Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (C) 2008 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/sched.h> |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 20 | #include <linux/slab.h> |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 21 | #include "ctree.h" |
| 22 | #include "transaction.h" |
| 23 | #include "disk-io.h" |
| 24 | #include "locking.h" |
| 25 | #include "print-tree.h" |
| 26 | #include "compat.h" |
Christoph Hellwig | b295086 | 2008-12-02 09:54:17 -0500 | [diff] [blame] | 27 | #include "tree-log.h" |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 28 | |
| 29 | /* magic values for the inode_only field in btrfs_log_inode: |
| 30 | * |
| 31 | * LOG_INODE_ALL means to log everything |
| 32 | * LOG_INODE_EXISTS means to log just enough to recreate the inode |
| 33 | * during log replay |
| 34 | */ |
| 35 | #define LOG_INODE_ALL 0 |
| 36 | #define LOG_INODE_EXISTS 1 |
| 37 | |
| 38 | /* |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 39 | * directory trouble cases |
| 40 | * |
| 41 | * 1) on rename or unlink, if the inode being unlinked isn't in the fsync |
| 42 | * log, we must force a full commit before doing an fsync of the directory |
| 43 | * where the unlink was done. |
| 44 | * ---> record transid of last unlink/rename per directory |
| 45 | * |
| 46 | * mkdir foo/some_dir |
| 47 | * normal commit |
| 48 | * rename foo/some_dir foo2/some_dir |
| 49 | * mkdir foo/some_dir |
| 50 | * fsync foo/some_dir/some_file |
| 51 | * |
| 52 | * The fsync above will unlink the original some_dir without recording |
| 53 | * it in its new location (foo2). After a crash, some_dir will be gone |
| 54 | * unless the fsync of some_file forces a full commit |
| 55 | * |
| 56 | * 2) we must log any new names for any file or dir that is in the fsync |
| 57 | * log. ---> check inode while renaming/linking. |
| 58 | * |
| 59 | * 2a) we must log any new names for any file or dir during rename |
| 60 | * when the directory they are being removed from was logged. |
| 61 | * ---> check inode and old parent dir during rename |
| 62 | * |
| 63 | * 2a is actually the more important variant. With the extra logging |
| 64 | * a crash might unlink the old name without recreating the new one |
| 65 | * |
| 66 | * 3) after a crash, we must go through any directories with a link count |
| 67 | * of zero and redo the rm -rf |
| 68 | * |
| 69 | * mkdir f1/foo |
| 70 | * normal commit |
| 71 | * rm -rf f1/foo |
| 72 | * fsync(f1) |
| 73 | * |
| 74 | * The directory f1 was fully removed from the FS, but fsync was never |
| 75 | * called on f1, only its parent dir. After a crash the rm -rf must |
| 76 | * be replayed. This must be able to recurse down the entire |
| 77 | * directory tree. The inode link count fixup code takes care of the |
| 78 | * ugly details. |
| 79 | */ |
| 80 | |
| 81 | /* |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 82 | * stages for the tree walking. The first |
| 83 | * stage (0) is to only pin down the blocks we find |
| 84 | * the second stage (1) is to make sure that all the inodes |
| 85 | * we find in the log are created in the subvolume. |
| 86 | * |
| 87 | * The last stage is to deal with directories and links and extents |
| 88 | * and all the other fun semantics |
| 89 | */ |
| 90 | #define LOG_WALK_PIN_ONLY 0 |
| 91 | #define LOG_WALK_REPLAY_INODES 1 |
| 92 | #define LOG_WALK_REPLAY_ALL 2 |
| 93 | |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 94 | static int btrfs_log_inode(struct btrfs_trans_handle *trans, |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 95 | struct btrfs_root *root, struct inode *inode, |
| 96 | int inode_only); |
Yan Zheng | ec051c0 | 2009-01-05 15:43:42 -0500 | [diff] [blame] | 97 | static int link_to_fixup_dir(struct btrfs_trans_handle *trans, |
| 98 | struct btrfs_root *root, |
| 99 | struct btrfs_path *path, u64 objectid); |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 100 | static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans, |
| 101 | struct btrfs_root *root, |
| 102 | struct btrfs_root *log, |
| 103 | struct btrfs_path *path, |
| 104 | u64 dirid, int del_all); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 105 | |
| 106 | /* |
| 107 | * tree logging is a special write ahead log used to make sure that |
| 108 | * fsyncs and O_SYNCs can happen without doing full tree commits. |
| 109 | * |
| 110 | * Full tree commits are expensive because they require commonly |
| 111 | * modified blocks to be recowed, creating many dirty pages in the |
| 112 | * extent tree an 4x-6x higher write load than ext3. |
| 113 | * |
| 114 | * Instead of doing a tree commit on every fsync, we use the |
| 115 | * key ranges and transaction ids to find items for a given file or directory |
| 116 | * that have changed in this transaction. Those items are copied into |
| 117 | * a special tree (one per subvolume root), that tree is written to disk |
| 118 | * and then the fsync is considered complete. |
| 119 | * |
| 120 | * After a crash, items are copied out of the log-tree back into the |
| 121 | * subvolume tree. Any file data extents found are recorded in the extent |
| 122 | * allocation tree, and the log-tree freed. |
| 123 | * |
| 124 | * The log tree is read three times, once to pin down all the extents it is |
| 125 | * using in ram and once, once to create all the inodes logged in the tree |
| 126 | * and once to do all the other items. |
| 127 | */ |
| 128 | |
| 129 | /* |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 130 | * start a sub transaction and setup the log tree |
| 131 | * this increments the log tree writer count to make the people |
| 132 | * syncing the tree wait for us to finish |
| 133 | */ |
| 134 | static int start_log_trans(struct btrfs_trans_handle *trans, |
| 135 | struct btrfs_root *root) |
| 136 | { |
| 137 | int ret; |
Yan, Zheng | 4a500fd | 2010-05-16 10:49:59 -0400 | [diff] [blame] | 138 | int err = 0; |
Yan Zheng | 7237f18 | 2009-01-21 12:54:03 -0500 | [diff] [blame] | 139 | |
| 140 | mutex_lock(&root->log_mutex); |
| 141 | if (root->log_root) { |
Josef Bacik | ff782e0 | 2009-10-08 15:30:04 -0400 | [diff] [blame] | 142 | if (!root->log_start_pid) { |
| 143 | root->log_start_pid = current->pid; |
| 144 | root->log_multiple_pids = false; |
| 145 | } else if (root->log_start_pid != current->pid) { |
| 146 | root->log_multiple_pids = true; |
| 147 | } |
| 148 | |
Yan Zheng | 7237f18 | 2009-01-21 12:54:03 -0500 | [diff] [blame] | 149 | root->log_batch++; |
| 150 | atomic_inc(&root->log_writers); |
| 151 | mutex_unlock(&root->log_mutex); |
| 152 | return 0; |
| 153 | } |
Josef Bacik | ff782e0 | 2009-10-08 15:30:04 -0400 | [diff] [blame] | 154 | root->log_multiple_pids = false; |
| 155 | root->log_start_pid = current->pid; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 156 | mutex_lock(&root->fs_info->tree_log_mutex); |
| 157 | if (!root->fs_info->log_root_tree) { |
| 158 | ret = btrfs_init_log_root_tree(trans, root->fs_info); |
Yan, Zheng | 4a500fd | 2010-05-16 10:49:59 -0400 | [diff] [blame] | 159 | if (ret) |
| 160 | err = ret; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 161 | } |
Yan, Zheng | 4a500fd | 2010-05-16 10:49:59 -0400 | [diff] [blame] | 162 | if (err == 0 && !root->log_root) { |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 163 | ret = btrfs_add_log_tree(trans, root); |
Yan, Zheng | 4a500fd | 2010-05-16 10:49:59 -0400 | [diff] [blame] | 164 | if (ret) |
| 165 | err = ret; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 166 | } |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 167 | mutex_unlock(&root->fs_info->tree_log_mutex); |
Yan Zheng | 7237f18 | 2009-01-21 12:54:03 -0500 | [diff] [blame] | 168 | root->log_batch++; |
| 169 | atomic_inc(&root->log_writers); |
| 170 | mutex_unlock(&root->log_mutex); |
Yan, Zheng | 4a500fd | 2010-05-16 10:49:59 -0400 | [diff] [blame] | 171 | return err; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 172 | } |
| 173 | |
| 174 | /* |
| 175 | * returns 0 if there was a log transaction running and we were able |
| 176 | * to join, or returns -ENOENT if there were not transactions |
| 177 | * in progress |
| 178 | */ |
| 179 | static int join_running_log_trans(struct btrfs_root *root) |
| 180 | { |
| 181 | int ret = -ENOENT; |
| 182 | |
| 183 | smp_mb(); |
| 184 | if (!root->log_root) |
| 185 | return -ENOENT; |
| 186 | |
Yan Zheng | 7237f18 | 2009-01-21 12:54:03 -0500 | [diff] [blame] | 187 | mutex_lock(&root->log_mutex); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 188 | if (root->log_root) { |
| 189 | ret = 0; |
Yan Zheng | 7237f18 | 2009-01-21 12:54:03 -0500 | [diff] [blame] | 190 | atomic_inc(&root->log_writers); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 191 | } |
Yan Zheng | 7237f18 | 2009-01-21 12:54:03 -0500 | [diff] [blame] | 192 | mutex_unlock(&root->log_mutex); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 193 | return ret; |
| 194 | } |
| 195 | |
| 196 | /* |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 197 | * This either makes the current running log transaction wait |
| 198 | * until you call btrfs_end_log_trans() or it makes any future |
| 199 | * log transactions wait until you call btrfs_end_log_trans() |
| 200 | */ |
| 201 | int btrfs_pin_log_trans(struct btrfs_root *root) |
| 202 | { |
| 203 | int ret = -ENOENT; |
| 204 | |
| 205 | mutex_lock(&root->log_mutex); |
| 206 | atomic_inc(&root->log_writers); |
| 207 | mutex_unlock(&root->log_mutex); |
| 208 | return ret; |
| 209 | } |
| 210 | |
| 211 | /* |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 212 | * indicate we're done making changes to the log tree |
| 213 | * and wake up anyone waiting to do a sync |
| 214 | */ |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 215 | int btrfs_end_log_trans(struct btrfs_root *root) |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 216 | { |
Yan Zheng | 7237f18 | 2009-01-21 12:54:03 -0500 | [diff] [blame] | 217 | if (atomic_dec_and_test(&root->log_writers)) { |
| 218 | smp_mb(); |
| 219 | if (waitqueue_active(&root->log_writer_wait)) |
| 220 | wake_up(&root->log_writer_wait); |
| 221 | } |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 222 | return 0; |
| 223 | } |
| 224 | |
| 225 | |
| 226 | /* |
| 227 | * the walk control struct is used to pass state down the chain when |
| 228 | * processing the log tree. The stage field tells us which part |
| 229 | * of the log tree processing we are currently doing. The others |
| 230 | * are state fields used for that specific part |
| 231 | */ |
| 232 | struct walk_control { |
| 233 | /* should we free the extent on disk when done? This is used |
| 234 | * at transaction commit time while freeing a log tree |
| 235 | */ |
| 236 | int free; |
| 237 | |
| 238 | /* should we write out the extent buffer? This is used |
| 239 | * while flushing the log tree to disk during a sync |
| 240 | */ |
| 241 | int write; |
| 242 | |
| 243 | /* should we wait for the extent buffer io to finish? Also used |
| 244 | * while flushing the log tree to disk for a sync |
| 245 | */ |
| 246 | int wait; |
| 247 | |
| 248 | /* pin only walk, we record which extents on disk belong to the |
| 249 | * log trees |
| 250 | */ |
| 251 | int pin; |
| 252 | |
| 253 | /* what stage of the replay code we're currently in */ |
| 254 | int stage; |
| 255 | |
| 256 | /* the root we are currently replaying */ |
| 257 | struct btrfs_root *replay_dest; |
| 258 | |
| 259 | /* the trans handle for the current replay */ |
| 260 | struct btrfs_trans_handle *trans; |
| 261 | |
| 262 | /* the function that gets used to process blocks we find in the |
| 263 | * tree. Note the extent_buffer might not be up to date when it is |
| 264 | * passed in, and it must be checked or read if you need the data |
| 265 | * inside it |
| 266 | */ |
| 267 | int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb, |
| 268 | struct walk_control *wc, u64 gen); |
| 269 | }; |
| 270 | |
| 271 | /* |
| 272 | * process_func used to pin down extents, write them or wait on them |
| 273 | */ |
| 274 | static int process_one_buffer(struct btrfs_root *log, |
| 275 | struct extent_buffer *eb, |
| 276 | struct walk_control *wc, u64 gen) |
| 277 | { |
Josef Bacik | 04018de | 2009-04-03 10:14:18 -0400 | [diff] [blame] | 278 | if (wc->pin) |
Chris Mason | e688b72 | 2011-10-31 20:52:39 -0400 | [diff] [blame] | 279 | btrfs_pin_extent_for_log_replay(wc->trans, |
| 280 | log->fs_info->extent_root, |
| 281 | eb->start, eb->len); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 282 | |
| 283 | if (btrfs_buffer_uptodate(eb, gen)) { |
| 284 | if (wc->write) |
| 285 | btrfs_write_tree_block(eb); |
| 286 | if (wc->wait) |
| 287 | btrfs_wait_tree_block_writeback(eb); |
| 288 | } |
| 289 | return 0; |
| 290 | } |
| 291 | |
| 292 | /* |
| 293 | * Item overwrite used by replay and tree logging. eb, slot and key all refer |
| 294 | * to the src data we are copying out. |
| 295 | * |
| 296 | * root is the tree we are copying into, and path is a scratch |
| 297 | * path for use in this function (it should be released on entry and |
| 298 | * will be released on exit). |
| 299 | * |
| 300 | * If the key is already in the destination tree the existing item is |
| 301 | * overwritten. If the existing item isn't big enough, it is extended. |
| 302 | * If it is too large, it is truncated. |
| 303 | * |
| 304 | * If the key isn't in the destination yet, a new item is inserted. |
| 305 | */ |
| 306 | static noinline int overwrite_item(struct btrfs_trans_handle *trans, |
| 307 | struct btrfs_root *root, |
| 308 | struct btrfs_path *path, |
| 309 | struct extent_buffer *eb, int slot, |
| 310 | struct btrfs_key *key) |
| 311 | { |
| 312 | int ret; |
| 313 | u32 item_size; |
| 314 | u64 saved_i_size = 0; |
| 315 | int save_old_i_size = 0; |
| 316 | unsigned long src_ptr; |
| 317 | unsigned long dst_ptr; |
| 318 | int overwrite_root = 0; |
| 319 | |
| 320 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) |
| 321 | overwrite_root = 1; |
| 322 | |
| 323 | item_size = btrfs_item_size_nr(eb, slot); |
| 324 | src_ptr = btrfs_item_ptr_offset(eb, slot); |
| 325 | |
| 326 | /* look for the key in the destination tree */ |
| 327 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
| 328 | if (ret == 0) { |
| 329 | char *src_copy; |
| 330 | char *dst_copy; |
| 331 | u32 dst_size = btrfs_item_size_nr(path->nodes[0], |
| 332 | path->slots[0]); |
| 333 | if (dst_size != item_size) |
| 334 | goto insert; |
| 335 | |
| 336 | if (item_size == 0) { |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 337 | btrfs_release_path(path); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 338 | return 0; |
| 339 | } |
| 340 | dst_copy = kmalloc(item_size, GFP_NOFS); |
| 341 | src_copy = kmalloc(item_size, GFP_NOFS); |
liubo | 2a29edc | 2011-01-26 06:22:08 +0000 | [diff] [blame] | 342 | if (!dst_copy || !src_copy) { |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 343 | btrfs_release_path(path); |
liubo | 2a29edc | 2011-01-26 06:22:08 +0000 | [diff] [blame] | 344 | kfree(dst_copy); |
| 345 | kfree(src_copy); |
| 346 | return -ENOMEM; |
| 347 | } |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 348 | |
| 349 | read_extent_buffer(eb, src_copy, src_ptr, item_size); |
| 350 | |
| 351 | dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]); |
| 352 | read_extent_buffer(path->nodes[0], dst_copy, dst_ptr, |
| 353 | item_size); |
| 354 | ret = memcmp(dst_copy, src_copy, item_size); |
| 355 | |
| 356 | kfree(dst_copy); |
| 357 | kfree(src_copy); |
| 358 | /* |
| 359 | * they have the same contents, just return, this saves |
| 360 | * us from cowing blocks in the destination tree and doing |
| 361 | * extra writes that may not have been done by a previous |
| 362 | * sync |
| 363 | */ |
| 364 | if (ret == 0) { |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 365 | btrfs_release_path(path); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 366 | return 0; |
| 367 | } |
| 368 | |
| 369 | } |
| 370 | insert: |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 371 | btrfs_release_path(path); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 372 | /* try to insert the key into the destination tree */ |
| 373 | ret = btrfs_insert_empty_item(trans, root, path, |
| 374 | key, item_size); |
| 375 | |
| 376 | /* make sure any existing item is the correct size */ |
| 377 | if (ret == -EEXIST) { |
| 378 | u32 found_size; |
| 379 | found_size = btrfs_item_size_nr(path->nodes[0], |
| 380 | path->slots[0]); |
| 381 | if (found_size > item_size) { |
| 382 | btrfs_truncate_item(trans, root, path, item_size, 1); |
| 383 | } else if (found_size < item_size) { |
Yan Zheng | 87b29b2 | 2008-12-17 10:21:48 -0500 | [diff] [blame] | 384 | ret = btrfs_extend_item(trans, root, path, |
| 385 | item_size - found_size); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 386 | } |
| 387 | } else if (ret) { |
Yan, Zheng | 4a500fd | 2010-05-16 10:49:59 -0400 | [diff] [blame] | 388 | return ret; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 389 | } |
| 390 | dst_ptr = btrfs_item_ptr_offset(path->nodes[0], |
| 391 | path->slots[0]); |
| 392 | |
| 393 | /* don't overwrite an existing inode if the generation number |
| 394 | * was logged as zero. This is done when the tree logging code |
| 395 | * is just logging an inode to make sure it exists after recovery. |
| 396 | * |
| 397 | * Also, don't overwrite i_size on directories during replay. |
| 398 | * log replay inserts and removes directory items based on the |
| 399 | * state of the tree found in the subvolume, and i_size is modified |
| 400 | * as it goes |
| 401 | */ |
| 402 | if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) { |
| 403 | struct btrfs_inode_item *src_item; |
| 404 | struct btrfs_inode_item *dst_item; |
| 405 | |
| 406 | src_item = (struct btrfs_inode_item *)src_ptr; |
| 407 | dst_item = (struct btrfs_inode_item *)dst_ptr; |
| 408 | |
| 409 | if (btrfs_inode_generation(eb, src_item) == 0) |
| 410 | goto no_copy; |
| 411 | |
| 412 | if (overwrite_root && |
| 413 | S_ISDIR(btrfs_inode_mode(eb, src_item)) && |
| 414 | S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) { |
| 415 | save_old_i_size = 1; |
| 416 | saved_i_size = btrfs_inode_size(path->nodes[0], |
| 417 | dst_item); |
| 418 | } |
| 419 | } |
| 420 | |
| 421 | copy_extent_buffer(path->nodes[0], eb, dst_ptr, |
| 422 | src_ptr, item_size); |
| 423 | |
| 424 | if (save_old_i_size) { |
| 425 | struct btrfs_inode_item *dst_item; |
| 426 | dst_item = (struct btrfs_inode_item *)dst_ptr; |
| 427 | btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size); |
| 428 | } |
| 429 | |
| 430 | /* make sure the generation is filled in */ |
| 431 | if (key->type == BTRFS_INODE_ITEM_KEY) { |
| 432 | struct btrfs_inode_item *dst_item; |
| 433 | dst_item = (struct btrfs_inode_item *)dst_ptr; |
| 434 | if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) { |
| 435 | btrfs_set_inode_generation(path->nodes[0], dst_item, |
| 436 | trans->transid); |
| 437 | } |
| 438 | } |
| 439 | no_copy: |
| 440 | btrfs_mark_buffer_dirty(path->nodes[0]); |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 441 | btrfs_release_path(path); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 442 | return 0; |
| 443 | } |
| 444 | |
| 445 | /* |
| 446 | * simple helper to read an inode off the disk from a given root |
| 447 | * This can only be called for subvolume roots and not for the log |
| 448 | */ |
| 449 | static noinline struct inode *read_one_inode(struct btrfs_root *root, |
| 450 | u64 objectid) |
| 451 | { |
