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 #define _XOPEN_SOURCE 600
  20 #define __USE_XOPEN2K
  21 #include <stdio.h>
  22 #include <stdlib.h>
  23 #include <sys/types.h>
  24 #include <sys/stat.h>
  25 #include <fcntl.h>
  26 #include <unistd.h>
  27 #include "kerncompat.h"
  28 #include "radix-tree.h"
  29 #include "ctree.h"
  30 #include "disk-io.h"
  31 #include "transaction.h"
  32
  33 static int allocated_blocks = 0;
  34 int cache_max = 10000;
  35
  36 int btrfs_map_bh_to_logical(struct btrfs_root *root, struct btrfs_buffer *bh,
  37                              u64 logical)
  38 {
  39         bh->fd = root->fs_info->fp;
  40         bh->dev_blocknr = logical;
  41         return 0;
  42 }
  43
  44 static int check_tree_block(struct btrfs_root *root, struct btrfs_buffer *buf)
  45 {
  46         if (buf->blocknr != btrfs_header_blocknr(&buf->node.header))
  47                 BUG();
  48         if (memcmp(root->fs_info->disk_super->fsid, buf->node.header.fsid,
  49                    sizeof(buf->node.header.fsid)))
  50                 BUG();
  51         return 0;
  52 }
  53
  54 static int free_some_buffers(struct btrfs_root *root)
  55 {
  56         struct list_head *node, *next;
  57         struct btrfs_buffer *b;
  58         if (root->fs_info->cache_size < cache_max)
  59                 return 0;
  60         list_for_each_safe(node, next, &root->fs_info->cache) {
  61                 b = list_entry(node, struct btrfs_buffer, cache);
  62                 if (b->count == 1) {
  63                         BUG_ON(!list_empty(&b->dirty));
  64                         list_del_init(&b->cache);
  65                         btrfs_block_release(root, b);
  66                         if (root->fs_info->cache_size < cache_max)
  67                                 break;
  68                 }
  69         }
  70         return 0;
  71 }
  72
  73 struct btrfs_buffer *alloc_tree_block(struct btrfs_root *root, u64 blocknr)
  74 {
  75         struct btrfs_buffer *buf;
  76         int ret;
  77
  78         buf = malloc(sizeof(struct btrfs_buffer) + root->blocksize);
  79         if (!buf)
  80                 return buf;
  81         allocated_blocks++;
  82         buf->blocknr = blocknr;
  83         buf->count = 2;
  84         INIT_LIST_HEAD(&buf->dirty);
  85         free_some_buffers(root);
  86         radix_tree_preload(GFP_KERNEL);
  87         ret = radix_tree_insert(&root->fs_info->cache_radix, blocknr, buf);
  88         radix_tree_preload_end();
  89         list_add_tail(&buf->cache, &root->fs_info->cache);
  90         root->fs_info->cache_size++;
  91         if (ret) {
  92                 free(buf);
  93                 return NULL;
  94         }
  95         return buf;
  96 }
  97
  98 struct btrfs_buffer *find_tree_block(struct btrfs_root *root, u64 blocknr)
  99 {
 100         struct btrfs_buffer *buf;
 101         buf = radix_tree_lookup(&root->fs_info->cache_radix, blocknr);
 102         if (buf) {
 103                 buf->count++;
 104         } else {
 105                 buf = alloc_tree_block(root, blocknr);
 106                 if (!buf) {
 107                         BUG();
 108                         return NULL;
 109                 }
 110         }
 111         return buf;
 112 }
 113
 114 struct btrfs_buffer *read_tree_block(struct btrfs_root *root, u64 blocknr)
 115 {
 116         struct btrfs_buffer *buf;
 117         int ret;
 118         buf = radix_tree_lookup(&root->fs_info->cache_radix, blocknr);
 119         if (buf) {
 120                 buf->count++;
 121                 if (check_tree_block(root, buf))
 122                         BUG();
 123         } else {
 124                 buf = alloc_tree_block(root, blocknr);
 125                 if (!buf)
 126                         return NULL;
 127                 btrfs_map_bh_to_logical(root, buf, blocknr);
 128                 ret = pread(buf->fd, &buf->node, root->blocksize,
 129                             buf->dev_blocknr * root->blocksize);
 130                 if (ret != root->blocksize) {
