail_sb_buffer;
1204
1205         if (btrfs_super_num_devices(disk_super) != fs_devices->num_devices) {
1206                 printk("Btrfs: wanted %llu devices, but found %llu\n",
1207                        (unsigned long long)btrfs_super_num_devices(disk_super),
1208                        (unsigned long long)fs_devices->num_devices);
1209                 goto fail_sb_buffer;
1210         }
1211         nodesize = btrfs_super_nodesize(disk_super);
1212         leafsize = btrfs_super_leafsize(disk_super);
1213         sectorsize = btrfs_super_sectorsize(disk_super);
1214         stripesize = btrfs_super_stripesize(disk_super);
1215         tree_root->nodesize = nodesize;
1216         tree_root->leafsize = leafsize;
1217         tree_root->sectorsize = sectorsize;
1218         tree_root->stripesize = stripesize;
1219         sb_set_blocksize(sb, sectorsize);
1220
1221         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1222                     sizeof(disk_super->magic))) {
1223                 printk("btrfs: valid FS not found on %s\n", sb->s_id);
1224                 goto fail_sb_buffer;
1225         }
1226
1227         mutex_lock(&fs_info->fs_mutex);
1228
1229         ret = btrfs_read_sys_array(tree_root);
1230         BUG_ON(ret);
1231
1232         blocksize = btrfs_level_size(tree_root,
1233                                      btrfs_super_chunk_root_level(disk_super));
1234
1235         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1236                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1237
1238         chunk_root->node = read_tree_block(chunk_root,
1239                                            btrfs_super_chunk_root(disk_super),
1240                                            blocksize);
1241         BUG_ON(!chunk_root->node);
1242
1243         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1244                  (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1245                  BTRFS_UUID_SIZE);
1246
1247         ret = btrfs_read_chunk_tree(chunk_root);
1248         BUG_ON(ret);
1249
1250         blocksize = btrfs_level_size(tree_root,
1251                                      btrfs_super_root_level(disk_super));
1252
1253
1254         tree_root->node = read_tree_block(tree_root,
1255                                           btrfs_super_root(disk_super),
1256                                           blocksize);
1257         if (!tree_root->node)
1258                 goto fail_sb_buffer;
1259
1260
1261         ret = find_and_setup_root(tree_root, fs_info,
1262                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1263         if (ret)
1264                 goto fail_tree_root;
1265         extent_root->track_dirty = 1;
1266
1267         ret = find_and_setup_root(tree_root, fs_info,
1268                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
1269         dev_root->track_dirty = 1;
1270
1271         if (ret)
1272                 goto fail_extent_root;
1273
1274         btrfs_read_block_groups(extent_root);
1275
1276         fs_info->generation = btrfs_super_generation(disk_super) + 1;
1277         fs_info->data_alloc_profile = (u64)-1;
1278         fs_info->metadata_alloc_profile = (u64)-1;
1279         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1280
1281         mutex_unlock(&fs_info->fs_mutex);
1282         return tree_root;
1283
1284 fail_extent_root:
1285         free_extent_buffer(extent_root->node);
1286 fail_tree_root:
1287         mutex_unlock(&fs_info->fs_mutex);
1288         free_extent_buffer(tree_root->node);
1289 fail_sb_buffer:
1290         free_extent_buffer(fs_info->sb_buffer);
1291         extent_io_tree_empty_lru(&BTRFS_I(fs_info->btree_inode)->io_tree);
1292 fail_iput:
1293         iput(fs_info->btree_inode);
1294 fail:
1295         close_all_devices(fs_info);
1296         kfree(extent_root);
1297         kfree(tree_root);
1298 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,23)
1299         bdi_destroy(&fs_info->bdi);
1300 #endif
1301         kfree(fs_info);
1302         return ERR_PTR(err);
1303 }
1304
1305 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1306 {
1307         char b[BDEVNAME_SIZE];
1308
1309         if (uptodate) {
1310                 set_buffer_uptodate(bh);
1311         } else {
1312                 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1313                         printk(KERN_WARNING "lost page write due to "
1314                                         "I/O error on %s\n",
1315                                        bdevname(bh->b_bdev, b));
1316                 }
1317                 set_buffer_write_io_error(bh);
1318                 clear_buffer_uptodate(bh);
1319         }
1320         unlock_buffer(bh);
1321         put_bh(bh);
1322 }
1323
