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 "ctree.h"
  27 #include "disk-io.h"
  28 #include "transaction.h"
  29 #include "btrfs_inode.h"
  30
  31 u64 bh_blocknr(struct buffer_head *bh)
  32 {
  33         return bh->b_blocknr;
  34 }
  35
  36 static int check_tree_block(struct btrfs_root *root, struct buffer_head *buf)
  37 {
  38         struct btrfs_node *node = btrfs_buffer_node(buf);
  39         if (bh_blocknr(buf) != btrfs_header_blocknr(&node->header)) {
  40                 printk(KERN_CRIT "bh_blocknr(buf) is %llu, header is %llu\n",
  41                        (unsigned long long)bh_blocknr(buf),
  42                        (unsigned long long)btrfs_header_blocknr(&node->header));
  43                 return 1;
  44         }
  45         return 0;
  46 }
  47
  48 struct buffer_head *btrfs_find_tree_block(struct btrfs_root *root, u64 blocknr)
  49 {
  50         struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
  51         int blockbits = root->fs_info->sb->s_blocksize_bits;
  52         unsigned long index = blocknr >> (PAGE_CACHE_SHIFT - blockbits);
  53         struct page *page;
  54         struct buffer_head *bh;
  55         struct buffer_head *head;
  56         struct buffer_head *ret = NULL;
  57
  58
  59         page = find_lock_page(mapping, index);
  60         if (!page)
  61                 return NULL;
  62
  63         if (!page_has_buffers(page))
  64                 goto out_unlock;
  65
  66         head = page_buffers(page);
  67         bh = head;
  68         do {
  69                 if (buffer_mapped(bh) && bh_blocknr(bh) == blocknr) {
  70                         ret = bh;
  71                         get_bh(bh);
  72                         goto out_unlock;
  73                 }
  74                 bh = bh->b_this_page;
  75         } while (bh != head);
  76 out_unlock:
  77         unlock_page(page);
  78         page_cache_release(page);
  79         return ret;
  80 }
  81
  82 int btrfs_map_bh_to_logical(struct btrfs_root *root, struct buffer_head *bh,
  83                              u64 logical)
  84 {
  85         if (logical == 0) {
  86                 bh->b_bdev = NULL;
  87                 bh->b_blocknr = 0;
  88                 set_buffer_mapped(bh);
  89         } else {
  90                 map_bh(bh, root->fs_info->sb, logical);
  91         }
  92         return 0;
  93 }
  94
  95 struct buffer_head *btrfs_find_create_tree_block(struct btrfs_root *root,
  96                                                  u64 blocknr)
  97 {
  98         struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
  99         int blockbits = root->fs_info->sb->s_blocksize_bits;
 100         unsigned long index = blocknr >> (PAGE_CACHE_SHIFT - blockbits);
 101         struct page *page;
 102         struct buffer_head *bh;
 103         struct buffer_head *head;
 104         struct buffer_head *ret = NULL;
 105         int err;
 106         u64 first_block = index << (PAGE_CACHE_SHIFT - blockbits);
 107
 108         page = find_or_create_page(mapping, index, GFP_NOFS);
 109         if (!page)
 110                 return NULL;
 111
 112         if (!page_has_buffers(page))
 113                 create_empty_buffers(page, root->fs_info->sb->s_blocksize, 0);
 114         head = page_buffers(page);
 115         bh = head;
 116         do {
 117                 if (!buffer_mapped(bh)) {
 118                         err = btrfs_map_bh_to_logical(root, bh, first_block);
 119                         BUG_ON(err);
 120                 }
 121                 if (bh_blocknr(bh) == blocknr) {
 122                         ret = bh;
 123                         get_bh(bh);
 124                         goto out_unlock;
 125                 }
 126                 bh = bh->b_this_page;
 127                 first_block++;
 128         } while (bh != head);
 129 out_unlock:
 130         unlock_page(page);
 131         if (ret)
 132                 touch_buffer(ret);
 133         page_cache_release(page);
 134         return ret;
 135 }
 136
 137 static int btree_get_block(struct inode *inode, sector_t iblock,
 138                            struct buffer_head *bh, int create)
 139 {
 140         int err;
 141         struct btrfs_root *root = BTRFS_I(bh->b_page->mapping->host)->root;
 142         err = btrfs_map_bh_to_logical(root, bh, iblock);
 143         return err;
 144 }
 145
 146 int btrfs_csum_data(struct btrfs_root * root, char *data, size_t len,
 147                     char *result)
 148 {
 149         u32 crc;
 150         crc = crc32c(0, data, len);
 151         memcpy(result, &crc, BTRFS_CRC32_SIZE);
 152         return 0;