Yan Zheng | 5d4f98a | 2009-06-10 10:45:14 -0400 | [diff] [blame] | 452 | struct btrfs_key key; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 453 | struct inode *inode; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 454 | |
Yan Zheng | 5d4f98a | 2009-06-10 10:45:14 -0400 | [diff] [blame] | 455 | key.objectid = objectid; |
| 456 | key.type = BTRFS_INODE_ITEM_KEY; |
| 457 | key.offset = 0; |
Josef Bacik | 73f7341 | 2009-12-04 17:38:27 +0000 | [diff] [blame] | 458 | inode = btrfs_iget(root->fs_info->sb, &key, root, NULL); |
Yan Zheng | 5d4f98a | 2009-06-10 10:45:14 -0400 | [diff] [blame] | 459 | if (IS_ERR(inode)) { |
| 460 | inode = NULL; |
| 461 | } else if (is_bad_inode(inode)) { |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 462 | iput(inode); |
| 463 | inode = NULL; |
| 464 | } |
| 465 | return inode; |
| 466 | } |
| 467 | |
| 468 | /* replays a single extent in 'eb' at 'slot' with 'key' into the |
| 469 | * subvolume 'root'. path is released on entry and should be released |
| 470 | * on exit. |
| 471 | * |
| 472 | * extents in the log tree have not been allocated out of the extent |
| 473 | * tree yet. So, this completes the allocation, taking a reference |
| 474 | * as required if the extent already exists or creating a new extent |
| 475 | * if it isn't in the extent allocation tree yet. |
| 476 | * |
| 477 | * The extent is inserted into the file, dropping any existing extents |
| 478 | * from the file that overlap the new one. |
| 479 | */ |
| 480 | static noinline int replay_one_extent(struct btrfs_trans_handle *trans, |
| 481 | struct btrfs_root *root, |
| 482 | struct btrfs_path *path, |
| 483 | struct extent_buffer *eb, int slot, |
| 484 | struct btrfs_key *key) |
| 485 | { |
| 486 | int found_type; |
| 487 | u64 mask = root->sectorsize - 1; |
| 488 | u64 extent_end; |
| 489 | u64 alloc_hint; |
| 490 | u64 start = key->offset; |
Yan Zheng | 07d400a | 2009-01-06 11:42:00 -0500 | [diff] [blame] | 491 | u64 saved_nbytes; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 492 | struct btrfs_file_extent_item *item; |
| 493 | struct inode *inode = NULL; |
| 494 | unsigned long size; |
| 495 | int ret = 0; |
| 496 | |
| 497 | item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); |
| 498 | found_type = btrfs_file_extent_type(eb, item); |
| 499 | |
Yan Zheng | d899e05 | 2008-10-30 14:25:28 -0400 | [diff] [blame] | 500 | if (found_type == BTRFS_FILE_EXTENT_REG || |
| 501 | found_type == BTRFS_FILE_EXTENT_PREALLOC) |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 502 | extent_end = start + btrfs_file_extent_num_bytes(eb, item); |
| 503 | else if (found_type == BTRFS_FILE_EXTENT_INLINE) { |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 504 | size = btrfs_file_extent_inline_len(eb, item); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 505 | extent_end = (start + size + mask) & ~mask; |
| 506 | } else { |
| 507 | ret = 0; |
| 508 | goto out; |
| 509 | } |
| 510 | |
| 511 | inode = read_one_inode(root, key->objectid); |
| 512 | if (!inode) { |
| 513 | ret = -EIO; |
| 514 | goto out; |
| 515 | } |
| 516 | |
| 517 | /* |
| 518 | * first check to see if we already have this extent in the |
| 519 | * file. This must be done before the btrfs_drop_extents run |
| 520 | * so we don't try to drop this extent. |
| 521 | */ |
Li Zefan | 33345d01 | 2011-04-20 10:31:50 +0800 | [diff] [blame] | 522 | ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode), |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 523 | start, 0); |
| 524 | |
Yan Zheng | d899e05 | 2008-10-30 14:25:28 -0400 | [diff] [blame] | 525 | if (ret == 0 && |
| 526 | (found_type == BTRFS_FILE_EXTENT_REG || |
| 527 | found_type == BTRFS_FILE_EXTENT_PREALLOC)) { |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 528 | struct btrfs_file_extent_item cmp1; |
| 529 | struct btrfs_file_extent_item cmp2; |
| 530 | struct btrfs_file_extent_item *existing; |
| 531 | struct extent_buffer *leaf; |
| 532 | |
| 533 | leaf = path->nodes[0]; |
| 534 | existing = btrfs_item_ptr(leaf, path->slots[0], |
| 535 | struct btrfs_file_extent_item); |
| 536 | |
| 537 | read_extent_buffer(eb, &cmp1, (unsigned long)item, |
| 538 | sizeof(cmp1)); |
| 539 | read_extent_buffer(leaf, &cmp2, (unsigned long)existing, |
| 540 | sizeof(cmp2)); |
| 541 | |
| 542 | /* |
| 543 | * we already have a pointer to this exact extent, |
| 544 | * we don't have to do anything |
| 545 | */ |
| 546 | if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) { |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 547 | btrfs_release_path(path); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 548 | goto out; |
| 549 | } |
| 550 | } |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 551 | btrfs_release_path(path); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 552 | |
Yan Zheng | 07d400a | 2009-01-06 11:42:00 -0500 | [diff] [blame] | 553 | saved_nbytes = inode_get_bytes(inode); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 554 | /* drop any overlapping extents */ |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 555 | ret = btrfs_drop_extents(trans, inode, start, extent_end, |
| 556 | &alloc_hint, 1); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 557 | BUG_ON(ret); |
| 558 | |
Yan Zheng | 07d400a | 2009-01-06 11:42:00 -0500 | [diff] [blame] | 559 | if (found_type == BTRFS_FILE_EXTENT_REG || |
| 560 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { |
Yan Zheng | 5d4f98a | 2009-06-10 10:45:14 -0400 | [diff] [blame] | 561 | u64 offset; |
Yan Zheng | 07d400a | 2009-01-06 11:42:00 -0500 | [diff] [blame] | 562 | unsigned long dest_offset; |
| 563 | struct btrfs_key ins; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 564 | |
Yan Zheng | 07d400a | 2009-01-06 11:42:00 -0500 | [diff] [blame] | 565 | ret = btrfs_insert_empty_item(trans, root, path, key, |
| 566 | sizeof(*item)); |
| 567 | BUG_ON(ret); |
| 568 | dest_offset = btrfs_item_ptr_offset(path->nodes[0], |
| 569 | path->slots[0]); |
| 570 | copy_extent_buffer(path->nodes[0], eb, dest_offset, |
| 571 | (unsigned long)item, sizeof(*item)); |
| 572 | |
| 573 | ins.objectid = btrfs_file_extent_disk_bytenr(eb, item); |
| 574 | ins.offset = btrfs_file_extent_disk_num_bytes(eb, item); |
| 575 | ins.type = BTRFS_EXTENT_ITEM_KEY; |
Yan Zheng | 5d4f98a | 2009-06-10 10:45:14 -0400 | [diff] [blame] | 576 | offset = key->offset - btrfs_file_extent_offset(eb, item); |
Yan Zheng | 07d400a | 2009-01-06 11:42:00 -0500 | [diff] [blame] | 577 | |
| 578 | if (ins.objectid > 0) { |
| 579 | u64 csum_start; |
| 580 | u64 csum_end; |
| 581 | LIST_HEAD(ordered_sums); |
| 582 | /* |
| 583 | * is this extent already allocated in the extent |
| 584 | * allocation tree? If so, just add a reference |
| 585 | */ |
| 586 | ret = btrfs_lookup_extent(root, ins.objectid, |
| 587 | ins.offset); |
| 588 | if (ret == 0) { |
| 589 | ret = btrfs_inc_extent_ref(trans, root, |
| 590 | ins.objectid, ins.offset, |
Yan Zheng | 5d4f98a | 2009-06-10 10:45:14 -0400 | [diff] [blame] | 591 | 0, root->root_key.objectid, |
Arne Jansen | 66d7e7f | 2011-09-12 15:26:38 +0200 | [diff] [blame] | 592 | key->objectid, offset, 0); |
Tsutomu Itoh | 37daa4f | 2011-04-28 09:18:21 +0000 | [diff] [blame] | 593 | BUG_ON(ret); |
Yan Zheng | 07d400a | 2009-01-06 11:42:00 -0500 | [diff] [blame] | 594 | } else { |
| 595 | /* |
| 596 | * insert the extent pointer in the extent |
| 597 | * allocation tree |
| 598 | */ |
Yan Zheng | 5d4f98a | 2009-06-10 10:45:14 -0400 | [diff] [blame] | 599 | ret = btrfs_alloc_logged_file_extent(trans, |
| 600 | root, root->root_key.objectid, |
| 601 | key->objectid, offset, &ins); |
Yan Zheng | 07d400a | 2009-01-06 11:42:00 -0500 | [diff] [blame] | 602 | BUG_ON(ret); |
| 603 | } |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 604 | btrfs_release_path(path); |
Yan Zheng | 07d400a | 2009-01-06 11:42:00 -0500 | [diff] [blame] | 605 | |
| 606 | if (btrfs_file_extent_compression(eb, item)) { |
| 607 | csum_start = ins.objectid; |
| 608 | csum_end = csum_start + ins.offset; |
| 609 | } else { |
| 610 | csum_start = ins.objectid + |
| 611 | btrfs_file_extent_offset(eb, item); |
| 612 | csum_end = csum_start + |
| 613 | btrfs_file_extent_num_bytes(eb, item); |
| 614 | } |
| 615 | |
| 616 | ret = btrfs_lookup_csums_range(root->log_root, |
| 617 | csum_start, csum_end - 1, |
Arne Jansen | a2de733 | 2011-03-08 14:14:00 +0100 | [diff] [blame] | 618 | &ordered_sums, 0); |
Yan Zheng | 07d400a | 2009-01-06 11:42:00 -0500 | [diff] [blame] | 619 | BUG_ON(ret); |
| 620 | while (!list_empty(&ordered_sums)) { |
| 621 | struct btrfs_ordered_sum *sums; |
| 622 | sums = list_entry(ordered_sums.next, |
| 623 | struct btrfs_ordered_sum, |
| 624 | list); |
| 625 | ret = btrfs_csum_file_blocks(trans, |
| 626 | root->fs_info->csum_root, |
| 627 | sums); |
| 628 | BUG_ON(ret); |
| 629 | list_del(&sums->list); |
| 630 | kfree(sums); |
| 631 | } |
| 632 | } else { |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 633 | btrfs_release_path(path); |
Yan Zheng | 07d400a | 2009-01-06 11:42:00 -0500 | [diff] [blame] | 634 | } |
| 635 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { |
| 636 | /* inline extents are easy, we just overwrite them */ |
| 637 | ret = overwrite_item(trans, root, path, eb, slot, key); |
| 638 | BUG_ON(ret); |
| 639 | } |
| 640 | |
| 641 | inode_set_bytes(inode, saved_nbytes); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 642 | btrfs_update_inode(trans, root, inode); |
| 643 | out: |
| 644 | if (inode) |
| 645 | iput(inode); |
| 646 | return ret; |
| 647 | } |
| 648 | |
| 649 | /* |
| 650 | * when cleaning up conflicts between the directory names in the |
| 651 | * subvolume, directory names in the log and directory names in the |
| 652 | * inode back references, we may have to unlink inodes from directories. |
| 653 | * |
| 654 | * This is a helper function to do the unlink of a specific directory |
| 655 | * item |
| 656 | */ |
| 657 | static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans, |
| 658 | struct btrfs_root *root, |
| 659 | struct btrfs_path *path, |
| 660 | struct inode *dir, |
| 661 | struct btrfs_dir_item *di) |
| 662 | { |
| 663 | struct inode *inode; |
| 664 | char *name; |
| 665 | int name_len; |
| 666 | struct extent_buffer *leaf; |
| 667 | struct btrfs_key location; |
| 668 | int ret; |
| 669 | |
| 670 | leaf = path->nodes[0]; |
| 671 | |
| 672 | btrfs_dir_item_key_to_cpu(leaf, di, &location); |
| 673 | name_len = btrfs_dir_name_len(leaf, di); |
| 674 | name = kmalloc(name_len, GFP_NOFS); |
liubo | 2a29edc | 2011-01-26 06:22:08 +0000 | [diff] [blame] | 675 | if (!name) |
| 676 | return -ENOMEM; |
| 677 | |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 678 | read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len); |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 679 | btrfs_release_path(path); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 680 | |
| 681 | inode = read_one_inode(root, location.objectid); |
Tsutomu Itoh | c00e949 | 2011-04-28 09:10:23 +0000 | [diff] [blame] | 682 | if (!inode) { |
| 683 | kfree(name); |
| 684 | return -EIO; |
| 685 | } |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 686 | |
Yan Zheng | ec051c0 | 2009-01-05 15:43:42 -0500 | [diff] [blame] | 687 | ret = link_to_fixup_dir(trans, root, path, location.objectid); |
| 688 | BUG_ON(ret); |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 689 | |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 690 | ret = btrfs_unlink_inode(trans, root, dir, inode, name, name_len); |
Yan Zheng | ec051c0 | 2009-01-05 15:43:42 -0500 | [diff] [blame] | 691 | BUG_ON(ret); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 692 | kfree(name); |
| 693 | |
| 694 | iput(inode); |
| 695 | return ret; |
| 696 | } |
| 697 | |
| 698 | /* |
| 699 | * helper function to see if a given name and sequence number found |
| 700 | * in an inode back reference are already in a directory and correctly |
| 701 | * point to this inode |
| 702 | */ |
| 703 | static noinline int inode_in_dir(struct btrfs_root *root, |
| 704 | struct btrfs_path *path, |
| 705 | u64 dirid, u64 objectid, u64 index, |
| 706 | const char *name, int name_len) |
| 707 | { |
| 708 | struct btrfs_dir_item *di; |
| 709 | struct btrfs_key location; |
| 710 | int match = 0; |
| 711 | |
| 712 | di = btrfs_lookup_dir_index_item(NULL, root, path, dirid, |
| 713 | index, name, name_len, 0); |
| 714 | if (di && !IS_ERR(di)) { |
| 715 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location); |
| 716 | if (location.objectid != objectid) |
| 717 | goto out; |
| 718 | } else |
| 719 | goto out; |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 720 | btrfs_release_path(path); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 721 | |
| 722 | di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0); |
| 723 | if (di && !IS_ERR(di)) { |
| 724 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location); |
| 725 | if (location.objectid != objectid) |
| 726 | goto out; |
| 727 | } else |
| 728 | goto out; |
| 729 | match = 1; |
| 730 | out: |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 731 | btrfs_release_path(path); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 732 | return match; |
| 733 | } |
| 734 | |
| 735 | /* |
| 736 | * helper function to check a log tree for a named back reference in |
| 737 | * an inode. This is used to decide if a back reference that is |
| 738 | * found in the subvolume conflicts with what we find in the log. |
| 739 | * |
| 740 | * inode backreferences may have multiple refs in a single item, |
| 741 | * during replay we process one reference at a time, and we don't |
| 742 | * want to delete valid links to a file from the subvolume if that |
| 743 | * link is also in the log. |
| 744 | */ |
| 745 | static noinline int backref_in_log(struct btrfs_root *log, |
| 746 | struct btrfs_key *key, |
| 747 | char *name, int namelen) |
| 748 | { |
| 749 | struct btrfs_path *path; |
| 750 | struct btrfs_inode_ref *ref; |
| 751 | unsigned long ptr; |
| 752 | unsigned long ptr_end; |
| 753 | unsigned long name_ptr; |
| 754 | int found_name_len; |
| 755 | int item_size; |
| 756 | int ret; |
| 757 | int match = 0; |
| 758 | |
| 759 | path = btrfs_alloc_path(); |
liubo | 2a29edc | 2011-01-26 06:22:08 +0000 | [diff] [blame] | 760 | if (!path) |
| 761 | return -ENOMEM; |
| 762 | |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 763 | ret = btrfs_search_slot(NULL, log, key, path, 0, 0); |
| 764 | if (ret != 0) |
| 765 | goto out; |
| 766 | |
| 767 | item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]); |
| 768 | ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]); |
| 769 | ptr_end = ptr + item_size; |
| 770 | while (ptr < ptr_end) { |
| 771 | ref = (struct btrfs_inode_ref *)ptr; |
| 772 | found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref); |
| 773 | if (found_name_len == namelen) { |
| 774 | name_ptr = (unsigned long)(ref + 1); |
| 775 | ret = memcmp_extent_buffer(path->nodes[0], name, |
| 776 | name_ptr, namelen); |
| 777 | if (ret == 0) { |
| 778 | match = 1; |
| 779 | goto out; |
| 780 | } |
| 781 | } |
| 782 | ptr = (unsigned long)(ref + 1) + found_name_len; |
| 783 | } |
| 784 | out: |
| 785 | btrfs_free_path(path); |
| 786 | return match; |
| 787 | } |
| 788 | |
| 789 | |
| 790 | /* |
| 791 | * replay one inode back reference item found in the log tree. |
| 792 | * eb, slot and key refer to the buffer and key found in the log tree. |
| 793 | * root is the destination we are replaying into, and path is for temp |
| 794 | * use by this function. (it should be released on return). |
| 795 | */ |
| 796 | static noinline int add_inode_ref(struct btrfs_trans_handle *trans, |
| 797 | struct btrfs_root *root, |
| 798 | struct btrfs_root *log, |
| 799 | struct btrfs_path *path, |
| 800 | struct extent_buffer *eb, int slot, |
| 801 | struct btrfs_key *key) |
| 802 | { |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 803 | struct btrfs_inode_ref *ref; |
liubo | 34f3e4f | 2011-08-06 08:35:23 +0000 | [diff] [blame] | 804 | struct btrfs_dir_item *di; |
| 805 | struct inode *dir; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 806 | struct inode *inode; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 807 | unsigned long ref_ptr; |
| 808 | unsigned long ref_end; |
liubo | 34f3e4f | 2011-08-06 08:35:23 +0000 | [diff] [blame] | 809 | char *name; |
| 810 | int namelen; |
| 811 | int ret; |
liubo | c622ae6 | 2011-03-26 08:01:12 -0400 | [diff] [blame] | 812 | int search_done = 0; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 813 | |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 814 | /* |
| 815 | * it is possible that we didn't log all the parent directories |
| 816 | * for a given inode. If we don't find the dir, just don't |
| 817 | * copy the back ref in. The link count fixup code will take |
| 818 | * care of the rest |
| 819 | */ |
| 820 | dir = read_one_inode(root, key->offset); |
| 821 | if (!dir) |
| 822 | return -ENOENT; |
| 823 | |
| 824 | inode = read_one_inode(root, key->objectid); |
Tsutomu Itoh | c00e949 | 2011-04-28 09:10:23 +0000 | [diff] [blame] | 825 | if (!inode) { |
| 826 | iput(dir); |
| 827 | return -EIO; |
| 828 | } |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 829 | |
| 830 | ref_ptr = btrfs_item_ptr_offset(eb, slot); |
| 831 | ref_end = ref_ptr + btrfs_item_size_nr(eb, slot); |
| 832 | |
| 833 | again: |
| 834 | ref = (struct btrfs_inode_ref *)ref_ptr; |
| 835 | |
| 836 | namelen = btrfs_inode_ref_name_len(eb, ref); |
| 837 | name = kmalloc(namelen, GFP_NOFS); |
| 838 | BUG_ON(!name); |
| 839 | |
| 840 | read_extent_buffer(eb, name, (unsigned long)(ref + 1), namelen); |
| 841 | |
| 842 | /* if we already have a perfect match, we're done */ |
Li Zefan | 33345d01 | 2011-04-20 10:31:50 +0800 | [diff] [blame] | 843 | if (inode_in_dir(root, path, btrfs_ino(dir), btrfs_ino(inode), |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 844 | btrfs_inode_ref_index(eb, ref), |
| 845 | name, namelen)) { |
| 846 | goto out; |
| 847 | } |
| 848 | |
| 849 | /* |
| 850 | * look for a conflicting back reference in the metadata. |
| 851 | * if we find one we have to unlink that name of the file |
| 852 | * before we add our new link. Later on, we overwrite any |
| 853 | * existing back reference, and we don't want to create |
| 854 | * dangling pointers in the directory. |
| 855 | */ |
liubo | c622ae6 | 2011-03-26 08:01:12 -0400 | [diff] [blame] | 856 | |
| 857 | if (search_done) |
| 858 | goto insert; |
| 859 | |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 860 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
| 861 | if (ret == 0) { |
| 862 | char *victim_name; |
| 863 | int victim_name_len; |
| 864 | struct btrfs_inode_ref *victim_ref; |
| 865 | unsigned long ptr; |
| 866 | unsigned long ptr_end; |
| 867 | struct extent_buffer *leaf = path->nodes[0]; |
| 868 | |
| 869 | /* are we trying to overwrite a back ref for the root directory |
| 870 | * if so, just jump out, we're done |
| 871 | */ |
| 872 | if (key->objectid == key->offset) |
| 873 | goto out_nowrite; |
| 874 | |
| 875 | /* check all the names in this back reference to see |
| 876 | * if they are in the log. if so, we allow them to stay |
| 877 | * otherwise they must be unlinked as a conflict |
| 878 | */ |
| 879 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); |
| 880 | ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]); |
Chris Mason | d397712 | 2009-01-05 21:25:51 -0500 | [diff] [blame] | 881 | while (ptr < ptr_end) { |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 882 | victim_ref = (struct btrfs_inode_ref *)ptr; |
