]>
Commit | Line | Data |
---|---|---|
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 | #include <linux/time.h> | |
32 | #include <linux/random.h> | |
33 | #include <linux/delay.h> | |
34 | ||
35 | #include <cluster/masklog.h> | |
36 | ||
37 | #include "ocfs2.h" | |
38 | ||
39 | #include "alloc.h" | |
40 | #include "blockcheck.h" | |
41 | #include "dir.h" | |
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" | |
48 | #include "slot_map.h" | |
49 | #include "super.h" | |
50 | #include "sysfile.h" | |
51 | #include "uptodate.h" | |
52 | #include "quota.h" | |
53 | #include "file.h" | |
54 | #include "namei.h" | |
55 | ||
56 | #include "buffer_head_io.h" | |
57 | #include "ocfs2_trace.h" | |
58 | ||
59 | DEFINE_SPINLOCK(trans_inc_lock); | |
60 | ||
61 | #define ORPHAN_SCAN_SCHEDULE_TIMEOUT 300000 | |
62 | ||
63 | static int ocfs2_force_read_journal(struct inode *inode); | |
64 | static int ocfs2_recover_node(struct ocfs2_super *osb, | |
65 | int node_num, int slot_num); | |
66 | static int __ocfs2_recovery_thread(void *arg); | |
67 | static int ocfs2_commit_cache(struct ocfs2_super *osb); | |
68 | static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota); | |
69 | static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb, | |
70 | int dirty, int replayed); | |
71 | static int ocfs2_trylock_journal(struct ocfs2_super *osb, | |
72 | int slot_num); | |
73 | static int ocfs2_recover_orphans(struct ocfs2_super *osb, | |
74 | int slot, | |
75 | enum ocfs2_orphan_reco_type orphan_reco_type); | |
76 | static int ocfs2_commit_thread(void *arg); | |
77 | static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal, | |
78 | int slot_num, | |
79 | struct ocfs2_dinode *la_dinode, | |
80 | struct ocfs2_dinode *tl_dinode, | |
81 | struct ocfs2_quota_recovery *qrec, | |
82 | enum ocfs2_orphan_reco_type orphan_reco_type); | |
83 | ||
84 | static inline int ocfs2_wait_on_mount(struct ocfs2_super *osb) | |
85 | { | |
86 | return __ocfs2_wait_on_mount(osb, 0); | |
87 | } | |
88 | ||
89 | static inline int ocfs2_wait_on_quotas(struct ocfs2_super *osb) | |
90 | { | |
91 | return __ocfs2_wait_on_mount(osb, 1); | |
92 | } | |
93 | ||
94 | /* | |
95 | * This replay_map is to track online/offline slots, so we could recover | |
96 | * offline slots during recovery and mount | |
97 | */ | |
98 | ||
99 | enum ocfs2_replay_state { | |
100 | REPLAY_UNNEEDED = 0, /* Replay is not needed, so ignore this map */ | |
101 | REPLAY_NEEDED, /* Replay slots marked in rm_replay_slots */ | |
102 | REPLAY_DONE /* Replay was already queued */ | |
103 | }; | |
104 | ||
105 | struct ocfs2_replay_map { | |
106 | unsigned int rm_slots; | |
107 | enum ocfs2_replay_state rm_state; | |
108 | unsigned char rm_replay_slots[0]; | |
109 | }; | |
110 | ||
111 | void ocfs2_replay_map_set_state(struct ocfs2_super *osb, int state) | |
112 | { | |
113 | if (!osb->replay_map) | |
114 | return; | |
115 | ||
116 | /* If we've already queued the replay, we don't have any more to do */ | |
117 | if (osb->replay_map->rm_state == REPLAY_DONE) | |
118 | return; | |
119 | ||
120 | osb->replay_map->rm_state = state; | |
121 | } | |
122 | ||
123 | int ocfs2_compute_replay_slots(struct ocfs2_super *osb) | |
124 | { | |
125 | struct ocfs2_replay_map *replay_map; | |
126 | int i, node_num; | |
127 | ||
128 | /* If replay map is already set, we don't do it again */ | |
129 | if (osb->replay_map) | |
130 | return 0; | |
131 | ||
132 | replay_map = kzalloc(sizeof(struct ocfs2_replay_map) + | |
133 | (osb->max_slots * sizeof(char)), GFP_KERNEL); | |
134 | ||
135 | if (!replay_map) { | |
136 | mlog_errno(-ENOMEM); | |
137 | return -ENOMEM; | |
138 | } | |
139 | ||
140 | spin_lock(&osb->osb_lock); | |
141 | ||
142 | replay_map->rm_slots = osb->max_slots; | |
143 | replay_map->rm_state = REPLAY_UNNEEDED; | |
144 | ||
145 | /* set rm_replay_slots for offline slot(s) */ | |
146 | for (i = 0; i < replay_map->rm_slots; i++) { | |
147 | if (ocfs2_slot_to_node_num_locked(osb, i, &node_num) == -ENOENT) | |
148 | replay_map->rm_replay_slots[i] = 1; | |
149 | } | |
150 | ||
151 | osb->replay_map = replay_map; | |
152 | spin_unlock(&osb->osb_lock); | |
153 | return 0; | |
154 | } | |
155 | ||
156 | void ocfs2_queue_replay_slots(struct ocfs2_super *osb, | |
157 | enum ocfs2_orphan_reco_type orphan_reco_type) | |
158 | { | |
159 | struct ocfs2_replay_map *replay_map = osb->replay_map; | |
160 | int i; | |
161 | ||
162 | if (!replay_map) | |
163 | return; | |
164 | ||
165 | if (replay_map->rm_state != REPLAY_NEEDED) | |
166 | return; | |
167 | ||
168 | for (i = 0; i < replay_map->rm_slots; i++) | |
169 | if (replay_map->rm_replay_slots[i]) | |
170 | ocfs2_queue_recovery_completion(osb->journal, i, NULL, | |
171 | NULL, NULL, | |
172 | orphan_reco_type); | |
173 | replay_map->rm_state = REPLAY_DONE; | |
174 | } | |
175 | ||
176 | void ocfs2_free_replay_slots(struct ocfs2_super *osb) | |
177 | { | |
178 | struct ocfs2_replay_map *replay_map = osb->replay_map; | |
179 | ||
180 | if (!osb->replay_map) | |
181 | return; | |
182 | ||
183 | kfree(replay_map); | |
184 | osb->replay_map = NULL; | |
185 | } | |
186 | ||
187 | int ocfs2_recovery_init(struct ocfs2_super *osb) | |
188 | { | |
189 | struct ocfs2_recovery_map *rm; | |
190 | ||
191 | mutex_init(&osb->recovery_lock); | |
192 | osb->disable_recovery = 0; | |
193 | osb->recovery_thread_task = NULL; | |
194 | init_waitqueue_head(&osb->recovery_event); | |
195 | ||
196 | rm = kzalloc(sizeof(struct ocfs2_recovery_map) + | |
197 | osb->max_slots * sizeof(unsigned int), | |
198 | GFP_KERNEL); | |
199 | if (!rm) { | |
200 | mlog_errno(-ENOMEM); | |
201 | return -ENOMEM; | |
202 | } | |
203 | ||
204 | rm->rm_entries = (unsigned int *)((char *)rm + | |
205 | sizeof(struct ocfs2_recovery_map)); | |
206 | osb->recovery_map = rm; | |
207 | ||
208 | return 0; | |
209 | } | |
210 | ||
211 | /* we can't grab the goofy sem lock from inside wait_event, so we use | |
212 | * memory barriers to make sure that we'll see the null task before | |
213 | * being woken up */ | |
214 | static int ocfs2_recovery_thread_running(struct ocfs2_super *osb) | |
215 | { | |
216 | mb(); | |
217 | return osb->recovery_thread_task != NULL; | |
218 | } | |
219 | ||
220 | void ocfs2_recovery_exit(struct ocfs2_super *osb) | |
221 | { | |
222 | struct ocfs2_recovery_map *rm; | |
223 | ||
224 | /* disable any new recovery threads and wait for any currently | |
225 | * running ones to exit. Do this before setting the vol_state. */ | |
226 | mutex_lock(&osb->recovery_lock); | |
227 | osb->disable_recovery = 1; | |
228 | mutex_unlock(&osb->recovery_lock); | |
229 | wait_event(osb->recovery_event, !ocfs2_recovery_thread_running(osb)); | |
230 | ||
231 | /* At this point, we know that no more recovery threads can be | |
232 | * launched, so wait for any recovery completion work to | |
233 | * complete. */ | |
234 | flush_workqueue(ocfs2_wq); | |
235 | ||
236 | /* | |
237 | * Now that recovery is shut down, and the osb is about to be | |
238 | * freed, the osb_lock is not taken here. | |
239 | */ | |
240 | rm = osb->recovery_map; | |
241 | /* XXX: Should we bug if there are dirty entries? */ | |
242 | ||
243 | kfree(rm); | |
244 | } | |
245 | ||
246 | static int __ocfs2_recovery_map_test(struct ocfs2_super *osb, | |
247 | unsigned int node_num) | |
248 | { | |
249 | int i; | |
250 | struct ocfs2_recovery_map *rm = osb->recovery_map; | |
251 | ||
252 | assert_spin_locked(&osb->osb_lock); | |
253 | ||
254 | for (i = 0; i < rm->rm_used; i++) { | |
255 | if (rm->rm_entries[i] == node_num) | |
256 | return 1; | |
257 | } | |
258 | ||
259 | return 0; | |
260 | } | |
261 | ||
262 | /* Behaves like test-and-set. Returns the previous value */ | |
263 | static int ocfs2_recovery_map_set(struct ocfs2_super *osb, | |
264 | unsigned int node_num) | |
265 | { | |
266 | struct ocfs2_recovery_map *rm = osb->recovery_map; | |
267 | ||
268 | spin_lock(&osb->osb_lock); | |
269 | if (__ocfs2_recovery_map_test(osb, node_num)) { | |
270 | spin_unlock(&osb->osb_lock); | |
271 | return 1; | |
272 | } | |
273 | ||
274 | /* XXX: Can this be exploited? Not from o2dlm... */ | |
275 | BUG_ON(rm->rm_used >= osb->max_slots); | |
276 | ||
277 | rm->rm_entries[rm->rm_used] = node_num; | |
278 | rm->rm_used++; | |
279 | spin_unlock(&osb->osb_lock); | |
280 | ||
281 | return 0; | |
282 | } | |
283 | ||
284 | static void ocfs2_recovery_map_clear(struct ocfs2_super *osb, | |
285 | unsigned int node_num) | |
286 | { | |
287 | int i; | |
288 | struct ocfs2_recovery_map *rm = osb->recovery_map; | |
289 | ||
290 | spin_lock(&osb->osb_lock); | |
291 | ||
292 | for (i = 0; i < rm->rm_used; i++) { | |
293 | if (rm->rm_entries[i] == node_num) | |
294 | break; | |
295 | } | |
296 | ||
297 | if (i < rm->rm_used) { | |
298 | /* XXX: be careful with the pointer math */ | |
299 | memmove(&(rm->rm_entries[i]), &(rm->rm_entries[i + 1]), | |
300 | (rm->rm_used - i - 1) * sizeof(unsigned int)); | |
301 | rm->rm_used--; | |
302 | } | |
303 | ||
304 | spin_unlock(&osb->osb_lock); | |
305 | } | |
306 | ||
307 | static int ocfs2_commit_cache(struct ocfs2_super *osb) | |
308 | { | |
309 | int status = 0; | |
310 | unsigned int flushed; | |
311 | struct ocfs2_journal *journal = NULL; | |
312 | ||
313 | journal = osb->journal; | |
314 | ||
315 | /* Flush all pending commits and checkpoint the journal. */ | |
316 | down_write(&journal->j_trans_barrier); | |
317 | ||
318 | flushed = atomic_read(&journal->j_num_trans); | |
319 | trace_ocfs2_commit_cache_begin(flushed); | |
320 | if (flushed == 0) { | |
321 | up_write(&journal->j_trans_barrier); | |
322 | goto finally; | |
323 | } | |
324 | ||
325 | jbd2_journal_lock_updates(journal->j_journal); | |
326 | status = jbd2_journal_flush(journal->j_journal); | |
327 | jbd2_journal_unlock_updates(journal->j_journal); | |
328 | if (status < 0) { | |
329 | up_write(&journal->j_trans_barrier); | |
330 | mlog_errno(status); | |
331 | goto finally; | |
332 | } | |
333 | ||
334 | ocfs2_inc_trans_id(journal); | |
335 | ||
336 | flushed = atomic_read(&journal->j_num_trans); | |
337 | atomic_set(&journal->j_num_trans, 0); | |
338 | up_write(&journal->j_trans_barrier); | |