 153 }
 154
 155 static int csum_tree_block(struct btrfs_root *root, struct buffer_head *bh,
 156                            int verify)
 157 {
 158         char result[BTRFS_CRC32_SIZE];
 159         int ret;
 160         struct btrfs_node *node;
 161
 162         ret = btrfs_csum_data(root, bh->b_data + BTRFS_CSUM_SIZE,
 163                               bh->b_size - BTRFS_CSUM_SIZE, result);
 164         if (ret)
 165                 return ret;
 166         if (verify) {
 167                 if (memcmp(bh->b_data, result, BTRFS_CRC32_SIZE)) {
 168                         printk("btrfs: %s checksum verify failed on %llu\n",
 169                                root->fs_info->sb->s_id,
 170                                (unsigned long long)bh_blocknr(bh));
 171                         return 1;
 172                 }
 173         } else {
 174                 node = btrfs_buffer_node(bh);
 175                 memcpy(node->header.csum, result, BTRFS_CRC32_SIZE);
 176         }
 177         return 0;
 178 }
 179
 180 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 181 {
 182         struct buffer_head *bh;
 183         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 184         struct buffer_head *head;
 185         if (!page_has_buffers(page)) {
 186                 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
 187                                         (1 << BH_Dirty)|(1 << BH_Uptodate));
 188         }
 189         head = page_buffers(page);
 190         bh = head;
 191         do {
 192                 if (buffer_dirty(bh))
 193                         csum_tree_block(root, bh, 0);
 194                 bh = bh->b_this_page;
 195         } while (bh != head);
 196         return block_write_full_page(page, btree_get_block, wbc);
 197 }
 198
 199 static int btree_readpage(struct file * file, struct page * page)
 200 {
 201         return block_read_full_page(page, btree_get_block);
 202 }
 203
 204 static struct address_space_operations btree_aops = {
 205         .readpage       = btree_readpage,
 206         .writepage      = btree_writepage,
 207         .sync_page      = block_sync_page,
 208 };
 209
 210 int readahead_tree_block(struct btrfs_root *root, u64 blocknr)
 211 {
 212         struct buffer_head *bh = NULL;
 213         int ret = 0;
 214
 215         bh = btrfs_find_create_tree_block(root, blocknr);
 216         if (!bh)
 217                 return 0;
 218         if (buffer_uptodate(bh)) {
 219                 ret = 1;
 220                 goto done;
 221         }
 222         if (test_set_buffer_locked(bh)) {
 223                 ret = 1;
 224                 goto done;
 225         }
 226         if (!buffer_uptodate(bh)) {
 227                 get_bh(bh);
 228                 bh->b_end_io = end_buffer_read_sync;
 229                 submit_bh(READ, bh);
 230         } else {
 231                 unlock_buffer(bh);
 232                 ret = 1;
 233         }
 234 done:
 235         brelse(bh);
 236         return ret;
 237 }
 238
 239 struct buffer_head *read_tree_block(struct btrfs_root *root, u64 blocknr)
 240 {
 241         struct buffer_head *bh = NULL;
 242
 243         bh = btrfs_find_create_tree_block(root, blocknr);
 244         if (!bh)
 245                 return bh;
 246         if (buffer_uptodate(bh))
 247                 goto uptodate;
 248         lock_buffer(bh);
 249         if (!buffer_uptodate(bh)) {
 250                 get_bh(bh);
 251                 bh->b_end_io = end_buffer_read_sync;
 252                 submit_bh(READ, bh);
 253                 wait_on_buffer(bh);
 254                 if (!buffer_uptodate(bh))
 255                         goto fail;
 256         } else {
 257                 unlock_buffer(bh);
 258         }
 259 uptodate:
 260         if (!buffer_checked(bh)) {
 261                 csum_tree_block(root, bh, 1);
 262                 set_buffer_checked(bh);
 263         }
 264         if (check_tree_block(root, bh))
 265                 goto fail;
 266         return bh;
 267 fail:
 268         brelse(bh);
 269         return NULL;
 270 }
 271
 272 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 273                      struct buffer_head *buf)
 274 {
 275         WARN_ON(atomic_read(&buf->b_count) == 0);
 276         clear_buffer_dirty(buf);
 277         return 0;
 278 }
 279
 280 static int __setup_root(int blocksize,
 281                         struct btrfs_root *root,
 282                         struct btrfs_fs_info *fs_info,
 283                         u64 objectid)
 284 {
 285         root->node = NULL;
 286         root->inode = NULL;
 287         root->commit_root = NULL;
