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 "ctree.h"
   8 #include "disk-io.h"
   9 #include "transaction.h"
  10
  11 static int check_tree_block(struct btrfs_root *root, struct buffer_head *buf)
  12 {
  13         struct btrfs_node *node = btrfs_buffer_node(buf);
  14         if (buf->b_blocknr != btrfs_header_blocknr(&node->header)) {
  15                 BUG();
  16         }
  17         if (root->node && btrfs_header_parentid(&node->header) !=
  18             btrfs_header_parentid(btrfs_buffer_header(root->node))) {
  19                 BUG();
  20         }
  21         return 0;
  22 }
  23
  24 struct buffer_head *btrfs_find_tree_block(struct btrfs_root *root, u64 blocknr)
  25 {
  26         struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
  27         int blockbits = root->fs_info->sb->s_blocksize_bits;
  28         unsigned long index = blocknr >> (PAGE_CACHE_SHIFT - blockbits);
  29         struct page *page;
  30         struct buffer_head *bh;
  31         struct buffer_head *head;
  32         struct buffer_head *ret = NULL;
  33
  34
  35         page = find_lock_page(mapping, index);
  36         if (!page)
  37                 return NULL;
  38
  39         if (!page_has_buffers(page))
  40                 goto out_unlock;
  41
  42         head = page_buffers(page);
  43         bh = head;
  44         do {
  45                 if (buffer_mapped(bh) && bh->b_blocknr == blocknr) {
  46                         ret = bh;
  47                         get_bh(bh);
  48                         goto out_unlock;
  49                 }
  50                 bh = bh->b_this_page;
  51         } while (bh != head);
  52 out_unlock:
  53         unlock_page(page);
  54         if (ret) {
  55                 touch_buffer(ret);
  56         }
  57         page_cache_release(page);
  58         return ret;
  59 }
  60
  61 struct buffer_head *btrfs_find_create_tree_block(struct btrfs_root *root,
  62                                                  u64 blocknr)
  63 {
  64         struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
  65         int blockbits = root->fs_info->sb->s_blocksize_bits;
  66         unsigned long index = blocknr >> (PAGE_CACHE_SHIFT - blockbits);
  67         struct page *page;
  68         struct buffer_head *bh;
  69         struct buffer_head *head;
  70         struct buffer_head *ret = NULL;
  71         u64 first_block = index << (PAGE_CACHE_SHIFT - blockbits);
  72
  73         page = grab_cache_page(mapping, index);
  74         if (!page)
  75                 return NULL;
  76
  77         if (!page_has_buffers(page))
  78                 create_empty_buffers(page, root->fs_info->sb->s_blocksize, 0);
  79         head = page_buffers(page);
  80         bh = head;
  81         do {
  82                 if (!buffer_mapped(bh)) {
  83                         bh->b_bdev = root->fs_info->sb->s_bdev;
  84                         bh->b_blocknr = first_block;
  85                         set_buffer_mapped(bh);
  86                 }
  87                 if (bh->b_blocknr == blocknr) {
  88                         ret = bh;
  89                         get_bh(bh);
  90                         goto out_unlock;
  91                 }
  92                 bh = bh->b_this_page;
  93                 first_block++;
  94         } while (bh != head);
  95 out_unlock:
  96         unlock_page(page);
  97         if (ret)
  98                 touch_buffer(ret);
  99         page_cache_release(page);
 100         return ret;
 101 }
 102
 103 static sector_t max_block(struct block_device *bdev)
 104 {
 105         sector_t retval = ~((sector_t)0);
 106         loff_t sz = i_size_read(bdev->bd_inode);
 107
 108         if (sz) {
 109                 unsigned int size = block_size(bdev);
 110                 unsigned int sizebits = blksize_bits(size);
 111                 retval = (sz >> sizebits);
 112         }
 113         return retval;
 114 }
 115
 116 static int btree_get_block(struct inode *inode, sector_t iblock,
 117                            struct buffer_head *bh, int create)
 118 {
 119         if (iblock >= max_block(inode->i_sb->s_bdev)) {
 120                 if (create)
 121                         return -EIO;
 122
 123                 /*
 124                  * for reads, we're just trying to fill a partial page.
