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/crc32c.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         u32 crc;
 151         crc = crc32c(0, data, len);
 152         memcpy(result, &crc, BTRFS_CRC32_SIZE);
 153         return 0;
 154 }
 155
 156 static int csum_tree_block(struct btrfs_root *root, struct buffer_head *bh,
 157                            int verify)
 158 {
 159         char result[BTRFS_CRC32_SIZE];
 160         int ret;
 161         struct btrfs_node *node;
 162
 163         ret = btrfs_csum_data(root, bh->b_data + BTRFS_CSUM_SIZE,
 164                               bh->b_size - BTRFS_CSUM_SIZE, result);
 165         if (ret)
 166                 return ret;
 167         if (verify) {
 168                 if (memcmp(bh->b_data, result, BTRFS_CRC32_SIZE)) {
 169                         printk("btrfs: %s checksum verify failed on %llu\n",
 170                                root->fs_info->sb->s_id,
 171                                (unsigned long long)bh_blocknr(bh));
 172                         return 1;
 173                 }
 174         } else {
 175                 node = btrfs_buffer_node(bh);
 176                 memcpy(node->header.csum, result, BTRFS_CRC32_SIZE);
 177         }
 178         return 0;
 179 }
 180
 181 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 182 {
 183         struct buffer_head *bh;
 184         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 185         struct buffer_head *head;
 186         if (!page_has_buffers(page)) {
 187                 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
 188                                         (1 << BH_Dirty)|(1 << BH_Uptodate));
 189         }
 190         head = page_buffers(page);
 191         bh = head;
 192         do {
 193                 if (buffer_dirty(bh))
 194                         csum_tree_block(root, bh, 0);
 195                 bh = bh->b_this_page;
 196         } while (bh != head);
 197         return block_write_full_page(page, btree_get_block, wbc);
 198 }
 199
 200 static int btree_readpage(struct file * file, struct page * page)
 201 {
 202         return block_read_full_page(page, btree_get_block);
 203 }
 204
 205 static struct address_space_operations btree_aops = {
 206         .readpage       = btree_readpage,
 207         .writepage      = btree_writepage,
 208         .sync_page      = block_sync_page,
 209 };
 210
 211 int readahead_tree_block(struct btrfs_root *root, u64 blocknr)
 212 {
 213         struct buffer_head *bh = NULL;
 214         int ret = 0;
 215
 216         bh = btrfs_find_create_tree_block(root, blocknr);
 217         if (!bh)
 218                 return 0;
 219         if (buffer_uptodate(bh)) {
 220                 ret = 1;
 221                 goto done;
 222         }
 223         if (test_set_buffer_locked(bh)) {
 224                 ret = 1;
 225                 goto done;
 226         }
 227         if (!buffer_uptodate(bh)) {
 228                 get_bh(bh);
 229                 bh->b_end_io = end_buffer_read_sync;
 230                 submit_bh(READ, bh);
 231         } else {
 232                 unlock_buffer(bh);
 233                 ret = 1;
 234         }
 235 done:
 236         brelse(bh);
 237         return ret;
 238 }
 239
 240 struct buffer_head *read_tree_block(struct btrfs_root *root, u64 blocknr)
 241 {
 242         struct buffer_head *bh = NULL;
 243
 244         bh = btrfs_find_create_tree_block(root, blocknr);
 245         if (!bh)
 246                 return bh;
 247         if (buffer_uptodate(bh))
 248                 goto uptodate;
 249         lock_buffer(bh);
 250         if (!buffer_uptodate(bh)) {
 251                 get_bh(bh);
 252                 bh->b_end_io = end_buffer_read_sync;
 253                 submit_bh(READ, bh);
 254                 wait_on_buffer(bh);
 255                 if (!buffer_uptodate(bh))
 256                         goto fail;
 257         } else {
 258                 unlock_buffer(bh);
 259         }
 260 uptodate:
 261         if (!buffer_checked(bh)) {
 262                 csum_tree_block(root, bh, 1);
 263                 set_buffer_checked(bh);
 264         }
 265         if (check_tree_block(root, bh))
 266                 goto fail;
 267         return bh;
 268 fail:
 269         brelse(bh);
 270         return NULL;
 271 }
 272
 273 int dirty_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 274                      struct buffer_head *buf)
 275 {
 276         WARN_ON(atomic_read(&buf->b_count) == 0);
 277         mark_buffer_dirty(buf);
 278         return 0;
 279 }
 280
 281 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 282                      struct buffer_head *buf)
 283 {
 284         WARN_ON(atomic_read(&buf->b_count) == 0);
 285         clear_buffer_dirty(buf);
 286         return 0;
 287 }
 288
 289 static int __setup_root(int blocksize,
 290                         struct btrfs_root *root,
 291                         struct btrfs_fs_info *fs_info,
 292                         u64 objectid)
 293 {
 294         root->node = NULL;
 295         root->inode = NULL;
 296         root->commit_root = NULL;
