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> | |
83273932 SE |
31 | #include <linux/time.h> |
32 | #include <linux/random.h> | |
55b465b6 | 33 | #include <linux/delay.h> |
ccd979bd | 34 | |
ccd979bd MF |
35 | #include <cluster/masklog.h> |
36 | ||
37 | #include "ocfs2.h" | |
38 | ||
39 | #include "alloc.h" | |
50655ae9 | 40 | #include "blockcheck.h" |
316f4b9f | 41 | #include "dir.h" |
ccd979bd MF |
42 | #include "dlmglue.h" |
43 | #include "extent_map.h" | |
44 | #include "heartbeat.h" | |
45 | #include "inode.h" | |
46 | #include "journal.h" | |
47 | #include "localalloc.h" | |
ccd979bd MF |
48 | #include "slot_map.h" |
49 | #include "super.h" | |
ccd979bd | 50 | #include "sysfile.h" |
0cf2f763 | 51 | #include "uptodate.h" |
2205363d | 52 | #include "quota.h" |
ccd979bd MF |
53 | |
54 | #include "buffer_head_io.h" | |
b4107950 | 55 | #include "ocfs2_trace.h" |
ccd979bd | 56 | |
34af946a | 57 | DEFINE_SPINLOCK(trans_inc_lock); |
ccd979bd | 58 | |
83273932 SE |
59 | #define ORPHAN_SCAN_SCHEDULE_TIMEOUT 300000 |
60 | ||
ccd979bd MF |
61 | static int ocfs2_force_read_journal(struct inode *inode); |
62 | static int ocfs2_recover_node(struct ocfs2_super *osb, | |
2205363d | 63 | int node_num, int slot_num); |
ccd979bd MF |
64 | static int __ocfs2_recovery_thread(void *arg); |
65 | static int ocfs2_commit_cache(struct ocfs2_super *osb); | |
19ece546 | 66 | static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota); |
ccd979bd | 67 | static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb, |
539d8264 | 68 | int dirty, int replayed); |
ccd979bd MF |
69 | static int ocfs2_trylock_journal(struct ocfs2_super *osb, |
70 | int slot_num); | |
71 | static int ocfs2_recover_orphans(struct ocfs2_super *osb, | |
72 | int slot); | |
73 | static int ocfs2_commit_thread(void *arg); | |
9140db04 SE |
74 | static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal, |
75 | int slot_num, | |
76 | struct ocfs2_dinode *la_dinode, | |
77 | struct ocfs2_dinode *tl_dinode, | |
78 | struct ocfs2_quota_recovery *qrec); | |
ccd979bd | 79 | |
19ece546 JK |
80 | static inline int ocfs2_wait_on_mount(struct ocfs2_super *osb) |
81 | { | |
82 | return __ocfs2_wait_on_mount(osb, 0); | |
83 | } | |
84 | ||
85 | static inline int ocfs2_wait_on_quotas(struct ocfs2_super *osb) | |
86 | { | |
87 | return __ocfs2_wait_on_mount(osb, 1); | |
88 | } | |
89 | ||
9140db04 SE |
90 | /* |
91 | * This replay_map is to track online/offline slots, so we could recover | |
92 | * offline slots during recovery and mount | |
93 | */ | |
94 | ||
95 | enum ocfs2_replay_state { | |
96 | REPLAY_UNNEEDED = 0, /* Replay is not needed, so ignore this map */ | |
97 | REPLAY_NEEDED, /* Replay slots marked in rm_replay_slots */ | |
98 | REPLAY_DONE /* Replay was already queued */ | |
99 | }; | |
100 | ||
101 | struct ocfs2_replay_map { | |
102 | unsigned int rm_slots; | |
103 | enum ocfs2_replay_state rm_state; | |
104 | unsigned char rm_replay_slots[0]; | |
105 | }; | |
106 | ||
107 | void ocfs2_replay_map_set_state(struct ocfs2_super *osb, int state) | |
108 | { | |
109 | if (!osb->replay_map) | |
110 | return; | |
111 | ||
112 | /* If we've already queued the replay, we don't have any more to do */ | |
113 | if (osb->replay_map->rm_state == REPLAY_DONE) | |
114 | return; | |
115 | ||
116 | osb->replay_map->rm_state = state; | |
117 | } | |
118 | ||
119 | int ocfs2_compute_replay_slots(struct ocfs2_super *osb) | |
120 | { | |
121 | struct ocfs2_replay_map *replay_map; | |
122 | int i, node_num; | |
123 | ||
124 | /* If replay map is already set, we don't do it again */ | |
125 | if (osb->replay_map) | |
126 | return 0; | |
127 | ||
128 | replay_map = kzalloc(sizeof(struct ocfs2_replay_map) + | |
129 | (osb->max_slots * sizeof(char)), GFP_KERNEL); | |
130 | ||
131 | if (!replay_map) { | |
132 | mlog_errno(-ENOMEM); | |
133 | return -ENOMEM; | |
134 | } | |
135 | ||
136 | spin_lock(&osb->osb_lock); | |
137 | ||
138 | replay_map->rm_slots = osb->max_slots; | |
139 | replay_map->rm_state = REPLAY_UNNEEDED; | |
140 | ||
141 | /* set rm_replay_slots for offline slot(s) */ | |
142 | for (i = 0; i < replay_map->rm_slots; i++) { | |
143 | if (ocfs2_slot_to_node_num_locked(osb, i, &node_num) == -ENOENT) | |
144 | replay_map->rm_replay_slots[i] = 1; | |
145 | } | |
146 | ||
147 | osb->replay_map = replay_map; | |
148 | spin_unlock(&osb->osb_lock); | |
149 | return 0; | |
150 | } | |
151 | ||
152 | void ocfs2_queue_replay_slots(struct ocfs2_super *osb) | |
153 | { | |
154 | struct ocfs2_replay_map *replay_map = osb->replay_map; | |
155 | int i; | |
156 | ||
157 | if (!replay_map) | |
158 | return; | |
159 | ||
160 | if (replay_map->rm_state != REPLAY_NEEDED) | |
161 | return; | |
162 | ||
163 | for (i = 0; i < replay_map->rm_slots; i++) | |
164 | if (replay_map->rm_replay_slots[i]) | |
165 | ocfs2_queue_recovery_completion(osb->journal, i, NULL, | |
166 | NULL, NULL); | |
167 | replay_map->rm_state = REPLAY_DONE; | |
168 | } | |
169 | ||
170 | void ocfs2_free_replay_slots(struct ocfs2_super *osb) | |
171 | { | |
172 | struct ocfs2_replay_map *replay_map = osb->replay_map; | |
173 | ||
174 | if (!osb->replay_map) | |
175 | return; | |
176 | ||
177 | kfree(replay_map); | |
178 | osb->replay_map = NULL; | |
179 | } | |
180 | ||
553abd04 JB |
181 | int ocfs2_recovery_init(struct ocfs2_super *osb) |
182 | { | |
183 | struct ocfs2_recovery_map *rm; | |
184 | ||
185 | mutex_init(&osb->recovery_lock); | |
186 | osb->disable_recovery = 0; | |
187 | osb->recovery_thread_task = NULL; | |
188 | init_waitqueue_head(&osb->recovery_event); | |
189 | ||
190 | rm = kzalloc(sizeof(struct ocfs2_recovery_map) + | |
191 | osb->max_slots * sizeof(unsigned int), | |
192 | GFP_KERNEL); | |
193 | if (!rm) { | |
194 | mlog_errno(-ENOMEM); | |
195 | return -ENOMEM; | |
196 | } | |
197 | ||
198 | rm->rm_entries = (unsigned int *)((char *)rm + | |
199 | sizeof(struct ocfs2_recovery_map)); | |
200 | osb->recovery_map = rm; | |
201 | ||
202 | return 0; | |
203 | } | |
204 | ||
205 | /* we can't grab the goofy sem lock from inside wait_event, so we use | |
206 | * memory barriers to make sure that we'll see the null task before | |
207 | * being woken up */ | |
208 | static int ocfs2_recovery_thread_running(struct ocfs2_super *osb) | |
209 | { | |
210 | mb(); | |
211 | return osb->recovery_thread_task != NULL; | |
212 | } | |
213 | ||
214 | void ocfs2_recovery_exit(struct ocfs2_super *osb) | |
215 | { | |
216 | struct ocfs2_recovery_map *rm; | |
217 | ||
218 | /* disable any new recovery threads and wait for any currently | |
219 | * running ones to exit. Do this before setting the vol_state. */ | |
220 | mutex_lock(&osb->recovery_lock); | |
221 | osb->disable_recovery = 1; | |
222 | mutex_unlock(&osb->recovery_lock); | |
223 | wait_event(osb->recovery_event, !ocfs2_recovery_thread_running(osb)); | |
224 | ||
225 | /* At this point, we know that no more recovery threads can be | |
226 | * launched, so wait for any recovery completion work to | |
227 | * complete. */ | |
228 | flush_workqueue(ocfs2_wq); | |
229 | ||
230 | /* | |
231 | * Now that recovery is shut down, and the osb is about to be | |
232 | * freed, the osb_lock is not taken here. | |
233 | */ | |
234 | rm = osb->recovery_map; | |
235 | /* XXX: Should we bug if there are dirty entries? */ | |
236 | ||
237 | kfree(rm); | |
238 | } | |
239 | ||
240 | static int __ocfs2_recovery_map_test(struct ocfs2_super *osb, | |
241 | unsigned int node_num) | |
242 | { | |
243 | int i; | |
244 | struct ocfs2_recovery_map *rm = osb->recovery_map; | |
245 | ||
246 | assert_spin_locked(&osb->osb_lock); | |
247 | ||
248 | for (i = 0; i < rm->rm_used; i++) { | |
249 | if (rm->rm_entries[i] == node_num) | |
250 | return 1; | |
251 | } | |
252 | ||
253 | return 0; | |
254 | } | |
255 | ||
256 | /* Behaves like test-and-set. Returns the previous value */ | |
257 | static int ocfs2_recovery_map_set(struct ocfs2_super *osb, | |
258 | unsigned int node_num) | |
259 | { | |
260 | struct ocfs2_recovery_map *rm = osb->recovery_map; | |
261 | ||
262 | spin_lock(&osb->osb_lock); | |
263 | if (__ocfs2_recovery_map_test(osb, node_num)) { | |
264 | spin_unlock(&osb->osb_lock); | |
265 | return 1; | |
266 | } | |
267 | ||
268 | /* XXX: Can this be exploited? Not from o2dlm... */ | |
269 | BUG_ON(rm->rm_used >= osb->max_slots); | |
270 | ||
271 | rm->rm_entries[rm->rm_used] = node_num; | |
272 | rm->rm_used++; | |
273 | spin_unlock(&osb->osb_lock); | |
274 | ||
275 | return 0; | |
276 | } | |
277 | ||
278 | static void ocfs2_recovery_map_clear(struct ocfs2_super *osb, | |
279 | unsigned int node_num) | |
280 | { | |
281 | int i; | |
282 | struct ocfs2_recovery_map *rm = osb->recovery_map; | |
283 | ||
284 | spin_lock(&osb->osb_lock); | |
285 | ||
286 | for (i = 0; i < rm->rm_used; i++) { | |
287 | if (rm->rm_entries[i] == node_num) | |
288 | break; | |
289 | } | |
290 | ||
291 | if (i < rm->rm_used) { | |
292 | /* XXX: be careful with the pointer math */ | |
293 | memmove(&(rm->rm_entries[i]), &(rm->rm_entries[i + 1]), | |
294 | (rm->rm_used - i - 1) * sizeof(unsigned int)); | |
295 | rm->rm_used--; | |
296 | } | |
297 | ||
298 | spin_unlock(&osb->osb_lock); | |
299 | } | |
300 | ||
ccd979bd MF |
301 | static int ocfs2_commit_cache(struct ocfs2_super *osb) |
302 | { | |
303 | int status = 0; | |
304 | unsigned int flushed; | |
ccd979bd MF |
305 | struct ocfs2_journal *journal = NULL; |
306 | ||
ccd979bd MF |
307 | journal = osb->journal; |
308 | ||
309 | /* Flush all pending commits and checkpoint the journal. */ | |
310 | down_write(&journal->j_trans_barrier); | |
311 | ||
b4107950 TM |
312 | flushed = atomic_read(&journal->j_num_trans); |
313 | trace_ocfs2_commit_cache_begin(flushed); | |
314 | if (flushed == 0) { | |
ccd979bd | 315 | up_write(&journal->j_trans_barrier); |
ccd979bd MF |
316 | goto finally; |
317 | } | |
318 | ||
2b4e30fb JB |
319 | jbd2_journal_lock_updates(journal->j_journal); |
320 | status = jbd2_journal_flush(journal->j_journal); | |
321 | jbd2_journal_unlock_updates(journal->j_journal); | |
ccd979bd MF |
322 | if (status < 0) { |
323 | up_write(&journal->j_trans_barrier); | |
324 | mlog_errno(status); | |
325 | goto finally; | |
326 | } | |
327 | ||
f9c57ada | 328 | ocfs2_inc_trans_id(journal); |
ccd979bd MF |
329 | |
330 | flushed = atomic_read(&journal->j_num_trans); | |
331 | atomic_set(&journal->j_num_trans, 0); | |
332 | up_write(&journal->j_trans_barrier); | |
333 | ||
b4107950 | 334 | trace_ocfs2_commit_cache_end(journal->j_trans_id, flushed); |
ccd979bd | 335 | |
34d024f8 | 336 | ocfs2_wake_downconvert_thread(osb); |
ccd979bd MF |