 131                         free(buf);
 132                         return NULL;
 133                 }
 134                 if (check_tree_block(root, buf))
 135                         BUG();
 136         }
 137         return buf;
 138 }
 139
 140 int dirty_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 141                      struct btrfs_buffer *buf)
 142 {
 143         if (!list_empty(&buf->dirty))
 144                 return 0;
 145         list_add_tail(&buf->dirty, &root->fs_info->trans);
 146         buf->count++;
 147         if (check_tree_block(root, buf))
 148                 BUG();
 149         return 0;
 150 }
 151
 152 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 153                      struct btrfs_buffer *buf)
 154 {
 155         if (!list_empty(&buf->dirty)) {
 156                 list_del_init(&buf->dirty);
 157                 btrfs_block_release(root, buf);
 158         }
 159         return 0;
 160 }
 161
 162 int write_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 163                      struct btrfs_buffer *buf)
 164 {
 165         int ret;
 166
 167         if (buf->blocknr != btrfs_header_blocknr(&buf->node.header))
 168                 BUG();
 169         btrfs_map_bh_to_logical(root, buf, buf->blocknr);
 170         if (check_tree_block(root, buf))
 171                 BUG();
 172         ret = pwrite(buf->fd, &buf->node, root->blocksize,
 173                      buf->dev_blocknr * root->blocksize);
 174         if (ret != root->blocksize)
 175                 return ret;
 176         return 0;
 177 }
 178
 179 static int __commit_transaction(struct btrfs_trans_handle *trans, struct
 180                                 btrfs_root *root)
 181 {
 182         struct btrfs_buffer *b;
 183         int ret = 0;
 184         int wret;
 185         while(!list_empty(&root->fs_info->trans)) {
 186                 b = list_entry(root->fs_info->trans.next, struct btrfs_buffer,
 187                                dirty);
 188                 list_del_init(&b->dirty);
 189                 wret = write_tree_block(trans, root, b);
 190                 if (wret)
 191                         ret = wret;
 192                 btrfs_block_release(root, b);
 193         }
 194         return ret;
 195 }
 196
 197 static int commit_tree_roots(struct btrfs_trans_handle *trans,
 198                              struct btrfs_fs_info *fs_info)
 199 {
 200         int ret;
 201         u64 old_extent_block;
 202         struct btrfs_root *tree_root = fs_info->tree_root;
 203         struct btrfs_root *extent_root = fs_info->extent_root;
 204
 205         btrfs_write_dirty_block_groups(trans, fs_info->extent_root);
 206         while(1) {
 207                 old_extent_block = btrfs_root_blocknr(&extent_root->root_item);
 208                 if (old_extent_block == extent_root->node->blocknr)
 209                         break;
 210                 btrfs_set_root_blocknr(&extent_root->root_item,
 211                                        extent_root->node->blocknr);
 212                 ret = btrfs_update_root(trans, tree_root,
 213                                         &extent_root->root_key,
 214                                         &extent_root->root_item);
 215                 BUG_ON(ret);
 216                 btrfs_write_dirty_block_groups(trans, fs_info->extent_root);
 217         }
 218         return 0;
 219 }
 220
 221 int btrfs_commit_transaction(struct btrfs_trans_handle *trans, struct
 222                              btrfs_root *root, struct btrfs_super_block *s)
 223 {
 224         int ret = 0;
 225         struct btrfs_buffer *snap = root->commit_root;
 226         struct btrfs_key snap_key;
 227
 228         if (root->commit_root == root->node)
 229                 return 0;
 230
 231         memcpy(&snap_key, &root->root_key, sizeof(snap_key));
 232         root->root_key.offset++;
 233
 234         btrfs_set_root_blocknr(&root->root_item, root->node->blocknr);
 235         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