339 | ||
340 | trace_ocfs2_commit_cache_end(journal->j_trans_id, flushed); | |
341 | ||
342 | ocfs2_wake_downconvert_thread(osb); | |
343 | wake_up(&journal->j_checkpointed); | |
344 | finally: | |
345 | return status; | |
346 | } | |
347 | ||
348 | handle_t *ocfs2_start_trans(struct ocfs2_super *osb, int max_buffs) | |
349 | { | |
350 | journal_t *journal = osb->journal->j_journal; | |
351 | handle_t *handle; | |
352 | ||
353 | BUG_ON(!osb || !osb->journal->j_journal); | |
354 | ||
355 | if (ocfs2_is_hard_readonly(osb)) | |
356 | return ERR_PTR(-EROFS); | |
357 | ||
358 | BUG_ON(osb->journal->j_state == OCFS2_JOURNAL_FREE); | |
359 | BUG_ON(max_buffs <= 0); | |
360 | ||
361 | /* Nested transaction? Just return the handle... */ | |
362 | if (journal_current_handle()) | |
363 | return jbd2_journal_start(journal, max_buffs); | |
364 | ||
365 | sb_start_intwrite(osb->sb); | |
366 | ||
367 | down_read(&osb->journal->j_trans_barrier); | |
368 | ||
369 | handle = jbd2_journal_start(journal, max_buffs); | |
370 | if (IS_ERR(handle)) { | |
371 | up_read(&osb->journal->j_trans_barrier); | |
372 | sb_end_intwrite(osb->sb); | |
373 | ||
374 | mlog_errno(PTR_ERR(handle)); | |
375 | ||
376 | if (is_journal_aborted(journal)) { | |
377 | ocfs2_abort(osb->sb, "Detected aborted journal"); | |
378 | handle = ERR_PTR(-EROFS); | |
379 | } | |
380 | } else { | |
381 | if (!ocfs2_mount_local(osb)) | |
382 | atomic_inc(&(osb->journal->j_num_trans)); | |
383 | } | |
384 | ||
385 | return handle; | |
386 | } | |
387 | ||
388 | int ocfs2_commit_trans(struct ocfs2_super *osb, | |
389 | handle_t *handle) | |
390 | { | |
391 | int ret, nested; | |
392 | struct ocfs2_journal *journal = osb->journal; | |
393 | ||
394 | BUG_ON(!handle); | |
395 | ||
396 | nested = handle->h_ref > 1; | |
397 | ret = jbd2_journal_stop(handle); | |
398 | if (ret < 0) | |
399 | mlog_errno(ret); | |
400 | ||
401 | if (!nested) { | |
402 | up_read(&journal->j_trans_barrier); | |
403 | sb_end_intwrite(osb->sb); | |
404 | } | |
405 | ||
406 | return ret; | |
407 | } | |
408 | ||
409 | /* | |
410 | * 'nblocks' is what you want to add to the current transaction. | |
411 | * | |
412 | * This might call jbd2_journal_restart() which will commit dirty buffers | |
413 | * and then restart the transaction. Before calling | |
414 | * ocfs2_extend_trans(), any changed blocks should have been | |
415 | * dirtied. After calling it, all blocks which need to be changed must | |
416 | * go through another set of journal_access/journal_dirty calls. | |
417 | * | |
418 | * WARNING: This will not release any semaphores or disk locks taken | |
419 | * during the transaction, so make sure they were taken *before* | |
420 | * start_trans or we'll have ordering deadlocks. | |
421 | * | |
422 | * WARNING2: Note that we do *not* drop j_trans_barrier here. This is | |
423 | * good because transaction ids haven't yet been recorded on the | |
424 | * cluster locks associated with this handle. | |
425 | */ | |
426 | int ocfs2_extend_trans(handle_t *handle, int nblocks) | |
427 | { | |
428 | int status, old_nblocks; | |
429 | ||
430 | BUG_ON(!handle); | |
431 | BUG_ON(nblocks < 0); | |
432 | ||
433 | if (!nblocks) | |
434 | return 0; | |
435 | ||
436 | old_nblocks = handle->h_buffer_credits; | |
437 | ||
438 | trace_ocfs2_extend_trans(old_nblocks, nblocks); | |
439 | ||
440 | #ifdef CONFIG_OCFS2_DEBUG_FS | |
441 | status = 1; | |
442 | #else | |
443 | status = jbd2_journal_extend(handle, nblocks); | |
444 | if (status < 0) { | |
445 | mlog_errno(status); | |
446 | goto bail; | |
447 | } | |
448 | #endif | |
449 | ||
450 | if (status > 0) { | |
451 | trace_ocfs2_extend_trans_restart(old_nblocks + nblocks); | |
452 | status = jbd2_journal_restart(handle, | |
453 | old_nblocks + nblocks); | |
454 | if (status < 0) { | |
455 | mlog_errno(status); | |
456 | goto bail; | |
457 | } | |
458 | } | |
459 | ||
460 | status = 0; | |
461 | bail: | |
462 | return status; | |
463 | } | |
464 | ||
465 | /* | |
466 | * If we have fewer than thresh credits, extend by OCFS2_MAX_TRANS_DATA. | |
467 | * If that fails, restart the transaction & regain write access for the | |
468 | * buffer head which is used for metadata modifications. | |
469 | * Taken from Ext4: extend_or_restart_transaction() | |
470 | */ | |
471 | int ocfs2_allocate_extend_trans(handle_t *handle, int thresh) | |
472 | { | |
473 | int status, old_nblks; | |
474 | ||
475 | BUG_ON(!handle); | |
476 | ||
477 | old_nblks = handle->h_buffer_credits; | |
478 | trace_ocfs2_allocate_extend_trans(old_nblks, thresh); | |
479 | ||
480 | if (old_nblks < thresh) | |
481 | return 0; | |
482 | ||
483 | status = jbd2_journal_extend(handle, OCFS2_MAX_TRANS_DATA); | |
484 | if (status < 0) { | |
485 | mlog_errno(status); | |
486 | goto bail; | |
487 | } | |
488 | ||
489 | if (status > 0) { | |
490 | status = jbd2_journal_restart(handle, OCFS2_MAX_TRANS_DATA); | |
491 | if (status < 0) | |
492 | mlog_errno(status); | |
493 | } | |
494 | ||
495 | bail: | |
496 | return status; | |
497 | } | |
498 | ||
499 | ||
500 | struct ocfs2_triggers { | |
501 | struct jbd2_buffer_trigger_type ot_triggers; | |
502 | int ot_offset; | |
503 | }; | |
504 | ||
505 | static inline struct ocfs2_triggers *to_ocfs2_trigger(struct jbd2_buffer_trigger_type *triggers) | |
506 | { | |
507 | return container_of(triggers, struct ocfs2_triggers, ot_triggers); | |
508 | } | |
509 | ||
510 | static void ocfs2_frozen_trigger(struct jbd2_buffer_trigger_type *triggers, | |
511 | struct buffer_head *bh, | |
512 | void *data, size_t size) | |
513 | { | |
514 | struct ocfs2_triggers *ot = to_ocfs2_trigger(triggers); | |
515 | ||
516 | /* | |
517 | * We aren't guaranteed to have the superblock here, so we | |
518 | * must unconditionally compute the ecc data. | |
519 | * __ocfs2_journal_access() will only set the triggers if | |
520 | * metaecc is enabled. | |
521 | */ | |
522 | ocfs2_block_check_compute(data, size, data + ot->ot_offset); | |
523 | } | |
524 | ||
525 | /* | |
526 | * Quota blocks have their own trigger because the struct ocfs2_block_check | |
527 | * offset depends on the blocksize. | |
528 | */ | |
529 | static void ocfs2_dq_frozen_trigger(struct jbd2_buffer_trigger_type *triggers, | |
530 | struct buffer_head *bh, | |
531 | void *data, size_t size) | |
532 | { | |
533 | struct ocfs2_disk_dqtrailer *dqt = | |
534 | ocfs2_block_dqtrailer(size, data); | |
535 | ||
536 | /* | |
537 | * We aren't guaranteed to have the superblock here, so we | |
538 | * must unconditionally compute the ecc data. | |
539 | * __ocfs2_journal_access() will only set the triggers if | |
540 | * metaecc is enabled. | |
541 | */ | |
542 | ocfs2_block_check_compute(data, size, &dqt->dq_check); | |
543 | } | |
544 | ||
545 | /* | |
546 | * Directory blocks also have their own trigger because the | |
547 | * struct ocfs2_block_check offset depends on the blocksize. | |
548 | */ | |
549 | static void ocfs2_db_frozen_trigger(struct jbd2_buffer_trigger_type *triggers, | |
550 | struct buffer_head *bh, | |
551 | void *data, size_t size) | |
552 | { | |
553 | struct ocfs2_dir_block_trailer *trailer = | |
554 | ocfs2_dir_trailer_from_size(size, data); | |
555 | ||
556 | /* | |
557 | * We aren't guaranteed to have the superblock here, so we | |
558 | * must unconditionally compute the ecc data. | |
559 | * __ocfs2_journal_access() will only set the triggers if | |
560 | * metaecc is enabled. | |
561 | */ | |
562 | ocfs2_block_check_compute(data, size, &trailer->db_check); | |
563 | } | |
564 | ||
565 | static void ocfs2_abort_trigger(struct jbd2_buffer_trigger_type *triggers, | |
566 | struct buffer_head *bh) | |
567 | { | |
568 | mlog(ML_ERROR, | |
569 | "ocfs2_abort_trigger called by JBD2. bh = 0x%lx, " | |
570 | "bh->b_blocknr = %llu\n", | |
571 | (unsigned long)bh, | |
572 | (unsigned long long)bh->b_blocknr); | |
573 | ||
574 | /* We aren't guaranteed to have the superblock here - but if we | |
575 | * don't, it'll just crash. */ | |
576 | ocfs2_error(bh->b_assoc_map->host->i_sb, | |
577 | "JBD2 has aborted our journal, ocfs2 cannot continue\n"); | |
578 | } | |
579 | ||
580 | static struct ocfs2_triggers di_triggers = { | |
581 | .ot_triggers = { | |
582 | .t_frozen = ocfs2_frozen_trigger, | |
583 | .t_abort = ocfs2_abort_trigger, | |
584 | }, | |
585 | .ot_offset = offsetof(struct ocfs2_dinode, i_check), | |
586 | }; | |
587 | ||
588 | static struct ocfs2_triggers eb_triggers = { | |
589 | .ot_triggers = { | |
590 | .t_frozen = ocfs2_frozen_trigger, | |
591 | .t_abort = ocfs2_abort_trigger, | |
592 | }, | |
593 | .ot_offset = offsetof(struct ocfs2_extent_block, h_check), | |
594 | }; | |
595 | ||
596 | static struct ocfs2_triggers rb_triggers = { | |
597 | .ot_triggers = { | |
598 | .t_frozen = ocfs2_frozen_trigger, | |
599 | .t_abort = ocfs2_abort_trigger, | |
600 | }, | |
601 | .ot_offset = offsetof(struct ocfs2_refcount_block, rf_check), | |
602 | }; | |
603 | ||
604 | static struct ocfs2_triggers gd_triggers = { | |
605 | .ot_triggers = { | |
606 | .t_frozen = ocfs2_frozen_trigger, | |
607 | .t_abort = ocfs2_abort_trigger, | |
608 | }, | |
609 | .ot_offset = offsetof(struct ocfs2_group_desc, bg_check), | |
610 | }; | |
611 | ||
612 | static struct ocfs2_triggers db_triggers = { | |
613 | .ot_triggers = { | |
614 | .t_frozen = ocfs2_db_frozen_trigger, | |
615 | .t_abort = ocfs2_abort_trigger, | |
616 | }, | |
617 | }; | |
618 | ||
619 | static struct ocfs2_triggers xb_triggers = { | |
620 | .ot_triggers = { | |
621 | .t_frozen = ocfs2_frozen_trigger, | |
622 | .t_abort = ocfs2_abort_trigger, | |
623 | }, | |
624 | .ot_offset = offsetof(struct ocfs2_xattr_block, xb_check), | |
625 | }; | |
626 | ||
627 | static struct ocfs2_triggers dq_triggers = { | |
628 | .ot_triggers = { | |
629 | .t_frozen = ocfs2_dq_frozen_trigger, | |
630 | .t_abort = ocfs2_abort_trigger, | |
631 | }, | |
632 | }; | |
633 | ||
634 | static struct ocfs2_triggers dr_triggers = { | |
635 | .ot_triggers = { | |
636 | .t_frozen = ocfs2_frozen_trigger, | |
637 | .t_abort = ocfs2_abort_trigger, | |
638 | }, | |
639 | .ot_offset = offsetof(struct ocfs2_dx_root_block, dr_check), | |
640 | }; | |