| 883 | victim_name_len = btrfs_inode_ref_name_len(leaf, |
| 884 | victim_ref); |
| 885 | victim_name = kmalloc(victim_name_len, GFP_NOFS); |
| 886 | BUG_ON(!victim_name); |
| 887 | |
| 888 | read_extent_buffer(leaf, victim_name, |
| 889 | (unsigned long)(victim_ref + 1), |
| 890 | victim_name_len); |
| 891 | |
| 892 | if (!backref_in_log(log, key, victim_name, |
| 893 | victim_name_len)) { |
| 894 | btrfs_inc_nlink(inode); |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 895 | btrfs_release_path(path); |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 896 | |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 897 | ret = btrfs_unlink_inode(trans, root, dir, |
| 898 | inode, victim_name, |
| 899 | victim_name_len); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 900 | } |
| 901 | kfree(victim_name); |
| 902 | ptr = (unsigned long)(victim_ref + 1) + victim_name_len; |
| 903 | } |
| 904 | BUG_ON(ret); |
liubo | c622ae6 | 2011-03-26 08:01:12 -0400 | [diff] [blame] | 905 | |
| 906 | /* |
| 907 | * NOTE: we have searched root tree and checked the |
| 908 | * coresponding ref, it does not need to check again. |
| 909 | */ |
| 910 | search_done = 1; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 911 | } |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 912 | btrfs_release_path(path); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 913 | |
liubo | 34f3e4f | 2011-08-06 08:35:23 +0000 | [diff] [blame] | 914 | /* look for a conflicting sequence number */ |
| 915 | di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir), |
| 916 | btrfs_inode_ref_index(eb, ref), |
| 917 | name, namelen, 0); |
| 918 | if (di && !IS_ERR(di)) { |
| 919 | ret = drop_one_dir_item(trans, root, path, dir, di); |
| 920 | BUG_ON(ret); |
| 921 | } |
| 922 | btrfs_release_path(path); |
| 923 | |
| 924 | /* look for a conflicing name */ |
| 925 | di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir), |
| 926 | name, namelen, 0); |
| 927 | if (di && !IS_ERR(di)) { |
| 928 | ret = drop_one_dir_item(trans, root, path, dir, di); |
| 929 | BUG_ON(ret); |
| 930 | } |
| 931 | btrfs_release_path(path); |
| 932 | |
liubo | c622ae6 | 2011-03-26 08:01:12 -0400 | [diff] [blame] | 933 | insert: |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 934 | /* insert our name */ |
| 935 | ret = btrfs_add_link(trans, dir, inode, name, namelen, 0, |
| 936 | btrfs_inode_ref_index(eb, ref)); |
| 937 | BUG_ON(ret); |
| 938 | |
| 939 | btrfs_update_inode(trans, root, inode); |
| 940 | |
| 941 | out: |
| 942 | ref_ptr = (unsigned long)(ref + 1) + namelen; |
| 943 | kfree(name); |
| 944 | if (ref_ptr < ref_end) |
| 945 | goto again; |
| 946 | |
| 947 | /* finally write the back reference in the inode */ |
| 948 | ret = overwrite_item(trans, root, path, eb, slot, key); |
| 949 | BUG_ON(ret); |
| 950 | |
| 951 | out_nowrite: |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 952 | btrfs_release_path(path); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 953 | iput(dir); |
| 954 | iput(inode); |
| 955 | return 0; |
| 956 | } |
| 957 | |
Yan, Zheng | c71bf09 | 2009-11-12 09:34:40 +0000 | [diff] [blame] | 958 | static int insert_orphan_item(struct btrfs_trans_handle *trans, |
| 959 | struct btrfs_root *root, u64 offset) |
| 960 | { |
| 961 | int ret; |
| 962 | ret = btrfs_find_orphan_item(root, offset); |
| 963 | if (ret > 0) |
| 964 | ret = btrfs_insert_orphan_item(trans, root, offset); |
| 965 | return ret; |
| 966 | } |
| 967 | |
| 968 | |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 969 | /* |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 970 | * There are a few corners where the link count of the file can't |
| 971 | * be properly maintained during replay. So, instead of adding |
| 972 | * lots of complexity to the log code, we just scan the backrefs |
| 973 | * for any file that has been through replay. |
| 974 | * |
| 975 | * The scan will update the link count on the inode to reflect the |
| 976 | * number of back refs found. If it goes down to zero, the iput |
| 977 | * will free the inode. |
| 978 | */ |
| 979 | static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans, |
| 980 | struct btrfs_root *root, |
| 981 | struct inode *inode) |
| 982 | { |
| 983 | struct btrfs_path *path; |
| 984 | int ret; |
| 985 | struct btrfs_key key; |
| 986 | u64 nlink = 0; |
| 987 | unsigned long ptr; |
| 988 | unsigned long ptr_end; |
| 989 | int name_len; |
Li Zefan | 33345d01 | 2011-04-20 10:31:50 +0800 | [diff] [blame] | 990 | u64 ino = btrfs_ino(inode); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 991 | |
Li Zefan | 33345d01 | 2011-04-20 10:31:50 +0800 | [diff] [blame] | 992 | key.objectid = ino; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 993 | key.type = BTRFS_INODE_REF_KEY; |
| 994 | key.offset = (u64)-1; |
| 995 | |
| 996 | path = btrfs_alloc_path(); |
liubo | 2a29edc | 2011-01-26 06:22:08 +0000 | [diff] [blame] | 997 | if (!path) |
| 998 | return -ENOMEM; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 999 | |
Chris Mason | d397712 | 2009-01-05 21:25:51 -0500 | [diff] [blame] | 1000 | while (1) { |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1001 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
| 1002 | if (ret < 0) |
| 1003 | break; |
| 1004 | if (ret > 0) { |
| 1005 | if (path->slots[0] == 0) |
| 1006 | break; |
| 1007 | path->slots[0]--; |
| 1008 | } |
| 1009 | btrfs_item_key_to_cpu(path->nodes[0], &key, |
| 1010 | path->slots[0]); |
Li Zefan | 33345d01 | 2011-04-20 10:31:50 +0800 | [diff] [blame] | 1011 | if (key.objectid != ino || |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1012 | key.type != BTRFS_INODE_REF_KEY) |
| 1013 | break; |
| 1014 | ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]); |
| 1015 | ptr_end = ptr + btrfs_item_size_nr(path->nodes[0], |
| 1016 | path->slots[0]); |
Chris Mason | d397712 | 2009-01-05 21:25:51 -0500 | [diff] [blame] | 1017 | while (ptr < ptr_end) { |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1018 | struct btrfs_inode_ref *ref; |
| 1019 | |
| 1020 | ref = (struct btrfs_inode_ref *)ptr; |
| 1021 | name_len = btrfs_inode_ref_name_len(path->nodes[0], |
| 1022 | ref); |
| 1023 | ptr = (unsigned long)(ref + 1) + name_len; |
| 1024 | nlink++; |
| 1025 | } |
| 1026 | |
| 1027 | if (key.offset == 0) |
| 1028 | break; |
| 1029 | key.offset--; |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 1030 | btrfs_release_path(path); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1031 | } |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 1032 | btrfs_release_path(path); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1033 | if (nlink != inode->i_nlink) { |
Miklos Szeredi | bfe8684 | 2011-10-28 14:13:29 +0200 | [diff] [blame] | 1034 | set_nlink(inode, nlink); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1035 | btrfs_update_inode(trans, root, inode); |
| 1036 | } |
Chris Mason | 8d5bf1c | 2008-09-11 15:51:21 -0400 | [diff] [blame] | 1037 | BTRFS_I(inode)->index_cnt = (u64)-1; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1038 | |
Yan, Zheng | c71bf09 | 2009-11-12 09:34:40 +0000 | [diff] [blame] | 1039 | if (inode->i_nlink == 0) { |
| 1040 | if (S_ISDIR(inode->i_mode)) { |
| 1041 | ret = replay_dir_deletes(trans, root, NULL, path, |
Li Zefan | 33345d01 | 2011-04-20 10:31:50 +0800 | [diff] [blame] | 1042 | ino, 1); |
Yan, Zheng | c71bf09 | 2009-11-12 09:34:40 +0000 | [diff] [blame] | 1043 | BUG_ON(ret); |
| 1044 | } |
Li Zefan | 33345d01 | 2011-04-20 10:31:50 +0800 | [diff] [blame] | 1045 | ret = insert_orphan_item(trans, root, ino); |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 1046 | BUG_ON(ret); |
| 1047 | } |
| 1048 | btrfs_free_path(path); |
| 1049 | |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1050 | return 0; |
| 1051 | } |
| 1052 | |
| 1053 | static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans, |
| 1054 | struct btrfs_root *root, |
| 1055 | struct btrfs_path *path) |
| 1056 | { |
| 1057 | int ret; |
| 1058 | struct btrfs_key key; |
| 1059 | struct inode *inode; |
| 1060 | |
| 1061 | key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID; |
| 1062 | key.type = BTRFS_ORPHAN_ITEM_KEY; |
| 1063 | key.offset = (u64)-1; |
Chris Mason | d397712 | 2009-01-05 21:25:51 -0500 | [diff] [blame] | 1064 | while (1) { |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1065 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
| 1066 | if (ret < 0) |
| 1067 | break; |
| 1068 | |
| 1069 | if (ret == 1) { |
| 1070 | if (path->slots[0] == 0) |
| 1071 | break; |
| 1072 | path->slots[0]--; |
| 1073 | } |
| 1074 | |
| 1075 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); |
| 1076 | if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID || |
| 1077 | key.type != BTRFS_ORPHAN_ITEM_KEY) |
| 1078 | break; |
| 1079 | |
| 1080 | ret = btrfs_del_item(trans, root, path); |
Tsutomu Itoh | 65a246c | 2011-05-19 04:37:44 +0000 | [diff] [blame] | 1081 | if (ret) |
| 1082 | goto out; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1083 | |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 1084 | btrfs_release_path(path); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1085 | inode = read_one_inode(root, key.offset); |
Tsutomu Itoh | c00e949 | 2011-04-28 09:10:23 +0000 | [diff] [blame] | 1086 | if (!inode) |
| 1087 | return -EIO; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1088 | |
| 1089 | ret = fixup_inode_link_count(trans, root, inode); |
| 1090 | BUG_ON(ret); |
| 1091 | |
| 1092 | iput(inode); |
| 1093 | |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 1094 | /* |
| 1095 | * fixup on a directory may create new entries, |
| 1096 | * make sure we always look for the highset possible |
| 1097 | * offset |
| 1098 | */ |
| 1099 | key.offset = (u64)-1; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1100 | } |
Tsutomu Itoh | 65a246c | 2011-05-19 04:37:44 +0000 | [diff] [blame] | 1101 | ret = 0; |
| 1102 | out: |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 1103 | btrfs_release_path(path); |
Tsutomu Itoh | 65a246c | 2011-05-19 04:37:44 +0000 | [diff] [blame] | 1104 | return ret; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1105 | } |
| 1106 | |
| 1107 | |
| 1108 | /* |
| 1109 | * record a given inode in the fixup dir so we can check its link |
| 1110 | * count when replay is done. The link count is incremented here |
| 1111 | * so the inode won't go away until we check it |
| 1112 | */ |
| 1113 | static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans, |
| 1114 | struct btrfs_root *root, |
| 1115 | struct btrfs_path *path, |
| 1116 | u64 objectid) |
| 1117 | { |
| 1118 | struct btrfs_key key; |
| 1119 | int ret = 0; |
| 1120 | struct inode *inode; |
| 1121 | |
| 1122 | inode = read_one_inode(root, objectid); |
Tsutomu Itoh | c00e949 | 2011-04-28 09:10:23 +0000 | [diff] [blame] | 1123 | if (!inode) |
| 1124 | return -EIO; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1125 | |
| 1126 | key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID; |
| 1127 | btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY); |
| 1128 | key.offset = objectid; |
| 1129 | |
| 1130 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); |
| 1131 | |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 1132 | btrfs_release_path(path); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1133 | if (ret == 0) { |
| 1134 | btrfs_inc_nlink(inode); |
| 1135 | btrfs_update_inode(trans, root, inode); |
| 1136 | } else if (ret == -EEXIST) { |
| 1137 | ret = 0; |
| 1138 | } else { |
| 1139 | BUG(); |
| 1140 | } |
| 1141 | iput(inode); |
| 1142 | |
| 1143 | return ret; |
| 1144 | } |
| 1145 | |
| 1146 | /* |
| 1147 | * when replaying the log for a directory, we only insert names |
| 1148 | * for inodes that actually exist. This means an fsync on a directory |
| 1149 | * does not implicitly fsync all the new files in it |
| 1150 | */ |
| 1151 | static noinline int insert_one_name(struct btrfs_trans_handle *trans, |
| 1152 | struct btrfs_root *root, |
| 1153 | struct btrfs_path *path, |
| 1154 | u64 dirid, u64 index, |
| 1155 | char *name, int name_len, u8 type, |
| 1156 | struct btrfs_key *location) |
| 1157 | { |
| 1158 | struct inode *inode; |
| 1159 | struct inode *dir; |
| 1160 | int ret; |
| 1161 | |
| 1162 | inode = read_one_inode(root, location->objectid); |
| 1163 | if (!inode) |
| 1164 | return -ENOENT; |
| 1165 | |
| 1166 | dir = read_one_inode(root, dirid); |
| 1167 | if (!dir) { |
| 1168 | iput(inode); |
| 1169 | return -EIO; |
| 1170 | } |
| 1171 | ret = btrfs_add_link(trans, dir, inode, name, name_len, 1, index); |
| 1172 | |
| 1173 | /* FIXME, put inode into FIXUP list */ |
| 1174 | |
| 1175 | iput(inode); |
| 1176 | iput(dir); |
| 1177 | return ret; |
| 1178 | } |
| 1179 | |
| 1180 | /* |
| 1181 | * take a single entry in a log directory item and replay it into |
| 1182 | * the subvolume. |
| 1183 | * |
| 1184 | * if a conflicting item exists in the subdirectory already, |
| 1185 | * the inode it points to is unlinked and put into the link count |
| 1186 | * fix up tree. |
| 1187 | * |
| 1188 | * If a name from the log points to a file or directory that does |
| 1189 | * not exist in the FS, it is skipped. fsyncs on directories |
| 1190 | * do not force down inodes inside that directory, just changes to the |
| 1191 | * names or unlinks in a directory. |
| 1192 | */ |
| 1193 | static noinline int replay_one_name(struct btrfs_trans_handle *trans, |
| 1194 | struct btrfs_root *root, |
| 1195 | struct btrfs_path *path, |
| 1196 | struct extent_buffer *eb, |
| 1197 | struct btrfs_dir_item *di, |
| 1198 | struct btrfs_key *key) |
| 1199 | { |
| 1200 | char *name; |
| 1201 | int name_len; |
| 1202 | struct btrfs_dir_item *dst_di; |
| 1203 | struct btrfs_key found_key; |
| 1204 | struct btrfs_key log_key; |
| 1205 | struct inode *dir; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1206 | u8 log_type; |
Chris Mason | 4bef084 | 2008-09-08 11:18:08 -0400 | [diff] [blame] | 1207 | int exists; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1208 | int ret; |
| 1209 | |
| 1210 | dir = read_one_inode(root, key->objectid); |
Tsutomu Itoh | c00e949 | 2011-04-28 09:10:23 +0000 | [diff] [blame] | 1211 | if (!dir) |
| 1212 | return -EIO; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1213 | |
| 1214 | name_len = btrfs_dir_name_len(eb, di); |
| 1215 | name = kmalloc(name_len, GFP_NOFS); |
liubo | 2a29edc | 2011-01-26 06:22:08 +0000 | [diff] [blame] | 1216 | if (!name) |
| 1217 | return -ENOMEM; |
| 1218 | |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1219 | log_type = btrfs_dir_type(eb, di); |
| 1220 | read_extent_buffer(eb, name, (unsigned long)(di + 1), |
| 1221 | name_len); |
| 1222 | |
| 1223 | btrfs_dir_item_key_to_cpu(eb, di, &log_key); |
Chris Mason | 4bef084 | 2008-09-08 11:18:08 -0400 | [diff] [blame] | 1224 | exists = btrfs_lookup_inode(trans, root, path, &log_key, 0); |
| 1225 | if (exists == 0) |
| 1226 | exists = 1; |
| 1227 | else |
| 1228 | exists = 0; |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 1229 | btrfs_release_path(path); |
Chris Mason | 4bef084 | 2008-09-08 11:18:08 -0400 | [diff] [blame] | 1230 | |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1231 | if (key->type == BTRFS_DIR_ITEM_KEY) { |
| 1232 | dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid, |
| 1233 | name, name_len, 1); |
Chris Mason | d397712 | 2009-01-05 21:25:51 -0500 | [diff] [blame] | 1234 | } else if (key->type == BTRFS_DIR_INDEX_KEY) { |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1235 | dst_di = btrfs_lookup_dir_index_item(trans, root, path, |
| 1236 | key->objectid, |
| 1237 | key->offset, name, |
| 1238 | name_len, 1); |
| 1239 | } else { |
| 1240 | BUG(); |
| 1241 | } |
David Sterba | c704005 | 2011-04-19 18:00:01 +0200 | [diff] [blame] | 1242 | if (IS_ERR_OR_NULL(dst_di)) { |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1243 | /* we need a sequence number to insert, so we only |
| 1244 | * do inserts for the BTRFS_DIR_INDEX_KEY types |
| 1245 | */ |
| 1246 | if (key->type != BTRFS_DIR_INDEX_KEY) |
| 1247 | goto out; |
| 1248 | goto insert; |
| 1249 | } |
| 1250 | |
| 1251 | btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key); |
| 1252 | /* the existing item matches the logged item */ |
| 1253 | if (found_key.objectid == log_key.objectid && |
| 1254 | found_key.type == log_key.type && |
| 1255 | found_key.offset == log_key.offset && |
| 1256 | btrfs_dir_type(path->nodes[0], dst_di) == log_type) { |
| 1257 | goto out; |
| 1258 | } |
| 1259 | |
| 1260 | /* |
| 1261 | * don't drop the conflicting directory entry if the inode |
| 1262 | * for the new entry doesn't exist |
| 1263 | */ |
Chris Mason | 4bef084 | 2008-09-08 11:18:08 -0400 | [diff] [blame] | 1264 | if (!exists) |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1265 | goto out; |
| 1266 | |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1267 | ret = drop_one_dir_item(trans, root, path, dir, dst_di); |
| 1268 | BUG_ON(ret); |
| 1269 | |
| 1270 | if (key->type == BTRFS_DIR_INDEX_KEY) |
| 1271 | goto insert; |
| 1272 | out: |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 1273 | btrfs_release_path(path); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1274 | kfree(name); |
| 1275 | iput(dir); |
| 1276 | return 0; |
| 1277 | |
| 1278 | insert: |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 1279 | btrfs_release_path(path); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1280 | ret = insert_one_name(trans, root, path, key->objectid, key->offset, |
| 1281 | name, name_len, log_type, &log_key); |
| 1282 | |
Stoyan Gaydarov | c293498 | 2009-04-02 17:05:11 -0400 | [diff] [blame] | 1283 | BUG_ON(ret && ret != -ENOENT); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1284 | goto out; |
| 1285 | } |
| 1286 | |
| 1287 | /* |
| 1288 | * find all the names in a directory item and reconcile them into |
| 1289 | * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than |
| 1290 | * one name in a directory item, but the same code gets used for |
| 1291 | * both directory index types |
| 1292 | */ |
| 1293 | static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans, |
| 1294 | struct btrfs_root *root, |
| 1295 | struct btrfs_path *path, |
| 1296 | struct extent_buffer *eb, int slot, |
| 1297 | struct btrfs_key *key) |
| 1298 | { |
| 1299 | int ret; |
| 1300 | u32 item_size = btrfs_item_size_nr(eb, slot); |
| 1301 | struct btrfs_dir_item *di; |
| 1302 | int name_len; |
| 1303 | unsigned long ptr; |
| 1304 | unsigned long ptr_end; |
| 1305 | |
| 1306 | ptr = btrfs_item_ptr_offset(eb, slot); |
| 1307 | ptr_end = ptr + item_size; |
Chris Mason | d397712 | 2009-01-05 21:25:51 -0500 | [diff] [blame] | 1308 | while (ptr < ptr_end) { |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1309 | di = (struct btrfs_dir_item *)ptr; |
Josef Bacik | 22a94d4 | 2011-03-16 16:47:17 -0400 | [diff] [blame] | 1310 | if (verify_dir_item(root, eb, di)) |
| 1311 | return -EIO; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1312 | name_len = btrfs_dir_name_len(eb, di); |
| 1313 | ret = replay_one_name(trans, root, path, eb, di, key); |
| 1314 | BUG_ON(ret); |
| 1315 | ptr = (unsigned long)(di + 1); |
| 1316 | ptr += name_len; |
| 1317 | } |
| 1318 | return 0; |
| 1319 | } |
| 1320 | |
| 1321 | /* |
| 1322 | * directory replay has two parts. There are the standard directory |
| 1323 | * items in the log copied from the subvolume, and range items |
| 1324 | * created in the log while the subvolume was logged. |
| 1325 | * |
| 1326 | * The range items tell us which parts of the key space the log |