337 | wake_up(&journal->j_checkpointed); |
338 | finally: | |
ccd979bd MF |
339 | return status; |
340 | } | |
341 | ||
1fabe148 | 342 | handle_t *ocfs2_start_trans(struct ocfs2_super *osb, int max_buffs) |
ccd979bd | 343 | { |
ccd979bd | 344 | journal_t *journal = osb->journal->j_journal; |
1fabe148 | 345 | handle_t *handle; |
ccd979bd | 346 | |
ebdec83b | 347 | BUG_ON(!osb || !osb->journal->j_journal); |
ccd979bd | 348 | |
65eff9cc MF |
349 | if (ocfs2_is_hard_readonly(osb)) |
350 | return ERR_PTR(-EROFS); | |
ccd979bd MF |
351 | |
352 | BUG_ON(osb->journal->j_state == OCFS2_JOURNAL_FREE); | |
353 | BUG_ON(max_buffs <= 0); | |
354 | ||
90e86a63 JK |
355 | /* Nested transaction? Just return the handle... */ |
356 | if (journal_current_handle()) | |
357 | return jbd2_journal_start(journal, max_buffs); | |
ccd979bd | 358 | |
fef6925c JK |
359 | sb_start_intwrite(osb->sb); |
360 | ||
ccd979bd MF |
361 | down_read(&osb->journal->j_trans_barrier); |
362 | ||
2b4e30fb | 363 | handle = jbd2_journal_start(journal, max_buffs); |
1fabe148 | 364 | if (IS_ERR(handle)) { |
ccd979bd | 365 | up_read(&osb->journal->j_trans_barrier); |
fef6925c | 366 | sb_end_intwrite(osb->sb); |
ccd979bd | 367 | |
1fabe148 | 368 | mlog_errno(PTR_ERR(handle)); |
ccd979bd MF |
369 | |
370 | if (is_journal_aborted(journal)) { | |
371 | ocfs2_abort(osb->sb, "Detected aborted journal"); | |
1fabe148 | 372 | handle = ERR_PTR(-EROFS); |
ccd979bd | 373 | } |
c271c5c2 SM |
374 | } else { |
375 | if (!ocfs2_mount_local(osb)) | |
376 | atomic_inc(&(osb->journal->j_num_trans)); | |
377 | } | |
ccd979bd | 378 | |
ccd979bd | 379 | return handle; |
ccd979bd MF |
380 | } |
381 | ||
1fabe148 MF |
382 | int ocfs2_commit_trans(struct ocfs2_super *osb, |
383 | handle_t *handle) | |
ccd979bd | 384 | { |
90e86a63 | 385 | int ret, nested; |
02dc1af4 | 386 | struct ocfs2_journal *journal = osb->journal; |
ccd979bd MF |
387 | |
388 | BUG_ON(!handle); | |
389 | ||
90e86a63 | 390 | nested = handle->h_ref > 1; |
2b4e30fb | 391 | ret = jbd2_journal_stop(handle); |
1fabe148 MF |
392 | if (ret < 0) |
393 | mlog_errno(ret); | |
ccd979bd | 394 | |
fef6925c | 395 | if (!nested) { |
90e86a63 | 396 | up_read(&journal->j_trans_barrier); |
fef6925c JK |
397 | sb_end_intwrite(osb->sb); |
398 | } | |
ccd979bd | 399 | |
1fabe148 | 400 | return ret; |
ccd979bd MF |
401 | } |
402 | ||
403 | /* | |
c901fb00 | 404 | * 'nblocks' is what you want to add to the current transaction. |
ccd979bd | 405 | * |
2b4e30fb | 406 | * This might call jbd2_journal_restart() which will commit dirty buffers |
e8aed345 MF |
407 | * and then restart the transaction. Before calling |
408 | * ocfs2_extend_trans(), any changed blocks should have been | |
409 | * dirtied. After calling it, all blocks which need to be changed must | |
410 | * go through another set of journal_access/journal_dirty calls. | |
411 | * | |
ccd979bd MF |
412 | * WARNING: This will not release any semaphores or disk locks taken |
413 | * during the transaction, so make sure they were taken *before* | |
414 | * start_trans or we'll have ordering deadlocks. | |
415 | * | |
416 | * WARNING2: Note that we do *not* drop j_trans_barrier here. This is | |
417 | * good because transaction ids haven't yet been recorded on the | |
418 | * cluster locks associated with this handle. | |
419 | */ | |
1fc58146 | 420 | int ocfs2_extend_trans(handle_t *handle, int nblocks) |
ccd979bd | 421 | { |
c901fb00 | 422 | int status, old_nblocks; |
ccd979bd MF |
423 | |
424 | BUG_ON(!handle); | |
c901fb00 | 425 | BUG_ON(nblocks < 0); |
ccd979bd | 426 | |
c901fb00 TM |
427 | if (!nblocks) |
428 | return 0; | |
429 | ||
430 | old_nblocks = handle->h_buffer_credits; | |
ccd979bd | 431 | |
b4107950 | 432 | trace_ocfs2_extend_trans(old_nblocks, nblocks); |
ccd979bd | 433 | |
e407e397 | 434 | #ifdef CONFIG_OCFS2_DEBUG_FS |
0879c584 MF |
435 | status = 1; |
436 | #else | |
2b4e30fb | 437 | status = jbd2_journal_extend(handle, nblocks); |
ccd979bd MF |
438 | if (status < 0) { |
439 | mlog_errno(status); | |
440 | goto bail; | |
441 | } | |
0879c584 | 442 | #endif |
ccd979bd MF |
443 | |
444 | if (status > 0) { | |
b4107950 | 445 | trace_ocfs2_extend_trans_restart(old_nblocks + nblocks); |
c901fb00 TM |
446 | status = jbd2_journal_restart(handle, |
447 | old_nblocks + nblocks); | |
ccd979bd | 448 | if (status < 0) { |
ccd979bd MF |
449 | mlog_errno(status); |
450 | goto bail; | |
451 | } | |
01ddf1e1 | 452 | } |
ccd979bd MF |
453 | |
454 | status = 0; | |
455 | bail: | |
ccd979bd MF |
456 | return status; |
457 | } | |
458 | ||
2b1e55c3 YL |
459 | /* |
460 | * If we have fewer than thresh credits, extend by OCFS2_MAX_TRANS_DATA. | |
461 | * If that fails, restart the transaction & regain write access for the | |
462 | * buffer head which is used for metadata modifications. | |
463 | * Taken from Ext4: extend_or_restart_transaction() | |
464 | */ | |
465 | int ocfs2_allocate_extend_trans(handle_t *handle, int thresh) | |
466 | { | |
467 | int status, old_nblks; | |
468 | ||
469 | BUG_ON(!handle); | |
470 | ||
471 | old_nblks = handle->h_buffer_credits; | |
472 | trace_ocfs2_allocate_extend_trans(old_nblks, thresh); | |
473 | ||
474 | if (old_nblks < thresh) | |
475 | return 0; | |
476 | ||
477 | status = jbd2_journal_extend(handle, OCFS2_MAX_TRANS_DATA); | |
478 | if (status < 0) { | |
479 | mlog_errno(status); | |
480 | goto bail; | |
481 | } | |
482 | ||
483 | if (status > 0) { | |
484 | status = jbd2_journal_restart(handle, OCFS2_MAX_TRANS_DATA); | |
485 | if (status < 0) | |
486 | mlog_errno(status); | |
487 | } | |
488 | ||
489 | bail: | |
490 | return status; | |
491 | } | |
492 | ||
493 | ||
50655ae9 JB |
494 | struct ocfs2_triggers { |
495 | struct jbd2_buffer_trigger_type ot_triggers; | |
496 | int ot_offset; | |
497 | }; | |
498 | ||
499 | static inline struct ocfs2_triggers *to_ocfs2_trigger(struct jbd2_buffer_trigger_type *triggers) | |
500 | { | |
501 | return container_of(triggers, struct ocfs2_triggers, ot_triggers); | |
502 | } | |
503 | ||
13ceef09 | 504 | static void ocfs2_frozen_trigger(struct jbd2_buffer_trigger_type *triggers, |
50655ae9 JB |
505 | struct buffer_head *bh, |
506 | void *data, size_t size) | |
507 | { | |
508 | struct ocfs2_triggers *ot = to_ocfs2_trigger(triggers); | |
509 | ||
510 | /* | |
511 | * We aren't guaranteed to have the superblock here, so we | |
512 | * must unconditionally compute the ecc data. | |
513 | * __ocfs2_journal_access() will only set the triggers if | |
514 | * metaecc is enabled. | |
515 | */ | |
516 | ocfs2_block_check_compute(data, size, data + ot->ot_offset); | |
517 | } | |
518 | ||
519 | /* | |
520 | * Quota blocks have their own trigger because the struct ocfs2_block_check | |
521 | * offset depends on the blocksize. | |
522 | */ | |
13ceef09 | 523 | static void ocfs2_dq_frozen_trigger(struct jbd2_buffer_trigger_type *triggers, |
50655ae9 JB |
524 | struct buffer_head *bh, |
525 | void *data, size_t size) | |
526 | { | |
527 | struct ocfs2_disk_dqtrailer *dqt = | |
528 | ocfs2_block_dqtrailer(size, data); | |
529 | ||
530 | /* | |
531 | * We aren't guaranteed to have the superblock here, so we | |
532 | * must unconditionally compute the ecc data. | |
533 | * __ocfs2_journal_access() will only set the triggers if | |
534 | * metaecc is enabled. | |
535 | */ | |
536 | ocfs2_block_check_compute(data, size, &dqt->dq_check); | |
537 | } | |
538 | ||
c175a518 JB |
539 | /* |
540 | * Directory blocks also have their own trigger because the | |
541 | * struct ocfs2_block_check offset depends on the blocksize. | |
542 | */ | |
13ceef09 | 543 | static void ocfs2_db_frozen_trigger(struct jbd2_buffer_trigger_type *triggers, |
c175a518 JB |
544 | struct buffer_head *bh, |
545 | void *data, size_t size) | |
546 | { | |
547 | struct ocfs2_dir_block_trailer *trailer = | |
548 | ocfs2_dir_trailer_from_size(size, data); | |
549 | ||
550 | /* | |
551 | * We aren't guaranteed to have the superblock here, so we | |
552 | * must unconditionally compute the ecc data. | |
553 | * __ocfs2_journal_access() will only set the triggers if | |
554 | * metaecc is enabled. | |
555 | */ | |
556 | ocfs2_block_check_compute(data, size, &trailer->db_check); | |
557 | } | |
558 | ||
50655ae9 JB |
559 | static void ocfs2_abort_trigger(struct jbd2_buffer_trigger_type *triggers, |
560 | struct buffer_head *bh) | |
561 | { | |
562 | mlog(ML_ERROR, | |
563 | "ocfs2_abort_trigger called by JBD2. bh = 0x%lx, " | |
564 | "bh->b_blocknr = %llu\n", | |
565 | (unsigned long)bh, | |
566 | (unsigned long long)bh->b_blocknr); | |
567 | ||
568 | /* We aren't guaranteed to have the superblock here - but if we | |
569 | * don't, it'll just crash. */ | |
570 | ocfs2_error(bh->b_assoc_map->host->i_sb, | |
571 | "JBD2 has aborted our journal, ocfs2 cannot continue\n"); | |
572 | } | |
573 | ||
574 | static struct ocfs2_triggers di_triggers = { | |
575 | .ot_triggers = { | |
13ceef09 | 576 | .t_frozen = ocfs2_frozen_trigger, |
50655ae9 JB |
577 | .t_abort = ocfs2_abort_trigger, |
578 | }, | |
579 | .ot_offset = offsetof(struct ocfs2_dinode, i_check), | |
580 | }; | |
581 | ||
582 | static struct ocfs2_triggers eb_triggers = { | |
583 | .ot_triggers = { | |
13ceef09 | 584 | .t_frozen = ocfs2_frozen_trigger, |
50655ae9 JB |
585 | .t_abort = ocfs2_abort_trigger, |
586 | }, | |
587 | .ot_offset = offsetof(struct ocfs2_extent_block, h_check), | |
588 | }; | |
589 | ||
93c97087 TM |
590 | static struct ocfs2_triggers rb_triggers = { |
591 | .ot_triggers = { | |
13ceef09 | 592 | .t_frozen = ocfs2_frozen_trigger, |
93c97087 TM |
593 | .t_abort = ocfs2_abort_trigger, |
594 | }, | |
595 | .ot_offset = offsetof(struct ocfs2_refcount_block, rf_check), | |
596 | }; | |
597 | ||
50655ae9 JB |
598 | static struct ocfs2_triggers gd_triggers = { |
599 | .ot_triggers = { | |
13ceef09 | 600 | .t_frozen = ocfs2_frozen_trigger, |
50655ae9 JB |
601 | .t_abort = ocfs2_abort_trigger, |
602 | }, | |
603 | .ot_offset = offsetof(struct ocfs2_group_desc, bg_check), | |
604 | }; | |
605 | ||
c175a518 JB |
606 | static struct ocfs2_triggers db_triggers = { |
607 | .ot_triggers = { | |
13ceef09 | 608 | .t_frozen = ocfs2_db_frozen_trigger, |
c175a518 JB |
609 | .t_abort = ocfs2_abort_trigger, |
610 | }, | |
611 | }; | |
612 | ||
50655ae9 JB |
613 | static struct ocfs2_triggers xb_triggers = { |
614 | .ot_triggers = { | |
13ceef09 | 615 | .t_frozen = ocfs2_frozen_trigger, |
50655ae9 JB |
616 | .t_abort = ocfs2_abort_trigger, |
617 | }, | |
618 | .ot_offset = offsetof(struct ocfs2_xattr_block, xb_check), | |
619 | }; | |
620 | ||
621 | static struct ocfs2_triggers dq_triggers = { | |
622 | .ot_triggers = { | |
