SAFE public projects git trees. - safe/jmp/linux-2.6/blob - fs/btrfs/disk-io.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/fs.h>
  20 #include <linux/blkdev.h>
  21 #include <linux/scatterlist.h>
  22 #include <linux/swap.h>
  23 #include <linux/radix-tree.h>
  24 #include <linux/writeback.h>
  25 #include <linux/buffer_head.h> // for block_sync_page
  26 #include <linux/workqueue.h>
  27 #include "crc32c.h"
  28 #include "ctree.h"
  29 #include "disk-io.h"
  30 #include "transaction.h"
  31 #include "btrfs_inode.h"
  32 #include "volumes.h"
  33 #include "print-tree.h"
  34
  35 #if 0
  36 static int check_tree_block(struct btrfs_root *root, struct extent_buffer *buf)
  37 {
  38         if (extent_buffer_blocknr(buf) != btrfs_header_blocknr(buf)) {
  39                 printk(KERN_CRIT "buf blocknr(buf) is %llu, header is %llu\n",
  40                        (unsigned long long)extent_buffer_blocknr(buf),
  41                        (unsigned long long)btrfs_header_blocknr(buf));
  42                 return 1;
  43         }
  44         return 0;
  45 }
  46 #endif
  47
  48 static struct extent_io_ops btree_extent_io_ops;
  49 static struct workqueue_struct *end_io_workqueue;
  50 static struct workqueue_struct *async_submit_workqueue;
  51
  52 struct end_io_wq {
  53         struct bio *bio;
  54         bio_end_io_t *end_io;
  55         void *private;
  56         struct btrfs_fs_info *info;
  57         int error;
  58         int metadata;
  59         struct list_head list;
  60 };
  61
  62 struct async_submit_bio {
  63         struct inode *inode;
  64         struct bio *bio;
  65         struct list_head list;
  66         extent_submit_bio_hook_t *submit_bio_hook;
  67         int rw;
  68         int mirror_num;
  69 };
  70
  71 struct extent_map *btree_get_extent(struct inode *inode, struct page *page,
  72                                     size_t page_offset, u64 start, u64 len,
  73                                     int create)
  74 {
  75         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
  76         struct extent_map *em;
  77         int ret;
  78
  79         spin_lock(&em_tree->lock);
  80         em = lookup_extent_mapping(em_tree, start, len);
  81         if (em) {
  82                 em->bdev =
  83                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
  84                 spin_unlock(&em_tree->lock);
  85                 goto out;
  86         }
  87         spin_unlock(&em_tree->lock);
  88
  89         em = alloc_extent_map(GFP_NOFS);
  90         if (!em) {
  91                 em = ERR_PTR(-ENOMEM);
  92                 goto out;
  93         }
  94         em->start = 0;
  95         em->len = (u64)-1;
  96         em->block_start = 0;
  97         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
  98
  99         spin_lock(&em_tree->lock);
 100         ret = add_extent_mapping(em_tree, em);
 101         if (ret == -EEXIST) {
 102                 u64 failed_start = em->start;
 103                 u64 failed_len = em->len;
 104
 105                 printk("failed to insert %Lu %Lu -> %Lu into tree\n",
 106                        em->start, em->len, em->block_start);
 107                 free_extent_map(em);
 108                 em = lookup_extent_mapping(em_tree, start, len);
 109                 if (em) {
 110                         printk("after failing, found %Lu %Lu %Lu\n",
 111                                em->start, em->len, em->block_start);
 112                         ret = 0;
 113                 } else {
 114                         em = lookup_extent_mapping(em_tree, failed_start,
 115                                                    failed_len);
 116                         if (em) {
 117                                 printk("double failure lookup gives us "
 118                                        "%Lu %Lu -> %Lu\n", em->start,
 119                                        em->len, em->block_start);
 120                                 free_extent_map(em);
 121                         }
 122                         ret = -EIO;
 123                 }
 124         } else if (ret) {
 125                 free_extent_map(em);
 126                 em = NULL;
 127         }
 128         spin_unlock(&em_tree->lock);
 129
 130         if (ret)
 131                 em = ERR_PTR(ret);
 132 out:
 133         return em;
 134 }
 135
 136 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
 137 {
 138         return btrfs_crc32c(seed, data, len);
 139 }
 140
 141 void btrfs_csum_final(u32 crc, char *result)
 142 {
 143         *(__le32 *)result = ~cpu_to_le32(crc);
 144 }
 145
 146 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 147                            int verify)
 148 {
 149         char result[BTRFS_CRC32_SIZE];
 150         unsigned long len;
 151         unsigned long cur_len;
 152         unsigned long offset = BTRFS_CSUM_SIZE;
 153         char *map_token = NULL;
 154         char *kaddr;
 155         unsigned long map_start;
 156         unsigned long map_len;
 157         int err;
 158         u32 crc = ~(u32)0;
 159
 160         len = buf->len - offset;
 161         while(len > 0) {
 162                 err = map_private_extent_buffer(buf, offset, 32,
 163                                         &map_token, &kaddr,
 164                                         &map_start, &map_len, KM_USER0);
 165                 if (err) {
 166                         printk("failed to map extent buffer! %lu\n",
 167                                offset);
 168                         return 1;
 169                 }
 170                 cur_len = min(len, map_len - (offset - map_start));
 171                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
 172                                       crc, cur_len);
 173                 len -= cur_len;
 174                 offset += cur_len;
 175                 unmap_extent_buffer(buf, map_token, KM_USER0);
 176         }
 177         btrfs_csum_final(crc, result);
 178
 179         if (verify) {
 180                 int from_this_trans = 0;
 181
 182                 if (root->fs_info->running_transaction &&
 183                     btrfs_header_generation(buf) ==
 184                     root->fs_info->running_transaction->transid)
 185                         from_this_trans = 1;
 186
 187                 /* FIXME, this is not good */
 188                 if (memcmp_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE)) {
 189                         u32 val;
 190                         u32 found = 0;
 191                         memcpy(&found, result, BTRFS_CRC32_SIZE);
 192
 193                         read_extent_buffer(buf, &val, 0, BTRFS_CRC32_SIZE);
 194                         printk("btrfs: %s checksum verify failed on %llu "
 195                                "wanted %X found %X from_this_trans %d "
 196                                "level %d\n",
 197                                root->fs_info->sb->s_id,
 198                                buf->start, val, found, from_this_trans,
 199                                btrfs_header_level(buf));
 200                         return 1;
 201                 }
 202         } else {
 203                 write_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE);
 204         }
 205         return 0;
 206 }
 207
 208 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 209                                           struct extent_buffer *eb,
 210                                           u64 start)
 211 {
 212         struct extent_io_tree *io_tree;
 213         int ret;
 214         int num_copies = 0;
 215         int mirror_num = 0;
 216
 217         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 218         while (1) {
 219                 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
 220                                                btree_get_extent, mirror_num);
 221                 if (!ret)
 222                         return ret;
 223
 224                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
 225                                               eb->start, eb->len);
 226                 if (num_copies == 1)
 227                         return ret;
 228
