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/crc32c.h>
  22 #include <linux/scatterlist.h>
  23 #include <linux/swap.h>
  24 #include <linux/radix-tree.h>
  25 #include <linux/writeback.h>
  26 #include <linux/buffer_head.h> // for block_sync_page
  27 #include "ctree.h"
  28 #include "disk-io.h"
  29 #include "transaction.h"
  30 #include "btrfs_inode.h"
  31 #include "volumes.h"
  32 #include "print-tree.h"
  33
  34 #if 0
  35 static int check_tree_block(struct btrfs_root *root, struct extent_buffer *buf)
  36 {
  37         if (extent_buffer_blocknr(buf) != btrfs_header_blocknr(buf)) {
  38                 printk(KERN_CRIT "buf blocknr(buf) is %llu, header is %llu\n",
  39                        (unsigned long long)extent_buffer_blocknr(buf),
  40                        (unsigned long long)btrfs_header_blocknr(buf));
  41                 return 1;
  42         }
  43         return 0;
  44 }
  45 #endif
  46
  47 static struct extent_io_ops btree_extent_io_ops;
  48
  49 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
  50                                             u64 bytenr, u32 blocksize)
  51 {
  52         struct inode *btree_inode = root->fs_info->btree_inode;
  53         struct extent_buffer *eb;
  54         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
  55                                 bytenr, blocksize, GFP_NOFS);
  56         return eb;
  57 }
  58
  59 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
  60                                                  u64 bytenr, u32 blocksize)
  61 {
  62         struct inode *btree_inode = root->fs_info->btree_inode;
  63         struct extent_buffer *eb;
  64
  65         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
  66                                  bytenr, blocksize, NULL, GFP_NOFS);
  67         return eb;
  68 }
  69
  70 struct extent_map *btree_get_extent(struct inode *inode, struct page *page,
  71                                     size_t page_offset, u64 start, u64 len,
  72                                     int create)
  73 {
  74         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
  75         struct extent_map *em;
  76         int ret;
  77
  78 again:
  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 = i_size_read(inode);
  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         spin_unlock(&em_tree->lock);
  98
  99         if (ret == -EEXIST) {
 100                 free_extent_map(em);
 101                 em = NULL;
 102                 goto again;
 103         } else if (ret) {
 104                 em = ERR_PTR(ret);
 105         }
 106 out:
 107         return em;
 108 }
 109
 110 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
 111 {
 112         return crc32c(seed, data, len);
 113 }
 114
 115 void btrfs_csum_final(u32 crc, char *result)
 116 {
 117         *(__le32 *)result = ~cpu_to_le32(crc);
 118 }
 119
 120 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 121                            int verify)
 122 {
 123         char result[BTRFS_CRC32_SIZE];
 124         unsigned long len;
 125         unsigned long cur_len;
 126         unsigned long offset = BTRFS_CSUM_SIZE;
 127         char *map_token = NULL;
 128         char *kaddr;
 129         unsigned long map_start;
 130         unsigned long map_len;
 131         int err;
 132         u32 crc = ~(u32)0;
 133
 134         len = buf->len - offset;
 135         while(len > 0) {
 136                 err = map_private_extent_buffer(buf, offset, 32,
 137                                         &map_token, &kaddr,
 138                                         &map_start, &map_len, KM_USER0);
 139                 if (err) {
 140                         printk("failed to map extent buffer! %lu\n",
 141                                offset);
 142                         return 1;
 143                 }
 144                 cur_len = min(len, map_len - (offset - map_start));
 145                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
 146                                       crc, cur_len);
 147                 len -= cur_len;
 148                 offset += cur_len;
 149                 unmap_extent_buffer(buf, map_token, KM_USER0);
 150         }
 151         btrfs_csum_final(crc, result);
 152
 153         if (verify) {
 154                 int from_this_trans = 0;
 155
 156                 if (root->fs_info->running_transaction &&
 157                     btrfs_header_generation(buf) ==
 158                     root->fs_info->running_transaction->transid)
 159                         from_this_trans = 1;
 160
 161                 /* FIXME, this is not good */
 162                 if (memcmp_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE)) {
 163                         u32 val;
 164                         u32 found = 0;
 165                         memcpy(&found, result, BTRFS_CRC32_SIZE);
 166
 167                         read_extent_buffer(buf, &val, 0, BTRFS_CRC32_SIZE);
 168                         WARN_ON(1);
 169                         printk("btrfs: %s checksum verify failed on %llu "
 170                                "wanted %X found %X from_this_trans %d "
 171                                "level %d\n",
 172                                root->fs_info->sb->s_id,
 173                                buf->start, val, found, from_this_trans,
 174                                btrfs_header_level(buf));
 175                         return 1;
 176                 }
 177         } else {
 178                 write_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE);
 179         }
 180         return 0;
 181 }
 182
 183
