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/module.h>
  20 #include <linux/fs.h>
  21 #include <linux/blkdev.h>
  22 #include <linux/crypto.h>
  23 #include <linux/scatterlist.h>
  24 #include <linux/swap.h>
  25 #include <linux/radix-tree.h>
  26 #include <linux/writeback.h>
  27 #include "ctree.h"
  28 #include "disk-io.h"
  29 #include "transaction.h"
  30 #include "btrfs_inode.h"
  31
  32 u64 bh_blocknr(struct buffer_head *bh)
  33 {
  34         return bh->b_blocknr;
  35 }
  36
  37 static int check_tree_block(struct btrfs_root *root, struct buffer_head *buf)
  38 {
  39         struct btrfs_node *node = btrfs_buffer_node(buf);
  40         if (bh_blocknr(buf) != btrfs_header_blocknr(&node->header)) {
  41                 printk(KERN_CRIT "bh_blocknr(buf) is %llu, header is %llu\n",
  42                        (unsigned long long)bh_blocknr(buf),
  43                        (unsigned long long)btrfs_header_blocknr(&node->header));
  44                 return 1;
  45         }
  46         return 0;
  47 }
  48
  49 struct buffer_head *btrfs_find_tree_block(struct btrfs_root *root, u64 blocknr)
  50 {
  51         struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
  52         int blockbits = root->fs_info->sb->s_blocksize_bits;
  53         unsigned long index = blocknr >> (PAGE_CACHE_SHIFT - blockbits);
  54         struct page *page;
  55         struct buffer_head *bh;
  56         struct buffer_head *head;
  57         struct buffer_head *ret = NULL;
  58
  59
  60         page = find_lock_page(mapping, index);
  61         if (!page)
  62                 return NULL;
  63
  64         if (!page_has_buffers(page))
  65                 goto out_unlock;
  66
  67         head = page_buffers(page);
  68         bh = head;
  69         do {
  70                 if (buffer_mapped(bh) && bh_blocknr(bh) == blocknr) {
  71                         ret = bh;
  72                         get_bh(bh);
  73                         goto out_unlock;
  74                 }
  75                 bh = bh->b_this_page;
  76         } while (bh != head);
  77 out_unlock:
  78         unlock_page(page);
  79         page_cache_release(page);
  80         return ret;
  81 }
  82
  83 int btrfs_map_bh_to_logical(struct btrfs_root *root, struct buffer_head *bh,
  84                              u64 logical)
  85 {
  86         if (logical == 0) {
  87                 bh->b_bdev = NULL;
  88                 bh->b_blocknr = 0;
  89                 set_buffer_mapped(bh);
  90         } else {
  91                 map_bh(bh, root->fs_info->sb, logical);
  92         }
  93         return 0;
  94 }
  95
  96 struct buffer_head *btrfs_find_create_tree_block(struct btrfs_root *root,
  97                                                  u64 blocknr)
  98 {
  99         struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
 100         int blockbits = root->fs_info->sb->s_blocksize_bits;
 101         unsigned long index = blocknr >> (PAGE_CACHE_SHIFT - blockbits);
 102         struct page *page;
 103         struct buffer_head *bh;
 104         struct buffer_head *head;
 105         struct buffer_head *ret = NULL;
 106         int err;
 107         u64 first_block = index << (PAGE_CACHE_SHIFT - blockbits);
 108
 109         page = find_or_create_page(mapping, index, GFP_NOFS);
 110         if (!page)
 111                 return NULL;
 112
 113         if (!page_has_buffers(page))
 114                 create_empty_buffers(page, root->fs_info->sb->s_blocksize, 0);
 115         head = page_buffers(page);
 116         bh = head;
 117         do {
 118                 if (!buffer_mapped(bh)) {
 119                         err = btrfs_map_bh_to_logical(root, bh, first_block);
 120                         BUG_ON(err);
 121                 }
 122                 if (bh_blocknr(bh) == blocknr) {
 123                         ret = bh;
 124                         get_bh(bh);
 125                         goto out_unlock;
 126                 }
 127                 bh = bh->b_this_page;
 128                 first_block++;
 129         } while (bh != head);
 130 out_unlock:
 131         unlock_page(page);
 132         if (ret)
 133                 touch_buffer(ret);
 134         page_cache_release(page);
 135         return ret;
 136 }
 137
 138 static int btree_get_block(struct inode *inode, sector_t iblock,
 139                            struct buffer_head *bh, int create)
 140 {
 141         int err;
 142         struct btrfs_root *root = BTRFS_I(bh->b_page->mapping->host)->root;
 143         err = btrfs_map_bh_to_logical(root, bh, iblock);
 144         return err;
 145 }
 146