 229                 mirror_num++;
 230                 if (mirror_num > num_copies)
 231                         return ret;
 232         }
 233         return -EIO;
 234 }
 235
 236 int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 237 {
 238         struct extent_io_tree *tree;
 239         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
 240         u64 found_start;
 241         int found_level;
 242         unsigned long len;
 243         struct extent_buffer *eb;
 244         int ret;
 245
 246         tree = &BTRFS_I(page->mapping->host)->io_tree;
 247
 248         if (page->private == EXTENT_PAGE_PRIVATE)
 249                 goto out;
 250         if (!page->private)
 251                 goto out;
 252         len = page->private >> 2;
 253         if (len == 0) {
 254                 WARN_ON(1);
 255         }
 256         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 257         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE);
 258         BUG_ON(ret);
 259         btrfs_clear_buffer_defrag(eb);
 260         found_start = btrfs_header_bytenr(eb);
 261         if (found_start != start) {
 262                 printk("warning: eb start incorrect %Lu buffer %Lu len %lu\n",
 263                        start, found_start, len);
 264                 WARN_ON(1);
 265                 goto err;
 266         }
 267         if (eb->first_page != page) {
 268                 printk("bad first page %lu %lu\n", eb->first_page->index,
 269                        page->index);
 270                 WARN_ON(1);
 271                 goto err;
 272         }
 273         if (!PageUptodate(page)) {
 274                 printk("csum not up to date page %lu\n", page->index);
 275                 WARN_ON(1);
 276                 goto err;
 277         }
 278         found_level = btrfs_header_level(eb);
 279         spin_lock(&root->fs_info->hash_lock);
 280         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
 281         spin_unlock(&root->fs_info->hash_lock);
 282         csum_tree_block(root, eb, 0);
 283 err:
 284         free_extent_buffer(eb);
 285 out:
 286         return 0;
 287 }
 288
 289 static int btree_writepage_io_hook(struct page *page, u64 start, u64 end)
 290 {
 291         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 292
 293         csum_dirty_buffer(root, page);
 294         return 0;
 295 }
 296
 297 int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 298                                struct extent_state *state)
 299 {
 300         struct extent_io_tree *tree;
 301         u64 found_start;
 302         int found_level;
 303         unsigned long len;
 304         struct extent_buffer *eb;
 305         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 306         int ret = 0;
 307
 308         tree = &BTRFS_I(page->mapping->host)->io_tree;
 309         if (page->private == EXTENT_PAGE_PRIVATE)
 310                 goto out;
 311         if (!page->private)
 312                 goto out;
 313         len = page->private >> 2;
 314         if (len == 0) {
 315                 WARN_ON(1);
 316         }
 317         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 318
 319         btrfs_clear_buffer_defrag(eb);
 320         found_start = btrfs_header_bytenr(eb);
 321         if (found_start != start) {
 322                 ret = -EIO;
 323                 goto err;
 324         }
 325         if (eb->first_page != page) {
 326                 printk("bad first page %lu %lu\n", eb->first_page->index,
 327                        page->index);
 328                 WARN_ON(1);
 329                 ret = -EIO;
 330                 goto err;
 331         }
 332         found_level = btrfs_header_level(eb);
 333
 334         ret = csum_tree_block(root, eb, 1);
 335         if (ret)
 336                 ret = -EIO;
 337
 338         end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
 339         end = eb->start + end - 1;
 340         release_extent_buffer_tail_pages(eb);
 341 err:
 342         free_extent_buffer(eb);
 343 out:
 344         return ret;
 345 }
 346
 347 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
 348 static void end_workqueue_bio(struct bio *bio, int err)
 349 #else
 350 static int end_workqueue_bio(struct bio *bio,
 351                                    unsigned int bytes_done, int err)
 352 #endif
 353 {
 354         struct end_io_wq *end_io_wq = bio->bi_private;
 355         struct btrfs_fs_info *fs_info;
 356         unsigned long flags;
 357
 358 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
 359         if (bio->bi_size)
 360                 return 1;
 361 #endif
 362
 363         fs_info = end_io_wq->info;
 364         spin_lock_irqsave(&fs_info->end_io_work_lock, flags);
 365         end_io_wq->error = err;
 366         list_add_tail(&end_io_wq->list, &fs_info->end_io_work_list);
 367         spin_unlock_irqrestore(&fs_info->end_io_work_lock, flags);
 368         queue_work(end_io_workqueue, &fs_info->end_io_work);
 369
 370 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
 371         return 0;
 372 #endif
 373 }
 374
 375 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 376                         int metadata)
 377 {
 378         struct end_io_wq *end_io_wq;
 379         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
 380         if (!end_io_wq)
 381                 return -ENOMEM;
 382
 383         end_io_wq->private = bio->bi_private;
 384         end_io_wq->end_io = bio->bi_end_io;
 385         end_io_wq->info = info;
 386         end_io_wq->error = 0;
 387         end_io_wq->bio = bio;
 388         end_io_wq->metadata = metadata;
 389
 390         bio->bi_private = end_io_wq;
 391         bio->bi_end_io = end_workqueue_bio;
 392         return 0;
 393 }
 394
 395 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 396                         int rw, struct bio *bio, int mirror_num,
 397                         extent_submit_bio_hook_t *submit_bio_hook)
 398 {
 399         struct async_submit_bio *async;
 400
 401         /*
 402          * inline writerback should stay inline, only hop to the async
 403          * queue if we're pdflush
 404          */
 405         if (!current_is_pdflush())
 406                 return submit_bio_hook(inode, rw, bio, mirror_num);
 407
 408         async = kmalloc(sizeof(*async), GFP_NOFS);
 409         if (!async)
 410                 return -ENOMEM;
 411
 412         async->inode = inode;
 413         async->rw = rw;
 414         async->bio = bio;
 415         async->mirror_num = mirror_num;
 416         async->submit_bio_hook = submit_bio_hook;
 417
 418         spin_lock(&fs_info->async_submit_work_lock);
 419         list_add_tail(&async->list, &fs_info->async_submit_work_list);
 420         spin_unlock(&fs_info->async_submit_work_lock);
 421
 422         queue_work(async_submit_workqueue, &fs_info->async_submit_work);
 423         return 0;
 424 }
 425
 426 static int __btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 427                                  int mirror_num)
 428 {
 429         struct btrfs_root *root = BTRFS_I(inode)->root;
 430         u64 offset;
 431         int ret;
 432
 433         offset = bio->bi_sector << 9;
 434
 435         if (rw & (1 << BIO_RW)) {
 436                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num);
 437         }
 438
 439         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 1);
 440         BUG_ON(ret);
 441
 442         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num);
 443 }
 444
 445 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 446                                  int mirror_num)
 447 {
 448         if (!(rw & (1 << BIO_RW))) {
 449                 return __btree_submit_bio_hook(inode, rw, bio, mirror_num);
 450         }
 451         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 452                                    inode, rw, bio, mirror_num,
 453                                    __btree_submit_bio_hook);
 454 }
 455
 456 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 457 {
 458         struct extent_io_tree *tree;
 459         tree = &BTRFS_I(page->mapping->host)->io_tree;
 460         return extent_write_full_page(tree, page, btree_get_extent, wbc);
