]>
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" | |
50655ae9 | 38 | #include "blockcheck.h" |
316f4b9f | 39 | #include "dir.h" |
ccd979bd MF |
40 | #include "dlmglue.h" |
41 | #include "extent_map.h" | |
42 | #include "heartbeat.h" | |
43 | #include "inode.h" | |
44 | #include "journal.h" | |
45 | #include "localalloc.h" | |
ccd979bd MF |
46 | #include "slot_map.h" |
47 | #include "super.h" | |
ccd979bd | 48 | #include "sysfile.h" |
2205363d | 49 | #include "quota.h" |
ccd979bd MF |
50 | |
51 | #include "buffer_head_io.h" | |
52 | ||
34af946a | 53 | DEFINE_SPINLOCK(trans_inc_lock); |
ccd979bd MF |
54 | |
55 | static int ocfs2_force_read_journal(struct inode *inode); | |
56 | static int ocfs2_recover_node(struct ocfs2_super *osb, | |
2205363d | 57 | int node_num, int slot_num); |
ccd979bd MF |
58 | static int __ocfs2_recovery_thread(void *arg); |
59 | static int ocfs2_commit_cache(struct ocfs2_super *osb); | |
19ece546 | 60 | static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota); |
ccd979bd | 61 | static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb, |
539d8264 | 62 | int dirty, int replayed); |
ccd979bd MF |
63 | static int ocfs2_trylock_journal(struct ocfs2_super *osb, |
64 | int slot_num); | |
65 | static int ocfs2_recover_orphans(struct ocfs2_super *osb, | |
66 | int slot); | |
67 | static int ocfs2_commit_thread(void *arg); | |
9140db04 SE |
68 | static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal, |
69 | int slot_num, | |
70 | struct ocfs2_dinode *la_dinode, | |
71 | struct ocfs2_dinode *tl_dinode, | |
72 | struct ocfs2_quota_recovery *qrec); | |
ccd979bd | 73 | |
19ece546 JK |
74 | static inline int ocfs2_wait_on_mount(struct ocfs2_super *osb) |
75 | { | |
76 | return __ocfs2_wait_on_mount(osb, 0); | |
77 | } | |
78 | ||
79 | static inline int ocfs2_wait_on_quotas(struct ocfs2_super *osb) | |
80 | { | |
81 | return __ocfs2_wait_on_mount(osb, 1); | |
82 | } | |
83 | ||
9140db04 SE |
84 | /* |
85 | * This replay_map is to track online/offline slots, so we could recover | |
86 | * offline slots during recovery and mount | |
87 | */ | |
88 | ||
89 | enum ocfs2_replay_state { | |
90 | REPLAY_UNNEEDED = 0, /* Replay is not needed, so ignore this map */ | |
91 | REPLAY_NEEDED, /* Replay slots marked in rm_replay_slots */ | |
92 | REPLAY_DONE /* Replay was already queued */ | |
93 | }; | |
94 | ||
95 | struct ocfs2_replay_map { | |
96 | unsigned int rm_slots; | |
97 | enum ocfs2_replay_state rm_state; | |
98 | unsigned char rm_replay_slots[0]; | |
99 | }; | |
100 | ||
101 | void ocfs2_replay_map_set_state(struct ocfs2_super *osb, int state) | |
102 | { | |
103 | if (!osb->replay_map) | |
104 | return; | |
105 | ||
106 | /* If we've already queued the replay, we don't have any more to do */ | |
107 | if (osb->replay_map->rm_state == REPLAY_DONE) | |
108 | return; | |
109 | ||
110 | osb->replay_map->rm_state = state; | |
111 | } | |
112 | ||
113 | int ocfs2_compute_replay_slots(struct ocfs2_super *osb) | |
114 | { | |
115 | struct ocfs2_replay_map *replay_map; | |
116 | int i, node_num; | |
117 | ||
118 | /* If replay map is already set, we don't do it again */ | |
119 | if (osb->replay_map) | |
120 | return 0; | |
121 | ||
122 | replay_map = kzalloc(sizeof(struct ocfs2_replay_map) + | |
123 | (osb->max_slots * sizeof(char)), GFP_KERNEL); | |
124 | ||
125 | if (!replay_map) { | |
126 | mlog_errno(-ENOMEM); | |
127 | return -ENOMEM; | |
128 | } | |
129 | ||
130 | spin_lock(&osb->osb_lock); | |
131 | ||
132 | replay_map->rm_slots = osb->max_slots; | |
133 | replay_map->rm_state = REPLAY_UNNEEDED; | |
134 | ||
135 | /* set rm_replay_slots for offline slot(s) */ | |
136 | for (i = 0; i < replay_map->rm_slots; i++) { | |
137 | if (ocfs2_slot_to_node_num_locked(osb, i, &node_num) == -ENOENT) | |
138 | replay_map->rm_replay_slots[i] = 1; | |
139 | } | |
140 | ||
141 | osb->replay_map = replay_map; | |
142 | spin_unlock(&osb->osb_lock); | |
143 | return 0; | |
144 | } | |
145 | ||
146 | void ocfs2_queue_replay_slots(struct ocfs2_super *osb) | |
147 | { | |
148 | struct ocfs2_replay_map *replay_map = osb->replay_map; | |
149 | int i; | |
150 | ||
151 | if (!replay_map) | |
152 | return; | |
153 | ||
154 | if (replay_map->rm_state != REPLAY_NEEDED) | |
155 | return; | |
156 | ||
157 | for (i = 0; i < replay_map->rm_slots; i++) | |
158 | if (replay_map->rm_replay_slots[i]) | |
159 | ocfs2_queue_recovery_completion(osb->journal, i, NULL, | |
160 | NULL, NULL); | |
161 | replay_map->rm_state = REPLAY_DONE; | |
162 | } | |
163 | ||
164 | void ocfs2_free_replay_slots(struct ocfs2_super *osb) | |
165 | { | |
166 | struct ocfs2_replay_map *replay_map = osb->replay_map; | |
167 | ||
168 | if (!osb->replay_map) | |
169 | return; | |
170 | ||
171 | kfree(replay_map); | |
172 | osb->replay_map = NULL; | |
173 | } | |
174 | ||
553abd04 JB |
175 | int ocfs2_recovery_init(struct ocfs2_super *osb) |
176 | { | |
177 | struct ocfs2_recovery_map *rm; | |
178 | ||
179 | mutex_init(&osb->recovery_lock); | |
180 | osb->disable_recovery = 0; | |
181 | osb->recovery_thread_task = NULL; | |
182 | init_waitqueue_head(&osb->recovery_event); | |
183 | ||
184 | rm = kzalloc(sizeof(struct ocfs2_recovery_map) + | |
185 | osb->max_slots * sizeof(unsigned int), | |
186 | GFP_KERNEL); | |
187 | if (!rm) { | |
188 | mlog_errno(-ENOMEM); | |
189 | return -ENOMEM; | |
190 | } | |
191 | ||
192 | rm->rm_entries = (unsigned int *)((char *)rm + | |
193 | sizeof(struct ocfs2_recovery_map)); | |
194 | osb->recovery_map = rm; | |
195 | ||
196 | return 0; | |
197 | } | |
198 | ||
199 | /* we can't grab the goofy sem lock from inside wait_event, so we use | |
200 | * memory barriers to make sure that we'll see the null task before | |
201 | * being woken up */ | |
202 | static int ocfs2_recovery_thread_running(struct ocfs2_super *osb) | |
203 | { | |
204 | mb(); | |
205 | return osb->recovery_thread_task != NULL; | |
206 | } | |
207 | ||
208 | void ocfs2_recovery_exit(struct ocfs2_super *osb) | |
209 | { | |
210 | struct ocfs2_recovery_map *rm; | |
211 | ||
212 | /* disable any new recovery threads and wait for any currently | |
213 | * running ones to exit. Do this before setting the vol_state. */ | |
214 | mutex_lock(&osb->recovery_lock); | |
215 | osb->disable_recovery = 1; | |
216 | mutex_unlock(&osb->recovery_lock); | |
217 | wait_event(osb->recovery_event, !ocfs2_recovery_thread_running(osb)); | |
218 | ||
219 | /* At this point, we know that no more recovery threads can be | |
220 | * launched, so wait for any recovery completion work to | |
221 | * complete. */ | |
222 | flush_workqueue(ocfs2_wq); | |
223 | ||
224 | /* | |
225 | * Now that recovery is shut down, and the osb is about to be | |
226 | * freed, the osb_lock is not taken here. | |
227 | */ | |
228 | rm = osb->recovery_map; | |
229 | /* XXX: Should we bug if there are dirty entries? */ | |
230 | ||
231 | kfree(rm); | |
232 | } | |
233 | ||
234 | static int __ocfs2_recovery_map_test(struct ocfs2_super *osb, | |
235 | unsigned int node_num) | |
236 | { | |
237 | int i; | |
238 | struct ocfs2_recovery_map *rm = osb->recovery_map; | |
239 | ||
240 | assert_spin_locked(&osb->osb_lock); | |
241 | ||
242 | for (i = 0; i < rm->rm_used; i++) { | |
243 | if (rm->rm_entries[i] == node_num) | |
244 | return 1; | |
245 | } | |
246 | ||
247 | return 0; | |
248 | } | |
249 | ||
250 | /* Behaves like test-and-set. Returns the previous value */ | |
251 | static int ocfs2_recovery_map_set(struct ocfs2_super *osb, | |
252 | unsigned int node_num) | |
253 | { | |
254 | struct ocfs2_recovery_map *rm = osb->recovery_map; | |
255 | ||
256 | spin_lock(&osb->osb_lock); | |
257 | if (__ocfs2_recovery_map_test(osb, node_num)) { | |
258 | spin_unlock(&osb->osb_lock); | |
259 | return 1; | |
260 | } | |
261 | ||
262 | /* XXX: Can this be exploited? Not from o2dlm... */ | |
263 | BUG_ON(rm->rm_used >= osb->max_slots); | |
264 | ||
265 | rm->rm_entries[rm->rm_used] = node_num; | |
266 | rm->rm_used++; | |
267 | spin_unlock(&osb->osb_lock); | |
268 | ||
269 | return 0; | |
270 | } | |
271 | ||
272 | static void ocfs2_recovery_map_clear(struct ocfs2_super *osb, | |
273 | unsigned int node_num) | |
274 | { | |
275 | int i; | |
276 | struct ocfs2_recovery_map *rm = osb->recovery_map; | |
277 | ||
278 | spin_lock(&osb->osb_lock); | |
279 | ||
280 | for (i = 0; i < rm->rm_used; i++) { | |
281 | if (rm->rm_entries[i] == node_num) | |
282 | break; | |
283 | } | |
284 | ||
285 | if (i < rm->rm_used) { | |
286 | /* XXX: be careful with the pointer math */ | |
287 | memmove(&(rm->rm_entries[i]), &(rm->rm_entries[i + 1]), | |
288 | (rm->rm_used - i - 1) * sizeof(unsigned int)); | |
289 | rm->rm_used--; | |
290 | } | |
291 | ||
292 | spin_unlock(&osb->osb_lock); | |
293 | } | |
294 | ||
ccd979bd MF |
295 | static int ocfs2_commit_cache(struct ocfs2_super *osb) |
296 | { | |
297 | int status = 0; | |
298 | unsigned int flushed; | |
299 | unsigned long old_id; | |
300 | struct ocfs2_journal *journal = NULL; | |
301 | ||
302 | mlog_entry_void(); | |
303 | ||
304 | journal = osb->journal; | |
305 | ||
306 | /* Flush all pending commits and checkpoint the journal. */ | |
307 | down_write(&journal->j_trans_barrier); | |
308 | ||
309 | if (atomic_read(&journal->j_num_trans) == 0) { | |
310 | up_write(&journal->j_trans_barrier); | |
311 | mlog(0, "No transactions for me to flush!\n"); | |
312 | goto finally; | |
313 | } | |
314 | ||
2b4e30fb JB |
315 | jbd2_journal_lock_updates(journal->j_journal); |
316 | status = jbd2_journal_flush(journal->j_journal); | |
317 | jbd2_journal_unlock_updates(journal->j_journal); | |