641 | ||
642 | static struct ocfs2_triggers dl_triggers = { | |
643 | .ot_triggers = { | |
644 | .t_frozen = ocfs2_frozen_trigger, | |
645 | .t_abort = ocfs2_abort_trigger, | |
646 | }, | |
647 | .ot_offset = offsetof(struct ocfs2_dx_leaf, dl_check), | |
648 | }; | |
649 | ||
650 | static int __ocfs2_journal_access(handle_t *handle, | |
651 | struct ocfs2_caching_info *ci, | |
652 | struct buffer_head *bh, | |
653 | struct ocfs2_triggers *triggers, | |
654 | int type) | |
655 | { | |
656 | int status; | |
657 | struct ocfs2_super *osb = | |
658 | OCFS2_SB(ocfs2_metadata_cache_get_super(ci)); | |
659 | ||
660 | BUG_ON(!ci || !ci->ci_ops); | |
661 | BUG_ON(!handle); | |
662 | BUG_ON(!bh); | |
663 | ||
664 | trace_ocfs2_journal_access( | |
665 | (unsigned long long)ocfs2_metadata_cache_owner(ci), | |
666 | (unsigned long long)bh->b_blocknr, type, bh->b_size); | |
667 | ||
668 | /* we can safely remove this assertion after testing. */ | |
669 | if (!buffer_uptodate(bh)) { | |
670 | mlog(ML_ERROR, "giving me a buffer that's not uptodate!\n"); | |
671 | mlog(ML_ERROR, "b_blocknr=%llu\n", | |
672 | (unsigned long long)bh->b_blocknr); | |
673 | BUG(); | |
674 | } | |
675 | ||
676 | /* Set the current transaction information on the ci so | |
677 | * that the locking code knows whether it can drop it's locks | |
678 | * on this ci or not. We're protected from the commit | |
679 | * thread updating the current transaction id until | |
680 | * ocfs2_commit_trans() because ocfs2_start_trans() took | |
681 | * j_trans_barrier for us. */ | |
682 | ocfs2_set_ci_lock_trans(osb->journal, ci); | |
683 | ||
684 | ocfs2_metadata_cache_io_lock(ci); | |
685 | switch (type) { | |
686 | case OCFS2_JOURNAL_ACCESS_CREATE: | |
687 | case OCFS2_JOURNAL_ACCESS_WRITE: | |
688 | status = jbd2_journal_get_write_access(handle, bh); | |
689 | break; | |
690 | ||
691 | case OCFS2_JOURNAL_ACCESS_UNDO: | |
692 | status = jbd2_journal_get_undo_access(handle, bh); | |
693 | break; | |
694 | ||
695 | default: | |
696 | status = -EINVAL; | |
697 | mlog(ML_ERROR, "Unknown access type!\n"); | |
698 | } | |
699 | if (!status && ocfs2_meta_ecc(osb) && triggers) | |
700 | jbd2_journal_set_triggers(bh, &triggers->ot_triggers); | |
701 | ocfs2_metadata_cache_io_unlock(ci); | |
702 | ||
703 | if (status < 0) | |
704 | mlog(ML_ERROR, "Error %d getting %d access to buffer!\n", | |
705 | status, type); | |
706 | ||
707 | return status; | |
708 | } | |
709 | ||
710 | int ocfs2_journal_access_di(handle_t *handle, struct ocfs2_caching_info *ci, | |
711 | struct buffer_head *bh, int type) | |
712 | { | |
713 | return __ocfs2_journal_access(handle, ci, bh, &di_triggers, type); | |
714 | } | |
715 | ||
716 | int ocfs2_journal_access_eb(handle_t *handle, struct ocfs2_caching_info *ci, | |
717 | struct buffer_head *bh, int type) | |
718 | { | |
719 | return __ocfs2_journal_access(handle, ci, bh, &eb_triggers, type); | |
720 | } | |
721 | ||
722 | int ocfs2_journal_access_rb(handle_t *handle, struct ocfs2_caching_info *ci, | |
723 | struct buffer_head *bh, int type) | |
724 | { | |
725 | return __ocfs2_journal_access(handle, ci, bh, &rb_triggers, | |
726 | type); | |
727 | } | |
728 | ||
729 | int ocfs2_journal_access_gd(handle_t *handle, struct ocfs2_caching_info *ci, | |
730 | struct buffer_head *bh, int type) | |
731 | { | |
732 | return __ocfs2_journal_access(handle, ci, bh, &gd_triggers, type); | |
733 | } | |
734 | ||
735 | int ocfs2_journal_access_db(handle_t *handle, struct ocfs2_caching_info *ci, | |
736 | struct buffer_head *bh, int type) | |
737 | { | |
738 | return __ocfs2_journal_access(handle, ci, bh, &db_triggers, type); | |
739 | } | |
740 | ||
741 | int ocfs2_journal_access_xb(handle_t *handle, struct ocfs2_caching_info *ci, | |
742 | struct buffer_head *bh, int type) | |
743 | { | |
744 | return __ocfs2_journal_access(handle, ci, bh, &xb_triggers, type); | |
745 | } | |
746 | ||
747 | int ocfs2_journal_access_dq(handle_t *handle, struct ocfs2_caching_info *ci, | |
748 | struct buffer_head *bh, int type) | |
749 | { | |
750 | return __ocfs2_journal_access(handle, ci, bh, &dq_triggers, type); | |
751 | } | |
752 | ||
753 | int ocfs2_journal_access_dr(handle_t *handle, struct ocfs2_caching_info *ci, | |
754 | struct buffer_head *bh, int type) | |
755 | { | |
756 | return __ocfs2_journal_access(handle, ci, bh, &dr_triggers, type); | |
757 | } | |
758 | ||
759 | int ocfs2_journal_access_dl(handle_t *handle, struct ocfs2_caching_info *ci, | |
760 | struct buffer_head *bh, int type) | |
761 | { | |
762 | return __ocfs2_journal_access(handle, ci, bh, &dl_triggers, type); | |
763 | } | |
764 | ||
765 | int ocfs2_journal_access(handle_t *handle, struct ocfs2_caching_info *ci, | |
766 | struct buffer_head *bh, int type) | |
767 | { | |
768 | return __ocfs2_journal_access(handle, ci, bh, NULL, type); | |
769 | } | |
770 | ||
771 | void ocfs2_journal_dirty(handle_t *handle, struct buffer_head *bh) | |
772 | { | |
773 | int status; | |
774 | ||
775 | trace_ocfs2_journal_dirty((unsigned long long)bh->b_blocknr); | |
776 | ||
777 | status = jbd2_journal_dirty_metadata(handle, bh); | |
778 | if (status) { | |
779 | mlog_errno(status); | |
780 | if (!is_handle_aborted(handle)) { | |
781 | journal_t *journal = handle->h_transaction->t_journal; | |
782 | struct super_block *sb = bh->b_bdev->bd_super; | |
783 | ||
784 | mlog(ML_ERROR, "jbd2_journal_dirty_metadata failed. " | |
785 | "Aborting transaction and journal.\n"); | |
786 | handle->h_err = status; | |
787 | jbd2_journal_abort_handle(handle); | |
788 | jbd2_journal_abort(journal, status); | |
789 | ocfs2_abort(sb, "Journal already aborted.\n"); | |
790 | } | |
791 | } | |
792 | } | |
793 | ||
794 | #define OCFS2_DEFAULT_COMMIT_INTERVAL (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE) | |
795 | ||
796 | void ocfs2_set_journal_params(struct ocfs2_super *osb) | |
797 | { | |
798 | journal_t *journal = osb->journal->j_journal; | |
799 | unsigned long commit_interval = OCFS2_DEFAULT_COMMIT_INTERVAL; | |
800 | ||
801 | if (osb->osb_commit_interval) | |
802 | commit_interval = osb->osb_commit_interval; | |
803 | ||
804 | write_lock(&journal->j_state_lock); | |
805 | journal->j_commit_interval = commit_interval; | |
806 | if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER) | |
807 | journal->j_flags |= JBD2_BARRIER; | |
808 | else | |
809 | journal->j_flags &= ~JBD2_BARRIER; | |
810 | write_unlock(&journal->j_state_lock); | |
811 | } | |
812 | ||
813 | int ocfs2_journal_init(struct ocfs2_journal *journal, int *dirty) | |
814 | { | |
815 | int status = -1; | |
816 | struct inode *inode = NULL; /* the journal inode */ | |
817 | journal_t *j_journal = NULL; | |
818 | struct ocfs2_dinode *di = NULL; | |
819 | struct buffer_head *bh = NULL; | |
820 | struct ocfs2_super *osb; | |
821 | int inode_lock = 0; | |
822 | ||
823 | BUG_ON(!journal); | |
824 | ||
825 | osb = journal->j_osb; | |
826 | ||
827 | /* already have the inode for our journal */ | |
828 | inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, | |
829 | osb->slot_num); | |
830 | if (inode == NULL) { | |
831 | status = -EACCES; | |
832 | mlog_errno(status); | |
833 | goto done; | |
834 | } | |
835 | if (is_bad_inode(inode)) { | |
836 | mlog(ML_ERROR, "access error (bad inode)\n"); | |
837 | iput(inode); | |
838 | inode = NULL; | |
839 | status = -EACCES; | |
840 | goto done; | |
841 | } | |
842 | ||
843 | SET_INODE_JOURNAL(inode); | |
844 | OCFS2_I(inode)->ip_open_count++; | |
845 | ||
846 | /* Skip recovery waits here - journal inode metadata never | |
847 | * changes in a live cluster so it can be considered an | |
848 | * exception to the rule. */ | |
849 | status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY); | |
850 | if (status < 0) { | |
851 | if (status != -ERESTARTSYS) | |
852 | mlog(ML_ERROR, "Could not get lock on journal!\n"); | |
853 | goto done; | |
854 | } | |
855 | ||
856 | inode_lock = 1; | |
857 | di = (struct ocfs2_dinode *)bh->b_data; | |
858 | ||
859 | if (i_size_read(inode) < OCFS2_MIN_JOURNAL_SIZE) { | |
860 | mlog(ML_ERROR, "Journal file size (%lld) is too small!\n", | |
861 | i_size_read(inode)); | |
862 | status = -EINVAL; | |
863 | goto done; | |
864 | } | |
865 | ||
866 | trace_ocfs2_journal_init(i_size_read(inode), | |
867 | (unsigned long long)inode->i_blocks, | |
868 | OCFS2_I(inode)->ip_clusters); | |
869 | ||
870 | /* call the kernels journal init function now */ | |
871 | j_journal = jbd2_journal_init_inode(inode); | |
872 | if (j_journal == NULL) { | |
873 | mlog(ML_ERROR, "Linux journal layer error\n"); | |
874 | status = -EINVAL; | |
875 | goto done; | |
876 | } | |
877 | ||
878 | trace_ocfs2_journal_init_maxlen(j_journal->j_maxlen); | |
879 | ||
880 | *dirty = (le32_to_cpu(di->id1.journal1.ij_flags) & | |
881 | OCFS2_JOURNAL_DIRTY_FL); | |
882 | ||
883 | journal->j_journal = j_journal; | |
884 | journal->j_inode = inode; | |
885 | journal->j_bh = bh; | |
886 | ||
887 | ocfs2_set_journal_params(osb); | |
888 | ||
889 | journal->j_state = OCFS2_JOURNAL_LOADED; | |
890 | ||
891 | status = 0; | |
892 | done: | |
893 | if (status < 0) { | |
894 | if (inode_lock) | |
895 | ocfs2_inode_unlock(inode, 1); | |
896 | brelse(bh); | |
897 | if (inode) { | |
898 | OCFS2_I(inode)->ip_open_count--; | |
899 | iput(inode); | |
900 | } | |
901 | } | |
902 | ||
903 | return status; | |
904 | } | |
905 | ||
906 | static void ocfs2_bump_recovery_generation(struct ocfs2_dinode *di) | |
907 | { | |
908 | le32_add_cpu(&(di->id1.journal1.ij_recovery_generation), 1); | |
909 | } | |
910 | ||
911 | static u32 ocfs2_get_recovery_generation(struct ocfs2_dinode *di) | |
912 | { | |
913 | return le32_to_cpu(di->id1.journal1.ij_recovery_generation); | |
914 | } | |
915 | ||
916 | static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb, | |
917 | int dirty, int replayed) | |
918 | { | |
919 | int status; | |
920 | unsigned int flags; | |
921 | struct ocfs2_journal *journal = osb->journal; | |
922 | struct buffer_head *bh = journal->j_bh; | |
923 | struct ocfs2_dinode *fe; | |
924 | ||
925 | fe = (struct ocfs2_dinode *)bh->b_data; | |
926 | ||
927 | /* The journal bh on the osb always comes from ocfs2_journal_init() | |