1324 int write_all_supers(struct btrfs_root *root)
1325 {
1326         struct list_head *cur;
1327         struct list_head *head = &root->fs_info->fs_devices->devices;
1328         struct btrfs_device *dev;
1329         struct extent_buffer *sb;
1330         struct btrfs_dev_item *dev_item;
1331         struct buffer_head *bh;
1332         int ret;
1333         int do_barriers;
1334
1335         do_barriers = !btrfs_test_opt(root, NOBARRIER);
1336
1337         sb = root->fs_info->sb_buffer;
1338         dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
1339                                                       dev_item);
1340         list_for_each(cur, head) {
1341                 dev = list_entry(cur, struct btrfs_device, dev_list);
1342                 btrfs_set_device_type(sb, dev_item, dev->type);
1343                 btrfs_set_device_id(sb, dev_item, dev->devid);
1344                 btrfs_set_device_total_bytes(sb, dev_item, dev->total_bytes);
1345                 btrfs_set_device_bytes_used(sb, dev_item, dev->bytes_used);
1346                 btrfs_set_device_io_align(sb, dev_item, dev->io_align);
1347                 btrfs_set_device_io_width(sb, dev_item, dev->io_width);
1348                 btrfs_set_device_sector_size(sb, dev_item, dev->sector_size);
1349                 write_extent_buffer(sb, dev->uuid,
1350                                     (unsigned long)btrfs_device_uuid(dev_item),
1351                                     BTRFS_UUID_SIZE);
1352
1353                 btrfs_set_header_flag(sb, BTRFS_HEADER_FLAG_WRITTEN);
1354                 csum_tree_block(root, sb, 0);
1355
1356                 bh = __getblk(dev->bdev, BTRFS_SUPER_INFO_OFFSET /
1357                               root->fs_info->sb->s_blocksize,
1358                               BTRFS_SUPER_INFO_SIZE);
1359
1360                 read_extent_buffer(sb, bh->b_data, 0, BTRFS_SUPER_INFO_SIZE);
1361                 dev->pending_io = bh;
1362
1363                 get_bh(bh);
1364                 set_buffer_uptodate(bh);
1365                 lock_buffer(bh);
1366                 bh->b_end_io = btrfs_end_buffer_write_sync;
1367
1368                 if (do_barriers && dev->barriers) {
1369                         ret = submit_bh(WRITE_BARRIER, bh);
1370                         if (ret == -EOPNOTSUPP) {
1371                                 printk("btrfs: disabling barriers on dev %s\n",
1372                                        dev->name);
1373                                 set_buffer_uptodate(bh);
1374                                 dev->barriers = 0;
1375                                 get_bh(bh);
1376                                 lock_buffer(bh);
1377                                 ret = submit_bh(WRITE, bh);
1378                         }
1379                 } else {
1380                         ret = submit_bh(WRITE, bh);
1381                 }
1382                 BUG_ON(ret);
1383         }
1384
1385         list_for_each(cur, head) {
1386                 dev = list_entry(cur, struct btrfs_device, dev_list);
1387                 BUG_ON(!dev->pending_io);
1388                 bh = dev->pending_io;
1389                 wait_on_buffer(bh);
1390                 if (!buffer_uptodate(dev->pending_io)) {
1391                         if (do_barriers && dev->barriers) {
1392                                 printk("btrfs: disabling barriers on dev %s\n",
1393                                        dev->name);
1394                                 set_buffer_uptodate(bh);
1395                                 get_bh(bh);
1396                                 lock_buffer(bh);
1397                                 dev->barriers = 0;
1398                                 ret = submit_bh(WRITE, bh);
1399                                 BUG_ON(ret);
1400                                 wait_on_buffer(bh);
1401                                 BUG_ON(!buffer_uptodate(bh));
1402                         } else {
1403                                 BUG();
1404                         }
1405
1406                 }
1407                 dev->pending_io = NULL;
1408                 brelse(bh);
1409         }
1410         return 0;
1411 }
1412
1413 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
1414                       *root)
1415 {
1416         int ret;
1417
1418         ret = write_all_supers(root);
1419 #if 0
1420         if (!btrfs_test_opt(root, NOBARRIER))
1421                 blkdev_issue_flush(sb->s_bdev, NULL);
1422         set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, super);
1423         ret = sync_page_range_nolock(btree_inode, btree_inode->i_mapping,
1424                                      super->start, super->len);
1425         if (!btrfs_test_opt(root, NOBARRIER))
1426                 blkdev_issue_flush(sb->s_bdev, NULL);