 125                  * return a hole, they will have to call get_block again
 126                  * before they can fill it, and they will get -EIO at that
 127                  * time
 128                  */
 129                 return 0;
 130         }
 131         bh->b_bdev = inode->i_sb->s_bdev;
 132         bh->b_blocknr = iblock;
 133         set_buffer_mapped(bh);
 134         return 0;
 135 }
 136
 137 int btrfs_csum_data(struct btrfs_root * root, char *data, size_t len,
 138                     char *result)
 139 {
 140         struct scatterlist sg;
 141         struct crypto_hash *tfm = root->fs_info->hash_tfm;
 142         struct hash_desc desc;
 143         int ret;
 144
 145         desc.tfm = tfm;
 146         desc.flags = 0;
 147         sg_init_one(&sg, data, len);
 148         spin_lock(&root->fs_info->hash_lock);
 149         ret = crypto_hash_digest(&desc, &sg, 1, result);
 150         spin_unlock(&root->fs_info->hash_lock);
 151         if (ret) {
 152                 printk("sha256 digest failed\n");
 153         }
 154         return ret;
 155 }
 156 static int csum_tree_block(struct btrfs_root *root, struct buffer_head *bh,
 157                            int verify)
 158 {
 159         char result[BTRFS_CSUM_SIZE];
 160         int ret;
 161         struct btrfs_node *node;
 162
 163         return 0;
 164         ret = btrfs_csum_data(root, bh->b_data + BTRFS_CSUM_SIZE,
 165                               bh->b_size - BTRFS_CSUM_SIZE, result);
 166         if (ret)
 167                 return ret;
 168         if (verify) {
 169                 if (memcmp(bh->b_data, result, BTRFS_CSUM_SIZE)) {
 170                         printk("checksum verify failed on %lu\n",
 171                                bh->b_blocknr);
 172                         return 1;
 173                 }
 174         } else {
 175                 node = btrfs_buffer_node(bh);
 176                 memcpy(node->header.csum, result, BTRFS_CSUM_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_sb(page->mapping->host->i_sb);
 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 struct buffer_head *read_tree_block(struct btrfs_root *root, u64 blocknr)
 212 {
 213         struct buffer_head *bh = NULL;
 214
 215         bh = btrfs_find_create_tree_block(root, blocknr);
 216         if (!bh)
 217                 return bh;
 218         lock_buffer(bh);
 219         if (!buffer_uptodate(bh)) {
 220                 get_bh(bh);
 221                 bh->b_end_io = end_buffer_read_sync;
 222                 submit_bh(READ, bh);
 223                 wait_on_buffer(bh);
 224                 if (!buffer_uptodate(bh))
 225                         goto fail;
 226                 csum_tree_block(root, bh, 1);
 227         } else {
 228                 unlock_buffer(bh);
 229         }
 230         if (check_tree_block(root, bh))
 231                 BUG();
 232         return bh;
 233 fail:
 234         brelse(bh);
 235         return NULL;
 236 }
 237
 238 int dirty_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 239                      struct buffer_head *buf)
 240 {
 241         WARN_ON(atomic_read(&buf->b_count) == 0);
 242         mark_buffer_dirty(buf);
 243         return 0;
 244 }
 245
 246 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 247                      struct buffer_head *buf)
 248 {
 249         WARN_ON(atomic_read(&buf->b_count) == 0);
 250         clear_buffer_dirty(buf);
 251         return 0;
 252 }
 253
 254 static int __setup_root(int blocksize,
 255                         struct btrfs_root *root,
 256                         struct btrfs_fs_info *fs_info,
 257                         u64 objectid)
 258 {
 259         root->node = NULL;
 260         root->commit_root = NULL;
 261         root->blocksize = blocksize;
 262         root->ref_cows = 0;
 263         root->fs_info = fs_info;
