Commit | Line | Data |
---|---|---|
ccd979bd MF |
1 | /* -*- mode: c; c-basic-offset: 8; -*- |
2 | * vim: noexpandtab sw=8 ts=8 sts=0: | |
3 | * | |
4 | * journal.c | |
5 | * | |
6 | * Defines functions of journalling api | |
7 | * | |
8 | * Copyright (C) 2003, 2004 Oracle. All rights reserved. | |
9 | * | |
10 | * This program is free software; you can redistribute it and/or | |
11 | * modify it under the terms of the GNU General Public | |
12 | * License as published by the Free Software Foundation; either | |
13 | * version 2 of the License, or (at your option) any later version. | |
14 | * | |
15 | * This program is distributed in the hope that it will be useful, | |
16 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
17 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
18 | * General Public License for more details. | |
19 | * | |
20 | * You should have received a copy of the GNU General Public | |
21 | * License along with this program; if not, write to the | |
22 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
23 | * Boston, MA 021110-1307, USA. | |
24 | */ | |
25 | ||
26 | #include <linux/fs.h> | |
27 | #include <linux/types.h> | |
28 | #include <linux/slab.h> | |
29 | #include <linux/highmem.h> | |
30 | #include <linux/kthread.h> | |
31 | ||
32 | #define MLOG_MASK_PREFIX ML_JOURNAL | |
33 | #include <cluster/masklog.h> | |
34 | ||
35 | #include "ocfs2.h" | |
36 | ||
37 | #include "alloc.h" | |
316f4b9f | 38 | #include "dir.h" |
ccd979bd MF |
39 | #include "dlmglue.h" |
40 | #include "extent_map.h" | |
41 | #include "heartbeat.h" | |
42 | #include "inode.h" | |
43 | #include "journal.h" | |
44 | #include "localalloc.h" | |
ccd979bd MF |
45 | #include "slot_map.h" |
46 | #include "super.h" | |
47 | #include "vote.h" | |
48 | #include "sysfile.h" | |
49 | ||
50 | #include "buffer_head_io.h" | |
51 | ||
34af946a | 52 | DEFINE_SPINLOCK(trans_inc_lock); |
ccd979bd MF |
53 | |
54 | static int ocfs2_force_read_journal(struct inode *inode); | |
55 | static int ocfs2_recover_node(struct ocfs2_super *osb, | |
56 | int node_num); | |
57 | static int __ocfs2_recovery_thread(void *arg); | |
58 | static int ocfs2_commit_cache(struct ocfs2_super *osb); | |
59 | static int ocfs2_wait_on_mount(struct ocfs2_super *osb); | |
ccd979bd MF |
60 | static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb, |
61 | int dirty); | |
62 | static int ocfs2_trylock_journal(struct ocfs2_super *osb, | |
63 | int slot_num); | |
64 | static int ocfs2_recover_orphans(struct ocfs2_super *osb, | |
65 | int slot); | |
66 | static int ocfs2_commit_thread(void *arg); | |
67 | ||
68 | static int ocfs2_commit_cache(struct ocfs2_super *osb) | |
69 | { | |
70 | int status = 0; | |
71 | unsigned int flushed; | |
72 | unsigned long old_id; | |
73 | struct ocfs2_journal *journal = NULL; | |
74 | ||
75 | mlog_entry_void(); | |
76 | ||
77 | journal = osb->journal; | |
78 | ||
79 | /* Flush all pending commits and checkpoint the journal. */ | |
80 | down_write(&journal->j_trans_barrier); | |
81 | ||
82 | if (atomic_read(&journal->j_num_trans) == 0) { | |
83 | up_write(&journal->j_trans_barrier); | |
84 | mlog(0, "No transactions for me to flush!\n"); | |
85 | goto finally; | |
86 | } | |
87 | ||
88 | journal_lock_updates(journal->j_journal); | |
89 | status = journal_flush(journal->j_journal); | |
90 | journal_unlock_updates(journal->j_journal); | |
91 | if (status < 0) { | |
92 | up_write(&journal->j_trans_barrier); | |
93 | mlog_errno(status); | |
94 | goto finally; | |
95 | } | |
96 | ||
97 | old_id = ocfs2_inc_trans_id(journal); | |
98 | ||
99 | flushed = atomic_read(&journal->j_num_trans); | |
100 | atomic_set(&journal->j_num_trans, 0); | |
101 | up_write(&journal->j_trans_barrier); | |
102 | ||
103 | mlog(0, "commit_thread: flushed transaction %lu (%u handles)\n", | |
104 | journal->j_trans_id, flushed); | |
105 | ||
106 | ocfs2_kick_vote_thread(osb); | |
107 | wake_up(&journal->j_checkpointed); | |
108 | finally: | |
109 | mlog_exit(status); | |
110 | return status; | |
111 | } | |
112 | ||
ccd979bd MF |
113 | /* pass it NULL and it will allocate a new handle object for you. If |
114 | * you pass it a handle however, it may still return error, in which | |
115 | * case it has free'd the passed handle for you. */ | |
1fabe148 | 116 | handle_t *ocfs2_start_trans(struct ocfs2_super *osb, int max_buffs) |
ccd979bd | 117 | { |
ccd979bd | 118 | journal_t *journal = osb->journal->j_journal; |
1fabe148 | 119 | handle_t *handle; |
ccd979bd | 120 | |
ebdec83b | 121 | BUG_ON(!osb || !osb->journal->j_journal); |
ccd979bd | 122 | |
65eff9cc MF |
123 | if (ocfs2_is_hard_readonly(osb)) |
124 | return ERR_PTR(-EROFS); | |
ccd979bd MF |
125 | |
126 | BUG_ON(osb->journal->j_state == OCFS2_JOURNAL_FREE); | |
127 | BUG_ON(max_buffs <= 0); | |
128 | ||
129 | /* JBD might support this, but our journalling code doesn't yet. */ | |
130 | if (journal_current_handle()) { | |
131 | mlog(ML_ERROR, "Recursive transaction attempted!\n"); | |
132 | BUG(); | |
133 | } | |
134 | ||
ccd979bd MF |
135 | down_read(&osb->journal->j_trans_barrier); |
136 | ||
1fabe148 MF |
137 | handle = journal_start(journal, max_buffs); |
138 | if (IS_ERR(handle)) { | |
ccd979bd MF |
139 | up_read(&osb->journal->j_trans_barrier); |
140 | ||
1fabe148 | 141 | mlog_errno(PTR_ERR(handle)); |
ccd979bd MF |
142 | |
143 | if (is_journal_aborted(journal)) { | |
144 | ocfs2_abort(osb->sb, "Detected aborted journal"); | |
1fabe148 | 145 | handle = ERR_PTR(-EROFS); |
ccd979bd | 146 | } |
c271c5c2 SM |
147 | } else { |
148 | if (!ocfs2_mount_local(osb)) | |
149 | atomic_inc(&(osb->journal->j_num_trans)); | |
150 | } | |
ccd979bd | 151 | |
ccd979bd | 152 | return handle; |
ccd979bd MF |
153 | } |
154 | ||
1fabe148 MF |
155 | int ocfs2_commit_trans(struct ocfs2_super *osb, |
156 | handle_t *handle) | |
ccd979bd | 157 | { |
1fabe148 | 158 | int ret; |
02dc1af4 | 159 | struct ocfs2_journal *journal = osb->journal; |
ccd979bd MF |
160 | |
161 | BUG_ON(!handle); | |
162 | ||
1fabe148 MF |
163 | ret = journal_stop(handle); |
164 | if (ret < 0) | |
165 | mlog_errno(ret); | |
ccd979bd MF |
166 | |
167 | up_read(&journal->j_trans_barrier); | |
168 | ||
1fabe148 | 169 | return ret; |
ccd979bd MF |
170 | } |
171 | ||
172 | /* | |
173 | * 'nblocks' is what you want to add to the current | |
174 | * transaction. extend_trans will either extend the current handle by | |
175 | * nblocks, or commit it and start a new one with nblocks credits. | |
176 | * | |
e8aed345 MF |
177 | * This might call journal_restart() which will commit dirty buffers |