928 | * and was validated there inside ocfs2_inode_lock_full(). It's a | |
929 | * code bug if we mess it up. */ | |
930 | BUG_ON(!OCFS2_IS_VALID_DINODE(fe)); | |
931 | ||
932 | flags = le32_to_cpu(fe->id1.journal1.ij_flags); | |
933 | if (dirty) | |
934 | flags |= OCFS2_JOURNAL_DIRTY_FL; | |
935 | else | |
936 | flags &= ~OCFS2_JOURNAL_DIRTY_FL; | |
937 | fe->id1.journal1.ij_flags = cpu_to_le32(flags); | |
938 | ||
939 | if (replayed) | |
940 | ocfs2_bump_recovery_generation(fe); | |
941 | ||
942 | ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check); | |
943 | status = ocfs2_write_block(osb, bh, INODE_CACHE(journal->j_inode)); | |
944 | if (status < 0) | |
945 | mlog_errno(status); | |
946 | ||
947 | return status; | |
948 | } | |
949 | ||
950 | /* | |
951 | * If the journal has been kmalloc'd it needs to be freed after this | |
952 | * call. | |
953 | */ | |
954 | void ocfs2_journal_shutdown(struct ocfs2_super *osb) | |
955 | { | |
956 | struct ocfs2_journal *journal = NULL; | |
957 | int status = 0; | |
958 | struct inode *inode = NULL; | |
959 | int num_running_trans = 0; | |
960 | ||
961 | BUG_ON(!osb); | |
962 | ||
963 | journal = osb->journal; | |
964 | if (!journal) | |
965 | goto done; | |
966 | ||
967 | inode = journal->j_inode; | |
968 | ||
969 | if (journal->j_state != OCFS2_JOURNAL_LOADED) | |
970 | goto done; | |
971 | ||
972 | /* need to inc inode use count - jbd2_journal_destroy will iput. */ | |
973 | if (!igrab(inode)) | |
974 | BUG(); | |
975 | ||
976 | num_running_trans = atomic_read(&(osb->journal->j_num_trans)); | |
977 | trace_ocfs2_journal_shutdown(num_running_trans); | |
978 | ||
979 | /* Do a commit_cache here. It will flush our journal, *and* | |
980 | * release any locks that are still held. | |
981 | * set the SHUTDOWN flag and release the trans lock. | |
982 | * the commit thread will take the trans lock for us below. */ | |
983 | journal->j_state = OCFS2_JOURNAL_IN_SHUTDOWN; | |
984 | ||
985 | /* The OCFS2_JOURNAL_IN_SHUTDOWN will signal to commit_cache to not | |
986 | * drop the trans_lock (which we want to hold until we | |
987 | * completely destroy the journal. */ | |
988 | if (osb->commit_task) { | |
989 | /* Wait for the commit thread */ | |
990 | trace_ocfs2_journal_shutdown_wait(osb->commit_task); | |
991 | kthread_stop(osb->commit_task); | |
992 | osb->commit_task = NULL; | |
993 | } | |
994 | ||
995 | BUG_ON(atomic_read(&(osb->journal->j_num_trans)) != 0); | |
996 | ||
997 | if (ocfs2_mount_local(osb)) { | |
998 | jbd2_journal_lock_updates(journal->j_journal); | |
999 | status = jbd2_journal_flush(journal->j_journal); | |
1000 | jbd2_journal_unlock_updates(journal->j_journal); | |
1001 | if (status < 0) | |
1002 | mlog_errno(status); | |
1003 | } | |
1004 | ||
1005 | if (status == 0) { | |
1006 | /* | |
1007 | * Do not toggle if flush was unsuccessful otherwise | |
1008 | * will leave dirty metadata in a "clean" journal | |
1009 | */ | |
1010 | status = ocfs2_journal_toggle_dirty(osb, 0, 0); | |
1011 | if (status < 0) | |
1012 | mlog_errno(status); | |
1013 | } | |
1014 | ||
1015 | /* Shutdown the kernel journal system */ | |
1016 | jbd2_journal_destroy(journal->j_journal); | |
1017 | journal->j_journal = NULL; | |
1018 | ||
1019 | OCFS2_I(inode)->ip_open_count--; | |
1020 | ||
1021 | /* unlock our journal */ | |
1022 | ocfs2_inode_unlock(inode, 1); | |
1023 | ||
1024 | brelse(journal->j_bh); | |
1025 | journal->j_bh = NULL; | |
1026 | ||
1027 | journal->j_state = OCFS2_JOURNAL_FREE; | |
1028 | ||
1029 | // up_write(&journal->j_trans_barrier); | |
1030 | done: | |
1031 | if (inode) | |
1032 | iput(inode); | |
1033 | } | |
1034 | ||
1035 | static void ocfs2_clear_journal_error(struct super_block *sb, | |
1036 | journal_t *journal, | |
1037 | int slot) | |
1038 | { | |
1039 | int olderr; | |
1040 | ||
1041 | olderr = jbd2_journal_errno(journal); | |
1042 | if (olderr) { | |
1043 | mlog(ML_ERROR, "File system error %d recorded in " | |
1044 | "journal %u.\n", olderr, slot); | |
1045 | mlog(ML_ERROR, "File system on device %s needs checking.\n", | |
1046 | sb->s_id); | |
1047 | ||
1048 | jbd2_journal_ack_err(journal); | |
1049 | jbd2_journal_clear_err(journal); | |
1050 | } | |
1051 | } | |
1052 | ||
1053 | int ocfs2_journal_load(struct ocfs2_journal *journal, int local, int replayed) | |
1054 | { | |
1055 | int status = 0; | |
1056 | struct ocfs2_super *osb; | |
1057 | ||
1058 | BUG_ON(!journal); | |
1059 | ||
1060 | osb = journal->j_osb; | |
1061 | ||
1062 | status = jbd2_journal_load(journal->j_journal); | |
1063 | if (status < 0) { | |
1064 | mlog(ML_ERROR, "Failed to load journal!\n"); | |
1065 | goto done; | |
1066 | } | |
1067 | ||
1068 | ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num); | |
1069 | ||
1070 | status = ocfs2_journal_toggle_dirty(osb, 1, replayed); | |
1071 | if (status < 0) { | |
1072 | mlog_errno(status); | |
1073 | goto done; | |
1074 | } | |
1075 | ||
1076 | /* Launch the commit thread */ | |
1077 | if (!local) { | |
1078 | osb->commit_task = kthread_run(ocfs2_commit_thread, osb, | |
1079 | "ocfs2cmt"); | |
1080 | if (IS_ERR(osb->commit_task)) { | |
1081 | status = PTR_ERR(osb->commit_task); | |
1082 | osb->commit_task = NULL; | |
1083 | mlog(ML_ERROR, "unable to launch ocfs2commit thread, " | |
1084 | "error=%d", status); | |
1085 | goto done; | |
1086 | } | |
1087 | } else | |
1088 | osb->commit_task = NULL; | |
1089 | ||
1090 | done: | |
1091 | return status; | |
1092 | } | |
1093 | ||
1094 | ||
1095 | /* 'full' flag tells us whether we clear out all blocks or if we just | |
1096 | * mark the journal clean */ | |
1097 | int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full) | |
1098 | { | |
1099 | int status; | |
1100 | ||
1101 | BUG_ON(!journal); | |
1102 | ||
1103 | status = jbd2_journal_wipe(journal->j_journal, full); | |
1104 | if (status < 0) { | |
1105 | mlog_errno(status); | |
1106 | goto bail; | |
1107 | } | |
1108 | ||
1109 | status = ocfs2_journal_toggle_dirty(journal->j_osb, 0, 0); | |
1110 | if (status < 0) | |
1111 | mlog_errno(status); | |
1112 | ||
1113 | bail: | |
1114 | return status; | |
1115 | } | |
1116 | ||
1117 | static int ocfs2_recovery_completed(struct ocfs2_super *osb) | |
1118 | { | |
1119 | int empty; | |
1120 | struct ocfs2_recovery_map *rm = osb->recovery_map; | |
1121 | ||
1122 | spin_lock(&osb->osb_lock); | |
1123 | empty = (rm->rm_used == 0); | |
1124 | spin_unlock(&osb->osb_lock); | |
1125 | ||
1126 | return empty; | |
1127 | } | |
1128 | ||
1129 | void ocfs2_wait_for_recovery(struct ocfs2_super *osb) | |
1130 | { | |
1131 | wait_event(osb->recovery_event, ocfs2_recovery_completed(osb)); | |
1132 | } | |
1133 | ||
1134 | /* | |
1135 | * JBD Might read a cached version of another nodes journal file. We | |
1136 | * don't want this as this file changes often and we get no | |
1137 | * notification on those changes. The only way to be sure that we've | |
1138 | * got the most up to date version of those blocks then is to force | |
1139 | * read them off disk. Just searching through the buffer cache won't | |
1140 | * work as there may be pages backing this file which are still marked | |
1141 | * up to date. We know things can't change on this file underneath us | |
1142 | * as we have the lock by now :) | |
1143 | */ | |
1144 | static int ocfs2_force_read_journal(struct inode *inode) | |
1145 | { | |
1146 | int status = 0; | |
1147 | int i; | |
1148 | u64 v_blkno, p_blkno, p_blocks, num_blocks; | |
1149 | #define CONCURRENT_JOURNAL_FILL 32ULL | |
1150 | struct buffer_head *bhs[CONCURRENT_JOURNAL_FILL]; | |
1151 | ||
1152 | memset(bhs, 0, sizeof(struct buffer_head *) * CONCURRENT_JOURNAL_FILL); | |
1153 | ||
1154 | num_blocks = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode)); | |
1155 | v_blkno = 0; | |
1156 | while (v_blkno < num_blocks) { | |
1157 | status = ocfs2_extent_map_get_blocks(inode, v_blkno, | |
1158 | &p_blkno, &p_blocks, NULL); | |
1159 | if (status < 0) { | |
1160 | mlog_errno(status); | |
1161 | goto bail; | |
1162 | } | |
1163 | ||
1164 | if (p_blocks > CONCURRENT_JOURNAL_FILL) | |
1165 | p_blocks = CONCURRENT_JOURNAL_FILL; | |
1166 | ||
1167 | /* We are reading journal data which should not | |
1168 | * be put in the uptodate cache */ | |
1169 | status = ocfs2_read_blocks_sync(OCFS2_SB(inode->i_sb), | |
1170 | p_blkno, p_blocks, bhs); | |
1171 | if (status < 0) { | |
1172 | mlog_errno(status); | |
1173 | goto bail; | |
1174 | } | |
1175 | ||
1176 | for(i = 0; i < p_blocks; i++) { | |
1177 | brelse(bhs[i]); | |
1178 | bhs[i] = NULL; | |
1179 | } | |
1180 | ||
1181 | v_blkno += p_blocks; | |
1182 | } | |
1183 | ||
1184 | bail: | |
1185 | for(i = 0; i < CONCURRENT_JOURNAL_FILL; i++) | |
1186 | brelse(bhs[i]); | |
1187 | return status; | |
1188 | } | |
1189 | ||
1190 | struct ocfs2_la_recovery_item { | |
1191 | struct list_head lri_list; | |
1192 | int lri_slot; | |
1193 | struct ocfs2_dinode *lri_la_dinode; | |
1194 | struct ocfs2_dinode *lri_tl_dinode; | |
1195 | struct ocfs2_quota_recovery *lri_qrec; | |
1196 | enum ocfs2_orphan_reco_type lri_orphan_reco_type; | |
1197 | }; | |
1198 | ||
1199 | /* Does the second half of the recovery process. By this point, the | |
1200 | * node is marked clean and can actually be considered recovered, | |
1201 | * hence it's no longer in the recovery map, but there's still some | |
1202 | * cleanup we can do which shouldn't happen within the recovery thread | |
1203 | * as locking in that context becomes very difficult if we are to take | |
1204 | * recovering nodes into account. | |
1205 | * | |
1206 | * NOTE: This function can and will sleep on recovery of other nodes | |
1207 | * during cluster locking, just like any other ocfs2 process. | |
1208 | */ | |
1209 | void ocfs2_complete_recovery(struct work_struct *work) | |
1210 | { | |
1211 | int ret = 0; | |
1212 | struct ocfs2_journal *journal = | |
1213 | container_of(work, struct ocfs2_journal, j_recovery_work); | |
1214 | struct ocfs2_super *osb = journal->j_osb; | |
1215 | struct ocfs2_dinode *la_dinode, *tl_dinode; | |
1216 | struct ocfs2_la_recovery_item *item, *n; | |
1217 | struct ocfs2_quota_recovery *qrec; | |