| 1327 | * is authoritative for. During replay, if a key in the subvolume |
| 1328 | * directory is in a logged range item, but not actually in the log |
| 1329 | * that means it was deleted from the directory before the fsync |
| 1330 | * and should be removed. |
| 1331 | */ |
| 1332 | static noinline int find_dir_range(struct btrfs_root *root, |
| 1333 | struct btrfs_path *path, |
| 1334 | u64 dirid, int key_type, |
| 1335 | u64 *start_ret, u64 *end_ret) |
| 1336 | { |
| 1337 | struct btrfs_key key; |
| 1338 | u64 found_end; |
| 1339 | struct btrfs_dir_log_item *item; |
| 1340 | int ret; |
| 1341 | int nritems; |
| 1342 | |
| 1343 | if (*start_ret == (u64)-1) |
| 1344 | return 1; |
| 1345 | |
| 1346 | key.objectid = dirid; |
| 1347 | key.type = key_type; |
| 1348 | key.offset = *start_ret; |
| 1349 | |
| 1350 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
| 1351 | if (ret < 0) |
| 1352 | goto out; |
| 1353 | if (ret > 0) { |
| 1354 | if (path->slots[0] == 0) |
| 1355 | goto out; |
| 1356 | path->slots[0]--; |
| 1357 | } |
| 1358 | if (ret != 0) |
| 1359 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); |
| 1360 | |
| 1361 | if (key.type != key_type || key.objectid != dirid) { |
| 1362 | ret = 1; |
| 1363 | goto next; |
| 1364 | } |
| 1365 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
| 1366 | struct btrfs_dir_log_item); |
| 1367 | found_end = btrfs_dir_log_end(path->nodes[0], item); |
| 1368 | |
| 1369 | if (*start_ret >= key.offset && *start_ret <= found_end) { |
| 1370 | ret = 0; |
| 1371 | *start_ret = key.offset; |
| 1372 | *end_ret = found_end; |
| 1373 | goto out; |
| 1374 | } |
| 1375 | ret = 1; |
| 1376 | next: |
| 1377 | /* check the next slot in the tree to see if it is a valid item */ |
| 1378 | nritems = btrfs_header_nritems(path->nodes[0]); |
| 1379 | if (path->slots[0] >= nritems) { |
| 1380 | ret = btrfs_next_leaf(root, path); |
| 1381 | if (ret) |
| 1382 | goto out; |
| 1383 | } else { |
| 1384 | path->slots[0]++; |
| 1385 | } |
| 1386 | |
| 1387 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); |
| 1388 | |
| 1389 | if (key.type != key_type || key.objectid != dirid) { |
| 1390 | ret = 1; |
| 1391 | goto out; |
| 1392 | } |
| 1393 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
| 1394 | struct btrfs_dir_log_item); |
| 1395 | found_end = btrfs_dir_log_end(path->nodes[0], item); |
| 1396 | *start_ret = key.offset; |
| 1397 | *end_ret = found_end; |
| 1398 | ret = 0; |
| 1399 | out: |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 1400 | btrfs_release_path(path); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1401 | return ret; |
| 1402 | } |
| 1403 | |
| 1404 | /* |
| 1405 | * this looks for a given directory item in the log. If the directory |
| 1406 | * item is not in the log, the item is removed and the inode it points |
| 1407 | * to is unlinked |
| 1408 | */ |
| 1409 | static noinline int check_item_in_log(struct btrfs_trans_handle *trans, |
| 1410 | struct btrfs_root *root, |
| 1411 | struct btrfs_root *log, |
| 1412 | struct btrfs_path *path, |
| 1413 | struct btrfs_path *log_path, |
| 1414 | struct inode *dir, |
| 1415 | struct btrfs_key *dir_key) |
| 1416 | { |
| 1417 | int ret; |
| 1418 | struct extent_buffer *eb; |
| 1419 | int slot; |
| 1420 | u32 item_size; |
| 1421 | struct btrfs_dir_item *di; |
| 1422 | struct btrfs_dir_item *log_di; |
| 1423 | int name_len; |
| 1424 | unsigned long ptr; |
| 1425 | unsigned long ptr_end; |
| 1426 | char *name; |
| 1427 | struct inode *inode; |
| 1428 | struct btrfs_key location; |
| 1429 | |
| 1430 | again: |
| 1431 | eb = path->nodes[0]; |
| 1432 | slot = path->slots[0]; |
| 1433 | item_size = btrfs_item_size_nr(eb, slot); |
| 1434 | ptr = btrfs_item_ptr_offset(eb, slot); |
| 1435 | ptr_end = ptr + item_size; |
Chris Mason | d397712 | 2009-01-05 21:25:51 -0500 | [diff] [blame] | 1436 | while (ptr < ptr_end) { |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1437 | di = (struct btrfs_dir_item *)ptr; |
Josef Bacik | 22a94d4 | 2011-03-16 16:47:17 -0400 | [diff] [blame] | 1438 | if (verify_dir_item(root, eb, di)) { |
| 1439 | ret = -EIO; |
| 1440 | goto out; |
| 1441 | } |
| 1442 | |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1443 | name_len = btrfs_dir_name_len(eb, di); |
| 1444 | name = kmalloc(name_len, GFP_NOFS); |
| 1445 | if (!name) { |
| 1446 | ret = -ENOMEM; |
| 1447 | goto out; |
| 1448 | } |
| 1449 | read_extent_buffer(eb, name, (unsigned long)(di + 1), |
| 1450 | name_len); |
| 1451 | log_di = NULL; |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 1452 | if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) { |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1453 | log_di = btrfs_lookup_dir_item(trans, log, log_path, |
| 1454 | dir_key->objectid, |
| 1455 | name, name_len, 0); |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 1456 | } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) { |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1457 | log_di = btrfs_lookup_dir_index_item(trans, log, |
| 1458 | log_path, |
| 1459 | dir_key->objectid, |
| 1460 | dir_key->offset, |
| 1461 | name, name_len, 0); |
| 1462 | } |
David Sterba | c704005 | 2011-04-19 18:00:01 +0200 | [diff] [blame] | 1463 | if (IS_ERR_OR_NULL(log_di)) { |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1464 | btrfs_dir_item_key_to_cpu(eb, di, &location); |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 1465 | btrfs_release_path(path); |
| 1466 | btrfs_release_path(log_path); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1467 | inode = read_one_inode(root, location.objectid); |
Tsutomu Itoh | c00e949 | 2011-04-28 09:10:23 +0000 | [diff] [blame] | 1468 | if (!inode) { |
| 1469 | kfree(name); |
| 1470 | return -EIO; |
| 1471 | } |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1472 | |
| 1473 | ret = link_to_fixup_dir(trans, root, |
| 1474 | path, location.objectid); |
| 1475 | BUG_ON(ret); |
| 1476 | btrfs_inc_nlink(inode); |
| 1477 | ret = btrfs_unlink_inode(trans, root, dir, inode, |
| 1478 | name, name_len); |
| 1479 | BUG_ON(ret); |
| 1480 | kfree(name); |
| 1481 | iput(inode); |
| 1482 | |
| 1483 | /* there might still be more names under this key |
| 1484 | * check and repeat if required |
| 1485 | */ |
| 1486 | ret = btrfs_search_slot(NULL, root, dir_key, path, |
| 1487 | 0, 0); |
| 1488 | if (ret == 0) |
| 1489 | goto again; |
| 1490 | ret = 0; |
| 1491 | goto out; |
| 1492 | } |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 1493 | btrfs_release_path(log_path); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1494 | kfree(name); |
| 1495 | |
| 1496 | ptr = (unsigned long)(di + 1); |
| 1497 | ptr += name_len; |
| 1498 | } |
| 1499 | ret = 0; |
| 1500 | out: |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 1501 | btrfs_release_path(path); |
| 1502 | btrfs_release_path(log_path); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1503 | return ret; |
| 1504 | } |
| 1505 | |
| 1506 | /* |
| 1507 | * deletion replay happens before we copy any new directory items |
| 1508 | * out of the log or out of backreferences from inodes. It |
| 1509 | * scans the log to find ranges of keys that log is authoritative for, |
| 1510 | * and then scans the directory to find items in those ranges that are |
| 1511 | * not present in the log. |
| 1512 | * |
| 1513 | * Anything we don't find in the log is unlinked and removed from the |
| 1514 | * directory. |
| 1515 | */ |
| 1516 | static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans, |
| 1517 | struct btrfs_root *root, |
| 1518 | struct btrfs_root *log, |
| 1519 | struct btrfs_path *path, |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 1520 | u64 dirid, int del_all) |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1521 | { |
| 1522 | u64 range_start; |
| 1523 | u64 range_end; |
| 1524 | int key_type = BTRFS_DIR_LOG_ITEM_KEY; |
| 1525 | int ret = 0; |
| 1526 | struct btrfs_key dir_key; |
| 1527 | struct btrfs_key found_key; |
| 1528 | struct btrfs_path *log_path; |
| 1529 | struct inode *dir; |
| 1530 | |
| 1531 | dir_key.objectid = dirid; |
| 1532 | dir_key.type = BTRFS_DIR_ITEM_KEY; |
| 1533 | log_path = btrfs_alloc_path(); |
| 1534 | if (!log_path) |
| 1535 | return -ENOMEM; |
| 1536 | |
| 1537 | dir = read_one_inode(root, dirid); |
| 1538 | /* it isn't an error if the inode isn't there, that can happen |
| 1539 | * because we replay the deletes before we copy in the inode item |
| 1540 | * from the log |
| 1541 | */ |
| 1542 | if (!dir) { |
| 1543 | btrfs_free_path(log_path); |
| 1544 | return 0; |
| 1545 | } |
| 1546 | again: |
| 1547 | range_start = 0; |
| 1548 | range_end = 0; |
Chris Mason | d397712 | 2009-01-05 21:25:51 -0500 | [diff] [blame] | 1549 | while (1) { |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 1550 | if (del_all) |
| 1551 | range_end = (u64)-1; |
| 1552 | else { |
| 1553 | ret = find_dir_range(log, path, dirid, key_type, |
| 1554 | &range_start, &range_end); |
| 1555 | if (ret != 0) |
| 1556 | break; |
| 1557 | } |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1558 | |
| 1559 | dir_key.offset = range_start; |
Chris Mason | d397712 | 2009-01-05 21:25:51 -0500 | [diff] [blame] | 1560 | while (1) { |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1561 | int nritems; |
| 1562 | ret = btrfs_search_slot(NULL, root, &dir_key, path, |
| 1563 | 0, 0); |
| 1564 | if (ret < 0) |
| 1565 | goto out; |
| 1566 | |
| 1567 | nritems = btrfs_header_nritems(path->nodes[0]); |
| 1568 | if (path->slots[0] >= nritems) { |
| 1569 | ret = btrfs_next_leaf(root, path); |
| 1570 | if (ret) |
| 1571 | break; |
| 1572 | } |
| 1573 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, |
| 1574 | path->slots[0]); |
| 1575 | if (found_key.objectid != dirid || |
| 1576 | found_key.type != dir_key.type) |
| 1577 | goto next_type; |
| 1578 | |
| 1579 | if (found_key.offset > range_end) |
| 1580 | break; |
| 1581 | |
| 1582 | ret = check_item_in_log(trans, root, log, path, |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 1583 | log_path, dir, |
| 1584 | &found_key); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1585 | BUG_ON(ret); |
| 1586 | if (found_key.offset == (u64)-1) |
| 1587 | break; |
| 1588 | dir_key.offset = found_key.offset + 1; |
| 1589 | } |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 1590 | btrfs_release_path(path); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1591 | if (range_end == (u64)-1) |
| 1592 | break; |
| 1593 | range_start = range_end + 1; |
| 1594 | } |
| 1595 | |
| 1596 | next_type: |
| 1597 | ret = 0; |
| 1598 | if (key_type == BTRFS_DIR_LOG_ITEM_KEY) { |
| 1599 | key_type = BTRFS_DIR_LOG_INDEX_KEY; |
| 1600 | dir_key.type = BTRFS_DIR_INDEX_KEY; |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 1601 | btrfs_release_path(path); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1602 | goto again; |
| 1603 | } |
| 1604 | out: |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 1605 | btrfs_release_path(path); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1606 | btrfs_free_path(log_path); |
| 1607 | iput(dir); |
| 1608 | return ret; |
| 1609 | } |
| 1610 | |
| 1611 | /* |
| 1612 | * the process_func used to replay items from the log tree. This |
| 1613 | * gets called in two different stages. The first stage just looks |
| 1614 | * for inodes and makes sure they are all copied into the subvolume. |
| 1615 | * |
| 1616 | * The second stage copies all the other item types from the log into |
| 1617 | * the subvolume. The two stage approach is slower, but gets rid of |
| 1618 | * lots of complexity around inodes referencing other inodes that exist |
| 1619 | * only in the log (references come from either directory items or inode |
| 1620 | * back refs). |
| 1621 | */ |
| 1622 | static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb, |
| 1623 | struct walk_control *wc, u64 gen) |
| 1624 | { |
| 1625 | int nritems; |
| 1626 | struct btrfs_path *path; |
| 1627 | struct btrfs_root *root = wc->replay_dest; |
| 1628 | struct btrfs_key key; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1629 | int level; |
| 1630 | int i; |
| 1631 | int ret; |
| 1632 | |
| 1633 | btrfs_read_buffer(eb, gen); |
| 1634 | |
| 1635 | level = btrfs_header_level(eb); |
| 1636 | |
| 1637 | if (level != 0) |
| 1638 | return 0; |
| 1639 | |
| 1640 | path = btrfs_alloc_path(); |
Mark Fasheh | 1e5063d | 2011-07-12 10:46:06 -0700 | [diff] [blame] | 1641 | if (!path) |
| 1642 | return -ENOMEM; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1643 | |
| 1644 | nritems = btrfs_header_nritems(eb); |
| 1645 | for (i = 0; i < nritems; i++) { |
| 1646 | btrfs_item_key_to_cpu(eb, &key, i); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1647 | |
| 1648 | /* inode keys are done during the first stage */ |
| 1649 | if (key.type == BTRFS_INODE_ITEM_KEY && |
| 1650 | wc->stage == LOG_WALK_REPLAY_INODES) { |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1651 | struct btrfs_inode_item *inode_item; |
| 1652 | u32 mode; |
| 1653 | |
| 1654 | inode_item = btrfs_item_ptr(eb, i, |
| 1655 | struct btrfs_inode_item); |
| 1656 | mode = btrfs_inode_mode(eb, inode_item); |
| 1657 | if (S_ISDIR(mode)) { |
| 1658 | ret = replay_dir_deletes(wc->trans, |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 1659 | root, log, path, key.objectid, 0); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1660 | BUG_ON(ret); |
| 1661 | } |
| 1662 | ret = overwrite_item(wc->trans, root, path, |
| 1663 | eb, i, &key); |
| 1664 | BUG_ON(ret); |
| 1665 | |
Yan, Zheng | c71bf09 | 2009-11-12 09:34:40 +0000 | [diff] [blame] | 1666 | /* for regular files, make sure corresponding |
| 1667 | * orhpan item exist. extents past the new EOF |
| 1668 | * will be truncated later by orphan cleanup. |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1669 | */ |
| 1670 | if (S_ISREG(mode)) { |
Yan, Zheng | c71bf09 | 2009-11-12 09:34:40 +0000 | [diff] [blame] | 1671 | ret = insert_orphan_item(wc->trans, root, |
| 1672 | key.objectid); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1673 | BUG_ON(ret); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1674 | } |
Yan, Zheng | c71bf09 | 2009-11-12 09:34:40 +0000 | [diff] [blame] | 1675 | |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1676 | ret = link_to_fixup_dir(wc->trans, root, |
| 1677 | path, key.objectid); |
| 1678 | BUG_ON(ret); |
| 1679 | } |
| 1680 | if (wc->stage < LOG_WALK_REPLAY_ALL) |
| 1681 | continue; |
| 1682 | |
| 1683 | /* these keys are simply copied */ |
| 1684 | if (key.type == BTRFS_XATTR_ITEM_KEY) { |
| 1685 | ret = overwrite_item(wc->trans, root, path, |
| 1686 | eb, i, &key); |
| 1687 | BUG_ON(ret); |
| 1688 | } else if (key.type == BTRFS_INODE_REF_KEY) { |
| 1689 | ret = add_inode_ref(wc->trans, root, log, path, |
| 1690 | eb, i, &key); |
| 1691 | BUG_ON(ret && ret != -ENOENT); |
| 1692 | } else if (key.type == BTRFS_EXTENT_DATA_KEY) { |
| 1693 | ret = replay_one_extent(wc->trans, root, path, |
| 1694 | eb, i, &key); |
| 1695 | BUG_ON(ret); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1696 | } else if (key.type == BTRFS_DIR_ITEM_KEY || |
| 1697 | key.type == BTRFS_DIR_INDEX_KEY) { |
| 1698 | ret = replay_one_dir_item(wc->trans, root, path, |
| 1699 | eb, i, &key); |
| 1700 | BUG_ON(ret); |
| 1701 | } |
| 1702 | } |
| 1703 | btrfs_free_path(path); |
| 1704 | return 0; |
| 1705 | } |
| 1706 | |
Chris Mason | d397712 | 2009-01-05 21:25:51 -0500 | [diff] [blame] | 1707 | static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans, |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1708 | struct btrfs_root *root, |
| 1709 | struct btrfs_path *path, int *level, |
| 1710 | struct walk_control *wc) |
| 1711 | { |
| 1712 | u64 root_owner; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1713 | u64 bytenr; |
| 1714 | u64 ptr_gen; |
| 1715 | struct extent_buffer *next; |
| 1716 | struct extent_buffer *cur; |
| 1717 | struct extent_buffer *parent; |
| 1718 | u32 blocksize; |
| 1719 | int ret = 0; |
| 1720 | |
| 1721 | WARN_ON(*level < 0); |
| 1722 | WARN_ON(*level >= BTRFS_MAX_LEVEL); |
| 1723 | |
Chris Mason | d397712 | 2009-01-05 21:25:51 -0500 | [diff] [blame] | 1724 | while (*level > 0) { |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1725 | WARN_ON(*level < 0); |
| 1726 | WARN_ON(*level >= BTRFS_MAX_LEVEL); |
| 1727 | cur = path->nodes[*level]; |
| 1728 | |
| 1729 | if (btrfs_header_level(cur) != *level) |
| 1730 | WARN_ON(1); |
| 1731 | |
| 1732 | if (path->slots[*level] >= |
| 1733 | btrfs_header_nritems(cur)) |
| 1734 | break; |
| 1735 | |
| 1736 | bytenr = btrfs_node_blockptr(cur, path->slots[*level]); |
| 1737 | ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]); |
| 1738 | blocksize = btrfs_level_size(root, *level - 1); |
| 1739 | |
| 1740 | parent = path->nodes[*level]; |
| 1741 | root_owner = btrfs_header_owner(parent); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1742 | |
| 1743 | next = btrfs_find_create_tree_block(root, bytenr, blocksize); |
liubo | 2a29edc | 2011-01-26 06:22:08 +0000 | [diff] [blame] | 1744 | if (!next) |
| 1745 | return -ENOMEM; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1746 | |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1747 | if (*level == 1) { |
Mark Fasheh | 1e5063d | 2011-07-12 10:46:06 -0700 | [diff] [blame] | 1748 | ret = wc->process_func(root, next, wc, ptr_gen); |
| 1749 | if (ret) |
| 1750 | return ret; |
Yan, Zheng | 4a500fd | 2010-05-16 10:49:59 -0400 | [diff] [blame] | 1751 | |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1752 | path->slots[*level]++; |
| 1753 | if (wc->free) { |
| 1754 | btrfs_read_buffer(next, ptr_gen); |
| 1755 | |
| 1756 | btrfs_tree_lock(next); |
Chris Mason | b4ce94d | 2009-02-04 09:25:08 -0500 | [diff] [blame] | 1757 | btrfs_set_lock_blocking(next); |
Chris Mason | bd68151 | 2011-07-16 15:23:14 -0400 | [diff] [blame] | 1758 | clean_tree_block(trans, root, next); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1759 | btrfs_wait_tree_block_writeback(next); |
| 1760 | btrfs_tree_unlock(next); |
| 1761 | |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1762 | WARN_ON(root_owner != |
| 1763 | BTRFS_TREE_LOG_OBJECTID); |
Chris Mason | e688b72 | 2011-10-31 20:52:39 -0400 | [diff] [blame] | 1764 | ret = btrfs_free_and_pin_reserved_extent(root, |
Chris Mason | d00aff0 | 2008-09-11 15:54:42 -0400 | [diff] [blame] | 1765 | bytenr, blocksize); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1766 | BUG_ON(ret); |
| 1767 | } |
| 1768 | free_extent_buffer(next); |
| 1769 | continue; |
| 1770 | } |
| 1771 | btrfs_read_buffer(next, ptr_gen); |
| 1772 | |
| 1773 | WARN_ON(*level <= 0); |
| 1774 | if (path->nodes[*level-1]) |
| 1775 | free_extent_buffer(path->nodes[*level-1]); |
| 1776 | path->nodes[*level-1] = next; |
| 1777 | *level = btrfs_header_level(next); |
| 1778 | path->slots[*level] = 0; |
| 1779 | cond_resched(); |
| 1780 | } |
| 1781 | WARN_ON(*level < 0); |
| 1782 | WARN_ON(*level >= BTRFS_MAX_LEVEL); |
| 1783 | |
Yan, Zheng | 4a500fd | 2010-05-16 10:49:59 -0400 | [diff] [blame] | 1784 | path->slots[*level] = btrfs_header_nritems(path->nodes[*level]); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1785 | |
| 1786 | cond_resched(); |
| 1787 | return 0; |
| 1788 | } |
| 1789 | |
Chris Mason | d397712 | 2009-01-05 21:25:51 -0500 | [diff] [blame] | 1790 | static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans, |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1791 | struct btrfs_root *root, |
| 1792 | struct btrfs_path *path, int *level, |
| 1793 | struct walk_control *wc) |
| 1794 | { |