 461 }
 462
 463 static int btree_writepages(struct address_space *mapping,
 464                             struct writeback_control *wbc)
 465 {
 466         struct extent_io_tree *tree;
 467         tree = &BTRFS_I(mapping->host)->io_tree;
 468         if (wbc->sync_mode == WB_SYNC_NONE) {
 469                 u64 num_dirty;
 470                 u64 start = 0;
 471                 unsigned long thresh = 96 * 1024 * 1024;
 472
 473                 if (wbc->for_kupdate)
 474                         return 0;
 475
 476                 if (current_is_pdflush()) {
 477                         thresh = 96 * 1024 * 1024;
 478                 } else {
 479                         thresh = 8 * 1024 * 1024;
 480                 }
 481                 num_dirty = count_range_bits(tree, &start, (u64)-1,
 482                                              thresh, EXTENT_DIRTY);
 483                 if (num_dirty < thresh) {
 484                         return 0;
 485                 }
 486         }
 487         return extent_writepages(tree, mapping, btree_get_extent, wbc);
 488 }
 489
 490 int btree_readpage(struct file *file, struct page *page)
 491 {
 492         struct extent_io_tree *tree;
 493         tree = &BTRFS_I(page->mapping->host)->io_tree;
 494         return extent_read_full_page(tree, page, btree_get_extent);
 495 }
 496
 497 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 498 {
 499         struct extent_io_tree *tree;
 500         struct extent_map_tree *map;
 501         int ret;
 502
 503         if (page_count(page) > 3) {
 504                 /* once for page->private, once for the caller, once
 505                  * once for the page cache
 506                  */
 507                 return 0;
 508         }
 509         tree = &BTRFS_I(page->mapping->host)->io_tree;
 510         map = &BTRFS_I(page->mapping->host)->extent_tree;
 511         ret = try_release_extent_state(map, tree, page, gfp_flags);
 512         if (ret == 1) {
 513                 invalidate_extent_lru(tree, page_offset(page), PAGE_CACHE_SIZE);
 514                 ClearPagePrivate(page);
 515                 set_page_private(page, 0);
 516                 page_cache_release(page);
 517         }
 518         return ret;
 519 }
 520
 521 static void btree_invalidatepage(struct page *page, unsigned long offset)
 522 {
 523         struct extent_io_tree *tree;
 524         tree = &BTRFS_I(page->mapping->host)->io_tree;
 525         extent_invalidatepage(tree, page, offset);
 526         btree_releasepage(page, GFP_NOFS);
 527         if (PagePrivate(page)) {
 528                 invalidate_extent_lru(tree, page_offset(page), PAGE_CACHE_SIZE);
 529                 ClearPagePrivate(page);
 530                 set_page_private(page, 0);
 531                 page_cache_release(page);
 532         }
 533 }
 534
 535 #if 0
 536 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 537 {
 538         struct buffer_head *bh;
 539         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 540         struct buffer_head *head;
 541         if (!page_has_buffers(page)) {
 542                 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
 543                                         (1 << BH_Dirty)|(1 << BH_Uptodate));
 544         }
 545         head = page_buffers(page);
 546         bh = head;
 547         do {
 548                 if (buffer_dirty(bh))
 549                         csum_tree_block(root, bh, 0);
 550                 bh = bh->b_this_page;
 551         } while (bh != head);
 552         return block_write_full_page(page, btree_get_block, wbc);
 553 }
 554 #endif
 555
 556 static struct address_space_operations btree_aops = {
 557         .readpage       = btree_readpage,
 558         .writepage      = btree_writepage,
 559         .writepages     = btree_writepages,
 560         .releasepage    = btree_releasepage,
 561         .invalidatepage = btree_invalidatepage,
 562         .sync_page      = block_sync_page,
 563 };
 564
 565 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize)
 566 {
 567         struct extent_buffer *buf = NULL;
 568         struct inode *btree_inode = root->fs_info->btree_inode;
 569         int ret = 0;
 570
 571         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 572         if (!buf)
 573                 return 0;
 574         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
 575                                  buf, 0, 0, btree_get_extent, 0);
 576         free_extent_buffer(buf);
 577         return ret;
 578 }
 579
 580 static int close_all_devices(struct btrfs_fs_info *fs_info)
 581 {
 582         struct list_head *list;
 583         struct list_head *next;
 584         struct btrfs_device *device;
 585
 586         list = &fs_info->fs_devices->devices;
 587         list_for_each(next, list) {
 588                 device = list_entry(next, struct btrfs_device, dev_list);
 589                 close_bdev_excl(device->bdev);
 590                 device->bdev = NULL;
 591         }
 592         return 0;
 593 }
 594
 595 int btrfs_verify_block_csum(struct btrfs_root *root,
 596                             struct extent_buffer *buf)
 597 {
 598         return btrfs_buffer_uptodate(buf);
 599 }
 600
 601 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
 602                                             u64 bytenr, u32 blocksize)
 603 {
 604         struct inode *btree_inode = root->fs_info->btree_inode;
 605         struct extent_buffer *eb;
 606         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 607                                 bytenr, blocksize, GFP_NOFS);
 608         return eb;
 609 }
 610
 611 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
 612                                                  u64 bytenr, u32 blocksize)
 613 {
 614         struct inode *btree_inode = root->fs_info->btree_inode;
 615         struct extent_buffer *eb;
 616
 617         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 618                                  bytenr, blocksize, NULL, GFP_NOFS);
 619         return eb;
 620 }
 621
 622
 623 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
 624                                       u32 blocksize)
 625 {
 626         struct extent_buffer *buf = NULL;
 627         struct inode *btree_inode = root->fs_info->btree_inode;
 628         struct extent_io_tree *io_tree;
 629         int ret;
 630
 631         io_tree = &BTRFS_I(btree_inode)->io_tree;
 632
 633         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 634         if (!buf)
 635                 return NULL;
 636
 637         ret = btree_read_extent_buffer_pages(root, buf, 0);
 638
 639         if (ret == 0) {
 640                 buf->flags |= EXTENT_UPTODATE;
 641         }
 642         return buf;
 643
 644 }
 645
 646 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 647                      struct extent_buffer *buf)
 648 {
 649         struct inode *btree_inode = root->fs_info->btree_inode;
 650         if (btrfs_header_generation(buf) ==
 651             root->fs_info->running_transaction->transid)
 652                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
 653                                           buf);
 654         return 0;
 655 }
 656
 657 int wait_on_tree_block_writeback(struct btrfs_root *root,
 658                                  struct extent_buffer *buf)
 659 {
 660         struct inode *btree_inode = root->fs_info->btree_inode;
 661         wait_on_extent_buffer_writeback(&BTRFS_I(btree_inode)->io_tree,
 662                                         buf);
 663         return 0;
 664 }
 665
 666 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
 667                         u32 stripesize, struct btrfs_root *root,
 668                         struct btrfs_fs_info *fs_info,
 669                         u64 objectid)
 670 {
 671         root->node = NULL;
 672         root->inode = NULL;
 673         root->commit_root = NULL;
 674         root->sectorsize = sectorsize;
 675         root->nodesize = nodesize;
 676         root->leafsize = leafsize;
 677         root->stripesize = stripesize;
 678         root->ref_cows = 0;
 679         root->track_dirty = 0;