 297         root->blocksize = blocksize;
 298         root->ref_cows = 0;
 299         root->fs_info = fs_info;
 300         root->objectid = objectid;
 301         root->last_trans = 0;
 302         root->highest_inode = 0;
 303         root->last_inode_alloc = 0;
 304         memset(&root->root_key, 0, sizeof(root->root_key));
 305         memset(&root->root_item, 0, sizeof(root->root_item));
 306         root->root_key.objectid = objectid;
 307         return 0;
 308 }
 309
 310 static int find_and_setup_root(int blocksize,
 311                                struct btrfs_root *tree_root,
 312                                struct btrfs_fs_info *fs_info,
 313                                u64 objectid,
 314                                struct btrfs_root *root)
 315 {
 316         int ret;
 317
 318         __setup_root(blocksize, root, fs_info, objectid);
 319         ret = btrfs_find_last_root(tree_root, objectid,
 320                                    &root->root_item, &root->root_key);
 321         BUG_ON(ret);
 322
 323         root->node = read_tree_block(root,
 324                                      btrfs_root_blocknr(&root->root_item));
 325         BUG_ON(!root->node);
 326         return 0;
 327 }
 328
 329 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_fs_info *fs_info,
 330                                                struct btrfs_key *location)
 331 {
 332         struct btrfs_root *root;
 333         struct btrfs_root *tree_root = fs_info->tree_root;
 334         struct btrfs_path *path;
 335         struct btrfs_leaf *l;
 336         u64 highest_inode;
 337         int ret = 0;
 338
 339         root = kzalloc(sizeof(*root), GFP_NOFS);
 340         if (!root)
 341                 return ERR_PTR(-ENOMEM);
 342         if (location->offset == (u64)-1) {
 343                 ret = find_and_setup_root(fs_info->sb->s_blocksize,
 344                                           fs_info->tree_root, fs_info,
 345                                           location->objectid, root);
 346                 if (ret) {
 347                         kfree(root);
 348                         return ERR_PTR(ret);
 349                 }
 350                 goto insert;
 351         }
 352
 353         __setup_root(fs_info->sb->s_blocksize, root, fs_info,
 354                      location->objectid);
 355
 356         path = btrfs_alloc_path();
 357         BUG_ON(!path);
 358         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
 359         if (ret != 0) {
 360                 if (ret > 0)
 361                         ret = -ENOENT;
 362                 goto out;
 363         }
 364         l = btrfs_buffer_leaf(path->nodes[0]);
 365         memcpy(&root->root_item,
 366                btrfs_item_ptr(l, path->slots[0], struct btrfs_root_item),
 367                sizeof(root->root_item));
 368         memcpy(&root->root_key, location, sizeof(*location));
 369         ret = 0;
 370 out:
 371         btrfs_release_path(root, path);
 372         btrfs_free_path(path);
 373         if (ret) {
 374                 kfree(root);
 375                 return ERR_PTR(ret);
 376         }
 377         root->node = read_tree_block(root,
 378                                      btrfs_root_blocknr(&root->root_item));
 379         BUG_ON(!root->node);
 380 insert:
 381         root->ref_cows = 1;
 382         ret = btrfs_find_highest_inode(root, &highest_inode);
 383         if (ret == 0) {
 384                 root->highest_inode = highest_inode;
 385                 root->last_inode_alloc = highest_inode;
 386         }
 387         return root;
 388 }
 389
 390 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
 391                                       struct btrfs_key *location)
 392 {
 393         struct btrfs_root *root;
 394         int ret;
 395
 396         root = radix_tree_lookup(&fs_info->fs_roots_radix,
 397                                  (unsigned long)location->objectid);
 398         if (root)
 399                 return root;
 400
 401         root = btrfs_read_fs_root_no_radix(fs_info, location);
 402         if (IS_ERR(root))
 403                 return root;
 404         ret = radix_tree_insert(&fs_info->fs_roots_radix,
 405                                 (unsigned long)root->root_key.objectid,
 406                                 root);
 407         if (ret) {
 408                 brelse(root->node);
 409                 kfree(root);
 410                 return ERR_PTR(ret);
 411         }
 412         return root;
 413 }
 414
 415 struct btrfs_root *open_ctree(struct super_block *sb)
 416 {
 417         struct btrfs_root *extent_root = kmalloc(sizeof(struct btrfs_root),
 418                                                  GFP_NOFS);
 419         struct btrfs_root *tree_root = kmalloc(sizeof(struct btrfs_root),
 420                                                GFP_NOFS);
 421         struct btrfs_fs_info *fs_info = kmalloc(sizeof(*fs_info),
 422                                                 GFP_NOFS);
 423         int ret;
 424         int err = -EIO;
 425         struct btrfs_super_block *disk_super;
 426
 427         if (!extent_root || !tree_root || !fs_info) {