13ceef09 | 623 | .t_frozen = ocfs2_dq_frozen_trigger, |
50655ae9 JB |
624 | .t_abort = ocfs2_abort_trigger, |
625 | }, | |
626 | }; | |
627 | ||
9b7895ef MF |
628 | static struct ocfs2_triggers dr_triggers = { |
629 | .ot_triggers = { | |
13ceef09 | 630 | .t_frozen = ocfs2_frozen_trigger, |
9b7895ef MF |
631 | .t_abort = ocfs2_abort_trigger, |
632 | }, | |
633 | .ot_offset = offsetof(struct ocfs2_dx_root_block, dr_check), | |
634 | }; | |
635 | ||
636 | static struct ocfs2_triggers dl_triggers = { | |
637 | .ot_triggers = { | |
13ceef09 | 638 | .t_frozen = ocfs2_frozen_trigger, |
9b7895ef MF |
639 | .t_abort = ocfs2_abort_trigger, |
640 | }, | |
641 | .ot_offset = offsetof(struct ocfs2_dx_leaf, dl_check), | |
642 | }; | |
643 | ||
50655ae9 | 644 | static int __ocfs2_journal_access(handle_t *handle, |
0cf2f763 | 645 | struct ocfs2_caching_info *ci, |
50655ae9 JB |
646 | struct buffer_head *bh, |
647 | struct ocfs2_triggers *triggers, | |
648 | int type) | |
ccd979bd MF |
649 | { |
650 | int status; | |
0cf2f763 JB |
651 | struct ocfs2_super *osb = |
652 | OCFS2_SB(ocfs2_metadata_cache_get_super(ci)); | |
ccd979bd | 653 | |
0cf2f763 | 654 | BUG_ON(!ci || !ci->ci_ops); |
ccd979bd MF |
655 | BUG_ON(!handle); |
656 | BUG_ON(!bh); | |
ccd979bd | 657 | |
b4107950 TM |
658 | trace_ocfs2_journal_access( |
659 | (unsigned long long)ocfs2_metadata_cache_owner(ci), | |
660 | (unsigned long long)bh->b_blocknr, type, bh->b_size); | |
ccd979bd MF |
661 | |
662 | /* we can safely remove this assertion after testing. */ | |
663 | if (!buffer_uptodate(bh)) { | |
664 | mlog(ML_ERROR, "giving me a buffer that's not uptodate!\n"); | |
665 | mlog(ML_ERROR, "b_blocknr=%llu\n", | |
666 | (unsigned long long)bh->b_blocknr); | |
667 | BUG(); | |
668 | } | |
669 | ||
0cf2f763 | 670 | /* Set the current transaction information on the ci so |
ccd979bd | 671 | * that the locking code knows whether it can drop it's locks |
0cf2f763 | 672 | * on this ci or not. We're protected from the commit |
ccd979bd MF |
673 | * thread updating the current transaction id until |
674 | * ocfs2_commit_trans() because ocfs2_start_trans() took | |
675 | * j_trans_barrier for us. */ | |
0cf2f763 | 676 | ocfs2_set_ci_lock_trans(osb->journal, ci); |
ccd979bd | 677 | |
0cf2f763 | 678 | ocfs2_metadata_cache_io_lock(ci); |
ccd979bd MF |
679 | switch (type) { |
680 | case OCFS2_JOURNAL_ACCESS_CREATE: | |
681 | case OCFS2_JOURNAL_ACCESS_WRITE: | |
2b4e30fb | 682 | status = jbd2_journal_get_write_access(handle, bh); |
ccd979bd MF |
683 | break; |
684 | ||
685 | case OCFS2_JOURNAL_ACCESS_UNDO: | |
2b4e30fb | 686 | status = jbd2_journal_get_undo_access(handle, bh); |
ccd979bd MF |
687 | break; |
688 | ||
689 | default: | |
690 | status = -EINVAL; | |
af901ca1 | 691 | mlog(ML_ERROR, "Unknown access type!\n"); |
ccd979bd | 692 | } |
0cf2f763 | 693 | if (!status && ocfs2_meta_ecc(osb) && triggers) |
50655ae9 | 694 | jbd2_journal_set_triggers(bh, &triggers->ot_triggers); |
0cf2f763 | 695 | ocfs2_metadata_cache_io_unlock(ci); |
ccd979bd MF |
696 | |
697 | if (status < 0) | |
698 | mlog(ML_ERROR, "Error %d getting %d access to buffer!\n", | |
699 | status, type); | |
700 | ||
ccd979bd MF |
701 | return status; |
702 | } | |
703 | ||
0cf2f763 JB |
704 | int ocfs2_journal_access_di(handle_t *handle, struct ocfs2_caching_info *ci, |
705 | struct buffer_head *bh, int type) | |
50655ae9 | 706 | { |
0cf2f763 | 707 | return __ocfs2_journal_access(handle, ci, bh, &di_triggers, type); |
50655ae9 JB |
708 | } |
709 | ||
0cf2f763 | 710 | int ocfs2_journal_access_eb(handle_t *handle, struct ocfs2_caching_info *ci, |
50655ae9 JB |
711 | struct buffer_head *bh, int type) |
712 | { | |
0cf2f763 | 713 | return __ocfs2_journal_access(handle, ci, bh, &eb_triggers, type); |
50655ae9 JB |
714 | } |
715 | ||
93c97087 TM |
716 | int ocfs2_journal_access_rb(handle_t *handle, struct ocfs2_caching_info *ci, |
717 | struct buffer_head *bh, int type) | |
718 | { | |
719 | return __ocfs2_journal_access(handle, ci, bh, &rb_triggers, | |
720 | type); | |
721 | } | |
722 | ||
0cf2f763 | 723 | int ocfs2_journal_access_gd(handle_t *handle, struct ocfs2_caching_info *ci, |
50655ae9 JB |
724 | struct buffer_head *bh, int type) |
725 | { | |
0cf2f763 | 726 | return __ocfs2_journal_access(handle, ci, bh, &gd_triggers, type); |
50655ae9 JB |
727 | } |
728 | ||
0cf2f763 | 729 | int ocfs2_journal_access_db(handle_t *handle, struct ocfs2_caching_info *ci, |
50655ae9 JB |
730 | struct buffer_head *bh, int type) |
731 | { | |
0cf2f763 | 732 | return __ocfs2_journal_access(handle, ci, bh, &db_triggers, type); |
50655ae9 JB |
733 | } |
734 | ||
0cf2f763 | 735 | int ocfs2_journal_access_xb(handle_t *handle, struct ocfs2_caching_info *ci, |
50655ae9 JB |
736 | struct buffer_head *bh, int type) |
737 | { | |
0cf2f763 | 738 | return __ocfs2_journal_access(handle, ci, bh, &xb_triggers, type); |
50655ae9 JB |
739 | } |
740 | ||
0cf2f763 | 741 | int ocfs2_journal_access_dq(handle_t *handle, struct ocfs2_caching_info *ci, |
50655ae9 JB |
742 | struct buffer_head *bh, int type) |
743 | { | |
0cf2f763 | 744 | return __ocfs2_journal_access(handle, ci, bh, &dq_triggers, type); |
50655ae9 JB |
745 | } |
746 | ||
0cf2f763 | 747 | int ocfs2_journal_access_dr(handle_t *handle, struct ocfs2_caching_info *ci, |
9b7895ef MF |
748 | struct buffer_head *bh, int type) |
749 | { | |
0cf2f763 | 750 | return __ocfs2_journal_access(handle, ci, bh, &dr_triggers, type); |
9b7895ef MF |
751 | } |
752 | ||
0cf2f763 | 753 | int ocfs2_journal_access_dl(handle_t *handle, struct ocfs2_caching_info *ci, |
9b7895ef MF |
754 | struct buffer_head *bh, int type) |
755 | { | |
0cf2f763 | 756 | return __ocfs2_journal_access(handle, ci, bh, &dl_triggers, type); |
9b7895ef MF |
757 | } |
758 | ||
0cf2f763 | 759 | int ocfs2_journal_access(handle_t *handle, struct ocfs2_caching_info *ci, |
50655ae9 JB |
760 | struct buffer_head *bh, int type) |
761 | { | |
0cf2f763 | 762 | return __ocfs2_journal_access(handle, ci, bh, NULL, type); |
50655ae9 JB |
763 | } |
764 | ||
ec20cec7 | 765 | void ocfs2_journal_dirty(handle_t *handle, struct buffer_head *bh) |
ccd979bd MF |
766 | { |
767 | int status; | |
768 | ||
b4107950 | 769 | trace_ocfs2_journal_dirty((unsigned long long)bh->b_blocknr); |
ccd979bd | 770 | |
2b4e30fb | 771 | status = jbd2_journal_dirty_metadata(handle, bh); |
ec20cec7 | 772 | BUG_ON(status); |
ccd979bd MF |
773 | } |
774 | ||
2b4e30fb | 775 | #define OCFS2_DEFAULT_COMMIT_INTERVAL (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE) |
ccd979bd MF |
776 | |
777 | void ocfs2_set_journal_params(struct ocfs2_super *osb) | |
778 | { | |
779 | journal_t *journal = osb->journal->j_journal; | |
d147b3d6 MF |
780 | unsigned long commit_interval = OCFS2_DEFAULT_COMMIT_INTERVAL; |
781 | ||
782 | if (osb->osb_commit_interval) | |
783 | commit_interval = osb->osb_commit_interval; | |
ccd979bd | 784 | |
a931da6a | 785 | write_lock(&journal->j_state_lock); |
d147b3d6 | 786 | journal->j_commit_interval = commit_interval; |
ccd979bd | 787 | if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER) |
2b4e30fb | 788 | journal->j_flags |= JBD2_BARRIER; |
ccd979bd | 789 | else |
2b4e30fb | 790 | journal->j_flags &= ~JBD2_BARRIER; |
a931da6a | 791 | write_unlock(&journal->j_state_lock); |
ccd979bd MF |
792 | } |
793 | ||
794 | int ocfs2_journal_init(struct ocfs2_journal *journal, int *dirty) | |
795 | { | |
796 | int status = -1; | |
797 | struct inode *inode = NULL; /* the journal inode */ | |
798 | journal_t *j_journal = NULL; | |
799 | struct ocfs2_dinode *di = NULL; | |
800 | struct buffer_head *bh = NULL; | |
801 | struct ocfs2_super *osb; | |
e63aecb6 | 802 | int inode_lock = 0; |
ccd979bd | 803 | |
ccd979bd MF |
804 | BUG_ON(!journal); |
805 | ||
806 | osb = journal->j_osb; | |
807 | ||
808 | /* already have the inode for our journal */ | |
809 | inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, | |
810 | osb->slot_num); | |
811 | if (inode == NULL) { | |
812 | status = -EACCES; | |
813 | mlog_errno(status); | |
814 | goto done; | |
815 | } | |
816 | if (is_bad_inode(inode)) { | |
817 | mlog(ML_ERROR, "access error (bad inode)\n"); | |
818 | iput(inode); | |
819 | inode = NULL; | |
820 | status = -EACCES; | |
821 | goto done; | |
822 | } | |
823 | ||
824 | SET_INODE_JOURNAL(inode); | |
825 | OCFS2_I(inode)->ip_open_count++; | |
826 | ||
6eff5790 MF |
827 | /* Skip recovery waits here - journal inode metadata never |
828 | * changes in a live cluster so it can be considered an | |
829 | * exception to the rule. */ | |
e63aecb6 | 830 | status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY); |
ccd979bd MF |
831 | if (status < 0) { |
832 | if (status != -ERESTARTSYS) | |
833 | mlog(ML_ERROR, "Could not get lock on journal!\n"); | |
834 | goto done; | |
835 | } | |
836 | ||
e63aecb6 | 837 | inode_lock = 1; |
ccd979bd MF |
838 | di = (struct ocfs2_dinode *)bh->b_data; |
839 | ||
f17c20dd | 840 | if (i_size_read(inode) < OCFS2_MIN_JOURNAL_SIZE) { |
ccd979bd | 841 | mlog(ML_ERROR, "Journal file size (%lld) is too small!\n", |
f17c20dd | 842 | i_size_read(inode)); |
ccd979bd MF |
843 | status = -EINVAL; |
844 | goto done; | |
845 | } | |
846 | ||
f17c20dd | 847 | trace_ocfs2_journal_init(i_size_read(inode), |
b4107950 TM |
848 | (unsigned long long)inode->i_blocks, |
849 | OCFS2_I(inode)->ip_clusters); | |
ccd979bd MF |
850 | |
851 | /* call the kernels journal init function now */ | |
2b4e30fb | 852 | j_journal = jbd2_journal_init_inode(inode); |
ccd979bd MF |
853 | if (j_journal == NULL) { |
854 | mlog(ML_ERROR, "Linux journal layer error\n"); | |
855 | status = -EINVAL; | |
856 | goto done; | |
857 | } | |
858 | ||
b4107950 | 859 | trace_ocfs2_journal_init_maxlen(j_journal->j_maxlen); |
ccd979bd MF |
860 | |
861 | *dirty = (le32_to_cpu(di->id1.journal1.ij_flags) & | |
862 | OCFS2_JOURNAL_DIRTY_FL); | |
863 | ||
864 | journal->j_journal = j_journal; | |
865 | journal->j_inode = inode; | |
866 | journal->j_bh = bh; | |
867 | ||
868 | ocfs2_set_journal_params(osb); | |
869 | ||
870 | journal->j_state = OCFS2_JOURNAL_LOADED; | |
871 | ||
872 | status = 0; | |
873 | done: | |
874 | if (status < 0) { | |
e63aecb6 MF |
875 | if (inode_lock) |
876 | ocfs2_inode_unlock(inode, 1); | |
a81cb88b | 877 | brelse(bh); |
ccd979bd MF |
878 | if (inode) { |
879 | OCFS2_I(inode)->ip_open_count--; | |
880 | iput(inode); | |
881 | } | |
882 | } | |
883 | ||
ccd979bd MF |
884 | return status; |
885 | } | |
886 | ||
539d8264 SM |
887 | static void ocfs2_bump_recovery_generation(struct ocfs2_dinode *di) |
888 | { | |
889 | le32_add_cpu(&(di->id1.journal1.ij_recovery_generation), 1); | |