 147 int btrfs_csum_data(struct btrfs_root * root, char *data, size_t len,
 148                     char *result)
 149 {
 150         struct scatterlist sg;
 151         struct crypto_hash *tfm = root->fs_info->hash_tfm;
 152         struct hash_desc desc;
 153         int ret;
 154
 155         desc.tfm = tfm;
 156         desc.flags = 0;
 157         sg_init_one(&sg, data, len);
 158         spin_lock(&root->fs_info->hash_lock);
 159         ret = crypto_hash_digest(&desc, &sg, 1, result);
 160         spin_unlock(&root->fs_info->hash_lock);
 161         if (ret) {
 162                 printk("digest failed\n");
 163         }
 164         return ret;
 165 }
 166 static int csum_tree_block(struct btrfs_root *root, struct buffer_head *bh,
 167                            int verify)
 168 {
 169         char result[BTRFS_CRC32_SIZE];
 170         int ret;
 171         struct btrfs_node *node;
 172
 173         ret = btrfs_csum_data(root, bh->b_data + BTRFS_CSUM_SIZE,
 174                               bh->b_size - BTRFS_CSUM_SIZE, result);
 175         if (ret)
 176                 return ret;
 177         if (verify) {
 178                 if (memcmp(bh->b_data, result, BTRFS_CRC32_SIZE)) {
 179                         printk("btrfs: %s checksum verify failed on %llu\n",
 180                                root->fs_info->sb->s_id,
 181                                (unsigned long long)bh_blocknr(bh));
 182                         return 1;
 183                 }
 184         } else {
 185                 node = btrfs_buffer_node(bh);
 186                 memcpy(node->header.csum, result, BTRFS_CRC32_SIZE);
 187         }
 188         return 0;
 189 }
 190
 191 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 192 {
 193         struct buffer_head *bh;
 194         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 195         struct buffer_head *head;
 196         if (!page_has_buffers(page)) {
 197                 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
 198                                         (1 << BH_Dirty)|(1 << BH_Uptodate));
 199         }
 200         head = page_buffers(page);
 201         bh = head;
 202         do {
 203                 if (buffer_dirty(bh))
 204                         csum_tree_block(root, bh, 0);
 205                 bh = bh->b_this_page;
 206         } while (bh != head);
 207         return block_write_full_page(page, btree_get_block, wbc);
 208 }
 209
 210 static int btree_readpage(struct file * file, struct page * page)
 211 {
 212         return block_read_full_page(page, btree_get_block);
 213 }
 214
 215 static struct address_space_operations btree_aops = {
 216         .readpage       = btree_readpage,
 217         .writepage      = btree_writepage,
 218         .sync_page      = block_sync_page,
 219 };
 220
 221 int readahead_tree_block(struct btrfs_root *root, u64 blocknr)
 222 {
 223         struct buffer_head *bh = NULL;
 224         int ret = 0;
 225
 226         bh = btrfs_find_create_tree_block(root, blocknr);
 227         if (!bh)
 228                 return 0;
 229         if (buffer_uptodate(bh)) {
 230                 ret = 1;
 231                 goto done;
 232         }
 233         if (test_set_buffer_locked(bh)) {
 234                 ret = 1;
 235                 goto done;
 236         }
 237         if (!buffer_uptodate(bh)) {
 238                 get_bh(bh);
 239                 bh->b_end_io = end_buffer_read_sync;
 240                 submit_bh(READ, bh);
 241         } else {
 242                 unlock_buffer(bh);
 243                 ret = 1;
 244         }
 245 done:
 246         brelse(bh);
 247         return ret;
 248 }
 249
 250 struct buffer_head *read_tree_block(struct btrfs_root *root, u64 blocknr)
 251 {
 252         struct buffer_head *bh = NULL;
 253
 254         bh = btrfs_find_create_tree_block(root, blocknr);
 255         if (!bh)
 256                 return bh;
 257         if (buffer_uptodate(bh))
 258                 goto uptodate;
 259         lock_buffer(bh);
 260         if (!buffer_uptodate(bh)) {
 261                 get_bh(bh);
 262                 bh->b_end_io = end_buffer_read_sync;
 263                 submit_bh(READ, bh);
 264                 wait_on_buffer(bh);
 265                 if (!buffer_uptodate(bh))
 266                         goto fail;
 267         } else {
 268                 unlock_buffer(bh);
 269         }
 270 uptodate:
 271         if (!buffer_checked(bh)) {
 272                 csum_tree_block(root, bh, 1);
 273                 set_buffer_checked(bh);
 274         }
 275         if (check_tree_block(root, bh))
 276                 goto fail;
 277         return bh;
 278 fail:
 279         brelse(bh);
