]>
Commit | Line | Data |
---|---|---|
328970de | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
ccd979bd MF |
2 | /* -*- mode: c; c-basic-offset: 8; -*- |
3 | * vim: noexpandtab sw=8 ts=8 sts=0: | |
4 | * | |
5 | * journal.c | |
6 | * | |
7 | * Defines functions of journalling api | |
8 | * | |
9 | * Copyright (C) 2003, 2004 Oracle. All rights reserved. | |
ccd979bd MF |
10 | */ |
11 | ||
12 | #include <linux/fs.h> | |
13 | #include <linux/types.h> | |
14 | #include <linux/slab.h> | |
15 | #include <linux/highmem.h> | |
16 | #include <linux/kthread.h> | |
83273932 SE |
17 | #include <linux/time.h> |
18 | #include <linux/random.h> | |
55b465b6 | 19 | #include <linux/delay.h> |
ccd979bd | 20 | |
ccd979bd MF |
21 | #include <cluster/masklog.h> |
22 | ||
23 | #include "ocfs2.h" | |
24 | ||
25 | #include "alloc.h" | |
50655ae9 | 26 | #include "blockcheck.h" |
316f4b9f | 27 | #include "dir.h" |
ccd979bd MF |
28 | #include "dlmglue.h" |
29 | #include "extent_map.h" | |
30 | #include "heartbeat.h" | |
31 | #include "inode.h" | |
32 | #include "journal.h" | |
33 | #include "localalloc.h" | |
ccd979bd MF |
34 | #include "slot_map.h" |
35 | #include "super.h" | |
ccd979bd | 36 | #include "sysfile.h" |
0cf2f763 | 37 | #include "uptodate.h" |
2205363d | 38 | #include "quota.h" |
ed460cff JQ |
39 | #include "file.h" |
40 | #include "namei.h" | |
ccd979bd MF |
41 | |
42 | #include "buffer_head_io.h" | |
b4107950 | 43 | #include "ocfs2_trace.h" |
ccd979bd | 44 | |
34af946a | 45 | DEFINE_SPINLOCK(trans_inc_lock); |
ccd979bd | 46 | |
83273932 SE |
47 | #define ORPHAN_SCAN_SCHEDULE_TIMEOUT 300000 |
48 | ||
ccd979bd MF |
49 | static int ocfs2_force_read_journal(struct inode *inode); |
50 | static int ocfs2_recover_node(struct ocfs2_super *osb, | |
2205363d | 51 | int node_num, int slot_num); |
ccd979bd MF |
52 | static int __ocfs2_recovery_thread(void *arg); |
53 | static int ocfs2_commit_cache(struct ocfs2_super *osb); | |
19ece546 | 54 | static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota); |
ccd979bd | 55 | static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb, |
539d8264 | 56 | int dirty, int replayed); |
ccd979bd MF |
57 | static int ocfs2_trylock_journal(struct ocfs2_super *osb, |
58 | int slot_num); | |
59 | static int ocfs2_recover_orphans(struct ocfs2_super *osb, | |
ed460cff JQ |
60 | int slot, |
61 | enum ocfs2_orphan_reco_type orphan_reco_type); | |
ccd979bd | 62 | static int ocfs2_commit_thread(void *arg); |
9140db04 SE |
63 | static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal, |
64 | int slot_num, | |
65 | struct ocfs2_dinode *la_dinode, | |
66 | struct ocfs2_dinode *tl_dinode, | |
ed460cff JQ |
67 | struct ocfs2_quota_recovery *qrec, |
68 | enum ocfs2_orphan_reco_type orphan_reco_type); | |
ccd979bd | 69 | |
19ece546 JK |
70 | static inline int ocfs2_wait_on_mount(struct ocfs2_super *osb) |
71 | { | |
72 | return __ocfs2_wait_on_mount(osb, 0); | |
73 | } | |
74 | ||
75 | static inline int ocfs2_wait_on_quotas(struct ocfs2_super *osb) | |
76 | { | |
77 | return __ocfs2_wait_on_mount(osb, 1); | |