 264         memset(&root->root_key, 0, sizeof(root->root_key));
 265         memset(&root->root_item, 0, sizeof(root->root_item));
 266         return 0;
 267 }
 268
 269 static int find_and_setup_root(int blocksize,
 270                                struct btrfs_root *tree_root,
 271                                struct btrfs_fs_info *fs_info,
 272                                u64 objectid,
 273                                struct btrfs_root *root)
 274 {
 275         int ret;
 276
 277         __setup_root(blocksize, root, fs_info, objectid);
 278         ret = btrfs_find_last_root(tree_root, objectid,
 279                                    &root->root_item, &root->root_key);
 280         BUG_ON(ret);
 281
 282         root->node = read_tree_block(root,
 283                                      btrfs_root_blocknr(&root->root_item));
 284         BUG_ON(!root->node);
 285         return 0;
 286 }
 287
 288 struct btrfs_root *open_ctree(struct super_block *sb)
 289 {
 290         struct btrfs_root *root = kmalloc(sizeof(struct btrfs_root),
 291                                           GFP_NOFS);
 292         struct btrfs_root *extent_root = kmalloc(sizeof(struct btrfs_root),
 293                                                  GFP_NOFS);
 294         struct btrfs_root *tree_root = kmalloc(sizeof(struct btrfs_root),
 295                                                GFP_NOFS);
 296         struct btrfs_root *inode_root = kmalloc(sizeof(struct btrfs_root),
 297                                                 GFP_NOFS);
 298         struct btrfs_fs_info *fs_info = kmalloc(sizeof(*fs_info),
 299                                                 GFP_NOFS);
 300         int ret;
 301         struct btrfs_super_block *disk_super;
 302
 303         init_bit_radix(&fs_info->pinned_radix);
 304         init_bit_radix(&fs_info->pending_del_radix);
 305         sb_set_blocksize(sb, 4096);
 306         fs_info->running_transaction = NULL;
 307         fs_info->fs_root = root;
 308         fs_info->tree_root = tree_root;
 309         fs_info->extent_root = extent_root;
 310         fs_info->inode_root = inode_root;
 311         fs_info->last_inode_alloc = 0;
 312         fs_info->last_inode_alloc_dirid = 0;
 313         fs_info->sb = sb;
 314         fs_info->btree_inode = new_inode(sb);
 315         fs_info->btree_inode->i_ino = 1;
 316         fs_info->btree_inode->i_nlink = 1;
 317         fs_info->btree_inode->i_size = sb->s_bdev->bd_inode->i_size;
 318         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
 319         insert_inode_hash(fs_info->btree_inode);
 320         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
 321         fs_info->hash_tfm = crypto_alloc_hash("sha256", 0, CRYPTO_ALG_ASYNC);
 322         spin_lock_init(&fs_info->hash_lock);
 323         if (!fs_info->hash_tfm || IS_ERR(fs_info->hash_tfm)) {
 324                 printk("failed to allocate sha256 hash\n");
 325                 return NULL;
 326         }
 327         mutex_init(&fs_info->trans_mutex);
 328         mutex_init(&fs_info->fs_mutex);
 329         memset(&fs_info->current_insert, 0, sizeof(fs_info->current_insert));
 330         memset(&fs_info->last_insert, 0, sizeof(fs_info->last_insert));
 331
 332         __setup_root(sb->s_blocksize, tree_root,
 333                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
 334         fs_info->sb_buffer = read_tree_block(tree_root,
 335                                              BTRFS_SUPER_INFO_OFFSET /
 336                                              sb->s_blocksize);
 337
 338         if (!fs_info->sb_buffer) {
 339 printk("failed2\n");
 340                 return NULL;
 341         }
 342         disk_super = (struct btrfs_super_block *)fs_info->sb_buffer->b_data;
 343         if (!btrfs_super_root(disk_super)) {