890 | } | |
891 | ||
892 | static u32 ocfs2_get_recovery_generation(struct ocfs2_dinode *di) | |
893 | { | |
894 | return le32_to_cpu(di->id1.journal1.ij_recovery_generation); | |
895 | } | |
896 | ||
ccd979bd | 897 | static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb, |
539d8264 | 898 | int dirty, int replayed) |
ccd979bd MF |
899 | { |
900 | int status; | |
901 | unsigned int flags; | |
902 | struct ocfs2_journal *journal = osb->journal; | |
903 | struct buffer_head *bh = journal->j_bh; | |
904 | struct ocfs2_dinode *fe; | |
905 | ||
ccd979bd | 906 | fe = (struct ocfs2_dinode *)bh->b_data; |
10995aa2 JB |
907 | |
908 | /* The journal bh on the osb always comes from ocfs2_journal_init() | |
909 | * and was validated there inside ocfs2_inode_lock_full(). It's a | |
910 | * code bug if we mess it up. */ | |
911 | BUG_ON(!OCFS2_IS_VALID_DINODE(fe)); | |
ccd979bd MF |
912 | |
913 | flags = le32_to_cpu(fe->id1.journal1.ij_flags); | |
914 | if (dirty) | |
915 | flags |= OCFS2_JOURNAL_DIRTY_FL; | |
916 | else | |
917 | flags &= ~OCFS2_JOURNAL_DIRTY_FL; | |
918 | fe->id1.journal1.ij_flags = cpu_to_le32(flags); | |
919 | ||
539d8264 SM |
920 | if (replayed) |
921 | ocfs2_bump_recovery_generation(fe); | |
922 | ||
13723d00 | 923 | ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check); |
8cb471e8 | 924 | status = ocfs2_write_block(osb, bh, INODE_CACHE(journal->j_inode)); |
ccd979bd MF |
925 | if (status < 0) |
926 | mlog_errno(status); | |
927 | ||
ccd979bd MF |
928 | return status; |
929 | } | |
930 | ||
931 | /* | |
932 | * If the journal has been kmalloc'd it needs to be freed after this | |
933 | * call. | |
934 | */ | |
935 | void ocfs2_journal_shutdown(struct ocfs2_super *osb) | |
936 | { | |
937 | struct ocfs2_journal *journal = NULL; | |
938 | int status = 0; | |
939 | struct inode *inode = NULL; | |
940 | int num_running_trans = 0; | |
941 | ||
ebdec83b | 942 | BUG_ON(!osb); |
ccd979bd MF |
943 | |
944 | journal = osb->journal; | |
945 | if (!journal) | |
946 | goto done; | |
947 | ||
948 | inode = journal->j_inode; | |
949 | ||
950 | if (journal->j_state != OCFS2_JOURNAL_LOADED) | |
951 | goto done; | |
952 | ||
2b4e30fb | 953 | /* need to inc inode use count - jbd2_journal_destroy will iput. */ |
ccd979bd MF |
954 | if (!igrab(inode)) |
955 | BUG(); | |
956 | ||
957 | num_running_trans = atomic_read(&(osb->journal->j_num_trans)); | |
b4107950 | 958 | trace_ocfs2_journal_shutdown(num_running_trans); |
ccd979bd MF |
959 | |
960 | /* Do a commit_cache here. It will flush our journal, *and* | |
961 | * release any locks that are still held. | |
962 | * set the SHUTDOWN flag and release the trans lock. | |
963 | * the commit thread will take the trans lock for us below. */ | |
964 | journal->j_state = OCFS2_JOURNAL_IN_SHUTDOWN; | |
965 | ||
966 | /* The OCFS2_JOURNAL_IN_SHUTDOWN will signal to commit_cache to not | |
967 | * drop the trans_lock (which we want to hold until we | |
968 | * completely destroy the journal. */ | |
969 | if (osb->commit_task) { | |
970 | /* Wait for the commit thread */ | |
b4107950 | 971 | trace_ocfs2_journal_shutdown_wait(osb->commit_task); |
ccd979bd MF |
972 | kthread_stop(osb->commit_task); |
973 | osb->commit_task = NULL; | |
974 | } | |
975 | ||
976 | BUG_ON(atomic_read(&(osb->journal->j_num_trans)) != 0); | |
977 | ||
c271c5c2 | 978 | if (ocfs2_mount_local(osb)) { |
2b4e30fb JB |
979 | jbd2_journal_lock_updates(journal->j_journal); |
980 | status = jbd2_journal_flush(journal->j_journal); | |
981 | jbd2_journal_unlock_updates(journal->j_journal); | |
c271c5c2 SM |
982 | if (status < 0) |
983 | mlog_errno(status); | |
984 | } | |
985 | ||
986 | if (status == 0) { | |
987 | /* | |
988 | * Do not toggle if flush was unsuccessful otherwise | |
989 | * will leave dirty metadata in a "clean" journal | |
990 | */ | |
539d8264 | 991 | status = ocfs2_journal_toggle_dirty(osb, 0, 0); |
c271c5c2 SM |
992 | if (status < 0) |
993 | mlog_errno(status); | |
994 | } | |
ccd979bd MF |
995 | |
996 | /* Shutdown the kernel journal system */ | |
2b4e30fb | 997 | jbd2_journal_destroy(journal->j_journal); |
ae0dff68 | 998 | journal->j_journal = NULL; |
ccd979bd MF |
999 | |
1000 | OCFS2_I(inode)->ip_open_count--; | |
1001 | ||
1002 | /* unlock our journal */ | |
e63aecb6 | 1003 | ocfs2_inode_unlock(inode, 1); |
ccd979bd MF |
1004 | |
1005 | brelse(journal->j_bh); | |
1006 | journal->j_bh = NULL; | |
1007 | ||
1008 | journal->j_state = OCFS2_JOURNAL_FREE; | |
1009 | ||
1010 | // up_write(&journal->j_trans_barrier); | |
1011 | done: | |
1012 | if (inode) | |
1013 | iput(inode); | |
ccd979bd MF |
1014 | } |
1015 | ||
1016 | static void ocfs2_clear_journal_error(struct super_block *sb, | |
1017 | journal_t *journal, | |
1018 | int slot) | |
1019 | { | |
1020 | int olderr; | |
1021 | ||
2b4e30fb | 1022 | olderr = jbd2_journal_errno(journal); |
ccd979bd MF |
1023 | if (olderr) { |
1024 | mlog(ML_ERROR, "File system error %d recorded in " | |
1025 | "journal %u.\n", olderr, slot); | |
1026 | mlog(ML_ERROR, "File system on device %s needs checking.\n", | |
1027 | sb->s_id); | |
1028 | ||
2b4e30fb JB |
1029 | jbd2_journal_ack_err(journal); |
1030 | jbd2_journal_clear_err(journal); | |
ccd979bd MF |
1031 | } |
1032 | } | |
1033 | ||
539d8264 | 1034 | int ocfs2_journal_load(struct ocfs2_journal *journal, int local, int replayed) |
ccd979bd MF |
1035 | { |
1036 | int status = 0; | |
1037 | struct ocfs2_super *osb; | |
1038 | ||
b1f3550f | 1039 | BUG_ON(!journal); |
ccd979bd MF |
1040 | |
1041 | osb = journal->j_osb; | |
1042 | ||
2b4e30fb | 1043 | status = jbd2_journal_load(journal->j_journal); |
ccd979bd MF |
1044 | if (status < 0) { |
1045 | mlog(ML_ERROR, "Failed to load journal!\n"); | |
1046 | goto done; | |
1047 | } | |
1048 | ||
1049 | ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num); | |
1050 | ||
539d8264 | 1051 | status = ocfs2_journal_toggle_dirty(osb, 1, replayed); |
ccd979bd MF |
1052 | if (status < 0) { |
1053 | mlog_errno(status); | |
1054 | goto done; | |
1055 | } | |
1056 | ||
1057 | /* Launch the commit thread */ | |
c271c5c2 SM |
1058 | if (!local) { |
1059 | osb->commit_task = kthread_run(ocfs2_commit_thread, osb, | |
1060 | "ocfs2cmt"); | |
1061 | if (IS_ERR(osb->commit_task)) { | |
1062 | status = PTR_ERR(osb->commit_task); | |
1063 | osb->commit_task = NULL; | |
1064 | mlog(ML_ERROR, "unable to launch ocfs2commit thread, " | |
1065 | "error=%d", status); | |
1066 | goto done; | |
1067 | } | |
1068 | } else | |
ccd979bd | 1069 | osb->commit_task = NULL; |
ccd979bd MF |
1070 | |
1071 | done: | |
ccd979bd MF |
1072 | return status; |
1073 | } | |
1074 | ||
1075 | ||
1076 | /* 'full' flag tells us whether we clear out all blocks or if we just | |
1077 | * mark the journal clean */ | |
1078 | int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full) | |
1079 | { | |
1080 | int status; | |
1081 | ||
ebdec83b | 1082 | BUG_ON(!journal); |
ccd979bd | 1083 | |
2b4e30fb | 1084 | status = jbd2_journal_wipe(journal->j_journal, full); |
ccd979bd MF |
1085 | if (status < 0) { |
1086 | mlog_errno(status); | |
1087 | goto bail; | |
1088 | } | |
1089 | ||
539d8264 | 1090 | status = ocfs2_journal_toggle_dirty(journal->j_osb, 0, 0); |
ccd979bd MF |
1091 | if (status < 0) |
1092 | mlog_errno(status); | |
1093 | ||
1094 | bail: | |
ccd979bd MF |
1095 | return status; |
1096 | } | |
1097 | ||
553abd04 JB |
1098 | static int ocfs2_recovery_completed(struct ocfs2_super *osb) |
1099 | { | |
1100 | int empty; | |
1101 | struct ocfs2_recovery_map *rm = osb->recovery_map; | |
1102 | ||
1103 | spin_lock(&osb->osb_lock); | |
1104 | empty = (rm->rm_used == 0); | |
1105 | spin_unlock(&osb->osb_lock); | |
1106 | ||
1107 | return empty; | |
1108 | } | |
1109 | ||
1110 | void ocfs2_wait_for_recovery(struct ocfs2_super *osb) | |
1111 | { | |
1112 | wait_event(osb->recovery_event, ocfs2_recovery_completed(osb)); | |
1113 | } | |
1114 | ||
ccd979bd MF |
1115 | /* |
1116 | * JBD Might read a cached version of another nodes journal file. We | |
1117 | * don't want this as this file changes often and we get no | |
1118 | * notification on those changes. The only way to be sure that we've | |
1119 | * got the most up to date version of those blocks then is to force | |
1120 | * read them off disk. Just searching through the buffer cache won't | |
1121 | * work as there may be pages backing this file which are still marked | |
1122 | * up to date. We know things can't change on this file underneath us | |
1123 | * as we have the lock by now :) | |
1124 | */ | |
1125 | static int ocfs2_force_read_journal(struct inode *inode) | |
1126 | { | |
1127 | int status = 0; | |
4f902c37 | 1128 | int i; |
8110b073 | 1129 | u64 v_blkno, p_blkno, p_blocks, num_blocks; |
4f902c37 | 1130 | #define CONCURRENT_JOURNAL_FILL 32ULL |
ccd979bd MF |
1131 | struct buffer_head *bhs[CONCURRENT_JOURNAL_FILL]; |
1132 | ||
ccd979bd MF |
1133 | memset(bhs, 0, sizeof(struct buffer_head *) * CONCURRENT_JOURNAL_FILL); |
1134 | ||
f17c20dd | 1135 | num_blocks = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode)); |
ccd979bd | 1136 | v_blkno = 0; |
8110b073 | 1137 | while (v_blkno < num_blocks) { |
ccd979bd | 1138 | status = ocfs2_extent_map_get_blocks(inode, v_blkno, |
49cb8d2d | 1139 | &p_blkno, &p_blocks, NULL); |
ccd979bd MF |
1140 | if (status < 0) { |
1141 | mlog_errno(status); | |
1142 | goto bail; | |
1143 | } | |
1144 | ||
1145 | if (p_blocks > CONCURRENT_JOURNAL_FILL) | |
1146 | p_blocks = CONCURRENT_JOURNAL_FILL; | |
1147 | ||
dd4a2c2b MF |
1148 | /* We are reading journal data which should not |
1149 | * be put in the uptodate cache */ | |
da1e9098 JB |
1150 | status = ocfs2_read_blocks_sync(OCFS2_SB(inode->i_sb), |
1151 | p_blkno, p_blocks, bhs); | |
ccd979bd MF |
1152 | if (status < 0) { |
1153 | mlog_errno(status); | |
1154 | goto bail; | |
1155 | } | |
1156 | ||
1157 | for(i = 0; i < p_blocks; i++) { | |
1158 | brelse(bhs[i]); | |
1159 | bhs[i] = NULL; | |
1160 | } | |
1161 | ||
1162 | v_blkno += p_blocks; | |
1163 | } | |
1164 | ||
1165 | bail: | |
1166 | for(i = 0; i < CONCURRENT_JOURNAL_FILL; i++) | |
a81cb88b | 1167 | brelse(bhs[i]); |
ccd979bd MF |
1168 | return status; |
1169 | } | |
1170 | ||
1171 | struct ocfs2_la_recovery_item { | |
1172 | struct list_head lri_list; | |
1173 | int lri_slot; | |
1174 | struct ocfs2_dinode *lri_la_dinode; | |
1175 | struct ocfs2_dinode *lri_tl_dinode; | |
2205363d | 1176 | struct ocfs2_quota_recovery *lri_qrec; |
ccd979bd MF |
1177 | }; |
1178 | ||
1179 | /* Does the second half of the recovery process. By this point, the | |