1218 | enum ocfs2_orphan_reco_type orphan_reco_type; | |
1219 | LIST_HEAD(tmp_la_list); | |
1220 | ||
1221 | trace_ocfs2_complete_recovery( | |
1222 | (unsigned long long)OCFS2_I(journal->j_inode)->ip_blkno); | |
1223 | ||
1224 | spin_lock(&journal->j_lock); | |
1225 | list_splice_init(&journal->j_la_cleanups, &tmp_la_list); | |
1226 | spin_unlock(&journal->j_lock); | |
1227 | ||
1228 | list_for_each_entry_safe(item, n, &tmp_la_list, lri_list) { | |
1229 | list_del_init(&item->lri_list); | |
1230 | ||
1231 | ocfs2_wait_on_quotas(osb); | |
1232 | ||
1233 | la_dinode = item->lri_la_dinode; | |
1234 | tl_dinode = item->lri_tl_dinode; | |
1235 | qrec = item->lri_qrec; | |
1236 | orphan_reco_type = item->lri_orphan_reco_type; | |
1237 | ||
1238 | trace_ocfs2_complete_recovery_slot(item->lri_slot, | |
1239 | la_dinode ? le64_to_cpu(la_dinode->i_blkno) : 0, | |
1240 | tl_dinode ? le64_to_cpu(tl_dinode->i_blkno) : 0, | |
1241 | qrec); | |
1242 | ||
1243 | if (la_dinode) { | |
1244 | ret = ocfs2_complete_local_alloc_recovery(osb, | |
1245 | la_dinode); | |
1246 | if (ret < 0) | |
1247 | mlog_errno(ret); | |
1248 | ||
1249 | kfree(la_dinode); | |
1250 | } | |
1251 | ||
1252 | if (tl_dinode) { | |
1253 | ret = ocfs2_complete_truncate_log_recovery(osb, | |
1254 | tl_dinode); | |
1255 | if (ret < 0) | |
1256 | mlog_errno(ret); | |
1257 | ||
1258 | kfree(tl_dinode); | |
1259 | } | |
1260 | ||
1261 | ret = ocfs2_recover_orphans(osb, item->lri_slot, | |
1262 | orphan_reco_type); | |
1263 | if (ret < 0) | |
1264 | mlog_errno(ret); | |
1265 | ||
1266 | if (qrec) { | |
1267 | ret = ocfs2_finish_quota_recovery(osb, qrec, | |
1268 | item->lri_slot); | |
1269 | if (ret < 0) | |
1270 | mlog_errno(ret); | |
1271 | /* Recovery info is already freed now */ | |
1272 | } | |
1273 | ||
1274 | kfree(item); | |
1275 | } | |
1276 | ||
1277 | trace_ocfs2_complete_recovery_end(ret); | |
1278 | } | |
1279 | ||
1280 | /* NOTE: This function always eats your references to la_dinode and | |
1281 | * tl_dinode, either manually on error, or by passing them to | |
1282 | * ocfs2_complete_recovery */ | |
1283 | static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal, | |
1284 | int slot_num, | |
1285 | struct ocfs2_dinode *la_dinode, | |
1286 | struct ocfs2_dinode *tl_dinode, | |
1287 | struct ocfs2_quota_recovery *qrec, | |
1288 | enum ocfs2_orphan_reco_type orphan_reco_type) | |
1289 | { | |
1290 | struct ocfs2_la_recovery_item *item; | |
1291 | ||
1292 | item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS); | |
1293 | if (!item) { | |
1294 | /* Though we wish to avoid it, we are in fact safe in | |
1295 | * skipping local alloc cleanup as fsck.ocfs2 is more | |
1296 | * than capable of reclaiming unused space. */ | |
1297 | kfree(la_dinode); | |
1298 | kfree(tl_dinode); | |
1299 | ||
1300 | if (qrec) | |
1301 | ocfs2_free_quota_recovery(qrec); | |
1302 | ||
1303 | mlog_errno(-ENOMEM); | |
1304 | return; | |
1305 | } | |
1306 | ||
1307 | INIT_LIST_HEAD(&item->lri_list); | |
1308 | item->lri_la_dinode = la_dinode; | |
1309 | item->lri_slot = slot_num; | |
1310 | item->lri_tl_dinode = tl_dinode; | |
1311 | item->lri_qrec = qrec; | |
1312 | item->lri_orphan_reco_type = orphan_reco_type; | |
1313 | ||
1314 | spin_lock(&journal->j_lock); | |
1315 | list_add_tail(&item->lri_list, &journal->j_la_cleanups); | |
1316 | queue_work(ocfs2_wq, &journal->j_recovery_work); | |
1317 | spin_unlock(&journal->j_lock); | |
1318 | } | |
1319 | ||
1320 | /* Called by the mount code to queue recovery the last part of | |
1321 | * recovery for it's own and offline slot(s). */ | |
1322 | void ocfs2_complete_mount_recovery(struct ocfs2_super *osb) | |
1323 | { | |
1324 | struct ocfs2_journal *journal = osb->journal; | |
1325 | ||
1326 | if (ocfs2_is_hard_readonly(osb)) | |
1327 | return; | |
1328 | ||
1329 | /* No need to queue up our truncate_log as regular cleanup will catch | |
1330 | * that */ | |
1331 | ocfs2_queue_recovery_completion(journal, osb->slot_num, | |
1332 | osb->local_alloc_copy, NULL, NULL, | |
1333 | ORPHAN_NEED_TRUNCATE); | |
1334 | ocfs2_schedule_truncate_log_flush(osb, 0); | |
1335 | ||
1336 | osb->local_alloc_copy = NULL; | |
1337 | osb->dirty = 0; | |
1338 | ||
1339 | /* queue to recover orphan slots for all offline slots */ | |
1340 | ocfs2_replay_map_set_state(osb, REPLAY_NEEDED); | |
1341 | ocfs2_queue_replay_slots(osb, ORPHAN_NEED_TRUNCATE); | |
1342 | ocfs2_free_replay_slots(osb); | |
1343 | } | |
1344 | ||
1345 | void ocfs2_complete_quota_recovery(struct ocfs2_super *osb) | |
1346 | { | |
1347 | if (osb->quota_rec) { | |
1348 | ocfs2_queue_recovery_completion(osb->journal, | |
1349 | osb->slot_num, | |
1350 | NULL, | |
1351 | NULL, | |
1352 | osb->quota_rec, | |
1353 | ORPHAN_NEED_TRUNCATE); | |
1354 | osb->quota_rec = NULL; | |
1355 | } | |
1356 | } | |
1357 | ||
1358 | static int __ocfs2_recovery_thread(void *arg) | |
1359 | { | |
1360 | int status, node_num, slot_num; | |
1361 | struct ocfs2_super *osb = arg; | |
1362 | struct ocfs2_recovery_map *rm = osb->recovery_map; | |
1363 | int *rm_quota = NULL; | |
1364 | int rm_quota_used = 0, i; | |
1365 | struct ocfs2_quota_recovery *qrec; | |
1366 | ||
1367 | status = ocfs2_wait_on_mount(osb); | |
1368 | if (status < 0) { | |
1369 | goto bail; | |
1370 | } | |
1371 | ||
1372 | rm_quota = kzalloc(osb->max_slots * sizeof(int), GFP_NOFS); | |
1373 | if (!rm_quota) { | |
1374 | status = -ENOMEM; | |
1375 | goto bail; | |
1376 | } | |
1377 | restart: | |
1378 | status = ocfs2_super_lock(osb, 1); | |
1379 | if (status < 0) { | |
1380 | mlog_errno(status); | |
1381 | goto bail; | |
1382 | } | |
1383 | ||
1384 | status = ocfs2_compute_replay_slots(osb); | |
1385 | if (status < 0) | |
1386 | mlog_errno(status); | |
1387 | ||
1388 | /* queue recovery for our own slot */ | |
1389 | ocfs2_queue_recovery_completion(osb->journal, osb->slot_num, NULL, | |
1390 | NULL, NULL, ORPHAN_NO_NEED_TRUNCATE); | |
1391 | ||
1392 | spin_lock(&osb->osb_lock); | |
1393 | while (rm->rm_used) { | |
1394 | /* It's always safe to remove entry zero, as we won't | |
1395 | * clear it until ocfs2_recover_node() has succeeded. */ | |
1396 | node_num = rm->rm_entries[0]; | |
1397 | spin_unlock(&osb->osb_lock); | |
1398 | slot_num = ocfs2_node_num_to_slot(osb, node_num); | |
1399 | trace_ocfs2_recovery_thread_node(node_num, slot_num); | |
1400 | if (slot_num == -ENOENT) { | |
1401 | status = 0; | |
1402 | goto skip_recovery; | |
1403 | } | |
1404 | ||
1405 | /* It is a bit subtle with quota recovery. We cannot do it | |
1406 | * immediately because we have to obtain cluster locks from | |
1407 | * quota files and we also don't want to just skip it because | |
1408 | * then quota usage would be out of sync until some node takes | |
1409 | * the slot. So we remember which nodes need quota recovery | |
1410 | * and when everything else is done, we recover quotas. */ | |
1411 | for (i = 0; i < rm_quota_used && rm_quota[i] != slot_num; i++); | |
1412 | if (i == rm_quota_used) | |
1413 | rm_quota[rm_quota_used++] = slot_num; | |
1414 | ||
1415 | status = ocfs2_recover_node(osb, node_num, slot_num); | |
1416 | skip_recovery: | |
1417 | if (!status) { | |
1418 | ocfs2_recovery_map_clear(osb, node_num); | |
1419 | } else { | |
1420 | mlog(ML_ERROR, | |
1421 | "Error %d recovering node %d on device (%u,%u)!\n", | |
1422 | status, node_num, | |
1423 | MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev)); | |
1424 | mlog(ML_ERROR, "Volume requires unmount.\n"); | |
1425 | } | |
1426 | ||
1427 | spin_lock(&osb->osb_lock); | |
1428 | } | |
1429 | spin_unlock(&osb->osb_lock); | |
1430 | trace_ocfs2_recovery_thread_end(status); | |
1431 | ||
1432 | /* Refresh all journal recovery generations from disk */ | |
1433 | status = ocfs2_check_journals_nolocks(osb); | |
1434 | status = (status == -EROFS) ? 0 : status; | |
1435 | if (status < 0) | |
1436 | mlog_errno(status); | |
1437 | ||
1438 | /* Now it is right time to recover quotas... We have to do this under | |
1439 | * superblock lock so that no one can start using the slot (and crash) | |
1440 | * before we recover it */ | |
1441 | for (i = 0; i < rm_quota_used; i++) { | |
1442 | qrec = ocfs2_begin_quota_recovery(osb, rm_quota[i]); | |
1443 | if (IS_ERR(qrec)) { | |
1444 | status = PTR_ERR(qrec); | |
1445 | mlog_errno(status); | |
1446 | continue; | |
1447 | } | |
1448 | ocfs2_queue_recovery_completion(osb->journal, rm_quota[i], | |
1449 | NULL, NULL, qrec, | |
1450 | ORPHAN_NEED_TRUNCATE); | |
1451 | } | |
1452 | ||
1453 | ocfs2_super_unlock(osb, 1); | |
1454 | ||
1455 | /* queue recovery for offline slots */ | |
1456 | ocfs2_queue_replay_slots(osb, ORPHAN_NEED_TRUNCATE); | |
1457 | ||
1458 | bail: | |
1459 | mutex_lock(&osb->recovery_lock); | |
1460 | if (!status && !ocfs2_recovery_completed(osb)) { | |
1461 | mutex_unlock(&osb->recovery_lock); | |
1462 | goto restart; | |
1463 | } | |
1464 | ||
1465 | ocfs2_free_replay_slots(osb); | |
1466 | osb->recovery_thread_task = NULL; | |
1467 | mb(); /* sync with ocfs2_recovery_thread_running */ | |
1468 | wake_up(&osb->recovery_event); | |
1469 | ||
1470 | mutex_unlock(&osb->recovery_lock); | |
1471 | ||
1472 | kfree(rm_quota); | |
1473 | ||
1474 | /* no one is callint kthread_stop() for us so the kthread() api | |
1475 | * requires that we call do_exit(). And it isn't exported, but | |
1476 | * complete_and_exit() seems to be a minimal wrapper around it. */ | |
1477 | complete_and_exit(NULL, status); | |
1478 | } | |
1479 | ||
1480 | void ocfs2_recovery_thread(struct ocfs2_super *osb, int node_num) | |
1481 | { | |
1482 | mutex_lock(&osb->recovery_lock); | |
1483 | ||
1484 | trace_ocfs2_recovery_thread(node_num, osb->node_num, | |
1485 | osb->disable_recovery, osb->recovery_thread_task, | |
1486 | osb->disable_recovery ? | |
1487 | -1 : ocfs2_recovery_map_set(osb, node_num)); | |
1488 | ||
1489 | if (osb->disable_recovery) | |
1490 | goto out; | |
1491 | ||
1492 | if (osb->recovery_thread_task) | |
1493 | goto out; | |
1494 | ||
1495 | osb->recovery_thread_task = kthread_run(__ocfs2_recovery_thread, osb, | |
1496 | "ocfs2rec"); | |
1497 | if (IS_ERR(osb->recovery_thread_task)) { | |