 288         root->blocksize = blocksize;
 289         root->ref_cows = 0;
 290         root->fs_info = fs_info;
 291         root->objectid = objectid;
 292         root->last_trans = 0;
 293         root->highest_inode = 0;
 294         root->last_inode_alloc = 0;
 295         memset(&root->root_key, 0, sizeof(root->root_key));
 296         memset(&root->root_item, 0, sizeof(root->root_item));
 297         root->root_key.objectid = objectid;
 298         return 0;
 299 }
 300
 301 static int find_and_setup_root(int blocksize,
 302                                struct btrfs_root *tree_root,
 303                                struct btrfs_fs_info *fs_info,
 304                                u64 objectid,
 305                                struct btrfs_root *root)
 306 {
 307         int ret;
 308
 309         __setup_root(blocksize, root, fs_info, objectid);
 310         ret = btrfs_find_last_root(tree_root, objectid,
 311                                    &root->root_item, &root->root_key);
 312         BUG_ON(ret);
 313
 314         root->node = read_tree_block(root,
 315                                      btrfs_root_blocknr(&root->root_item));
 316         BUG_ON(!root->node);
 317         return 0;
 318 }
 319
 320 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_fs_info *fs_info,
 321                                                struct btrfs_key *location)
 322 {
 323         struct btrfs_root *root;
 324         struct btrfs_root *tree_root = fs_info->tree_root;
 325         struct btrfs_path *path;
 326         struct btrfs_leaf *l;
 327         u64 highest_inode;
 328         int ret = 0;
 329
 330         root = kzalloc(sizeof(*root), GFP_NOFS);
 331         if (!root)
 332                 return ERR_PTR(-ENOMEM);
 333         if (location->offset == (u64)-1) {
 334                 ret = find_and_setup_root(fs_info->sb->s_blocksize,
 335                                           fs_info->tree_root, fs_info,
 336                                           location->objectid, root);
 337                 if (ret) {
 338                         kfree(root);
 339                         return ERR_PTR(ret);
 340                 }
 341                 goto insert;
 342         }
 343
 344         __setup_root(fs_info->sb->s_blocksize, root, fs_info,
 345                      location->objectid);
 346
 347         path = btrfs_alloc_path();
 348         BUG_ON(!path);
 349         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
 350         if (ret != 0) {
 351                 if (ret > 0)
 352                         ret = -ENOENT;
 353                 goto out;
 354         }
 355         l = btrfs_buffer_leaf(path->nodes[0]);
 356         memcpy(&root->root_item,
 357                btrfs_item_ptr(l, path->slots[0], struct btrfs_root_item),
 358                sizeof(root->root_item));
 359         memcpy(&root->root_key, location, sizeof(*location));
 360         ret = 0;
 361 out:
 362         btrfs_release_path(root, path);
 363         btrfs_free_path(path);
 364         if (ret) {
 365                 kfree(root);
 366                 return ERR_PTR(ret);
 367         }
 368         root->node = read_tree_block(root,
 369                                      btrfs_root_blocknr(&root->root_item));
 370         BUG_ON(!root->node);
 371 insert:
 372         root->ref_cows = 1;
 373         ret = btrfs_find_highest_inode(root, &highest_inode);
 374         if (ret == 0) {
 375                 root->highest_inode = highest_inode;
 376                 root->last_inode_alloc = highest_inode;
 377         }
 378         return root;
 379 }
 380
 381 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
 382                                       struct btrfs_key *location)
 383 {
 384         struct btrfs_root *root;
 385         int ret;
 386
 387         root = radix_tree_lookup(&fs_info->fs_roots_radix,
 388                                  (unsigned long)location->objectid);
 389         if (root)
 390                 return root;
 391
 392         root = btrfs_read_fs_root_no_radix(fs_info, location);
 393         if (IS_ERR(root))
 394                 return root;
 395         ret = radix_tree_insert(&fs_info->fs_roots_radix,
 396                                 (unsigned long)root->root_key.objectid,
 397                                 root);
 398         if (ret) {
 399                 brelse(root->node);
 400                 kfree(root);
 401                 return ERR_PTR(ret);
 402         }
 403         return root;
 404 }
 405
 406 struct btrfs_root *open_ctree(struct super_block *sb)
 407 {