 236                                 &root->root_key, &root->root_item);
 237         BUG_ON(ret);
 238
 239         ret = commit_tree_roots(trans, root->fs_info);
 240         BUG_ON(ret);
 241
 242         ret = __commit_transaction(trans, root);
 243         BUG_ON(ret);
 244
 245         write_ctree_super(trans, root, s);
 246         btrfs_finish_extent_commit(trans, root->fs_info->extent_root);
 247         btrfs_finish_extent_commit(trans, root->fs_info->tree_root);
 248
 249         root->commit_root = root->node;
 250         root->node->count++;
 251         ret = btrfs_drop_snapshot(trans, root, snap);
 252         BUG_ON(ret);
 253
 254         ret = btrfs_del_root(trans, root->fs_info->tree_root, &snap_key);
 255         BUG_ON(ret);
 256         root->fs_info->generation = root->root_key.offset + 1;
 257
 258         return ret;
 259 }
 260
 261 static int __setup_root(struct btrfs_super_block *super,
 262                         struct btrfs_root *root,
 263                         struct btrfs_fs_info *fs_info,
 264                         u64 objectid, int fp)
 265 {
 266         root->node = NULL;
 267         root->commit_root = NULL;
 268         root->blocksize = btrfs_super_blocksize(super);
 269         root->ref_cows = 0;
 270         root->fs_info = fs_info;
 271         memset(&root->root_key, 0, sizeof(root->root_key));
 272         memset(&root->root_item, 0, sizeof(root->root_item));
 273         root->root_key.objectid = objectid;
 274         return 0;
 275 }
 276
 277 static int find_and_setup_root(struct btrfs_super_block *super,
 278                                struct btrfs_root *tree_root,
 279                                struct btrfs_fs_info *fs_info,
 280                                u64 objectid,
 281                                struct btrfs_root *root, int fp)
 282 {
 283         int ret;
 284
 285         __setup_root(super, root, fs_info, objectid, fp);
 286         ret = btrfs_find_last_root(tree_root, objectid,
 287                                    &root->root_item, &root->root_key);
 288         BUG_ON(ret);
 289
 290         root->node = read_tree_block(root,
 291                                      btrfs_root_blocknr(&root->root_item));
 292         BUG_ON(!root->node);
 293         return 0;
 294 }
 295
 296 struct btrfs_root *open_ctree(char *filename, struct btrfs_super_block *super)
 297 {
 298         int fp;
 299
 300         fp = open(filename, O_CREAT | O_RDWR, 0600);
 301         if (fp < 0) {
 302                 return NULL;
 303         }
 304         return open_ctree_fd(fp, super);
 305 }
 306
 307 struct btrfs_root *open_ctree_fd(int fp, struct btrfs_super_block *super)
 308 {
 309         struct btrfs_root *root = malloc(sizeof(struct btrfs_root));
 310         struct btrfs_root *extent_root = malloc(sizeof(struct btrfs_root));
 311         struct btrfs_root *tree_root = malloc(sizeof(struct btrfs_root));
 312         struct btrfs_fs_info *fs_info = malloc(sizeof(*fs_info));
 313         int ret;
 314
 315         INIT_RADIX_TREE(&fs_info->cache_radix, GFP_KERNEL);
 316         INIT_RADIX_TREE(&fs_info->pinned_radix, GFP_KERNEL);
 317         INIT_RADIX_TREE(&fs_info->block_group_radix, GFP_KERNEL);
 318         INIT_LIST_HEAD(&fs_info->trans);
 319         INIT_LIST_HEAD(&fs_info->cache);
 320         fs_info->cache_size = 0;
 321         fs_info->fp = fp;
 322         fs_info->running_transaction = NULL;
 323         fs_info->fs_root = root;
 324         fs_info->tree_root = tree_root;
 325         fs_info->extent_root = extent_root;
 326         fs_info->last_inode_alloc = 0;
 327         fs_info->last_inode_alloc_dirid = 0;
 328         fs_info->disk_super = super;
 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         ret = pread(fp, super, sizeof(struct btrfs_super_block),
 333                      BTRFS_SUPER_INFO_OFFSET);
 334         if (ret == 0 || btrfs_super_root(super) == 0) {
 335                 BUG();
 336                 return NULL;
 337         }
 338         BUG_ON(ret < 0);
 339