 184 int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 185 {
 186         struct extent_io_tree *tree;
 187         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
 188         u64 found_start;
 189         int found_level;
 190         unsigned long len;
 191         struct extent_buffer *eb;
 192         tree = &BTRFS_I(page->mapping->host)->io_tree;
 193
 194         if (page->private == EXTENT_PAGE_PRIVATE)
 195                 goto out;
 196         if (!page->private)
 197                 goto out;
 198         len = page->private >> 2;
 199         if (len == 0) {
 200                 WARN_ON(1);
 201         }
 202         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 203         read_extent_buffer_pages(tree, eb, start + PAGE_CACHE_SIZE, 1,
 204                                  btree_get_extent);
 205         btrfs_clear_buffer_defrag(eb);
 206         found_start = btrfs_header_bytenr(eb);
 207         if (found_start != start) {
 208                 printk("warning: eb start incorrect %Lu buffer %Lu len %lu\n",
 209                        start, found_start, len);
 210                 WARN_ON(1);
 211                 goto err;
 212         }
 213         if (eb->first_page != page) {
 214                 printk("bad first page %lu %lu\n", eb->first_page->index,
 215                        page->index);
 216                 WARN_ON(1);
 217                 goto err;
 218         }
 219         if (!PageUptodate(page)) {
 220                 printk("csum not up to date page %lu\n", page->index);
 221                 WARN_ON(1);
 222                 goto err;
 223         }
 224         found_level = btrfs_header_level(eb);
 225         spin_lock(&root->fs_info->hash_lock);
 226         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
 227         spin_unlock(&root->fs_info->hash_lock);
 228         csum_tree_block(root, eb, 0);
 229 err:
 230         free_extent_buffer(eb);
 231 out:
 232         return 0;
 233 }
 234
 235 static int btree_writepage_io_hook(struct page *page, u64 start, u64 end)
 236 {
 237         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 238
 239         csum_dirty_buffer(root, page);
 240         return 0;
 241 }
 242
 243 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio)
 244 {
 245         struct btrfs_root *root = BTRFS_I(inode)->root;
 246         u64 offset;
 247         offset = bio->bi_sector << 9;
 248         if (offset == BTRFS_SUPER_INFO_OFFSET) {
 249                 bio->bi_bdev = root->fs_info->sb->s_bdev;
 250                 submit_bio(rw, bio);
 251                 return 0;
 252         }
 253         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio);
 254 }
 255
 256 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 257 {
 258         struct extent_io_tree *tree;
 259         tree = &BTRFS_I(page->mapping->host)->io_tree;
 260         return extent_write_full_page(tree, page, btree_get_extent, wbc);
 261 }
 262
 263 static int btree_writepages(struct address_space *mapping,
 264                             struct writeback_control *wbc)
 265 {
 266         struct extent_io_tree *tree;
 267         tree = &BTRFS_I(mapping->host)->io_tree;
 268         if (wbc->sync_mode == WB_SYNC_NONE) {
 269                 u64 num_dirty;
 270                 u64 start = 0;
 271                 unsigned long thresh = 96 * 1024 * 1024;
 272
 273                 if (wbc->for_kupdate)
 274                         return 0;
 275
 276                 if (current_is_pdflush()) {
 277                         thresh = 96 * 1024 * 1024;
 278                 } else {
 279                         thresh = 8 * 1024 * 1024;
 280                 }
 281                 num_dirty = count_range_bits(tree, &start, (u64)-1,
 282                                              thresh, EXTENT_DIRTY);
 283                 if (num_dirty < thresh) {
 284                         return 0;
 285                 }
 286         }
 287         return extent_writepages(tree, mapping, btree_get_extent, wbc);
 288 }
 289
 290 int btree_readpage(struct file *file, struct page *page)
 291 {
 292         struct extent_io_tree *tree;
 293         tree = &BTRFS_I(page->mapping->host)->io_tree;
 294         return extent_read_full_page(tree, page, btree_get_extent);
 295 }
 296
 297 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 298 {
 299         struct extent_io_tree *tree;
 300         struct extent_map_tree *map;
 301         int ret;
 302
 303         tree = &BTRFS_I(page->mapping->host)->io_tree;
 304         map = &BTRFS_I(page->mapping->host)->extent_tree;
 305         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
 306         if (ret == 1) {
 307                 ClearPagePrivate(page);
 308                 set_page_private(page, 0);
 309                 page_cache_release(page);
 310         }
 311         return ret;
 312 }
 313
 314 static void btree_invalidatepage(struct page *page, unsigned long offset)
 315 {
 316         struct extent_io_tree *tree;
 317         tree = &BTRFS_I(page->mapping->host)->io_tree;
 318         extent_invalidatepage(tree, page, offset);
 319         btree_releasepage(page, GFP_NOFS);
 320 }
 321
 322 #if 0
 323 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 324 {
 325         struct buffer_head *bh;
 326         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 327         struct buffer_head *head;
 328         if (!page_has_buffers(page)) {
 329                 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
 330                                         (1 << BH_Dirty)|(1 << BH_Uptodate));
 331         }
 332         head = page_buffers(page);
 333         bh = head;
 334         do {