78 | } | |
79 | ||
9140db04 SE |
80 | /* |
81 | * This replay_map is to track online/offline slots, so we could recover | |
82 | * offline slots during recovery and mount | |
83 | */ | |
84 | ||
85 | enum ocfs2_replay_state { | |
86 | REPLAY_UNNEEDED = 0, /* Replay is not needed, so ignore this map */ | |
87 | REPLAY_NEEDED, /* Replay slots marked in rm_replay_slots */ | |
88 | REPLAY_DONE /* Replay was already queued */ | |
89 | }; | |
90 | ||
91 | struct ocfs2_replay_map { | |
92 | unsigned int rm_slots; | |
93 | enum ocfs2_replay_state rm_state; | |
94 | unsigned char rm_replay_slots[0]; | |
95 | }; | |
96 | ||
b519ea6d | 97 | static void ocfs2_replay_map_set_state(struct ocfs2_super *osb, int state) |
9140db04 SE |
98 | { |
99 | if (!osb->replay_map) | |
100 | return; | |
101 | ||
102 | /* If we've already queued the replay, we don't have any more to do */ | |
103 | if (osb->replay_map->rm_state == REPLAY_DONE) | |
104 | return; | |
105 | ||
106 | osb->replay_map->rm_state = state; | |
107 | } | |
108 | ||
109 | int ocfs2_compute_replay_slots(struct ocfs2_super *osb) | |
110 | { | |
111 | struct ocfs2_replay_map *replay_map; | |
112 | int i, node_num; | |
113 | ||
114 | /* If replay map is already set, we don't do it again */ | |
115 | if (osb->replay_map) | |
116 | return 0; | |
117 | ||
118 | replay_map = kzalloc(sizeof(struct ocfs2_replay_map) + | |
119 | (osb->max_slots * sizeof(char)), GFP_KERNEL); | |
120 | ||
121 | if (!replay_map) { | |
122 | mlog_errno(-ENOMEM); | |
123 | return -ENOMEM; | |
124 | } | |
125 | ||
126 | spin_lock(&osb->osb_lock); | |
127 | ||
128 | replay_map->rm_slots = osb->max_slots; | |
129 | replay_map->rm_state = REPLAY_UNNEEDED; | |
130 | ||
131 | /* set rm_replay_slots for offline slot(s) */ | |
132 | for (i = 0; i < replay_map->rm_slots; i++) { | |
133 | if (ocfs2_slot_to_node_num_locked(osb, i, &node_num) == -ENOENT) | |
134 | replay_map->rm_replay_slots[i] = 1; | |
135 | } | |
136 | ||
137 | osb->replay_map = replay_map; | |
138 | spin_unlock(&osb->osb_lock); | |
139 | return 0; | |
140 | } | |
141 | ||
b519ea6d | 142 | static void ocfs2_queue_replay_slots(struct ocfs2_super *osb, |
ed460cff | 143 | enum ocfs2_orphan_reco_type orphan_reco_type) |
9140db04 SE |
144 | { |
145 | struct ocfs2_replay_map *replay_map = osb->replay_map; | |
146 | int i; | |
147 | ||
148 | if (!replay_map) | |
149 | return; | |
150 | ||
151 | if (replay_map->rm_state != REPLAY_NEEDED) | |
152 | return; | |
153 | ||
154 | for (i = 0; i < replay_map->rm_slots; i++) | |
155 | if (replay_map->rm_replay_slots[i]) | |
156 | ocfs2_queue_recovery_completion(osb->journal, i, NULL, | |
ed460cff JQ |
157 | NULL, NULL, |
158 | orphan_reco_type); | |
9140db04 SE |
159 | replay_map->rm_state = REPLAY_DONE; |
160 | } | |
161 | ||
b519ea6d | 162 | static void ocfs2_free_replay_slots(struct ocfs2_super *osb) |
9140db04 SE |
163 | { |
164 | struct ocfs2_replay_map *replay_map = osb->replay_map; | |
165 | ||
166 | if (!osb->replay_map) | |
167 | return; | |
168 | ||
169 | kfree(replay_map); | |