 680
 681         root->fs_info = fs_info;
 682         root->objectid = objectid;
 683         root->last_trans = 0;
 684         root->highest_inode = 0;
 685         root->last_inode_alloc = 0;
 686         root->name = NULL;
 687         root->in_sysfs = 0;
 688
 689         INIT_LIST_HEAD(&root->dirty_list);
 690         memset(&root->root_key, 0, sizeof(root->root_key));
 691         memset(&root->root_item, 0, sizeof(root->root_item));
 692         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
 693         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
 694         init_completion(&root->kobj_unregister);
 695         root->defrag_running = 0;
 696         root->defrag_level = 0;
 697         root->root_key.objectid = objectid;
 698         return 0;
 699 }
 700
 701 static int find_and_setup_root(struct btrfs_root *tree_root,
 702                                struct btrfs_fs_info *fs_info,
 703                                u64 objectid,
 704                                struct btrfs_root *root)
 705 {
 706         int ret;
 707         u32 blocksize;
 708
 709         __setup_root(tree_root->nodesize, tree_root->leafsize,
 710                      tree_root->sectorsize, tree_root->stripesize,
 711                      root, fs_info, objectid);
 712         ret = btrfs_find_last_root(tree_root, objectid,
 713                                    &root->root_item, &root->root_key);
 714         BUG_ON(ret);
 715
 716         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 717         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 718                                      blocksize);
 719         BUG_ON(!root->node);
 720         return 0;
 721 }
 722
 723 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_fs_info *fs_info,
 724                                                struct btrfs_key *location)
 725 {
 726         struct btrfs_root *root;
 727         struct btrfs_root *tree_root = fs_info->tree_root;
 728         struct btrfs_path *path;
 729         struct extent_buffer *l;
 730         u64 highest_inode;
 731         u32 blocksize;
 732         int ret = 0;
 733
 734         root = kzalloc(sizeof(*root), GFP_NOFS);
 735         if (!root)
 736                 return ERR_PTR(-ENOMEM);
 737         if (location->offset == (u64)-1) {
 738                 ret = find_and_setup_root(tree_root, fs_info,
 739                                           location->objectid, root);
 740                 if (ret) {
 741                         kfree(root);
 742                         return ERR_PTR(ret);
 743                 }
 744                 goto insert;
 745         }
 746
 747         __setup_root(tree_root->nodesize, tree_root->leafsize,
 748                      tree_root->sectorsize, tree_root->stripesize,
 749                      root, fs_info, location->objectid);
 750
 751         path = btrfs_alloc_path();
 752         BUG_ON(!path);
 753         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
 754         if (ret != 0) {
 755                 if (ret > 0)
 756                         ret = -ENOENT;
 757                 goto out;
 758         }
 759         l = path->nodes[0];
 760         read_extent_buffer(l, &root->root_item,
 761                btrfs_item_ptr_offset(l, path->slots[0]),
 762                sizeof(root->root_item));
 763         memcpy(&root->root_key, location, sizeof(*location));
 764         ret = 0;
 765 out:
 766         btrfs_release_path(root, path);
 767         btrfs_free_path(path);
 768         if (ret) {
 769                 kfree(root);
 770                 return ERR_PTR(ret);
 771         }
 772         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 773         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 774                                      blocksize);
 775         BUG_ON(!root->node);
 776 insert:
 777         root->ref_cows = 1;
 778         ret = btrfs_find_highest_inode(root, &highest_inode);
 779         if (ret == 0) {
 780                 root->highest_inode = highest_inode;
 781                 root->last_inode_alloc = highest_inode;
 782         }
 783         return root;
 784 }
 785
 786 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
 787                                         u64 root_objectid)
 788 {
 789         struct btrfs_root *root;
 790
 791         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
 792                 return fs_info->tree_root;
 793         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
 794                 return fs_info->extent_root;
 795
 796         root = radix_tree_lookup(&fs_info->fs_roots_radix,
 797                                  (unsigned long)root_objectid);
 798         return root;
 799 }
 800
 801 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
 802                                               struct btrfs_key *location)
 803 {
 804         struct btrfs_root *root;
 805         int ret;
 806
 807         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
 808                 return fs_info->tree_root;
 809         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
 810                 return fs_info->extent_root;
 811         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
 812                 return fs_info->chunk_root;
 813         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
 814                 return fs_info->dev_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 /*
 909  * this unplugs every device on the box, and it is only used when page
 910  * is null
 911  */
 912 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
 913 {
 914         struct list_head *cur;
 915         struct btrfs_device *device;
 916         struct btrfs_fs_info *info;
 917
 918         info = (struct btrfs_fs_info *)bdi->unplug_io_data;
 919         list_for_each(cur, &info->fs_devices->devices) {
 920                 device = list_entry(cur, struct btrfs_device, dev_list);
 921                 bdi = blk_get_backing_dev_info(device->bdev);
 922                 if (bdi->unplug_io_fn) {
 923                         bdi->unplug_io_fn(bdi, page);
 924                 }
 925         }
 926 }
 927
 928 void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
 929 {
 930         struct inode *inode;
 931         struct extent_map_tree *em_tree;
 932         struct extent_map *em;
 933         struct address_space *mapping;
 934         u64 offset;
 935
 936         /* the generic O_DIRECT read code does this */
 937         if (!page) {
 938                 __unplug_io_fn(bdi, page);
 939                 return;
 940         }
 941
 942         /*
 943          * page->mapping may change at any time.  Get a consistent copy
 944          * and use that for everything below
 945          */
 946         smp_mb();
 947         mapping = page->mapping;
 948         if (!mapping)
 949                 return;
 950
 951         inode = mapping->host;
 952         offset = page_offset(page);
 953
 954         em_tree = &BTRFS_I(inode)->extent_tree;
 955         spin_lock(&em_tree->lock);
 956         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
 957         spin_unlock(&em_tree->lock);
 958         if (!em)
 959                 return;
 960
 961         offset = offset - em->start;
 962         btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
 963                           em->block_start + offset, page);
 964         free_extent_map(em);
 965 }
 966
 967 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
 968 {
 969 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,23)
 970         bdi_init(bdi);
 971 #endif
 972         bdi->ra_pages   = default_backing_dev_info.ra_pages;
 973         bdi->state              = 0;
 974         bdi->capabilities       = default_backing_dev_info.capabilities;
 975         bdi->unplug_io_fn       = btrfs_unplug_io_fn;
 976         bdi->unplug_io_data     = info;
 977         bdi->congested_fn       = btrfs_congested_fn;
 978         bdi->congested_data     = info;
 979         return 0;
 980 }
 981
 982 static int bio_ready_for_csum(struct bio *bio)
 983 {
 984         u64 length = 0;
 985         u64 buf_len = 0;
 986         u64 start = 0;