1498 | mlog_errno((int)PTR_ERR(osb->recovery_thread_task)); | |
1499 | osb->recovery_thread_task = NULL; | |
1500 | } | |
1501 | ||
1502 | out: | |
1503 | mutex_unlock(&osb->recovery_lock); | |
1504 | wake_up(&osb->recovery_event); | |
1505 | } | |
1506 | ||
1507 | static int ocfs2_read_journal_inode(struct ocfs2_super *osb, | |
1508 | int slot_num, | |
1509 | struct buffer_head **bh, | |
1510 | struct inode **ret_inode) | |
1511 | { | |
1512 | int status = -EACCES; | |
1513 | struct inode *inode = NULL; | |
1514 | ||
1515 | BUG_ON(slot_num >= osb->max_slots); | |
1516 | ||
1517 | inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, | |
1518 | slot_num); | |
1519 | if (!inode || is_bad_inode(inode)) { | |
1520 | mlog_errno(status); | |
1521 | goto bail; | |
1522 | } | |
1523 | SET_INODE_JOURNAL(inode); | |
1524 | ||
1525 | status = ocfs2_read_inode_block_full(inode, bh, OCFS2_BH_IGNORE_CACHE); | |
1526 | if (status < 0) { | |
1527 | mlog_errno(status); | |
1528 | goto bail; | |
1529 | } | |
1530 | ||
1531 | status = 0; | |
1532 | ||
1533 | bail: | |
1534 | if (inode) { | |
1535 | if (status || !ret_inode) | |
1536 | iput(inode); | |
1537 | else | |
1538 | *ret_inode = inode; | |
1539 | } | |
1540 | return status; | |
1541 | } | |
1542 | ||
1543 | /* Does the actual journal replay and marks the journal inode as | |
1544 | * clean. Will only replay if the journal inode is marked dirty. */ | |
1545 | static int ocfs2_replay_journal(struct ocfs2_super *osb, | |
1546 | int node_num, | |
1547 | int slot_num) | |
1548 | { | |
1549 | int status; | |
1550 | int got_lock = 0; | |
1551 | unsigned int flags; | |
1552 | struct inode *inode = NULL; | |
1553 | struct ocfs2_dinode *fe; | |
1554 | journal_t *journal = NULL; | |
1555 | struct buffer_head *bh = NULL; | |
1556 | u32 slot_reco_gen; | |
1557 | ||
1558 | status = ocfs2_read_journal_inode(osb, slot_num, &bh, &inode); | |
1559 | if (status) { | |
1560 | mlog_errno(status); | |
1561 | goto done; | |
1562 | } | |
1563 | ||
1564 | fe = (struct ocfs2_dinode *)bh->b_data; | |
1565 | slot_reco_gen = ocfs2_get_recovery_generation(fe); | |
1566 | brelse(bh); | |
1567 | bh = NULL; | |
1568 | ||
1569 | /* | |
1570 | * As the fs recovery is asynchronous, there is a small chance that | |
1571 | * another node mounted (and recovered) the slot before the recovery | |
1572 | * thread could get the lock. To handle that, we dirty read the journal | |
1573 | * inode for that slot to get the recovery generation. If it is | |
1574 | * different than what we expected, the slot has been recovered. | |
1575 | * If not, it needs recovery. | |
1576 | */ | |
1577 | if (osb->slot_recovery_generations[slot_num] != slot_reco_gen) { | |
1578 | trace_ocfs2_replay_journal_recovered(slot_num, | |
1579 | osb->slot_recovery_generations[slot_num], slot_reco_gen); | |
1580 | osb->slot_recovery_generations[slot_num] = slot_reco_gen; | |
1581 | status = -EBUSY; | |
1582 | goto done; | |
1583 | } | |
1584 | ||
1585 | /* Continue with recovery as the journal has not yet been recovered */ | |
1586 | ||
1587 | status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY); | |
1588 | if (status < 0) { | |
1589 | trace_ocfs2_replay_journal_lock_err(status); | |
1590 | if (status != -ERESTARTSYS) | |
1591 | mlog(ML_ERROR, "Could not lock journal!\n"); | |
1592 | goto done; | |
1593 | } | |
1594 | got_lock = 1; | |
1595 | ||
1596 | fe = (struct ocfs2_dinode *) bh->b_data; | |
1597 | ||
1598 | flags = le32_to_cpu(fe->id1.journal1.ij_flags); | |
1599 | slot_reco_gen = ocfs2_get_recovery_generation(fe); | |
1600 | ||
1601 | if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) { | |
1602 | trace_ocfs2_replay_journal_skip(node_num); | |
1603 | /* Refresh recovery generation for the slot */ | |
1604 | osb->slot_recovery_generations[slot_num] = slot_reco_gen; | |
1605 | goto done; | |
1606 | } | |
1607 | ||
1608 | /* we need to run complete recovery for offline orphan slots */ | |
1609 | ocfs2_replay_map_set_state(osb, REPLAY_NEEDED); | |
1610 | ||
1611 | printk(KERN_NOTICE "ocfs2: Begin replay journal (node %d, slot %d) on "\ | |
1612 | "device (%u,%u)\n", node_num, slot_num, MAJOR(osb->sb->s_dev), | |
1613 | MINOR(osb->sb->s_dev)); | |
1614 | ||
1615 | OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters); | |
1616 | ||
1617 | status = ocfs2_force_read_journal(inode); | |
1618 | if (status < 0) { | |
1619 | mlog_errno(status); | |
1620 | goto done; | |
1621 | } | |
1622 | ||
1623 | journal = jbd2_journal_init_inode(inode); | |
1624 | if (journal == NULL) { | |
1625 | mlog(ML_ERROR, "Linux journal layer error\n"); | |
1626 | status = -EIO; | |
1627 | goto done; | |
1628 | } | |
1629 | ||
1630 | status = jbd2_journal_load(journal); | |
1631 | if (status < 0) { | |
1632 | mlog_errno(status); | |
1633 | if (!igrab(inode)) | |
1634 | BUG(); | |
1635 | jbd2_journal_destroy(journal); | |
1636 | goto done; | |
1637 | } | |
1638 | ||
1639 | ocfs2_clear_journal_error(osb->sb, journal, slot_num); | |
1640 | ||
1641 | /* wipe the journal */ | |
1642 | jbd2_journal_lock_updates(journal); | |
1643 | status = jbd2_journal_flush(journal); | |
1644 | jbd2_journal_unlock_updates(journal); | |
1645 | if (status < 0) | |
1646 | mlog_errno(status); | |
1647 | ||
1648 | /* This will mark the node clean */ | |
1649 | flags = le32_to_cpu(fe->id1.journal1.ij_flags); | |
1650 | flags &= ~OCFS2_JOURNAL_DIRTY_FL; | |
1651 | fe->id1.journal1.ij_flags = cpu_to_le32(flags); | |
1652 | ||
1653 | /* Increment recovery generation to indicate successful recovery */ | |
1654 | ocfs2_bump_recovery_generation(fe); | |
1655 | osb->slot_recovery_generations[slot_num] = | |
1656 | ocfs2_get_recovery_generation(fe); | |
1657 | ||
1658 | ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check); | |
1659 | status = ocfs2_write_block(osb, bh, INODE_CACHE(inode)); | |
1660 | if (status < 0) | |
1661 | mlog_errno(status); | |
1662 | ||
1663 | if (!igrab(inode)) | |
1664 | BUG(); | |
1665 | ||
1666 | jbd2_journal_destroy(journal); | |
1667 | ||
1668 | printk(KERN_NOTICE "ocfs2: End replay journal (node %d, slot %d) on "\ | |
1669 | "device (%u,%u)\n", node_num, slot_num, MAJOR(osb->sb->s_dev), | |
1670 | MINOR(osb->sb->s_dev)); | |
1671 | done: | |
1672 | /* drop the lock on this nodes journal */ | |
1673 | if (got_lock) | |
1674 | ocfs2_inode_unlock(inode, 1); | |
1675 | ||
1676 | if (inode) | |
1677 | iput(inode); | |
1678 | ||
1679 | brelse(bh); | |
1680 | ||
1681 | return status; | |
1682 | } | |
1683 | ||
1684 | /* | |
1685 | * Do the most important parts of node recovery: | |
1686 | * - Replay it's journal | |
1687 | * - Stamp a clean local allocator file | |
1688 | * - Stamp a clean truncate log | |
1689 | * - Mark the node clean | |
1690 | * | |
1691 | * If this function completes without error, a node in OCFS2 can be | |
1692 | * said to have been safely recovered. As a result, failure during the | |
1693 | * second part of a nodes recovery process (local alloc recovery) is | |
1694 | * far less concerning. | |
1695 | */ | |
1696 | static int ocfs2_recover_node(struct ocfs2_super *osb, | |
1697 | int node_num, int slot_num) | |
1698 | { | |
1699 | int status = 0; | |
1700 | struct ocfs2_dinode *la_copy = NULL; | |
1701 | struct ocfs2_dinode *tl_copy = NULL; | |
1702 | ||
1703 | trace_ocfs2_recover_node(node_num, slot_num, osb->node_num); | |
1704 | ||
1705 | /* Should not ever be called to recover ourselves -- in that | |
1706 | * case we should've called ocfs2_journal_load instead. */ | |
1707 | BUG_ON(osb->node_num == node_num); | |
1708 | ||
1709 | status = ocfs2_replay_journal(osb, node_num, slot_num); | |
1710 | if (status < 0) { | |
1711 | if (status == -EBUSY) { | |
1712 | trace_ocfs2_recover_node_skip(slot_num, node_num); | |
1713 | status = 0; | |
1714 | goto done; | |
1715 | } | |
1716 | mlog_errno(status); | |
1717 | goto done; | |
1718 | } | |
1719 | ||
1720 | /* Stamp a clean local alloc file AFTER recovering the journal... */ | |
1721 | status = ocfs2_begin_local_alloc_recovery(osb, slot_num, &la_copy); | |
1722 | if (status < 0) { | |
1723 | mlog_errno(status); | |
1724 | goto done; | |
1725 | } | |
1726 | ||
1727 | /* An error from begin_truncate_log_recovery is not | |
1728 | * serious enough to warrant halting the rest of | |
1729 | * recovery. */ | |
1730 | status = ocfs2_begin_truncate_log_recovery(osb, slot_num, &tl_copy); | |
1731 | if (status < 0) | |
1732 | mlog_errno(status); | |
1733 | ||
1734 | /* Likewise, this would be a strange but ultimately not so | |
1735 | * harmful place to get an error... */ | |
1736 | status = ocfs2_clear_slot(osb, slot_num); | |
1737 | if (status < 0) | |
1738 | mlog_errno(status); | |
1739 | ||
1740 | /* This will kfree the memory pointed to by la_copy and tl_copy */ | |
1741 | ocfs2_queue_recovery_completion(osb->journal, slot_num, la_copy, | |
1742 | tl_copy, NULL, ORPHAN_NEED_TRUNCATE); | |
1743 | ||
1744 | status = 0; | |
1745 | done: | |
1746 | ||
1747 | return status; | |
1748 | } | |
1749 | ||
1750 | /* Test node liveness by trylocking his journal. If we get the lock, | |
1751 | * we drop it here. Return 0 if we got the lock, -EAGAIN if node is | |
1752 | * still alive (we couldn't get the lock) and < 0 on error. */ | |
1753 | static int ocfs2_trylock_journal(struct ocfs2_super *osb, | |
1754 | int slot_num) | |
1755 | { | |
1756 | int status, flags; | |
1757 | struct inode *inode = NULL; | |
1758 | ||
1759 | inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, | |
1760 | slot_num); | |
1761 | if (inode == NULL) { | |
1762 | mlog(ML_ERROR, "access error\n"); | |
1763 | status = -EACCES; | |
1764 | goto bail; | |
1765 | } | |
1766 | if (is_bad_inode(inode)) { | |
1767 | mlog(ML_ERROR, "access error (bad inode)\n"); | |
1768 | iput(inode); | |
1769 | inode = NULL; | |
1770 | status = -EACCES; | |
1771 | goto bail; | |
1772 | } | |
1773 | SET_INODE_JOURNAL(inode); | |
1774 | ||
1775 | flags = OCFS2_META_LOCK_RECOVERY | OCFS2_META_LOCK_NOQUEUE; | |
1776 | status = ocfs2_inode_lock_full(inode, NULL, 1, flags); | |
1777 | if (status < 0) { | |
1778 | if (status != -EAGAIN) | |
1779 | mlog_errno(status); | |
1780 | goto bail; | |
1781 | } | |
