fs/btrfs/disk-io.c

   1 #include <linux/module.h>
   2 #include <linux/fs.h>
   3 #include <linux/blkdev.h>
   4 #include <linux/crypto.h>
   5 #include <linux/scatterlist.h>
   6 #include <linux/swap.h>
   7 #include <linux/radix-tree.h>
   8 #include <linux/writeback.h>
   9 #include "ctree.h"
  10 #include "disk-io.h"
  11 #include "transaction.h"
  12 #include "btrfs_inode.h"
  13
  14 u64 bh_blocknr(struct buffer_head *bh)
  15 {
  16         return bh->b_blocknr;
  17 }
  18
  19 static int check_tree_block(struct btrfs_root *root, struct buffer_head *buf)
  20 {
  21         struct btrfs_node *node = btrfs_buffer_node(buf);
  22         if (bh_blocknr(buf) != btrfs_header_blocknr(&node->header)) {
  23                 printk(KERN_CRIT "bh_blocknr(buf) is %llu, header is %llu\n",
  24                        (unsigned long long)bh_blocknr(buf),
  25                        (unsigned long long)btrfs_header_blocknr(&node->header));
  26                 return 1;
  27         }
  28         return 0;
  29 }
  30
  31 struct buffer_head *btrfs_find_tree_block(struct btrfs_root *root, u64 blocknr)
  32 {
  33         struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
  34         int blockbits = root->fs_info->sb->s_blocksize_bits;
  35         unsigned long index = blocknr >> (PAGE_CACHE_SHIFT - blockbits);
  36         struct page *page;
  37         struct buffer_head *bh;
  38         struct buffer_head *head;
  39         struct buffer_head *ret = NULL;
  40
  41
  42         page = find_lock_page(mapping, index);
  43         if (!page)
  44                 return NULL;
  45
  46         if (!page_has_buffers(page))
  47                 goto out_unlock;
  48
  49         head = page_buffers(page);
  50         bh = head;
  51         do {
  52                 if (buffer_mapped(bh) && bh_blocknr(bh) == blocknr) {
  53                         ret = bh;
  54                         get_bh(bh);
  55                         goto out_unlock;
  56                 }
  57                 bh = bh->b_this_page;
  58         } while (bh != head);
  59 out_unlock:
  60         unlock_page(page);
  61         page_cache_release(page);
  62         return ret;
  63 }
  64
  65 int btrfs_map_bh_to_logical(struct btrfs_root *root, struct buffer_head *bh,
  66                              u64 logical)
  67 {
  68         if (logical == 0) {
  69                 bh->b_bdev = NULL;
  70                 bh->b_blocknr = 0;
  71                 set_buffer_mapped(bh);
  72         } else {
  73                 map_bh(bh, root->fs_info->sb, logical);
  74         }
  75         return 0;
  76 }
  77
  78 struct buffer_head *btrfs_find_create_tree_block(struct btrfs_root *root,
  79                                                  u64 blocknr)
  80 {
  81         struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
  82         int blockbits = root->fs_info->sb->s_blocksize_bits;
  83         unsigned long index = blocknr >> (PAGE_CACHE_SHIFT - blockbits);
  84         struct page *page;
  85         struct buffer_head *bh;
  86         struct buffer_head *head;
  87         struct buffer_head *ret = NULL;
  88         int err;
  89         u64 first_block = index << (PAGE_CACHE_SHIFT - blockbits);
  90
  91         page = grab_cache_page(mapping, index);
  92         if (!page)
  93                 return NULL;
  94
  95         if (!page_has_buffers(page))
  96                 create_empty_buffers(page, root->fs_info->sb->s_blocksize, 0);
  97         head = page_buffers(page);
  98         bh = head;
  99         do {
 100                 if (!buffer_mapped(bh)) {
 101                         err = btrfs_map_bh_to_logical(root, bh, first_block);
 102                         BUG_ON(err);
 103                 }
 104                 if (bh_blocknr(bh) == blocknr) {
 105                         ret = bh;
 106                         get_bh(bh);
 107                         goto out_unlock;
 108                 }
 109                 bh = bh->b_this_page;
 110                 first_block++;
 111         } while (bh != head);
 112 out_unlock:
 113         unlock_page(page);
 114         if (ret)
 115                 touch_buffer(ret);
 116         page_cache_release(page);
 117         return ret;
 118 }
 119
 120 static int btree_get_block(struct inode *inode, sector_t iblock,
 121                            struct buffer_head *bh, int create)
 122 {
 123         int err;
 124         struct btrfs_root *root = BTRFS_I(bh->b_page->mapping->host)->root;
 125         err = btrfs_map_bh_to_logical(root, bh, iblock);
 126         return err;
 127 }
 128