 987         struct page *page;
 988         struct extent_io_tree *io_tree = NULL;
 989         struct btrfs_fs_info *info = NULL;
 990         struct bio_vec *bvec;
 991         int i;
 992         int ret;
 993
 994         bio_for_each_segment(bvec, bio, i) {
 995                 page = bvec->bv_page;
 996                 if (page->private == EXTENT_PAGE_PRIVATE) {
 997                         length += bvec->bv_len;
 998                         continue;
 999                 }
1000                 if (!page->private) {
1001                         length += bvec->bv_len;
1002                         continue;
1003                 }
1004                 length = bvec->bv_len;
1005                 buf_len = page->private >> 2;
1006                 start = page_offset(page) + bvec->bv_offset;
1007                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1008                 info = BTRFS_I(page->mapping->host)->root->fs_info;
1009         }
1010         /* are we fully contained in this bio? */
1011         if (buf_len <= length)
1012                 return 1;
1013
1014         ret = extent_range_uptodate(io_tree, start + length,
1015                                     start + buf_len - 1);
1016         if (ret == 1)
1017                 return ret;
1018         return ret;
1019 }
1020
1021 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1022 static void btrfs_end_io_csum(void *p)
1023 #else
1024 static void btrfs_end_io_csum(struct work_struct *work)
1025 #endif
1026 {
1027 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1028         struct btrfs_fs_info *fs_info = p;
1029 #else
1030         struct btrfs_fs_info *fs_info = container_of(work,
1031                                                      struct btrfs_fs_info,
1032                                                      end_io_work);
1033 #endif
1034         unsigned long flags;
1035         struct end_io_wq *end_io_wq;
1036         struct bio *bio;
1037         struct list_head *next;
1038         int error;
1039         int was_empty;
1040
1041         while(1) {
1042                 spin_lock_irqsave(&fs_info->end_io_work_lock, flags);
1043                 if (list_empty(&fs_info->end_io_work_list)) {
1044                         spin_unlock_irqrestore(&fs_info->end_io_work_lock,
1045                                                flags);
1046                         return;
1047                 }
1048                 next = fs_info->end_io_work_list.next;
1049                 list_del(next);
1050                 spin_unlock_irqrestore(&fs_info->end_io_work_lock, flags);
1051
1052                 end_io_wq = list_entry(next, struct end_io_wq, list);
1053
1054                 bio = end_io_wq->bio;
1055                 if (end_io_wq->metadata && !bio_ready_for_csum(bio)) {
1056                         spin_lock_irqsave(&fs_info->end_io_work_lock, flags);
1057                         was_empty = list_empty(&fs_info->end_io_work_list);
1058                         list_add_tail(&end_io_wq->list,
1059                                       &fs_info->end_io_work_list);
1060                         spin_unlock_irqrestore(&fs_info->end_io_work_lock,
1061                                                flags);
1062                         if (was_empty)
1063                                 return;
1064                         continue;
1065                 }
1066                 error = end_io_wq->error;
1067                 bio->bi_private = end_io_wq->private;
1068                 bio->bi_end_io = end_io_wq->end_io;
1069                 kfree(end_io_wq);
1070 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1071                 bio_endio(bio, bio->bi_size, error);
1072 #else
1073                 bio_endio(bio, error);
1074 #endif
1075         }
1076 }
1077
1078 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1079 static void btrfs_async_submit_work(void *p)
1080 #else
1081 static void btrfs_async_submit_work(struct work_struct *work)
1082 #endif
1083 {
1084 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1085         struct btrfs_fs_info *fs_info = p;
1086 #else
1087         struct btrfs_fs_info *fs_info = container_of(work,
1088                                                      struct btrfs_fs_info,
1089                                                      async_submit_work);
1090 #endif
1091         struct async_submit_bio *async;
1092         struct list_head *next;
1093
1094         while(1) {
1095                 spin_lock(&fs_info->async_submit_work_lock);
1096                 if (list_empty(&fs_info->async_submit_work_list)) {
1097                         spin_unlock(&fs_info->async_submit_work_lock);
1098                         return;
1099                 }
1100                 next = fs_info->async_submit_work_list.next;
1101                 list_del(next);
1102                 spin_unlock(&fs_info->async_submit_work_lock);
1103
1104                 async = list_entry(next, struct async_submit_bio, list);
1105                 async->submit_bio_hook(async->inode, async->rw, async->bio,
1106                                        async->mirror_num);
1107                 kfree(async);
1108         }
1109 }
1110
1111 struct btrfs_root *open_ctree(struct super_block *sb,
1112                               struct btrfs_fs_devices *fs_devices)
1113 {
1114         u32 sectorsize;
1115         u32 nodesize;
1116         u32 leafsize;
1117         u32 blocksize;
1118         u32 stripesize;
1119         struct buffer_head *bh;
1120         struct btrfs_root *extent_root = kmalloc(sizeof(struct btrfs_root),
1121                                                  GFP_NOFS);
1122         struct btrfs_root *tree_root = kmalloc(sizeof(struct btrfs_root),
1123                                                GFP_NOFS);
1124         struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1125                                                 GFP_NOFS);
1126         struct btrfs_root *chunk_root = kmalloc(sizeof(struct btrfs_root),
1127                                                 GFP_NOFS);
1128         struct btrfs_root *dev_root = kmalloc(sizeof(struct btrfs_root),
1129                                               GFP_NOFS);
1130         int ret;
1131         int err = -EINVAL;
1132         struct btrfs_super_block *disk_super;
1133
1134         if (!extent_root || !tree_root || !fs_info) {
1135                 err = -ENOMEM;
1136                 goto fail;
1137         }
1138         end_io_workqueue = create_workqueue("btrfs-end-io");
1139         BUG_ON(!end_io_workqueue);
1140         async_submit_workqueue = create_workqueue("btrfs-async-submit");
1141
1142         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1143         INIT_LIST_HEAD(&fs_info->trans_list);
1144         INIT_LIST_HEAD(&fs_info->dead_roots);
1145         INIT_LIST_HEAD(&fs_info->hashers);
1146         INIT_LIST_HEAD(&fs_info->end_io_work_list);
1147         INIT_LIST_HEAD(&fs_info->async_submit_work_list);
1148         spin_lock_init(&fs_info->hash_lock);
1149         spin_lock_init(&fs_info->end_io_work_lock);
1150         spin_lock_init(&fs_info->async_submit_work_lock);
1151         spin_lock_init(&fs_info->delalloc_lock);
1152         spin_lock_init(&fs_info->new_trans_lock);
1153
1154         init_completion(&fs_info->kobj_unregister);
1155         fs_info->tree_root = tree_root;
1156         fs_info->extent_root = extent_root;
1157         fs_info->chunk_root = chunk_root;
1158         fs_info->dev_root = dev_root;
1159         fs_info->fs_devices = fs_devices;
1160         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1161         INIT_LIST_HEAD(&fs_info->space_info);
1162         btrfs_mapping_init(&fs_info->mapping_tree);
1163         fs_info->sb = sb;
1164         fs_info->max_extent = (u64)-1;
1165         fs_info->max_inline = 8192 * 1024;
1166         setup_bdi(fs_info, &fs_info->bdi);
1167         fs_info->btree_inode = new_inode(sb);
1168         fs_info->btree_inode->i_ino = 1;
1169         fs_info->btree_inode->i_nlink = 1;
1170
1171         sb->s_blocksize = 4096;
1172         sb->s_blocksize_bits = blksize_bits(4096);
1173
1174         /*
1175          * we set the i_size on the btree inode to the max possible int.
