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