 129 int btrfs_csum_data(struct btrfs_root * root, char *data, size_t len,
 130                     char *result)
 131 {
 132         struct scatterlist sg;
 133         struct crypto_hash *tfm = root->fs_info->hash_tfm;
 134         struct hash_desc desc;
 135         int ret;
 136
 137         desc.tfm = tfm;
 138         desc.flags = 0;
 139         sg_init_one(&sg, data, len);
 140         spin_lock(&root->fs_info->hash_lock);
 141         ret = crypto_hash_digest(&desc, &sg, 1, result);
 142         spin_unlock(&root->fs_info->hash_lock);
 143         if (ret) {
 144                 printk("digest failed\n");
 145         }
 146         return ret;
 147 }
 148 static int csum_tree_block(struct btrfs_root *root, struct buffer_head *bh,
 149                            int verify)
 150 {
 151         char result[BTRFS_CRC32_SIZE];
 152         int ret;
 153         struct btrfs_node *node;
 154
 155         ret = btrfs_csum_data(root, bh->b_data + BTRFS_CSUM_SIZE,
 156                               bh->b_size - BTRFS_CSUM_SIZE, result);
 157         if (ret)
 158                 return ret;
 159         if (verify) {
 160                 if (memcmp(bh->b_data, result, BTRFS_CRC32_SIZE)) {
 161                         printk("btrfs: %s checksum verify failed on %llu\n",
 162                                root->fs_info->sb->s_id,
 163                                (unsigned long long)bh_blocknr(bh));
 164                         return 1;
 165                 }
 166         } else {
 167                 node = btrfs_buffer_node(bh);
 168                 memcpy(node->header.csum, result, BTRFS_CRC32_SIZE);
 169         }
 170         return 0;
 171 }
 172
 173 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 174 {
 175         struct buffer_head *bh;
 176         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 177         struct buffer_head *head;
 178         if (!page_has_buffers(page)) {
 179                 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
 180                                         (1 << BH_Dirty)|(1 << BH_Uptodate));
 181         }
 182         head = page_buffers(page);
 183         bh = head;
 184         do {
 185                 if (buffer_dirty(bh))
 186                         csum_tree_block(root, bh, 0);
 187                 bh = bh->b_this_page;
 188         } while (bh != head);
 189         return block_write_full_page(page, btree_get_block, wbc);
 190 }
 191
 192 static int btree_readpage(struct file * file, struct page * page)
 193 {
 194         return block_read_full_page(page, btree_get_block);
 195 }
 196
 197 static struct address_space_operations btree_aops = {
 198         .readpage       = btree_readpage,
 199         .writepage      = btree_writepage,
 200         .sync_page      = block_sync_page,
 201 };
 202
 203 int readahead_tree_block(struct btrfs_root *root, u64 blocknr)
 204 {
 205         struct buffer_head *bh = NULL;
 206         int ret = 0;
 207
 208         bh = btrfs_find_create_tree_block(root, blocknr);
 209         if (!bh)
 210                 return 0;
 211         if (buffer_uptodate(bh)) {
 212                 ret = 1;
 213                 goto done;
 214         }
 215         if (test_set_buffer_locked(bh)) {
 216                 ret = 1;
 217                 goto done;
 218         }
 219         if (!buffer_uptodate(bh)) {
 220                 get_bh(bh);
 221                 bh->b_end_io = end_buffer_read_sync;
 222                 submit_bh(READ, bh);
 223         } else {
 224                 unlock_buffer(bh);
 225                 ret = 1;
 226         }
 227 done:
 228         brelse(bh);
 229         return ret;
 230 }
 231
 232 struct buffer_head *read_tree_block(struct btrfs_root *root, u64 blocknr)
 233 {
 234         struct buffer_head *bh = NULL;
 235
 236         bh = btrfs_find_create_tree_block(root, blocknr);
 237         if (!bh)
 238                 return bh;
 239         if (buffer_uptodate(bh))
 240                 goto uptodate;
 241         lock_buffer(bh);
 242         if (!buffer_uptodate(bh)) {
 243                 get_bh(bh);
 244                 bh->b_end_io = end_buffer_read_sync;
 245                 submit_bh(READ, bh);
 246                 wait_on_buffer(bh);
 247                 if (!buffer_uptodate(bh))
 248                         goto fail;
 249         } else {
 250                 unlock_buffer(bh);
 251         }
 252 uptodate:
 253         if (!buffer_checked(bh)) {
 254                 csum_tree_block(root, bh, 1);
 255                 set_buffer_checked(bh);
 256         }
 257         if (check_tree_block(root, bh))
 258                 goto fail;
 259         return bh;
 260 fail:
 261         brelse(bh);
 262         return NULL;
 263 }
 264
 265 int dirty_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 266                      struct buffer_head *buf)
 267 {
 268         WARN_ON(atomic_read(&buf->b_count) == 0);
 269         mark_buffer_dirty(buf);
 270         return 0;
 271 }
 272
 273 int clean_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         clear_buffer_dirty(buf);
 278         return 0;
 279 }
 280
 281 static int __setup_root(int blocksize,
 282                         struct btrfs_root *root,
 283                         struct btrfs_fs_info *fs_info,
 284                         u64 objectid)
 285 {
 286         root->node = NULL;
 287         root->inode = NULL;
 288         root->commit_root = NULL;
 289         root->blocksize = blocksize;
 290         root->ref_cows = 0;
 291         root->fs_info = fs_info;
 292         root->objectid = objectid;
 293         root->last_trans = 0;
 294         root->highest_inode = 0;
 295         root->last_inode_alloc = 0;
 296         memset(&root->root_key, 0, sizeof(root->root_key));
 297         memset(&root->root_item, 0, sizeof(root->root_item));
 298         root->root_key.objectid = objectid;
 299         return 0;
 300 }
 301
 302 static int find_and_setup_root(int blocksize,
 303                                struct btrfs_root *tree_root,
 304                                struct btrfs_fs_info *fs_info,
 305                                u64 objectid,
 306                                struct btrfs_root *root)
 307 {
 308         int ret;
 309
 310         __setup_root(blocksize, root, fs_info, objectid);
 311         ret = btrfs_find_last_root(tree_root, objectid,
 312                                    &root->root_item, &root->root_key);
 313         BUG_ON(ret);
 314
 315         root->node = read_tree_block(root,
 316                                      btrfs_root_blocknr(&root->root_item));
 317         BUG_ON(!root->node);
 318         return 0;
 319 }
 320
 321 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
 322                                       struct btrfs_key *location)
 323 {
 324         struct btrfs_root *root;
 325         struct btrfs_root *tree_root = fs_info->tree_root;
 326         struct btrfs_path *path;
 327         struct btrfs_leaf *l;
 328         u64 highest_inode;
 329         int ret = 0;
 330
 331         root = radix_tree_lookup(&fs_info->fs_roots_radix,
 332                                  (unsigned long)location->objectid);
 333         if (root)
 334                 return root;
 335         root = kmalloc(sizeof(*root), GFP_NOFS);
 336         if (!root)
 337                 return ERR_PTR(-ENOMEM);
 338         if (location->offset == (u64)-1) {
 339                 ret = find_and_setup_root(fs_info->sb->s_blocksize,
 340                                           fs_info->tree_root, fs_info,
 341                                           location->objectid, root);
 342                 if (ret) {
 343                         kfree(root);
 344                         return ERR_PTR(ret);
 345                 }
 346                 goto insert;
 347         }
 348
 349         __setup_root(fs_info->sb->s_blocksize, root, fs_info,
 350                      location->objectid);
 351
 352         path = btrfs_alloc_path();
 353         BUG_ON(!path);
 354         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
 355         if (ret != 0) {
 356                 if (ret > 0)
 357                         ret = -ENOENT;
 358                 goto out;
 359         }
 360         l = btrfs_buffer_leaf(path->nodes[0]);
 361         memcpy(&root->root_item,
 362                btrfs_item_ptr(l, path->slots[0], struct btrfs_root_item),
 363                sizeof(root->root_item));
 364         memcpy(&root->root_key, location, sizeof(*location));
 365         ret = 0;
 366 out:
 367         btrfs_release_path(root, path);
 368         btrfs_free_path(path);
 369         if (ret) {
 370                 kfree(root);
 371                 return ERR_PTR(ret);
 372         }
 373         root->node = read_tree_block(root,
 374                                      btrfs_root_blocknr(&root->root_item));
 375         BUG_ON(!root->node);
 376 insert:
 377         root->ref_cows = 1;
 378         ret = radix_tree_insert(&fs_info->fs_roots_radix,
 379                                 (unsigned long)root->root_key.objectid,
 380                                 root);
 381         if (ret) {
 382                 brelse(root->node);
 383                 kfree(root);
 384                 return ERR_PTR(ret);
 385         }