170 | osb->replay_map = NULL; | |
171 | } | |
172 | ||
553abd04 JB |
173 | int ocfs2_recovery_init(struct ocfs2_super *osb) |
174 | { | |
175 | struct ocfs2_recovery_map *rm; | |
176 | ||
177 | mutex_init(&osb->recovery_lock); | |
178 | osb->disable_recovery = 0; | |
179 | osb->recovery_thread_task = NULL; | |
180 | init_waitqueue_head(&osb->recovery_event); | |
181 | ||
182 | rm = kzalloc(sizeof(struct ocfs2_recovery_map) + | |
183 | osb->max_slots * sizeof(unsigned int), | |
184 | GFP_KERNEL); | |
185 | if (!rm) { | |
186 | mlog_errno(-ENOMEM); | |
187 | return -ENOMEM; | |
188 | } | |
189 | ||
190 | rm->rm_entries = (unsigned int *)((char *)rm + | |
191 | sizeof(struct ocfs2_recovery_map)); | |
192 | osb->recovery_map = rm; | |
193 | ||
194 | return 0; | |
195 | } | |
196 | ||
197 | /* we can't grab the goofy sem lock from inside wait_event, so we use | |
198 | * memory barriers to make sure that we'll see the null task before | |
199 | * being woken up */ | |
200 | static int ocfs2_recovery_thread_running(struct ocfs2_super *osb) | |
201 | { | |
202 | mb(); | |
203 | return osb->recovery_thread_task != NULL; | |
204 | } | |
205 | ||
206 | void ocfs2_recovery_exit(struct ocfs2_super *osb) | |
207 | { | |
208 | struct ocfs2_recovery_map *rm; | |
209 | ||
210 | /* disable any new recovery threads and wait for any currently | |
211 | * running ones to exit. Do this before setting the vol_state. */ | |
212 | mutex_lock(&osb->recovery_lock); | |
213 | osb->disable_recovery = 1; | |
214 | mutex_unlock(&osb->recovery_lock); | |
215 | wait_event(osb->recovery_event, !ocfs2_recovery_thread_running(osb)); | |
216 | ||
217 | /* At this point, we know that no more recovery threads can be | |
218 | * launched, so wait for any recovery completion work to | |
219 | * complete. */ | |
35ddf78e | 220 | flush_workqueue(osb->ocfs2_wq); |
553abd04 JB |
221 | |
222 | /* | |
223 | * Now that recovery is shut down, and the osb is about to be | |
224 | * freed, the osb_lock is not taken here. | |
225 | */ | |
226 | rm = osb->recovery_map; | |
227 | /* XXX: Should we bug if there are dirty entries? */ | |
228 | ||
229 | kfree(rm); | |
230 | } | |
231 | ||
232 | static int __ocfs2_recovery_map_test(struct ocfs2_super *osb, | |
233 | unsigned int node_num) | |
234 | { | |
235 | int i; | |
236 | struct ocfs2_recovery_map *rm = osb->recovery_map; | |
237 | ||
238 | assert_spin_locked(&osb->osb_lock); | |
239 | ||
240 | for (i = 0; i < rm->rm_used; i++) { | |
241 | if (rm->rm_entries[i] == node_num) | |
242 | return 1; | |
243 | } | |
244 | ||
245 | return 0; | |
246 | } | |
247 | ||
248 | /* Behaves like test-and-set. Returns the previous value */ | |
249 | static int ocfs2_recovery_map_set(struct ocfs2_super *osb, | |
250 | unsigned int node_num) | |
251 | { | |
252 | struct ocfs2_recovery_map *rm = osb->recovery_map; | |
253 | ||
254 | spin_lock(&osb->osb_lock); | |
255 | if (__ocfs2_recovery_map_test(osb, node_num)) { | |
256 | spin_unlock(&osb->osb_lock); | |
257 | return 1; | |
258 | } | |
259 | ||
260 | /* XXX: Can this be exploited? Not from o2dlm... */ | |