ccd979bd MF |
318 | if (status < 0) { |
319 | up_write(&journal->j_trans_barrier); | |
320 | mlog_errno(status); | |
321 | goto finally; | |
322 | } | |
323 | ||
324 | old_id = ocfs2_inc_trans_id(journal); | |
325 | ||
326 | flushed = atomic_read(&journal->j_num_trans); | |
327 | atomic_set(&journal->j_num_trans, 0); | |
328 | up_write(&journal->j_trans_barrier); | |
329 | ||
330 | mlog(0, "commit_thread: flushed transaction %lu (%u handles)\n", | |
331 | journal->j_trans_id, flushed); | |
332 | ||
34d024f8 | 333 | ocfs2_wake_downconvert_thread(osb); |
ccd979bd MF |
334 | wake_up(&journal->j_checkpointed); |
335 | finally: | |
336 | mlog_exit(status); | |
337 | return status; | |
338 | } | |
339 | ||
ccd979bd MF |
340 | /* pass it NULL and it will allocate a new handle object for you. If |
341 | * you pass it a handle however, it may still return error, in which | |
342 | * case it has free'd the passed handle for you. */ | |
1fabe148 | 343 | handle_t *ocfs2_start_trans(struct ocfs2_super *osb, int max_buffs) |
ccd979bd | 344 | { |
ccd979bd | 345 | journal_t *journal = osb->journal->j_journal; |
1fabe148 | 346 | handle_t *handle; |
ccd979bd | 347 | |
ebdec83b | 348 | BUG_ON(!osb || !osb->journal->j_journal); |
ccd979bd | 349 | |
65eff9cc MF |
350 | if (ocfs2_is_hard_readonly(osb)) |
351 | return ERR_PTR(-EROFS); | |
ccd979bd MF |
352 | |
353 | BUG_ON(osb->journal->j_state == OCFS2_JOURNAL_FREE); | |
354 | BUG_ON(max_buffs <= 0); | |
355 | ||
90e86a63 JK |
356 | /* Nested transaction? Just return the handle... */ |
357 | if (journal_current_handle()) | |
358 | return jbd2_journal_start(journal, max_buffs); | |
ccd979bd | 359 | |
ccd979bd MF |
360 | down_read(&osb->journal->j_trans_barrier); |
361 | ||
2b4e30fb | 362 | handle = jbd2_journal_start(journal, max_buffs); |
1fabe148 | 363 | if (IS_ERR(handle)) { |
ccd979bd MF |
364 | up_read(&osb->journal->j_trans_barrier); |
365 | ||
1fabe148 | 366 | mlog_errno(PTR_ERR(handle)); |
ccd979bd MF |
367 | |
368 | if (is_journal_aborted(journal)) { | |
369 | ocfs2_abort(osb->sb, "Detected aborted journal"); | |
1fabe148 | 370 | handle = ERR_PTR(-EROFS); |
ccd979bd | 371 | } |
c271c5c2 SM |
372 | } else { |
373 | if (!ocfs2_mount_local(osb)) | |
374 | atomic_inc(&(osb->journal->j_num_trans)); | |
375 | } | |
ccd979bd | 376 | |
ccd979bd | 377 | return handle; |
ccd979bd MF |
378 | } |
379 | ||
1fabe148 MF |
380 | int ocfs2_commit_trans(struct ocfs2_super *osb, |
381 | handle_t *handle) | |
ccd979bd | 382 | { |
90e86a63 | 383 | int ret, nested; |
02dc1af4 | 384 | struct ocfs2_journal *journal = osb->journal; |
ccd979bd MF |
385 | |
386 | BUG_ON(!handle); | |
387 | ||
90e86a63 | 388 | nested = handle->h_ref > 1; |
2b4e30fb | 389 | ret = jbd2_journal_stop(handle); |
1fabe148 MF |
390 | if (ret < 0) |
391 | mlog_errno(ret); | |
ccd979bd | 392 | |
90e86a63 JK |
393 | if (!nested) |
394 | up_read(&journal->j_trans_barrier); | |
ccd979bd | 395 | |
1fabe148 | 396 | return ret; |
ccd979bd MF |
397 | } |
398 | ||
399 | /* | |
400 | * 'nblocks' is what you want to add to the current | |
401 | * transaction. extend_trans will either extend the current handle by | |
402 | * nblocks, or commit it and start a new one with nblocks credits. | |
403 | * | |
2b4e30fb | 404 | * This might call jbd2_journal_restart() which will commit dirty buffers |
e8aed345 MF |
405 | * and then restart the transaction. Before calling |
406 | * ocfs2_extend_trans(), any changed blocks should have been | |
407 | * dirtied. After calling it, all blocks which need to be changed must | |
408 | * go through another set of journal_access/journal_dirty calls. | |
409 | * | |
ccd979bd MF |
410 | * WARNING: This will not release any semaphores or disk locks taken |
411 | * during the transaction, so make sure they were taken *before* | |
412 | * start_trans or we'll have ordering deadlocks. | |
413 | * | |
414 | * WARNING2: Note that we do *not* drop j_trans_barrier here. This is | |
415 | * good because transaction ids haven't yet been recorded on the | |
416 | * cluster locks associated with this handle. | |
417 | */ | |
1fc58146 | 418 | int ocfs2_extend_trans(handle_t *handle, int nblocks) |
ccd979bd MF |
419 | { |
420 | int status; | |
421 | ||
422 | BUG_ON(!handle); | |
ccd979bd MF |
423 | BUG_ON(!nblocks); |
424 | ||
425 | mlog_entry_void(); | |
426 | ||
427 | mlog(0, "Trying to extend transaction by %d blocks\n", nblocks); | |
428 | ||
e407e397 | 429 | #ifdef CONFIG_OCFS2_DEBUG_FS |
0879c584 MF |
430 | status = 1; |
431 | #else | |
2b4e30fb | 432 | status = jbd2_journal_extend(handle, nblocks); |
ccd979bd MF |
433 | if (status < 0) { |
434 | mlog_errno(status); | |
435 | goto bail; | |
436 | } | |
0879c584 | 437 | #endif |
ccd979bd MF |
438 | |
439 | if (status > 0) { | |
2b4e30fb JB |
440 | mlog(0, |
441 | "jbd2_journal_extend failed, trying " | |
442 | "jbd2_journal_restart\n"); | |
443 | status = jbd2_journal_restart(handle, nblocks); | |
ccd979bd | 444 | if (status < 0) { |
ccd979bd MF |
445 | mlog_errno(status); |
446 | goto bail; | |
447 | } | |
01ddf1e1 | 448 | } |
ccd979bd MF |
449 | |
450 | status = 0; | |
451 | bail: | |
452 | ||
453 | mlog_exit(status); | |
454 | return status; | |
455 | } | |
456 | ||
50655ae9 JB |
457 | struct ocfs2_triggers { |
458 | struct jbd2_buffer_trigger_type ot_triggers; | |
459 | int ot_offset; | |
460 | }; | |
461 | ||
462 | static inline struct ocfs2_triggers *to_ocfs2_trigger(struct jbd2_buffer_trigger_type *triggers) | |
463 | { | |
464 | return container_of(triggers, struct ocfs2_triggers, ot_triggers); | |
465 | } | |
466 | ||
467 | static void ocfs2_commit_trigger(struct jbd2_buffer_trigger_type *triggers, | |
468 | struct buffer_head *bh, | |
469 | void *data, size_t size) | |
470 | { | |
471 | struct ocfs2_triggers *ot = to_ocfs2_trigger(triggers); | |
472 | ||
473 | /* | |
474 | * We aren't guaranteed to have the superblock here, so we | |
475 | * must unconditionally compute the ecc data. | |
476 | * __ocfs2_journal_access() will only set the triggers if | |
477 | * metaecc is enabled. | |
478 | */ | |
479 | ocfs2_block_check_compute(data, size, data + ot->ot_offset); | |
480 | } | |
481 | ||
482 | /* | |
483 | * Quota blocks have their own trigger because the struct ocfs2_block_check | |
484 | * offset depends on the blocksize. | |
485 | */ | |
486 | static void ocfs2_dq_commit_trigger(struct jbd2_buffer_trigger_type *triggers, | |
487 | struct buffer_head *bh, | |
488 | void *data, size_t size) | |
489 | { | |
490 | struct ocfs2_disk_dqtrailer *dqt = | |
491 | ocfs2_block_dqtrailer(size, data); | |
492 | ||
493 | /* | |
494 | * We aren't guaranteed to have the superblock here, so we | |
495 | * must unconditionally compute the ecc data. | |
496 | * __ocfs2_journal_access() will only set the triggers if | |
497 | * metaecc is enabled. | |
498 | */ | |
499 | ocfs2_block_check_compute(data, size, &dqt->dq_check); | |
500 | } | |
501 | ||
c175a518 JB |
502 | /* |
503 | * Directory blocks also have their own trigger because the | |
504 | * struct ocfs2_block_check offset depends on the blocksize. | |
505 | */ | |
506 | static void ocfs2_db_commit_trigger(struct jbd2_buffer_trigger_type *triggers, | |
507 | struct buffer_head *bh, | |
508 | void *data, size_t size) | |
509 | { | |
510 | struct ocfs2_dir_block_trailer *trailer = | |
511 | ocfs2_dir_trailer_from_size(size, data); | |
512 | ||
513 | /* | |
514 | * We aren't guaranteed to have the superblock here, so we | |
515 | * must unconditionally compute the ecc data. | |
516 | * __ocfs2_journal_access() will only set the triggers if | |
517 | * metaecc is enabled. | |
518 | */ | |
519 | ocfs2_block_check_compute(data, size, &trailer->db_check); | |
520 | } | |
521 | ||
50655ae9 JB |
522 | static void ocfs2_abort_trigger(struct jbd2_buffer_trigger_type *triggers, |
523 | struct buffer_head *bh) | |
524 | { | |
525 | mlog(ML_ERROR, | |
526 | "ocfs2_abort_trigger called by JBD2. bh = 0x%lx, " | |
527 | "bh->b_blocknr = %llu\n", | |
528 | (unsigned long)bh, | |
529 | (unsigned long long)bh->b_blocknr); | |
530 | ||
531 | /* We aren't guaranteed to have the superblock here - but if we | |
532 | * don't, it'll just crash. */ | |
533 | ocfs2_error(bh->b_assoc_map->host->i_sb, | |
534 | "JBD2 has aborted our journal, ocfs2 cannot continue\n"); | |
535 | } | |
536 | ||
537 | static struct ocfs2_triggers di_triggers = { | |
538 | .ot_triggers = { | |
539 | .t_commit = ocfs2_commit_trigger, | |
540 | .t_abort = ocfs2_abort_trigger, | |
541 | }, | |
542 | .ot_offset = offsetof(struct ocfs2_dinode, i_check), | |
543 | }; | |
544 | ||
545 | static struct ocfs2_triggers eb_triggers = { | |
546 | .ot_triggers = { | |
547 | .t_commit = ocfs2_commit_trigger, | |
548 | .t_abort = ocfs2_abort_trigger, | |
549 | }, | |
550 | .ot_offset = offsetof(struct ocfs2_extent_block, h_check), | |
551 | }; | |
552 | ||
553 | static struct ocfs2_triggers gd_triggers = { | |
554 | .ot_triggers = { | |
555 | .t_commit = ocfs2_commit_trigger, | |
556 | .t_abort = ocfs2_abort_trigger, | |
557 | }, | |
558 | .ot_offset = offsetof(struct ocfs2_group_desc, bg_check), | |
559 | }; | |
560 | ||
c175a518 JB |
561 | static struct ocfs2_triggers db_triggers = { |
562 | .ot_triggers = { | |
563 | .t_commit = ocfs2_db_commit_trigger, | |
564 | .t_abort = ocfs2_abort_trigger, | |
565 | }, | |
566 | }; | |
567 | ||
50655ae9 JB |
568 | static struct ocfs2_triggers xb_triggers = { |
569 | .ot_triggers = { | |
570 | .t_commit = ocfs2_commit_trigger, | |
571 | .t_abort = ocfs2_abort_trigger, | |
572 | }, | |
573 | .ot_offset = offsetof(struct ocfs2_xattr_block, xb_check), | |