178 | * and then restart the transaction. Before calling | |
179 | * ocfs2_extend_trans(), any changed blocks should have been | |
180 | * dirtied. After calling it, all blocks which need to be changed must | |
181 | * go through another set of journal_access/journal_dirty calls. | |
182 | * | |
ccd979bd MF |
183 | * WARNING: This will not release any semaphores or disk locks taken |
184 | * during the transaction, so make sure they were taken *before* | |
185 | * start_trans or we'll have ordering deadlocks. | |
186 | * | |
187 | * WARNING2: Note that we do *not* drop j_trans_barrier here. This is | |
188 | * good because transaction ids haven't yet been recorded on the | |
189 | * cluster locks associated with this handle. | |
190 | */ | |
1fc58146 | 191 | int ocfs2_extend_trans(handle_t *handle, int nblocks) |
ccd979bd MF |
192 | { |
193 | int status; | |
194 | ||
195 | BUG_ON(!handle); | |
ccd979bd MF |
196 | BUG_ON(!nblocks); |
197 | ||
198 | mlog_entry_void(); | |
199 | ||
200 | mlog(0, "Trying to extend transaction by %d blocks\n", nblocks); | |
201 | ||
0879c584 MF |
202 | #ifdef OCFS2_DEBUG_FS |
203 | status = 1; | |
204 | #else | |
1fc58146 | 205 | status = journal_extend(handle, nblocks); |
ccd979bd MF |
206 | if (status < 0) { |
207 | mlog_errno(status); | |
208 | goto bail; | |
209 | } | |
0879c584 | 210 | #endif |
ccd979bd MF |
211 | |
212 | if (status > 0) { | |
213 | mlog(0, "journal_extend failed, trying journal_restart\n"); | |
1fc58146 | 214 | status = journal_restart(handle, nblocks); |
ccd979bd | 215 | if (status < 0) { |
ccd979bd MF |
216 | mlog_errno(status); |
217 | goto bail; | |
218 | } | |
01ddf1e1 | 219 | } |
ccd979bd MF |
220 | |
221 | status = 0; | |
222 | bail: | |
223 | ||
224 | mlog_exit(status); | |
225 | return status; | |
226 | } | |
227 | ||
1fabe148 | 228 | int ocfs2_journal_access(handle_t *handle, |
ccd979bd MF |
229 | struct inode *inode, |
230 | struct buffer_head *bh, | |
231 | int type) | |
232 | { | |
233 | int status; | |
234 | ||
235 | BUG_ON(!inode); | |
236 | BUG_ON(!handle); | |
237 | BUG_ON(!bh); | |
ccd979bd | 238 | |
205f87f6 | 239 | mlog_entry("bh->b_blocknr=%llu, type=%d (\"%s\"), bh->b_size = %zu\n", |
ccd979bd MF |
240 | (unsigned long long)bh->b_blocknr, type, |
241 | (type == OCFS2_JOURNAL_ACCESS_CREATE) ? | |
242 | "OCFS2_JOURNAL_ACCESS_CREATE" : | |
243 | "OCFS2_JOURNAL_ACCESS_WRITE", | |
244 | bh->b_size); | |
245 | ||
246 | /* we can safely remove this assertion after testing. */ | |
247 | if (!buffer_uptodate(bh)) { | |
248 | mlog(ML_ERROR, "giving me a buffer that's not uptodate!\n"); | |
249 | mlog(ML_ERROR, "b_blocknr=%llu\n", | |
250 | (unsigned long long)bh->b_blocknr); | |
251 | BUG(); | |
252 | } | |
253 | ||
254 | /* Set the current transaction information on the inode so | |
255 | * that the locking code knows whether it can drop it's locks | |
256 | * on this inode or not. We're protected from the commit | |
257 | * thread updating the current transaction id until | |
258 | * ocfs2_commit_trans() because ocfs2_start_trans() took | |
259 | * j_trans_barrier for us. */ | |
260 | ocfs2_set_inode_lock_trans(OCFS2_SB(inode->i_sb)->journal, inode); | |
261 | ||
251b6ecc | 262 | mutex_lock(&OCFS2_I(inode)->ip_io_mutex); |
ccd979bd MF |
263 | switch (type) { |
264 | case OCFS2_JOURNAL_ACCESS_CREATE: | |
265 | case OCFS2_JOURNAL_ACCESS_WRITE: | |
1fabe148 | 266 | status = journal_get_write_access(handle, bh); |
ccd979bd MF |
267 | break; |
268 | ||
269 | case OCFS2_JOURNAL_ACCESS_UNDO: | |
1fabe148 | 270 | status = journal_get_undo_access(handle, bh); |
ccd979bd MF |
271 | break; |
272 | ||
273 | default: | |
274 | status = -EINVAL; | |
275 | mlog(ML_ERROR, "Uknown access type!\n"); | |
276 | } | |
251b6ecc | 277 | mutex_unlock(&OCFS2_I(inode)->ip_io_mutex); |
ccd979bd MF |
278 | |
279 | if (status < 0) | |
280 | mlog(ML_ERROR, "Error %d getting %d access to buffer!\n", | |
281 | status, type); | |
282 | ||
283 | mlog_exit(status); | |
284 | return status; | |
285 | } | |
286 | ||
1fabe148 | 287 | int ocfs2_journal_dirty(handle_t *handle, |
ccd979bd MF |
288 | struct buffer_head *bh) |
289 | { | |
290 | int status; | |
291 | ||
ccd979bd MF |
292 | mlog_entry("(bh->b_blocknr=%llu)\n", |
293 | (unsigned long long)bh->b_blocknr); | |
294 | ||
1fabe148 | 295 | status = journal_dirty_metadata(handle, bh); |
ccd979bd MF |
296 | if (status < 0) |
297 | mlog(ML_ERROR, "Could not dirty metadata buffer. " | |
298 | "(bh->b_blocknr=%llu)\n", | |
299 | (unsigned long long)bh->b_blocknr); | |
300 | ||
301 | mlog_exit(status); | |
302 | return status; | |
303 | } | |
304 | ||
305 | int ocfs2_journal_dirty_data(handle_t *handle, | |
306 | struct buffer_head *bh) | |
307 | { | |
308 | int err = journal_dirty_data(handle, bh); | |
309 | if (err) | |
310 | mlog_errno(err); | |
311 | /* TODO: When we can handle it, abort the handle and go RO on | |
312 | * error here. */ | |
313 | ||
314 | return err; | |
315 | } | |
316 | ||
ccd979bd MF |
317 | #define OCFS2_DEFAULT_COMMIT_INTERVAL (HZ * 5) |
318 | ||
319 | void ocfs2_set_journal_params(struct ocfs2_super *osb) | |
320 | { | |
321 | journal_t *journal = osb->journal->j_journal; | |
322 | ||
323 | spin_lock(&journal->j_state_lock); | |
324 | journal->j_commit_interval = OCFS2_DEFAULT_COMMIT_INTERVAL; | |
325 | if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER) | |
326 | journal->j_flags |= JFS_BARRIER; | |
327 | else | |
328 | journal->j_flags &= ~JFS_BARRIER; | |
329 | spin_unlock(&journal->j_state_lock); | |
330 | } | |
331 | ||
332 | int ocfs2_journal_init(struct ocfs2_journal *journal, int *dirty) | |
333 | { | |
334 | int status = -1; | |
335 | struct inode *inode = NULL; /* the journal inode */ | |
336 | journal_t *j_journal = NULL; | |
337 | struct ocfs2_dinode *di = NULL; | |
338 | struct buffer_head *bh = NULL; | |
339 | struct ocfs2_super *osb; | |
340 | int meta_lock = 0; | |
341 | ||
342 | mlog_entry_void(); | |
343 | ||
344 | BUG_ON(!journal); | |
345 | ||
346 | osb = journal->j_osb; | |
347 | ||
348 | /* already have the inode for our journal */ | |
349 | inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, | |
350 | osb->slot_num); | |
351 | if (inode == NULL) { | |
352 | status = -EACCES; | |
353 | mlog_errno(status); | |
354 | goto done; | |
355 | } | |
356 | if (is_bad_inode(inode)) { | |
357 | mlog(ML_ERROR, "access error (bad inode)\n"); | |
358 | iput(inode); | |
359 | inode = NULL; | |
360 | status = -EACCES; | |
361 | goto done; | |
362 | } | |
363 | ||
364 | SET_INODE_JOURNAL(inode); | |
365 | OCFS2_I(inode)->ip_open_count++; | |
366 | ||