 428                 err = -ENOMEM;
 429                 goto fail;
 430         }
 431         init_bit_radix(&fs_info->pinned_radix);
 432         init_bit_radix(&fs_info->pending_del_radix);
 433         init_bit_radix(&fs_info->extent_map_radix);
 434         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
 435         INIT_RADIX_TREE(&fs_info->block_group_radix, GFP_KERNEL);
 436         INIT_RADIX_TREE(&fs_info->block_group_data_radix, GFP_KERNEL);
 437         INIT_LIST_HEAD(&fs_info->trans_list);
 438         INIT_LIST_HEAD(&fs_info->dead_roots);
 439         sb_set_blocksize(sb, 4096);
 440         fs_info->running_transaction = NULL;
 441         fs_info->tree_root = tree_root;
 442         fs_info->extent_root = extent_root;
 443         fs_info->sb = sb;
 444         fs_info->btree_inode = new_inode(sb);
 445         fs_info->btree_inode->i_ino = 1;
 446         fs_info->btree_inode->i_nlink = 1;
 447         fs_info->btree_inode->i_size = sb->s_bdev->bd_inode->i_size;
 448         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
 449         fs_info->do_barriers = 1;
 450         fs_info->extent_tree_insert_nr = 0;
 451         fs_info->extent_tree_prealloc_nr = 0;
 452         fs_info->closing = 0;
 453
 454         INIT_DELAYED_WORK(&fs_info->trans_work, btrfs_transaction_cleaner);
 455         BTRFS_I(fs_info->btree_inode)->root = tree_root;
 456         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
 457                sizeof(struct btrfs_key));
 458         insert_inode_hash(fs_info->btree_inode);
 459         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
 460
 461         mutex_init(&fs_info->trans_mutex);
 462         mutex_init(&fs_info->fs_mutex);
 463
 464         __setup_root(sb->s_blocksize, tree_root,
 465                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
 466
 467         fs_info->sb_buffer = read_tree_block(tree_root,
 468                                              BTRFS_SUPER_INFO_OFFSET /
 469                                              sb->s_blocksize);
 470
 471         if (!fs_info->sb_buffer)
 472                 goto fail_iput;
 473         disk_super = (struct btrfs_super_block *)fs_info->sb_buffer->b_data;
 474         fs_info->disk_super = disk_super;
 475         memcpy(&fs_info->super_copy, disk_super, sizeof(fs_info->super_copy));
 476
 477         if (!btrfs_super_root(disk_super))
 478                 goto fail_sb_buffer;
 479
 480         i_size_write(fs_info->btree_inode,
 481                      btrfs_super_total_blocks(disk_super) <<
 482                      fs_info->btree_inode->i_blkbits);
 483
 484
 485         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
 486                     sizeof(disk_super->magic))) {
 487                 printk("btrfs: valid FS not found on %s\n", sb->s_id);
 488                 goto fail_sb_buffer;
 489         }
 490         tree_root->node = read_tree_block(tree_root,
 491                                           btrfs_super_root(disk_super));
 492         if (!tree_root->node)
 493                 goto fail_sb_buffer;
 494
 495         mutex_lock(&fs_info->fs_mutex);
 496         ret = find_and_setup_root(sb->s_blocksize, tree_root, fs_info,
 497                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
 498         if (ret) {
 499                 mutex_unlock(&fs_info->fs_mutex);
 500                 goto fail_tree_root;
 501         }
 502
 503         btrfs_read_block_groups(extent_root);
 504
 505         fs_info->generation = btrfs_super_generation(disk_super) + 1;
 506         ret = btrfs_find_dead_roots(tree_root);
 507         if (ret)
 508                 goto fail_tree_root;
 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         lock_buffer(bh);
 532         WARN_ON(atomic_read(&bh->b_count) < 1);
 533         clear_buffer_dirty(bh);
 534         csum_tree_block(root, bh, 0);
 535         bh->b_end_io = end_buffer_write_sync;
 536         get_bh(bh);
 537         if (root->fs_info->do_barriers)
 538                 ret = submit_bh(WRITE_BARRIER, bh);
 539         else
 540                 ret = submit_bh(WRITE, bh);
 541         if (ret == -EOPNOTSUPP) {
 542                 get_bh(bh);
 543                 lock_buffer(bh);
 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 int btrfs_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                         btrfs_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         ret = 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         truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
 615         iput(fs_info->btree_inode);
 616
 617         btrfs_free_block_groups(root->fs_info);
 618         del_fs_roots(fs_info);
 619         kfree(fs_info->extent_root);
 620         kfree(fs_info->tree_root);
 621         return 0;
 622 }
 623
 624 void btrfs_block_release(struct btrfs_root *root, struct buffer_head *buf)
 625 {
 626         brelse(buf);
 627 }
 628
 629 void btrfs_btree_balance_dirty(struct btrfs_root *root)
 630 {
 631         balance_dirty_pages_ratelimited(root->fs_info->btree_inode->i_mapping);
 632 }