1176          * the real end of the address space is determined by all of
1177          * the devices in the system
1178          */
1179         fs_info->btree_inode->i_size = OFFSET_MAX;
1180         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1181         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1182
1183         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1184                              fs_info->btree_inode->i_mapping,
1185                              GFP_NOFS);
1186         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1187                              GFP_NOFS);
1188
1189         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1190
1191         extent_io_tree_init(&fs_info->free_space_cache,
1192                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1193         extent_io_tree_init(&fs_info->block_group_cache,
1194                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1195         extent_io_tree_init(&fs_info->pinned_extents,
1196                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1197         extent_io_tree_init(&fs_info->pending_del,
1198                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1199         extent_io_tree_init(&fs_info->extent_ins,
1200                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1201         fs_info->do_barriers = 1;
1202
1203 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1204         INIT_WORK(&fs_info->end_io_work, btrfs_end_io_csum, fs_info);
1205         INIT_WORK(&fs_info->async_submit_work, btrfs_async_submit_work,
1206                   fs_info);
1207         INIT_WORK(&fs_info->trans_work, btrfs_transaction_cleaner, fs_info);
1208 #else
1209         INIT_WORK(&fs_info->end_io_work, btrfs_end_io_csum);
1210         INIT_WORK(&fs_info->async_submit_work, btrfs_async_submit_work);
1211         INIT_DELAYED_WORK(&fs_info->trans_work, btrfs_transaction_cleaner);
1212 #endif
1213         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1214         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1215                sizeof(struct btrfs_key));
1216         insert_inode_hash(fs_info->btree_inode);
1217         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1218
1219         mutex_init(&fs_info->trans_mutex);
1220         mutex_init(&fs_info->fs_mutex);
1221
1222 #if 0
1223         ret = add_hasher(fs_info, "crc32c");
1224         if (ret) {
1225                 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
1226                 err = -ENOMEM;
1227                 goto fail_iput;
1228         }
1229 #endif
1230         __setup_root(4096, 4096, 4096, 4096, tree_root,
1231                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1232
1233
1234         bh = __bread(fs_devices->latest_bdev,
1235                      BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
1236         if (!bh)
1237                 goto fail_iput;
1238
1239         memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1240         brelse(bh);
1241
1242         memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1243
1244         disk_super = &fs_info->super_copy;
1245         if (!btrfs_super_root(disk_super))
1246                 goto fail_sb_buffer;
1247
1248         if (btrfs_super_num_devices(disk_super) != fs_devices->num_devices) {
1249                 printk("Btrfs: wanted %llu devices, but found %llu\n",
1250                        (unsigned long long)btrfs_super_num_devices(disk_super),
1251                        (unsigned long long)fs_devices->num_devices);
1252                 goto fail_sb_buffer;
1253         }
1254         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1255
1256         nodesize = btrfs_super_nodesize(disk_super);
1257         leafsize = btrfs_super_leafsize(disk_super);
1258         sectorsize = btrfs_super_sectorsize(disk_super);
1259         stripesize = btrfs_super_stripesize(disk_super);
1260         tree_root->nodesize = nodesize;
1261         tree_root->leafsize = leafsize;
1262         tree_root->sectorsize = sectorsize;
1263         tree_root->stripesize = stripesize;
1264
1265         sb->s_blocksize = sectorsize;
1266         sb->s_blocksize_bits = blksize_bits(sectorsize);
1267
1268         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1269                     sizeof(disk_super->magic))) {
1270                 printk("btrfs: valid FS not found on %s\n", sb->s_id);
1271                 goto fail_sb_buffer;
1272         }
1273
1274         mutex_lock(&fs_info->fs_mutex);
1275
1276         ret = btrfs_read_sys_array(tree_root);
1277         if (ret) {
1278                 printk("btrfs: failed to read the system array on %s\n",
1279                        sb->s_id);
1280                 goto fail_sys_array;
1281         }
1282
1283         blocksize = btrfs_level_size(tree_root,
1284                                      btrfs_super_chunk_root_level(disk_super));
1285
1286         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1287                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1288
1289         chunk_root->node = read_tree_block(chunk_root,
1290                                            btrfs_super_chunk_root(disk_super),
1291                                            blocksize);
1292         BUG_ON(!chunk_root->node);
1293
1294         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1295                  (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1296                  BTRFS_UUID_SIZE);
1297
1298         ret = btrfs_read_chunk_tree(chunk_root);
1299         BUG_ON(ret);
1300
1301         blocksize = btrfs_level_size(tree_root,
1302                                      btrfs_super_root_level(disk_super));
1303
1304
1305         tree_root->node = read_tree_block(tree_root,
1306                                           btrfs_super_root(disk_super),
1307                                           blocksize);
1308         if (!tree_root->node)
1309                 goto fail_sb_buffer;
1310
1311
1312         ret = find_and_setup_root(tree_root, fs_info,
1313                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1314         if (ret)
1315                 goto fail_tree_root;
1316         extent_root->track_dirty = 1;
1317
1318         ret = find_and_setup_root(tree_root, fs_info,
1319                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
1320         dev_root->track_dirty = 1;
1321
1322         if (ret)
1323                 goto fail_extent_root;
1324
1325         btrfs_read_block_groups(extent_root);
1326
1327         fs_info->generation = btrfs_super_generation(disk_super) + 1;
1328         fs_info->data_alloc_profile = (u64)-1;
1329         fs_info->metadata_alloc_profile = (u64)-1;
1330         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1331
1332         mutex_unlock(&fs_info->fs_mutex);
1333         return tree_root;
1334
1335 fail_extent_root:
1336         free_extent_buffer(extent_root->node);
1337 fail_tree_root:
1338         free_extent_buffer(tree_root->node);
1339 fail_sys_array:
1340         mutex_unlock(&fs_info->fs_mutex);
1341 fail_sb_buffer:
1342         extent_io_tree_empty_lru(&BTRFS_I(fs_info->btree_inode)->io_tree);
1343 fail_iput:
1344         iput(fs_info->btree_inode);
1345 fail:
1346         close_all_devices(fs_info);
1347         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1348
1349         kfree(extent_root);
1350         kfree(tree_root);
1351 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,23)
1352         bdi_destroy(&fs_info->bdi);
1353 #endif
1354         kfree(fs_info);
1355         return ERR_PTR(err);
1356 }
1357
1358 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1359 {
1360         char b[BDEVNAME_SIZE];
1361