 335                 if (buffer_dirty(bh))
 336                         csum_tree_block(root, bh, 0);
 337                 bh = bh->b_this_page;
 338         } while (bh != head);
 339         return block_write_full_page(page, btree_get_block, wbc);
 340 }
 341 #endif
 342
 343 static struct address_space_operations btree_aops = {
 344         .readpage       = btree_readpage,
 345         .writepage      = btree_writepage,
 346         .writepages     = btree_writepages,
 347         .releasepage    = btree_releasepage,
 348         .invalidatepage = btree_invalidatepage,
 349         .sync_page      = block_sync_page,
 350 };
 351
 352 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize)
 353 {
 354         struct extent_buffer *buf = NULL;
 355         struct inode *btree_inode = root->fs_info->btree_inode;
 356         int ret = 0;
 357
 358         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 359         if (!buf)
 360                 return 0;
 361         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
 362                                  buf, 0, 0, btree_get_extent);
 363         free_extent_buffer(buf);
 364         return ret;
 365 }
 366
 367 static int close_all_devices(struct btrfs_fs_info *fs_info)
 368 {
 369         struct list_head *list;
 370         struct list_head *next;
 371         struct btrfs_device *device;
 372
 373         list = &fs_info->fs_devices->devices;
 374         list_for_each(next, list) {
 375                 device = list_entry(next, struct btrfs_device, dev_list);
 376                 if (device->bdev && device->bdev != fs_info->sb->s_bdev)
 377                         close_bdev_excl(device->bdev);
 378                 device->bdev = NULL;
 379         }
 380         return 0;
 381 }
 382
 383 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
 384                                       u32 blocksize)
 385 {
 386         struct extent_buffer *buf = NULL;
 387         struct inode *btree_inode = root->fs_info->btree_inode;
 388         struct extent_io_tree *io_tree;
 389         u64 end;
 390         int ret;
 391
 392         io_tree = &BTRFS_I(btree_inode)->io_tree;
 393
 394         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 395         if (!buf)
 396                 return NULL;
 397         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree, buf, 0, 1,
 398                                  btree_get_extent);
 399
 400         if (buf->flags & EXTENT_CSUM)
 401                 return buf;
 402
 403         end = buf->start + PAGE_CACHE_SIZE - 1;
 404         if (test_range_bit(io_tree, buf->start, end, EXTENT_CSUM, 1)) {
 405                 buf->flags |= EXTENT_CSUM;
 406                 return buf;
 407         }
 408
 409         lock_extent(io_tree, buf->start, end, GFP_NOFS);
 410
 411         if (test_range_bit(io_tree, buf->start, end, EXTENT_CSUM, 1)) {
 412                 buf->flags |= EXTENT_CSUM;
 413                 goto out_unlock;
 414         }
 415
 416         ret = csum_tree_block(root, buf, 1);
 417         set_extent_bits(io_tree, buf->start, end, EXTENT_CSUM, GFP_NOFS);
 418         buf->flags |= EXTENT_CSUM;
 419
 420 out_unlock:
 421         unlock_extent(io_tree, buf->start, end, GFP_NOFS);
 422         return buf;
 423 }
 424
 425 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 426                      struct extent_buffer *buf)
 427 {
 428         struct inode *btree_inode = root->fs_info->btree_inode;
 429         if (btrfs_header_generation(buf) ==
 430             root->fs_info->running_transaction->transid)
 431                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
 432                                           buf);
 433         return 0;
 434 }
 435
 436 int wait_on_tree_block_writeback(struct btrfs_root *root,
 437                                  struct extent_buffer *buf)
 438 {
 439         struct inode *btree_inode = root->fs_info->btree_inode;
 440         wait_on_extent_buffer_writeback(&BTRFS_I(btree_inode)->io_tree,
 441                                         buf);
 442         return 0;
 443 }
 444
 445 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
 446                         u32 stripesize, struct btrfs_root *root,
 447                         struct btrfs_fs_info *fs_info,
 448                         u64 objectid)
 449 {
 450         root->node = NULL;
 451         root->inode = NULL;
 452         root->commit_root = NULL;
 453         root->sectorsize = sectorsize;
 454         root->nodesize = nodesize;
 455         root->leafsize = leafsize;
 456         root->stripesize = stripesize;
 457         root->ref_cows = 0;
 458         root->track_dirty = 0;
 459
 460         root->fs_info = fs_info;
 461         root->objectid = objectid;
 462         root->last_trans = 0;
 463         root->highest_inode = 0;
 464         root->last_inode_alloc = 0;
 465         root->name = NULL;
 466         root->in_sysfs = 0;
 467
 468         INIT_LIST_HEAD(&root->dirty_list);
 469         memset(&root->root_key, 0, sizeof(root->root_key));
 470         memset(&root->root_item, 0, sizeof(root->root_item));
 471         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
 472         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
 473         init_completion(&root->kobj_unregister);
 474         root->defrag_running = 0;
 475         root->defrag_level = 0;
 476         root->root_key.objectid = objectid;
 477         return 0;
 478 }
 479
 480 static int find_and_setup_root(struct btrfs_root *tree_root,
 481                                struct btrfs_fs_info *fs_info,
 482                                u64 objectid,