574 | }; | |
575 | ||
576 | static struct ocfs2_triggers dq_triggers = { | |
577 | .ot_triggers = { | |
578 | .t_commit = ocfs2_dq_commit_trigger, | |
579 | .t_abort = ocfs2_abort_trigger, | |
580 | }, | |
581 | }; | |
582 | ||
9b7895ef MF |
583 | static struct ocfs2_triggers dr_triggers = { |
584 | .ot_triggers = { | |
585 | .t_commit = ocfs2_commit_trigger, | |
586 | .t_abort = ocfs2_abort_trigger, | |
587 | }, | |
588 | .ot_offset = offsetof(struct ocfs2_dx_root_block, dr_check), | |
589 | }; | |
590 | ||
591 | static struct ocfs2_triggers dl_triggers = { | |
592 | .ot_triggers = { | |
593 | .t_commit = ocfs2_commit_trigger, | |
594 | .t_abort = ocfs2_abort_trigger, | |
595 | }, | |
596 | .ot_offset = offsetof(struct ocfs2_dx_leaf, dl_check), | |
597 | }; | |
598 | ||
50655ae9 JB |
599 | static int __ocfs2_journal_access(handle_t *handle, |
600 | struct inode *inode, | |
601 | struct buffer_head *bh, | |
602 | struct ocfs2_triggers *triggers, | |
603 | int type) | |
ccd979bd MF |
604 | { |
605 | int status; | |
606 | ||
607 | BUG_ON(!inode); | |
608 | BUG_ON(!handle); | |
609 | BUG_ON(!bh); | |
ccd979bd | 610 | |
205f87f6 | 611 | mlog_entry("bh->b_blocknr=%llu, type=%d (\"%s\"), bh->b_size = %zu\n", |
ccd979bd MF |
612 | (unsigned long long)bh->b_blocknr, type, |
613 | (type == OCFS2_JOURNAL_ACCESS_CREATE) ? | |
614 | "OCFS2_JOURNAL_ACCESS_CREATE" : | |
615 | "OCFS2_JOURNAL_ACCESS_WRITE", | |
616 | bh->b_size); | |
617 | ||
618 | /* we can safely remove this assertion after testing. */ | |
619 | if (!buffer_uptodate(bh)) { | |
620 | mlog(ML_ERROR, "giving me a buffer that's not uptodate!\n"); | |
621 | mlog(ML_ERROR, "b_blocknr=%llu\n", | |
622 | (unsigned long long)bh->b_blocknr); | |
623 | BUG(); | |
624 | } | |
625 | ||
626 | /* Set the current transaction information on the inode so | |
627 | * that the locking code knows whether it can drop it's locks | |
628 | * on this inode or not. We're protected from the commit | |
629 | * thread updating the current transaction id until | |
630 | * ocfs2_commit_trans() because ocfs2_start_trans() took | |
631 | * j_trans_barrier for us. */ | |
632 | ocfs2_set_inode_lock_trans(OCFS2_SB(inode->i_sb)->journal, inode); | |
633 | ||
251b6ecc | 634 | mutex_lock(&OCFS2_I(inode)->ip_io_mutex); |
ccd979bd MF |
635 | switch (type) { |
636 | case OCFS2_JOURNAL_ACCESS_CREATE: | |
637 | case OCFS2_JOURNAL_ACCESS_WRITE: | |
2b4e30fb | 638 | status = jbd2_journal_get_write_access(handle, bh); |
ccd979bd MF |
639 | break; |
640 | ||
641 | case OCFS2_JOURNAL_ACCESS_UNDO: | |
2b4e30fb | 642 | status = jbd2_journal_get_undo_access(handle, bh); |
ccd979bd MF |
643 | break; |
644 | ||
645 | default: | |
646 | status = -EINVAL; | |
647 | mlog(ML_ERROR, "Uknown access type!\n"); | |
648 | } | |
50655ae9 JB |
649 | if (!status && ocfs2_meta_ecc(OCFS2_SB(inode->i_sb)) && triggers) |
650 | jbd2_journal_set_triggers(bh, &triggers->ot_triggers); | |
251b6ecc | 651 | mutex_unlock(&OCFS2_I(inode)->ip_io_mutex); |
ccd979bd MF |
652 | |
653 | if (status < 0) | |
654 | mlog(ML_ERROR, "Error %d getting %d access to buffer!\n", | |
655 | status, type); | |
656 | ||
657 | mlog_exit(status); | |
658 | return status; | |
659 | } | |
660 | ||
50655ae9 JB |
661 | int ocfs2_journal_access_di(handle_t *handle, struct inode *inode, |
662 | struct buffer_head *bh, int type) | |
663 | { | |
664 | return __ocfs2_journal_access(handle, inode, bh, &di_triggers, | |
665 | type); | |
666 | } | |
667 | ||
668 | int ocfs2_journal_access_eb(handle_t *handle, struct inode *inode, | |
669 | struct buffer_head *bh, int type) | |
670 | { | |
671 | return __ocfs2_journal_access(handle, inode, bh, &eb_triggers, | |
672 | type); | |
673 | } | |
674 | ||
675 | int ocfs2_journal_access_gd(handle_t *handle, struct inode *inode, | |
676 | struct buffer_head *bh, int type) | |
677 | { | |
678 | return __ocfs2_journal_access(handle, inode, bh, &gd_triggers, | |
679 | type); | |
680 | } | |
681 | ||
682 | int ocfs2_journal_access_db(handle_t *handle, struct inode *inode, | |
683 | struct buffer_head *bh, int type) | |
684 | { | |
c175a518 JB |
685 | return __ocfs2_journal_access(handle, inode, bh, &db_triggers, |
686 | type); | |
50655ae9 JB |
687 | } |
688 | ||
689 | int ocfs2_journal_access_xb(handle_t *handle, struct inode *inode, | |
690 | struct buffer_head *bh, int type) | |
691 | { | |
692 | return __ocfs2_journal_access(handle, inode, bh, &xb_triggers, | |
693 | type); | |
694 | } | |
695 | ||
696 | int ocfs2_journal_access_dq(handle_t *handle, struct inode *inode, | |
697 | struct buffer_head *bh, int type) | |
698 | { | |
699 | return __ocfs2_journal_access(handle, inode, bh, &dq_triggers, | |
700 | type); | |
701 | } | |
702 | ||
9b7895ef MF |
703 | int ocfs2_journal_access_dr(handle_t *handle, struct inode *inode, |
704 | struct buffer_head *bh, int type) | |
705 | { | |
706 | return __ocfs2_journal_access(handle, inode, bh, &dr_triggers, | |
707 | type); | |
708 | } | |
709 | ||
710 | int ocfs2_journal_access_dl(handle_t *handle, struct inode *inode, | |
711 | struct buffer_head *bh, int type) | |
712 | { | |
713 | return __ocfs2_journal_access(handle, inode, bh, &dl_triggers, | |
714 | type); | |
715 | } | |
716 | ||
50655ae9 JB |
717 | int ocfs2_journal_access(handle_t *handle, struct inode *inode, |
718 | struct buffer_head *bh, int type) | |
719 | { | |
720 | return __ocfs2_journal_access(handle, inode, bh, NULL, type); | |
721 | } | |
722 | ||
1fabe148 | 723 | int ocfs2_journal_dirty(handle_t *handle, |
ccd979bd MF |
724 | struct buffer_head *bh) |
725 | { | |
726 | int status; | |
727 | ||
ccd979bd MF |
728 | mlog_entry("(bh->b_blocknr=%llu)\n", |
729 | (unsigned long long)bh->b_blocknr); | |
730 | ||
2b4e30fb | 731 | status = jbd2_journal_dirty_metadata(handle, bh); |
ccd979bd MF |
732 | if (status < 0) |
733 | mlog(ML_ERROR, "Could not dirty metadata buffer. " | |
734 | "(bh->b_blocknr=%llu)\n", | |
735 | (unsigned long long)bh->b_blocknr); | |
736 | ||
737 | mlog_exit(status); | |
738 | return status; | |
739 | } | |
740 | ||
2b4e30fb | 741 | #define OCFS2_DEFAULT_COMMIT_INTERVAL (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE) |
ccd979bd MF |
742 | |
743 | void ocfs2_set_journal_params(struct ocfs2_super *osb) | |
744 | { | |
745 | journal_t *journal = osb->journal->j_journal; | |
d147b3d6 MF |
746 | unsigned long commit_interval = OCFS2_DEFAULT_COMMIT_INTERVAL; |
747 | ||
748 | if (osb->osb_commit_interval) | |
749 | commit_interval = osb->osb_commit_interval; | |
ccd979bd MF |
750 | |
751 | spin_lock(&journal->j_state_lock); | |
d147b3d6 | 752 | journal->j_commit_interval = commit_interval; |
ccd979bd | 753 | if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER) |
2b4e30fb | 754 | journal->j_flags |= JBD2_BARRIER; |
ccd979bd | 755 | else |
2b4e30fb | 756 | journal->j_flags &= ~JBD2_BARRIER; |
ccd979bd MF |
757 | spin_unlock(&journal->j_state_lock); |
758 | } | |
759 | ||
760 | int ocfs2_journal_init(struct ocfs2_journal *journal, int *dirty) | |
761 | { | |
762 | int status = -1; | |
763 | struct inode *inode = NULL; /* the journal inode */ | |
764 | journal_t *j_journal = NULL; | |
765 | struct ocfs2_dinode *di = NULL; | |
766 | struct buffer_head *bh = NULL; | |
767 | struct ocfs2_super *osb; | |
e63aecb6 | 768 | int inode_lock = 0; |
ccd979bd MF |
769 | |
770 | mlog_entry_void(); | |
771 | ||
772 | BUG_ON(!journal); | |
773 | ||
774 | osb = journal->j_osb; | |
775 | ||
776 | /* already have the inode for our journal */ | |
777 | inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, | |
778 | osb->slot_num); | |
779 | if (inode == NULL) { | |
780 | status = -EACCES; | |
781 | mlog_errno(status); | |
782 | goto done; | |
783 | } | |
784 | if (is_bad_inode(inode)) { | |
785 | mlog(ML_ERROR, "access error (bad inode)\n"); | |
786 | iput(inode); | |
787 | inode = NULL; | |
788 | status = -EACCES; | |
789 | goto done; | |
790 | } | |
791 | ||
792 | SET_INODE_JOURNAL(inode); | |
793 | OCFS2_I(inode)->ip_open_count++; | |
794 | ||
6eff5790 MF |
795 | /* Skip recovery waits here - journal inode metadata never |
796 | * changes in a live cluster so it can be considered an | |
797 | * exception to the rule. */ | |
e63aecb6 | 798 | status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY); |
ccd979bd MF |
799 | if (status < 0) { |
800 | if (status != -ERESTARTSYS) | |
801 | mlog(ML_ERROR, "Could not get lock on journal!\n"); | |
802 | goto done; | |
803 | } | |
804 | ||
e63aecb6 | 805 | inode_lock = 1; |
ccd979bd MF |
806 | di = (struct ocfs2_dinode *)bh->b_data; |
807 | ||
808 | if (inode->i_size < OCFS2_MIN_JOURNAL_SIZE) { | |
809 | mlog(ML_ERROR, "Journal file size (%lld) is too small!\n", | |
810 | inode->i_size); | |
811 | status = -EINVAL; | |
812 | goto done; | |
813 | } | |
814 | ||
815 | mlog(0, "inode->i_size = %lld\n", inode->i_size); | |
5515eff8 AM |
816 | mlog(0, "inode->i_blocks = %llu\n", |
817 | (unsigned long long)inode->i_blocks); | |
ccd979bd MF |
818 | mlog(0, "inode->ip_clusters = %u\n", OCFS2_I(inode)->ip_clusters); |
819 | ||
820 | /* call the kernels journal init function now */ | |
2b4e30fb | 821 | j_journal = jbd2_journal_init_inode(inode); |
ccd979bd MF |
822 | if (j_journal == NULL) { |
823 | mlog(ML_ERROR, "Linux journal layer error\n"); | |
824 | status = -EINVAL; | |
825 | goto done; | |
826 | } | |
827 | ||
2b4e30fb | 828 | mlog(0, "Returned from jbd2_journal_init_inode\n"); |
ccd979bd MF |
829 | mlog(0, "j_journal->j_maxlen = %u\n", j_journal->j_maxlen); |
830 | ||
831 | *dirty = (le32_to_cpu(di->id1.journal1.ij_flags) & | |