1782 | ||
1783 | ocfs2_inode_unlock(inode, 1); | |
1784 | bail: | |
1785 | if (inode) | |
1786 | iput(inode); | |
1787 | ||
1788 | return status; | |
1789 | } | |
1790 | ||
1791 | /* Call this underneath ocfs2_super_lock. It also assumes that the | |
1792 | * slot info struct has been updated from disk. */ | |
1793 | int ocfs2_mark_dead_nodes(struct ocfs2_super *osb) | |
1794 | { | |
1795 | unsigned int node_num; | |
1796 | int status, i; | |
1797 | u32 gen; | |
1798 | struct buffer_head *bh = NULL; | |
1799 | struct ocfs2_dinode *di; | |
1800 | ||
1801 | /* This is called with the super block cluster lock, so we | |
1802 | * know that the slot map can't change underneath us. */ | |
1803 | ||
1804 | for (i = 0; i < osb->max_slots; i++) { | |
1805 | /* Read journal inode to get the recovery generation */ | |
1806 | status = ocfs2_read_journal_inode(osb, i, &bh, NULL); | |
1807 | if (status) { | |
1808 | mlog_errno(status); | |
1809 | goto bail; | |
1810 | } | |
1811 | di = (struct ocfs2_dinode *)bh->b_data; | |
1812 | gen = ocfs2_get_recovery_generation(di); | |
1813 | brelse(bh); | |
1814 | bh = NULL; | |
1815 | ||
1816 | spin_lock(&osb->osb_lock); | |
1817 | osb->slot_recovery_generations[i] = gen; | |
1818 | ||
1819 | trace_ocfs2_mark_dead_nodes(i, | |
1820 | osb->slot_recovery_generations[i]); | |
1821 | ||
1822 | if (i == osb->slot_num) { | |
1823 | spin_unlock(&osb->osb_lock); | |
1824 | continue; | |
1825 | } | |
1826 | ||
1827 | status = ocfs2_slot_to_node_num_locked(osb, i, &node_num); | |
1828 | if (status == -ENOENT) { | |
1829 | spin_unlock(&osb->osb_lock); | |
1830 | continue; | |
1831 | } | |
1832 | ||
1833 | if (__ocfs2_recovery_map_test(osb, node_num)) { | |
1834 | spin_unlock(&osb->osb_lock); | |
1835 | continue; | |
1836 | } | |
1837 | spin_unlock(&osb->osb_lock); | |
1838 | ||
1839 | /* Ok, we have a slot occupied by another node which | |
1840 | * is not in the recovery map. We trylock his journal | |
1841 | * file here to test if he's alive. */ | |
1842 | status = ocfs2_trylock_journal(osb, i); | |
1843 | if (!status) { | |
1844 | /* Since we're called from mount, we know that | |
1845 | * the recovery thread can't race us on | |
1846 | * setting / checking the recovery bits. */ | |
1847 | ocfs2_recovery_thread(osb, node_num); | |
1848 | } else if ((status < 0) && (status != -EAGAIN)) { | |
1849 | mlog_errno(status); | |
1850 | goto bail; | |
1851 | } | |
1852 | } | |
1853 | ||
1854 | status = 0; | |
1855 | bail: | |
1856 | return status; | |
1857 | } | |
1858 | ||
1859 | /* | |
1860 | * Scan timer should get fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT. Add some | |
1861 | * randomness to the timeout to minimize multple nodes firing the timer at the | |
1862 | * same time. | |
1863 | */ | |
1864 | static inline unsigned long ocfs2_orphan_scan_timeout(void) | |
1865 | { | |
1866 | unsigned long time; | |
1867 | ||
1868 | get_random_bytes(&time, sizeof(time)); | |
1869 | time = ORPHAN_SCAN_SCHEDULE_TIMEOUT + (time % 5000); | |
1870 | return msecs_to_jiffies(time); | |
1871 | } | |
1872 | ||
1873 | /* | |
1874 | * ocfs2_queue_orphan_scan calls ocfs2_queue_recovery_completion for | |
1875 | * every slot, queuing a recovery of the slot on the ocfs2_wq thread. This | |
1876 | * is done to catch any orphans that are left over in orphan directories. | |
1877 | * | |
1878 | * It scans all slots, even ones that are in use. It does so to handle the | |
1879 | * case described below: | |
1880 | * | |
1881 | * Node 1 has an inode it was using. The dentry went away due to memory | |
1882 | * pressure. Node 1 closes the inode, but it's on the free list. The node | |
1883 | * has the open lock. | |
1884 | * Node 2 unlinks the inode. It grabs the dentry lock to notify others, | |
1885 | * but node 1 has no dentry and doesn't get the message. It trylocks the | |
1886 | * open lock, sees that another node has a PR, and does nothing. | |
1887 | * Later node 2 runs its orphan dir. It igets the inode, trylocks the | |
1888 | * open lock, sees the PR still, and does nothing. | |
1889 | * Basically, we have to trigger an orphan iput on node 1. The only way | |
1890 | * for this to happen is if node 1 runs node 2's orphan dir. | |
1891 | * | |
1892 | * ocfs2_queue_orphan_scan gets called every ORPHAN_SCAN_SCHEDULE_TIMEOUT | |
1893 | * seconds. It gets an EX lock on os_lockres and checks sequence number | |
1894 | * stored in LVB. If the sequence number has changed, it means some other | |
1895 | * node has done the scan. This node skips the scan and tracks the | |
1896 | * sequence number. If the sequence number didn't change, it means a scan | |
1897 | * hasn't happened. The node queues a scan and increments the | |
1898 | * sequence number in the LVB. | |
1899 | */ | |
1900 | void ocfs2_queue_orphan_scan(struct ocfs2_super *osb) | |
1901 | { | |
1902 | struct ocfs2_orphan_scan *os; | |
1903 | int status, i; | |
1904 | u32 seqno = 0; | |
1905 | ||
1906 | os = &osb->osb_orphan_scan; | |
1907 | ||
1908 | if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE) | |
1909 | goto out; | |
1910 | ||
1911 | trace_ocfs2_queue_orphan_scan_begin(os->os_count, os->os_seqno, | |
1912 | atomic_read(&os->os_state)); | |
1913 | ||
1914 | status = ocfs2_orphan_scan_lock(osb, &seqno); | |
1915 | if (status < 0) { | |
1916 | if (status != -EAGAIN) | |
1917 | mlog_errno(status); | |
1918 | goto out; | |
1919 | } | |
1920 | ||
1921 | /* Do no queue the tasks if the volume is being umounted */ | |
1922 | if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE) | |
1923 | goto unlock; | |
1924 | ||
1925 | if (os->os_seqno != seqno) { | |
1926 | os->os_seqno = seqno; | |
1927 | goto unlock; | |
1928 | } | |
1929 | ||
1930 | for (i = 0; i < osb->max_slots; i++) | |
1931 | ocfs2_queue_recovery_completion(osb->journal, i, NULL, NULL, | |
1932 | NULL, ORPHAN_NO_NEED_TRUNCATE); | |
1933 | /* | |
1934 | * We queued a recovery on orphan slots, increment the sequence | |
1935 | * number and update LVB so other node will skip the scan for a while | |
1936 | */ | |
1937 | seqno++; | |
1938 | os->os_count++; | |
1939 | os->os_scantime = CURRENT_TIME; | |
1940 | unlock: | |
1941 | ocfs2_orphan_scan_unlock(osb, seqno); | |
1942 | out: | |
1943 | trace_ocfs2_queue_orphan_scan_end(os->os_count, os->os_seqno, | |
1944 | atomic_read(&os->os_state)); | |
1945 | return; | |
1946 | } | |
1947 | ||
1948 | /* Worker task that gets fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT millsec */ | |
1949 | void ocfs2_orphan_scan_work(struct work_struct *work) | |
1950 | { | |
1951 | struct ocfs2_orphan_scan *os; | |
1952 | struct ocfs2_super *osb; | |
1953 | ||
1954 | os = container_of(work, struct ocfs2_orphan_scan, | |
1955 | os_orphan_scan_work.work); | |
1956 | osb = os->os_osb; | |
1957 | ||
1958 | mutex_lock(&os->os_lock); | |
1959 | ocfs2_queue_orphan_scan(osb); | |
1960 | if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE) | |
1961 | queue_delayed_work(ocfs2_wq, &os->os_orphan_scan_work, | |
1962 | ocfs2_orphan_scan_timeout()); | |
1963 | mutex_unlock(&os->os_lock); | |
1964 | } | |
1965 | ||
1966 | void ocfs2_orphan_scan_stop(struct ocfs2_super *osb) | |
1967 | { | |
1968 | struct ocfs2_orphan_scan *os; | |
1969 | ||
1970 | os = &osb->osb_orphan_scan; | |
1971 | if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE) { | |
1972 | atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE); | |
1973 | mutex_lock(&os->os_lock); | |
1974 | cancel_delayed_work(&os->os_orphan_scan_work); | |
1975 | mutex_unlock(&os->os_lock); | |
1976 | } | |
1977 | } | |
1978 | ||
1979 | void ocfs2_orphan_scan_init(struct ocfs2_super *osb) | |
1980 | { | |
1981 | struct ocfs2_orphan_scan *os; | |
1982 | ||
1983 | os = &osb->osb_orphan_scan; | |
1984 | os->os_osb = osb; | |
1985 | os->os_count = 0; | |
1986 | os->os_seqno = 0; | |
1987 | mutex_init(&os->os_lock); | |
1988 | INIT_DELAYED_WORK(&os->os_orphan_scan_work, ocfs2_orphan_scan_work); | |
1989 | } | |
1990 | ||
1991 | void ocfs2_orphan_scan_start(struct ocfs2_super *osb) | |
1992 | { | |
1993 | struct ocfs2_orphan_scan *os; | |
1994 | ||
1995 | os = &osb->osb_orphan_scan; | |
1996 | os->os_scantime = CURRENT_TIME; | |
1997 | if (ocfs2_is_hard_readonly(osb) || ocfs2_mount_local(osb)) | |
1998 | atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE); | |
1999 | else { | |
2000 | atomic_set(&os->os_state, ORPHAN_SCAN_ACTIVE); | |
2001 | queue_delayed_work(ocfs2_wq, &os->os_orphan_scan_work, | |
2002 | ocfs2_orphan_scan_timeout()); | |
2003 | } | |
2004 | } | |
2005 | ||
2006 | struct ocfs2_orphan_filldir_priv { | |
2007 | struct dir_context ctx; | |
2008 | struct inode *head; | |
2009 | struct ocfs2_super *osb; | |
2010 | }; | |
2011 | ||
2012 | static int ocfs2_orphan_filldir(struct dir_context *ctx, const char *name, | |
2013 | int name_len, loff_t pos, u64 ino, | |
2014 | unsigned type) | |
2015 | { | |
2016 | struct ocfs2_orphan_filldir_priv *p = | |
2017 | container_of(ctx, struct ocfs2_orphan_filldir_priv, ctx); | |
2018 | struct inode *iter; | |
2019 | ||
2020 | if (name_len == 1 && !strncmp(".", name, 1)) | |
2021 | return 0; | |
2022 | if (name_len == 2 && !strncmp("..", name, 2)) | |
2023 | return 0; | |
2024 | ||
2025 | /* Skip bad inodes so that recovery can continue */ | |
2026 | iter = ocfs2_iget(p->osb, ino, | |
2027 | OCFS2_FI_FLAG_ORPHAN_RECOVERY, 0); | |
2028 | if (IS_ERR(iter)) | |
2029 | return 0; | |
2030 | ||
2031 | /* Skip inodes which are already added to recover list, since dio may | |
2032 | * happen concurrently with unlink/rename */ | |
2033 | if (OCFS2_I(iter)->ip_next_orphan) { | |
2034 | iput(iter); | |
2035 | return 0; | |
2036 | } | |
2037 | ||
2038 | trace_ocfs2_orphan_filldir((unsigned long long)OCFS2_I(iter)->ip_blkno); | |
2039 | /* No locking is required for the next_orphan queue as there | |
2040 | * is only ever a single process doing orphan recovery. */ | |
2041 | OCFS2_I(iter)->ip_next_orphan = p->head; | |
2042 | p->head = iter; | |
2043 | ||
2044 | return 0; | |
2045 | } | |
2046 | ||
2047 | static int ocfs2_queue_orphans(struct ocfs2_super *osb, | |
2048 | int slot, | |
2049 | struct inode **head) | |