 483                                struct btrfs_root *root)
 484 {
 485         int ret;
 486         u32 blocksize;
 487
 488         __setup_root(tree_root->nodesize, tree_root->leafsize,
 489                      tree_root->sectorsize, tree_root->stripesize,
 490                      root, fs_info, objectid);
 491         ret = btrfs_find_last_root(tree_root, objectid,
 492                                    &root->root_item, &root->root_key);
 493         BUG_ON(ret);
 494
 495         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 496         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 497                                      blocksize);
 498         BUG_ON(!root->node);
 499         return 0;
 500 }
 501
 502 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_fs_info *fs_info,
 503                                                struct btrfs_key *location)
 504 {
 505         struct btrfs_root *root;
 506         struct btrfs_root *tree_root = fs_info->tree_root;
 507         struct btrfs_path *path;
 508         struct extent_buffer *l;
 509         u64 highest_inode;
 510         u32 blocksize;
 511         int ret = 0;
 512
 513         root = kzalloc(sizeof(*root), GFP_NOFS);
 514         if (!root)
 515                 return ERR_PTR(-ENOMEM);
 516         if (location->offset == (u64)-1) {
 517                 ret = find_and_setup_root(tree_root, fs_info,
 518                                           location->objectid, root);
 519                 if (ret) {
 520                         kfree(root);
 521                         return ERR_PTR(ret);
 522                 }
 523                 goto insert;
 524         }
 525
 526         __setup_root(tree_root->nodesize, tree_root->leafsize,
 527                      tree_root->sectorsize, tree_root->stripesize,
 528                      root, fs_info, location->objectid);
 529
 530         path = btrfs_alloc_path();
 531         BUG_ON(!path);
 532         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
 533         if (ret != 0) {
 534                 if (ret > 0)
 535                         ret = -ENOENT;
 536                 goto out;
 537         }
 538         l = path->nodes[0];
 539         read_extent_buffer(l, &root->root_item,
 540                btrfs_item_ptr_offset(l, path->slots[0]),
 541                sizeof(root->root_item));
 542         memcpy(&root->root_key, location, sizeof(*location));
 543         ret = 0;
 544 out:
 545         btrfs_release_path(root, path);
 546         btrfs_free_path(path);
 547         if (ret) {
 548                 kfree(root);
 549                 return ERR_PTR(ret);
 550         }
 551         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 552         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 553                                      blocksize);
 554         BUG_ON(!root->node);
 555 insert:
 556         root->ref_cows = 1;
 557         ret = btrfs_find_highest_inode(root, &highest_inode);
 558         if (ret == 0) {
 559                 root->highest_inode = highest_inode;
 560                 root->last_inode_alloc = highest_inode;
 561         }
 562         return root;
 563 }
 564
 565 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
 566                                         u64 root_objectid)
 567 {
 568         struct btrfs_root *root;
 569
 570         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
 571                 return fs_info->tree_root;
 572         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
 573                 return fs_info->extent_root;
 574
 575         root = radix_tree_lookup(&fs_info->fs_roots_radix,
 576                                  (unsigned long)root_objectid);
 577         return root;
 578 }
 579
 580 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
 581                                               struct btrfs_key *location)
 582 {
 583         struct btrfs_root *root;
 584         int ret;
 585
 586         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
 587                 return fs_info->tree_root;
 588         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
 589                 return fs_info->extent_root;
 590
 591         root = radix_tree_lookup(&fs_info->fs_roots_radix,
 592                                  (unsigned long)location->objectid);
 593         if (root)
 594                 return root;
 595
 596         root = btrfs_read_fs_root_no_radix(fs_info, location);
 597         if (IS_ERR(root))
 598                 return root;
 599         ret = radix_tree_insert(&fs_info->fs_roots_radix,
 600                                 (unsigned long)root->root_key.objectid,
 601                                 root);
 602         if (ret) {
 603                 free_extent_buffer(root->node);
 604                 kfree(root);
 605                 return ERR_PTR(ret);
 606         }
 607         ret = btrfs_find_dead_roots(fs_info->tree_root,
 608                                     root->root_key.objectid, root);
 609         BUG_ON(ret);
 610
 611         return root;
 612 }
 613
 614 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
 615                                       struct btrfs_key *location,
 616                                       const char *name, int namelen)
 617 {
 618         struct btrfs_root *root;
 619         int ret;
 620
 621         root = btrfs_read_fs_root_no_name(fs_info, location);
 622         if (!root)
 623                 return NULL;
 624
 625         if (root->in_sysfs)
 626                 return root;
 627
 628         ret = btrfs_set_root_name(root, name, namelen);
 629         if (ret) {