| 1795 | u64 root_owner; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1796 | int i; |
| 1797 | int slot; |
| 1798 | int ret; |
| 1799 | |
Chris Mason | d397712 | 2009-01-05 21:25:51 -0500 | [diff] [blame] | 1800 | for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) { |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1801 | slot = path->slots[i]; |
Yan, Zheng | 4a500fd | 2010-05-16 10:49:59 -0400 | [diff] [blame] | 1802 | if (slot + 1 < btrfs_header_nritems(path->nodes[i])) { |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1803 | path->slots[i]++; |
| 1804 | *level = i; |
| 1805 | WARN_ON(*level == 0); |
| 1806 | return 0; |
| 1807 | } else { |
Zheng Yan | 31840ae | 2008-09-23 13:14:14 -0400 | [diff] [blame] | 1808 | struct extent_buffer *parent; |
| 1809 | if (path->nodes[*level] == root->node) |
| 1810 | parent = path->nodes[*level]; |
| 1811 | else |
| 1812 | parent = path->nodes[*level + 1]; |
| 1813 | |
| 1814 | root_owner = btrfs_header_owner(parent); |
Mark Fasheh | 1e5063d | 2011-07-12 10:46:06 -0700 | [diff] [blame] | 1815 | ret = wc->process_func(root, path->nodes[*level], wc, |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1816 | btrfs_header_generation(path->nodes[*level])); |
Mark Fasheh | 1e5063d | 2011-07-12 10:46:06 -0700 | [diff] [blame] | 1817 | if (ret) |
| 1818 | return ret; |
| 1819 | |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1820 | if (wc->free) { |
| 1821 | struct extent_buffer *next; |
| 1822 | |
| 1823 | next = path->nodes[*level]; |
| 1824 | |
| 1825 | btrfs_tree_lock(next); |
Chris Mason | b4ce94d | 2009-02-04 09:25:08 -0500 | [diff] [blame] | 1826 | btrfs_set_lock_blocking(next); |
Chris Mason | bd68151 | 2011-07-16 15:23:14 -0400 | [diff] [blame] | 1827 | clean_tree_block(trans, root, next); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1828 | btrfs_wait_tree_block_writeback(next); |
| 1829 | btrfs_tree_unlock(next); |
| 1830 | |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1831 | WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID); |
Chris Mason | e688b72 | 2011-10-31 20:52:39 -0400 | [diff] [blame] | 1832 | ret = btrfs_free_and_pin_reserved_extent(root, |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1833 | path->nodes[*level]->start, |
Chris Mason | d00aff0 | 2008-09-11 15:54:42 -0400 | [diff] [blame] | 1834 | path->nodes[*level]->len); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1835 | BUG_ON(ret); |
| 1836 | } |
| 1837 | free_extent_buffer(path->nodes[*level]); |
| 1838 | path->nodes[*level] = NULL; |
| 1839 | *level = i + 1; |
| 1840 | } |
| 1841 | } |
| 1842 | return 1; |
| 1843 | } |
| 1844 | |
| 1845 | /* |
| 1846 | * drop the reference count on the tree rooted at 'snap'. This traverses |
| 1847 | * the tree freeing any blocks that have a ref count of zero after being |
| 1848 | * decremented. |
| 1849 | */ |
| 1850 | static int walk_log_tree(struct btrfs_trans_handle *trans, |
| 1851 | struct btrfs_root *log, struct walk_control *wc) |
| 1852 | { |
| 1853 | int ret = 0; |
| 1854 | int wret; |
| 1855 | int level; |
| 1856 | struct btrfs_path *path; |
| 1857 | int i; |
| 1858 | int orig_level; |
| 1859 | |
| 1860 | path = btrfs_alloc_path(); |
Tsutomu Itoh | db5b493 | 2011-03-23 08:14:16 +0000 | [diff] [blame] | 1861 | if (!path) |
| 1862 | return -ENOMEM; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1863 | |
| 1864 | level = btrfs_header_level(log->node); |
| 1865 | orig_level = level; |
| 1866 | path->nodes[level] = log->node; |
| 1867 | extent_buffer_get(log->node); |
| 1868 | path->slots[level] = 0; |
| 1869 | |
Chris Mason | d397712 | 2009-01-05 21:25:51 -0500 | [diff] [blame] | 1870 | while (1) { |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1871 | wret = walk_down_log_tree(trans, log, path, &level, wc); |
| 1872 | if (wret > 0) |
| 1873 | break; |
| 1874 | if (wret < 0) |
| 1875 | ret = wret; |
| 1876 | |
| 1877 | wret = walk_up_log_tree(trans, log, path, &level, wc); |
| 1878 | if (wret > 0) |
| 1879 | break; |
| 1880 | if (wret < 0) |
| 1881 | ret = wret; |
| 1882 | } |
| 1883 | |
| 1884 | /* was the root node processed? if not, catch it here */ |
| 1885 | if (path->nodes[orig_level]) { |
| 1886 | wc->process_func(log, path->nodes[orig_level], wc, |
| 1887 | btrfs_header_generation(path->nodes[orig_level])); |
| 1888 | if (wc->free) { |
| 1889 | struct extent_buffer *next; |
| 1890 | |
| 1891 | next = path->nodes[orig_level]; |
| 1892 | |
| 1893 | btrfs_tree_lock(next); |
Chris Mason | b4ce94d | 2009-02-04 09:25:08 -0500 | [diff] [blame] | 1894 | btrfs_set_lock_blocking(next); |
Chris Mason | bd68151 | 2011-07-16 15:23:14 -0400 | [diff] [blame] | 1895 | clean_tree_block(trans, log, next); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1896 | btrfs_wait_tree_block_writeback(next); |
| 1897 | btrfs_tree_unlock(next); |
| 1898 | |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1899 | WARN_ON(log->root_key.objectid != |
| 1900 | BTRFS_TREE_LOG_OBJECTID); |
Chris Mason | e688b72 | 2011-10-31 20:52:39 -0400 | [diff] [blame] | 1901 | ret = btrfs_free_and_pin_reserved_extent(log, next->start, |
Chris Mason | d00aff0 | 2008-09-11 15:54:42 -0400 | [diff] [blame] | 1902 | next->len); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1903 | BUG_ON(ret); |
| 1904 | } |
| 1905 | } |
| 1906 | |
| 1907 | for (i = 0; i <= orig_level; i++) { |
| 1908 | if (path->nodes[i]) { |
| 1909 | free_extent_buffer(path->nodes[i]); |
| 1910 | path->nodes[i] = NULL; |
| 1911 | } |
| 1912 | } |
| 1913 | btrfs_free_path(path); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1914 | return ret; |
| 1915 | } |
| 1916 | |
Yan Zheng | 7237f18 | 2009-01-21 12:54:03 -0500 | [diff] [blame] | 1917 | /* |
| 1918 | * helper function to update the item for a given subvolumes log root |
| 1919 | * in the tree of log roots |
| 1920 | */ |
| 1921 | static int update_log_root(struct btrfs_trans_handle *trans, |
| 1922 | struct btrfs_root *log) |
| 1923 | { |
| 1924 | int ret; |
| 1925 | |
| 1926 | if (log->log_transid == 1) { |
| 1927 | /* insert root item on the first sync */ |
| 1928 | ret = btrfs_insert_root(trans, log->fs_info->log_root_tree, |
| 1929 | &log->root_key, &log->root_item); |
| 1930 | } else { |
| 1931 | ret = btrfs_update_root(trans, log->fs_info->log_root_tree, |
| 1932 | &log->root_key, &log->root_item); |
| 1933 | } |
| 1934 | return ret; |
| 1935 | } |
| 1936 | |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 1937 | static int wait_log_commit(struct btrfs_trans_handle *trans, |
| 1938 | struct btrfs_root *root, unsigned long transid) |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1939 | { |
| 1940 | DEFINE_WAIT(wait); |
Yan Zheng | 7237f18 | 2009-01-21 12:54:03 -0500 | [diff] [blame] | 1941 | int index = transid % 2; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1942 | |
Yan Zheng | 7237f18 | 2009-01-21 12:54:03 -0500 | [diff] [blame] | 1943 | /* |
| 1944 | * we only allow two pending log transactions at a time, |
| 1945 | * so we know that if ours is more than 2 older than the |
| 1946 | * current transaction, we're done |
| 1947 | */ |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1948 | do { |
Yan Zheng | 7237f18 | 2009-01-21 12:54:03 -0500 | [diff] [blame] | 1949 | prepare_to_wait(&root->log_commit_wait[index], |
| 1950 | &wait, TASK_UNINTERRUPTIBLE); |
| 1951 | mutex_unlock(&root->log_mutex); |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 1952 | |
| 1953 | if (root->fs_info->last_trans_log_full_commit != |
| 1954 | trans->transid && root->log_transid < transid + 2 && |
Yan Zheng | 7237f18 | 2009-01-21 12:54:03 -0500 | [diff] [blame] | 1955 | atomic_read(&root->log_commit[index])) |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1956 | schedule(); |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 1957 | |
Yan Zheng | 7237f18 | 2009-01-21 12:54:03 -0500 | [diff] [blame] | 1958 | finish_wait(&root->log_commit_wait[index], &wait); |
| 1959 | mutex_lock(&root->log_mutex); |
Jan Kara | 6dd70ce | 2012-01-26 15:01:11 -0500 | [diff] [blame] | 1960 | } while (root->fs_info->last_trans_log_full_commit != |
| 1961 | trans->transid && root->log_transid < transid + 2 && |
Yan Zheng | 7237f18 | 2009-01-21 12:54:03 -0500 | [diff] [blame] | 1962 | atomic_read(&root->log_commit[index])); |
| 1963 | return 0; |
| 1964 | } |
| 1965 | |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 1966 | static int wait_for_writer(struct btrfs_trans_handle *trans, |
| 1967 | struct btrfs_root *root) |
Yan Zheng | 7237f18 | 2009-01-21 12:54:03 -0500 | [diff] [blame] | 1968 | { |
| 1969 | DEFINE_WAIT(wait); |
Jan Kara | 6dd70ce | 2012-01-26 15:01:11 -0500 | [diff] [blame] | 1970 | while (root->fs_info->last_trans_log_full_commit != |
| 1971 | trans->transid && atomic_read(&root->log_writers)) { |
Yan Zheng | 7237f18 | 2009-01-21 12:54:03 -0500 | [diff] [blame] | 1972 | prepare_to_wait(&root->log_writer_wait, |
| 1973 | &wait, TASK_UNINTERRUPTIBLE); |
| 1974 | mutex_unlock(&root->log_mutex); |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 1975 | if (root->fs_info->last_trans_log_full_commit != |
| 1976 | trans->transid && atomic_read(&root->log_writers)) |
Yan Zheng | 7237f18 | 2009-01-21 12:54:03 -0500 | [diff] [blame] | 1977 | schedule(); |
| 1978 | mutex_lock(&root->log_mutex); |
| 1979 | finish_wait(&root->log_writer_wait, &wait); |
| 1980 | } |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1981 | return 0; |
| 1982 | } |
| 1983 | |
| 1984 | /* |
| 1985 | * btrfs_sync_log does sends a given tree log down to the disk and |
| 1986 | * updates the super blocks to record it. When this call is done, |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 1987 | * you know that any inodes previously logged are safely on disk only |
| 1988 | * if it returns 0. |
| 1989 | * |
| 1990 | * Any other return value means you need to call btrfs_commit_transaction. |
| 1991 | * Some of the edge cases for fsyncing directories that have had unlinks |
| 1992 | * or renames done in the past mean that sometimes the only safe |
| 1993 | * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN, |
| 1994 | * that has happened. |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 1995 | */ |
| 1996 | int btrfs_sync_log(struct btrfs_trans_handle *trans, |
| 1997 | struct btrfs_root *root) |
| 1998 | { |
Yan Zheng | 7237f18 | 2009-01-21 12:54:03 -0500 | [diff] [blame] | 1999 | int index1; |
| 2000 | int index2; |
Yan, Zheng | 8cef4e1 | 2009-11-12 09:33:26 +0000 | [diff] [blame] | 2001 | int mark; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2002 | int ret; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2003 | struct btrfs_root *log = root->log_root; |
Yan Zheng | 7237f18 | 2009-01-21 12:54:03 -0500 | [diff] [blame] | 2004 | struct btrfs_root *log_root_tree = root->fs_info->log_root_tree; |
Yan, Zheng | 8cef4e1 | 2009-11-12 09:33:26 +0000 | [diff] [blame] | 2005 | unsigned long log_transid = 0; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2006 | |
Yan Zheng | 7237f18 | 2009-01-21 12:54:03 -0500 | [diff] [blame] | 2007 | mutex_lock(&root->log_mutex); |
| 2008 | index1 = root->log_transid % 2; |
| 2009 | if (atomic_read(&root->log_commit[index1])) { |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 2010 | wait_log_commit(trans, root, root->log_transid); |
Yan Zheng | 7237f18 | 2009-01-21 12:54:03 -0500 | [diff] [blame] | 2011 | mutex_unlock(&root->log_mutex); |
| 2012 | return 0; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2013 | } |
Yan Zheng | 7237f18 | 2009-01-21 12:54:03 -0500 | [diff] [blame] | 2014 | atomic_set(&root->log_commit[index1], 1); |
| 2015 | |
| 2016 | /* wait for previous tree log sync to complete */ |
| 2017 | if (atomic_read(&root->log_commit[(index1 + 1) % 2])) |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 2018 | wait_log_commit(trans, root, root->log_transid - 1); |
Yan, Zheng | 86df7eb | 2009-10-14 09:24:59 -0400 | [diff] [blame] | 2019 | while (1) { |
Yan Zheng | 7237f18 | 2009-01-21 12:54:03 -0500 | [diff] [blame] | 2020 | unsigned long batch = root->log_batch; |
Chris Mason | cd354ad | 2011-10-20 15:45:37 -0400 | [diff] [blame] | 2021 | /* when we're on an ssd, just kick the log commit out */ |
| 2022 | if (!btrfs_test_opt(root, SSD) && root->log_multiple_pids) { |
Yan, Zheng | 86df7eb | 2009-10-14 09:24:59 -0400 | [diff] [blame] | 2023 | mutex_unlock(&root->log_mutex); |
| 2024 | schedule_timeout_uninterruptible(1); |
| 2025 | mutex_lock(&root->log_mutex); |
| 2026 | } |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 2027 | wait_for_writer(trans, root); |
Yan Zheng | 7237f18 | 2009-01-21 12:54:03 -0500 | [diff] [blame] | 2028 | if (batch == root->log_batch) |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2029 | break; |
| 2030 | } |
Chris Mason | d0c803c | 2008-09-11 16:17:57 -0400 | [diff] [blame] | 2031 | |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 2032 | /* bail out if we need to do a full commit */ |
| 2033 | if (root->fs_info->last_trans_log_full_commit == trans->transid) { |
| 2034 | ret = -EAGAIN; |
| 2035 | mutex_unlock(&root->log_mutex); |
| 2036 | goto out; |
| 2037 | } |
| 2038 | |
Yan, Zheng | 8cef4e1 | 2009-11-12 09:33:26 +0000 | [diff] [blame] | 2039 | log_transid = root->log_transid; |
| 2040 | if (log_transid % 2 == 0) |
| 2041 | mark = EXTENT_DIRTY; |
| 2042 | else |
| 2043 | mark = EXTENT_NEW; |
| 2044 | |
Chris Mason | 690587d | 2009-10-13 13:29:19 -0400 | [diff] [blame] | 2045 | /* we start IO on all the marked extents here, but we don't actually |
| 2046 | * wait for them until later. |
| 2047 | */ |
Yan, Zheng | 8cef4e1 | 2009-11-12 09:33:26 +0000 | [diff] [blame] | 2048 | ret = btrfs_write_marked_extents(log, &log->dirty_log_pages, mark); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2049 | BUG_ON(ret); |
Yan Zheng | 7237f18 | 2009-01-21 12:54:03 -0500 | [diff] [blame] | 2050 | |
Yan Zheng | 5d4f98a | 2009-06-10 10:45:14 -0400 | [diff] [blame] | 2051 | btrfs_set_root_node(&log->root_item, log->node); |
Yan Zheng | 7237f18 | 2009-01-21 12:54:03 -0500 | [diff] [blame] | 2052 | |
| 2053 | root->log_batch = 0; |
| 2054 | root->log_transid++; |
| 2055 | log->log_transid = root->log_transid; |
Josef Bacik | ff782e0 | 2009-10-08 15:30:04 -0400 | [diff] [blame] | 2056 | root->log_start_pid = 0; |
Yan Zheng | 7237f18 | 2009-01-21 12:54:03 -0500 | [diff] [blame] | 2057 | smp_mb(); |
| 2058 | /* |
Yan, Zheng | 8cef4e1 | 2009-11-12 09:33:26 +0000 | [diff] [blame] | 2059 | * IO has been started, blocks of the log tree have WRITTEN flag set |
| 2060 | * in their headers. new modifications of the log will be written to |
| 2061 | * new positions. so it's safe to allow log writers to go in. |
Yan Zheng | 7237f18 | 2009-01-21 12:54:03 -0500 | [diff] [blame] | 2062 | */ |
| 2063 | mutex_unlock(&root->log_mutex); |
| 2064 | |
| 2065 | mutex_lock(&log_root_tree->log_mutex); |
| 2066 | log_root_tree->log_batch++; |
| 2067 | atomic_inc(&log_root_tree->log_writers); |
| 2068 | mutex_unlock(&log_root_tree->log_mutex); |
| 2069 | |
| 2070 | ret = update_log_root(trans, log); |
Yan Zheng | 7237f18 | 2009-01-21 12:54:03 -0500 | [diff] [blame] | 2071 | |
| 2072 | mutex_lock(&log_root_tree->log_mutex); |
| 2073 | if (atomic_dec_and_test(&log_root_tree->log_writers)) { |
| 2074 | smp_mb(); |
| 2075 | if (waitqueue_active(&log_root_tree->log_writer_wait)) |
| 2076 | wake_up(&log_root_tree->log_writer_wait); |
| 2077 | } |
| 2078 | |
Yan, Zheng | 4a500fd | 2010-05-16 10:49:59 -0400 | [diff] [blame] | 2079 | if (ret) { |
| 2080 | BUG_ON(ret != -ENOSPC); |
| 2081 | root->fs_info->last_trans_log_full_commit = trans->transid; |
| 2082 | btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark); |
| 2083 | mutex_unlock(&log_root_tree->log_mutex); |
| 2084 | ret = -EAGAIN; |
| 2085 | goto out; |
| 2086 | } |
| 2087 | |
Yan Zheng | 7237f18 | 2009-01-21 12:54:03 -0500 | [diff] [blame] | 2088 | index2 = log_root_tree->log_transid % 2; |
| 2089 | if (atomic_read(&log_root_tree->log_commit[index2])) { |
Yan, Zheng | 8cef4e1 | 2009-11-12 09:33:26 +0000 | [diff] [blame] | 2090 | btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark); |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 2091 | wait_log_commit(trans, log_root_tree, |
| 2092 | log_root_tree->log_transid); |
Yan Zheng | 7237f18 | 2009-01-21 12:54:03 -0500 | [diff] [blame] | 2093 | mutex_unlock(&log_root_tree->log_mutex); |
Chris Mason | b31eabd | 2011-01-31 16:48:24 -0500 | [diff] [blame] | 2094 | ret = 0; |
Yan Zheng | 7237f18 | 2009-01-21 12:54:03 -0500 | [diff] [blame] | 2095 | goto out; |
| 2096 | } |
| 2097 | atomic_set(&log_root_tree->log_commit[index2], 1); |
| 2098 | |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 2099 | if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) { |
| 2100 | wait_log_commit(trans, log_root_tree, |
| 2101 | log_root_tree->log_transid - 1); |
| 2102 | } |
Yan Zheng | 7237f18 | 2009-01-21 12:54:03 -0500 | [diff] [blame] | 2103 | |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 2104 | wait_for_writer(trans, log_root_tree); |
| 2105 | |
| 2106 | /* |
| 2107 | * now that we've moved on to the tree of log tree roots, |
| 2108 | * check the full commit flag again |
| 2109 | */ |
| 2110 | if (root->fs_info->last_trans_log_full_commit == trans->transid) { |
Yan, Zheng | 8cef4e1 | 2009-11-12 09:33:26 +0000 | [diff] [blame] | 2111 | btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark); |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 2112 | mutex_unlock(&log_root_tree->log_mutex); |
| 2113 | ret = -EAGAIN; |
| 2114 | goto out_wake_log_root; |
| 2115 | } |
Yan Zheng | 7237f18 | 2009-01-21 12:54:03 -0500 | [diff] [blame] | 2116 | |
| 2117 | ret = btrfs_write_and_wait_marked_extents(log_root_tree, |
Yan, Zheng | 8cef4e1 | 2009-11-12 09:33:26 +0000 | [diff] [blame] | 2118 | &log_root_tree->dirty_log_pages, |
| 2119 | EXTENT_DIRTY | EXTENT_NEW); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2120 | BUG_ON(ret); |
Yan, Zheng | 8cef4e1 | 2009-11-12 09:33:26 +0000 | [diff] [blame] | 2121 | btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2122 | |
David Sterba | 6c41761 | 2011-04-13 15:41:04 +0200 | [diff] [blame] | 2123 | btrfs_set_super_log_root(root->fs_info->super_for_commit, |
Yan Zheng | 7237f18 | 2009-01-21 12:54:03 -0500 | [diff] [blame] | 2124 | log_root_tree->node->start); |
David Sterba | 6c41761 | 2011-04-13 15:41:04 +0200 | [diff] [blame] | 2125 | btrfs_set_super_log_root_level(root->fs_info->super_for_commit, |
Yan Zheng | 7237f18 | 2009-01-21 12:54:03 -0500 | [diff] [blame] | 2126 | btrfs_header_level(log_root_tree->node)); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2127 | |
Yan Zheng | 7237f18 | 2009-01-21 12:54:03 -0500 | [diff] [blame] | 2128 | log_root_tree->log_batch = 0; |
| 2129 | log_root_tree->log_transid++; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2130 | smp_mb(); |
Yan Zheng | 7237f18 | 2009-01-21 12:54:03 -0500 | [diff] [blame] | 2131 | |
| 2132 | mutex_unlock(&log_root_tree->log_mutex); |
| 2133 | |
| 2134 | /* |
| 2135 | * nobody else is going to jump in and write the the ctree |
| 2136 | * super here because the log_commit atomic below is protecting |
| 2137 | * us. We must be called with a transaction handle pinning |
| 2138 | * the running transaction open, so a full commit can't hop |
| 2139 | * in and cause problems either. |
| 2140 | */ |
Arne Jansen | a2de733 | 2011-03-08 14:14:00 +0100 | [diff] [blame] | 2141 | btrfs_scrub_pause_super(root); |
Chris Mason | 4722607 | 2009-10-13 12:55:09 -0400 | [diff] [blame] | 2142 | write_ctree_super(trans, root->fs_info->tree_root, 1); |