6eff5790 MF |
367 | /* Skip recovery waits here - journal inode metadata never |
368 | * changes in a live cluster so it can be considered an | |
369 | * exception to the rule. */ | |
4bcec184 | 370 | status = ocfs2_meta_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY); |
ccd979bd MF |
371 | if (status < 0) { |
372 | if (status != -ERESTARTSYS) | |
373 | mlog(ML_ERROR, "Could not get lock on journal!\n"); | |
374 | goto done; | |
375 | } | |
376 | ||
377 | meta_lock = 1; | |
378 | di = (struct ocfs2_dinode *)bh->b_data; | |
379 | ||
380 | if (inode->i_size < OCFS2_MIN_JOURNAL_SIZE) { | |
381 | mlog(ML_ERROR, "Journal file size (%lld) is too small!\n", | |
382 | inode->i_size); | |
383 | status = -EINVAL; | |
384 | goto done; | |
385 | } | |
386 | ||
387 | mlog(0, "inode->i_size = %lld\n", inode->i_size); | |
5515eff8 AM |
388 | mlog(0, "inode->i_blocks = %llu\n", |
389 | (unsigned long long)inode->i_blocks); | |
ccd979bd MF |
390 | mlog(0, "inode->ip_clusters = %u\n", OCFS2_I(inode)->ip_clusters); |
391 | ||
392 | /* call the kernels journal init function now */ | |
393 | j_journal = journal_init_inode(inode); | |
394 | if (j_journal == NULL) { | |
395 | mlog(ML_ERROR, "Linux journal layer error\n"); | |
396 | status = -EINVAL; | |
397 | goto done; | |
398 | } | |
399 | ||
400 | mlog(0, "Returned from journal_init_inode\n"); | |
401 | mlog(0, "j_journal->j_maxlen = %u\n", j_journal->j_maxlen); | |
402 | ||
403 | *dirty = (le32_to_cpu(di->id1.journal1.ij_flags) & | |
404 | OCFS2_JOURNAL_DIRTY_FL); | |
405 | ||
406 | journal->j_journal = j_journal; | |
407 | journal->j_inode = inode; | |
408 | journal->j_bh = bh; | |
409 | ||
410 | ocfs2_set_journal_params(osb); | |
411 | ||
412 | journal->j_state = OCFS2_JOURNAL_LOADED; | |
413 | ||
414 | status = 0; | |
415 | done: | |
416 | if (status < 0) { | |
417 | if (meta_lock) | |
418 | ocfs2_meta_unlock(inode, 1); | |
419 | if (bh != NULL) | |
420 | brelse(bh); | |
421 | if (inode) { | |
422 | OCFS2_I(inode)->ip_open_count--; | |
423 | iput(inode); | |
424 | } | |
425 | } | |
426 | ||
427 | mlog_exit(status); | |
428 | return status; | |
429 | } | |
430 | ||
431 | static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb, | |
432 | int dirty) | |
433 | { | |
434 | int status; | |
435 | unsigned int flags; | |
436 | struct ocfs2_journal *journal = osb->journal; | |
437 | struct buffer_head *bh = journal->j_bh; | |
438 | struct ocfs2_dinode *fe; | |
439 | ||
440 | mlog_entry_void(); | |
441 | ||
442 | fe = (struct ocfs2_dinode *)bh->b_data; | |
443 | if (!OCFS2_IS_VALID_DINODE(fe)) { | |
444 | /* This is called from startup/shutdown which will | |
445 | * handle the errors in a specific manner, so no need | |
446 | * to call ocfs2_error() here. */ | |
b0697053 | 447 | mlog(ML_ERROR, "Journal dinode %llu has invalid " |
1ca1a111 MF |
448 | "signature: %.*s", |
449 | (unsigned long long)le64_to_cpu(fe->i_blkno), 7, | |
b0697053 | 450 | fe->i_signature); |
ccd979bd MF |
451 | status = -EIO; |
452 | goto out; | |
453 | } | |
454 | ||
455 | flags = le32_to_cpu(fe->id1.journal1.ij_flags); | |
456 | if (dirty) | |
457 | flags |= OCFS2_JOURNAL_DIRTY_FL; | |
458 | else | |
459 | flags &= ~OCFS2_JOURNAL_DIRTY_FL; | |
460 | fe->id1.journal1.ij_flags = cpu_to_le32(flags); | |
461 | ||
462 | status = ocfs2_write_block(osb, bh, journal->j_inode); | |
463 | if (status < 0) | |
464 | mlog_errno(status); | |
465 | ||
466 | out: | |
467 | mlog_exit(status); | |
468 | return status; | |
469 | } | |
470 | ||
471 | /* | |
472 | * If the journal has been kmalloc'd it needs to be freed after this | |
473 | * call. | |
474 | */ | |
475 | void ocfs2_journal_shutdown(struct ocfs2_super *osb) | |
476 | { | |
477 | struct ocfs2_journal *journal = NULL; | |
478 | int status = 0; | |
479 | struct inode *inode = NULL; | |
480 | int num_running_trans = 0; | |
481 | ||
482 | mlog_entry_void(); | |
483 | ||
ebdec83b | 484 | BUG_ON(!osb); |
ccd979bd MF |
485 | |
486 | journal = osb->journal; | |
487 | if (!journal) | |
488 | goto done; | |
489 | ||
490 | inode = journal->j_inode; | |
491 | ||
492 | if (journal->j_state != OCFS2_JOURNAL_LOADED) | |
493 | goto done; | |
494 | ||
495 | /* need to inc inode use count as journal_destroy will iput. */ | |
496 | if (!igrab(inode)) | |
497 | BUG(); | |
498 | ||
499 | num_running_trans = atomic_read(&(osb->journal->j_num_trans)); | |
500 | if (num_running_trans > 0) | |
501 | mlog(0, "Shutting down journal: must wait on %d " | |
502 | "running transactions!\n", | |
503 | num_running_trans); | |
504 | ||
505 | /* Do a commit_cache here. It will flush our journal, *and* | |
506 | * release any locks that are still held. | |
507 | * set the SHUTDOWN flag and release the trans lock. | |
508 | * the commit thread will take the trans lock for us below. */ | |
509 | journal->j_state = OCFS2_JOURNAL_IN_SHUTDOWN; | |
510 | ||
511 | /* The OCFS2_JOURNAL_IN_SHUTDOWN will signal to commit_cache to not | |
512 | * drop the trans_lock (which we want to hold until we | |
513 | * completely destroy the journal. */ | |
514 | if (osb->commit_task) { | |
515 | /* Wait for the commit thread */ | |
516 | mlog(0, "Waiting for ocfs2commit to exit....\n"); | |
517 | kthread_stop(osb->commit_task); | |
518 | osb->commit_task = NULL; | |
519 | } | |
520 | ||
521 | BUG_ON(atomic_read(&(osb->journal->j_num_trans)) != 0); | |
522 | ||
c271c5c2 SM |
523 | if (ocfs2_mount_local(osb)) { |
524 | journal_lock_updates(journal->j_journal); | |
525 | status = journal_flush(journal->j_journal); | |
526 | journal_unlock_updates(journal->j_journal); | |
527 | if (status < 0) | |
528 | mlog_errno(status); | |
529 | } | |
530 | ||
531 | if (status == 0) { | |
532 | /* | |
533 | * Do not toggle if flush was unsuccessful otherwise | |
534 | * will leave dirty metadata in a "clean" journal | |
535 | */ | |
536 | status = ocfs2_journal_toggle_dirty(osb, 0); | |
537 | if (status < 0) | |
538 | mlog_errno(status); | |
539 | } | |
ccd979bd MF |
540 | |
541 | /* Shutdown the kernel journal system */ | |
542 | journal_destroy(journal->j_journal); | |
543 | ||
544 | OCFS2_I(inode)->ip_open_count--; | |
545 | ||
546 | /* unlock our journal */ | |
547 | ocfs2_meta_unlock(inode, 1); | |
548 | ||
549 | brelse(journal->j_bh); | |
550 | journal->j_bh = NULL; | |
551 | ||
552 | journal->j_state = OCFS2_JOURNAL_FREE; | |
553 | ||
554 | // up_write(&journal->j_trans_barrier); | |
555 | done: | |
556 | if (inode) | |
557 | iput(inode); | |
558 | mlog_exit_void(); | |
559 | } | |
560 | ||
561 | static void ocfs2_clear_journal_error(struct super_block *sb, | |
562 | journal_t *journal, | |