 386         ret = btrfs_find_highest_inode(root, &highest_inode);
 387         if (ret == 0) {
 388                 root->highest_inode = highest_inode;
 389                 root->last_inode_alloc = highest_inode;
 390         }
 391         return root;
 392 }
 393
 394 struct btrfs_root *open_ctree(struct super_block *sb)
 395 {
 396         struct btrfs_root *extent_root = kmalloc(sizeof(struct btrfs_root),
 397                                                  GFP_NOFS);
 398         struct btrfs_root *tree_root = kmalloc(sizeof(struct btrfs_root),
 399                                                GFP_NOFS);
 400         struct btrfs_fs_info *fs_info = kmalloc(sizeof(*fs_info),
 401                                                 GFP_NOFS);
 402         int ret;
 403         int err = -EIO;
 404         struct btrfs_super_block *disk_super;
 405
 406         if (!extent_root || !tree_root || !fs_info) {
 407                 err = -ENOMEM;
 408                 goto fail;
 409         }
 410         init_bit_radix(&fs_info->pinned_radix);
 411         init_bit_radix(&fs_info->pending_del_radix);
 412         init_bit_radix(&fs_info->extent_map_radix);
 413         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
 414         INIT_RADIX_TREE(&fs_info->block_group_radix, GFP_KERNEL);
 415         INIT_RADIX_TREE(&fs_info->block_group_data_radix, GFP_KERNEL);
 416         INIT_LIST_HEAD(&fs_info->trans_list);
 417         INIT_LIST_HEAD(&fs_info->dead_roots);
 418         sb_set_blocksize(sb, 4096);
 419         fs_info->running_transaction = NULL;
 420         fs_info->tree_root = tree_root;
 421         fs_info->extent_root = extent_root;
 422         fs_info->sb = sb;
 423         fs_info->btree_inode = new_inode(sb);
 424         fs_info->btree_inode->i_ino = 1;
 425         fs_info->btree_inode->i_nlink = 1;
 426         fs_info->btree_inode->i_size = sb->s_bdev->bd_inode->i_size;
 427         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
 428         fs_info->do_barriers = 1;
 429         fs_info->extent_tree_insert_nr = 0;
 430         fs_info->extent_tree_prealloc_nr = 0;
 431         fs_info->closing = 0;
 432
 433         INIT_DELAYED_WORK(&fs_info->trans_work, btrfs_transaction_cleaner);
 434         BTRFS_I(fs_info->btree_inode)->root = tree_root;
 435         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
 436                sizeof(struct btrfs_key));
 437         insert_inode_hash(fs_info->btree_inode);
 438         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
 439         fs_info->hash_tfm = crypto_alloc_hash("crc32c", 0, CRYPTO_ALG_ASYNC);
 440         spin_lock_init(&fs_info->hash_lock);
 441
 442         if (!fs_info->hash_tfm || IS_ERR(fs_info->hash_tfm)) {
 443                 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
 444                 err = -ENOMEM;
 445                 goto fail_iput;
 446         }
 447         mutex_init(&fs_info->trans_mutex);
 448         mutex_init(&fs_info->fs_mutex);
 449
 450         __setup_root(sb->s_blocksize, tree_root,
 451                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
 452
 453         fs_info->sb_buffer = read_tree_block(tree_root,
 454                                              BTRFS_SUPER_INFO_OFFSET /
 455                                              sb->s_blocksize);
 456
 457         if (!fs_info->sb_buffer)
 458                 goto fail_iput;
 459         disk_super = (struct btrfs_super_block *)fs_info->sb_buffer->b_data;
 460
 461         if (!btrfs_super_root(disk_super))
 462                 goto fail_sb_buffer;
 463
 464         i_size_write(fs_info->btree_inode,
 465                      btrfs_super_total_blocks(disk_super) <<
 466                      fs_info->btree_inode->i_blkbits);
 467
 468         fs_info->disk_super = disk_super;
 469
 470         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
 471                     sizeof(disk_super->magic))) {
 472                 printk("btrfs: valid FS not found on %s\n", sb->s_id);
 473                 goto fail_sb_buffer;
 474         }
 475         tree_root->node = read_tree_block(tree_root,
 476                                           btrfs_super_root(disk_super));
 477         if (!tree_root->node)
 478                 goto fail_sb_buffer;
 479
 480         mutex_lock(&fs_info->fs_mutex);
 481         ret = find_and_setup_root(sb->s_blocksize, tree_root, fs_info,
 482                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
 483         if (ret) {
 484                 mutex_unlock(&fs_info->fs_mutex);
 485                 goto fail_tree_root;
 486         }
 487