Arne Jansen | a2de733 | 2011-03-08 14:14:00 +0100 | [diff] [blame] | 2143 | btrfs_scrub_continue_super(root); |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 2144 | ret = 0; |
Yan Zheng | 7237f18 | 2009-01-21 12:54:03 -0500 | [diff] [blame] | 2145 | |
Chris Mason | 257c62e | 2009-10-13 13:21:08 -0400 | [diff] [blame] | 2146 | mutex_lock(&root->log_mutex); |
| 2147 | if (root->last_log_commit < log_transid) |
| 2148 | root->last_log_commit = log_transid; |
| 2149 | mutex_unlock(&root->log_mutex); |
| 2150 | |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 2151 | out_wake_log_root: |
Yan Zheng | 7237f18 | 2009-01-21 12:54:03 -0500 | [diff] [blame] | 2152 | atomic_set(&log_root_tree->log_commit[index2], 0); |
| 2153 | smp_mb(); |
| 2154 | if (waitqueue_active(&log_root_tree->log_commit_wait[index2])) |
| 2155 | wake_up(&log_root_tree->log_commit_wait[index2]); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2156 | out: |
Yan Zheng | 7237f18 | 2009-01-21 12:54:03 -0500 | [diff] [blame] | 2157 | atomic_set(&root->log_commit[index1], 0); |
| 2158 | smp_mb(); |
| 2159 | if (waitqueue_active(&root->log_commit_wait[index1])) |
| 2160 | wake_up(&root->log_commit_wait[index1]); |
Chris Mason | b31eabd | 2011-01-31 16:48:24 -0500 | [diff] [blame] | 2161 | return ret; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2162 | } |
| 2163 | |
Yan, Zheng | 4a500fd | 2010-05-16 10:49:59 -0400 | [diff] [blame] | 2164 | static void free_log_tree(struct btrfs_trans_handle *trans, |
| 2165 | struct btrfs_root *log) |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2166 | { |
| 2167 | int ret; |
Chris Mason | d0c803c | 2008-09-11 16:17:57 -0400 | [diff] [blame] | 2168 | u64 start; |
| 2169 | u64 end; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2170 | struct walk_control wc = { |
| 2171 | .free = 1, |
| 2172 | .process_func = process_one_buffer |
| 2173 | }; |
| 2174 | |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2175 | ret = walk_log_tree(trans, log, &wc); |
| 2176 | BUG_ON(ret); |
| 2177 | |
Chris Mason | d397712 | 2009-01-05 21:25:51 -0500 | [diff] [blame] | 2178 | while (1) { |
Chris Mason | d0c803c | 2008-09-11 16:17:57 -0400 | [diff] [blame] | 2179 | ret = find_first_extent_bit(&log->dirty_log_pages, |
Yan, Zheng | 8cef4e1 | 2009-11-12 09:33:26 +0000 | [diff] [blame] | 2180 | 0, &start, &end, EXTENT_DIRTY | EXTENT_NEW); |
Chris Mason | d0c803c | 2008-09-11 16:17:57 -0400 | [diff] [blame] | 2181 | if (ret) |
| 2182 | break; |
| 2183 | |
Yan, Zheng | 8cef4e1 | 2009-11-12 09:33:26 +0000 | [diff] [blame] | 2184 | clear_extent_bits(&log->dirty_log_pages, start, end, |
| 2185 | EXTENT_DIRTY | EXTENT_NEW, GFP_NOFS); |
Chris Mason | d0c803c | 2008-09-11 16:17:57 -0400 | [diff] [blame] | 2186 | } |
| 2187 | |
Yan Zheng | 7237f18 | 2009-01-21 12:54:03 -0500 | [diff] [blame] | 2188 | free_extent_buffer(log->node); |
| 2189 | kfree(log); |
Yan, Zheng | 4a500fd | 2010-05-16 10:49:59 -0400 | [diff] [blame] | 2190 | } |
| 2191 | |
| 2192 | /* |
| 2193 | * free all the extents used by the tree log. This should be called |
| 2194 | * at commit time of the full transaction |
| 2195 | */ |
| 2196 | int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root) |
| 2197 | { |
| 2198 | if (root->log_root) { |
| 2199 | free_log_tree(trans, root->log_root); |
| 2200 | root->log_root = NULL; |
| 2201 | } |
| 2202 | return 0; |
| 2203 | } |
| 2204 | |
| 2205 | int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans, |
| 2206 | struct btrfs_fs_info *fs_info) |
| 2207 | { |
| 2208 | if (fs_info->log_root_tree) { |
| 2209 | free_log_tree(trans, fs_info->log_root_tree); |
| 2210 | fs_info->log_root_tree = NULL; |
| 2211 | } |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2212 | return 0; |
| 2213 | } |
| 2214 | |
| 2215 | /* |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2216 | * If both a file and directory are logged, and unlinks or renames are |
| 2217 | * mixed in, we have a few interesting corners: |
| 2218 | * |
| 2219 | * create file X in dir Y |
| 2220 | * link file X to X.link in dir Y |
| 2221 | * fsync file X |
| 2222 | * unlink file X but leave X.link |
| 2223 | * fsync dir Y |
| 2224 | * |
| 2225 | * After a crash we would expect only X.link to exist. But file X |
| 2226 | * didn't get fsync'd again so the log has back refs for X and X.link. |
| 2227 | * |
| 2228 | * We solve this by removing directory entries and inode backrefs from the |
| 2229 | * log when a file that was logged in the current transaction is |
| 2230 | * unlinked. Any later fsync will include the updated log entries, and |
| 2231 | * we'll be able to reconstruct the proper directory items from backrefs. |
| 2232 | * |
| 2233 | * This optimizations allows us to avoid relogging the entire inode |
| 2234 | * or the entire directory. |
| 2235 | */ |
| 2236 | int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans, |
| 2237 | struct btrfs_root *root, |
| 2238 | const char *name, int name_len, |
| 2239 | struct inode *dir, u64 index) |
| 2240 | { |
| 2241 | struct btrfs_root *log; |
| 2242 | struct btrfs_dir_item *di; |
| 2243 | struct btrfs_path *path; |
| 2244 | int ret; |
Yan, Zheng | 4a500fd | 2010-05-16 10:49:59 -0400 | [diff] [blame] | 2245 | int err = 0; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2246 | int bytes_del = 0; |
Li Zefan | 33345d01 | 2011-04-20 10:31:50 +0800 | [diff] [blame] | 2247 | u64 dir_ino = btrfs_ino(dir); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2248 | |
Chris Mason | 3a5f1d4 | 2008-09-11 15:53:37 -0400 | [diff] [blame] | 2249 | if (BTRFS_I(dir)->logged_trans < trans->transid) |
| 2250 | return 0; |
| 2251 | |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2252 | ret = join_running_log_trans(root); |
| 2253 | if (ret) |
| 2254 | return 0; |
| 2255 | |
| 2256 | mutex_lock(&BTRFS_I(dir)->log_mutex); |
| 2257 | |
| 2258 | log = root->log_root; |
| 2259 | path = btrfs_alloc_path(); |
Tsutomu Itoh | a62f44a | 2011-04-25 19:43:51 -0400 | [diff] [blame] | 2260 | if (!path) { |
| 2261 | err = -ENOMEM; |
| 2262 | goto out_unlock; |
| 2263 | } |
liubo | 2a29edc | 2011-01-26 06:22:08 +0000 | [diff] [blame] | 2264 | |
Li Zefan | 33345d01 | 2011-04-20 10:31:50 +0800 | [diff] [blame] | 2265 | di = btrfs_lookup_dir_item(trans, log, path, dir_ino, |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2266 | name, name_len, -1); |
Yan, Zheng | 4a500fd | 2010-05-16 10:49:59 -0400 | [diff] [blame] | 2267 | if (IS_ERR(di)) { |
| 2268 | err = PTR_ERR(di); |
| 2269 | goto fail; |
| 2270 | } |
| 2271 | if (di) { |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2272 | ret = btrfs_delete_one_dir_name(trans, log, path, di); |
| 2273 | bytes_del += name_len; |
| 2274 | BUG_ON(ret); |
| 2275 | } |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 2276 | btrfs_release_path(path); |
Li Zefan | 33345d01 | 2011-04-20 10:31:50 +0800 | [diff] [blame] | 2277 | di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino, |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2278 | index, name, name_len, -1); |
Yan, Zheng | 4a500fd | 2010-05-16 10:49:59 -0400 | [diff] [blame] | 2279 | if (IS_ERR(di)) { |
| 2280 | err = PTR_ERR(di); |
| 2281 | goto fail; |
| 2282 | } |
| 2283 | if (di) { |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2284 | ret = btrfs_delete_one_dir_name(trans, log, path, di); |
| 2285 | bytes_del += name_len; |
| 2286 | BUG_ON(ret); |
| 2287 | } |
| 2288 | |
| 2289 | /* update the directory size in the log to reflect the names |
| 2290 | * we have removed |
| 2291 | */ |
| 2292 | if (bytes_del) { |
| 2293 | struct btrfs_key key; |
| 2294 | |
Li Zefan | 33345d01 | 2011-04-20 10:31:50 +0800 | [diff] [blame] | 2295 | key.objectid = dir_ino; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2296 | key.offset = 0; |
| 2297 | key.type = BTRFS_INODE_ITEM_KEY; |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 2298 | btrfs_release_path(path); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2299 | |
| 2300 | ret = btrfs_search_slot(trans, log, &key, path, 0, 1); |
Yan, Zheng | 4a500fd | 2010-05-16 10:49:59 -0400 | [diff] [blame] | 2301 | if (ret < 0) { |
| 2302 | err = ret; |
| 2303 | goto fail; |
| 2304 | } |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2305 | if (ret == 0) { |
| 2306 | struct btrfs_inode_item *item; |
| 2307 | u64 i_size; |
| 2308 | |
| 2309 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
| 2310 | struct btrfs_inode_item); |
| 2311 | i_size = btrfs_inode_size(path->nodes[0], item); |
| 2312 | if (i_size > bytes_del) |
| 2313 | i_size -= bytes_del; |
| 2314 | else |
| 2315 | i_size = 0; |
| 2316 | btrfs_set_inode_size(path->nodes[0], item, i_size); |
| 2317 | btrfs_mark_buffer_dirty(path->nodes[0]); |
| 2318 | } else |
| 2319 | ret = 0; |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 2320 | btrfs_release_path(path); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2321 | } |
Yan, Zheng | 4a500fd | 2010-05-16 10:49:59 -0400 | [diff] [blame] | 2322 | fail: |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2323 | btrfs_free_path(path); |
Tsutomu Itoh | a62f44a | 2011-04-25 19:43:51 -0400 | [diff] [blame] | 2324 | out_unlock: |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2325 | mutex_unlock(&BTRFS_I(dir)->log_mutex); |
Yan, Zheng | 4a500fd | 2010-05-16 10:49:59 -0400 | [diff] [blame] | 2326 | if (ret == -ENOSPC) { |
| 2327 | root->fs_info->last_trans_log_full_commit = trans->transid; |
| 2328 | ret = 0; |
| 2329 | } |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 2330 | btrfs_end_log_trans(root); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2331 | |
Andi Kleen | 411fc6b | 2010-10-29 15:14:31 -0400 | [diff] [blame] | 2332 | return err; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2333 | } |
| 2334 | |
| 2335 | /* see comments for btrfs_del_dir_entries_in_log */ |
| 2336 | int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans, |
| 2337 | struct btrfs_root *root, |
| 2338 | const char *name, int name_len, |
| 2339 | struct inode *inode, u64 dirid) |
| 2340 | { |
| 2341 | struct btrfs_root *log; |
| 2342 | u64 index; |
| 2343 | int ret; |
| 2344 | |
Chris Mason | 3a5f1d4 | 2008-09-11 15:53:37 -0400 | [diff] [blame] | 2345 | if (BTRFS_I(inode)->logged_trans < trans->transid) |
| 2346 | return 0; |
| 2347 | |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2348 | ret = join_running_log_trans(root); |
| 2349 | if (ret) |
| 2350 | return 0; |
| 2351 | log = root->log_root; |
| 2352 | mutex_lock(&BTRFS_I(inode)->log_mutex); |
| 2353 | |
Li Zefan | 33345d01 | 2011-04-20 10:31:50 +0800 | [diff] [blame] | 2354 | ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode), |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2355 | dirid, &index); |
| 2356 | mutex_unlock(&BTRFS_I(inode)->log_mutex); |
Yan, Zheng | 4a500fd | 2010-05-16 10:49:59 -0400 | [diff] [blame] | 2357 | if (ret == -ENOSPC) { |
| 2358 | root->fs_info->last_trans_log_full_commit = trans->transid; |
| 2359 | ret = 0; |
| 2360 | } |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 2361 | btrfs_end_log_trans(root); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2362 | |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2363 | return ret; |
| 2364 | } |
| 2365 | |
| 2366 | /* |
| 2367 | * creates a range item in the log for 'dirid'. first_offset and |
| 2368 | * last_offset tell us which parts of the key space the log should |
| 2369 | * be considered authoritative for. |
| 2370 | */ |
| 2371 | static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans, |
| 2372 | struct btrfs_root *log, |
| 2373 | struct btrfs_path *path, |
| 2374 | int key_type, u64 dirid, |
| 2375 | u64 first_offset, u64 last_offset) |
| 2376 | { |
| 2377 | int ret; |
| 2378 | struct btrfs_key key; |
| 2379 | struct btrfs_dir_log_item *item; |
| 2380 | |
| 2381 | key.objectid = dirid; |
| 2382 | key.offset = first_offset; |
| 2383 | if (key_type == BTRFS_DIR_ITEM_KEY) |
| 2384 | key.type = BTRFS_DIR_LOG_ITEM_KEY; |
| 2385 | else |
| 2386 | key.type = BTRFS_DIR_LOG_INDEX_KEY; |
| 2387 | ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item)); |
Yan, Zheng | 4a500fd | 2010-05-16 10:49:59 -0400 | [diff] [blame] | 2388 | if (ret) |
| 2389 | return ret; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2390 | |
| 2391 | item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
| 2392 | struct btrfs_dir_log_item); |
| 2393 | btrfs_set_dir_log_end(path->nodes[0], item, last_offset); |
| 2394 | btrfs_mark_buffer_dirty(path->nodes[0]); |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 2395 | btrfs_release_path(path); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2396 | return 0; |
| 2397 | } |
| 2398 | |
| 2399 | /* |
| 2400 | * log all the items included in the current transaction for a given |
| 2401 | * directory. This also creates the range items in the log tree required |
| 2402 | * to replay anything deleted before the fsync |
| 2403 | */ |
| 2404 | static noinline int log_dir_items(struct btrfs_trans_handle *trans, |
| 2405 | struct btrfs_root *root, struct inode *inode, |
| 2406 | struct btrfs_path *path, |
| 2407 | struct btrfs_path *dst_path, int key_type, |
| 2408 | u64 min_offset, u64 *last_offset_ret) |
| 2409 | { |
| 2410 | struct btrfs_key min_key; |
| 2411 | struct btrfs_key max_key; |
| 2412 | struct btrfs_root *log = root->log_root; |
| 2413 | struct extent_buffer *src; |
Yan, Zheng | 4a500fd | 2010-05-16 10:49:59 -0400 | [diff] [blame] | 2414 | int err = 0; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2415 | int ret; |
| 2416 | int i; |
| 2417 | int nritems; |
| 2418 | u64 first_offset = min_offset; |
| 2419 | u64 last_offset = (u64)-1; |
Li Zefan | 33345d01 | 2011-04-20 10:31:50 +0800 | [diff] [blame] | 2420 | u64 ino = btrfs_ino(inode); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2421 | |
| 2422 | log = root->log_root; |
Li Zefan | 33345d01 | 2011-04-20 10:31:50 +0800 | [diff] [blame] | 2423 | max_key.objectid = ino; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2424 | max_key.offset = (u64)-1; |
| 2425 | max_key.type = key_type; |
| 2426 | |
Li Zefan | 33345d01 | 2011-04-20 10:31:50 +0800 | [diff] [blame] | 2427 | min_key.objectid = ino; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2428 | min_key.type = key_type; |
| 2429 | min_key.offset = min_offset; |
| 2430 | |
| 2431 | path->keep_locks = 1; |
| 2432 | |
| 2433 | ret = btrfs_search_forward(root, &min_key, &max_key, |
| 2434 | path, 0, trans->transid); |
| 2435 | |
| 2436 | /* |
| 2437 | * we didn't find anything from this transaction, see if there |
| 2438 | * is anything at all |
| 2439 | */ |
Li Zefan | 33345d01 | 2011-04-20 10:31:50 +0800 | [diff] [blame] | 2440 | if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) { |
| 2441 | min_key.objectid = ino; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2442 | min_key.type = key_type; |
| 2443 | min_key.offset = (u64)-1; |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 2444 | btrfs_release_path(path); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2445 | ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0); |
| 2446 | if (ret < 0) { |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 2447 | btrfs_release_path(path); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2448 | return ret; |
| 2449 | } |
Li Zefan | 33345d01 | 2011-04-20 10:31:50 +0800 | [diff] [blame] | 2450 | ret = btrfs_previous_item(root, path, ino, key_type); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2451 | |
| 2452 | /* if ret == 0 there are items for this type, |
| 2453 | * create a range to tell us the last key of this type. |
| 2454 | * otherwise, there are no items in this directory after |
| 2455 | * *min_offset, and we create a range to indicate that. |
| 2456 | */ |
| 2457 | if (ret == 0) { |
| 2458 | struct btrfs_key tmp; |
| 2459 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, |
| 2460 | path->slots[0]); |
Chris Mason | d397712 | 2009-01-05 21:25:51 -0500 | [diff] [blame] | 2461 | if (key_type == tmp.type) |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2462 | first_offset = max(min_offset, tmp.offset) + 1; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2463 | } |
| 2464 | goto done; |
| 2465 | } |
| 2466 | |
| 2467 | /* go backward to find any previous key */ |
Li Zefan | 33345d01 | 2011-04-20 10:31:50 +0800 | [diff] [blame] | 2468 | ret = btrfs_previous_item(root, path, ino, key_type); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2469 | if (ret == 0) { |
| 2470 | struct btrfs_key tmp; |
| 2471 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]); |
| 2472 | if (key_type == tmp.type) { |
| 2473 | first_offset = tmp.offset; |
| 2474 | ret = overwrite_item(trans, log, dst_path, |
| 2475 | path->nodes[0], path->slots[0], |
| 2476 | &tmp); |
Yan, Zheng | 4a500fd | 2010-05-16 10:49:59 -0400 | [diff] [blame] | 2477 | if (ret) { |
| 2478 | err = ret; |
| 2479 | goto done; |
| 2480 | } |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2481 | } |
| 2482 | } |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 2483 | btrfs_release_path(path); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2484 | |
| 2485 | /* find the first key from this transaction again */ |
| 2486 | ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0); |
| 2487 | if (ret != 0) { |
| 2488 | WARN_ON(1); |
| 2489 | goto done; |
| 2490 | } |
| 2491 | |
| 2492 | /* |
| 2493 | * we have a block from this transaction, log every item in it |
| 2494 | * from our directory |
| 2495 | */ |
Chris Mason | d397712 | 2009-01-05 21:25:51 -0500 | [diff] [blame] | 2496 | while (1) { |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2497 | struct btrfs_key tmp; |
| 2498 | src = path->nodes[0]; |
| 2499 | nritems = btrfs_header_nritems(src); |
| 2500 | for (i = path->slots[0]; i < nritems; i++) { |
| 2501 | btrfs_item_key_to_cpu(src, &min_key, i); |
| 2502 | |
Li Zefan | 33345d01 | 2011-04-20 10:31:50 +0800 | [diff] [blame] | 2503 | if (min_key.objectid != ino || min_key.type != key_type) |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2504 | goto done; |
| 2505 | ret = overwrite_item(trans, log, dst_path, src, i, |
| 2506 | &min_key); |
Yan, Zheng | 4a500fd | 2010-05-16 10:49:59 -0400 | [diff] [blame] | 2507 | if (ret) { |
| 2508 | err = ret; |
| 2509 | goto done; |
| 2510 | } |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2511 | } |
| 2512 | path->slots[0] = nritems; |
| 2513 | |
| 2514 | /* |
| 2515 | * look ahead to the next item and see if it is also |
| 2516 | * from this directory and from this transaction |
| 2517 | */ |
| 2518 | ret = btrfs_next_leaf(root, path); |
| 2519 | if (ret == 1) { |
| 2520 | last_offset = (u64)-1; |
| 2521 | goto done; |
| 2522 | } |
| 2523 | btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]); |
Li Zefan | 33345d01 | 2011-04-20 10:31:50 +0800 | [diff] [blame] | 2524 | if (tmp.objectid != ino || tmp.type != key_type) { |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2525 | last_offset = (u64)-1; |
| 2526 | goto done; |
| 2527 | } |
| 2528 | if (btrfs_header_generation(path->nodes[0]) != trans->transid) { |
| 2529 | ret = overwrite_item(trans, log, dst_path, |
| 2530 | path->nodes[0], path->slots[0], |
| 2531 | &tmp); |
Yan, Zheng | 4a500fd | 2010-05-16 10:49:59 -0400 | [diff] [blame] | 2532 | if (ret) |
| 2533 | err = ret; |
| 2534 | else |
| 2535 | last_offset = tmp.offset; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2536 | goto done; |
| 2537 | } |
| 2538 | } |
| 2539 | done: |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 2540 | btrfs_release_path(path); |