563 | int slot) | |
564 | { | |
565 | int olderr; | |
566 | ||
567 | olderr = journal_errno(journal); | |
568 | if (olderr) { | |
569 | mlog(ML_ERROR, "File system error %d recorded in " | |
570 | "journal %u.\n", olderr, slot); | |
571 | mlog(ML_ERROR, "File system on device %s needs checking.\n", | |
572 | sb->s_id); | |
573 | ||
574 | journal_ack_err(journal); | |
575 | journal_clear_err(journal); | |
576 | } | |
577 | } | |
578 | ||
c271c5c2 | 579 | int ocfs2_journal_load(struct ocfs2_journal *journal, int local) |
ccd979bd MF |
580 | { |
581 | int status = 0; | |
582 | struct ocfs2_super *osb; | |
583 | ||
584 | mlog_entry_void(); | |
585 | ||
586 | if (!journal) | |
587 | BUG(); | |
588 | ||
589 | osb = journal->j_osb; | |
590 | ||
591 | status = journal_load(journal->j_journal); | |
592 | if (status < 0) { | |
593 | mlog(ML_ERROR, "Failed to load journal!\n"); | |
594 | goto done; | |
595 | } | |
596 | ||
597 | ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num); | |
598 | ||
599 | status = ocfs2_journal_toggle_dirty(osb, 1); | |
600 | if (status < 0) { | |
601 | mlog_errno(status); | |
602 | goto done; | |
603 | } | |
604 | ||
605 | /* Launch the commit thread */ | |
c271c5c2 SM |
606 | if (!local) { |
607 | osb->commit_task = kthread_run(ocfs2_commit_thread, osb, | |
608 | "ocfs2cmt"); | |
609 | if (IS_ERR(osb->commit_task)) { | |
610 | status = PTR_ERR(osb->commit_task); | |
611 | osb->commit_task = NULL; | |
612 | mlog(ML_ERROR, "unable to launch ocfs2commit thread, " | |
613 | "error=%d", status); | |
614 | goto done; | |
615 | } | |
616 | } else | |
ccd979bd | 617 | osb->commit_task = NULL; |
ccd979bd MF |
618 | |
619 | done: | |
620 | mlog_exit(status); | |
621 | return status; | |
622 | } | |
623 | ||
624 | ||
625 | /* 'full' flag tells us whether we clear out all blocks or if we just | |
626 | * mark the journal clean */ | |
627 | int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full) | |
628 | { | |
629 | int status; | |
630 | ||
631 | mlog_entry_void(); | |
632 | ||
ebdec83b | 633 | BUG_ON(!journal); |
ccd979bd MF |
634 | |
635 | status = journal_wipe(journal->j_journal, full); | |
636 | if (status < 0) { | |
637 | mlog_errno(status); | |
638 | goto bail; | |
639 | } | |
640 | ||
641 | status = ocfs2_journal_toggle_dirty(journal->j_osb, 0); | |
642 | if (status < 0) | |
643 | mlog_errno(status); | |
644 | ||
645 | bail: | |
646 | mlog_exit(status); | |
647 | return status; | |
648 | } | |
649 | ||
650 | /* | |
651 | * JBD Might read a cached version of another nodes journal file. We | |
652 | * don't want this as this file changes often and we get no | |
653 | * notification on those changes. The only way to be sure that we've | |
654 | * got the most up to date version of those blocks then is to force | |
655 | * read them off disk. Just searching through the buffer cache won't | |
656 | * work as there may be pages backing this file which are still marked | |
657 | * up to date. We know things can't change on this file underneath us | |
658 | * as we have the lock by now :) | |
659 | */ | |
660 | static int ocfs2_force_read_journal(struct inode *inode) | |
661 | { | |
662 | int status = 0; | |
4f902c37 | 663 | int i; |
8110b073 | 664 | u64 v_blkno, p_blkno, p_blocks, num_blocks; |
4f902c37 | 665 | #define CONCURRENT_JOURNAL_FILL 32ULL |
ccd979bd MF |
666 | struct buffer_head *bhs[CONCURRENT_JOURNAL_FILL]; |
667 | ||
668 | mlog_entry_void(); | |
669 | ||
ccd979bd MF |
670 | memset(bhs, 0, sizeof(struct buffer_head *) * CONCURRENT_JOURNAL_FILL); |
671 | ||
8110b073 | 672 | num_blocks = ocfs2_blocks_for_bytes(inode->i_sb, inode->i_size); |
ccd979bd | 673 | v_blkno = 0; |
8110b073 | 674 | while (v_blkno < num_blocks) { |
ccd979bd | 675 | status = ocfs2_extent_map_get_blocks(inode, v_blkno, |
49cb8d2d | 676 | &p_blkno, &p_blocks, NULL); |
ccd979bd MF |
677 | if (status < 0) { |
678 | mlog_errno(status); | |
679 | goto bail; | |
680 | } | |
681 | ||
682 | if (p_blocks > CONCURRENT_JOURNAL_FILL) | |
683 | p_blocks = CONCURRENT_JOURNAL_FILL; | |
684 | ||
dd4a2c2b MF |
685 | /* We are reading journal data which should not |
686 | * be put in the uptodate cache */ | |
ccd979bd MF |
687 | status = ocfs2_read_blocks(OCFS2_SB(inode->i_sb), |
688 | p_blkno, p_blocks, bhs, 0, | |
dd4a2c2b | 689 | NULL); |
ccd979bd MF |
690 | if (status < 0) { |
691 | mlog_errno(status); | |
692 | goto bail; | |
693 | } | |
694 | ||
695 | for(i = 0; i < p_blocks; i++) { | |
696 | brelse(bhs[i]); | |
697 | bhs[i] = NULL; | |
698 | } | |
699 | ||
700 | v_blkno += p_blocks; | |
701 | } | |
702 | ||
703 | bail: | |
704 | for(i = 0; i < CONCURRENT_JOURNAL_FILL; i++) | |
705 | if (bhs[i]) | |
706 | brelse(bhs[i]); | |
707 | mlog_exit(status); | |
708 | return status; | |
709 | } | |
710 | ||
711 | struct ocfs2_la_recovery_item { | |
712 | struct list_head lri_list; | |
713 | int lri_slot; | |
714 | struct ocfs2_dinode *lri_la_dinode; | |
715 | struct ocfs2_dinode *lri_tl_dinode; | |
716 | }; | |
717 | ||
718 | /* Does the second half of the recovery process. By this point, the | |
719 | * node is marked clean and can actually be considered recovered, | |
720 | * hence it's no longer in the recovery map, but there's still some | |
721 | * cleanup we can do which shouldn't happen within the recovery thread | |
722 | * as locking in that context becomes very difficult if we are to take | |
723 | * recovering nodes into account. | |
724 | * | |
725 | * NOTE: This function can and will sleep on recovery of other nodes | |
726 | * during cluster locking, just like any other ocfs2 process. | |
727 | */ | |
c4028958 | 728 | void ocfs2_complete_recovery(struct work_struct *work) |
ccd979bd MF |
729 | { |
730 | int ret; | |
c4028958 DH |
731 | struct ocfs2_journal *journal = |
732 | container_of(work, struct ocfs2_journal, j_recovery_work); | |
733 | struct ocfs2_super *osb = journal->j_osb; | |
ccd979bd | 734 | struct ocfs2_dinode *la_dinode, *tl_dinode; |
800deef3 | 735 | struct ocfs2_la_recovery_item *item, *n; |
ccd979bd MF |
736 | LIST_HEAD(tmp_la_list); |
737 | ||
738 | mlog_entry_void(); | |
739 | ||
740 | mlog(0, "completing recovery from keventd\n"); | |
741 | ||
742 | spin_lock(&journal->j_lock); | |
743 | list_splice_init(&journal->j_la_cleanups, &tmp_la_list); | |
744 | spin_unlock(&journal->j_lock); | |
745 | ||
800deef3 | 746 | list_for_each_entry_safe(item, n, &tmp_la_list, lri_list) { |
ccd979bd MF |
747 | list_del_init(&item->lri_list); |
748 | ||
749 | mlog(0, "Complete recovery for slot %d\n", item->lri_slot); | |
750 | ||
751 | la_dinode = item->lri_la_dinode; | |