 488         btrfs_read_block_groups(extent_root);
 489
 490         fs_info->generation = btrfs_super_generation(disk_super) + 1;
 491         mutex_unlock(&fs_info->fs_mutex);
 492         return tree_root;
 493
 494 fail_tree_root:
 495         btrfs_block_release(tree_root, tree_root->node);
 496 fail_sb_buffer:
 497         btrfs_block_release(tree_root, fs_info->sb_buffer);
 498 fail_iput:
 499         iput(fs_info->btree_inode);
 500 fail:
 501         kfree(extent_root);
 502         kfree(tree_root);
 503         kfree(fs_info);
 504         return ERR_PTR(err);
 505 }
 506
 507 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
 508                       *root)
 509 {
 510         int ret;
 511         struct buffer_head *bh = root->fs_info->sb_buffer;
 512
 513         btrfs_set_super_root(root->fs_info->disk_super,
 514                              bh_blocknr(root->fs_info->tree_root->node));
 515         lock_buffer(bh);
 516         WARN_ON(atomic_read(&bh->b_count) < 1);
 517         clear_buffer_dirty(bh);
 518         csum_tree_block(root, bh, 0);
 519         bh->b_end_io = end_buffer_write_sync;
 520         get_bh(bh);
 521         if (root->fs_info->do_barriers)
 522                 ret = submit_bh(WRITE_BARRIER, bh);
 523         else
 524                 ret = submit_bh(WRITE, bh);
 525         if (ret == -EOPNOTSUPP) {
 526                 set_buffer_uptodate(bh);
 527                 root->fs_info->do_barriers = 0;
 528                 ret = submit_bh(WRITE, bh);
 529         }
 530         wait_on_buffer(bh);
 531         if (!buffer_uptodate(bh)) {
 532                 WARN_ON(1);
 533                 return -EIO;
 534         }
 535         return 0;
 536 }
 537
 538 static int free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
 539 {
 540         radix_tree_delete(&fs_info->fs_roots_radix,
 541                           (unsigned long)root->root_key.objectid);
 542         if (root->inode)
 543                 iput(root->inode);
 544         if (root->node)
 545                 brelse(root->node);
 546         if (root->commit_root)
 547                 brelse(root->commit_root);
 548         kfree(root);
 549         return 0;
 550 }
 551
 552 static int del_fs_roots(struct btrfs_fs_info *fs_info)
 553 {
 554         int ret;
 555         struct btrfs_root *gang[8];
 556         int i;
 557
 558         while(1) {
 559                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
 560                                              (void **)gang, 0,
 561                                              ARRAY_SIZE(gang));
 562                 if (!ret)
 563                         break;
 564                 for (i = 0; i < ret; i++)
 565                         free_fs_root(fs_info, gang[i]);
 566         }
 567         return 0;
 568 }
 569
 570 int close_ctree(struct btrfs_root *root)
 571 {
 572         int ret;
 573         struct btrfs_trans_handle *trans;
 574         struct btrfs_fs_info *fs_info = root->fs_info;
 575
 576         fs_info->closing = 1;
 577         btrfs_transaction_flush_work(root);
 578         mutex_lock(&fs_info->fs_mutex);
 579         trans = btrfs_start_transaction(root, 1);
 580         btrfs_commit_transaction(trans, root);
 581         /* run commit again to  drop the original snapshot */
 582         trans = btrfs_start_transaction(root, 1);
 583         btrfs_commit_transaction(trans, root);
 584         ret = btrfs_write_and_wait_transaction(NULL, root);
 585         BUG_ON(ret);
 586         write_ctree_super(NULL, root);
 587         mutex_unlock(&fs_info->fs_mutex);
 588
 589         if (fs_info->extent_root->node)
 590                 btrfs_block_release(fs_info->extent_root,
 591                                     fs_info->extent_root->node);
 592         if (fs_info->tree_root->node)
 593                 btrfs_block_release(fs_info->tree_root,
 594                                     fs_info->tree_root->node);
 595         btrfs_block_release(root, fs_info->sb_buffer);
 596         crypto_free_hash(fs_info->hash_tfm);
 597         truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
 598         iput(fs_info->btree_inode);
 599
 600         btrfs_free_block_groups(root->fs_info);
 601         del_fs_roots(fs_info);
 602         kfree(fs_info->extent_root);
 603         kfree(fs_info->tree_root);
 604         return 0;
 605 }
 606
 607 void btrfs_block_release(struct btrfs_root *root, struct buffer_head *buf)
 608 {
 609         brelse(buf);
 610 }
 611
 612 void btrfs_btree_balance_dirty(struct btrfs_root *root)
 613 {
 614         balance_dirty_pages_ratelimited(root->fs_info->btree_inode->i_mapping);
 615 }