 630                 free_extent_buffer(root->node);
 631                 kfree(root);
 632                 return ERR_PTR(ret);
 633         }
 634
 635         ret = btrfs_sysfs_add_root(root);
 636         if (ret) {
 637                 free_extent_buffer(root->node);
 638                 kfree(root->name);
 639                 kfree(root);
 640                 return ERR_PTR(ret);
 641         }
 642         root->in_sysfs = 1;
 643         return root;
 644 }
 645 #if 0
 646 static int add_hasher(struct btrfs_fs_info *info, char *type) {
 647         struct btrfs_hasher *hasher;
 648
 649         hasher = kmalloc(sizeof(*hasher), GFP_NOFS);
 650         if (!hasher)
 651                 return -ENOMEM;
 652         hasher->hash_tfm = crypto_alloc_hash(type, 0, CRYPTO_ALG_ASYNC);
 653         if (!hasher->hash_tfm) {
 654                 kfree(hasher);
 655                 return -EINVAL;
 656         }
 657         spin_lock(&info->hash_lock);
 658         list_add(&hasher->list, &info->hashers);
 659         spin_unlock(&info->hash_lock);
 660         return 0;
 661 }
 662 #endif
 663
 664 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
 665 {
 666         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
 667         int ret = 0;
 668         struct list_head *cur;
 669         struct btrfs_device *device;
 670         struct backing_dev_info *bdi;
 671
 672         list_for_each(cur, &info->fs_devices->devices) {
 673                 device = list_entry(cur, struct btrfs_device, dev_list);
 674                 bdi = blk_get_backing_dev_info(device->bdev);
 675                 if (bdi && bdi_congested(bdi, bdi_bits)) {
 676                         ret = 1;
 677                         break;
 678                 }
 679         }
 680         return ret;
 681 }
 682
 683 void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
 684 {
 685         struct list_head *cur;
 686         struct btrfs_device *device;
 687         struct btrfs_fs_info *info;
 688
 689         info = (struct btrfs_fs_info *)bdi->unplug_io_data;
 690         list_for_each(cur, &info->fs_devices->devices) {
 691                 device = list_entry(cur, struct btrfs_device, dev_list);
 692                 bdi = blk_get_backing_dev_info(device->bdev);
 693                 if (bdi->unplug_io_fn) {
 694                         bdi->unplug_io_fn(bdi, page);
 695                 }
 696         }
 697 }
 698
 699 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
 700 {
 701         bdi_init(bdi);
 702         bdi->ra_pages   = default_backing_dev_info.ra_pages * 4;
 703         bdi->state              = 0;
 704         bdi->capabilities       = default_backing_dev_info.capabilities;
 705         bdi->unplug_io_fn       = btrfs_unplug_io_fn;
 706         bdi->unplug_io_data     = info;
 707         bdi->congested_fn       = btrfs_congested_fn;
 708         bdi->congested_data     = info;
 709         return 0;
 710 }
 711
 712 struct btrfs_root *open_ctree(struct super_block *sb,
 713                               struct btrfs_fs_devices *fs_devices)
 714 {
 715         u32 sectorsize;
 716         u32 nodesize;
 717         u32 leafsize;
 718         u32 blocksize;
 719         u32 stripesize;
 720         struct btrfs_root *extent_root = kmalloc(sizeof(struct btrfs_root),
 721                                                  GFP_NOFS);
 722         struct btrfs_root *tree_root = kmalloc(sizeof(struct btrfs_root),
 723                                                GFP_NOFS);
 724         struct btrfs_fs_info *fs_info = kmalloc(sizeof(*fs_info),
 725                                                 GFP_NOFS);
 726         struct btrfs_root *chunk_root = kmalloc(sizeof(struct btrfs_root),
 727                                                 GFP_NOFS);
 728         struct btrfs_root *dev_root = kmalloc(sizeof(struct btrfs_root),
 729                                               GFP_NOFS);
 730         int ret;
 731         int err = -EINVAL;
 732         struct btrfs_super_block *disk_super;
 733         if (!extent_root || !tree_root || !fs_info) {
 734                 err = -ENOMEM;
 735                 goto fail;
 736         }
 737         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
 738         INIT_LIST_HEAD(&fs_info->trans_list);
 739         INIT_LIST_HEAD(&fs_info->dead_roots);
 740         INIT_LIST_HEAD(&fs_info->hashers);
 741         spin_lock_init(&fs_info->hash_lock);
 742         spin_lock_init(&fs_info->delalloc_lock);
 743         spin_lock_init(&fs_info->new_trans_lock);
 744
 745         memset(&fs_info->super_kobj, 0, sizeof(fs_info->super_kobj));
 746         init_completion(&fs_info->kobj_unregister);
 747         sb_set_blocksize(sb, 4096);
 748         fs_info->running_transaction = NULL;
 749         fs_info->last_trans_committed = 0;
 750         fs_info->tree_root = tree_root;
 751         fs_info->extent_root = extent_root;
 752         fs_info->chunk_root = chunk_root;
 753         fs_info->dev_root = dev_root;
 754         fs_info->fs_devices = fs_devices;
 755         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
 756         INIT_LIST_HEAD(&fs_info->space_info);
 757         btrfs_mapping_init(&fs_info->mapping_tree);
 758         fs_info->sb = sb;
 759         fs_info->throttles = 0;
 760         fs_info->mount_opt = 0;
 761         fs_info->max_extent = (u64)-1;
 762         fs_info->max_inline = 8192 * 1024;
 763         fs_info->delalloc_bytes = 0;
 764         setup_bdi(fs_info, &fs_info->bdi);
 765         fs_info->btree_inode = new_inode(sb);
 766         fs_info->btree_inode->i_ino = 1;
 767         fs_info->btree_inode->i_nlink = 1;