1180 | * node is marked clean and can actually be considered recovered, | |
1181 | * hence it's no longer in the recovery map, but there's still some | |
1182 | * cleanup we can do which shouldn't happen within the recovery thread | |
1183 | * as locking in that context becomes very difficult if we are to take | |
1184 | * recovering nodes into account. | |
1185 | * | |
1186 | * NOTE: This function can and will sleep on recovery of other nodes | |
1187 | * during cluster locking, just like any other ocfs2 process. | |
1188 | */ | |
c4028958 | 1189 | void ocfs2_complete_recovery(struct work_struct *work) |
ccd979bd | 1190 | { |
b4107950 | 1191 | int ret = 0; |
c4028958 DH |
1192 | struct ocfs2_journal *journal = |
1193 | container_of(work, struct ocfs2_journal, j_recovery_work); | |
1194 | struct ocfs2_super *osb = journal->j_osb; | |
ccd979bd | 1195 | struct ocfs2_dinode *la_dinode, *tl_dinode; |
800deef3 | 1196 | struct ocfs2_la_recovery_item *item, *n; |
2205363d | 1197 | struct ocfs2_quota_recovery *qrec; |
ccd979bd MF |
1198 | LIST_HEAD(tmp_la_list); |
1199 | ||
b4107950 TM |
1200 | trace_ocfs2_complete_recovery( |
1201 | (unsigned long long)OCFS2_I(journal->j_inode)->ip_blkno); | |
ccd979bd MF |
1202 | |
1203 | spin_lock(&journal->j_lock); | |
1204 | list_splice_init(&journal->j_la_cleanups, &tmp_la_list); | |
1205 | spin_unlock(&journal->j_lock); | |
1206 | ||
800deef3 | 1207 | list_for_each_entry_safe(item, n, &tmp_la_list, lri_list) { |
ccd979bd MF |
1208 | list_del_init(&item->lri_list); |
1209 | ||
19ece546 JK |
1210 | ocfs2_wait_on_quotas(osb); |
1211 | ||
ccd979bd | 1212 | la_dinode = item->lri_la_dinode; |
b4107950 TM |
1213 | tl_dinode = item->lri_tl_dinode; |
1214 | qrec = item->lri_qrec; | |
ccd979bd | 1215 | |
b4107950 TM |
1216 | trace_ocfs2_complete_recovery_slot(item->lri_slot, |
1217 | la_dinode ? le64_to_cpu(la_dinode->i_blkno) : 0, | |
1218 | tl_dinode ? le64_to_cpu(tl_dinode->i_blkno) : 0, | |
1219 | qrec); | |
1220 | ||
1221 | if (la_dinode) { | |
ccd979bd MF |
1222 | ret = ocfs2_complete_local_alloc_recovery(osb, |
1223 | la_dinode); | |
1224 | if (ret < 0) | |
1225 | mlog_errno(ret); | |
1226 | ||
1227 | kfree(la_dinode); | |
1228 | } | |
1229 | ||
ccd979bd | 1230 | if (tl_dinode) { |
ccd979bd MF |
1231 | ret = ocfs2_complete_truncate_log_recovery(osb, |
1232 | tl_dinode); | |
1233 | if (ret < 0) | |
1234 | mlog_errno(ret); | |
1235 | ||
1236 | kfree(tl_dinode); | |
1237 | } | |
1238 | ||
1239 | ret = ocfs2_recover_orphans(osb, item->lri_slot); | |
1240 | if (ret < 0) | |
1241 | mlog_errno(ret); | |
1242 | ||
2205363d | 1243 | if (qrec) { |
2205363d JK |
1244 | ret = ocfs2_finish_quota_recovery(osb, qrec, |
1245 | item->lri_slot); | |
1246 | if (ret < 0) | |
1247 | mlog_errno(ret); | |
1248 | /* Recovery info is already freed now */ | |
1249 | } | |
1250 | ||
ccd979bd MF |
1251 | kfree(item); |
1252 | } | |
1253 | ||
b4107950 | 1254 | trace_ocfs2_complete_recovery_end(ret); |
ccd979bd MF |
1255 | } |
1256 | ||
1257 | /* NOTE: This function always eats your references to la_dinode and | |
1258 | * tl_dinode, either manually on error, or by passing them to | |
1259 | * ocfs2_complete_recovery */ | |
1260 | static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal, | |
1261 | int slot_num, | |
1262 | struct ocfs2_dinode *la_dinode, | |
2205363d JK |
1263 | struct ocfs2_dinode *tl_dinode, |
1264 | struct ocfs2_quota_recovery *qrec) | |
ccd979bd MF |
1265 | { |
1266 | struct ocfs2_la_recovery_item *item; | |
1267 | ||
afae00ab | 1268 | item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS); |
ccd979bd MF |
1269 | if (!item) { |
1270 | /* Though we wish to avoid it, we are in fact safe in | |
1271 | * skipping local alloc cleanup as fsck.ocfs2 is more | |
1272 | * than capable of reclaiming unused space. */ | |
d787ab09 TG |
1273 | kfree(la_dinode); |
1274 | kfree(tl_dinode); | |
ccd979bd | 1275 | |
2205363d JK |
1276 | if (qrec) |
1277 | ocfs2_free_quota_recovery(qrec); | |
1278 | ||
ccd979bd MF |
1279 | mlog_errno(-ENOMEM); |
1280 | return; | |
1281 | } | |
1282 | ||
1283 | INIT_LIST_HEAD(&item->lri_list); | |
1284 | item->lri_la_dinode = la_dinode; | |
1285 | item->lri_slot = slot_num; | |
1286 | item->lri_tl_dinode = tl_dinode; | |
2205363d | 1287 | item->lri_qrec = qrec; |
ccd979bd MF |
1288 | |
1289 | spin_lock(&journal->j_lock); | |
1290 | list_add_tail(&item->lri_list, &journal->j_la_cleanups); | |
1291 | queue_work(ocfs2_wq, &journal->j_recovery_work); | |
1292 | spin_unlock(&journal->j_lock); | |
1293 | } | |
1294 | ||
1295 | /* Called by the mount code to queue recovery the last part of | |
9140db04 | 1296 | * recovery for it's own and offline slot(s). */ |
ccd979bd MF |
1297 | void ocfs2_complete_mount_recovery(struct ocfs2_super *osb) |
1298 | { | |
1299 | struct ocfs2_journal *journal = osb->journal; | |
1300 | ||
10b3dd76 SM |
1301 | if (ocfs2_is_hard_readonly(osb)) |
1302 | return; | |
1303 | ||
9140db04 SE |
1304 | /* No need to queue up our truncate_log as regular cleanup will catch |
1305 | * that */ | |
1306 | ocfs2_queue_recovery_completion(journal, osb->slot_num, | |
1307 | osb->local_alloc_copy, NULL, NULL); | |
1308 | ocfs2_schedule_truncate_log_flush(osb, 0); | |
ccd979bd | 1309 | |
9140db04 SE |
1310 | osb->local_alloc_copy = NULL; |
1311 | osb->dirty = 0; | |
1312 | ||
1313 | /* queue to recover orphan slots for all offline slots */ | |
1314 | ocfs2_replay_map_set_state(osb, REPLAY_NEEDED); | |
1315 | ocfs2_queue_replay_slots(osb); | |
1316 | ocfs2_free_replay_slots(osb); | |
ccd979bd MF |
1317 | } |
1318 | ||
2205363d JK |
1319 | void ocfs2_complete_quota_recovery(struct ocfs2_super *osb) |
1320 | { | |
1321 | if (osb->quota_rec) { | |
1322 | ocfs2_queue_recovery_completion(osb->journal, | |
1323 | osb->slot_num, | |
1324 | NULL, | |
1325 | NULL, | |
1326 | osb->quota_rec); | |
1327 | osb->quota_rec = NULL; | |
1328 | } | |
1329 | } | |
1330 | ||
ccd979bd MF |
1331 | static int __ocfs2_recovery_thread(void *arg) |
1332 | { | |
2205363d | 1333 | int status, node_num, slot_num; |
ccd979bd | 1334 | struct ocfs2_super *osb = arg; |
553abd04 | 1335 | struct ocfs2_recovery_map *rm = osb->recovery_map; |
2205363d JK |
1336 | int *rm_quota = NULL; |
1337 | int rm_quota_used = 0, i; | |
1338 | struct ocfs2_quota_recovery *qrec; | |
ccd979bd | 1339 | |
ccd979bd MF |
1340 | status = ocfs2_wait_on_mount(osb); |
1341 | if (status < 0) { | |
1342 | goto bail; | |
1343 | } | |
1344 | ||
2205363d JK |
1345 | rm_quota = kzalloc(osb->max_slots * sizeof(int), GFP_NOFS); |
1346 | if (!rm_quota) { | |
1347 | status = -ENOMEM; | |
1348 | goto bail; | |
1349 | } | |
ccd979bd MF |
1350 | restart: |
1351 | status = ocfs2_super_lock(osb, 1); | |
1352 | if (status < 0) { | |
1353 | mlog_errno(status); | |
1354 | goto bail; | |
1355 | } | |
1356 | ||
9140db04 SE |
1357 | status = ocfs2_compute_replay_slots(osb); |
1358 | if (status < 0) | |
1359 | mlog_errno(status); | |
1360 | ||
1361 | /* queue recovery for our own slot */ | |
1362 | ocfs2_queue_recovery_completion(osb->journal, osb->slot_num, NULL, | |
1363 | NULL, NULL); | |
1364 | ||
553abd04 JB |
1365 | spin_lock(&osb->osb_lock); |
1366 | while (rm->rm_used) { | |
1367 | /* It's always safe to remove entry zero, as we won't | |
1368 | * clear it until ocfs2_recover_node() has succeeded. */ | |
1369 | node_num = rm->rm_entries[0]; | |
1370 | spin_unlock(&osb->osb_lock); | |
2205363d | 1371 | slot_num = ocfs2_node_num_to_slot(osb, node_num); |
b4107950 | 1372 | trace_ocfs2_recovery_thread_node(node_num, slot_num); |
2205363d JK |
1373 | if (slot_num == -ENOENT) { |
1374 | status = 0; | |
2205363d JK |
1375 | goto skip_recovery; |
1376 | } | |
2205363d JK |
1377 | |
1378 | /* It is a bit subtle with quota recovery. We cannot do it | |
1379 | * immediately because we have to obtain cluster locks from | |
1380 | * quota files and we also don't want to just skip it because | |
1381 | * then quota usage would be out of sync until some node takes | |
1382 | * the slot. So we remember which nodes need quota recovery | |
1383 | * and when everything else is done, we recover quotas. */ | |
1384 | for (i = 0; i < rm_quota_used && rm_quota[i] != slot_num; i++); | |
1385 | if (i == rm_quota_used) | |
1386 | rm_quota[rm_quota_used++] = slot_num; | |
1387 | ||
1388 | status = ocfs2_recover_node(osb, node_num, slot_num); | |
1389 | skip_recovery: | |
553abd04 JB |
1390 | if (!status) { |
1391 | ocfs2_recovery_map_clear(osb, node_num); | |
1392 | } else { | |
ccd979bd MF |
1393 | mlog(ML_ERROR, |
1394 | "Error %d recovering node %d on device (%u,%u)!\n", | |
1395 | status, node_num, | |
1396 | MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev)); | |
1397 | mlog(ML_ERROR, "Volume requires unmount.\n"); | |
ccd979bd MF |
1398 | } |
1399 | ||
553abd04 | 1400 | spin_lock(&osb->osb_lock); |
ccd979bd | 1401 | } |
553abd04 | 1402 | spin_unlock(&osb->osb_lock); |
b4107950 | 1403 | trace_ocfs2_recovery_thread_end(status); |
553abd04 | 1404 | |
539d8264 SM |
1405 | /* Refresh all journal recovery generations from disk */ |
1406 | status = ocfs2_check_journals_nolocks(osb); | |
1407 | status = (status == -EROFS) ? 0 : status; | |
1408 | if (status < 0) | |
1409 | mlog_errno(status); | |
1410 | ||
2205363d | 1411 | /* Now it is right time to recover quotas... We have to do this under |
25985edc | 1412 | * superblock lock so that no one can start using the slot (and crash) |
2205363d JK |
1413 | * before we recover it */ |
1414 | for (i = 0; i < rm_quota_used; i++) { | |
1415 | qrec = ocfs2_begin_quota_recovery(osb, rm_quota[i]); | |
1416 | if (IS_ERR(qrec)) { | |
1417 | status = PTR_ERR(qrec); | |
1418 | mlog_errno(status); | |
1419 | continue; | |
1420 | } | |
1421 | ocfs2_queue_recovery_completion(osb->journal, rm_quota[i], | |
1422 | NULL, NULL, qrec); | |
1423 | } | |
1424 | ||
ccd979bd MF |
1425 | ocfs2_super_unlock(osb, 1); |
1426 | ||
9140db04 SE |
1427 | /* queue recovery for offline slots */ |
1428 | ocfs2_queue_replay_slots(osb); | |
ccd979bd MF |
1429 | |
1430 | bail: | |
c74ec2f7 | 1431 | mutex_lock(&osb->recovery_lock); |
553abd04 | 1432 | if (!status && !ocfs2_recovery_completed(osb)) { |
c74ec2f7 | 1433 | mutex_unlock(&osb->recovery_lock); |
ccd979bd MF |
1434 | goto restart; |
1435 | } | |
1436 | ||
9140db04 | 1437 | ocfs2_free_replay_slots(osb); |
ccd979bd MF |
1438 | osb->recovery_thread_task = NULL; |
1439 | mb(); /* sync with ocfs2_recovery_thread_running */ | |
1440 | wake_up(&osb->recovery_event); | |
1441 | ||
c74ec2f7 | 1442 | mutex_unlock(&osb->recovery_lock); |
ccd979bd | 1443 | |
d787ab09 | 1444 | kfree(rm_quota); |
2205363d | 1445 | |
ccd979bd MF |