832 | OCFS2_JOURNAL_DIRTY_FL); | |
833 | ||
834 | journal->j_journal = j_journal; | |
835 | journal->j_inode = inode; | |
836 | journal->j_bh = bh; | |
837 | ||
838 | ocfs2_set_journal_params(osb); | |
839 | ||
840 | journal->j_state = OCFS2_JOURNAL_LOADED; | |
841 | ||
842 | status = 0; | |
843 | done: | |
844 | if (status < 0) { | |
e63aecb6 MF |
845 | if (inode_lock) |
846 | ocfs2_inode_unlock(inode, 1); | |
a81cb88b | 847 | brelse(bh); |
ccd979bd MF |
848 | if (inode) { |
849 | OCFS2_I(inode)->ip_open_count--; | |
850 | iput(inode); | |
851 | } | |
852 | } | |
853 | ||
854 | mlog_exit(status); | |
855 | return status; | |
856 | } | |
857 | ||
539d8264 SM |
858 | static void ocfs2_bump_recovery_generation(struct ocfs2_dinode *di) |
859 | { | |
860 | le32_add_cpu(&(di->id1.journal1.ij_recovery_generation), 1); | |
861 | } | |
862 | ||
863 | static u32 ocfs2_get_recovery_generation(struct ocfs2_dinode *di) | |
864 | { | |
865 | return le32_to_cpu(di->id1.journal1.ij_recovery_generation); | |
866 | } | |
867 | ||
ccd979bd | 868 | static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb, |
539d8264 | 869 | int dirty, int replayed) |
ccd979bd MF |
870 | { |
871 | int status; | |
872 | unsigned int flags; | |
873 | struct ocfs2_journal *journal = osb->journal; | |
874 | struct buffer_head *bh = journal->j_bh; | |
875 | struct ocfs2_dinode *fe; | |
876 | ||
877 | mlog_entry_void(); | |
878 | ||
879 | fe = (struct ocfs2_dinode *)bh->b_data; | |
10995aa2 JB |
880 | |
881 | /* The journal bh on the osb always comes from ocfs2_journal_init() | |
882 | * and was validated there inside ocfs2_inode_lock_full(). It's a | |
883 | * code bug if we mess it up. */ | |
884 | BUG_ON(!OCFS2_IS_VALID_DINODE(fe)); | |
ccd979bd MF |
885 | |
886 | flags = le32_to_cpu(fe->id1.journal1.ij_flags); | |
887 | if (dirty) | |
888 | flags |= OCFS2_JOURNAL_DIRTY_FL; | |
889 | else | |
890 | flags &= ~OCFS2_JOURNAL_DIRTY_FL; | |
891 | fe->id1.journal1.ij_flags = cpu_to_le32(flags); | |
892 | ||
539d8264 SM |
893 | if (replayed) |
894 | ocfs2_bump_recovery_generation(fe); | |
895 | ||
13723d00 | 896 | ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check); |
ccd979bd MF |
897 | status = ocfs2_write_block(osb, bh, journal->j_inode); |
898 | if (status < 0) | |
899 | mlog_errno(status); | |
900 | ||
ccd979bd MF |
901 | mlog_exit(status); |
902 | return status; | |
903 | } | |
904 | ||
905 | /* | |
906 | * If the journal has been kmalloc'd it needs to be freed after this | |
907 | * call. | |
908 | */ | |
909 | void ocfs2_journal_shutdown(struct ocfs2_super *osb) | |
910 | { | |
911 | struct ocfs2_journal *journal = NULL; | |
912 | int status = 0; | |
913 | struct inode *inode = NULL; | |
914 | int num_running_trans = 0; | |
915 | ||
916 | mlog_entry_void(); | |
917 | ||
ebdec83b | 918 | BUG_ON(!osb); |
ccd979bd MF |
919 | |
920 | journal = osb->journal; | |
921 | if (!journal) | |
922 | goto done; | |
923 | ||
924 | inode = journal->j_inode; | |
925 | ||
926 | if (journal->j_state != OCFS2_JOURNAL_LOADED) | |
927 | goto done; | |
928 | ||
2b4e30fb | 929 | /* need to inc inode use count - jbd2_journal_destroy will iput. */ |
ccd979bd MF |
930 | if (!igrab(inode)) |
931 | BUG(); | |
932 | ||
933 | num_running_trans = atomic_read(&(osb->journal->j_num_trans)); | |
934 | if (num_running_trans > 0) | |
935 | mlog(0, "Shutting down journal: must wait on %d " | |
936 | "running transactions!\n", | |
937 | num_running_trans); | |
938 | ||
939 | /* Do a commit_cache here. It will flush our journal, *and* | |
940 | * release any locks that are still held. | |
941 | * set the SHUTDOWN flag and release the trans lock. | |
942 | * the commit thread will take the trans lock for us below. */ | |
943 | journal->j_state = OCFS2_JOURNAL_IN_SHUTDOWN; | |
944 | ||
945 | /* The OCFS2_JOURNAL_IN_SHUTDOWN will signal to commit_cache to not | |
946 | * drop the trans_lock (which we want to hold until we | |
947 | * completely destroy the journal. */ | |
948 | if (osb->commit_task) { | |
949 | /* Wait for the commit thread */ | |
950 | mlog(0, "Waiting for ocfs2commit to exit....\n"); | |
951 | kthread_stop(osb->commit_task); | |
952 | osb->commit_task = NULL; | |
953 | } | |
954 | ||
955 | BUG_ON(atomic_read(&(osb->journal->j_num_trans)) != 0); | |
956 | ||
c271c5c2 | 957 | if (ocfs2_mount_local(osb)) { |
2b4e30fb JB |
958 | jbd2_journal_lock_updates(journal->j_journal); |
959 | status = jbd2_journal_flush(journal->j_journal); | |
960 | jbd2_journal_unlock_updates(journal->j_journal); | |
c271c5c2 SM |
961 | if (status < 0) |
962 | mlog_errno(status); | |
963 | } | |
964 | ||
965 | if (status == 0) { | |
966 | /* | |
967 | * Do not toggle if flush was unsuccessful otherwise | |
968 | * will leave dirty metadata in a "clean" journal | |
969 | */ | |
539d8264 | 970 | status = ocfs2_journal_toggle_dirty(osb, 0, 0); |
c271c5c2 SM |
971 | if (status < 0) |
972 | mlog_errno(status); | |
973 | } | |
ccd979bd MF |
974 | |
975 | /* Shutdown the kernel journal system */ | |
2b4e30fb | 976 | jbd2_journal_destroy(journal->j_journal); |
ae0dff68 | 977 | journal->j_journal = NULL; |
ccd979bd MF |
978 | |
979 | OCFS2_I(inode)->ip_open_count--; | |
980 | ||
981 | /* unlock our journal */ | |
e63aecb6 | 982 | ocfs2_inode_unlock(inode, 1); |
ccd979bd MF |
983 | |
984 | brelse(journal->j_bh); | |
985 | journal->j_bh = NULL; | |
986 | ||
987 | journal->j_state = OCFS2_JOURNAL_FREE; | |
988 | ||
989 | // up_write(&journal->j_trans_barrier); | |
990 | done: | |
991 | if (inode) | |
992 | iput(inode); | |
993 | mlog_exit_void(); | |
994 | } | |
995 | ||
996 | static void ocfs2_clear_journal_error(struct super_block *sb, | |
997 | journal_t *journal, | |
998 | int slot) | |
999 | { | |
1000 | int olderr; | |
1001 | ||
2b4e30fb | 1002 | olderr = jbd2_journal_errno(journal); |
ccd979bd MF |
1003 | if (olderr) { |
1004 | mlog(ML_ERROR, "File system error %d recorded in " | |
1005 | "journal %u.\n", olderr, slot); | |
1006 | mlog(ML_ERROR, "File system on device %s needs checking.\n", | |
1007 | sb->s_id); | |
1008 | ||
2b4e30fb JB |
1009 | jbd2_journal_ack_err(journal); |
1010 | jbd2_journal_clear_err(journal); | |
ccd979bd MF |
1011 | } |
1012 | } | |
1013 | ||
539d8264 | 1014 | int ocfs2_journal_load(struct ocfs2_journal *journal, int local, int replayed) |
ccd979bd MF |
1015 | { |
1016 | int status = 0; | |
1017 | struct ocfs2_super *osb; | |
1018 | ||
1019 | mlog_entry_void(); | |
1020 | ||
b1f3550f | 1021 | BUG_ON(!journal); |
ccd979bd MF |
1022 | |
1023 | osb = journal->j_osb; | |
1024 | ||
2b4e30fb | 1025 | status = jbd2_journal_load(journal->j_journal); |
ccd979bd MF |
1026 | if (status < 0) { |
1027 | mlog(ML_ERROR, "Failed to load journal!\n"); | |
1028 | goto done; | |
1029 | } | |
1030 | ||
1031 | ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num); | |
1032 | ||
539d8264 | 1033 | status = ocfs2_journal_toggle_dirty(osb, 1, replayed); |
ccd979bd MF |
1034 | if (status < 0) { |
1035 | mlog_errno(status); | |
1036 | goto done; | |
1037 | } | |
1038 | ||
1039 | /* Launch the commit thread */ | |
c271c5c2 SM |
1040 | if (!local) { |
1041 | osb->commit_task = kthread_run(ocfs2_commit_thread, osb, | |
1042 | "ocfs2cmt"); | |
1043 | if (IS_ERR(osb->commit_task)) { | |
1044 | status = PTR_ERR(osb->commit_task); | |
1045 | osb->commit_task = NULL; | |
1046 | mlog(ML_ERROR, "unable to launch ocfs2commit thread, " | |
1047 | "error=%d", status); | |
1048 | goto done; | |
1049 | } | |
1050 | } else | |
ccd979bd | 1051 | osb->commit_task = NULL; |
ccd979bd MF |
1052 | |
1053 | done: | |
1054 | mlog_exit(status); | |
1055 | return status; | |
1056 | } | |
1057 | ||
1058 | ||
1059 | /* 'full' flag tells us whether we clear out all blocks or if we just | |
1060 | * mark the journal clean */ | |
1061 | int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full) | |
1062 | { | |
1063 | int status; | |
1064 | ||
1065 | mlog_entry_void(); | |
1066 | ||
ebdec83b | 1067 | BUG_ON(!journal); |
ccd979bd | 1068 | |
2b4e30fb | 1069 | status = jbd2_journal_wipe(journal->j_journal, full); |
ccd979bd MF |
1070 | if (status < 0) { |
1071 | mlog_errno(status); | |
1072 | goto bail; | |
1073 | } | |
1074 | ||
539d8264 | 1075 | status = ocfs2_journal_toggle_dirty(journal->j_osb, 0, 0); |
ccd979bd MF |
1076 | if (status < 0) |
1077 | mlog_errno(status); | |
1078 | ||
1079 | bail: | |
1080 | mlog_exit(status); | |
1081 | return status; | |
1082 | } | |
1083 | ||
553abd04 JB |
1084 | static int ocfs2_recovery_completed(struct ocfs2_super *osb) |
1085 | { | |
1086 | int empty; | |
1087 | struct ocfs2_recovery_map *rm = osb->recovery_map; | |
1088 | ||
1089 | spin_lock(&osb->osb_lock); | |
1090 | empty = (rm->rm_used == 0); | |
1091 | spin_unlock(&osb->osb_lock); | |
1092 | ||
1093 | return empty; | |
1094 | } | |
1095 | ||
1096 | void ocfs2_wait_for_recovery(struct ocfs2_super *osb) | |
1097 | { | |
1098 | wait_event(osb->recovery_event, ocfs2_recovery_completed(osb)); | |
1099 | } | |
1100 | ||
ccd979bd MF |
1101 | /* |
1102 | * JBD Might read a cached version of another nodes journal file. We | |
1103 | * don't want this as this file changes often and we get no | |
1104 | * notification on those changes. The only way to be sure that we've | |
1105 | * got the most up to date version of those blocks then is to force | |
1106 | * read them off disk. Just searching through the buffer cache won't | |
1107 | * work as there may be pages backing this file which are still marked | |
1108 | * up to date. We know things can't change on this file underneath us | |
1109 | * as we have the lock by now :) | |