1362         if (uptodate) {
1363                 set_buffer_uptodate(bh);
1364         } else {
1365                 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1366                         printk(KERN_WARNING "lost page write due to "
1367                                         "I/O error on %s\n",
1368                                        bdevname(bh->b_bdev, b));
1369                 }
1370                 set_buffer_write_io_error(bh);
1371                 clear_buffer_uptodate(bh);
1372         }
1373         unlock_buffer(bh);
1374         put_bh(bh);
1375 }
1376
1377 int write_all_supers(struct btrfs_root *root)
1378 {
1379         struct list_head *cur;
1380         struct list_head *head = &root->fs_info->fs_devices->devices;
1381         struct btrfs_device *dev;
1382         struct btrfs_super_block *sb;
1383         struct btrfs_dev_item *dev_item;
1384         struct buffer_head *bh;
1385         int ret;
1386         int do_barriers;
1387         int max_errors;
1388         int total_errors = 0;
1389         u32 crc;
1390         u64 flags;
1391
1392         max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1393         do_barriers = !btrfs_test_opt(root, NOBARRIER);
1394
1395         sb = &root->fs_info->super_for_commit;
1396         dev_item = &sb->dev_item;
1397         list_for_each(cur, head) {
1398                 dev = list_entry(cur, struct btrfs_device, dev_list);
1399                 btrfs_set_stack_device_type(dev_item, dev->type);
1400                 btrfs_set_stack_device_id(dev_item, dev->devid);
1401                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
1402                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
1403                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
1404                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
1405                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
1406                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
1407                 flags = btrfs_super_flags(sb);
1408                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
1409
1410
1411                 crc = ~(u32)0;
1412                 crc = btrfs_csum_data(root, (char *)sb + BTRFS_CSUM_SIZE, crc,
1413                                       BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
1414                 btrfs_csum_final(crc, sb->csum);
1415
1416                 bh = __getblk(dev->bdev, BTRFS_SUPER_INFO_OFFSET / 4096,
1417                               BTRFS_SUPER_INFO_SIZE);
1418
1419                 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
1420                 dev->pending_io = bh;
1421
1422                 get_bh(bh);
1423                 set_buffer_uptodate(bh);
1424                 lock_buffer(bh);
1425                 bh->b_end_io = btrfs_end_buffer_write_sync;
1426
1427                 if (do_barriers && dev->barriers) {
1428                         ret = submit_bh(WRITE_BARRIER, bh);
1429                         if (ret == -EOPNOTSUPP) {
1430                                 printk("btrfs: disabling barriers on dev %s\n",
1431                                        dev->name);
1432                                 set_buffer_uptodate(bh);
1433                                 dev->barriers = 0;
1434                                 get_bh(bh);
1435                                 lock_buffer(bh);
1436                                 ret = submit_bh(WRITE, bh);
1437                         }
1438                 } else {
1439                         ret = submit_bh(WRITE, bh);
1440                 }
1441                 if (ret)
1442                         total_errors++;
1443         }
1444         if (total_errors > max_errors) {
1445                 printk("btrfs: %d errors while writing supers\n", total_errors);
1446                 BUG();
1447         }
1448         total_errors = 0;
1449
1450         list_for_each(cur, head) {
1451                 dev = list_entry(cur, struct btrfs_device, dev_list);
1452                 BUG_ON(!dev->pending_io);
1453                 bh = dev->pending_io;
1454                 wait_on_buffer(bh);
1455                 if (!buffer_uptodate(dev->pending_io)) {
1456                         if (do_barriers && dev->barriers) {
1457                                 printk("btrfs: disabling barriers on dev %s\n",
1458                                        dev->name);
1459                                 set_buffer_uptodate(bh);
1460                                 get_bh(bh);
1461                                 lock_buffer(bh);
1462                                 dev->barriers = 0;
1463                                 ret = submit_bh(WRITE, bh);
1464                                 BUG_ON(ret);
1465                                 wait_on_buffer(bh);
1466                                 BUG_ON(!buffer_uptodate(bh));
1467                         } else {
1468                                 total_errors++;
1469                         }
1470
1471                 }
1472                 dev->pending_io = NULL;
1473                 brelse(bh);
1474         }
1475         if (total_errors > max_errors) {
1476                 printk("btrfs: %d errors while writing supers\n", total_errors);
1477                 BUG();
1478         }
1479         return 0;
1480 }
1481
1482 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
1483                       *root)
1484 {
1485         int ret;
1486
1487         ret = write_all_supers(root);
1488         return ret;
1489 }
1490
1491 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
1492 {
1493         radix_tree_delete(&fs_info->fs_roots_radix,
1494                           (unsigned long)root->root_key.objectid);
1495         if (root->in_sysfs)
1496                 btrfs_sysfs_del_root(root);
1497         if (root->inode)
1498                 iput(root->inode);
1499         if (root->node)
1500                 free_extent_buffer(root->node);
1501         if (root->commit_root)
1502                 free_extent_buffer(root->commit_root);
1503         if (root->name)
1504                 kfree(root->name);
1505         kfree(root);
1506         return 0;
1507 }
1508
1509 static int del_fs_roots(struct btrfs_fs_info *fs_info)
1510 {
1511         int ret;
1512         struct btrfs_root *gang[8];
1513         int i;
1514
1515         while(1) {
1516                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1517                                              (void **)gang, 0,
1518                                              ARRAY_SIZE(gang));
1519                 if (!ret)
1520                         break;
1521                 for (i = 0; i < ret; i++)
1522                         btrfs_free_fs_root(fs_info, gang[i]);
1523         }
1524         return 0;
1525 }
1526
1527 int close_ctree(struct btrfs_root *root)
1528 {
1529         int ret;
1530         struct btrfs_trans_handle *trans;
1531         struct btrfs_fs_info *fs_info = root->fs_info;
1532
1533         fs_info->closing = 1;
1534         btrfs_transaction_flush_work(root);
1535         mutex_lock(&fs_info->fs_mutex);
1536         btrfs_defrag_dirty_roots(root->fs_info);
1537         trans = btrfs_start_transaction(root, 1);
1538         ret = btrfs_commit_transaction(trans, root);
1539         /* run commit again to  drop the original snapshot */
1540         trans = btrfs_start_transaction(root, 1);
1541         btrfs_commit_transaction(trans, root);
1542         ret = btrfs_write_and_wait_transaction(NULL, root);
1543         BUG_ON(ret);
1544
1545         write_ctree_super(NULL, root);
1546         mutex_unlock(&fs_info->fs_mutex);
1547
1548         btrfs_transaction_flush_work(root);
1549
1550         if (fs_info->delalloc_bytes) {
