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