| 2541 | btrfs_release_path(dst_path); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2542 | |
Yan, Zheng | 4a500fd | 2010-05-16 10:49:59 -0400 | [diff] [blame] | 2543 | if (err == 0) { |
| 2544 | *last_offset_ret = last_offset; |
| 2545 | /* |
| 2546 | * insert the log range keys to indicate where the log |
| 2547 | * is valid |
| 2548 | */ |
| 2549 | ret = insert_dir_log_key(trans, log, path, key_type, |
Li Zefan | 33345d01 | 2011-04-20 10:31:50 +0800 | [diff] [blame] | 2550 | ino, first_offset, last_offset); |
Yan, Zheng | 4a500fd | 2010-05-16 10:49:59 -0400 | [diff] [blame] | 2551 | if (ret) |
| 2552 | err = ret; |
| 2553 | } |
| 2554 | return err; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2555 | } |
| 2556 | |
| 2557 | /* |
| 2558 | * logging directories is very similar to logging inodes, We find all the items |
| 2559 | * from the current transaction and write them to the log. |
| 2560 | * |
| 2561 | * The recovery code scans the directory in the subvolume, and if it finds a |
| 2562 | * key in the range logged that is not present in the log tree, then it means |
| 2563 | * that dir entry was unlinked during the transaction. |
| 2564 | * |
| 2565 | * In order for that scan to work, we must include one key smaller than |
| 2566 | * the smallest logged by this transaction and one key larger than the largest |
| 2567 | * key logged by this transaction. |
| 2568 | */ |
| 2569 | static noinline int log_directory_changes(struct btrfs_trans_handle *trans, |
| 2570 | struct btrfs_root *root, struct inode *inode, |
| 2571 | struct btrfs_path *path, |
| 2572 | struct btrfs_path *dst_path) |
| 2573 | { |
| 2574 | u64 min_key; |
| 2575 | u64 max_key; |
| 2576 | int ret; |
| 2577 | int key_type = BTRFS_DIR_ITEM_KEY; |
| 2578 | |
| 2579 | again: |
| 2580 | min_key = 0; |
| 2581 | max_key = 0; |
Chris Mason | d397712 | 2009-01-05 21:25:51 -0500 | [diff] [blame] | 2582 | while (1) { |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2583 | ret = log_dir_items(trans, root, inode, path, |
| 2584 | dst_path, key_type, min_key, |
| 2585 | &max_key); |
Yan, Zheng | 4a500fd | 2010-05-16 10:49:59 -0400 | [diff] [blame] | 2586 | if (ret) |
| 2587 | return ret; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2588 | if (max_key == (u64)-1) |
| 2589 | break; |
| 2590 | min_key = max_key + 1; |
| 2591 | } |
| 2592 | |
| 2593 | if (key_type == BTRFS_DIR_ITEM_KEY) { |
| 2594 | key_type = BTRFS_DIR_INDEX_KEY; |
| 2595 | goto again; |
| 2596 | } |
| 2597 | return 0; |
| 2598 | } |
| 2599 | |
| 2600 | /* |
| 2601 | * a helper function to drop items from the log before we relog an |
| 2602 | * inode. max_key_type indicates the highest item type to remove. |
| 2603 | * This cannot be run for file data extents because it does not |
| 2604 | * free the extents they point to. |
| 2605 | */ |
| 2606 | static int drop_objectid_items(struct btrfs_trans_handle *trans, |
| 2607 | struct btrfs_root *log, |
| 2608 | struct btrfs_path *path, |
| 2609 | u64 objectid, int max_key_type) |
| 2610 | { |
| 2611 | int ret; |
| 2612 | struct btrfs_key key; |
| 2613 | struct btrfs_key found_key; |
| 2614 | |
| 2615 | key.objectid = objectid; |
| 2616 | key.type = max_key_type; |
| 2617 | key.offset = (u64)-1; |
| 2618 | |
Chris Mason | d397712 | 2009-01-05 21:25:51 -0500 | [diff] [blame] | 2619 | while (1) { |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2620 | ret = btrfs_search_slot(trans, log, &key, path, -1, 1); |
Yan, Zheng | 4a500fd | 2010-05-16 10:49:59 -0400 | [diff] [blame] | 2621 | BUG_ON(ret == 0); |
| 2622 | if (ret < 0) |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2623 | break; |
| 2624 | |
| 2625 | if (path->slots[0] == 0) |
| 2626 | break; |
| 2627 | |
| 2628 | path->slots[0]--; |
| 2629 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, |
| 2630 | path->slots[0]); |
| 2631 | |
| 2632 | if (found_key.objectid != objectid) |
| 2633 | break; |
| 2634 | |
| 2635 | ret = btrfs_del_item(trans, log, path); |
Tsutomu Itoh | 65a246c | 2011-05-19 04:37:44 +0000 | [diff] [blame] | 2636 | if (ret) |
| 2637 | break; |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 2638 | btrfs_release_path(path); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2639 | } |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 2640 | btrfs_release_path(path); |
Yan, Zheng | 4a500fd | 2010-05-16 10:49:59 -0400 | [diff] [blame] | 2641 | return ret; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2642 | } |
| 2643 | |
Chris Mason | 31ff1cd | 2008-09-11 16:17:57 -0400 | [diff] [blame] | 2644 | static noinline int copy_items(struct btrfs_trans_handle *trans, |
| 2645 | struct btrfs_root *log, |
| 2646 | struct btrfs_path *dst_path, |
| 2647 | struct extent_buffer *src, |
| 2648 | int start_slot, int nr, int inode_only) |
| 2649 | { |
| 2650 | unsigned long src_offset; |
| 2651 | unsigned long dst_offset; |
| 2652 | struct btrfs_file_extent_item *extent; |
| 2653 | struct btrfs_inode_item *inode_item; |
| 2654 | int ret; |
| 2655 | struct btrfs_key *ins_keys; |
| 2656 | u32 *ins_sizes; |
| 2657 | char *ins_data; |
| 2658 | int i; |
Chris Mason | d20f704 | 2008-12-08 16:58:54 -0500 | [diff] [blame] | 2659 | struct list_head ordered_sums; |
| 2660 | |
| 2661 | INIT_LIST_HEAD(&ordered_sums); |
Chris Mason | 31ff1cd | 2008-09-11 16:17:57 -0400 | [diff] [blame] | 2662 | |
| 2663 | ins_data = kmalloc(nr * sizeof(struct btrfs_key) + |
| 2664 | nr * sizeof(u32), GFP_NOFS); |
liubo | 2a29edc | 2011-01-26 06:22:08 +0000 | [diff] [blame] | 2665 | if (!ins_data) |
| 2666 | return -ENOMEM; |
| 2667 | |
Chris Mason | 31ff1cd | 2008-09-11 16:17:57 -0400 | [diff] [blame] | 2668 | ins_sizes = (u32 *)ins_data; |
| 2669 | ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32)); |
| 2670 | |
| 2671 | for (i = 0; i < nr; i++) { |
| 2672 | ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot); |
| 2673 | btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot); |
| 2674 | } |
| 2675 | ret = btrfs_insert_empty_items(trans, log, dst_path, |
| 2676 | ins_keys, ins_sizes, nr); |
Yan, Zheng | 4a500fd | 2010-05-16 10:49:59 -0400 | [diff] [blame] | 2677 | if (ret) { |
| 2678 | kfree(ins_data); |
| 2679 | return ret; |
| 2680 | } |
Chris Mason | 31ff1cd | 2008-09-11 16:17:57 -0400 | [diff] [blame] | 2681 | |
Yan Zheng | 5d4f98a | 2009-06-10 10:45:14 -0400 | [diff] [blame] | 2682 | for (i = 0; i < nr; i++, dst_path->slots[0]++) { |
Chris Mason | 31ff1cd | 2008-09-11 16:17:57 -0400 | [diff] [blame] | 2683 | dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0], |
| 2684 | dst_path->slots[0]); |
| 2685 | |
| 2686 | src_offset = btrfs_item_ptr_offset(src, start_slot + i); |
| 2687 | |
| 2688 | copy_extent_buffer(dst_path->nodes[0], src, dst_offset, |
| 2689 | src_offset, ins_sizes[i]); |
| 2690 | |
| 2691 | if (inode_only == LOG_INODE_EXISTS && |
| 2692 | ins_keys[i].type == BTRFS_INODE_ITEM_KEY) { |
| 2693 | inode_item = btrfs_item_ptr(dst_path->nodes[0], |
| 2694 | dst_path->slots[0], |
| 2695 | struct btrfs_inode_item); |
| 2696 | btrfs_set_inode_size(dst_path->nodes[0], inode_item, 0); |
| 2697 | |
| 2698 | /* set the generation to zero so the recover code |
| 2699 | * can tell the difference between an logging |
| 2700 | * just to say 'this inode exists' and a logging |
| 2701 | * to say 'update this inode with these values' |
| 2702 | */ |
| 2703 | btrfs_set_inode_generation(dst_path->nodes[0], |
| 2704 | inode_item, 0); |
| 2705 | } |
| 2706 | /* take a reference on file data extents so that truncates |
| 2707 | * or deletes of this inode don't have to relog the inode |
| 2708 | * again |
| 2709 | */ |
| 2710 | if (btrfs_key_type(ins_keys + i) == BTRFS_EXTENT_DATA_KEY) { |
| 2711 | int found_type; |
| 2712 | extent = btrfs_item_ptr(src, start_slot + i, |
| 2713 | struct btrfs_file_extent_item); |
| 2714 | |
liubo | 8e531cd | 2011-05-06 10:36:09 +0800 | [diff] [blame] | 2715 | if (btrfs_file_extent_generation(src, extent) < trans->transid) |
| 2716 | continue; |
| 2717 | |
Chris Mason | 31ff1cd | 2008-09-11 16:17:57 -0400 | [diff] [blame] | 2718 | found_type = btrfs_file_extent_type(src, extent); |
Yan Zheng | d899e05 | 2008-10-30 14:25:28 -0400 | [diff] [blame] | 2719 | if (found_type == BTRFS_FILE_EXTENT_REG || |
| 2720 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { |
Yan Zheng | 5d4f98a | 2009-06-10 10:45:14 -0400 | [diff] [blame] | 2721 | u64 ds, dl, cs, cl; |
| 2722 | ds = btrfs_file_extent_disk_bytenr(src, |
| 2723 | extent); |
| 2724 | /* ds == 0 is a hole */ |
| 2725 | if (ds == 0) |
| 2726 | continue; |
| 2727 | |
| 2728 | dl = btrfs_file_extent_disk_num_bytes(src, |
| 2729 | extent); |
| 2730 | cs = btrfs_file_extent_offset(src, extent); |
| 2731 | cl = btrfs_file_extent_num_bytes(src, |
Joe Perches | a419aef | 2009-08-18 11:18:35 -0700 | [diff] [blame] | 2732 | extent); |
Chris Mason | 580afd7 | 2008-12-08 19:15:39 -0500 | [diff] [blame] | 2733 | if (btrfs_file_extent_compression(src, |
| 2734 | extent)) { |
| 2735 | cs = 0; |
| 2736 | cl = dl; |
| 2737 | } |
Yan Zheng | 5d4f98a | 2009-06-10 10:45:14 -0400 | [diff] [blame] | 2738 | |
| 2739 | ret = btrfs_lookup_csums_range( |
| 2740 | log->fs_info->csum_root, |
| 2741 | ds + cs, ds + cs + cl - 1, |
Arne Jansen | a2de733 | 2011-03-08 14:14:00 +0100 | [diff] [blame] | 2742 | &ordered_sums, 0); |
Yan Zheng | 5d4f98a | 2009-06-10 10:45:14 -0400 | [diff] [blame] | 2743 | BUG_ON(ret); |
Chris Mason | 31ff1cd | 2008-09-11 16:17:57 -0400 | [diff] [blame] | 2744 | } |
| 2745 | } |
Chris Mason | 31ff1cd | 2008-09-11 16:17:57 -0400 | [diff] [blame] | 2746 | } |
| 2747 | |
| 2748 | btrfs_mark_buffer_dirty(dst_path->nodes[0]); |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 2749 | btrfs_release_path(dst_path); |
Chris Mason | 31ff1cd | 2008-09-11 16:17:57 -0400 | [diff] [blame] | 2750 | kfree(ins_data); |
Chris Mason | d20f704 | 2008-12-08 16:58:54 -0500 | [diff] [blame] | 2751 | |
| 2752 | /* |
| 2753 | * we have to do this after the loop above to avoid changing the |
| 2754 | * log tree while trying to change the log tree. |
| 2755 | */ |
Yan, Zheng | 4a500fd | 2010-05-16 10:49:59 -0400 | [diff] [blame] | 2756 | ret = 0; |
Chris Mason | d397712 | 2009-01-05 21:25:51 -0500 | [diff] [blame] | 2757 | while (!list_empty(&ordered_sums)) { |
Chris Mason | d20f704 | 2008-12-08 16:58:54 -0500 | [diff] [blame] | 2758 | struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next, |
| 2759 | struct btrfs_ordered_sum, |
| 2760 | list); |
Yan, Zheng | 4a500fd | 2010-05-16 10:49:59 -0400 | [diff] [blame] | 2761 | if (!ret) |
| 2762 | ret = btrfs_csum_file_blocks(trans, log, sums); |
Chris Mason | d20f704 | 2008-12-08 16:58:54 -0500 | [diff] [blame] | 2763 | list_del(&sums->list); |
| 2764 | kfree(sums); |
| 2765 | } |
Yan, Zheng | 4a500fd | 2010-05-16 10:49:59 -0400 | [diff] [blame] | 2766 | return ret; |
Chris Mason | 31ff1cd | 2008-09-11 16:17:57 -0400 | [diff] [blame] | 2767 | } |
| 2768 | |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2769 | /* log a single inode in the tree log. |
| 2770 | * At least one parent directory for this inode must exist in the tree |
| 2771 | * or be logged already. |
| 2772 | * |
| 2773 | * Any items from this inode changed by the current transaction are copied |
| 2774 | * to the log tree. An extra reference is taken on any extents in this |
| 2775 | * file, allowing us to avoid a whole pile of corner cases around logging |
| 2776 | * blocks that have been removed from the tree. |
| 2777 | * |
| 2778 | * See LOG_INODE_ALL and related defines for a description of what inode_only |
| 2779 | * does. |
| 2780 | * |
| 2781 | * This handles both files and directories. |
| 2782 | */ |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 2783 | static int btrfs_log_inode(struct btrfs_trans_handle *trans, |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2784 | struct btrfs_root *root, struct inode *inode, |
| 2785 | int inode_only) |
| 2786 | { |
| 2787 | struct btrfs_path *path; |
| 2788 | struct btrfs_path *dst_path; |
| 2789 | struct btrfs_key min_key; |
| 2790 | struct btrfs_key max_key; |
| 2791 | struct btrfs_root *log = root->log_root; |
Chris Mason | 31ff1cd | 2008-09-11 16:17:57 -0400 | [diff] [blame] | 2792 | struct extent_buffer *src = NULL; |
Yan, Zheng | 4a500fd | 2010-05-16 10:49:59 -0400 | [diff] [blame] | 2793 | int err = 0; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2794 | int ret; |
Chris Mason | 3a5f1d4 | 2008-09-11 15:53:37 -0400 | [diff] [blame] | 2795 | int nritems; |
Chris Mason | 31ff1cd | 2008-09-11 16:17:57 -0400 | [diff] [blame] | 2796 | int ins_start_slot = 0; |
| 2797 | int ins_nr; |
Li Zefan | 33345d01 | 2011-04-20 10:31:50 +0800 | [diff] [blame] | 2798 | u64 ino = btrfs_ino(inode); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2799 | |
| 2800 | log = root->log_root; |
| 2801 | |
| 2802 | path = btrfs_alloc_path(); |
Tsutomu Itoh | 5df6708 | 2011-02-01 09:17:35 +0000 | [diff] [blame] | 2803 | if (!path) |
| 2804 | return -ENOMEM; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2805 | dst_path = btrfs_alloc_path(); |
Tsutomu Itoh | 5df6708 | 2011-02-01 09:17:35 +0000 | [diff] [blame] | 2806 | if (!dst_path) { |
| 2807 | btrfs_free_path(path); |
| 2808 | return -ENOMEM; |
| 2809 | } |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2810 | |
Li Zefan | 33345d01 | 2011-04-20 10:31:50 +0800 | [diff] [blame] | 2811 | min_key.objectid = ino; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2812 | min_key.type = BTRFS_INODE_ITEM_KEY; |
| 2813 | min_key.offset = 0; |
| 2814 | |
Li Zefan | 33345d01 | 2011-04-20 10:31:50 +0800 | [diff] [blame] | 2815 | max_key.objectid = ino; |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 2816 | |
| 2817 | /* today the code can only do partial logging of directories */ |
| 2818 | if (!S_ISDIR(inode->i_mode)) |
| 2819 | inode_only = LOG_INODE_ALL; |
| 2820 | |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2821 | if (inode_only == LOG_INODE_EXISTS || S_ISDIR(inode->i_mode)) |
| 2822 | max_key.type = BTRFS_XATTR_ITEM_KEY; |
| 2823 | else |
| 2824 | max_key.type = (u8)-1; |
| 2825 | max_key.offset = (u64)-1; |
| 2826 | |
Miao Xie | 16cdcec | 2011-04-22 18:12:22 +0800 | [diff] [blame] | 2827 | ret = btrfs_commit_inode_delayed_items(trans, inode); |
| 2828 | if (ret) { |
| 2829 | btrfs_free_path(path); |
| 2830 | btrfs_free_path(dst_path); |
| 2831 | return ret; |
| 2832 | } |
| 2833 | |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2834 | mutex_lock(&BTRFS_I(inode)->log_mutex); |
| 2835 | |
| 2836 | /* |
| 2837 | * a brute force approach to making sure we get the most uptodate |
| 2838 | * copies of everything. |
| 2839 | */ |
| 2840 | if (S_ISDIR(inode->i_mode)) { |
| 2841 | int max_key_type = BTRFS_DIR_LOG_INDEX_KEY; |
| 2842 | |
| 2843 | if (inode_only == LOG_INODE_EXISTS) |
| 2844 | max_key_type = BTRFS_XATTR_ITEM_KEY; |
Li Zefan | 33345d01 | 2011-04-20 10:31:50 +0800 | [diff] [blame] | 2845 | ret = drop_objectid_items(trans, log, path, ino, max_key_type); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2846 | } else { |
| 2847 | ret = btrfs_truncate_inode_items(trans, log, inode, 0, 0); |
| 2848 | } |
Yan, Zheng | 4a500fd | 2010-05-16 10:49:59 -0400 | [diff] [blame] | 2849 | if (ret) { |
| 2850 | err = ret; |
| 2851 | goto out_unlock; |
| 2852 | } |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2853 | path->keep_locks = 1; |
| 2854 | |
Chris Mason | d397712 | 2009-01-05 21:25:51 -0500 | [diff] [blame] | 2855 | while (1) { |
Chris Mason | 31ff1cd | 2008-09-11 16:17:57 -0400 | [diff] [blame] | 2856 | ins_nr = 0; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2857 | ret = btrfs_search_forward(root, &min_key, &max_key, |
| 2858 | path, 0, trans->transid); |
| 2859 | if (ret != 0) |
| 2860 | break; |
Chris Mason | 3a5f1d4 | 2008-09-11 15:53:37 -0400 | [diff] [blame] | 2861 | again: |
Chris Mason | 31ff1cd | 2008-09-11 16:17:57 -0400 | [diff] [blame] | 2862 | /* note, ins_nr might be > 0 here, cleanup outside the loop */ |
Li Zefan | 33345d01 | 2011-04-20 10:31:50 +0800 | [diff] [blame] | 2863 | if (min_key.objectid != ino) |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2864 | break; |
| 2865 | if (min_key.type > max_key.type) |
| 2866 | break; |
Chris Mason | 31ff1cd | 2008-09-11 16:17:57 -0400 | [diff] [blame] | 2867 | |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2868 | src = path->nodes[0]; |
Chris Mason | 31ff1cd | 2008-09-11 16:17:57 -0400 | [diff] [blame] | 2869 | if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) { |
| 2870 | ins_nr++; |
| 2871 | goto next_slot; |
| 2872 | } else if (!ins_nr) { |
| 2873 | ins_start_slot = path->slots[0]; |
| 2874 | ins_nr = 1; |
| 2875 | goto next_slot; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2876 | } |
| 2877 | |
Chris Mason | 31ff1cd | 2008-09-11 16:17:57 -0400 | [diff] [blame] | 2878 | ret = copy_items(trans, log, dst_path, src, ins_start_slot, |
| 2879 | ins_nr, inode_only); |
Yan, Zheng | 4a500fd | 2010-05-16 10:49:59 -0400 | [diff] [blame] | 2880 | if (ret) { |
| 2881 | err = ret; |
| 2882 | goto out_unlock; |
| 2883 | } |
Chris Mason | 31ff1cd | 2008-09-11 16:17:57 -0400 | [diff] [blame] | 2884 | ins_nr = 1; |
| 2885 | ins_start_slot = path->slots[0]; |
| 2886 | next_slot: |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2887 | |
Chris Mason | 3a5f1d4 | 2008-09-11 15:53:37 -0400 | [diff] [blame] | 2888 | nritems = btrfs_header_nritems(path->nodes[0]); |
| 2889 | path->slots[0]++; |
| 2890 | if (path->slots[0] < nritems) { |
| 2891 | btrfs_item_key_to_cpu(path->nodes[0], &min_key, |
| 2892 | path->slots[0]); |
| 2893 | goto again; |
| 2894 | } |
Chris Mason | 31ff1cd | 2008-09-11 16:17:57 -0400 | [diff] [blame] | 2895 | if (ins_nr) { |
| 2896 | ret = copy_items(trans, log, dst_path, src, |
| 2897 | ins_start_slot, |
| 2898 | ins_nr, inode_only); |
Yan, Zheng | 4a500fd | 2010-05-16 10:49:59 -0400 | [diff] [blame] | 2899 | if (ret) { |
| 2900 | err = ret; |
| 2901 | goto out_unlock; |
| 2902 | } |
Chris Mason | 31ff1cd | 2008-09-11 16:17:57 -0400 | [diff] [blame] | 2903 | ins_nr = 0; |
| 2904 | } |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 2905 | btrfs_release_path(path); |
Chris Mason | 3a5f1d4 | 2008-09-11 15:53:37 -0400 | [diff] [blame] | 2906 | |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2907 | if (min_key.offset < (u64)-1) |
| 2908 | min_key.offset++; |
| 2909 | else if (min_key.type < (u8)-1) |
| 2910 | min_key.type++; |
| 2911 | else if (min_key.objectid < (u64)-1) |
| 2912 | min_key.objectid++; |
| 2913 | else |
| 2914 | break; |
| 2915 | } |
Chris Mason | 31ff1cd | 2008-09-11 16:17:57 -0400 | [diff] [blame] | 2916 | if (ins_nr) { |
| 2917 | ret = copy_items(trans, log, dst_path, src, |
| 2918 | ins_start_slot, |
| 2919 | ins_nr, inode_only); |
Yan, Zheng | 4a500fd | 2010-05-16 10:49:59 -0400 | [diff] [blame] | 2920 | if (ret) { |
| 2921 | err = ret; |
| 2922 | goto out_unlock; |
| 2923 | } |
Chris Mason | 31ff1cd | 2008-09-11 16:17:57 -0400 | [diff] [blame] | 2924 | ins_nr = 0; |
| 2925 | } |
| 2926 | WARN_ON(ins_nr); |
Chris Mason | 9623f9a | 2008-09-11 17:42:42 -0400 | [diff] [blame] | 2927 | if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) { |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 2928 | btrfs_release_path(path); |
| 2929 | btrfs_release_path(dst_path); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2930 | ret = log_directory_changes(trans, root, inode, path, dst_path); |
Yan, Zheng | 4a500fd | 2010-05-16 10:49:59 -0400 | [diff] [blame] | 2931 | if (ret) { |
| 2932 | err = ret; |
| 2933 | goto out_unlock; |
| 2934 | } |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2935 | } |