 768         fs_info->btree_inode->i_size = sb->s_bdev->bd_inode->i_size;
 769         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
 770         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
 771
 772         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
 773                              fs_info->btree_inode->i_mapping,
 774                              GFP_NOFS);
 775         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
 776                              GFP_NOFS);
 777
 778         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
 779
 780         extent_io_tree_init(&fs_info->free_space_cache,
 781                              fs_info->btree_inode->i_mapping, GFP_NOFS);
 782         extent_io_tree_init(&fs_info->block_group_cache,
 783                              fs_info->btree_inode->i_mapping, GFP_NOFS);
 784         extent_io_tree_init(&fs_info->pinned_extents,
 785                              fs_info->btree_inode->i_mapping, GFP_NOFS);
 786         extent_io_tree_init(&fs_info->pending_del,
 787                              fs_info->btree_inode->i_mapping, GFP_NOFS);
 788         extent_io_tree_init(&fs_info->extent_ins,
 789                              fs_info->btree_inode->i_mapping, GFP_NOFS);
 790         fs_info->do_barriers = 1;
 791         fs_info->closing = 0;
 792         fs_info->total_pinned = 0;
 793         fs_info->last_alloc = 0;
 794         fs_info->last_data_alloc = 0;
 795         fs_info->extra_alloc_bits = 0;
 796         fs_info->extra_data_alloc_bits = 0;
 797
 798 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
 799         INIT_WORK(&fs_info->trans_work, btrfs_transaction_cleaner, fs_info);
 800 #else
 801         INIT_DELAYED_WORK(&fs_info->trans_work, btrfs_transaction_cleaner);
 802 #endif
 803         BTRFS_I(fs_info->btree_inode)->root = tree_root;
 804         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
 805                sizeof(struct btrfs_key));
 806         insert_inode_hash(fs_info->btree_inode);
 807         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
 808
 809         mutex_init(&fs_info->trans_mutex);
 810         mutex_init(&fs_info->fs_mutex);
 811
 812 #if 0
 813         ret = add_hasher(fs_info, "crc32c");
 814         if (ret) {
 815                 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
 816                 err = -ENOMEM;
 817                 goto fail_iput;
 818         }
 819 #endif
 820         __setup_root(4096, 4096, 4096, 4096, tree_root,
 821                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
 822
 823         fs_info->sb_buffer = read_tree_block(tree_root,
 824                                              BTRFS_SUPER_INFO_OFFSET,
 825                                              4096);
 826
 827         if (!fs_info->sb_buffer)
 828                 goto fail_iput;
 829
 830         read_extent_buffer(fs_info->sb_buffer, &fs_info->super_copy, 0,
 831                            sizeof(fs_info->super_copy));
 832
 833         read_extent_buffer(fs_info->sb_buffer, fs_info->fsid,
 834                            (unsigned long)btrfs_super_fsid(fs_info->sb_buffer),
 835                            BTRFS_FSID_SIZE);
 836
 837         disk_super = &fs_info->super_copy;
 838         if (!btrfs_super_root(disk_super))
 839                 goto fail_sb_buffer;
 840
 841         if (btrfs_super_num_devices(disk_super) != fs_devices->num_devices) {
 842                 printk("Btrfs: wanted %llu devices, but found %llu\n",
 843                        (unsigned long long)btrfs_super_num_devices(disk_super),
 844                        (unsigned long long)fs_devices->num_devices);
 845                 goto fail_sb_buffer;
 846         }
 847         nodesize = btrfs_super_nodesize(disk_super);
 848         leafsize = btrfs_super_leafsize(disk_super);
 849         sectorsize = btrfs_super_sectorsize(disk_super);
 850         stripesize = btrfs_super_stripesize(disk_super);
 851         tree_root->nodesize = nodesize;
 852         tree_root->leafsize = leafsize;
 853         tree_root->sectorsize = sectorsize;
 854         tree_root->stripesize = stripesize;
 855         sb_set_blocksize(sb, sectorsize);
 856
 857         i_size_write(fs_info->btree_inode,
 858                      btrfs_super_total_bytes(disk_super));
 859
 860         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
 861                     sizeof(disk_super->magic))) {
 862                 printk("btrfs: valid FS not found on %s\n", sb->s_id);
 863                 goto fail_sb_buffer;
 864         }
 865
 866         mutex_lock(&fs_info->fs_mutex);
 867
 868         ret = btrfs_read_sys_array(tree_root);
 869         BUG_ON(ret);
 870
 871         blocksize = btrfs_level_size(tree_root,
 872                                      btrfs_super_chunk_root_level(disk_super));
 873
 874         __setup_root(nodesize, leafsize, sectorsize, stripesize,
 875                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
 876
 877         chunk_root->node = read_tree_block(chunk_root,
 878                                            btrfs_super_chunk_root(disk_super),
 879                                            blocksize);
 880         BUG_ON(!chunk_root->node);
 881
 882         ret = btrfs_read_chunk_tree(chunk_root);
 883         BUG_ON(ret);
 884
 885         blocksize = btrfs_level_size(tree_root,
 886                                      btrfs_super_root_level(disk_super));
 887
 888
 889         tree_root->node = read_tree_block(tree_root,
 890                                           btrfs_super_root(disk_super),
 891                                           blocksize);
 892         if (!tree_root->node)