1446 | /* no one is callint kthread_stop() for us so the kthread() api |
1447 | * requires that we call do_exit(). And it isn't exported, but | |
1448 | * complete_and_exit() seems to be a minimal wrapper around it. */ | |
1449 | complete_and_exit(NULL, status); | |
1450 | return status; | |
1451 | } | |
1452 | ||
1453 | void ocfs2_recovery_thread(struct ocfs2_super *osb, int node_num) | |
1454 | { | |
c74ec2f7 | 1455 | mutex_lock(&osb->recovery_lock); |
ccd979bd | 1456 | |
b4107950 TM |
1457 | trace_ocfs2_recovery_thread(node_num, osb->node_num, |
1458 | osb->disable_recovery, osb->recovery_thread_task, | |
1459 | osb->disable_recovery ? | |
1460 | -1 : ocfs2_recovery_map_set(osb, node_num)); | |
ccd979bd | 1461 | |
b4107950 TM |
1462 | if (osb->disable_recovery) |
1463 | goto out; | |
ccd979bd MF |
1464 | |
1465 | if (osb->recovery_thread_task) | |
1466 | goto out; | |
1467 | ||
1468 | osb->recovery_thread_task = kthread_run(__ocfs2_recovery_thread, osb, | |
78427043 | 1469 | "ocfs2rec"); |
ccd979bd MF |
1470 | if (IS_ERR(osb->recovery_thread_task)) { |
1471 | mlog_errno((int)PTR_ERR(osb->recovery_thread_task)); | |
1472 | osb->recovery_thread_task = NULL; | |
1473 | } | |
1474 | ||
1475 | out: | |
c74ec2f7 | 1476 | mutex_unlock(&osb->recovery_lock); |
ccd979bd | 1477 | wake_up(&osb->recovery_event); |
ccd979bd MF |
1478 | } |
1479 | ||
539d8264 SM |
1480 | static int ocfs2_read_journal_inode(struct ocfs2_super *osb, |
1481 | int slot_num, | |
1482 | struct buffer_head **bh, | |
1483 | struct inode **ret_inode) | |
1484 | { | |
1485 | int status = -EACCES; | |
1486 | struct inode *inode = NULL; | |
1487 | ||
1488 | BUG_ON(slot_num >= osb->max_slots); | |
1489 | ||
1490 | inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, | |
1491 | slot_num); | |
1492 | if (!inode || is_bad_inode(inode)) { | |
1493 | mlog_errno(status); | |
1494 | goto bail; | |
1495 | } | |
1496 | SET_INODE_JOURNAL(inode); | |
1497 | ||
b657c95c | 1498 | status = ocfs2_read_inode_block_full(inode, bh, OCFS2_BH_IGNORE_CACHE); |
539d8264 SM |
1499 | if (status < 0) { |
1500 | mlog_errno(status); | |
1501 | goto bail; | |
1502 | } | |
1503 | ||
1504 | status = 0; | |
1505 | ||
1506 | bail: | |
1507 | if (inode) { | |
1508 | if (status || !ret_inode) | |
1509 | iput(inode); | |
1510 | else | |
1511 | *ret_inode = inode; | |
1512 | } | |
1513 | return status; | |
1514 | } | |
1515 | ||
ccd979bd MF |
1516 | /* Does the actual journal replay and marks the journal inode as |
1517 | * clean. Will only replay if the journal inode is marked dirty. */ | |
1518 | static int ocfs2_replay_journal(struct ocfs2_super *osb, | |
1519 | int node_num, | |
1520 | int slot_num) | |
1521 | { | |
1522 | int status; | |
1523 | int got_lock = 0; | |
1524 | unsigned int flags; | |
1525 | struct inode *inode = NULL; | |
1526 | struct ocfs2_dinode *fe; | |
1527 | journal_t *journal = NULL; | |
1528 | struct buffer_head *bh = NULL; | |
539d8264 | 1529 | u32 slot_reco_gen; |
ccd979bd | 1530 | |
539d8264 SM |
1531 | status = ocfs2_read_journal_inode(osb, slot_num, &bh, &inode); |
1532 | if (status) { | |
ccd979bd MF |
1533 | mlog_errno(status); |
1534 | goto done; | |
1535 | } | |
539d8264 SM |
1536 | |
1537 | fe = (struct ocfs2_dinode *)bh->b_data; | |
1538 | slot_reco_gen = ocfs2_get_recovery_generation(fe); | |
1539 | brelse(bh); | |
1540 | bh = NULL; | |
1541 | ||
1542 | /* | |
1543 | * As the fs recovery is asynchronous, there is a small chance that | |
1544 | * another node mounted (and recovered) the slot before the recovery | |
1545 | * thread could get the lock. To handle that, we dirty read the journal | |
1546 | * inode for that slot to get the recovery generation. If it is | |
1547 | * different than what we expected, the slot has been recovered. | |
1548 | * If not, it needs recovery. | |
1549 | */ | |
1550 | if (osb->slot_recovery_generations[slot_num] != slot_reco_gen) { | |
b4107950 | 1551 | trace_ocfs2_replay_journal_recovered(slot_num, |
539d8264 SM |
1552 | osb->slot_recovery_generations[slot_num], slot_reco_gen); |
1553 | osb->slot_recovery_generations[slot_num] = slot_reco_gen; | |
1554 | status = -EBUSY; | |
ccd979bd MF |
1555 | goto done; |
1556 | } | |
539d8264 SM |
1557 | |
1558 | /* Continue with recovery as the journal has not yet been recovered */ | |
ccd979bd | 1559 | |
e63aecb6 | 1560 | status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY); |
ccd979bd | 1561 | if (status < 0) { |
b4107950 | 1562 | trace_ocfs2_replay_journal_lock_err(status); |
ccd979bd MF |
1563 | if (status != -ERESTARTSYS) |
1564 | mlog(ML_ERROR, "Could not lock journal!\n"); | |
1565 | goto done; | |
1566 | } | |
1567 | got_lock = 1; | |
1568 | ||
1569 | fe = (struct ocfs2_dinode *) bh->b_data; | |
1570 | ||
1571 | flags = le32_to_cpu(fe->id1.journal1.ij_flags); | |
539d8264 | 1572 | slot_reco_gen = ocfs2_get_recovery_generation(fe); |
ccd979bd MF |
1573 | |
1574 | if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) { | |
b4107950 | 1575 | trace_ocfs2_replay_journal_skip(node_num); |
539d8264 SM |
1576 | /* Refresh recovery generation for the slot */ |
1577 | osb->slot_recovery_generations[slot_num] = slot_reco_gen; | |
ccd979bd MF |
1578 | goto done; |
1579 | } | |
1580 | ||
9140db04 SE |
1581 | /* we need to run complete recovery for offline orphan slots */ |
1582 | ocfs2_replay_map_set_state(osb, REPLAY_NEEDED); | |
1583 | ||
619c200d SM |
1584 | printk(KERN_NOTICE "ocfs2: Begin replay journal (node %d, slot %d) on "\ |
1585 | "device (%u,%u)\n", node_num, slot_num, MAJOR(osb->sb->s_dev), | |
1586 | MINOR(osb->sb->s_dev)); | |
ccd979bd MF |
1587 | |
1588 | OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters); | |
1589 | ||
1590 | status = ocfs2_force_read_journal(inode); | |
1591 | if (status < 0) { | |
1592 | mlog_errno(status); | |
1593 | goto done; | |
1594 | } | |
1595 | ||
2b4e30fb | 1596 | journal = jbd2_journal_init_inode(inode); |
ccd979bd MF |
1597 | if (journal == NULL) { |
1598 | mlog(ML_ERROR, "Linux journal layer error\n"); | |
1599 | status = -EIO; | |
1600 | goto done; | |
1601 | } | |
1602 | ||
2b4e30fb | 1603 | status = jbd2_journal_load(journal); |
ccd979bd MF |
1604 | if (status < 0) { |
1605 | mlog_errno(status); | |
1606 | if (!igrab(inode)) | |
1607 | BUG(); | |
2b4e30fb | 1608 | jbd2_journal_destroy(journal); |
ccd979bd MF |
1609 | goto done; |
1610 | } | |
1611 | ||
1612 | ocfs2_clear_journal_error(osb->sb, journal, slot_num); | |
1613 | ||
1614 | /* wipe the journal */ | |
2b4e30fb JB |
1615 | jbd2_journal_lock_updates(journal); |
1616 | status = jbd2_journal_flush(journal); | |
1617 | jbd2_journal_unlock_updates(journal); | |
ccd979bd MF |
1618 | if (status < 0) |
1619 | mlog_errno(status); | |
1620 | ||
1621 | /* This will mark the node clean */ | |
1622 | flags = le32_to_cpu(fe->id1.journal1.ij_flags); | |
1623 | flags &= ~OCFS2_JOURNAL_DIRTY_FL; | |
1624 | fe->id1.journal1.ij_flags = cpu_to_le32(flags); | |
1625 | ||
539d8264 SM |
1626 | /* Increment recovery generation to indicate successful recovery */ |
1627 | ocfs2_bump_recovery_generation(fe); | |
1628 | osb->slot_recovery_generations[slot_num] = | |
1629 | ocfs2_get_recovery_generation(fe); | |
1630 | ||
13723d00 | 1631 | ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check); |
8cb471e8 | 1632 | status = ocfs2_write_block(osb, bh, INODE_CACHE(inode)); |
ccd979bd MF |
1633 | if (status < 0) |
1634 | mlog_errno(status); | |
1635 | ||
1636 | if (!igrab(inode)) | |
1637 | BUG(); | |
1638 | ||
2b4e30fb | 1639 | jbd2_journal_destroy(journal); |
ccd979bd | 1640 | |
619c200d SM |
1641 | printk(KERN_NOTICE "ocfs2: End replay journal (node %d, slot %d) on "\ |
1642 | "device (%u,%u)\n", node_num, slot_num, MAJOR(osb->sb->s_dev), | |
1643 | MINOR(osb->sb->s_dev)); | |
ccd979bd MF |
1644 | done: |
1645 | /* drop the lock on this nodes journal */ | |
1646 | if (got_lock) | |
e63aecb6 | 1647 | ocfs2_inode_unlock(inode, 1); |
ccd979bd MF |
1648 | |
1649 | if (inode) | |
1650 | iput(inode); | |
1651 | ||
a81cb88b | 1652 | brelse(bh); |
ccd979bd | 1653 | |
ccd979bd MF |
1654 | return status; |
1655 | } | |
1656 | ||
1657 | /* | |
1658 | * Do the most important parts of node recovery: | |
1659 | * - Replay it's journal | |
1660 | * - Stamp a clean local allocator file | |
1661 | * - Stamp a clean truncate log | |
1662 | * - Mark the node clean | |
1663 | * | |
1664 | * If this function completes without error, a node in OCFS2 can be | |
1665 | * said to have been safely recovered. As a result, failure during the | |
1666 | * second part of a nodes recovery process (local alloc recovery) is | |
1667 | * far less concerning. | |
1668 | */ | |
1669 | static int ocfs2_recover_node(struct ocfs2_super *osb, | |
2205363d | 1670 | int node_num, int slot_num) |
ccd979bd MF |
1671 | { |
1672 | int status = 0; | |
ccd979bd MF |
1673 | struct ocfs2_dinode *la_copy = NULL; |
1674 | struct ocfs2_dinode *tl_copy = NULL; | |
1675 | ||
b4107950 | 1676 | trace_ocfs2_recover_node(node_num, slot_num, osb->node_num); |
ccd979bd MF |
1677 | |
1678 | /* Should not ever be called to recover ourselves -- in that | |
1679 | * case we should've called ocfs2_journal_load instead. */ | |
ebdec83b | 1680 | BUG_ON(osb->node_num == node_num); |
ccd979bd | 1681 | |
ccd979bd MF |
1682 | status = ocfs2_replay_journal(osb, node_num, slot_num); |
1683 | if (status < 0) { | |
539d8264 | 1684 | if (status == -EBUSY) { |
b4107950 | 1685 | trace_ocfs2_recover_node_skip(slot_num, node_num); |
539d8264 SM |
1686 | status = 0; |
1687 | goto done; | |
1688 | } | |
ccd979bd MF |
1689 | mlog_errno(status); |
1690 | goto done; | |
1691 | } | |
1692 | ||
1693 | /* Stamp a clean local alloc file AFTER recovering the journal... */ | |
1694 | status = ocfs2_begin_local_alloc_recovery(osb, slot_num, &la_copy); | |
1695 | if (status < 0) { | |
1696 | mlog_errno(status); | |
1697 | goto done; | |
1698 | } | |
1699 | ||
1700 | /* An error from begin_truncate_log_recovery is not | |
1701 | * serious enough to warrant halting the rest of | |
1702 | * recovery. */ | |
1703 | status = ocfs2_begin_truncate_log_recovery(osb, slot_num, &tl_copy); | |
1704 | if (status < 0) | |
1705 | mlog_errno(status); | |
1706 | ||
1707 | /* Likewise, this would be a strange but ultimately not so | |
1708 | * harmful place to get an error... */ | |
8e8a4603 | 1709 | status = ocfs2_clear_slot(osb, slot_num); |
ccd979bd MF |
1710 | if (status < 0) |
1711 | mlog_errno(status); | |
1712 | ||
1713 | /* This will kfree the memory pointed to by la_copy and tl_copy */ | |
1714 | ocfs2_queue_recovery_completion(osb->journal, slot_num, la_copy, | |