1110 | */ | |
1111 | static int ocfs2_force_read_journal(struct inode *inode) | |
1112 | { | |
1113 | int status = 0; | |
4f902c37 | 1114 | int i; |
8110b073 | 1115 | u64 v_blkno, p_blkno, p_blocks, num_blocks; |
4f902c37 | 1116 | #define CONCURRENT_JOURNAL_FILL 32ULL |
ccd979bd MF |
1117 | struct buffer_head *bhs[CONCURRENT_JOURNAL_FILL]; |
1118 | ||
1119 | mlog_entry_void(); | |
1120 | ||
ccd979bd MF |
1121 | memset(bhs, 0, sizeof(struct buffer_head *) * CONCURRENT_JOURNAL_FILL); |
1122 | ||
8110b073 | 1123 | num_blocks = ocfs2_blocks_for_bytes(inode->i_sb, inode->i_size); |
ccd979bd | 1124 | v_blkno = 0; |
8110b073 | 1125 | while (v_blkno < num_blocks) { |
ccd979bd | 1126 | status = ocfs2_extent_map_get_blocks(inode, v_blkno, |
49cb8d2d | 1127 | &p_blkno, &p_blocks, NULL); |
ccd979bd MF |
1128 | if (status < 0) { |
1129 | mlog_errno(status); | |
1130 | goto bail; | |
1131 | } | |
1132 | ||
1133 | if (p_blocks > CONCURRENT_JOURNAL_FILL) | |
1134 | p_blocks = CONCURRENT_JOURNAL_FILL; | |
1135 | ||
dd4a2c2b MF |
1136 | /* We are reading journal data which should not |
1137 | * be put in the uptodate cache */ | |
da1e9098 JB |
1138 | status = ocfs2_read_blocks_sync(OCFS2_SB(inode->i_sb), |
1139 | p_blkno, p_blocks, bhs); | |
ccd979bd MF |
1140 | if (status < 0) { |
1141 | mlog_errno(status); | |
1142 | goto bail; | |
1143 | } | |
1144 | ||
1145 | for(i = 0; i < p_blocks; i++) { | |
1146 | brelse(bhs[i]); | |
1147 | bhs[i] = NULL; | |
1148 | } | |
1149 | ||
1150 | v_blkno += p_blocks; | |
1151 | } | |
1152 | ||
1153 | bail: | |
1154 | for(i = 0; i < CONCURRENT_JOURNAL_FILL; i++) | |
a81cb88b | 1155 | brelse(bhs[i]); |
ccd979bd MF |
1156 | mlog_exit(status); |
1157 | return status; | |
1158 | } | |
1159 | ||
1160 | struct ocfs2_la_recovery_item { | |
1161 | struct list_head lri_list; | |
1162 | int lri_slot; | |
1163 | struct ocfs2_dinode *lri_la_dinode; | |
1164 | struct ocfs2_dinode *lri_tl_dinode; | |
2205363d | 1165 | struct ocfs2_quota_recovery *lri_qrec; |
ccd979bd MF |
1166 | }; |
1167 | ||
1168 | /* Does the second half of the recovery process. By this point, the | |
1169 | * node is marked clean and can actually be considered recovered, | |
1170 | * hence it's no longer in the recovery map, but there's still some | |
1171 | * cleanup we can do which shouldn't happen within the recovery thread | |
1172 | * as locking in that context becomes very difficult if we are to take | |
1173 | * recovering nodes into account. | |
1174 | * | |
1175 | * NOTE: This function can and will sleep on recovery of other nodes | |
1176 | * during cluster locking, just like any other ocfs2 process. | |
1177 | */ | |
c4028958 | 1178 | void ocfs2_complete_recovery(struct work_struct *work) |
ccd979bd MF |
1179 | { |
1180 | int ret; | |
c4028958 DH |
1181 | struct ocfs2_journal *journal = |
1182 | container_of(work, struct ocfs2_journal, j_recovery_work); | |
1183 | struct ocfs2_super *osb = journal->j_osb; | |
ccd979bd | 1184 | struct ocfs2_dinode *la_dinode, *tl_dinode; |
800deef3 | 1185 | struct ocfs2_la_recovery_item *item, *n; |
2205363d | 1186 | struct ocfs2_quota_recovery *qrec; |
ccd979bd MF |
1187 | LIST_HEAD(tmp_la_list); |
1188 | ||
1189 | mlog_entry_void(); | |
1190 | ||
1191 | mlog(0, "completing recovery from keventd\n"); | |
1192 | ||
1193 | spin_lock(&journal->j_lock); | |
1194 | list_splice_init(&journal->j_la_cleanups, &tmp_la_list); | |
1195 | spin_unlock(&journal->j_lock); | |
1196 | ||
800deef3 | 1197 | list_for_each_entry_safe(item, n, &tmp_la_list, lri_list) { |
ccd979bd MF |
1198 | list_del_init(&item->lri_list); |
1199 | ||
1200 | mlog(0, "Complete recovery for slot %d\n", item->lri_slot); | |
1201 | ||
19ece546 JK |
1202 | ocfs2_wait_on_quotas(osb); |
1203 | ||
ccd979bd MF |
1204 | la_dinode = item->lri_la_dinode; |
1205 | if (la_dinode) { | |
b0697053 | 1206 | mlog(0, "Clean up local alloc %llu\n", |
1ca1a111 | 1207 | (unsigned long long)le64_to_cpu(la_dinode->i_blkno)); |
ccd979bd MF |
1208 | |
1209 | ret = ocfs2_complete_local_alloc_recovery(osb, | |
1210 | la_dinode); | |
1211 | if (ret < 0) | |
1212 | mlog_errno(ret); | |
1213 | ||
1214 | kfree(la_dinode); | |
1215 | } | |
1216 | ||
1217 | tl_dinode = item->lri_tl_dinode; | |
1218 | if (tl_dinode) { | |
b0697053 | 1219 | mlog(0, "Clean up truncate log %llu\n", |
1ca1a111 | 1220 | (unsigned long long)le64_to_cpu(tl_dinode->i_blkno)); |
ccd979bd MF |
1221 | |
1222 | ret = ocfs2_complete_truncate_log_recovery(osb, | |
1223 | tl_dinode); | |
1224 | if (ret < 0) | |
1225 | mlog_errno(ret); | |
1226 | ||
1227 | kfree(tl_dinode); | |
1228 | } | |
1229 | ||
1230 | ret = ocfs2_recover_orphans(osb, item->lri_slot); | |
1231 | if (ret < 0) | |
1232 | mlog_errno(ret); | |
1233 | ||
2205363d JK |
1234 | qrec = item->lri_qrec; |
1235 | if (qrec) { | |
1236 | mlog(0, "Recovering quota files"); | |
1237 | ret = ocfs2_finish_quota_recovery(osb, qrec, | |
1238 | item->lri_slot); | |
1239 | if (ret < 0) | |
1240 | mlog_errno(ret); | |
1241 | /* Recovery info is already freed now */ | |
1242 | } | |
1243 | ||
ccd979bd MF |
1244 | kfree(item); |
1245 | } | |
1246 | ||
1247 | mlog(0, "Recovery completion\n"); | |
1248 | mlog_exit_void(); | |
1249 | } | |
1250 | ||
1251 | /* NOTE: This function always eats your references to la_dinode and | |
1252 | * tl_dinode, either manually on error, or by passing them to | |
1253 | * ocfs2_complete_recovery */ | |
1254 | static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal, | |
1255 | int slot_num, | |
1256 | struct ocfs2_dinode *la_dinode, | |
2205363d JK |
1257 | struct ocfs2_dinode *tl_dinode, |
1258 | struct ocfs2_quota_recovery *qrec) | |
ccd979bd MF |
1259 | { |
1260 | struct ocfs2_la_recovery_item *item; | |
1261 | ||
afae00ab | 1262 | item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS); |
ccd979bd MF |
1263 | if (!item) { |
1264 | /* Though we wish to avoid it, we are in fact safe in | |
1265 | * skipping local alloc cleanup as fsck.ocfs2 is more | |
1266 | * than capable of reclaiming unused space. */ | |
1267 | if (la_dinode) | |
1268 | kfree(la_dinode); | |
1269 | ||
1270 | if (tl_dinode) | |
1271 | kfree(tl_dinode); | |
1272 | ||
2205363d JK |
1273 | if (qrec) |
1274 | ocfs2_free_quota_recovery(qrec); | |
1275 | ||
ccd979bd MF |
1276 | mlog_errno(-ENOMEM); |
1277 | return; | |
1278 | } | |
1279 | ||
1280 | INIT_LIST_HEAD(&item->lri_list); | |
1281 | item->lri_la_dinode = la_dinode; | |
1282 | item->lri_slot = slot_num; | |
1283 | item->lri_tl_dinode = tl_dinode; | |
2205363d | 1284 | item->lri_qrec = qrec; |
ccd979bd MF |
1285 | |
1286 | spin_lock(&journal->j_lock); | |
1287 | list_add_tail(&item->lri_list, &journal->j_la_cleanups); | |
1288 | queue_work(ocfs2_wq, &journal->j_recovery_work); | |
1289 | spin_unlock(&journal->j_lock); | |
1290 | } | |
1291 | ||
1292 | /* Called by the mount code to queue recovery the last part of | |
9140db04 | 1293 | * recovery for it's own and offline slot(s). */ |
ccd979bd MF |
1294 | void ocfs2_complete_mount_recovery(struct ocfs2_super *osb) |
1295 | { | |
1296 | struct ocfs2_journal *journal = osb->journal; | |
1297 | ||
9140db04 SE |
1298 | /* No need to queue up our truncate_log as regular cleanup will catch |
1299 | * that */ | |
1300 | ocfs2_queue_recovery_completion(journal, osb->slot_num, | |
1301 | osb->local_alloc_copy, NULL, NULL); | |
1302 | ocfs2_schedule_truncate_log_flush(osb, 0); | |
ccd979bd | 1303 | |
9140db04 SE |
1304 | osb->local_alloc_copy = NULL; |
1305 | osb->dirty = 0; | |
1306 | ||
1307 | /* queue to recover orphan slots for all offline slots */ | |
1308 | ocfs2_replay_map_set_state(osb, REPLAY_NEEDED); | |
1309 | ocfs2_queue_replay_slots(osb); | |
1310 | ocfs2_free_replay_slots(osb); | |
ccd979bd MF |
1311 | } |
1312 | ||
2205363d JK |
1313 | void ocfs2_complete_quota_recovery(struct ocfs2_super *osb) |
1314 | { | |
1315 | if (osb->quota_rec) { | |
1316 | ocfs2_queue_recovery_completion(osb->journal, | |
1317 | osb->slot_num, | |
1318 | NULL, | |
1319 | NULL, | |
1320 | osb->quota_rec); | |
1321 | osb->quota_rec = NULL; | |
1322 | } | |
1323 | } | |
1324 | ||
ccd979bd MF |
1325 | static int __ocfs2_recovery_thread(void *arg) |
1326 | { | |
2205363d | 1327 | int status, node_num, slot_num; |
ccd979bd | 1328 | struct ocfs2_super *osb = arg; |
553abd04 | 1329 | struct ocfs2_recovery_map *rm = osb->recovery_map; |
2205363d JK |
1330 | int *rm_quota = NULL; |
1331 | int rm_quota_used = 0, i; | |
1332 | struct ocfs2_quota_recovery *qrec; | |
ccd979bd MF |
1333 | |
1334 | mlog_entry_void(); | |
1335 | ||
1336 | status = ocfs2_wait_on_mount(osb); | |
1337 | if (status < 0) { | |
1338 | goto bail; | |
1339 | } | |
1340 | ||
2205363d JK |
1341 | rm_quota = kzalloc(osb->max_slots * sizeof(int), GFP_NOFS); |
1342 | if (!rm_quota) { | |
1343 | status = -ENOMEM; | |
1344 | goto bail; | |
1345 | } | |
ccd979bd MF |
1346 | restart: |
1347 | status = ocfs2_super_lock(osb, 1); | |
1348 | if (status < 0) { | |
1349 | mlog_errno(status); | |
1350 | goto bail; | |
1351 | } | |
1352 | ||
9140db04 SE |
1353 | status = ocfs2_compute_replay_slots(osb); |
1354 | if (status < 0) | |
1355 | mlog_errno(status); | |
1356 | ||
1357 | /* queue recovery for our own slot */ | |
1358 | ocfs2_queue_recovery_completion(osb->journal, osb->slot_num, NULL, | |
1359 | NULL, NULL); | |
1360 | ||
553abd04 JB |
1361 | spin_lock(&osb->osb_lock); |
1362 | while (rm->rm_used) { | |
1363 | /* It's always safe to remove entry zero, as we won't | |
1364 | * clear it until ocfs2_recover_node() has succeeded. */ | |
1365 | node_num = rm->rm_entries[0]; | |
1366 | spin_unlock(&osb->osb_lock); | |