Chris Mason | 3a5f1d4 | 2008-09-11 15:53:37 -0400 | [diff] [blame] | 2936 | BTRFS_I(inode)->logged_trans = trans->transid; |
Yan, Zheng | 4a500fd | 2010-05-16 10:49:59 -0400 | [diff] [blame] | 2937 | out_unlock: |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2938 | mutex_unlock(&BTRFS_I(inode)->log_mutex); |
| 2939 | |
| 2940 | btrfs_free_path(path); |
| 2941 | btrfs_free_path(dst_path); |
Yan, Zheng | 4a500fd | 2010-05-16 10:49:59 -0400 | [diff] [blame] | 2942 | return err; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2943 | } |
| 2944 | |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 2945 | /* |
| 2946 | * follow the dentry parent pointers up the chain and see if any |
| 2947 | * of the directories in it require a full commit before they can |
| 2948 | * be logged. Returns zero if nothing special needs to be done or 1 if |
| 2949 | * a full commit is required. |
| 2950 | */ |
| 2951 | static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans, |
| 2952 | struct inode *inode, |
| 2953 | struct dentry *parent, |
| 2954 | struct super_block *sb, |
| 2955 | u64 last_committed) |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2956 | { |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 2957 | int ret = 0; |
| 2958 | struct btrfs_root *root; |
Josef Bacik | 6a91221 | 2010-11-20 09:48:00 +0000 | [diff] [blame] | 2959 | struct dentry *old_parent = NULL; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2960 | |
Chris Mason | af4176b | 2009-03-24 10:24:31 -0400 | [diff] [blame] | 2961 | /* |
| 2962 | * for regular files, if its inode is already on disk, we don't |
| 2963 | * have to worry about the parents at all. This is because |
| 2964 | * we can use the last_unlink_trans field to record renames |
| 2965 | * and other fun in this file. |
| 2966 | */ |
| 2967 | if (S_ISREG(inode->i_mode) && |
| 2968 | BTRFS_I(inode)->generation <= last_committed && |
| 2969 | BTRFS_I(inode)->last_unlink_trans <= last_committed) |
| 2970 | goto out; |
| 2971 | |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 2972 | if (!S_ISDIR(inode->i_mode)) { |
| 2973 | if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb) |
| 2974 | goto out; |
| 2975 | inode = parent->d_inode; |
| 2976 | } |
| 2977 | |
| 2978 | while (1) { |
| 2979 | BTRFS_I(inode)->logged_trans = trans->transid; |
| 2980 | smp_mb(); |
| 2981 | |
| 2982 | if (BTRFS_I(inode)->last_unlink_trans > last_committed) { |
| 2983 | root = BTRFS_I(inode)->root; |
| 2984 | |
| 2985 | /* |
| 2986 | * make sure any commits to the log are forced |
| 2987 | * to be full commits |
| 2988 | */ |
| 2989 | root->fs_info->last_trans_log_full_commit = |
| 2990 | trans->transid; |
| 2991 | ret = 1; |
| 2992 | break; |
| 2993 | } |
| 2994 | |
| 2995 | if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb) |
| 2996 | break; |
| 2997 | |
Yan, Zheng | 76dda93 | 2009-09-21 16:00:26 -0400 | [diff] [blame] | 2998 | if (IS_ROOT(parent)) |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 2999 | break; |
| 3000 | |
Josef Bacik | 6a91221 | 2010-11-20 09:48:00 +0000 | [diff] [blame] | 3001 | parent = dget_parent(parent); |
| 3002 | dput(old_parent); |
| 3003 | old_parent = parent; |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 3004 | inode = parent->d_inode; |
| 3005 | |
| 3006 | } |
Josef Bacik | 6a91221 | 2010-11-20 09:48:00 +0000 | [diff] [blame] | 3007 | dput(old_parent); |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 3008 | out: |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 3009 | return ret; |
| 3010 | } |
| 3011 | |
Chris Mason | 257c62e | 2009-10-13 13:21:08 -0400 | [diff] [blame] | 3012 | static int inode_in_log(struct btrfs_trans_handle *trans, |
| 3013 | struct inode *inode) |
| 3014 | { |
| 3015 | struct btrfs_root *root = BTRFS_I(inode)->root; |
| 3016 | int ret = 0; |
| 3017 | |
| 3018 | mutex_lock(&root->log_mutex); |
| 3019 | if (BTRFS_I(inode)->logged_trans == trans->transid && |
| 3020 | BTRFS_I(inode)->last_sub_trans <= root->last_log_commit) |
| 3021 | ret = 1; |
| 3022 | mutex_unlock(&root->log_mutex); |
| 3023 | return ret; |
| 3024 | } |
| 3025 | |
| 3026 | |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 3027 | /* |
| 3028 | * helper function around btrfs_log_inode to make sure newly created |
| 3029 | * parent directories also end up in the log. A minimal inode and backref |
| 3030 | * only logging is done of any parent directories that are older than |
| 3031 | * the last committed transaction |
| 3032 | */ |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 3033 | int btrfs_log_inode_parent(struct btrfs_trans_handle *trans, |
| 3034 | struct btrfs_root *root, struct inode *inode, |
| 3035 | struct dentry *parent, int exists_only) |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 3036 | { |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 3037 | int inode_only = exists_only ? LOG_INODE_EXISTS : LOG_INODE_ALL; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 3038 | struct super_block *sb; |
Josef Bacik | 6a91221 | 2010-11-20 09:48:00 +0000 | [diff] [blame] | 3039 | struct dentry *old_parent = NULL; |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 3040 | int ret = 0; |
| 3041 | u64 last_committed = root->fs_info->last_trans_committed; |
| 3042 | |
| 3043 | sb = inode->i_sb; |
| 3044 | |
Sage Weil | 3a5e140 | 2009-04-02 16:49:40 -0400 | [diff] [blame] | 3045 | if (btrfs_test_opt(root, NOTREELOG)) { |
| 3046 | ret = 1; |
| 3047 | goto end_no_trans; |
| 3048 | } |
| 3049 | |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 3050 | if (root->fs_info->last_trans_log_full_commit > |
| 3051 | root->fs_info->last_trans_committed) { |
| 3052 | ret = 1; |
| 3053 | goto end_no_trans; |
| 3054 | } |
| 3055 | |
Yan, Zheng | 76dda93 | 2009-09-21 16:00:26 -0400 | [diff] [blame] | 3056 | if (root != BTRFS_I(inode)->root || |
| 3057 | btrfs_root_refs(&root->root_item) == 0) { |
| 3058 | ret = 1; |
| 3059 | goto end_no_trans; |
| 3060 | } |
| 3061 | |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 3062 | ret = check_parent_dirs_for_sync(trans, inode, parent, |
| 3063 | sb, last_committed); |
| 3064 | if (ret) |
| 3065 | goto end_no_trans; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 3066 | |
Chris Mason | 257c62e | 2009-10-13 13:21:08 -0400 | [diff] [blame] | 3067 | if (inode_in_log(trans, inode)) { |
| 3068 | ret = BTRFS_NO_LOG_SYNC; |
| 3069 | goto end_no_trans; |
| 3070 | } |
| 3071 | |
Yan, Zheng | 4a500fd | 2010-05-16 10:49:59 -0400 | [diff] [blame] | 3072 | ret = start_log_trans(trans, root); |
| 3073 | if (ret) |
| 3074 | goto end_trans; |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 3075 | |
| 3076 | ret = btrfs_log_inode(trans, root, inode, inode_only); |
Yan, Zheng | 4a500fd | 2010-05-16 10:49:59 -0400 | [diff] [blame] | 3077 | if (ret) |
| 3078 | goto end_trans; |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 3079 | |
Chris Mason | af4176b | 2009-03-24 10:24:31 -0400 | [diff] [blame] | 3080 | /* |
| 3081 | * for regular files, if its inode is already on disk, we don't |
| 3082 | * have to worry about the parents at all. This is because |
| 3083 | * we can use the last_unlink_trans field to record renames |
| 3084 | * and other fun in this file. |
| 3085 | */ |
| 3086 | if (S_ISREG(inode->i_mode) && |
| 3087 | BTRFS_I(inode)->generation <= last_committed && |
Yan, Zheng | 4a500fd | 2010-05-16 10:49:59 -0400 | [diff] [blame] | 3088 | BTRFS_I(inode)->last_unlink_trans <= last_committed) { |
| 3089 | ret = 0; |
| 3090 | goto end_trans; |
| 3091 | } |
Chris Mason | af4176b | 2009-03-24 10:24:31 -0400 | [diff] [blame] | 3092 | |
| 3093 | inode_only = LOG_INODE_EXISTS; |
Chris Mason | d397712 | 2009-01-05 21:25:51 -0500 | [diff] [blame] | 3094 | while (1) { |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 3095 | if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb) |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 3096 | break; |
| 3097 | |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 3098 | inode = parent->d_inode; |
Yan, Zheng | 76dda93 | 2009-09-21 16:00:26 -0400 | [diff] [blame] | 3099 | if (root != BTRFS_I(inode)->root) |
| 3100 | break; |
| 3101 | |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 3102 | if (BTRFS_I(inode)->generation > |
| 3103 | root->fs_info->last_trans_committed) { |
| 3104 | ret = btrfs_log_inode(trans, root, inode, inode_only); |
Yan, Zheng | 4a500fd | 2010-05-16 10:49:59 -0400 | [diff] [blame] | 3105 | if (ret) |
| 3106 | goto end_trans; |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 3107 | } |
Yan, Zheng | 76dda93 | 2009-09-21 16:00:26 -0400 | [diff] [blame] | 3108 | if (IS_ROOT(parent)) |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 3109 | break; |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 3110 | |
Josef Bacik | 6a91221 | 2010-11-20 09:48:00 +0000 | [diff] [blame] | 3111 | parent = dget_parent(parent); |
| 3112 | dput(old_parent); |
| 3113 | old_parent = parent; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 3114 | } |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 3115 | ret = 0; |
Yan, Zheng | 4a500fd | 2010-05-16 10:49:59 -0400 | [diff] [blame] | 3116 | end_trans: |
Josef Bacik | 6a91221 | 2010-11-20 09:48:00 +0000 | [diff] [blame] | 3117 | dput(old_parent); |
Yan, Zheng | 4a500fd | 2010-05-16 10:49:59 -0400 | [diff] [blame] | 3118 | if (ret < 0) { |
| 3119 | BUG_ON(ret != -ENOSPC); |
| 3120 | root->fs_info->last_trans_log_full_commit = trans->transid; |
| 3121 | ret = 1; |
| 3122 | } |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 3123 | btrfs_end_log_trans(root); |
| 3124 | end_no_trans: |
| 3125 | return ret; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 3126 | } |
| 3127 | |
| 3128 | /* |
| 3129 | * it is not safe to log dentry if the chunk root has added new |
| 3130 | * chunks. This returns 0 if the dentry was logged, and 1 otherwise. |
| 3131 | * If this returns 1, you must commit the transaction to safely get your |
| 3132 | * data on disk. |
| 3133 | */ |
| 3134 | int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans, |
| 3135 | struct btrfs_root *root, struct dentry *dentry) |
| 3136 | { |
Josef Bacik | 6a91221 | 2010-11-20 09:48:00 +0000 | [diff] [blame] | 3137 | struct dentry *parent = dget_parent(dentry); |
| 3138 | int ret; |
| 3139 | |
| 3140 | ret = btrfs_log_inode_parent(trans, root, dentry->d_inode, parent, 0); |
| 3141 | dput(parent); |
| 3142 | |
| 3143 | return ret; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 3144 | } |
| 3145 | |
| 3146 | /* |
| 3147 | * should be called during mount to recover any replay any log trees |
| 3148 | * from the FS |
| 3149 | */ |
| 3150 | int btrfs_recover_log_trees(struct btrfs_root *log_root_tree) |
| 3151 | { |
| 3152 | int ret; |
| 3153 | struct btrfs_path *path; |
| 3154 | struct btrfs_trans_handle *trans; |
| 3155 | struct btrfs_key key; |
| 3156 | struct btrfs_key found_key; |
| 3157 | struct btrfs_key tmp_key; |
| 3158 | struct btrfs_root *log; |
| 3159 | struct btrfs_fs_info *fs_info = log_root_tree->fs_info; |
| 3160 | struct walk_control wc = { |
| 3161 | .process_func = process_one_buffer, |
| 3162 | .stage = 0, |
| 3163 | }; |
| 3164 | |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 3165 | path = btrfs_alloc_path(); |
Tsutomu Itoh | db5b493 | 2011-03-23 08:14:16 +0000 | [diff] [blame] | 3166 | if (!path) |
| 3167 | return -ENOMEM; |
| 3168 | |
| 3169 | fs_info->log_root_recovering = 1; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 3170 | |
Yan, Zheng | 4a500fd | 2010-05-16 10:49:59 -0400 | [diff] [blame] | 3171 | trans = btrfs_start_transaction(fs_info->tree_root, 0); |
Tsutomu Itoh | 98d5dc1 | 2011-01-20 06:19:37 +0000 | [diff] [blame] | 3172 | BUG_ON(IS_ERR(trans)); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 3173 | |
| 3174 | wc.trans = trans; |
| 3175 | wc.pin = 1; |
| 3176 | |
Tsutomu Itoh | db5b493 | 2011-03-23 08:14:16 +0000 | [diff] [blame] | 3177 | ret = walk_log_tree(trans, log_root_tree, &wc); |
| 3178 | BUG_ON(ret); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 3179 | |
| 3180 | again: |
| 3181 | key.objectid = BTRFS_TREE_LOG_OBJECTID; |
| 3182 | key.offset = (u64)-1; |
| 3183 | btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY); |
| 3184 | |
Chris Mason | d397712 | 2009-01-05 21:25:51 -0500 | [diff] [blame] | 3185 | while (1) { |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 3186 | ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0); |
| 3187 | if (ret < 0) |
| 3188 | break; |
| 3189 | if (ret > 0) { |
| 3190 | if (path->slots[0] == 0) |
| 3191 | break; |
| 3192 | path->slots[0]--; |
| 3193 | } |
| 3194 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, |
| 3195 | path->slots[0]); |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 3196 | btrfs_release_path(path); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 3197 | if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID) |
| 3198 | break; |
| 3199 | |
| 3200 | log = btrfs_read_fs_root_no_radix(log_root_tree, |
| 3201 | &found_key); |
Tsutomu Itoh | db5b493 | 2011-03-23 08:14:16 +0000 | [diff] [blame] | 3202 | BUG_ON(IS_ERR(log)); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 3203 | |
| 3204 | tmp_key.objectid = found_key.offset; |
| 3205 | tmp_key.type = BTRFS_ROOT_ITEM_KEY; |
| 3206 | tmp_key.offset = (u64)-1; |
| 3207 | |
| 3208 | wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key); |
David Sterba | 3ed4498 | 2011-06-13 17:54:22 +0000 | [diff] [blame] | 3209 | BUG_ON(IS_ERR_OR_NULL(wc.replay_dest)); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 3210 | |
Yan Zheng | 07d400a | 2009-01-06 11:42:00 -0500 | [diff] [blame] | 3211 | wc.replay_dest->log_root = log; |
Yan Zheng | 5d4f98a | 2009-06-10 10:45:14 -0400 | [diff] [blame] | 3212 | btrfs_record_root_in_trans(trans, wc.replay_dest); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 3213 | ret = walk_log_tree(trans, log, &wc); |
| 3214 | BUG_ON(ret); |
| 3215 | |
| 3216 | if (wc.stage == LOG_WALK_REPLAY_ALL) { |
| 3217 | ret = fixup_inode_link_counts(trans, wc.replay_dest, |
| 3218 | path); |
| 3219 | BUG_ON(ret); |
| 3220 | } |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 3221 | |
| 3222 | key.offset = found_key.offset - 1; |
Yan Zheng | 07d400a | 2009-01-06 11:42:00 -0500 | [diff] [blame] | 3223 | wc.replay_dest->log_root = NULL; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 3224 | free_extent_buffer(log->node); |
Chris Mason | b263c2c | 2009-06-11 11:24:47 -0400 | [diff] [blame] | 3225 | free_extent_buffer(log->commit_root); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 3226 | kfree(log); |
| 3227 | |
| 3228 | if (found_key.offset == 0) |
| 3229 | break; |
| 3230 | } |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 3231 | btrfs_release_path(path); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 3232 | |
| 3233 | /* step one is to pin it all, step two is to replay just inodes */ |
| 3234 | if (wc.pin) { |
| 3235 | wc.pin = 0; |
| 3236 | wc.process_func = replay_one_buffer; |
| 3237 | wc.stage = LOG_WALK_REPLAY_INODES; |
| 3238 | goto again; |
| 3239 | } |
| 3240 | /* step three is to replay everything */ |
| 3241 | if (wc.stage < LOG_WALK_REPLAY_ALL) { |
| 3242 | wc.stage++; |
| 3243 | goto again; |
| 3244 | } |
| 3245 | |
| 3246 | btrfs_free_path(path); |
| 3247 | |
| 3248 | free_extent_buffer(log_root_tree->node); |
| 3249 | log_root_tree->log_root = NULL; |
| 3250 | fs_info->log_root_recovering = 0; |
| 3251 | |
| 3252 | /* step 4: commit the transaction, which also unpins the blocks */ |
| 3253 | btrfs_commit_transaction(trans, fs_info->tree_root); |
| 3254 | |
| 3255 | kfree(log_root_tree); |
| 3256 | return 0; |
| 3257 | } |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 3258 | |
| 3259 | /* |
| 3260 | * there are some corner cases where we want to force a full |
| 3261 | * commit instead of allowing a directory to be logged. |
| 3262 | * |
| 3263 | * They revolve around files there were unlinked from the directory, and |
| 3264 | * this function updates the parent directory so that a full commit is |
| 3265 | * properly done if it is fsync'd later after the unlinks are done. |
| 3266 | */ |
| 3267 | void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans, |
| 3268 | struct inode *dir, struct inode *inode, |
| 3269 | int for_rename) |
| 3270 | { |
| 3271 | /* |
Chris Mason | af4176b | 2009-03-24 10:24:31 -0400 | [diff] [blame] | 3272 | * when we're logging a file, if it hasn't been renamed |
| 3273 | * or unlinked, and its inode is fully committed on disk, |
| 3274 | * we don't have to worry about walking up the directory chain |
| 3275 | * to log its parents. |
| 3276 | * |
| 3277 | * So, we use the last_unlink_trans field to put this transid |
| 3278 | * into the file. When the file is logged we check it and |
| 3279 | * don't log the parents if the file is fully on disk. |
| 3280 | */ |
| 3281 | if (S_ISREG(inode->i_mode)) |
| 3282 | BTRFS_I(inode)->last_unlink_trans = trans->transid; |
| 3283 | |
| 3284 | /* |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 3285 | * if this directory was already logged any new |
| 3286 | * names for this file/dir will get recorded |
| 3287 | */ |
| 3288 | smp_mb(); |
| 3289 | if (BTRFS_I(dir)->logged_trans == trans->transid) |
| 3290 | return; |
| 3291 | |
| 3292 | /* |
| 3293 | * if the inode we're about to unlink was logged, |
| 3294 | * the log will be properly updated for any new names |
| 3295 | */ |
| 3296 | if (BTRFS_I(inode)->logged_trans == trans->transid) |
| 3297 | return; |
| 3298 | |
| 3299 | /* |
| 3300 | * when renaming files across directories, if the directory |
| 3301 | * there we're unlinking from gets fsync'd later on, there's |
| 3302 | * no way to find the destination directory later and fsync it |
| 3303 | * properly. So, we have to be conservative and force commits |
| 3304 | * so the new name gets discovered. |
| 3305 | */ |
| 3306 | if (for_rename) |
| 3307 | goto record; |
| 3308 | |
| 3309 | /* we can safely do the unlink without any special recording */ |
| 3310 | return; |
| 3311 | |
| 3312 | record: |
| 3313 | BTRFS_I(dir)->last_unlink_trans = trans->transid; |
| 3314 | } |
| 3315 | |
| 3316 | /* |
| 3317 | * Call this after adding a new name for a file and it will properly |
| 3318 | * update the log to reflect the new name. |
| 3319 | * |
| 3320 | * It will return zero if all goes well, and it will return 1 if a |
| 3321 | * full transaction commit is required. |
| 3322 | */ |
| 3323 | int btrfs_log_new_name(struct btrfs_trans_handle *trans, |
| 3324 | struct inode *inode, struct inode *old_dir, |
| 3325 | struct dentry *parent) |
| 3326 | { |
| 3327 | struct btrfs_root * root = BTRFS_I(inode)->root; |
| 3328 | |
| 3329 | /* |
Chris Mason | af4176b | 2009-03-24 10:24:31 -0400 | [diff] [blame] | 3330 | * this will force the logging code to walk the dentry chain |
| 3331 | * up for the file |
| 3332 | */ |
| 3333 | if (S_ISREG(inode->i_mode)) |
| 3334 | BTRFS_I(inode)->last_unlink_trans = trans->transid; |
| 3335 | |
| 3336 | /* |
Chris Mason | 12fcfd2 | 2009-03-24 10:24:20 -0400 | [diff] [blame] | 3337 | * if this inode hasn't been logged and directory we're renaming it |
| 3338 | * from hasn't been logged, we don't need to log it |
| 3339 | */ |
| 3340 | if (BTRFS_I(inode)->logged_trans <= |
| 3341 | root->fs_info->last_trans_committed && |
| 3342 | (!old_dir || BTRFS_I(old_dir)->logged_trans <= |
| 3343 | root->fs_info->last_trans_committed)) |
| 3344 | return 0; |
| 3345 | |
| 3346 | return btrfs_log_inode_parent(trans, root, inode, parent, 1); |
| 3347 | } |
| 3348 | |