 340         __setup_root(super, tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID, fp);
 341         tree_root->node = read_tree_block(tree_root, btrfs_super_root(super));
 342         BUG_ON(!tree_root->node);
 343
 344         ret = find_and_setup_root(super, tree_root, fs_info,
 345                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root, fp);
 346         BUG_ON(ret);
 347
 348         ret = find_and_setup_root(super, tree_root, fs_info,
 349                                   BTRFS_FS_TREE_OBJECTID, root, fp);
 350         BUG_ON(ret);
 351
 352         root->commit_root = root->node;
 353         root->node->count++;
 354         root->ref_cows = 1;
 355         root->fs_info->generation = root->root_key.offset + 1;
 356         btrfs_read_block_groups(root);
 357         return root;
 358 }
 359
 360 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
 361                       *root, struct btrfs_super_block *s)
 362 {
 363         int ret;
 364         btrfs_set_super_root(s, root->fs_info->tree_root->node->blocknr);
 365         ret = pwrite(root->fs_info->fp, s, sizeof(*s),
 366                      BTRFS_SUPER_INFO_OFFSET);
 367         if (ret != sizeof(*s)) {
 368                 fprintf(stderr, "failed to write new super block err %d\n", ret);
 369                 return ret;
 370         }
 371         return 0;
 372 }
 373
 374 static int drop_cache(struct btrfs_root *root)
 375 {
 376         while(!list_empty(&root->fs_info->cache)) {
 377                 struct btrfs_buffer *b = list_entry(root->fs_info->cache.next,
 378                                                     struct btrfs_buffer,
 379                                                     cache);
 380                 list_del_init(&b->cache);
 381                 btrfs_block_release(root, b);
 382         }
 383         return 0;
 384 }
 385
 386 int close_ctree(struct btrfs_root *root, struct btrfs_super_block *s)
 387 {
 388         int ret;
 389         struct btrfs_trans_handle *trans;
 390
 391         trans = root->fs_info->running_transaction;
 392         btrfs_commit_transaction(trans, root, s);
 393         ret = commit_tree_roots(trans, root->fs_info);
 394         BUG_ON(ret);
 395         ret = __commit_transaction(trans, root);
 396         BUG_ON(ret);
 397         write_ctree_super(trans, root, s);
 398         drop_cache(root);
 399         BUG_ON(!list_empty(&root->fs_info->trans));
 400
 401         btrfs_free_block_groups(root->fs_info);
 402         close(root->fs_info->fp);
 403         if (root->node)
 404                 btrfs_block_release(root, root->node);
 405         if (root->fs_info->extent_root->node)
 406                 btrfs_block_release(root->fs_info->extent_root,
 407                                     root->fs_info->extent_root->node);
 408         if (root->fs_info->tree_root->node)
 409                 btrfs_block_release(root->fs_info->tree_root,
 410                                     root->fs_info->tree_root->node);
 411         btrfs_block_release(root, root->commit_root);
 412         free(root);
 413         printf("on close %d blocks are allocated\n", allocated_blocks);
 414         return 0;
 415 }
 416
 417 void btrfs_block_release(struct btrfs_root *root, struct btrfs_buffer *buf)
 418 {
 419         buf->count--;
 420         if (buf->count < 0)
 421                 BUG();
 422         if (buf->count == 0) {
 423                 BUG_ON(!list_empty(&buf->cache));
 424                 BUG_ON(!list_empty(&buf->dirty));
 425                 if (!radix_tree_lookup(&root->fs_info->cache_radix,
 426                                        buf->blocknr))
 427                         BUG();
 428                 radix_tree_delete(&root->fs_info->cache_radix, buf->blocknr);
 429                 memset(buf, 0, sizeof(*buf));
 430                 free(buf);
 431                 BUG_ON(allocated_blocks == 0);
 432                 allocated_blocks--;
 433                 BUG_ON(root->fs_info->cache_size == 0);
 434                 root->fs_info->cache_size--;
 435         }
 436 }
 437