2205363d JK |
1367 | mlog(0, "checking node %d\n", node_num); |
1368 | slot_num = ocfs2_node_num_to_slot(osb, node_num); | |
1369 | if (slot_num == -ENOENT) { | |
1370 | status = 0; | |
1371 | mlog(0, "no slot for this node, so no recovery" | |
1372 | "required.\n"); | |
1373 | goto skip_recovery; | |
1374 | } | |
1375 | mlog(0, "node %d was using slot %d\n", node_num, slot_num); | |
1376 | ||
1377 | /* It is a bit subtle with quota recovery. We cannot do it | |
1378 | * immediately because we have to obtain cluster locks from | |
1379 | * quota files and we also don't want to just skip it because | |
1380 | * then quota usage would be out of sync until some node takes | |
1381 | * the slot. So we remember which nodes need quota recovery | |
1382 | * and when everything else is done, we recover quotas. */ | |
1383 | for (i = 0; i < rm_quota_used && rm_quota[i] != slot_num; i++); | |
1384 | if (i == rm_quota_used) | |
1385 | rm_quota[rm_quota_used++] = slot_num; | |
1386 | ||
1387 | status = ocfs2_recover_node(osb, node_num, slot_num); | |
1388 | skip_recovery: | |
553abd04 JB |
1389 | if (!status) { |
1390 | ocfs2_recovery_map_clear(osb, node_num); | |
1391 | } else { | |
ccd979bd MF |
1392 | mlog(ML_ERROR, |
1393 | "Error %d recovering node %d on device (%u,%u)!\n", | |
1394 | status, node_num, | |
1395 | MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev)); | |
1396 | mlog(ML_ERROR, "Volume requires unmount.\n"); | |
ccd979bd MF |
1397 | } |
1398 | ||
553abd04 | 1399 | spin_lock(&osb->osb_lock); |
ccd979bd | 1400 | } |
553abd04 JB |
1401 | spin_unlock(&osb->osb_lock); |
1402 | mlog(0, "All nodes recovered\n"); | |
1403 | ||
539d8264 SM |
1404 | /* Refresh all journal recovery generations from disk */ |
1405 | status = ocfs2_check_journals_nolocks(osb); | |
1406 | status = (status == -EROFS) ? 0 : status; | |
1407 | if (status < 0) | |
1408 | mlog_errno(status); | |
1409 | ||
2205363d JK |
1410 | /* Now it is right time to recover quotas... We have to do this under |
1411 | * superblock lock so that noone can start using the slot (and crash) | |
1412 | * before we recover it */ | |
1413 | for (i = 0; i < rm_quota_used; i++) { | |
1414 | qrec = ocfs2_begin_quota_recovery(osb, rm_quota[i]); | |
1415 | if (IS_ERR(qrec)) { | |
1416 | status = PTR_ERR(qrec); | |
1417 | mlog_errno(status); | |
1418 | continue; | |
1419 | } | |
1420 | ocfs2_queue_recovery_completion(osb->journal, rm_quota[i], | |
1421 | NULL, NULL, qrec); | |
1422 | } | |
1423 | ||
ccd979bd MF |
1424 | ocfs2_super_unlock(osb, 1); |
1425 | ||
9140db04 SE |
1426 | /* queue recovery for offline slots */ |
1427 | ocfs2_queue_replay_slots(osb); | |
ccd979bd MF |
1428 | |
1429 | bail: | |
c74ec2f7 | 1430 | mutex_lock(&osb->recovery_lock); |
553abd04 | 1431 | if (!status && !ocfs2_recovery_completed(osb)) { |
c74ec2f7 | 1432 | mutex_unlock(&osb->recovery_lock); |
ccd979bd MF |
1433 | goto restart; |
1434 | } | |
1435 | ||
9140db04 | 1436 | ocfs2_free_replay_slots(osb); |
ccd979bd MF |
1437 | osb->recovery_thread_task = NULL; |
1438 | mb(); /* sync with ocfs2_recovery_thread_running */ | |
1439 | wake_up(&osb->recovery_event); | |
1440 | ||
c74ec2f7 | 1441 | mutex_unlock(&osb->recovery_lock); |
ccd979bd | 1442 | |
2205363d JK |
1443 | if (rm_quota) |
1444 | kfree(rm_quota); | |
1445 | ||
ccd979bd MF |
1446 | mlog_exit(status); |
1447 | /* no one is callint kthread_stop() for us so the kthread() api | |
1448 | * requires that we call do_exit(). And it isn't exported, but | |
1449 | * complete_and_exit() seems to be a minimal wrapper around it. */ | |
1450 | complete_and_exit(NULL, status); | |
1451 | return status; | |
1452 | } | |
1453 | ||
1454 | void ocfs2_recovery_thread(struct ocfs2_super *osb, int node_num) | |
1455 | { | |
1456 | mlog_entry("(node_num=%d, osb->node_num = %d)\n", | |
1457 | node_num, osb->node_num); | |
1458 | ||
c74ec2f7 | 1459 | mutex_lock(&osb->recovery_lock); |
ccd979bd MF |
1460 | if (osb->disable_recovery) |
1461 | goto out; | |
1462 | ||
1463 | /* People waiting on recovery will wait on | |
1464 | * the recovery map to empty. */ | |
553abd04 JB |
1465 | if (ocfs2_recovery_map_set(osb, node_num)) |
1466 | mlog(0, "node %d already in recovery map.\n", node_num); | |
ccd979bd MF |
1467 | |
1468 | mlog(0, "starting recovery thread...\n"); | |
1469 | ||
1470 | if (osb->recovery_thread_task) | |
1471 | goto out; | |
1472 | ||
1473 | osb->recovery_thread_task = kthread_run(__ocfs2_recovery_thread, osb, | |
78427043 | 1474 | "ocfs2rec"); |
ccd979bd MF |
1475 | if (IS_ERR(osb->recovery_thread_task)) { |
1476 | mlog_errno((int)PTR_ERR(osb->recovery_thread_task)); | |
1477 | osb->recovery_thread_task = NULL; | |
1478 | } | |
1479 | ||
1480 | out: | |
c74ec2f7 | 1481 | mutex_unlock(&osb->recovery_lock); |
ccd979bd MF |
1482 | wake_up(&osb->recovery_event); |
1483 | ||
1484 | mlog_exit_void(); | |
1485 | } | |
1486 | ||
539d8264 SM |
1487 | static int ocfs2_read_journal_inode(struct ocfs2_super *osb, |
1488 | int slot_num, | |
1489 | struct buffer_head **bh, | |
1490 | struct inode **ret_inode) | |
1491 | { | |
1492 | int status = -EACCES; | |
1493 | struct inode *inode = NULL; | |
1494 | ||
1495 | BUG_ON(slot_num >= osb->max_slots); | |
1496 | ||
1497 | inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, | |
1498 | slot_num); | |
1499 | if (!inode || is_bad_inode(inode)) { | |
1500 | mlog_errno(status); | |
1501 | goto bail; | |
1502 | } | |
1503 | SET_INODE_JOURNAL(inode); | |
1504 | ||
b657c95c | 1505 | status = ocfs2_read_inode_block_full(inode, bh, OCFS2_BH_IGNORE_CACHE); |
539d8264 SM |
1506 | if (status < 0) { |
1507 | mlog_errno(status); | |
1508 | goto bail; | |
1509 | } | |
1510 | ||
1511 | status = 0; | |
1512 | ||
1513 | bail: | |
1514 | if (inode) { | |
1515 | if (status || !ret_inode) | |
1516 | iput(inode); | |
1517 | else | |
1518 | *ret_inode = inode; | |
1519 | } | |
1520 | return status; | |
1521 | } | |
1522 | ||
ccd979bd MF |
1523 | /* Does the actual journal replay and marks the journal inode as |
1524 | * clean. Will only replay if the journal inode is marked dirty. */ | |
1525 | static int ocfs2_replay_journal(struct ocfs2_super *osb, | |
1526 | int node_num, | |
1527 | int slot_num) | |
1528 | { | |
1529 | int status; | |
1530 | int got_lock = 0; | |
1531 | unsigned int flags; | |
1532 | struct inode *inode = NULL; | |
1533 | struct ocfs2_dinode *fe; | |
1534 | journal_t *journal = NULL; | |
1535 | struct buffer_head *bh = NULL; | |
539d8264 | 1536 | u32 slot_reco_gen; |
ccd979bd | 1537 | |
539d8264 SM |
1538 | status = ocfs2_read_journal_inode(osb, slot_num, &bh, &inode); |
1539 | if (status) { | |
ccd979bd MF |
1540 | mlog_errno(status); |
1541 | goto done; | |
1542 | } | |
539d8264 SM |
1543 | |
1544 | fe = (struct ocfs2_dinode *)bh->b_data; | |
1545 | slot_reco_gen = ocfs2_get_recovery_generation(fe); | |
1546 | brelse(bh); | |
1547 | bh = NULL; | |
1548 | ||
1549 | /* | |
1550 | * As the fs recovery is asynchronous, there is a small chance that | |
1551 | * another node mounted (and recovered) the slot before the recovery | |
1552 | * thread could get the lock. To handle that, we dirty read the journal | |
1553 | * inode for that slot to get the recovery generation. If it is | |
1554 | * different than what we expected, the slot has been recovered. | |
1555 | * If not, it needs recovery. | |
1556 | */ | |
1557 | if (osb->slot_recovery_generations[slot_num] != slot_reco_gen) { | |
1558 | mlog(0, "Slot %u already recovered (old/new=%u/%u)\n", slot_num, | |
1559 | osb->slot_recovery_generations[slot_num], slot_reco_gen); | |
1560 | osb->slot_recovery_generations[slot_num] = slot_reco_gen; | |
1561 | status = -EBUSY; | |
ccd979bd MF |
1562 | goto done; |
1563 | } | |
539d8264 SM |
1564 | |
1565 | /* Continue with recovery as the journal has not yet been recovered */ | |
ccd979bd | 1566 | |
e63aecb6 | 1567 | status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY); |
ccd979bd | 1568 | if (status < 0) { |
e63aecb6 | 1569 | mlog(0, "status returned from ocfs2_inode_lock=%d\n", status); |
ccd979bd MF |
1570 | if (status != -ERESTARTSYS) |
1571 | mlog(ML_ERROR, "Could not lock journal!\n"); | |
1572 | goto done; | |
1573 | } | |
1574 | got_lock = 1; | |
1575 | ||
1576 | fe = (struct ocfs2_dinode *) bh->b_data; | |
1577 | ||
1578 | flags = le32_to_cpu(fe->id1.journal1.ij_flags); | |
539d8264 | 1579 | slot_reco_gen = ocfs2_get_recovery_generation(fe); |
ccd979bd MF |
1580 | |
1581 | if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) { | |
1582 | mlog(0, "No recovery required for node %d\n", node_num); | |
539d8264 SM |
1583 | /* Refresh recovery generation for the slot */ |
1584 | osb->slot_recovery_generations[slot_num] = slot_reco_gen; | |
ccd979bd MF |
1585 | goto done; |
1586 | } | |
1587 | ||
9140db04 SE |
1588 | /* we need to run complete recovery for offline orphan slots */ |
1589 | ocfs2_replay_map_set_state(osb, REPLAY_NEEDED); | |
1590 | ||
ccd979bd MF |
1591 | mlog(ML_NOTICE, "Recovering node %d from slot %d on device (%u,%u)\n", |
1592 | node_num, slot_num, | |
1593 | MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev)); | |
1594 | ||
1595 | OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters); | |
1596 | ||
1597 | status = ocfs2_force_read_journal(inode); | |
1598 | if (status < 0) { | |
1599 | mlog_errno(status); | |
1600 | goto done; | |
1601 | } | |
1602 | ||
1603 | mlog(0, "calling journal_init_inode\n"); | |
2b4e30fb | 1604 | journal = jbd2_journal_init_inode(inode); |
ccd979bd MF |
1605 | if (journal == NULL) { |
1606 | mlog(ML_ERROR, "Linux journal layer error\n"); | |