261 | BUG_ON(rm->rm_used >= osb->max_slots); | |
262 | ||
263 | rm->rm_entries[rm->rm_used] = node_num; | |
264 | rm->rm_used++; | |
265 | spin_unlock(&osb->osb_lock); | |
266 | ||
267 | return 0; | |
268 | } | |
269 | ||
270 | static void ocfs2_recovery_map_clear(struct ocfs2_super *osb, | |
271 | unsigned int node_num) | |
272 | { | |
273 | int i; | |
274 | struct ocfs2_recovery_map *rm = osb->recovery_map; | |
275 | ||
276 | spin_lock(&osb->osb_lock); | |
277 | ||
278 | for (i = 0; i < rm->rm_used; i++) { | |
279 | if (rm->rm_entries[i] == node_num) | |
280 | break; | |
281 | } | |
282 | ||
283 | if (i < rm->rm_used) { | |
284 | /* XXX: be careful with the pointer math */ | |
285 | memmove(&(rm->rm_entries[i]), &(rm->rm_entries[i + 1]), | |
286 | (rm->rm_used - i - 1) * sizeof(unsigned int)); | |
287 | rm->rm_used--; | |
288 | } | |
289 | ||
290 | spin_unlock(&osb->osb_lock); | |
291 | } | |
292 | ||
ccd979bd MF |
293 | static int ocfs2_commit_cache(struct ocfs2_super *osb) |
294 | { | |
295 | int status = 0; | |
296 | unsigned int flushed; | |
ccd979bd MF |
297 | struct ocfs2_journal *journal = NULL; |
298 | ||
ccd979bd MF |
299 | journal = osb->journal; |
300 | ||
301 | /* Flush all pending commits and checkpoint the journal. */ | |
302 | down_write(&journal->j_trans_barrier); | |
303 | ||
b4107950 TM |
304 | flushed = atomic_read(&journal->j_num_trans); |
305 | trace_ocfs2_commit_cache_begin(flushed); | |
306 | if (flushed == 0) { | |
ccd979bd | 307 | up_write(&journal->j_trans_barrier); |
ccd979bd MF |
308 | goto finally; |
309 | } | |
310 | ||
2b4e30fb JB |
311 | jbd2_journal_lock_updates(journal->j_journal); |
312 | status = jbd2_journal_flush(journal->j_journal); | |
313 | jbd2_journal_unlock_updates(journal->j_journal); | |
ccd979bd MF |
314 | if (status < 0) { |
315 | up_write(&journal->j_trans_barrier); | |
316 | mlog_errno(status); | |
317 | goto finally; | |
318 | } | |
319 | ||
f9c57ada | 320 | ocfs2_inc_trans_id(journal); |
ccd979bd MF |
321 | |
322 | flushed = atomic_read(&journal->j_num_trans); | |
323 | atomic_set(&journal->j_num_trans, 0); | |
324 | up_write(&journal->j_trans_barrier); | |
325 | ||
b4107950 | 326 | trace_ocfs2_commit_cache_end(journal->j_trans_id, flushed); |
ccd979bd | 327 | |
34d024f8 | 328 | ocfs2_wake_downconvert_thread(osb); |
ccd979bd MF |
329 | wake_up(&journal->j_checkpointed); |
330 | finally: | |
ccd979bd MF |
331 | return status; |
332 | } | |
333 | ||
1fabe148 | 334 | handle_t *ocfs2_start_trans(struct ocfs2_super *osb, int max_buffs) |
ccd979bd | 335 | { |
ccd979bd | 336 | journal_t *journal = osb->journal->j_journal; |
1fabe148 | 337 | handle_t *handle; |
ccd979bd | 338 | |
ebdec83b | 339 | BUG_ON(!osb || !osb->journal->j_journal); |
ccd979bd | 340 | |
65eff9cc MF |
341 | if (ocfs2_is_hard_readonly(osb)) |
342 | return ERR_PTR(-EROFS); | |
ccd979bd MF |
343 | |
344 | BUG_ON(osb->journal->j_state == OCFS2_JOURNAL_FREE); | |
345 | BUG_ON(max_buffs <= 0); | |
346 | ||
90e86a63 JK |
347 | /* Nested transaction? Just return the handle... */ |