752 | if (la_dinode) { | |
b0697053 | 753 | mlog(0, "Clean up local alloc %llu\n", |
1ca1a111 | 754 | (unsigned long long)le64_to_cpu(la_dinode->i_blkno)); |
ccd979bd MF |
755 | |
756 | ret = ocfs2_complete_local_alloc_recovery(osb, | |
757 | la_dinode); | |
758 | if (ret < 0) | |
759 | mlog_errno(ret); | |
760 | ||
761 | kfree(la_dinode); | |
762 | } | |
763 | ||
764 | tl_dinode = item->lri_tl_dinode; | |
765 | if (tl_dinode) { | |
b0697053 | 766 | mlog(0, "Clean up truncate log %llu\n", |
1ca1a111 | 767 | (unsigned long long)le64_to_cpu(tl_dinode->i_blkno)); |
ccd979bd MF |
768 | |
769 | ret = ocfs2_complete_truncate_log_recovery(osb, | |
770 | tl_dinode); | |
771 | if (ret < 0) | |
772 | mlog_errno(ret); | |
773 | ||
774 | kfree(tl_dinode); | |
775 | } | |
776 | ||
777 | ret = ocfs2_recover_orphans(osb, item->lri_slot); | |
778 | if (ret < 0) | |
779 | mlog_errno(ret); | |
780 | ||
781 | kfree(item); | |
782 | } | |
783 | ||
784 | mlog(0, "Recovery completion\n"); | |
785 | mlog_exit_void(); | |
786 | } | |
787 | ||
788 | /* NOTE: This function always eats your references to la_dinode and | |
789 | * tl_dinode, either manually on error, or by passing them to | |
790 | * ocfs2_complete_recovery */ | |
791 | static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal, | |
792 | int slot_num, | |
793 | struct ocfs2_dinode *la_dinode, | |
794 | struct ocfs2_dinode *tl_dinode) | |
795 | { | |
796 | struct ocfs2_la_recovery_item *item; | |
797 | ||
afae00ab | 798 | item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS); |
ccd979bd MF |
799 | if (!item) { |
800 | /* Though we wish to avoid it, we are in fact safe in | |
801 | * skipping local alloc cleanup as fsck.ocfs2 is more | |
802 | * than capable of reclaiming unused space. */ | |
803 | if (la_dinode) | |
804 | kfree(la_dinode); | |
805 | ||
806 | if (tl_dinode) | |
807 | kfree(tl_dinode); | |
808 | ||
809 | mlog_errno(-ENOMEM); | |
810 | return; | |
811 | } | |
812 | ||
813 | INIT_LIST_HEAD(&item->lri_list); | |
814 | item->lri_la_dinode = la_dinode; | |
815 | item->lri_slot = slot_num; | |
816 | item->lri_tl_dinode = tl_dinode; | |
817 | ||
818 | spin_lock(&journal->j_lock); | |
819 | list_add_tail(&item->lri_list, &journal->j_la_cleanups); | |
820 | queue_work(ocfs2_wq, &journal->j_recovery_work); | |
821 | spin_unlock(&journal->j_lock); | |
822 | } | |
823 | ||
824 | /* Called by the mount code to queue recovery the last part of | |
825 | * recovery for it's own slot. */ | |
826 | void ocfs2_complete_mount_recovery(struct ocfs2_super *osb) | |
827 | { | |
828 | struct ocfs2_journal *journal = osb->journal; | |
829 | ||
830 | if (osb->dirty) { | |
831 | /* No need to queue up our truncate_log as regular | |
832 | * cleanup will catch that. */ | |
833 | ocfs2_queue_recovery_completion(journal, | |
834 | osb->slot_num, | |
835 | osb->local_alloc_copy, | |
836 | NULL); | |
837 | ocfs2_schedule_truncate_log_flush(osb, 0); | |
838 | ||
839 | osb->local_alloc_copy = NULL; | |
840 | osb->dirty = 0; | |
841 | } | |
842 | } | |
843 | ||
844 | static int __ocfs2_recovery_thread(void *arg) | |
845 | { | |
846 | int status, node_num; | |
847 | struct ocfs2_super *osb = arg; | |
848 | ||
849 | mlog_entry_void(); | |
850 | ||
851 | status = ocfs2_wait_on_mount(osb); | |
852 | if (status < 0) { | |
853 | goto bail; | |
854 | } | |
855 | ||
856 | restart: | |
857 | status = ocfs2_super_lock(osb, 1); | |
858 | if (status < 0) { | |
859 | mlog_errno(status); | |
860 | goto bail; | |
861 | } | |
862 | ||
863 | while(!ocfs2_node_map_is_empty(osb, &osb->recovery_map)) { | |
864 | node_num = ocfs2_node_map_first_set_bit(osb, | |
865 | &osb->recovery_map); | |
866 | if (node_num == O2NM_INVALID_NODE_NUM) { | |
867 | mlog(0, "Out of nodes to recover.\n"); | |
868 | break; | |
869 | } | |
870 | ||
871 | status = ocfs2_recover_node(osb, node_num); | |
872 | if (status < 0) { | |
873 | mlog(ML_ERROR, | |
874 | "Error %d recovering node %d on device (%u,%u)!\n", | |
875 | status, node_num, | |
876 | MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev)); | |
877 | mlog(ML_ERROR, "Volume requires unmount.\n"); | |
878 | continue; | |
879 | } | |
880 | ||
881 | ocfs2_recovery_map_clear(osb, node_num); | |
882 | } | |
883 | ocfs2_super_unlock(osb, 1); | |
884 | ||
885 | /* We always run recovery on our own orphan dir - the dead | |
886 | * node(s) may have voted "no" on an inode delete earlier. A | |
887 | * revote is therefore required. */ | |
888 | ocfs2_queue_recovery_completion(osb->journal, osb->slot_num, NULL, | |
889 | NULL); | |
890 | ||
891 | bail: | |
c74ec2f7 | 892 | mutex_lock(&osb->recovery_lock); |
ccd979bd MF |
893 | if (!status && |
894 | !ocfs2_node_map_is_empty(osb, &osb->recovery_map)) { | |
c74ec2f7 | 895 | mutex_unlock(&osb->recovery_lock); |
ccd979bd MF |
896 | goto restart; |
897 | } | |
898 | ||
899 | osb->recovery_thread_task = NULL; | |
900 | mb(); /* sync with ocfs2_recovery_thread_running */ | |
901 | wake_up(&osb->recovery_event); | |
902 | ||
c74ec2f7 | 903 | mutex_unlock(&osb->recovery_lock); |
ccd979bd MF |
904 | |
905 | mlog_exit(status); | |
906 | /* no one is callint kthread_stop() for us so the kthread() api | |
907 | * requires that we call do_exit(). And it isn't exported, but | |
908 | * complete_and_exit() seems to be a minimal wrapper around it. */ | |
909 | complete_and_exit(NULL, status); | |
910 | return status; | |
911 | } | |
912 | ||
913 | void ocfs2_recovery_thread(struct ocfs2_super *osb, int node_num) | |
914 | { | |
915 | mlog_entry("(node_num=%d, osb->node_num = %d)\n", | |
916 | node_num, osb->node_num); | |
917 | ||
c74ec2f7 | 918 | mutex_lock(&osb->recovery_lock); |
ccd979bd MF |
919 | if (osb->disable_recovery) |
920 | goto out; | |
921 | ||
922 | /* People waiting on recovery will wait on | |
923 | * the recovery map to empty. */ | |
924 | if (!ocfs2_recovery_map_set(osb, node_num)) | |
925 | mlog(0, "node %d already be in recovery.\n", node_num); | |
926 | ||
927 | mlog(0, "starting recovery thread...\n"); | |
928 | ||
929 | if (osb->recovery_thread_task) | |
930 | goto out; | |
931 | ||
932 | osb->recovery_thread_task = kthread_run(__ocfs2_recovery_thread, osb, | |
78427043 | 933 | "ocfs2rec"); |
ccd979bd MF |
934 | if (IS_ERR(osb->recovery_thread_task)) { |
935 | mlog_errno((int)PTR_ERR(osb->recovery_thread_task)); | |
936 | osb->recovery_thread_task = NULL; | |
937 | } | |
938 | ||
939 | out: | |
c74ec2f7 | 940 | mutex_unlock(&osb->recovery_lock); |
ccd979bd MF |
941 | wake_up(&osb->recovery_event); |
942 | ||
943 | mlog_exit_void(); | |
944 | } | |