1607 | status = -EIO; | |
1608 | goto done; | |
1609 | } | |
1610 | ||
2b4e30fb | 1611 | status = jbd2_journal_load(journal); |
ccd979bd MF |
1612 | if (status < 0) { |
1613 | mlog_errno(status); | |
1614 | if (!igrab(inode)) | |
1615 | BUG(); | |
2b4e30fb | 1616 | jbd2_journal_destroy(journal); |
ccd979bd MF |
1617 | goto done; |
1618 | } | |
1619 | ||
1620 | ocfs2_clear_journal_error(osb->sb, journal, slot_num); | |
1621 | ||
1622 | /* wipe the journal */ | |
1623 | mlog(0, "flushing the journal.\n"); | |
2b4e30fb JB |
1624 | jbd2_journal_lock_updates(journal); |
1625 | status = jbd2_journal_flush(journal); | |
1626 | jbd2_journal_unlock_updates(journal); | |
ccd979bd MF |
1627 | if (status < 0) |
1628 | mlog_errno(status); | |
1629 | ||
1630 | /* This will mark the node clean */ | |
1631 | flags = le32_to_cpu(fe->id1.journal1.ij_flags); | |
1632 | flags &= ~OCFS2_JOURNAL_DIRTY_FL; | |
1633 | fe->id1.journal1.ij_flags = cpu_to_le32(flags); | |
1634 | ||
539d8264 SM |
1635 | /* Increment recovery generation to indicate successful recovery */ |
1636 | ocfs2_bump_recovery_generation(fe); | |
1637 | osb->slot_recovery_generations[slot_num] = | |
1638 | ocfs2_get_recovery_generation(fe); | |
1639 | ||
13723d00 | 1640 | ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check); |
ccd979bd MF |
1641 | status = ocfs2_write_block(osb, bh, inode); |
1642 | if (status < 0) | |
1643 | mlog_errno(status); | |
1644 | ||
1645 | if (!igrab(inode)) | |
1646 | BUG(); | |
1647 | ||
2b4e30fb | 1648 | jbd2_journal_destroy(journal); |
ccd979bd MF |
1649 | |
1650 | done: | |
1651 | /* drop the lock on this nodes journal */ | |
1652 | if (got_lock) | |
e63aecb6 | 1653 | ocfs2_inode_unlock(inode, 1); |
ccd979bd MF |
1654 | |
1655 | if (inode) | |
1656 | iput(inode); | |
1657 | ||
a81cb88b | 1658 | brelse(bh); |
ccd979bd MF |
1659 | |
1660 | mlog_exit(status); | |
1661 | return status; | |
1662 | } | |
1663 | ||
1664 | /* | |
1665 | * Do the most important parts of node recovery: | |
1666 | * - Replay it's journal | |
1667 | * - Stamp a clean local allocator file | |
1668 | * - Stamp a clean truncate log | |
1669 | * - Mark the node clean | |
1670 | * | |
1671 | * If this function completes without error, a node in OCFS2 can be | |
1672 | * said to have been safely recovered. As a result, failure during the | |
1673 | * second part of a nodes recovery process (local alloc recovery) is | |
1674 | * far less concerning. | |
1675 | */ | |
1676 | static int ocfs2_recover_node(struct ocfs2_super *osb, | |
2205363d | 1677 | int node_num, int slot_num) |
ccd979bd MF |
1678 | { |
1679 | int status = 0; | |
ccd979bd MF |
1680 | struct ocfs2_dinode *la_copy = NULL; |
1681 | struct ocfs2_dinode *tl_copy = NULL; | |
1682 | ||
2205363d JK |
1683 | mlog_entry("(node_num=%d, slot_num=%d, osb->node_num = %d)\n", |
1684 | node_num, slot_num, osb->node_num); | |
ccd979bd MF |
1685 | |
1686 | /* Should not ever be called to recover ourselves -- in that | |
1687 | * case we should've called ocfs2_journal_load instead. */ | |
ebdec83b | 1688 | BUG_ON(osb->node_num == node_num); |
ccd979bd | 1689 | |
ccd979bd MF |
1690 | status = ocfs2_replay_journal(osb, node_num, slot_num); |
1691 | if (status < 0) { | |
539d8264 SM |
1692 | if (status == -EBUSY) { |
1693 | mlog(0, "Skipping recovery for slot %u (node %u) " | |
1694 | "as another node has recovered it\n", slot_num, | |
1695 | node_num); | |
1696 | status = 0; | |
1697 | goto done; | |
1698 | } | |
ccd979bd MF |
1699 | mlog_errno(status); |
1700 | goto done; | |
1701 | } | |
1702 | ||
1703 | /* Stamp a clean local alloc file AFTER recovering the journal... */ | |
1704 | status = ocfs2_begin_local_alloc_recovery(osb, slot_num, &la_copy); | |
1705 | if (status < 0) { | |
1706 | mlog_errno(status); | |
1707 | goto done; | |
1708 | } | |
1709 | ||
1710 | /* An error from begin_truncate_log_recovery is not | |
1711 | * serious enough to warrant halting the rest of | |
1712 | * recovery. */ | |
1713 | status = ocfs2_begin_truncate_log_recovery(osb, slot_num, &tl_copy); | |
1714 | if (status < 0) | |
1715 | mlog_errno(status); | |
1716 | ||
1717 | /* Likewise, this would be a strange but ultimately not so | |
1718 | * harmful place to get an error... */ | |
8e8a4603 | 1719 | status = ocfs2_clear_slot(osb, slot_num); |
ccd979bd MF |
1720 | if (status < 0) |
1721 | mlog_errno(status); | |
1722 | ||
1723 | /* This will kfree the memory pointed to by la_copy and tl_copy */ | |
1724 | ocfs2_queue_recovery_completion(osb->journal, slot_num, la_copy, | |
2205363d | 1725 | tl_copy, NULL); |
ccd979bd MF |
1726 | |
1727 | status = 0; | |
1728 | done: | |
1729 | ||
1730 | mlog_exit(status); | |
1731 | return status; | |
1732 | } | |
1733 | ||
1734 | /* Test node liveness by trylocking his journal. If we get the lock, | |
1735 | * we drop it here. Return 0 if we got the lock, -EAGAIN if node is | |
1736 | * still alive (we couldn't get the lock) and < 0 on error. */ | |
1737 | static int ocfs2_trylock_journal(struct ocfs2_super *osb, | |
1738 | int slot_num) | |
1739 | { | |
1740 | int status, flags; | |
1741 | struct inode *inode = NULL; | |
1742 | ||
1743 | inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, | |
1744 | slot_num); | |
1745 | if (inode == NULL) { | |
1746 | mlog(ML_ERROR, "access error\n"); | |
1747 | status = -EACCES; | |
1748 | goto bail; | |
1749 | } | |
1750 | if (is_bad_inode(inode)) { | |
1751 | mlog(ML_ERROR, "access error (bad inode)\n"); | |
1752 | iput(inode); | |
1753 | inode = NULL; | |
1754 | status = -EACCES; | |
1755 | goto bail; | |
1756 | } | |
1757 | SET_INODE_JOURNAL(inode); | |
1758 | ||
1759 | flags = OCFS2_META_LOCK_RECOVERY | OCFS2_META_LOCK_NOQUEUE; | |
e63aecb6 | 1760 | status = ocfs2_inode_lock_full(inode, NULL, 1, flags); |
ccd979bd MF |
1761 | if (status < 0) { |
1762 | if (status != -EAGAIN) | |
1763 | mlog_errno(status); | |
1764 | goto bail; | |
1765 | } | |
1766 | ||
e63aecb6 | 1767 | ocfs2_inode_unlock(inode, 1); |
ccd979bd MF |
1768 | bail: |
1769 | if (inode) | |
1770 | iput(inode); | |
1771 | ||
1772 | return status; | |
1773 | } | |
1774 | ||
1775 | /* Call this underneath ocfs2_super_lock. It also assumes that the | |
1776 | * slot info struct has been updated from disk. */ | |
1777 | int ocfs2_mark_dead_nodes(struct ocfs2_super *osb) | |
1778 | { | |
d85b20e4 JB |
1779 | unsigned int node_num; |
1780 | int status, i; | |
a1af7d15 | 1781 | u32 gen; |
539d8264 SM |
1782 | struct buffer_head *bh = NULL; |
1783 | struct ocfs2_dinode *di; | |
ccd979bd MF |
1784 | |
1785 | /* This is called with the super block cluster lock, so we | |
1786 | * know that the slot map can't change underneath us. */ | |
1787 | ||
d85b20e4 | 1788 | for (i = 0; i < osb->max_slots; i++) { |
539d8264 SM |
1789 | /* Read journal inode to get the recovery generation */ |
1790 | status = ocfs2_read_journal_inode(osb, i, &bh, NULL); | |
1791 | if (status) { | |
1792 | mlog_errno(status); | |
1793 | goto bail; | |
1794 | } | |
1795 | di = (struct ocfs2_dinode *)bh->b_data; | |
a1af7d15 | 1796 | gen = ocfs2_get_recovery_generation(di); |
539d8264 SM |
1797 | brelse(bh); |
1798 | bh = NULL; | |
1799 | ||
a1af7d15 MF |
1800 | spin_lock(&osb->osb_lock); |
1801 | osb->slot_recovery_generations[i] = gen; | |
1802 | ||
539d8264 SM |
1803 | mlog(0, "Slot %u recovery generation is %u\n", i, |
1804 | osb->slot_recovery_generations[i]); | |
1805 | ||
a1af7d15 MF |
1806 | if (i == osb->slot_num) { |
1807 | spin_unlock(&osb->osb_lock); | |
ccd979bd | 1808 | continue; |
a1af7d15 | 1809 | } |
d85b20e4 JB |
1810 | |
1811 | status = ocfs2_slot_to_node_num_locked(osb, i, &node_num); | |
a1af7d15 MF |
1812 | if (status == -ENOENT) { |
1813 | spin_unlock(&osb->osb_lock); | |
ccd979bd | 1814 | continue; |
a1af7d15 | 1815 | } |
ccd979bd | 1816 | |
a1af7d15 MF |
1817 | if (__ocfs2_recovery_map_test(osb, node_num)) { |
1818 | spin_unlock(&osb->osb_lock); | |
ccd979bd | 1819 | continue; |
a1af7d15 | 1820 | } |
d85b20e4 | 1821 | spin_unlock(&osb->osb_lock); |
ccd979bd MF |
1822 | |
1823 | /* Ok, we have a slot occupied by another node which | |
1824 | * is not in the recovery map. We trylock his journal | |
1825 | * file here to test if he's alive. */ | |
1826 | status = ocfs2_trylock_journal(osb, i); | |
1827 | if (!status) { | |
1828 | /* Since we're called from mount, we know that | |
1829 | * the recovery thread can't race us on | |
1830 | * setting / checking the recovery bits. */ | |
1831 | ocfs2_recovery_thread(osb, node_num); | |
1832 | } else if ((status < 0) && (status != -EAGAIN)) { | |
1833 | mlog_errno(status); | |
1834 | goto bail; | |
1835 | } | |
ccd979bd | 1836 | } |
ccd979bd MF |
1837 | |
1838 | status = 0; | |
1839 | bail: | |
1840 | mlog_exit(status); | |
1841 | return status; | |
1842 | } | |
1843 | ||
5eae5b96 MF |
1844 | struct ocfs2_orphan_filldir_priv { |
1845 | struct inode *head; | |
1846 | struct ocfs2_super *osb; | |
1847 | }; | |
1848 | ||
1849 | static int ocfs2_orphan_filldir(void *priv, const char *name, int name_len, | |
1850 | loff_t pos, u64 ino, unsigned type) | |
1851 | { | |
1852 | struct ocfs2_orphan_filldir_priv *p = priv; | |
1853 | struct inode *iter; | |
1854 | ||
1855 | if (name_len == 1 && !strncmp(".", name, 1)) | |
1856 | return 0; | |
1857 | if (name_len == 2 && !strncmp("..", name, 2)) | |
1858 | return 0; | |
1859 | ||
1860 | /* Skip bad inodes so that recovery can continue */ | |
1861 | iter = ocfs2_iget(p->osb, ino, | |
5fa0613e | 1862 | OCFS2_FI_FLAG_ORPHAN_RECOVERY, 0); |
5eae5b96 MF |
1863 | if (IS_ERR(iter)) |
1864 | return 0; | |
1865 | ||