2050 | { | |
2051 | int status; | |
2052 | struct inode *orphan_dir_inode = NULL; | |
2053 | struct ocfs2_orphan_filldir_priv priv = { | |
2054 | .ctx.actor = ocfs2_orphan_filldir, | |
2055 | .osb = osb, | |
2056 | .head = *head | |
2057 | }; | |
2058 | ||
2059 | orphan_dir_inode = ocfs2_get_system_file_inode(osb, | |
2060 | ORPHAN_DIR_SYSTEM_INODE, | |
2061 | slot); | |
2062 | if (!orphan_dir_inode) { | |
2063 | status = -ENOENT; | |
2064 | mlog_errno(status); | |
2065 | return status; | |
2066 | } | |
2067 | ||
2068 | mutex_lock(&orphan_dir_inode->i_mutex); | |
2069 | status = ocfs2_inode_lock(orphan_dir_inode, NULL, 0); | |
2070 | if (status < 0) { | |
2071 | mlog_errno(status); | |
2072 | goto out; | |
2073 | } | |
2074 | ||
2075 | status = ocfs2_dir_foreach(orphan_dir_inode, &priv.ctx); | |
2076 | if (status) { | |
2077 | mlog_errno(status); | |
2078 | goto out_cluster; | |
2079 | } | |
2080 | ||
2081 | *head = priv.head; | |
2082 | ||
2083 | out_cluster: | |
2084 | ocfs2_inode_unlock(orphan_dir_inode, 0); | |
2085 | out: | |
2086 | mutex_unlock(&orphan_dir_inode->i_mutex); | |
2087 | iput(orphan_dir_inode); | |
2088 | return status; | |
2089 | } | |
2090 | ||
2091 | static int ocfs2_orphan_recovery_can_continue(struct ocfs2_super *osb, | |
2092 | int slot) | |
2093 | { | |
2094 | int ret; | |
2095 | ||
2096 | spin_lock(&osb->osb_lock); | |
2097 | ret = !osb->osb_orphan_wipes[slot]; | |
2098 | spin_unlock(&osb->osb_lock); | |
2099 | return ret; | |
2100 | } | |
2101 | ||
2102 | static void ocfs2_mark_recovering_orphan_dir(struct ocfs2_super *osb, | |
2103 | int slot) | |
2104 | { | |
2105 | spin_lock(&osb->osb_lock); | |
2106 | /* Mark ourselves such that new processes in delete_inode() | |
2107 | * know to quit early. */ | |
2108 | ocfs2_node_map_set_bit(osb, &osb->osb_recovering_orphan_dirs, slot); | |
2109 | while (osb->osb_orphan_wipes[slot]) { | |
2110 | /* If any processes are already in the middle of an | |
2111 | * orphan wipe on this dir, then we need to wait for | |
2112 | * them. */ | |
2113 | spin_unlock(&osb->osb_lock); | |
2114 | wait_event_interruptible(osb->osb_wipe_event, | |
2115 | ocfs2_orphan_recovery_can_continue(osb, slot)); | |
2116 | spin_lock(&osb->osb_lock); | |
2117 | } | |
2118 | spin_unlock(&osb->osb_lock); | |
2119 | } | |
2120 | ||
2121 | static void ocfs2_clear_recovering_orphan_dir(struct ocfs2_super *osb, | |
2122 | int slot) | |
2123 | { | |
2124 | ocfs2_node_map_clear_bit(osb, &osb->osb_recovering_orphan_dirs, slot); | |
2125 | } | |
2126 | ||
2127 | /* | |
2128 | * Orphan recovery. Each mounted node has it's own orphan dir which we | |
2129 | * must run during recovery. Our strategy here is to build a list of | |
2130 | * the inodes in the orphan dir and iget/iput them. The VFS does | |
2131 | * (most) of the rest of the work. | |
2132 | * | |
2133 | * Orphan recovery can happen at any time, not just mount so we have a | |
2134 | * couple of extra considerations. | |
2135 | * | |
2136 | * - We grab as many inodes as we can under the orphan dir lock - | |
2137 | * doing iget() outside the orphan dir risks getting a reference on | |
2138 | * an invalid inode. | |
2139 | * - We must be sure not to deadlock with other processes on the | |
2140 | * system wanting to run delete_inode(). This can happen when they go | |
2141 | * to lock the orphan dir and the orphan recovery process attempts to | |
2142 | * iget() inside the orphan dir lock. This can be avoided by | |
2143 | * advertising our state to ocfs2_delete_inode(). | |
2144 | */ | |
2145 | static int ocfs2_recover_orphans(struct ocfs2_super *osb, | |
2146 | int slot, | |
2147 | enum ocfs2_orphan_reco_type orphan_reco_type) | |
2148 | { | |
2149 | int ret = 0; | |
2150 | struct inode *inode = NULL; | |
2151 | struct inode *iter; | |
2152 | struct ocfs2_inode_info *oi; | |
2153 | struct buffer_head *di_bh = NULL; | |
2154 | struct ocfs2_dinode *di = NULL; | |
2155 | ||
2156 | trace_ocfs2_recover_orphans(slot); | |
2157 | ||
2158 | ocfs2_mark_recovering_orphan_dir(osb, slot); | |
2159 | ret = ocfs2_queue_orphans(osb, slot, &inode); | |
2160 | ocfs2_clear_recovering_orphan_dir(osb, slot); | |
2161 | ||
2162 | /* Error here should be noted, but we want to continue with as | |
2163 | * many queued inodes as we've got. */ | |
2164 | if (ret) | |
2165 | mlog_errno(ret); | |
2166 | ||
2167 | while (inode) { | |
2168 | oi = OCFS2_I(inode); | |
2169 | trace_ocfs2_recover_orphans_iput( | |
2170 | (unsigned long long)oi->ip_blkno); | |
2171 | ||
2172 | iter = oi->ip_next_orphan; | |
2173 | oi->ip_next_orphan = NULL; | |
2174 | ||
2175 | ret = ocfs2_rw_lock(inode, 1); | |
2176 | if (ret < 0) { | |
2177 | mlog_errno(ret); | |
2178 | goto next; | |
2179 | } | |
2180 | /* | |
2181 | * We need to take and drop the inode lock to | |
2182 | * force read inode from disk. | |
2183 | */ | |
2184 | ret = ocfs2_inode_lock(inode, &di_bh, 1); | |
2185 | if (ret) { | |
2186 | mlog_errno(ret); | |
2187 | goto unlock_rw; | |
2188 | } | |
2189 | ||
2190 | di = (struct ocfs2_dinode *)di_bh->b_data; | |
2191 | ||
2192 | if (inode->i_nlink == 0) { | |
2193 | spin_lock(&oi->ip_lock); | |
2194 | /* Set the proper information to get us going into | |
2195 | * ocfs2_delete_inode. */ | |
2196 | oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED; | |
2197 | spin_unlock(&oi->ip_lock); | |
2198 | } else if ((orphan_reco_type == ORPHAN_NEED_TRUNCATE) && | |
2199 | (di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL))) { | |
2200 | ret = ocfs2_truncate_file(inode, di_bh, | |
2201 | i_size_read(inode)); | |
2202 | if (ret < 0) { | |
2203 | if (ret != -ENOSPC) | |
2204 | mlog_errno(ret); | |
2205 | goto unlock_inode; | |
2206 | } | |
2207 | ||
2208 | ret = ocfs2_del_inode_from_orphan(osb, inode, di_bh, 0, 0); | |
2209 | if (ret) | |
2210 | mlog_errno(ret); | |
2211 | ||
2212 | wake_up(&OCFS2_I(inode)->append_dio_wq); | |
2213 | } /* else if ORPHAN_NO_NEED_TRUNCATE, do nothing */ | |
2214 | unlock_inode: | |
2215 | ocfs2_inode_unlock(inode, 1); | |
2216 | unlock_rw: | |
2217 | ocfs2_rw_unlock(inode, 1); | |
2218 | next: | |
2219 | iput(inode); | |
2220 | brelse(di_bh); | |
2221 | di_bh = NULL; | |
2222 | inode = iter; | |
2223 | } | |
2224 | ||
2225 | return ret; | |
2226 | } | |
2227 | ||
2228 | static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota) | |
2229 | { | |
2230 | /* This check is good because ocfs2 will wait on our recovery | |
2231 | * thread before changing it to something other than MOUNTED | |
2232 | * or DISABLED. */ | |
2233 | wait_event(osb->osb_mount_event, | |
2234 | (!quota && atomic_read(&osb->vol_state) == VOLUME_MOUNTED) || | |
2235 | atomic_read(&osb->vol_state) == VOLUME_MOUNTED_QUOTAS || | |
2236 | atomic_read(&osb->vol_state) == VOLUME_DISABLED); | |
2237 | ||
2238 | /* If there's an error on mount, then we may never get to the | |
2239 | * MOUNTED flag, but this is set right before | |
2240 | * dismount_volume() so we can trust it. */ | |
2241 | if (atomic_read(&osb->vol_state) == VOLUME_DISABLED) { | |
2242 | trace_ocfs2_wait_on_mount(VOLUME_DISABLED); | |
2243 | mlog(0, "mount error, exiting!\n"); | |
2244 | return -EBUSY; | |
2245 | } | |
2246 | ||
2247 | return 0; | |
2248 | } | |
2249 | ||
2250 | static int ocfs2_commit_thread(void *arg) | |
2251 | { | |
2252 | int status; | |
2253 | struct ocfs2_super *osb = arg; | |
2254 | struct ocfs2_journal *journal = osb->journal; | |
2255 | ||
2256 | /* we can trust j_num_trans here because _should_stop() is only set in | |
2257 | * shutdown and nobody other than ourselves should be able to start | |
2258 | * transactions. committing on shutdown might take a few iterations | |
2259 | * as final transactions put deleted inodes on the list */ | |
2260 | while (!(kthread_should_stop() && | |
2261 | atomic_read(&journal->j_num_trans) == 0)) { | |
2262 | ||
2263 | wait_event_interruptible(osb->checkpoint_event, | |
2264 | atomic_read(&journal->j_num_trans) | |
2265 | || kthread_should_stop()); | |
2266 | ||
2267 | status = ocfs2_commit_cache(osb); | |
2268 | if (status < 0) { | |
2269 | static unsigned long abort_warn_time; | |
2270 | ||
2271 | /* Warn about this once per minute */ | |
2272 | if (printk_timed_ratelimit(&abort_warn_time, 60*HZ)) | |
2273 | mlog(ML_ERROR, "status = %d, journal is " | |
2274 | "already aborted.\n", status); | |
2275 | /* | |
2276 | * After ocfs2_commit_cache() fails, j_num_trans has a | |
2277 | * non-zero value. Sleep here to avoid a busy-wait | |
2278 | * loop. | |
2279 | */ | |
2280 | msleep_interruptible(1000); | |
2281 | } | |
2282 | ||
2283 | if (kthread_should_stop() && atomic_read(&journal->j_num_trans)){ | |
2284 | mlog(ML_KTHREAD, | |
2285 | "commit_thread: %u transactions pending on " | |
2286 | "shutdown\n", | |
2287 | atomic_read(&journal->j_num_trans)); | |
2288 | } | |
2289 | } | |
2290 | ||
2291 | return 0; | |
2292 | } | |
2293 | ||
2294 | /* Reads all the journal inodes without taking any cluster locks. Used | |
2295 | * for hard readonly access to determine whether any journal requires | |
2296 | * recovery. Also used to refresh the recovery generation numbers after | |
2297 | * a journal has been recovered by another node. | |
2298 | */ | |
2299 | int ocfs2_check_journals_nolocks(struct ocfs2_super *osb) | |
2300 | { | |
2301 | int ret = 0; | |
2302 | unsigned int slot; | |
2303 | struct buffer_head *di_bh = NULL; | |
2304 | struct ocfs2_dinode *di; | |
2305 | int journal_dirty = 0; | |
2306 | ||
2307 | for(slot = 0; slot < osb->max_slots; slot++) { | |
2308 | ret = ocfs2_read_journal_inode(osb, slot, &di_bh, NULL); | |
2309 | if (ret) { | |
2310 | mlog_errno(ret); | |
2311 | goto out; | |
2312 | } | |
2313 | ||
2314 | di = (struct ocfs2_dinode *) di_bh->b_data; | |
2315 | ||
2316 | osb->slot_recovery_generations[slot] = | |
2317 | ocfs2_get_recovery_generation(di); | |
2318 | ||
2319 | if (le32_to_cpu(di->id1.journal1.ij_flags) & | |
2320 | OCFS2_JOURNAL_DIRTY_FL) | |
2321 | journal_dirty = 1; | |
2322 | ||
2323 | brelse(di_bh); | |
2324 | di_bh = NULL; | |
2325 | } | |
2326 | ||
2327 | out: | |
2328 | if (journal_dirty) | |
2329 | ret = -EROFS; | |
2330 | return ret; | |
2331 | } |