1551                 printk("btrfs: at unmount delalloc count %Lu\n",
1552                        fs_info->delalloc_bytes);
1553         }
1554         if (fs_info->extent_root->node)
1555                 free_extent_buffer(fs_info->extent_root->node);
1556
1557         if (fs_info->tree_root->node)
1558                 free_extent_buffer(fs_info->tree_root->node);
1559
1560         if (root->fs_info->chunk_root->node);
1561                 free_extent_buffer(root->fs_info->chunk_root->node);
1562
1563         if (root->fs_info->dev_root->node);
1564                 free_extent_buffer(root->fs_info->dev_root->node);
1565
1566         btrfs_free_block_groups(root->fs_info);
1567         del_fs_roots(fs_info);
1568
1569         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1570
1571         extent_io_tree_empty_lru(&fs_info->free_space_cache);
1572         extent_io_tree_empty_lru(&fs_info->block_group_cache);
1573         extent_io_tree_empty_lru(&fs_info->pinned_extents);
1574         extent_io_tree_empty_lru(&fs_info->pending_del);
1575         extent_io_tree_empty_lru(&fs_info->extent_ins);
1576         extent_io_tree_empty_lru(&BTRFS_I(fs_info->btree_inode)->io_tree);
1577
1578         flush_workqueue(async_submit_workqueue);
1579         flush_workqueue(end_io_workqueue);
1580
1581         truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
1582
1583         flush_workqueue(async_submit_workqueue);
1584         destroy_workqueue(async_submit_workqueue);
1585
1586         flush_workqueue(end_io_workqueue);
1587         destroy_workqueue(end_io_workqueue);
1588
1589         iput(fs_info->btree_inode);
1590 #if 0
1591         while(!list_empty(&fs_info->hashers)) {
1592                 struct btrfs_hasher *hasher;
1593                 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
1594                                     hashers);
1595                 list_del(&hasher->hashers);
1596                 crypto_free_hash(&fs_info->hash_tfm);
1597                 kfree(hasher);
1598         }
1599 #endif
1600         close_all_devices(fs_info);
1601         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1602
1603 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,23)
1604         bdi_destroy(&fs_info->bdi);
1605 #endif
1606
1607         kfree(fs_info->extent_root);
1608         kfree(fs_info->tree_root);
1609         kfree(fs_info->chunk_root);
1610         kfree(fs_info->dev_root);
1611         return 0;
1612 }
1613
1614 int btrfs_buffer_uptodate(struct extent_buffer *buf)
1615 {
1616         struct inode *btree_inode = buf->first_page->mapping->host;
1617         return extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
1618 }
1619
1620 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
1621 {
1622         struct inode *btree_inode = buf->first_page->mapping->host;
1623         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
1624                                           buf);
1625 }
1626
1627 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
1628 {
1629         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1630         u64 transid = btrfs_header_generation(buf);
1631         struct inode *btree_inode = root->fs_info->btree_inode;
1632
1633         if (transid != root->fs_info->generation) {
1634                 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
1635                         (unsigned long long)buf->start,
1636                         transid, root->fs_info->generation);
1637                 WARN_ON(1);
1638         }
1639         set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
1640 }
1641
1642 void btrfs_throttle(struct btrfs_root *root)
1643 {
1644         struct backing_dev_info *bdi;
1645
1646         bdi = &root->fs_info->bdi;
1647         if (root->fs_info->throttles && bdi_write_congested(bdi)) {
1648 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,18)
1649                 congestion_wait(WRITE, HZ/20);
1650 #else
1651                 blk_congestion_wait(WRITE, HZ/20);
1652 #endif
1653         }
1654 }
1655
1656 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
1657 {
1658         struct extent_io_tree *tree;
1659         u64 num_dirty;
1660         u64 start = 0;
1661         unsigned long thresh = 16 * 1024 * 1024;
1662         tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
1663
1664         if (current_is_pdflush())
1665                 return;
1666
1667         num_dirty = count_range_bits(tree, &start, (u64)-1,
1668                                      thresh, EXTENT_DIRTY);
1669         if (num_dirty > thresh) {
1670                 balance_dirty_pages_ratelimited_nr(
1671                                    root->fs_info->btree_inode->i_mapping, 1);
1672         }
1673 }
1674
1675 void btrfs_set_buffer_defrag(struct extent_buffer *buf)
1676 {
1677         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1678         struct inode *btree_inode = root->fs_info->btree_inode;
1679         set_extent_bits(&BTRFS_I(btree_inode)->io_tree, buf->start,
1680                         buf->start + buf->len - 1, EXTENT_DEFRAG, GFP_NOFS);
1681 }
1682
1683 void btrfs_set_buffer_defrag_done(struct extent_buffer *buf)
1684 {
1685         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1686         struct inode *btree_inode = root->fs_info->btree_inode;
1687         set_extent_bits(&BTRFS_I(btree_inode)->io_tree, buf->start,
1688                         buf->start + buf->len - 1, EXTENT_DEFRAG_DONE,
1689                         GFP_NOFS);
1690 }
1691
1692 int btrfs_buffer_defrag(struct extent_buffer *buf)
1693 {
1694         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1695         struct inode *btree_inode = root->fs_info->btree_inode;
1696         return test_range_bit(&BTRFS_I(btree_inode)->io_tree,
1697                      buf->start, buf->start + buf->len - 1, EXTENT_DEFRAG, 0);
1698 }
1699
1700 int btrfs_buffer_defrag_done(struct extent_buffer *buf)
1701 {
1702         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1703         struct inode *btree_inode = root->fs_info->btree_inode;
1704         return test_range_bit(&BTRFS_I(btree_inode)->io_tree,
1705                      buf->start, buf->start + buf->len - 1,
1706                      EXTENT_DEFRAG_DONE, 0);
1707 }
1708
1709 int btrfs_clear_buffer_defrag_done(struct extent_buffer *buf)
1710 {
1711         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1712         struct inode *btree_inode = root->fs_info->btree_inode;
1713         return clear_extent_bits(&BTRFS_I(btree_inode)->io_tree,
1714                      buf->start, buf->start + buf->len - 1,
1715                      EXTENT_DEFRAG_DONE, GFP_NOFS);
1716 }
1717
1718 int btrfs_clear_buffer_defrag(struct extent_buffer *buf)
1719 {
1720         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1721         struct inode *btree_inode = root->fs_info->btree_inode;
1722         return clear_extent_bits(&BTRFS_I(btree_inode)->io_tree,
1723                      buf->start, buf->start + buf->len - 1,
1724                      EXTENT_DEFRAG, GFP_NOFS);
1725 }
1726
1727 int btrfs_read_buffer(struct extent_buffer *buf)
1728 {
1729         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1730         int ret;
1731         ret = btree_read_extent_buffer_pages(root, buf, 0);
1732         if (ret == 0) {
1733                 buf->flags |= EXTENT_UPTODATE;
1734         }
1735         return ret;
1736 }
1737
1738 static struct extent_io_ops btree_extent_io_ops = {
1739         .writepage_io_hook = btree_writepage_io_hook,
1740         .readpage_end_io_hook = btree_readpage_end_io_hook,
1741         .submit_bio_hook = btree_submit_bio_hook,
1742         /* note we're sharing with inode.c for the merge bio hook */
1743         .merge_bio_hook = btrfs_merge_bio_hook,
1744 };