 893                 goto fail_sb_buffer;
 894
 895
 896         ret = find_and_setup_root(tree_root, fs_info,
 897                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
 898         if (ret)
 899                 goto fail_tree_root;
 900         extent_root->track_dirty = 1;
 901
 902         ret = find_and_setup_root(tree_root, fs_info,
 903                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
 904         dev_root->track_dirty = 1;
 905
 906         if (ret)
 907                 goto fail_extent_root;
 908
 909         btrfs_read_block_groups(extent_root);
 910
 911         fs_info->generation = btrfs_super_generation(disk_super) + 1;
 912         mutex_unlock(&fs_info->fs_mutex);
 913         return tree_root;
 914
 915 fail_extent_root:
 916         free_extent_buffer(extent_root->node);
 917 fail_tree_root:
 918         mutex_unlock(&fs_info->fs_mutex);
 919         free_extent_buffer(tree_root->node);
 920 fail_sb_buffer:
 921         free_extent_buffer(fs_info->sb_buffer);
 922         extent_io_tree_empty_lru(&BTRFS_I(fs_info->btree_inode)->io_tree);
 923 fail_iput:
 924         iput(fs_info->btree_inode);
 925 fail:
 926         close_all_devices(fs_info);
 927         kfree(extent_root);
 928         kfree(tree_root);
 929         bdi_destroy(&fs_info->bdi);
 930         kfree(fs_info);
 931         return ERR_PTR(err);
 932 }
 933
 934 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
 935                       *root)
 936 {
 937         int ret;
 938         struct extent_buffer *super = root->fs_info->sb_buffer;
 939         struct inode *btree_inode = root->fs_info->btree_inode;
 940         struct super_block *sb = root->fs_info->sb;
 941
 942         if (!btrfs_test_opt(root, NOBARRIER))
 943                 blkdev_issue_flush(sb->s_bdev, NULL);
 944         set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, super);
 945         ret = sync_page_range_nolock(btree_inode, btree_inode->i_mapping,
 946                                      super->start, super->len);
 947         if (!btrfs_test_opt(root, NOBARRIER))
 948                 blkdev_issue_flush(sb->s_bdev, NULL);
 949         return ret;
 950 }
 951
 952 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
 953 {
 954         radix_tree_delete(&fs_info->fs_roots_radix,
 955                           (unsigned long)root->root_key.objectid);
 956         if (root->in_sysfs)
 957                 btrfs_sysfs_del_root(root);
 958         if (root->inode)
 959                 iput(root->inode);
 960         if (root->node)
 961                 free_extent_buffer(root->node);
 962         if (root->commit_root)
 963                 free_extent_buffer(root->commit_root);
 964         if (root->name)
 965                 kfree(root->name);
 966         kfree(root);
 967         return 0;
 968 }
 969
 970 static int del_fs_roots(struct btrfs_fs_info *fs_info)
 971 {
 972         int ret;
 973         struct btrfs_root *gang[8];
 974         int i;
 975
 976         while(1) {
 977                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
 978                                              (void **)gang, 0,
 979                                              ARRAY_SIZE(gang));
 980                 if (!ret)
 981                         break;
 982                 for (i = 0; i < ret; i++)
 983                         btrfs_free_fs_root(fs_info, gang[i]);
 984         }
 985         return 0;
 986 }
 987
 988 int close_ctree(struct btrfs_root *root)
 989 {
 990         int ret;
 991         struct btrfs_trans_handle *trans;
 992         struct btrfs_fs_info *fs_info = root->fs_info;
 993
 994         fs_info->closing = 1;
 995         btrfs_transaction_flush_work(root);
 996         mutex_lock(&fs_info->fs_mutex);
 997         btrfs_defrag_dirty_roots(root->fs_info);
 998         trans = btrfs_start_transaction(root, 1);
 999         ret = btrfs_commit_transaction(trans, root);
1000         /* run commit again to  drop the original snapshot */
1001         trans = btrfs_start_transaction(root, 1);
1002         btrfs_commit_transaction(trans, root);
1003         ret = btrfs_write_and_wait_transaction(NULL, root);
1004         BUG_ON(ret);
1005         write_ctree_super(NULL, root);
1006         mutex_unlock(&fs_info->fs_mutex);
1007
1008         if (fs_info->delalloc_bytes) {
1009                 printk("btrfs: at unmount delalloc count %Lu\n",
1010                        fs_info->delalloc_bytes);
1011         }
1012         if (fs_info->extent_root->node)
1013                 free_extent_buffer(fs_info->extent_root->node);
1014
1015         if (fs_info->tree_root->node)
1016                 free_extent_buffer(fs_info->tree_root->node);
1017
1018         if (root->fs_info->chunk_root->node);
1019                 free_extent_buffer(root->fs_info->chunk_root->node);
1020
1021         if (root->fs_info->dev_root->node);
1022                 free_extent_buffer(root->fs_info->dev_root->node);
1023
1024         free_extent_buffer(fs_info->sb_buffer);
1025
1026         btrfs_free_block_groups(root->fs_info);
1027         del_fs_roots(fs_info);
1028
1029         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1030
1031         extent_io_tree_empty_lru(&fs_info->free_space_cache);
1032         extent_io_tree_empty_lru(&fs_info->block_group_cache);
1033         extent_io_tree_empty_lru(&fs_info->pinned_extents);
1034         extent_io_tree_empty_lru(&fs_info->pending_del);
1035         extent_io_tree_empty_lru(&fs_info->extent_ins);
1036         extent_io_tree_empty_lru(&BTRFS_I(fs_info->btree_inode)->io_tree);
1037
1038         truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