945 | ||
946 | /* Does the actual journal replay and marks the journal inode as | |
947 | * clean. Will only replay if the journal inode is marked dirty. */ | |
948 | static int ocfs2_replay_journal(struct ocfs2_super *osb, | |
949 | int node_num, | |
950 | int slot_num) | |
951 | { | |
952 | int status; | |
953 | int got_lock = 0; | |
954 | unsigned int flags; | |
955 | struct inode *inode = NULL; | |
956 | struct ocfs2_dinode *fe; | |
957 | journal_t *journal = NULL; | |
958 | struct buffer_head *bh = NULL; | |
959 | ||
960 | inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, | |
961 | slot_num); | |
962 | if (inode == NULL) { | |
963 | status = -EACCES; | |
964 | mlog_errno(status); | |
965 | goto done; | |
966 | } | |
967 | if (is_bad_inode(inode)) { | |
968 | status = -EACCES; | |
969 | iput(inode); | |
970 | inode = NULL; | |
971 | mlog_errno(status); | |
972 | goto done; | |
973 | } | |
974 | SET_INODE_JOURNAL(inode); | |
975 | ||
4bcec184 | 976 | status = ocfs2_meta_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY); |
ccd979bd MF |
977 | if (status < 0) { |
978 | mlog(0, "status returned from ocfs2_meta_lock=%d\n", status); | |
979 | if (status != -ERESTARTSYS) | |
980 | mlog(ML_ERROR, "Could not lock journal!\n"); | |
981 | goto done; | |
982 | } | |
983 | got_lock = 1; | |
984 | ||
985 | fe = (struct ocfs2_dinode *) bh->b_data; | |
986 | ||
987 | flags = le32_to_cpu(fe->id1.journal1.ij_flags); | |
988 | ||
989 | if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) { | |
990 | mlog(0, "No recovery required for node %d\n", node_num); | |
991 | goto done; | |
992 | } | |
993 | ||
994 | mlog(ML_NOTICE, "Recovering node %d from slot %d on device (%u,%u)\n", | |
995 | node_num, slot_num, | |
996 | MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev)); | |
997 | ||
998 | OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters); | |
999 | ||
1000 | status = ocfs2_force_read_journal(inode); | |
1001 | if (status < 0) { | |
1002 | mlog_errno(status); | |
1003 | goto done; | |
1004 | } | |
1005 | ||
1006 | mlog(0, "calling journal_init_inode\n"); | |
1007 | journal = journal_init_inode(inode); | |
1008 | if (journal == NULL) { | |
1009 | mlog(ML_ERROR, "Linux journal layer error\n"); | |
1010 | status = -EIO; | |
1011 | goto done; | |
1012 | } | |
1013 | ||
1014 | status = journal_load(journal); | |
1015 | if (status < 0) { | |
1016 | mlog_errno(status); | |
1017 | if (!igrab(inode)) | |
1018 | BUG(); | |
1019 | journal_destroy(journal); | |
1020 | goto done; | |
1021 | } | |
1022 | ||
1023 | ocfs2_clear_journal_error(osb->sb, journal, slot_num); | |
1024 | ||
1025 | /* wipe the journal */ | |
1026 | mlog(0, "flushing the journal.\n"); | |
1027 | journal_lock_updates(journal); | |
1028 | status = journal_flush(journal); | |
1029 | journal_unlock_updates(journal); | |
1030 | if (status < 0) | |
1031 | mlog_errno(status); | |
1032 | ||
1033 | /* This will mark the node clean */ | |
1034 | flags = le32_to_cpu(fe->id1.journal1.ij_flags); | |
1035 | flags &= ~OCFS2_JOURNAL_DIRTY_FL; | |
1036 | fe->id1.journal1.ij_flags = cpu_to_le32(flags); | |
1037 | ||
1038 | status = ocfs2_write_block(osb, bh, inode); | |
1039 | if (status < 0) | |
1040 | mlog_errno(status); | |
1041 | ||
1042 | if (!igrab(inode)) | |
1043 | BUG(); | |
1044 | ||
1045 | journal_destroy(journal); | |
1046 | ||
1047 | done: | |
1048 | /* drop the lock on this nodes journal */ | |
1049 | if (got_lock) | |
1050 | ocfs2_meta_unlock(inode, 1); | |
1051 | ||
1052 | if (inode) | |
1053 | iput(inode); | |
1054 | ||
1055 | if (bh) | |
1056 | brelse(bh); | |
1057 | ||
1058 | mlog_exit(status); | |
1059 | return status; | |
1060 | } | |
1061 | ||
1062 | /* | |
1063 | * Do the most important parts of node recovery: | |
1064 | * - Replay it's journal | |
1065 | * - Stamp a clean local allocator file | |
1066 | * - Stamp a clean truncate log | |
1067 | * - Mark the node clean | |
1068 | * | |
1069 | * If this function completes without error, a node in OCFS2 can be | |
1070 | * said to have been safely recovered. As a result, failure during the | |
1071 | * second part of a nodes recovery process (local alloc recovery) is | |
1072 | * far less concerning. | |
1073 | */ | |
1074 | static int ocfs2_recover_node(struct ocfs2_super *osb, | |
1075 | int node_num) | |
1076 | { | |
1077 | int status = 0; | |
1078 | int slot_num; | |
1079 | struct ocfs2_slot_info *si = osb->slot_info; | |
1080 | struct ocfs2_dinode *la_copy = NULL; | |
1081 | struct ocfs2_dinode *tl_copy = NULL; | |
1082 | ||
1083 | mlog_entry("(node_num=%d, osb->node_num = %d)\n", | |
1084 | node_num, osb->node_num); | |
1085 | ||
1086 | mlog(0, "checking node %d\n", node_num); | |
1087 | ||
1088 | /* Should not ever be called to recover ourselves -- in that | |
1089 | * case we should've called ocfs2_journal_load instead. */ | |
ebdec83b | 1090 | BUG_ON(osb->node_num == node_num); |
ccd979bd MF |
1091 | |
1092 | slot_num = ocfs2_node_num_to_slot(si, node_num); | |
1093 | if (slot_num == OCFS2_INVALID_SLOT) { | |
1094 | status = 0; | |
1095 | mlog(0, "no slot for this node, so no recovery required.\n"); | |
1096 | goto done; | |
1097 | } | |
1098 | ||
1099 | mlog(0, "node %d was using slot %d\n", node_num, slot_num); | |
1100 | ||
1101 | status = ocfs2_replay_journal(osb, node_num, slot_num); | |
1102 | if (status < 0) { | |
1103 | mlog_errno(status); | |
1104 | goto done; | |
1105 | } | |
1106 | ||
1107 | /* Stamp a clean local alloc file AFTER recovering the journal... */ | |
1108 | status = ocfs2_begin_local_alloc_recovery(osb, slot_num, &la_copy); | |
1109 | if (status < 0) { | |
1110 | mlog_errno(status); | |
1111 | goto done; | |
1112 | } | |
1113 | ||
1114 | /* An error from begin_truncate_log_recovery is not | |
1115 | * serious enough to warrant halting the rest of | |
1116 | * recovery. */ | |
1117 | status = ocfs2_begin_truncate_log_recovery(osb, slot_num, &tl_copy); | |
1118 | if (status < 0) | |
1119 | mlog_errno(status); | |
1120 | ||
1121 | /* Likewise, this would be a strange but ultimately not so | |
1122 | * harmful place to get an error... */ | |
1123 | ocfs2_clear_slot(si, slot_num); | |
1124 | status = ocfs2_update_disk_slots(osb, si); | |
1125 | if (status < 0) | |
1126 | mlog_errno(status); | |
1127 | ||
1128 | /* This will kfree the memory pointed to by la_copy and tl_copy */ | |
1129 | ocfs2_queue_recovery_completion(osb->journal, slot_num, la_copy, | |
1130 | tl_copy); | |
1131 | ||
1132 | status = 0; | |
1133 | done: | |
1134 | ||
1135 | mlog_exit(status); | |
1136 | return status; | |
1137 | } | |