1427 #endif
1428         return ret;
1429 }
1430
1431 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
1432 {
1433         radix_tree_delete(&fs_info->fs_roots_radix,
1434                           (unsigned long)root->root_key.objectid);
1435         if (root->in_sysfs)
1436                 btrfs_sysfs_del_root(root);
1437         if (root->inode)
1438                 iput(root->inode);
1439         if (root->node)
1440                 free_extent_buffer(root->node);
1441         if (root->commit_root)
1442                 free_extent_buffer(root->commit_root);
1443         if (root->name)
1444                 kfree(root->name);
1445         kfree(root);
1446         return 0;
1447 }
1448
1449 static int del_fs_roots(struct btrfs_fs_info *fs_info)
1450 {
1451         int ret;
1452         struct btrfs_root *gang[8];
1453         int i;
1454
1455         while(1) {
1456                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1457                                              (void **)gang, 0,
1458                                              ARRAY_SIZE(gang));
1459                 if (!ret)
1460                         break;
1461                 for (i = 0; i < ret; i++)
1462                         btrfs_free_fs_root(fs_info, gang[i]);
1463         }
1464         return 0;
1465 }
1466
1467 int close_ctree(struct btrfs_root *root)
1468 {
1469         int ret;
1470         struct btrfs_trans_handle *trans;
1471         struct btrfs_fs_info *fs_info = root->fs_info;
1472
1473         fs_info->closing = 1;
1474         btrfs_transaction_flush_work(root);
1475         mutex_lock(&fs_info->fs_mutex);
1476         btrfs_defrag_dirty_roots(root->fs_info);
1477         trans = btrfs_start_transaction(root, 1);
1478         ret = btrfs_commit_transaction(trans, root);
1479         /* run commit again to  drop the original snapshot */
1480         trans = btrfs_start_transaction(root, 1);
1481         btrfs_commit_transaction(trans, root);
1482         ret = btrfs_write_and_wait_transaction(NULL, root);
1483         BUG_ON(ret);
1484         write_ctree_super(NULL, root);
1485         mutex_unlock(&fs_info->fs_mutex);
1486
1487         if (fs_info->delalloc_bytes) {
1488                 printk("btrfs: at unmount delalloc count %Lu\n",
1489                        fs_info->delalloc_bytes);
1490         }
1491         if (fs_info->extent_root->node)
1492                 free_extent_buffer(fs_info->extent_root->node);
1493
1494         if (fs_info->tree_root->node)
1495                 free_extent_buffer(fs_info->tree_root->node);
1496
1497         if (root->fs_info->chunk_root->node);
1498                 free_extent_buffer(root->fs_info->chunk_root->node);
1499
1500         if (root->fs_info->dev_root->node);
1501                 free_extent_buffer(root->fs_info->dev_root->node);
1502
1503         free_extent_buffer(fs_info->sb_buffer);
1504
1505         btrfs_free_block_groups(root->fs_info);
1506         del_fs_roots(fs_info);
1507
1508         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1509
1510         extent_io_tree_empty_lru(&fs_info->free_space_cache);
1511         extent_io_tree_empty_lru(&fs_info->block_group_cache);
1512         extent_io_tree_empty_lru(&fs_info->pinned_extents);
1513         extent_io_tree_empty_lru(&fs_info->pending_del);
1514         extent_io_tree_empty_lru(&fs_info->extent_ins);
1515         extent_io_tree_empty_lru(&BTRFS_I(fs_info->btree_inode)->io_tree);
1516
1517         truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
1518         flush_workqueue(end_io_workqueue);
1519         destroy_workqueue(end_io_workqueue);
1520
1521         flush_workqueue(async_submit_workqueue);
1522         destroy_workqueue(async_submit_workqueue);
1523
1524         iput(fs_info->btree_inode);
1525 #if 0
1526         while(!list_empty(&fs_info->hashers)) {
1527                 struct btrfs_hasher *hasher;
1528                 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
1529                                     hashers);
1530                 list_del(&hasher->hashers);
1531                 crypto_free_hash(&fs_info->hash_tfm);
1532                 kfree(hasher);
1533         }
1534 #endif
1535         close_all_devices(fs_info);
1536         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1537
1538 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,23)
1539         bdi_destroy(&fs_info->bdi);
1540 #endif
1541
1542         kfree(fs_info->extent_root);
1543         kfree(fs_info->tree_root);
1544         kfree(fs_info->chunk_root);
1545         kfree(fs_info->dev_root);
1546         return 0;
1547 }
1548
1549 int btrfs_buffer_uptodate(struct extent_buffer *buf)
1550 {
1551         struct inode *btree_inode = buf->first_page->mapping->host;