 280         return NULL;
 281 }
 282
 283 int dirty_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 284                      struct buffer_head *buf)
 285 {
 286         WARN_ON(atomic_read(&buf->b_count) == 0);
 287         mark_buffer_dirty(buf);
 288         return 0;
 289 }
 290
 291 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 292                      struct buffer_head *buf)
 293 {
 294         WARN_ON(atomic_read(&buf->b_count) == 0);
 295         clear_buffer_dirty(buf);
 296         return 0;
 297 }
 298
 299 static int __setup_root(int blocksize,
 300                         struct btrfs_root *root,
 301                         struct btrfs_fs_info *fs_info,
 302                         u64 objectid)
 303 {
 304         root->node = NULL;
 305         root->inode = NULL;
 306         root->commit_root = NULL;
 307         root->blocksize = blocksize;
 308         root->ref_cows = 0;
 309         root->fs_info = fs_info;
 310         root->objectid = objectid;
 311         root->last_trans = 0;
 312         root->highest_inode = 0;
 313         root->last_inode_alloc = 0;
 314         memset(&root->root_key, 0, sizeof(root->root_key));
 315         memset(&root->root_item, 0, sizeof(root->root_item));
 316         root->root_key.objectid = objectid;
 317         return 0;
 318 }
 319
 320 static int find_and_setup_root(int blocksize,
 321                                struct btrfs_root *tree_root,
 322                                struct btrfs_fs_info *fs_info,
 323                                u64 objectid,
 324                                struct btrfs_root *root)
 325 {
 326         int ret;
 327
 328         __setup_root(blocksize, root, fs_info, objectid);
 329         ret = btrfs_find_last_root(tree_root, objectid,
 330                                    &root->root_item, &root->root_key);
 331         BUG_ON(ret);
 332
 333         root->node = read_tree_block(root,
 334                                      btrfs_root_blocknr(&root->root_item));
 335         BUG_ON(!root->node);
 336         return 0;
 337 }
 338
 339 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
 340                                       struct btrfs_key *location)
 341 {
 342         struct btrfs_root *root;
 343         struct btrfs_root *tree_root = fs_info->tree_root;
 344         struct btrfs_path *path;
 345         struct btrfs_leaf *l;
 346         u64 highest_inode;
 347         int ret = 0;
 348
 349         root = radix_tree_lookup(&fs_info->fs_roots_radix,
 350                                  (unsigned long)location->objectid);
 351         if (root)
 352                 return root;
 353         root = kmalloc(sizeof(*root), GFP_NOFS);
 354         if (!root)
 355                 return ERR_PTR(-ENOMEM);
 356         if (location->offset == (u64)-1) {
 357                 ret = find_and_setup_root(fs_info->sb->s_blocksize,
 358                                           fs_info->tree_root, fs_info,
 359                                           location->objectid, root);
 360                 if (ret) {
 361                         kfree(root);
 362                         return ERR_PTR(ret);
 363                 }
 364                 goto insert;
 365         }
 366
 367         __setup_root(fs_info->sb->s_blocksize, root, fs_info,
 368                      location->objectid);
 369
 370         path = btrfs_alloc_path();
 371         BUG_ON(!path);
 372         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
 373         if (ret != 0) {
 374                 if (ret > 0)
 375                         ret = -ENOENT;
 376                 goto out;
 377         }
 378         l = btrfs_buffer_leaf(path->nodes[0]);
 379         memcpy(&root->root_item,
 380                btrfs_item_ptr(l, path->slots[0], struct btrfs_root_item),
 381                sizeof(root->root_item));
 382         memcpy(&root->root_key, location, sizeof(*location));
 383         ret = 0;
 384 out:
 385         btrfs_release_path(root, path);
 386         btrfs_free_path(path);
 387         if (ret) {
 388                 kfree(root);
 389                 return ERR_PTR(ret);
 390         }
 391         root->node = read_tree_block(root,
 392                                      btrfs_root_blocknr(&root->root_item));
 393         BUG_ON(!root->node);
 394 insert:
 395         root->ref_cows = 1;
 396         ret = radix_tree_insert(&fs_info->fs_roots_radix,
 397                                 (unsigned long)root->root_key.objectid,
 398                                 root);
 399         if (ret) {
 400                 brelse(root->node);
 401                 kfree(root);
 402                 return ERR_PTR(ret);