 408         struct btrfs_root *extent_root = kmalloc(sizeof(struct btrfs_root),
 409                                                  GFP_NOFS);
 410         struct btrfs_root *tree_root = kmalloc(sizeof(struct btrfs_root),
 411                                                GFP_NOFS);
 412         struct btrfs_fs_info *fs_info = kmalloc(sizeof(*fs_info),
 413                                                 GFP_NOFS);
 414         int ret;
 415         int err = -EIO;
 416         struct btrfs_super_block *disk_super;
 417
 418         if (!extent_root || !tree_root || !fs_info) {
 419                 err = -ENOMEM;
 420                 goto fail;
 421         }
 422         init_bit_radix(&fs_info->pinned_radix);
 423         init_bit_radix(&fs_info->pending_del_radix);
 424         init_bit_radix(&fs_info->extent_map_radix);
 425         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
 426         INIT_RADIX_TREE(&fs_info->block_group_radix, GFP_KERNEL);
 427         INIT_RADIX_TREE(&fs_info->block_group_data_radix, GFP_KERNEL);
 428         INIT_LIST_HEAD(&fs_info->trans_list);
 429         INIT_LIST_HEAD(&fs_info->dead_roots);
 430         sb_set_blocksize(sb, 4096);
 431         fs_info->running_transaction = NULL;
 432         fs_info->tree_root = tree_root;
 433         fs_info->extent_root = extent_root;
 434         fs_info->sb = sb;
 435         fs_info->btree_inode = new_inode(sb);
 436         fs_info->btree_inode->i_ino = 1;
 437         fs_info->btree_inode->i_nlink = 1;
 438         fs_info->btree_inode->i_size = sb->s_bdev->bd_inode->i_size;
 439         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
 440         fs_info->do_barriers = 1;
 441         fs_info->extent_tree_insert_nr = 0;
 442         fs_info->extent_tree_prealloc_nr = 0;
 443         fs_info->closing = 0;
 444
 445         INIT_DELAYED_WORK(&fs_info->trans_work, btrfs_transaction_cleaner);
 446         BTRFS_I(fs_info->btree_inode)->root = tree_root;
 447         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
 448                sizeof(struct btrfs_key));
 449         insert_inode_hash(fs_info->btree_inode);
 450         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
 451
 452         mutex_init(&fs_info->trans_mutex);
 453         mutex_init(&fs_info->fs_mutex);
 454
 455         __setup_root(sb->s_blocksize, tree_root,
 456                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
 457
 458         fs_info->sb_buffer = read_tree_block(tree_root,
 459                                              BTRFS_SUPER_INFO_OFFSET /
 460                                              sb->s_blocksize);
 461
 462         if (!fs_info->sb_buffer)
 463                 goto fail_iput;
 464         disk_super = (struct btrfs_super_block *)fs_info->sb_buffer->b_data;
 465         fs_info->disk_super = disk_super;
 466         memcpy(&fs_info->super_copy, disk_super, sizeof(fs_info->super_copy));
 467
 468         if (!btrfs_super_root(disk_super))
 469                 goto fail_sb_buffer;
 470
 471         i_size_write(fs_info->btree_inode,
 472                      btrfs_super_total_blocks(disk_super) <<
 473                      fs_info->btree_inode->i_blkbits);
 474
 475
 476         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
 477                     sizeof(disk_super->magic))) {
 478                 printk("btrfs: valid FS not found on %s\n", sb->s_id);
 479                 goto fail_sb_buffer;
 480         }
 481         tree_root->node = read_tree_block(tree_root,
 482                                           btrfs_super_root(disk_super));
 483         if (!tree_root->node)
 484                 goto fail_sb_buffer;
 485
 486         mutex_lock(&fs_info->fs_mutex);
 487         ret = find_and_setup_root(sb->s_blocksize, tree_root, fs_info,
 488                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
 489         if (ret) {
 490                 mutex_unlock(&fs_info->fs_mutex);
 491                 goto fail_tree_root;
 492         }
 493
 494         btrfs_read_block_groups(extent_root);
 495
 496         fs_info->generation = btrfs_super_generation(disk_super) + 1;
 497         ret = btrfs_find_dead_roots(tree_root);
 498         if (ret)
 499                 goto fail_tree_root;
 500         mutex_unlock(&fs_info->fs_mutex);
 501         return tree_root;
 502
 503 fail_tree_root:
 504         btrfs_block_release(tree_root, tree_root->node);
 505 fail_sb_buffer:
 506         btrfs_block_release(tree_root, fs_info->sb_buffer);
 507 fail_iput:
 508         iput(fs_info->btree_inode);
 509 fail:
 510         kfree(extent_root);
 511         kfree(tree_root);
 512         kfree(fs_info);