1039
1040         iput(fs_info->btree_inode);
1041 #if 0
1042         while(!list_empty(&fs_info->hashers)) {
1043                 struct btrfs_hasher *hasher;
1044                 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
1045                                     hashers);
1046                 list_del(&hasher->hashers);
1047                 crypto_free_hash(&fs_info->hash_tfm);
1048                 kfree(hasher);
1049         }
1050 #endif
1051         close_all_devices(fs_info);
1052         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1053         bdi_destroy(&fs_info->bdi);
1054
1055         kfree(fs_info->extent_root);
1056         kfree(fs_info->tree_root);
1057         kfree(fs_info->chunk_root);
1058         kfree(fs_info->dev_root);
1059         return 0;
1060 }
1061
1062 int btrfs_buffer_uptodate(struct extent_buffer *buf)
1063 {
1064         struct inode *btree_inode = buf->first_page->mapping->host;
1065         return extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
1066 }
1067
1068 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
1069 {
1070         struct inode *btree_inode = buf->first_page->mapping->host;
1071         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
1072                                           buf);
1073 }
1074
1075 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
1076 {
1077         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1078         u64 transid = btrfs_header_generation(buf);
1079         struct inode *btree_inode = root->fs_info->btree_inode;
1080
1081         if (transid != root->fs_info->generation) {
1082                 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
1083                         (unsigned long long)buf->start,
1084                         transid, root->fs_info->generation);
1085                 WARN_ON(1);
1086         }
1087         set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
1088 }
1089
1090 void btrfs_throttle(struct btrfs_root *root)
1091 {
1092         struct backing_dev_info *bdi;
1093
1094         bdi = root->fs_info->sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
1095         if (root->fs_info->throttles && bdi_write_congested(bdi)) {
1096 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,18)
1097                 congestion_wait(WRITE, HZ/20);
1098 #else
1099                 blk_congestion_wait(WRITE, HZ/20);
1100 #endif
1101         }
1102 }
1103
1104 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
1105 {
1106         balance_dirty_pages_ratelimited_nr(
1107                                    root->fs_info->btree_inode->i_mapping, 1);
1108 }
1109
1110 void btrfs_set_buffer_defrag(struct extent_buffer *buf)
1111 {
1112         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1113         struct inode *btree_inode = root->fs_info->btree_inode;
1114         set_extent_bits(&BTRFS_I(btree_inode)->io_tree, buf->start,
1115                         buf->start + buf->len - 1, EXTENT_DEFRAG, GFP_NOFS);
1116 }
1117
1118 void btrfs_set_buffer_defrag_done(struct extent_buffer *buf)
1119 {
1120         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1121         struct inode *btree_inode = root->fs_info->btree_inode;
1122         set_extent_bits(&BTRFS_I(btree_inode)->io_tree, buf->start,
1123                         buf->start + buf->len - 1, EXTENT_DEFRAG_DONE,
1124                         GFP_NOFS);
1125 }
1126
1127 int btrfs_buffer_defrag(struct extent_buffer *buf)
1128 {
1129         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1130         struct inode *btree_inode = root->fs_info->btree_inode;
1131         return test_range_bit(&BTRFS_I(btree_inode)->io_tree,
1132                      buf->start, buf->start + buf->len - 1, EXTENT_DEFRAG, 0);
1133 }
1134
1135 int btrfs_buffer_defrag_done(struct extent_buffer *buf)
1136 {
1137         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1138         struct inode *btree_inode = root->fs_info->btree_inode;
1139         return test_range_bit(&BTRFS_I(btree_inode)->io_tree,
1140                      buf->start, buf->start + buf->len - 1,
1141                      EXTENT_DEFRAG_DONE, 0);
1142 }
1143
1144 int btrfs_clear_buffer_defrag_done(struct extent_buffer *buf)
1145 {
1146         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1147         struct inode *btree_inode = root->fs_info->btree_inode;
1148         return clear_extent_bits(&BTRFS_I(btree_inode)->io_tree,
1149                      buf->start, buf->start + buf->len - 1,
1150                      EXTENT_DEFRAG_DONE, GFP_NOFS);
1151 }
1152
1153 int btrfs_clear_buffer_defrag(struct extent_buffer *buf)
1154 {
1155         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1156         struct inode *btree_inode = root->fs_info->btree_inode;
1157         return clear_extent_bits(&BTRFS_I(btree_inode)->io_tree,
1158                      buf->start, buf->start + buf->len - 1,
1159                      EXTENT_DEFRAG, GFP_NOFS);
1160 }
1161
1162 int btrfs_read_buffer(struct extent_buffer *buf)
1163 {
1164         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1165         struct inode *btree_inode = root->fs_info->btree_inode;
1166         return read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1167                                         buf, 0, 1, btree_get_extent);
1168 }
1169
1170 static struct extent_io_ops btree_extent_io_ops = {
1171         .writepage_io_hook = btree_writepage_io_hook,
1172         .submit_bio_hook = btree_submit_bio_hook,
1173         /* note we're sharing with inode.c for the merge bio hook */
1174         .merge_bio_hook = btrfs_merge_bio_hook,
1175 };