 403         }
 404         ret = btrfs_find_highest_inode(root, &highest_inode);
 405         if (ret == 0) {
 406                 root->highest_inode = highest_inode;
 407                 root->last_inode_alloc = highest_inode;
 408         }
 409         return root;
 410 }
 411
 412 struct btrfs_root *open_ctree(struct super_block *sb)
 413 {
 414         struct btrfs_root *extent_root = kmalloc(sizeof(struct btrfs_root),
 415                                                  GFP_NOFS);
 416         struct btrfs_root *tree_root = kmalloc(sizeof(struct btrfs_root),
 417                                                GFP_NOFS);
 418         struct btrfs_fs_info *fs_info = kmalloc(sizeof(*fs_info),
 419                                                 GFP_NOFS);
 420         int ret;
 421         int err = -EIO;
 422         struct btrfs_super_block *disk_super;
 423
 424         if (!extent_root || !tree_root || !fs_info) {
 425                 err = -ENOMEM;
 426                 goto fail;
 427         }
 428         init_bit_radix(&fs_info->pinned_radix);
 429         init_bit_radix(&fs_info->pending_del_radix);
 430         init_bit_radix(&fs_info->extent_map_radix);
 431         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
 432         INIT_RADIX_TREE(&fs_info->block_group_radix, GFP_KERNEL);
 433         INIT_RADIX_TREE(&fs_info->block_group_data_radix, GFP_KERNEL);
 434         INIT_LIST_HEAD(&fs_info->trans_list);
 435         INIT_LIST_HEAD(&fs_info->dead_roots);
 436         sb_set_blocksize(sb, 4096);
 437         fs_info->running_transaction = NULL;
 438         fs_info->tree_root = tree_root;
 439         fs_info->extent_root = extent_root;
 440         fs_info->sb = sb;
 441         fs_info->btree_inode = new_inode(sb);
 442         fs_info->btree_inode->i_ino = 1;
 443         fs_info->btree_inode->i_nlink = 1;
 444         fs_info->btree_inode->i_size = sb->s_bdev->bd_inode->i_size;
 445         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
 446         fs_info->do_barriers = 1;
 447         fs_info->extent_tree_insert_nr = 0;
 448         fs_info->extent_tree_prealloc_nr = 0;
 449         fs_info->closing = 0;
 450
 451         INIT_DELAYED_WORK(&fs_info->trans_work, btrfs_transaction_cleaner);
 452         BTRFS_I(fs_info->btree_inode)->root = tree_root;
 453         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
 454                sizeof(struct btrfs_key));
 455         insert_inode_hash(fs_info->btree_inode);
 456         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
 457         fs_info->hash_tfm = crypto_alloc_hash("crc32c", 0, CRYPTO_ALG_ASYNC);
 458         spin_lock_init(&fs_info->hash_lock);
 459
 460         if (!fs_info->hash_tfm || IS_ERR(fs_info->hash_tfm)) {
 461                 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
 462                 err = -ENOMEM;
 463                 goto fail_iput;
 464         }
 465         mutex_init(&fs_info->trans_mutex);
 466         mutex_init(&fs_info->fs_mutex);
 467
 468         __setup_root(sb->s_blocksize, tree_root,
 469                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
 470
 471         fs_info->sb_buffer = read_tree_block(tree_root,
 472                                              BTRFS_SUPER_INFO_OFFSET /
 473                                              sb->s_blocksize);
 474
 475         if (!fs_info->sb_buffer)
 476                 goto fail_iput;
 477         disk_super = (struct btrfs_super_block *)fs_info->sb_buffer->b_data;
 478
 479         if (!btrfs_super_root(disk_super))
 480                 goto fail_sb_buffer;
 481
 482         i_size_write(fs_info->btree_inode,
 483                      btrfs_super_total_blocks(disk_super) <<
 484                      fs_info->btree_inode->i_blkbits);
 485
 486         fs_info->disk_super = disk_super;
 487
 488         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
 489                     sizeof(disk_super->magic))) {
 490                 printk("btrfs: valid FS not found on %s\n", sb->s_id);
 491                 goto fail_sb_buffer;
 492         }
 493         tree_root->node = read_tree_block(tree_root,
 494                                           btrfs_super_root(disk_super));
 495         if (!tree_root->node)
 496                 goto fail_sb_buffer;
 497
 498         mutex_lock(&fs_info->fs_mutex);
 499         ret = find_and_setup_root(sb->s_blocksize, tree_root, fs_info,
 500                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
 501         if (ret) {
 502                 mutex_unlock(&fs_info->fs_mutex);
 503                 goto fail_tree_root;
 504         }