 513         return ERR_PTR(err);
 514 }
 515
 516 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
 517                       *root)
 518 {
 519         int ret;
 520         struct buffer_head *bh = root->fs_info->sb_buffer;
 521
 522         lock_buffer(bh);
 523         WARN_ON(atomic_read(&bh->b_count) < 1);
 524         clear_buffer_dirty(bh);
 525         csum_tree_block(root, bh, 0);
 526         bh->b_end_io = end_buffer_write_sync;
 527         get_bh(bh);
 528         if (root->fs_info->do_barriers)
 529                 ret = submit_bh(WRITE_BARRIER, bh);
 530         else
 531                 ret = submit_bh(WRITE, bh);
 532         if (ret == -EOPNOTSUPP) {
 533                 get_bh(bh);
 534                 lock_buffer(bh);
 535                 set_buffer_uptodate(bh);
 536                 root->fs_info->do_barriers = 0;
 537                 ret = submit_bh(WRITE, bh);
 538         }
 539         wait_on_buffer(bh);
 540         if (!buffer_uptodate(bh)) {
 541                 WARN_ON(1);
 542                 return -EIO;
 543         }
 544         return 0;
 545 }
 546
 547 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
 548 {
 549         radix_tree_delete(&fs_info->fs_roots_radix,
 550                           (unsigned long)root->root_key.objectid);
 551         if (root->inode)
 552                 iput(root->inode);
 553         if (root->node)
 554                 brelse(root->node);
 555         if (root->commit_root)
 556                 brelse(root->commit_root);
 557         kfree(root);
 558         return 0;
 559 }
 560
 561 static int del_fs_roots(struct btrfs_fs_info *fs_info)
 562 {
 563         int ret;
 564         struct btrfs_root *gang[8];
 565         int i;
 566
 567         while(1) {
 568                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
 569                                              (void **)gang, 0,
 570                                              ARRAY_SIZE(gang));
 571                 if (!ret)
 572                         break;
 573                 for (i = 0; i < ret; i++)
 574                         btrfs_free_fs_root(fs_info, gang[i]);
 575         }
 576         return 0;
 577 }
 578
 579 int close_ctree(struct btrfs_root *root)
 580 {
 581         int ret;
 582         struct btrfs_trans_handle *trans;
 583         struct btrfs_fs_info *fs_info = root->fs_info;
 584
 585         fs_info->closing = 1;
 586         btrfs_transaction_flush_work(root);
 587         mutex_lock(&fs_info->fs_mutex);
 588         trans = btrfs_start_transaction(root, 1);
 589         ret = btrfs_commit_transaction(trans, root);
 590         /* run commit again to  drop the original snapshot */
 591         trans = btrfs_start_transaction(root, 1);
 592         btrfs_commit_transaction(trans, root);
 593         ret = btrfs_write_and_wait_transaction(NULL, root);
 594         BUG_ON(ret);
 595         write_ctree_super(NULL, root);
 596         mutex_unlock(&fs_info->fs_mutex);
 597
 598         if (fs_info->extent_root->node)
 599                 btrfs_block_release(fs_info->extent_root,
 600                                     fs_info->extent_root->node);
 601         if (fs_info->tree_root->node)
 602                 btrfs_block_release(fs_info->tree_root,
 603                                     fs_info->tree_root->node);
 604         btrfs_block_release(root, fs_info->sb_buffer);
 605         truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
 606         iput(fs_info->btree_inode);
 607
 608         btrfs_free_block_groups(root->fs_info);
 609         del_fs_roots(fs_info);
 610         kfree(fs_info->extent_root);
 611         kfree(fs_info->tree_root);
 612         return 0;
 613 }
 614
 615 void btrfs_mark_buffer_dirty(struct buffer_head *bh)
 616 {
 617         struct btrfs_root *root = BTRFS_I(bh->b_page->mapping->host)->root;
 618         u64 transid = btrfs_header_generation(btrfs_buffer_header(bh));
 619         WARN_ON(!atomic_read(&bh->b_count));
 620         if (transid != root->fs_info->generation) {
 621                 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
 622                         (unsigned long long)bh->b_blocknr,
 623                         transid, root->fs_info->generation);
 624                 WARN_ON(1);
 625         }
 626         mark_buffer_dirty(bh);
 627 }
 628
 629 void btrfs_block_release(struct btrfs_root *root, struct buffer_head *buf)
 630 {
 631         brelse(buf);
 632 }
 633
 634 void btrfs_btree_balance_dirty(struct btrfs_root *root)
 635 {
 636         balance_dirty_pages_ratelimited(root->fs_info->btree_inode->i_mapping);
 637 }