1138 | ||
1139 | /* Test node liveness by trylocking his journal. If we get the lock, | |
1140 | * we drop it here. Return 0 if we got the lock, -EAGAIN if node is | |
1141 | * still alive (we couldn't get the lock) and < 0 on error. */ | |
1142 | static int ocfs2_trylock_journal(struct ocfs2_super *osb, | |
1143 | int slot_num) | |
1144 | { | |
1145 | int status, flags; | |
1146 | struct inode *inode = NULL; | |
1147 | ||
1148 | inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, | |
1149 | slot_num); | |
1150 | if (inode == NULL) { | |
1151 | mlog(ML_ERROR, "access error\n"); | |
1152 | status = -EACCES; | |
1153 | goto bail; | |
1154 | } | |
1155 | if (is_bad_inode(inode)) { | |
1156 | mlog(ML_ERROR, "access error (bad inode)\n"); | |
1157 | iput(inode); | |
1158 | inode = NULL; | |
1159 | status = -EACCES; | |
1160 | goto bail; | |
1161 | } | |
1162 | SET_INODE_JOURNAL(inode); | |
1163 | ||
1164 | flags = OCFS2_META_LOCK_RECOVERY | OCFS2_META_LOCK_NOQUEUE; | |
4bcec184 | 1165 | status = ocfs2_meta_lock_full(inode, NULL, 1, flags); |
ccd979bd MF |
1166 | if (status < 0) { |
1167 | if (status != -EAGAIN) | |
1168 | mlog_errno(status); | |
1169 | goto bail; | |
1170 | } | |
1171 | ||
1172 | ocfs2_meta_unlock(inode, 1); | |
1173 | bail: | |
1174 | if (inode) | |
1175 | iput(inode); | |
1176 | ||
1177 | return status; | |
1178 | } | |
1179 | ||
1180 | /* Call this underneath ocfs2_super_lock. It also assumes that the | |
1181 | * slot info struct has been updated from disk. */ | |
1182 | int ocfs2_mark_dead_nodes(struct ocfs2_super *osb) | |
1183 | { | |
1184 | int status, i, node_num; | |
1185 | struct ocfs2_slot_info *si = osb->slot_info; | |
1186 | ||
1187 | /* This is called with the super block cluster lock, so we | |
1188 | * know that the slot map can't change underneath us. */ | |
1189 | ||
1190 | spin_lock(&si->si_lock); | |
1191 | for(i = 0; i < si->si_num_slots; i++) { | |
1192 | if (i == osb->slot_num) | |
1193 | continue; | |
1194 | if (ocfs2_is_empty_slot(si, i)) | |
1195 | continue; | |
1196 | ||
1197 | node_num = si->si_global_node_nums[i]; | |
1198 | if (ocfs2_node_map_test_bit(osb, &osb->recovery_map, node_num)) | |
1199 | continue; | |
1200 | spin_unlock(&si->si_lock); | |
1201 | ||
1202 | /* Ok, we have a slot occupied by another node which | |
1203 | * is not in the recovery map. We trylock his journal | |
1204 | * file here to test if he's alive. */ | |
1205 | status = ocfs2_trylock_journal(osb, i); | |
1206 | if (!status) { | |
1207 | /* Since we're called from mount, we know that | |
1208 | * the recovery thread can't race us on | |
1209 | * setting / checking the recovery bits. */ | |
1210 | ocfs2_recovery_thread(osb, node_num); | |
1211 | } else if ((status < 0) && (status != -EAGAIN)) { | |
1212 | mlog_errno(status); | |
1213 | goto bail; | |
1214 | } | |
1215 | ||
1216 | spin_lock(&si->si_lock); | |
1217 | } | |
1218 | spin_unlock(&si->si_lock); | |
1219 | ||
1220 | status = 0; | |
1221 | bail: | |
1222 | mlog_exit(status); | |
1223 | return status; | |
1224 | } | |
1225 | ||
5eae5b96 MF |
1226 | struct ocfs2_orphan_filldir_priv { |
1227 | struct inode *head; | |
1228 | struct ocfs2_super *osb; | |
1229 | }; | |
1230 | ||
1231 | static int ocfs2_orphan_filldir(void *priv, const char *name, int name_len, | |
1232 | loff_t pos, u64 ino, unsigned type) | |
1233 | { | |
1234 | struct ocfs2_orphan_filldir_priv *p = priv; | |
1235 | struct inode *iter; | |
1236 | ||
1237 | if (name_len == 1 && !strncmp(".", name, 1)) | |
1238 | return 0; | |
1239 | if (name_len == 2 && !strncmp("..", name, 2)) | |
1240 | return 0; | |
1241 | ||
1242 | /* Skip bad inodes so that recovery can continue */ | |
1243 | iter = ocfs2_iget(p->osb, ino, | |
1244 | OCFS2_FI_FLAG_ORPHAN_RECOVERY); | |
1245 | if (IS_ERR(iter)) | |
1246 | return 0; | |
1247 | ||
1248 | mlog(0, "queue orphan %llu\n", | |
1249 | (unsigned long long)OCFS2_I(iter)->ip_blkno); | |
1250 | /* No locking is required for the next_orphan queue as there | |
1251 | * is only ever a single process doing orphan recovery. */ | |
1252 | OCFS2_I(iter)->ip_next_orphan = p->head; | |
1253 | p->head = iter; | |
1254 | ||
1255 | return 0; | |
1256 | } | |
1257 | ||
b4df6ed8 MF |
1258 | static int ocfs2_queue_orphans(struct ocfs2_super *osb, |
1259 | int slot, | |
1260 | struct inode **head) | |
ccd979bd | 1261 | { |
b4df6ed8 | 1262 | int status; |
ccd979bd | 1263 | struct inode *orphan_dir_inode = NULL; |
5eae5b96 MF |
1264 | struct ocfs2_orphan_filldir_priv priv; |
1265 | loff_t pos = 0; | |
1266 | ||
1267 | priv.osb = osb; | |
1268 | priv.head = *head; | |
ccd979bd MF |
1269 | |
1270 | orphan_dir_inode = ocfs2_get_system_file_inode(osb, | |
1271 | ORPHAN_DIR_SYSTEM_INODE, | |
1272 | slot); | |
1273 | if (!orphan_dir_inode) { | |
1274 | status = -ENOENT; | |
1275 | mlog_errno(status); | |
b4df6ed8 MF |
1276 | return status; |
1277 | } | |
ccd979bd | 1278 | |
1b1dcc1b | 1279 | mutex_lock(&orphan_dir_inode->i_mutex); |
4bcec184 | 1280 | status = ocfs2_meta_lock(orphan_dir_inode, NULL, 0); |
ccd979bd | 1281 | if (status < 0) { |
ccd979bd MF |
1282 | mlog_errno(status); |
1283 | goto out; | |
1284 | } | |
ccd979bd | 1285 | |
5eae5b96 MF |
1286 | status = ocfs2_dir_foreach(orphan_dir_inode, &pos, &priv, |
1287 | ocfs2_orphan_filldir); | |
1288 | if (status) { | |
1289 | mlog_errno(status); | |
a86370fb | 1290 | goto out_cluster; |
ccd979bd | 1291 | } |
ccd979bd | 1292 | |
5eae5b96 MF |
1293 | *head = priv.head; |
1294 | ||
a86370fb | 1295 | out_cluster: |
ccd979bd | 1296 | ocfs2_meta_unlock(orphan_dir_inode, 0); |
b4df6ed8 MF |
1297 | out: |
1298 | mutex_unlock(&orphan_dir_inode->i_mutex); | |
ccd979bd | 1299 | iput(orphan_dir_inode); |
b4df6ed8 MF |
1300 | return status; |
1301 | } | |
1302 | ||
1303 | static int ocfs2_orphan_recovery_can_continue(struct ocfs2_super *osb, | |
1304 | int slot) | |
1305 | { | |
1306 | int ret; | |
1307 | ||
1308 | spin_lock(&osb->osb_lock); | |
1309 | ret = !osb->osb_orphan_wipes[slot]; | |
1310 | spin_unlock(&osb->osb_lock); | |
1311 | return ret; | |
1312 | } | |
1313 | ||
1314 | static void ocfs2_mark_recovering_orphan_dir(struct ocfs2_super *osb, | |
1315 | int slot) | |
1316 | { | |
1317 | spin_lock(&osb->osb_lock); | |
1318 | /* Mark ourselves such that new processes in delete_inode() | |
1319 | * know to quit early. */ | |
1320 | ocfs2_node_map_set_bit(osb, &osb->osb_recovering_orphan_dirs, slot); | |
1321 | while (osb->osb_orphan_wipes[slot]) { | |
1322 | /* If any processes are already in the middle of an | |
1323 | * orphan wipe on this dir, then we need to wait for | |