 505
 506         btrfs_read_block_groups(extent_root);
 507
 508         fs_info->generation = btrfs_super_generation(disk_super) + 1;
 509         mutex_unlock(&fs_info->fs_mutex);
 510         return tree_root;
 511
 512 fail_tree_root:
 513         btrfs_block_release(tree_root, tree_root->node);
 514 fail_sb_buffer:
 515         btrfs_block_release(tree_root, fs_info->sb_buffer);
 516 fail_iput:
 517         iput(fs_info->btree_inode);
 518 fail:
 519         kfree(extent_root);
 520         kfree(tree_root);
 521         kfree(fs_info);
 522         return ERR_PTR(err);
 523 }
 524
 525 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
 526                       *root)
 527 {
 528         int ret;
 529         struct buffer_head *bh = root->fs_info->sb_buffer;
 530
 531         btrfs_set_super_root(root->fs_info->disk_super,
 532                              bh_blocknr(root->fs_info->tree_root->node));
 533         lock_buffer(bh);
 534         WARN_ON(atomic_read(&bh->b_count) < 1);
 535         clear_buffer_dirty(bh);
 536         csum_tree_block(root, bh, 0);
 537         bh->b_end_io = end_buffer_write_sync;
 538         get_bh(bh);
 539         if (root->fs_info->do_barriers)
 540                 ret = submit_bh(WRITE_BARRIER, bh);
 541         else
 542                 ret = submit_bh(WRITE, bh);
 543         if (ret == -EOPNOTSUPP) {
 544                 set_buffer_uptodate(bh);
 545                 root->fs_info->do_barriers = 0;
 546                 ret = submit_bh(WRITE, bh);
 547         }
 548         wait_on_buffer(bh);
 549         if (!buffer_uptodate(bh)) {
 550                 WARN_ON(1);
 551                 return -EIO;
 552         }
 553         return 0;
 554 }
 555
 556 static int free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
 557 {
 558         radix_tree_delete(&fs_info->fs_roots_radix,
 559                           (unsigned long)root->root_key.objectid);
 560         if (root->inode)
 561                 iput(root->inode);
 562         if (root->node)
 563                 brelse(root->node);
 564         if (root->commit_root)
 565                 brelse(root->commit_root);
 566         kfree(root);
 567         return 0;
 568 }
 569
 570 static int del_fs_roots(struct btrfs_fs_info *fs_info)
 571 {
 572         int ret;
 573         struct btrfs_root *gang[8];
 574         int i;
 575
 576         while(1) {
 577                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
 578                                              (void **)gang, 0,
 579                                              ARRAY_SIZE(gang));
 580                 if (!ret)
 581                         break;
 582                 for (i = 0; i < ret; i++)
 583                         free_fs_root(fs_info, gang[i]);
 584         }
 585         return 0;
 586 }
 587
 588 int close_ctree(struct btrfs_root *root)
 589 {
 590         int ret;
 591         struct btrfs_trans_handle *trans;
 592         struct btrfs_fs_info *fs_info = root->fs_info;
 593
 594         fs_info->closing = 1;
 595         btrfs_transaction_flush_work(root);
 596         mutex_lock(&fs_info->fs_mutex);
 597         trans = btrfs_start_transaction(root, 1);
 598         btrfs_commit_transaction(trans, root);
 599         /* run commit again to  drop the original snapshot */
 600         trans = btrfs_start_transaction(root, 1);
 601         btrfs_commit_transaction(trans, root);
 602         ret = btrfs_write_and_wait_transaction(NULL, root);
 603         BUG_ON(ret);
 604         write_ctree_super(NULL, root);
 605         mutex_unlock(&fs_info->fs_mutex);
 606
 607         if (fs_info->extent_root->node)
 608                 btrfs_block_release(fs_info->extent_root,
 609                                     fs_info->extent_root->node);
 610         if (fs_info->tree_root->node)
 611                 btrfs_block_release(fs_info->tree_root,
 612                                     fs_info->tree_root->node);
 613         btrfs_block_release(root, fs_info->sb_buffer);
 614         crypto_free_hash(fs_info->hash_tfm);
 615         truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
 616         iput(fs_info->btree_inode);
 617
 618         btrfs_free_block_groups(root->fs_info);
 619         del_fs_roots(fs_info);
 620         kfree(fs_info->extent_root);
 621         kfree(fs_info->tree_root);
 622         return 0;
 623 }
 624
 625 void btrfs_block_release(struct btrfs_root *root, struct buffer_head *buf)
 626 {
 627         brelse(buf);
 628 }
 629
 630 void btrfs_btree_balance_dirty(struct btrfs_root *root)
 631 {
 632         balance_dirty_pages_ratelimited(root->fs_info->btree_inode->i_mapping);
 633 }