1552         return extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
1553 }
1554
1555 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
1556 {
1557         struct inode *btree_inode = buf->first_page->mapping->host;
1558         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
1559                                           buf);
1560 }
1561
1562 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
1563 {
1564         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1565         u64 transid = btrfs_header_generation(buf);
1566         struct inode *btree_inode = root->fs_info->btree_inode;
1567
1568         if (transid != root->fs_info->generation) {
1569                 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
1570                         (unsigned long long)buf->start,
1571                         transid, root->fs_info->generation);
1572                 WARN_ON(1);
1573         }
1574         set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
1575 }
1576
1577 void btrfs_throttle(struct btrfs_root *root)
1578 {
1579         struct backing_dev_info *bdi;
1580
1581         bdi = root->fs_info->sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
1582         if (root->fs_info->throttles && bdi_write_congested(bdi)) {
1583 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,18)
1584                 congestion_wait(WRITE, HZ/20);
1585 #else
1586                 blk_congestion_wait(WRITE, HZ/20);
1587 #endif
1588         }
1589 }
1590
1591 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
1592 {
1593         balance_dirty_pages_ratelimited_nr(
1594                                    root->fs_info->btree_inode->i_mapping, 1);
1595 }
1596
1597 void btrfs_set_buffer_defrag(struct extent_buffer *buf)
1598 {
1599         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1600         struct inode *btree_inode = root->fs_info->btree_inode;
1601         set_extent_bits(&BTRFS_I(btree_inode)->io_tree, buf->start,
1602                         buf->start + buf->len - 1, EXTENT_DEFRAG, GFP_NOFS);
1603 }
1604
1605 void btrfs_set_buffer_defrag_done(struct extent_buffer *buf)
1606 {
1607         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1608         struct inode *btree_inode = root->fs_info->btree_inode;
1609         set_extent_bits(&BTRFS_I(btree_inode)->io_tree, buf->start,
1610                         buf->start + buf->len - 1, EXTENT_DEFRAG_DONE,
1611                         GFP_NOFS);
1612 }
1613
1614 int btrfs_buffer_defrag(struct extent_buffer *buf)
1615 {
1616         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1617         struct inode *btree_inode = root->fs_info->btree_inode;
1618         return test_range_bit(&BTRFS_I(btree_inode)->io_tree,
1619                      buf->start, buf->start + buf->len - 1, EXTENT_DEFRAG, 0);
1620 }
1621
1622 int btrfs_buffer_defrag_done(struct extent_buffer *buf)
1623 {
1624         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1625         struct inode *btree_inode = root->fs_info->btree_inode;
1626         return test_range_bit(&BTRFS_I(btree_inode)->io_tree,
1627                      buf->start, buf->start + buf->len - 1,
1628                      EXTENT_DEFRAG_DONE, 0);
1629 }
1630
1631 int btrfs_clear_buffer_defrag_done(struct extent_buffer *buf)
1632 {
1633         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1634         struct inode *btree_inode = root->fs_info->btree_inode;
1635         return clear_extent_bits(&BTRFS_I(btree_inode)->io_tree,
1636                      buf->start, buf->start + buf->len - 1,
1637                      EXTENT_DEFRAG_DONE, GFP_NOFS);
1638 }
1639
1640 int btrfs_clear_buffer_defrag(struct extent_buffer *buf)
1641 {
1642         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1643         struct inode *btree_inode = root->fs_info->btree_inode;
1644         return clear_extent_bits(&BTRFS_I(btree_inode)->io_tree,
1645                      buf->start, buf->start + buf->len - 1,
1646                      EXTENT_DEFRAG, GFP_NOFS);
1647 }
1648
1649 int btrfs_read_buffer(struct extent_buffer *buf)
1650 {
1651         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1652         int ret;
1653         ret = btree_read_extent_buffer_pages(root, buf, 0);
1654         if (ret == 0) {
1655                 buf->flags |= EXTENT_UPTODATE;
1656         }
1657         return ret;
1658 }
1659
1660 static struct extent_io_ops btree_extent_io_ops = {
1661         .writepage_io_hook = btree_writepage_io_hook,
1662         .readpage_end_io_hook = btree_readpage_end_io_hook,
1663         .submit_bio_hook = btree_submit_bio_hook,
1664         /* note we're sharing with inode.c for the merge bio hook */
1665         .merge_bio_hook = btrfs_merge_bio_hook,
1666 };