1866 | mlog(0, "queue orphan %llu\n", | |
1867 | (unsigned long long)OCFS2_I(iter)->ip_blkno); | |
1868 | /* No locking is required for the next_orphan queue as there | |
1869 | * is only ever a single process doing orphan recovery. */ | |
1870 | OCFS2_I(iter)->ip_next_orphan = p->head; | |
1871 | p->head = iter; | |
1872 | ||
1873 | return 0; | |
1874 | } | |
1875 | ||
b4df6ed8 MF |
1876 | static int ocfs2_queue_orphans(struct ocfs2_super *osb, |
1877 | int slot, | |
1878 | struct inode **head) | |
ccd979bd | 1879 | { |
b4df6ed8 | 1880 | int status; |
ccd979bd | 1881 | struct inode *orphan_dir_inode = NULL; |
5eae5b96 MF |
1882 | struct ocfs2_orphan_filldir_priv priv; |
1883 | loff_t pos = 0; | |
1884 | ||
1885 | priv.osb = osb; | |
1886 | priv.head = *head; | |
ccd979bd MF |
1887 | |
1888 | orphan_dir_inode = ocfs2_get_system_file_inode(osb, | |
1889 | ORPHAN_DIR_SYSTEM_INODE, | |
1890 | slot); | |
1891 | if (!orphan_dir_inode) { | |
1892 | status = -ENOENT; | |
1893 | mlog_errno(status); | |
b4df6ed8 MF |
1894 | return status; |
1895 | } | |
ccd979bd | 1896 | |
1b1dcc1b | 1897 | mutex_lock(&orphan_dir_inode->i_mutex); |
e63aecb6 | 1898 | status = ocfs2_inode_lock(orphan_dir_inode, NULL, 0); |
ccd979bd | 1899 | if (status < 0) { |
ccd979bd MF |
1900 | mlog_errno(status); |
1901 | goto out; | |
1902 | } | |
ccd979bd | 1903 | |
5eae5b96 MF |
1904 | status = ocfs2_dir_foreach(orphan_dir_inode, &pos, &priv, |
1905 | ocfs2_orphan_filldir); | |
1906 | if (status) { | |
1907 | mlog_errno(status); | |
a86370fb | 1908 | goto out_cluster; |
ccd979bd | 1909 | } |
ccd979bd | 1910 | |
5eae5b96 MF |
1911 | *head = priv.head; |
1912 | ||
a86370fb | 1913 | out_cluster: |
e63aecb6 | 1914 | ocfs2_inode_unlock(orphan_dir_inode, 0); |
b4df6ed8 MF |
1915 | out: |
1916 | mutex_unlock(&orphan_dir_inode->i_mutex); | |
ccd979bd | 1917 | iput(orphan_dir_inode); |
b4df6ed8 MF |
1918 | return status; |
1919 | } | |
1920 | ||
1921 | static int ocfs2_orphan_recovery_can_continue(struct ocfs2_super *osb, | |
1922 | int slot) | |
1923 | { | |
1924 | int ret; | |
1925 | ||
1926 | spin_lock(&osb->osb_lock); | |
1927 | ret = !osb->osb_orphan_wipes[slot]; | |
1928 | spin_unlock(&osb->osb_lock); | |
1929 | return ret; | |
1930 | } | |
1931 | ||
1932 | static void ocfs2_mark_recovering_orphan_dir(struct ocfs2_super *osb, | |
1933 | int slot) | |
1934 | { | |
1935 | spin_lock(&osb->osb_lock); | |
1936 | /* Mark ourselves such that new processes in delete_inode() | |
1937 | * know to quit early. */ | |
1938 | ocfs2_node_map_set_bit(osb, &osb->osb_recovering_orphan_dirs, slot); | |
1939 | while (osb->osb_orphan_wipes[slot]) { | |
1940 | /* If any processes are already in the middle of an | |
1941 | * orphan wipe on this dir, then we need to wait for | |
1942 | * them. */ | |
1943 | spin_unlock(&osb->osb_lock); | |
1944 | wait_event_interruptible(osb->osb_wipe_event, | |
1945 | ocfs2_orphan_recovery_can_continue(osb, slot)); | |
1946 | spin_lock(&osb->osb_lock); | |
1947 | } | |
1948 | spin_unlock(&osb->osb_lock); | |
1949 | } | |
1950 | ||
1951 | static void ocfs2_clear_recovering_orphan_dir(struct ocfs2_super *osb, | |
1952 | int slot) | |
1953 | { | |
1954 | ocfs2_node_map_clear_bit(osb, &osb->osb_recovering_orphan_dirs, slot); | |
1955 | } | |
1956 | ||
1957 | /* | |
1958 | * Orphan recovery. Each mounted node has it's own orphan dir which we | |
1959 | * must run during recovery. Our strategy here is to build a list of | |
1960 | * the inodes in the orphan dir and iget/iput them. The VFS does | |
1961 | * (most) of the rest of the work. | |
1962 | * | |
1963 | * Orphan recovery can happen at any time, not just mount so we have a | |
1964 | * couple of extra considerations. | |
1965 | * | |
1966 | * - We grab as many inodes as we can under the orphan dir lock - | |
1967 | * doing iget() outside the orphan dir risks getting a reference on | |
1968 | * an invalid inode. | |
1969 | * - We must be sure not to deadlock with other processes on the | |
1970 | * system wanting to run delete_inode(). This can happen when they go | |
1971 | * to lock the orphan dir and the orphan recovery process attempts to | |
1972 | * iget() inside the orphan dir lock. This can be avoided by | |
1973 | * advertising our state to ocfs2_delete_inode(). | |
1974 | */ | |
1975 | static int ocfs2_recover_orphans(struct ocfs2_super *osb, | |
1976 | int slot) | |
1977 | { | |
1978 | int ret = 0; | |
1979 | struct inode *inode = NULL; | |
1980 | struct inode *iter; | |
1981 | struct ocfs2_inode_info *oi; | |
1982 | ||
1983 | mlog(0, "Recover inodes from orphan dir in slot %d\n", slot); | |
1984 | ||
1985 | ocfs2_mark_recovering_orphan_dir(osb, slot); | |
1986 | ret = ocfs2_queue_orphans(osb, slot, &inode); | |
1987 | ocfs2_clear_recovering_orphan_dir(osb, slot); | |
1988 | ||
1989 | /* Error here should be noted, but we want to continue with as | |
1990 | * many queued inodes as we've got. */ | |
1991 | if (ret) | |
1992 | mlog_errno(ret); | |
ccd979bd MF |
1993 | |
1994 | while (inode) { | |
1995 | oi = OCFS2_I(inode); | |
b0697053 | 1996 | mlog(0, "iput orphan %llu\n", (unsigned long long)oi->ip_blkno); |
ccd979bd MF |
1997 | |
1998 | iter = oi->ip_next_orphan; | |
1999 | ||
2000 | spin_lock(&oi->ip_lock); | |
34d024f8 MF |
2001 | /* The remote delete code may have set these on the |
2002 | * assumption that the other node would wipe them | |
2003 | * successfully. If they are still in the node's | |
2004 | * orphan dir, we need to reset that state. */ | |
ccd979bd MF |
2005 | oi->ip_flags &= ~(OCFS2_INODE_DELETED|OCFS2_INODE_SKIP_DELETE); |
2006 | ||
2007 | /* Set the proper information to get us going into | |
2008 | * ocfs2_delete_inode. */ | |
2009 | oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED; | |
ccd979bd MF |
2010 | spin_unlock(&oi->ip_lock); |
2011 | ||
2012 | iput(inode); | |
2013 | ||
2014 | inode = iter; | |
2015 | } | |
2016 | ||
b4df6ed8 | 2017 | return ret; |
ccd979bd MF |
2018 | } |
2019 | ||
19ece546 | 2020 | static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota) |
ccd979bd MF |
2021 | { |
2022 | /* This check is good because ocfs2 will wait on our recovery | |
2023 | * thread before changing it to something other than MOUNTED | |
2024 | * or DISABLED. */ | |
2025 | wait_event(osb->osb_mount_event, | |
19ece546 JK |
2026 | (!quota && atomic_read(&osb->vol_state) == VOLUME_MOUNTED) || |
2027 | atomic_read(&osb->vol_state) == VOLUME_MOUNTED_QUOTAS || | |
ccd979bd MF |
2028 | atomic_read(&osb->vol_state) == VOLUME_DISABLED); |
2029 | ||
2030 | /* If there's an error on mount, then we may never get to the | |
2031 | * MOUNTED flag, but this is set right before | |
2032 | * dismount_volume() so we can trust it. */ | |
2033 | if (atomic_read(&osb->vol_state) == VOLUME_DISABLED) { | |
2034 | mlog(0, "mount error, exiting!\n"); | |
2035 | return -EBUSY; | |
2036 | } | |
2037 | ||
2038 | return 0; | |
2039 | } | |
2040 | ||
2041 | static int ocfs2_commit_thread(void *arg) | |
2042 | { | |
2043 | int status; | |
2044 | struct ocfs2_super *osb = arg; | |
2045 | struct ocfs2_journal *journal = osb->journal; | |
2046 | ||
2047 | /* we can trust j_num_trans here because _should_stop() is only set in | |
2048 | * shutdown and nobody other than ourselves should be able to start | |
2049 | * transactions. committing on shutdown might take a few iterations | |
2050 | * as final transactions put deleted inodes on the list */ | |
2051 | while (!(kthread_should_stop() && | |
2052 | atomic_read(&journal->j_num_trans) == 0)) { | |
2053 | ||
745ae8ba MF |
2054 | wait_event_interruptible(osb->checkpoint_event, |
2055 | atomic_read(&journal->j_num_trans) | |
2056 | || kthread_should_stop()); | |
ccd979bd MF |
2057 | |
2058 | status = ocfs2_commit_cache(osb); | |
2059 | if (status < 0) | |
2060 | mlog_errno(status); | |
2061 | ||
2062 | if (kthread_should_stop() && atomic_read(&journal->j_num_trans)){ | |
2063 | mlog(ML_KTHREAD, | |
2064 | "commit_thread: %u transactions pending on " | |
2065 | "shutdown\n", | |
2066 | atomic_read(&journal->j_num_trans)); | |
2067 | } | |
2068 | } | |
2069 | ||
2070 | return 0; | |
2071 | } | |
2072 | ||
539d8264 SM |
2073 | /* Reads all the journal inodes without taking any cluster locks. Used |
2074 | * for hard readonly access to determine whether any journal requires | |
2075 | * recovery. Also used to refresh the recovery generation numbers after | |
2076 | * a journal has been recovered by another node. | |
2077 | */ | |
ccd979bd MF |
2078 | int ocfs2_check_journals_nolocks(struct ocfs2_super *osb) |
2079 | { | |
2080 | int ret = 0; | |
2081 | unsigned int slot; | |
539d8264 | 2082 | struct buffer_head *di_bh = NULL; |
ccd979bd | 2083 | struct ocfs2_dinode *di; |
539d8264 | 2084 | int journal_dirty = 0; |
ccd979bd MF |
2085 | |
2086 | for(slot = 0; slot < osb->max_slots; slot++) { | |
539d8264 SM |
2087 | ret = ocfs2_read_journal_inode(osb, slot, &di_bh, NULL); |
2088 | if (ret) { | |
ccd979bd MF |
2089 | mlog_errno(ret); |
2090 | goto out; | |
2091 | } | |
2092 | ||
2093 | di = (struct ocfs2_dinode *) di_bh->b_data; | |
2094 | ||
539d8264 SM |
2095 | osb->slot_recovery_generations[slot] = |
2096 | ocfs2_get_recovery_generation(di); | |
2097 | ||
ccd979bd MF |
2098 | if (le32_to_cpu(di->id1.journal1.ij_flags) & |
2099 | OCFS2_JOURNAL_DIRTY_FL) | |
539d8264 | 2100 | journal_dirty = 1; |
ccd979bd MF |
2101 | |
2102 | brelse(di_bh); | |
539d8264 | 2103 | di_bh = NULL; |
ccd979bd MF |
2104 | } |
2105 | ||
2106 | out: | |
539d8264 SM |
2107 | if (journal_dirty) |
2108 | ret = -EROFS; | |
ccd979bd MF |
2109 | return ret; |
2110 | } |