2205363d | 1715 | tl_copy, NULL); |
ccd979bd MF |
1716 | |
1717 | status = 0; | |
1718 | done: | |
1719 | ||
ccd979bd MF |
1720 | return status; |
1721 | } | |
1722 | ||
1723 | /* Test node liveness by trylocking his journal. If we get the lock, | |
1724 | * we drop it here. Return 0 if we got the lock, -EAGAIN if node is | |
1725 | * still alive (we couldn't get the lock) and < 0 on error. */ | |
1726 | static int ocfs2_trylock_journal(struct ocfs2_super *osb, | |
1727 | int slot_num) | |
1728 | { | |
1729 | int status, flags; | |
1730 | struct inode *inode = NULL; | |
1731 | ||
1732 | inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, | |
1733 | slot_num); | |
1734 | if (inode == NULL) { | |
1735 | mlog(ML_ERROR, "access error\n"); | |
1736 | status = -EACCES; | |
1737 | goto bail; | |
1738 | } | |
1739 | if (is_bad_inode(inode)) { | |
1740 | mlog(ML_ERROR, "access error (bad inode)\n"); | |
1741 | iput(inode); | |
1742 | inode = NULL; | |
1743 | status = -EACCES; | |
1744 | goto bail; | |
1745 | } | |
1746 | SET_INODE_JOURNAL(inode); | |
1747 | ||
1748 | flags = OCFS2_META_LOCK_RECOVERY | OCFS2_META_LOCK_NOQUEUE; | |
e63aecb6 | 1749 | status = ocfs2_inode_lock_full(inode, NULL, 1, flags); |
ccd979bd MF |
1750 | if (status < 0) { |
1751 | if (status != -EAGAIN) | |
1752 | mlog_errno(status); | |
1753 | goto bail; | |
1754 | } | |
1755 | ||
e63aecb6 | 1756 | ocfs2_inode_unlock(inode, 1); |
ccd979bd MF |
1757 | bail: |
1758 | if (inode) | |
1759 | iput(inode); | |
1760 | ||
1761 | return status; | |
1762 | } | |
1763 | ||
1764 | /* Call this underneath ocfs2_super_lock. It also assumes that the | |
1765 | * slot info struct has been updated from disk. */ | |
1766 | int ocfs2_mark_dead_nodes(struct ocfs2_super *osb) | |
1767 | { | |
d85b20e4 JB |
1768 | unsigned int node_num; |
1769 | int status, i; | |
a1af7d15 | 1770 | u32 gen; |
539d8264 SM |
1771 | struct buffer_head *bh = NULL; |
1772 | struct ocfs2_dinode *di; | |
ccd979bd MF |
1773 | |
1774 | /* This is called with the super block cluster lock, so we | |
1775 | * know that the slot map can't change underneath us. */ | |
1776 | ||
d85b20e4 | 1777 | for (i = 0; i < osb->max_slots; i++) { |
539d8264 SM |
1778 | /* Read journal inode to get the recovery generation */ |
1779 | status = ocfs2_read_journal_inode(osb, i, &bh, NULL); | |
1780 | if (status) { | |
1781 | mlog_errno(status); | |
1782 | goto bail; | |
1783 | } | |
1784 | di = (struct ocfs2_dinode *)bh->b_data; | |
a1af7d15 | 1785 | gen = ocfs2_get_recovery_generation(di); |
539d8264 SM |
1786 | brelse(bh); |
1787 | bh = NULL; | |
1788 | ||
a1af7d15 MF |
1789 | spin_lock(&osb->osb_lock); |
1790 | osb->slot_recovery_generations[i] = gen; | |
1791 | ||
b4107950 TM |
1792 | trace_ocfs2_mark_dead_nodes(i, |
1793 | osb->slot_recovery_generations[i]); | |
539d8264 | 1794 | |
a1af7d15 MF |
1795 | if (i == osb->slot_num) { |
1796 | spin_unlock(&osb->osb_lock); | |
ccd979bd | 1797 | continue; |
a1af7d15 | 1798 | } |
d85b20e4 JB |
1799 | |
1800 | status = ocfs2_slot_to_node_num_locked(osb, i, &node_num); | |
a1af7d15 MF |
1801 | if (status == -ENOENT) { |
1802 | spin_unlock(&osb->osb_lock); | |
ccd979bd | 1803 | continue; |
a1af7d15 | 1804 | } |
ccd979bd | 1805 | |
a1af7d15 MF |
1806 | if (__ocfs2_recovery_map_test(osb, node_num)) { |
1807 | spin_unlock(&osb->osb_lock); | |
ccd979bd | 1808 | continue; |
a1af7d15 | 1809 | } |
d85b20e4 | 1810 | spin_unlock(&osb->osb_lock); |
ccd979bd MF |
1811 | |
1812 | /* Ok, we have a slot occupied by another node which | |
1813 | * is not in the recovery map. We trylock his journal | |
1814 | * file here to test if he's alive. */ | |
1815 | status = ocfs2_trylock_journal(osb, i); | |
1816 | if (!status) { | |
1817 | /* Since we're called from mount, we know that | |
1818 | * the recovery thread can't race us on | |
1819 | * setting / checking the recovery bits. */ | |
1820 | ocfs2_recovery_thread(osb, node_num); | |
1821 | } else if ((status < 0) && (status != -EAGAIN)) { | |
1822 | mlog_errno(status); | |
1823 | goto bail; | |
1824 | } | |
ccd979bd | 1825 | } |
ccd979bd MF |
1826 | |
1827 | status = 0; | |
1828 | bail: | |
ccd979bd MF |
1829 | return status; |
1830 | } | |
1831 | ||
83273932 SE |
1832 | /* |
1833 | * Scan timer should get fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT. Add some | |
1834 | * randomness to the timeout to minimize multple nodes firing the timer at the | |
1835 | * same time. | |
1836 | */ | |
1837 | static inline unsigned long ocfs2_orphan_scan_timeout(void) | |
1838 | { | |
1839 | unsigned long time; | |
1840 | ||
1841 | get_random_bytes(&time, sizeof(time)); | |
1842 | time = ORPHAN_SCAN_SCHEDULE_TIMEOUT + (time % 5000); | |
1843 | return msecs_to_jiffies(time); | |
1844 | } | |
1845 | ||
1846 | /* | |
1847 | * ocfs2_queue_orphan_scan calls ocfs2_queue_recovery_completion for | |
1848 | * every slot, queuing a recovery of the slot on the ocfs2_wq thread. This | |
1849 | * is done to catch any orphans that are left over in orphan directories. | |
1850 | * | |
a035bff6 SM |
1851 | * It scans all slots, even ones that are in use. It does so to handle the |
1852 | * case described below: | |
1853 | * | |
1854 | * Node 1 has an inode it was using. The dentry went away due to memory | |
1855 | * pressure. Node 1 closes the inode, but it's on the free list. The node | |
1856 | * has the open lock. | |
1857 | * Node 2 unlinks the inode. It grabs the dentry lock to notify others, | |
1858 | * but node 1 has no dentry and doesn't get the message. It trylocks the | |
1859 | * open lock, sees that another node has a PR, and does nothing. | |
1860 | * Later node 2 runs its orphan dir. It igets the inode, trylocks the | |
1861 | * open lock, sees the PR still, and does nothing. | |
1862 | * Basically, we have to trigger an orphan iput on node 1. The only way | |
1863 | * for this to happen is if node 1 runs node 2's orphan dir. | |
1864 | * | |
83273932 SE |
1865 | * ocfs2_queue_orphan_scan gets called every ORPHAN_SCAN_SCHEDULE_TIMEOUT |
1866 | * seconds. It gets an EX lock on os_lockres and checks sequence number | |
1867 | * stored in LVB. If the sequence number has changed, it means some other | |
1868 | * node has done the scan. This node skips the scan and tracks the | |
1869 | * sequence number. If the sequence number didn't change, it means a scan | |
1870 | * hasn't happened. The node queues a scan and increments the | |
1871 | * sequence number in the LVB. | |
1872 | */ | |
1873 | void ocfs2_queue_orphan_scan(struct ocfs2_super *osb) | |
1874 | { | |
1875 | struct ocfs2_orphan_scan *os; | |
1876 | int status, i; | |
1877 | u32 seqno = 0; | |
1878 | ||
1879 | os = &osb->osb_orphan_scan; | |
1880 | ||
692684e1 SM |
1881 | if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE) |
1882 | goto out; | |
1883 | ||
b4107950 TM |
1884 | trace_ocfs2_queue_orphan_scan_begin(os->os_count, os->os_seqno, |
1885 | atomic_read(&os->os_state)); | |
1886 | ||
df152c24 | 1887 | status = ocfs2_orphan_scan_lock(osb, &seqno); |
83273932 SE |
1888 | if (status < 0) { |
1889 | if (status != -EAGAIN) | |
1890 | mlog_errno(status); | |
1891 | goto out; | |
1892 | } | |
1893 | ||
692684e1 SM |
1894 | /* Do no queue the tasks if the volume is being umounted */ |
1895 | if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE) | |
1896 | goto unlock; | |
1897 | ||
83273932 SE |
1898 | if (os->os_seqno != seqno) { |
1899 | os->os_seqno = seqno; | |
1900 | goto unlock; | |
1901 | } | |
1902 | ||
1903 | for (i = 0; i < osb->max_slots; i++) | |
1904 | ocfs2_queue_recovery_completion(osb->journal, i, NULL, NULL, | |
1905 | NULL); | |
1906 | /* | |
1907 | * We queued a recovery on orphan slots, increment the sequence | |
1908 | * number and update LVB so other node will skip the scan for a while | |
1909 | */ | |
1910 | seqno++; | |
15633a22 SE |
1911 | os->os_count++; |
1912 | os->os_scantime = CURRENT_TIME; | |
83273932 | 1913 | unlock: |
df152c24 | 1914 | ocfs2_orphan_scan_unlock(osb, seqno); |
83273932 | 1915 | out: |
b4107950 TM |
1916 | trace_ocfs2_queue_orphan_scan_end(os->os_count, os->os_seqno, |
1917 | atomic_read(&os->os_state)); | |
83273932 SE |
1918 | return; |
1919 | } | |
1920 | ||
1921 | /* Worker task that gets fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT millsec */ | |
1922 | void ocfs2_orphan_scan_work(struct work_struct *work) | |
1923 | { | |
1924 | struct ocfs2_orphan_scan *os; | |
1925 | struct ocfs2_super *osb; | |
1926 | ||
1927 | os = container_of(work, struct ocfs2_orphan_scan, | |
1928 | os_orphan_scan_work.work); | |
1929 | osb = os->os_osb; | |
1930 | ||
1931 | mutex_lock(&os->os_lock); | |
1932 | ocfs2_queue_orphan_scan(osb); | |
692684e1 | 1933 | if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE) |
40f165f4 | 1934 | queue_delayed_work(ocfs2_wq, &os->os_orphan_scan_work, |
692684e1 | 1935 | ocfs2_orphan_scan_timeout()); |
83273932 SE |
1936 | mutex_unlock(&os->os_lock); |
1937 | } | |
1938 | ||
1939 | void ocfs2_orphan_scan_stop(struct ocfs2_super *osb) | |
1940 | { | |
1941 | struct ocfs2_orphan_scan *os; | |
1942 | ||
1943 | os = &osb->osb_orphan_scan; | |
df152c24 SM |
1944 | if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE) { |
1945 | atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE); | |
1946 | mutex_lock(&os->os_lock); | |
1947 | cancel_delayed_work(&os->os_orphan_scan_work); | |
1948 | mutex_unlock(&os->os_lock); | |
1949 | } | |
83273932 SE |
1950 | } |
1951 | ||
df152c24 | 1952 | void ocfs2_orphan_scan_init(struct ocfs2_super *osb) |
83273932 SE |
1953 | { |
1954 | struct ocfs2_orphan_scan *os; | |
1955 | ||
1956 | os = &osb->osb_orphan_scan; | |
1957 | os->os_osb = osb; | |
15633a22 | 1958 | os->os_count = 0; |
3211949f | 1959 | os->os_seqno = 0; |
83273932 | 1960 | mutex_init(&os->os_lock); |
df152c24 | 1961 | INIT_DELAYED_WORK(&os->os_orphan_scan_work, ocfs2_orphan_scan_work); |
8b712cd5 | 1962 | } |
83273932 | 1963 | |
8b712cd5 JM |
1964 | void ocfs2_orphan_scan_start(struct ocfs2_super *osb) |
1965 | { | |
1966 | struct ocfs2_orphan_scan *os; | |
1967 | ||
1968 | os = &osb->osb_orphan_scan; | |
1969 | os->os_scantime = CURRENT_TIME; | |
df152c24 SM |
1970 | if (ocfs2_is_hard_readonly(osb) || ocfs2_mount_local(osb)) |
1971 | atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE); | |
1972 | else { | |
1973 | atomic_set(&os->os_state, ORPHAN_SCAN_ACTIVE); | |
40f165f4 TM |
1974 | queue_delayed_work(ocfs2_wq, &os->os_orphan_scan_work, |
1975 | ocfs2_orphan_scan_timeout()); | |
df152c24 | 1976 | } |
83273932 SE |
1977 | } |
1978 | ||
5eae5b96 | 1979 | struct ocfs2_orphan_filldir_priv { |