 344                 return NULL;
 345         }
 346         fs_info->disk_super = disk_super;
 347         tree_root->node = read_tree_block(tree_root,
 348                                           btrfs_super_root(disk_super));
 349         BUG_ON(!tree_root->node);
 350
 351         mutex_lock(&fs_info->fs_mutex);
 352         ret = find_and_setup_root(sb->s_blocksize, tree_root, fs_info,
 353                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
 354         BUG_ON(ret);
 355
 356         ret = find_and_setup_root(sb->s_blocksize, tree_root, fs_info,
 357                                   BTRFS_INODE_MAP_OBJECTID, inode_root);
 358         BUG_ON(ret);
 359
 360         ret = find_and_setup_root(sb->s_blocksize, tree_root, fs_info,
 361                                   BTRFS_FS_TREE_OBJECTID, root);
 362         mutex_unlock(&fs_info->fs_mutex);
 363         BUG_ON(ret);
 364         root->commit_root = root->node;
 365         get_bh(root->node);
 366         root->ref_cows = 1;
 367         root->fs_info->generation = root->root_key.offset + 1;
 368         return root;
 369 }
 370
 371 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
 372                       *root)
 373 {
 374         struct buffer_head *bh = root->fs_info->sb_buffer;
 375
 376         btrfs_set_super_root(root->fs_info->disk_super,
 377                              root->fs_info->tree_root->node->b_blocknr);
 378         lock_buffer(bh);
 379         WARN_ON(atomic_read(&bh->b_count) < 1);
 380         clear_buffer_dirty(bh);
 381         csum_tree_block(root, bh, 0);
 382         bh->b_end_io = end_buffer_write_sync;
 383         get_bh(bh);
 384         submit_bh(WRITE, bh);
 385         wait_on_buffer(bh);
 386         if (!buffer_uptodate(bh)) {
 387                 WARN_ON(1);
 388                 return -EIO;
 389         }
 390         return 0;
 391 }
 392
 393 int close_ctree(struct btrfs_root *root)
 394 {
 395         int ret;
 396         struct btrfs_trans_handle *trans;
 397
 398         mutex_lock(&root->fs_info->fs_mutex);
 399         trans = btrfs_start_transaction(root, 1);
 400         btrfs_commit_transaction(trans, root);
 401         /* run commit again to  drop the original snapshot */
 402         trans = btrfs_start_transaction(root, 1);
 403         btrfs_commit_transaction(trans, root);
 404         ret = btrfs_write_and_wait_transaction(NULL, root);
 405         BUG_ON(ret);
 406         write_ctree_super(NULL, root);
 407         mutex_unlock(&root->fs_info->fs_mutex);
 408
 409         if (root->node)
 410                 btrfs_block_release(root, root->node);
 411         if (root->fs_info->extent_root->node)
 412                 btrfs_block_release(root->fs_info->extent_root,
 413                                     root->fs_info->extent_root->node);
 414         if (root->fs_info->inode_root->node)
 415                 btrfs_block_release(root->fs_info->inode_root,
 416                                     root->fs_info->inode_root->node);
 417         if (root->fs_info->tree_root->node)
 418                 btrfs_block_release(root->fs_info->tree_root,
 419                                     root->fs_info->tree_root->node);
 420         btrfs_block_release(root, root->commit_root);
 421         btrfs_block_release(root, root->fs_info->sb_buffer);
 422         crypto_free_hash(root->fs_info->hash_tfm);
 423         truncate_inode_pages(root->fs_info->btree_inode->i_mapping, 0);
 424         iput(root->fs_info->btree_inode);
 425         kfree(root->fs_info->extent_root);
 426         kfree(root->fs_info->inode_root);
 427         kfree(root->fs_info->tree_root);
 428         kfree(root->fs_info);
 429         kfree(root);
 430         return 0;
 431 }
 432
 433 void btrfs_block_release(struct btrfs_root *root, struct buffer_head *buf)
 434 {
 435         brelse(buf);
 436 }
 437