1324 | * them. */ | |
1325 | spin_unlock(&osb->osb_lock); | |
1326 | wait_event_interruptible(osb->osb_wipe_event, | |
1327 | ocfs2_orphan_recovery_can_continue(osb, slot)); | |
1328 | spin_lock(&osb->osb_lock); | |
1329 | } | |
1330 | spin_unlock(&osb->osb_lock); | |
1331 | } | |
1332 | ||
1333 | static void ocfs2_clear_recovering_orphan_dir(struct ocfs2_super *osb, | |
1334 | int slot) | |
1335 | { | |
1336 | ocfs2_node_map_clear_bit(osb, &osb->osb_recovering_orphan_dirs, slot); | |
1337 | } | |
1338 | ||
1339 | /* | |
1340 | * Orphan recovery. Each mounted node has it's own orphan dir which we | |
1341 | * must run during recovery. Our strategy here is to build a list of | |
1342 | * the inodes in the orphan dir and iget/iput them. The VFS does | |
1343 | * (most) of the rest of the work. | |
1344 | * | |
1345 | * Orphan recovery can happen at any time, not just mount so we have a | |
1346 | * couple of extra considerations. | |
1347 | * | |
1348 | * - We grab as many inodes as we can under the orphan dir lock - | |
1349 | * doing iget() outside the orphan dir risks getting a reference on | |
1350 | * an invalid inode. | |
1351 | * - We must be sure not to deadlock with other processes on the | |
1352 | * system wanting to run delete_inode(). This can happen when they go | |
1353 | * to lock the orphan dir and the orphan recovery process attempts to | |
1354 | * iget() inside the orphan dir lock. This can be avoided by | |
1355 | * advertising our state to ocfs2_delete_inode(). | |
1356 | */ | |
1357 | static int ocfs2_recover_orphans(struct ocfs2_super *osb, | |
1358 | int slot) | |
1359 | { | |
1360 | int ret = 0; | |
1361 | struct inode *inode = NULL; | |
1362 | struct inode *iter; | |
1363 | struct ocfs2_inode_info *oi; | |
1364 | ||
1365 | mlog(0, "Recover inodes from orphan dir in slot %d\n", slot); | |
1366 | ||
1367 | ocfs2_mark_recovering_orphan_dir(osb, slot); | |
1368 | ret = ocfs2_queue_orphans(osb, slot, &inode); | |
1369 | ocfs2_clear_recovering_orphan_dir(osb, slot); | |
1370 | ||
1371 | /* Error here should be noted, but we want to continue with as | |
1372 | * many queued inodes as we've got. */ | |
1373 | if (ret) | |
1374 | mlog_errno(ret); | |
ccd979bd MF |
1375 | |
1376 | while (inode) { | |
1377 | oi = OCFS2_I(inode); | |
b0697053 | 1378 | mlog(0, "iput orphan %llu\n", (unsigned long long)oi->ip_blkno); |
ccd979bd MF |
1379 | |
1380 | iter = oi->ip_next_orphan; | |
1381 | ||
1382 | spin_lock(&oi->ip_lock); | |
1383 | /* Delete voting may have set these on the assumption | |
1384 | * that the other node would wipe them successfully. | |
1385 | * If they are still in the node's orphan dir, we need | |
1386 | * to reset that state. */ | |
1387 | oi->ip_flags &= ~(OCFS2_INODE_DELETED|OCFS2_INODE_SKIP_DELETE); | |
1388 | ||
1389 | /* Set the proper information to get us going into | |
1390 | * ocfs2_delete_inode. */ | |
1391 | oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED; | |
ccd979bd MF |
1392 | spin_unlock(&oi->ip_lock); |
1393 | ||
1394 | iput(inode); | |
1395 | ||
1396 | inode = iter; | |
1397 | } | |
1398 | ||
b4df6ed8 | 1399 | return ret; |
ccd979bd MF |
1400 | } |
1401 | ||
1402 | static int ocfs2_wait_on_mount(struct ocfs2_super *osb) | |
1403 | { | |
1404 | /* This check is good because ocfs2 will wait on our recovery | |
1405 | * thread before changing it to something other than MOUNTED | |
1406 | * or DISABLED. */ | |
1407 | wait_event(osb->osb_mount_event, | |
1408 | atomic_read(&osb->vol_state) == VOLUME_MOUNTED || | |
1409 | atomic_read(&osb->vol_state) == VOLUME_DISABLED); | |
1410 | ||
1411 | /* If there's an error on mount, then we may never get to the | |
1412 | * MOUNTED flag, but this is set right before | |
1413 | * dismount_volume() so we can trust it. */ | |
1414 | if (atomic_read(&osb->vol_state) == VOLUME_DISABLED) { | |
1415 | mlog(0, "mount error, exiting!\n"); | |
1416 | return -EBUSY; | |
1417 | } | |
1418 | ||
1419 | return 0; | |
1420 | } | |
1421 | ||
1422 | static int ocfs2_commit_thread(void *arg) | |
1423 | { | |
1424 | int status; | |
1425 | struct ocfs2_super *osb = arg; | |
1426 | struct ocfs2_journal *journal = osb->journal; | |
1427 | ||
1428 | /* we can trust j_num_trans here because _should_stop() is only set in | |
1429 | * shutdown and nobody other than ourselves should be able to start | |
1430 | * transactions. committing on shutdown might take a few iterations | |
1431 | * as final transactions put deleted inodes on the list */ | |
1432 | while (!(kthread_should_stop() && | |
1433 | atomic_read(&journal->j_num_trans) == 0)) { | |
1434 | ||
745ae8ba MF |
1435 | wait_event_interruptible(osb->checkpoint_event, |
1436 | atomic_read(&journal->j_num_trans) | |
1437 | || kthread_should_stop()); | |
ccd979bd MF |
1438 | |
1439 | status = ocfs2_commit_cache(osb); | |
1440 | if (status < 0) | |
1441 | mlog_errno(status); | |
1442 | ||
1443 | if (kthread_should_stop() && atomic_read(&journal->j_num_trans)){ | |
1444 | mlog(ML_KTHREAD, | |
1445 | "commit_thread: %u transactions pending on " | |
1446 | "shutdown\n", | |
1447 | atomic_read(&journal->j_num_trans)); | |
1448 | } | |
1449 | } | |
1450 | ||
1451 | return 0; | |
1452 | } | |
1453 | ||
1454 | /* Look for a dirty journal without taking any cluster locks. Used for | |
1455 | * hard readonly access to determine whether the file system journals | |
1456 | * require recovery. */ | |
1457 | int ocfs2_check_journals_nolocks(struct ocfs2_super *osb) | |
1458 | { | |
1459 | int ret = 0; | |
1460 | unsigned int slot; | |
1461 | struct buffer_head *di_bh; | |
1462 | struct ocfs2_dinode *di; | |
1463 | struct inode *journal = NULL; | |
1464 | ||
1465 | for(slot = 0; slot < osb->max_slots; slot++) { | |
1466 | journal = ocfs2_get_system_file_inode(osb, | |
1467 | JOURNAL_SYSTEM_INODE, | |
1468 | slot); | |
1469 | if (!journal || is_bad_inode(journal)) { | |
1470 | ret = -EACCES; | |
1471 | mlog_errno(ret); | |
1472 | goto out; | |
1473 | } | |
1474 | ||
1475 | di_bh = NULL; | |
1476 | ret = ocfs2_read_block(osb, OCFS2_I(journal)->ip_blkno, &di_bh, | |
1477 | 0, journal); | |
1478 | if (ret < 0) { | |
1479 | mlog_errno(ret); | |
1480 | goto out; | |
1481 | } | |
1482 | ||
1483 | di = (struct ocfs2_dinode *) di_bh->b_data; | |
1484 | ||
1485 | if (le32_to_cpu(di->id1.journal1.ij_flags) & | |
1486 | OCFS2_JOURNAL_DIRTY_FL) | |
1487 | ret = -EROFS; | |
1488 | ||
1489 | brelse(di_bh); | |
1490 | if (ret) | |
1491 | break; | |
1492 | } | |
1493 | ||
1494 | out: | |
1495 | if (journal) | |
1496 | iput(journal); | |
1497 | ||
1498 | return ret; | |
1499 | } |