3704412b | 1980 | struct dir_context ctx; |
5eae5b96 MF |
1981 | struct inode *head; |
1982 | struct ocfs2_super *osb; | |
1983 | }; | |
1984 | ||
1985 | static int ocfs2_orphan_filldir(void *priv, const char *name, int name_len, | |
1986 | loff_t pos, u64 ino, unsigned type) | |
1987 | { | |
1988 | struct ocfs2_orphan_filldir_priv *p = priv; | |
1989 | struct inode *iter; | |
1990 | ||
1991 | if (name_len == 1 && !strncmp(".", name, 1)) | |
1992 | return 0; | |
1993 | if (name_len == 2 && !strncmp("..", name, 2)) | |
1994 | return 0; | |
1995 | ||
1996 | /* Skip bad inodes so that recovery can continue */ | |
1997 | iter = ocfs2_iget(p->osb, ino, | |
5fa0613e | 1998 | OCFS2_FI_FLAG_ORPHAN_RECOVERY, 0); |
5eae5b96 MF |
1999 | if (IS_ERR(iter)) |
2000 | return 0; | |
2001 | ||
b4107950 | 2002 | trace_ocfs2_orphan_filldir((unsigned long long)OCFS2_I(iter)->ip_blkno); |
5eae5b96 MF |
2003 | /* No locking is required for the next_orphan queue as there |
2004 | * is only ever a single process doing orphan recovery. */ | |
2005 | OCFS2_I(iter)->ip_next_orphan = p->head; | |
2006 | p->head = iter; | |
2007 | ||
2008 | return 0; | |
2009 | } | |
2010 | ||
b4df6ed8 MF |
2011 | static int ocfs2_queue_orphans(struct ocfs2_super *osb, |
2012 | int slot, | |
2013 | struct inode **head) | |
ccd979bd | 2014 | { |
b4df6ed8 | 2015 | int status; |
ccd979bd | 2016 | struct inode *orphan_dir_inode = NULL; |
3704412b AV |
2017 | struct ocfs2_orphan_filldir_priv priv = { |
2018 | .ctx.actor = ocfs2_orphan_filldir, | |
2019 | .osb = osb, | |
2020 | .head = *head | |
2021 | }; | |
ccd979bd MF |
2022 | |
2023 | orphan_dir_inode = ocfs2_get_system_file_inode(osb, | |
2024 | ORPHAN_DIR_SYSTEM_INODE, | |
2025 | slot); | |
2026 | if (!orphan_dir_inode) { | |
2027 | status = -ENOENT; | |
2028 | mlog_errno(status); | |
b4df6ed8 | 2029 | return status; |
2bd63216 | 2030 | } |
ccd979bd | 2031 | |
1b1dcc1b | 2032 | mutex_lock(&orphan_dir_inode->i_mutex); |
e63aecb6 | 2033 | status = ocfs2_inode_lock(orphan_dir_inode, NULL, 0); |
ccd979bd | 2034 | if (status < 0) { |
ccd979bd MF |
2035 | mlog_errno(status); |
2036 | goto out; | |
2037 | } | |
ccd979bd | 2038 | |
3704412b | 2039 | status = ocfs2_dir_foreach(orphan_dir_inode, &priv.ctx); |
5eae5b96 MF |
2040 | if (status) { |
2041 | mlog_errno(status); | |
a86370fb | 2042 | goto out_cluster; |
ccd979bd | 2043 | } |
ccd979bd | 2044 | |
5eae5b96 MF |
2045 | *head = priv.head; |
2046 | ||
a86370fb | 2047 | out_cluster: |
e63aecb6 | 2048 | ocfs2_inode_unlock(orphan_dir_inode, 0); |
b4df6ed8 MF |
2049 | out: |
2050 | mutex_unlock(&orphan_dir_inode->i_mutex); | |
ccd979bd | 2051 | iput(orphan_dir_inode); |
b4df6ed8 MF |
2052 | return status; |
2053 | } | |
2054 | ||
2055 | static int ocfs2_orphan_recovery_can_continue(struct ocfs2_super *osb, | |
2056 | int slot) | |
2057 | { | |
2058 | int ret; | |
2059 | ||
2060 | spin_lock(&osb->osb_lock); | |
2061 | ret = !osb->osb_orphan_wipes[slot]; | |
2062 | spin_unlock(&osb->osb_lock); | |
2063 | return ret; | |
2064 | } | |
2065 | ||
2066 | static void ocfs2_mark_recovering_orphan_dir(struct ocfs2_super *osb, | |
2067 | int slot) | |
2068 | { | |
2069 | spin_lock(&osb->osb_lock); | |
2070 | /* Mark ourselves such that new processes in delete_inode() | |
2071 | * know to quit early. */ | |
2072 | ocfs2_node_map_set_bit(osb, &osb->osb_recovering_orphan_dirs, slot); | |
2073 | while (osb->osb_orphan_wipes[slot]) { | |
2074 | /* If any processes are already in the middle of an | |
2075 | * orphan wipe on this dir, then we need to wait for | |
2076 | * them. */ | |
2077 | spin_unlock(&osb->osb_lock); | |
2078 | wait_event_interruptible(osb->osb_wipe_event, | |
2079 | ocfs2_orphan_recovery_can_continue(osb, slot)); | |
2080 | spin_lock(&osb->osb_lock); | |
2081 | } | |
2082 | spin_unlock(&osb->osb_lock); | |
2083 | } | |
2084 | ||
2085 | static void ocfs2_clear_recovering_orphan_dir(struct ocfs2_super *osb, | |
2086 | int slot) | |
2087 | { | |
2088 | ocfs2_node_map_clear_bit(osb, &osb->osb_recovering_orphan_dirs, slot); | |
2089 | } | |
2090 | ||
2091 | /* | |
2092 | * Orphan recovery. Each mounted node has it's own orphan dir which we | |
2093 | * must run during recovery. Our strategy here is to build a list of | |
2094 | * the inodes in the orphan dir and iget/iput them. The VFS does | |
2095 | * (most) of the rest of the work. | |
2096 | * | |
2097 | * Orphan recovery can happen at any time, not just mount so we have a | |
2098 | * couple of extra considerations. | |
2099 | * | |
2100 | * - We grab as many inodes as we can under the orphan dir lock - | |
2101 | * doing iget() outside the orphan dir risks getting a reference on | |
2102 | * an invalid inode. | |
2103 | * - We must be sure not to deadlock with other processes on the | |
2104 | * system wanting to run delete_inode(). This can happen when they go | |
2105 | * to lock the orphan dir and the orphan recovery process attempts to | |
2106 | * iget() inside the orphan dir lock. This can be avoided by | |
2107 | * advertising our state to ocfs2_delete_inode(). | |
2108 | */ | |
2109 | static int ocfs2_recover_orphans(struct ocfs2_super *osb, | |
2110 | int slot) | |
2111 | { | |
2112 | int ret = 0; | |
2113 | struct inode *inode = NULL; | |
2114 | struct inode *iter; | |
2115 | struct ocfs2_inode_info *oi; | |
2116 | ||
b4107950 | 2117 | trace_ocfs2_recover_orphans(slot); |
b4df6ed8 MF |
2118 | |
2119 | ocfs2_mark_recovering_orphan_dir(osb, slot); | |
2120 | ret = ocfs2_queue_orphans(osb, slot, &inode); | |
2121 | ocfs2_clear_recovering_orphan_dir(osb, slot); | |
2122 | ||
2123 | /* Error here should be noted, but we want to continue with as | |
2124 | * many queued inodes as we've got. */ | |
2125 | if (ret) | |
2126 | mlog_errno(ret); | |
ccd979bd MF |
2127 | |
2128 | while (inode) { | |
2129 | oi = OCFS2_I(inode); | |
b4107950 TM |
2130 | trace_ocfs2_recover_orphans_iput( |
2131 | (unsigned long long)oi->ip_blkno); | |
ccd979bd MF |
2132 | |
2133 | iter = oi->ip_next_orphan; | |
2134 | ||
2135 | spin_lock(&oi->ip_lock); | |
ccd979bd MF |
2136 | /* Set the proper information to get us going into |
2137 | * ocfs2_delete_inode. */ | |
2138 | oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED; | |
ccd979bd MF |
2139 | spin_unlock(&oi->ip_lock); |
2140 | ||
2141 | iput(inode); | |
2142 | ||
2143 | inode = iter; | |
2144 | } | |
2145 | ||
b4df6ed8 | 2146 | return ret; |
ccd979bd MF |
2147 | } |
2148 | ||
19ece546 | 2149 | static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota) |
ccd979bd MF |
2150 | { |
2151 | /* This check is good because ocfs2 will wait on our recovery | |
2152 | * thread before changing it to something other than MOUNTED | |
2153 | * or DISABLED. */ | |
2154 | wait_event(osb->osb_mount_event, | |
19ece546 JK |
2155 | (!quota && atomic_read(&osb->vol_state) == VOLUME_MOUNTED) || |
2156 | atomic_read(&osb->vol_state) == VOLUME_MOUNTED_QUOTAS || | |
ccd979bd MF |
2157 | atomic_read(&osb->vol_state) == VOLUME_DISABLED); |
2158 | ||
2159 | /* If there's an error on mount, then we may never get to the | |
2160 | * MOUNTED flag, but this is set right before | |
2161 | * dismount_volume() so we can trust it. */ | |
2162 | if (atomic_read(&osb->vol_state) == VOLUME_DISABLED) { | |
b4107950 | 2163 | trace_ocfs2_wait_on_mount(VOLUME_DISABLED); |
ccd979bd MF |
2164 | mlog(0, "mount error, exiting!\n"); |
2165 | return -EBUSY; | |
2166 | } | |
2167 | ||
2168 | return 0; | |
2169 | } | |
2170 | ||
2171 | static int ocfs2_commit_thread(void *arg) | |
2172 | { | |
2173 | int status; | |
2174 | struct ocfs2_super *osb = arg; | |
2175 | struct ocfs2_journal *journal = osb->journal; | |
2176 | ||
2177 | /* we can trust j_num_trans here because _should_stop() is only set in | |
2178 | * shutdown and nobody other than ourselves should be able to start | |
2179 | * transactions. committing on shutdown might take a few iterations | |
2180 | * as final transactions put deleted inodes on the list */ | |
2181 | while (!(kthread_should_stop() && | |
2182 | atomic_read(&journal->j_num_trans) == 0)) { | |
2183 | ||
745ae8ba MF |
2184 | wait_event_interruptible(osb->checkpoint_event, |
2185 | atomic_read(&journal->j_num_trans) | |
2186 | || kthread_should_stop()); | |
ccd979bd MF |
2187 | |
2188 | status = ocfs2_commit_cache(osb); | |
55b465b6 JQ |
2189 | if (status < 0) { |
2190 | static unsigned long abort_warn_time; | |
2191 | ||
2192 | /* Warn about this once per minute */ | |
2193 | if (printk_timed_ratelimit(&abort_warn_time, 60*HZ)) | |
2194 | mlog(ML_ERROR, "status = %d, journal is " | |
2195 | "already aborted.\n", status); | |
2196 | /* | |
2197 | * After ocfs2_commit_cache() fails, j_num_trans has a | |
2198 | * non-zero value. Sleep here to avoid a busy-wait | |
2199 | * loop. | |
2200 | */ | |
2201 | msleep_interruptible(1000); | |
2202 | } | |
ccd979bd MF |
2203 | |
2204 | if (kthread_should_stop() && atomic_read(&journal->j_num_trans)){ | |
2205 | mlog(ML_KTHREAD, | |
2206 | "commit_thread: %u transactions pending on " | |
2207 | "shutdown\n", | |
2208 | atomic_read(&journal->j_num_trans)); | |
2209 | } | |
2210 | } | |
2211 | ||
2212 | return 0; | |
2213 | } | |
2214 | ||
539d8264 SM |
2215 | /* Reads all the journal inodes without taking any cluster locks. Used |
2216 | * for hard readonly access to determine whether any journal requires | |
2217 | * recovery. Also used to refresh the recovery generation numbers after | |
2218 | * a journal has been recovered by another node. | |
2219 | */ | |
ccd979bd MF |
2220 | int ocfs2_check_journals_nolocks(struct ocfs2_super *osb) |
2221 | { | |
2222 | int ret = 0; | |
2223 | unsigned int slot; | |
539d8264 | 2224 | struct buffer_head *di_bh = NULL; |
ccd979bd | 2225 | struct ocfs2_dinode *di; |
539d8264 | 2226 | int journal_dirty = 0; |
ccd979bd MF |
2227 | |
2228 | for(slot = 0; slot < osb->max_slots; slot++) { | |
539d8264 SM |
2229 | ret = ocfs2_read_journal_inode(osb, slot, &di_bh, NULL); |
2230 | if (ret) { | |
ccd979bd MF |
2231 | mlog_errno(ret); |
2232 | goto out; | |
2233 | } | |
2234 | ||
2235 | di = (struct ocfs2_dinode *) di_bh->b_data; | |
2236 | ||
539d8264 SM |
2237 | osb->slot_recovery_generations[slot] = |
2238 | ocfs2_get_recovery_generation(di); | |
2239 | ||
ccd979bd MF |
2240 | if (le32_to_cpu(di->id1.journal1.ij_flags) & |
2241 | OCFS2_JOURNAL_DIRTY_FL) | |
539d8264 | 2242 | journal_dirty = 1; |
ccd979bd MF |
2243 | |
2244 | brelse(di_bh); | |
539d8264 | 2245 | di_bh = NULL; |
ccd979bd MF |
2246 | } |
2247 | ||
2248 | out: | |
539d8264 SM |
2249 | if (journal_dirty) |
2250 | ret = -EROFS; | |
ccd979bd MF |
2251 | return ret; |
2252 | } |