348 | if (journal_current_handle()) | |
349 | return jbd2_journal_start(journal, max_buffs); | |
ccd979bd | 350 | |
fef6925c JK |
351 | sb_start_intwrite(osb->sb); |
352 | ||
ccd979bd MF |
353 | down_read(&osb->journal->j_trans_barrier); |
354 | ||
2b4e30fb | 355 | handle = jbd2_journal_start(journal, max_buffs); |
1fabe148 | 356 | if (IS_ERR(handle)) { |
ccd979bd | 357 | up_read(&osb->journal->j_trans_barrier); |
fef6925c | 358 | sb_end_intwrite(osb->sb); |
ccd979bd | 359 | |
1fabe148 | 360 | mlog_errno(PTR_ERR(handle)); |
ccd979bd MF |
361 | |
362 | if (is_journal_aborted(journal)) { | |
7ecef14a | 363 | ocfs2_abort(osb->sb, "Detected aborted journal\n"); |
1fabe148 | 364 | handle = ERR_PTR(-EROFS); |
ccd979bd | 365 | } |
c271c5c2 SM |
366 | } else { |
367 | if (!ocfs2_mount_local(osb)) | |
368 | atomic_inc(&(osb->journal->j_num_trans)); | |
369 | } | |
ccd979bd | 370 | |
ccd979bd | 371 | return handle; |
ccd979bd MF |
372 | } |
373 | ||
1fabe148 MF |
374 | int ocfs2_commit_trans(struct ocfs2_super *osb, |
375 | handle_t *handle) | |
ccd979bd | 376 | { |
90e86a63 | 377 | int ret, nested; |
02dc1af4 | 378 | struct ocfs2_journal *journal = osb->journal; |
ccd979bd MF |
379 | |
380 | BUG_ON(!handle); | |
381 | ||
90e86a63 | 382 | nested = handle->h_ref > 1; |
2b4e30fb | 383 | ret = jbd2_journal_stop(handle); |
1fabe148 MF |
384 | if (ret < 0) |
385 | mlog_errno(ret); | |
ccd979bd | 386 | |
fef6925c | 387 | if (!nested) { |
90e86a63 | 388 | up_read(&journal->j_trans_barrier); |
fef6925c JK |
389 | sb_end_intwrite(osb->sb); |
390 | } | |
ccd979bd | 391 | |
1fabe148 | 392 | return ret; |
ccd979bd MF |
393 | } |
394 | ||
395 | /* | |
c901fb00 | 396 | * 'nblocks' is what you want to add to the current transaction. |
ccd979bd | 397 | * |
2b4e30fb | 398 | * This might call jbd2_journal_restart() which will commit dirty buffers |
e8aed345 MF |
399 | * and then restart the transaction. Before calling |
400 | * ocfs2_extend_trans(), any changed blocks should have been | |
401 | * dirtied. After calling it, all blocks which need to be changed must | |
402 | * go through another set of journal_access/journal_dirty calls. | |
403 | * | |
ccd979bd MF |
404 | * WARNING: This will not release any semaphores or disk locks taken |
405 | * during the transaction, so make sure they were taken *before* | |
406 | * start_trans or we'll have ordering deadlocks. | |
407 | * | |
408 | * WARNING2: Note that we do *not* drop j_trans_barrier here. This is | |
409 | * good because transaction ids haven't yet been recorded on the | |
410 | * cluster locks associated with this handle. | |
411 | */ | |
1fc58146 | 412 | int ocfs2_extend_trans(handle_t *handle, int nblocks) |
ccd979bd | 413 | { |
c901fb00 | 414 | int status, old_nblocks; |
ccd979bd MF |
415 | |
416 | BUG_ON(!handle); | |
c901fb00 | 417 | BUG_ON(nblocks < 0); |
ccd979bd | 418 | |
c901fb00 TM |
419 | if (!nblocks) |
420 | return 0; | |
421 | ||
9797a902 | 422 | old_nblocks = jbd2_handle_buffer_credits(handle); |
ccd979bd | 423 | |
b4107950 | 424 | trace_ocfs2_extend_trans(old_nblocks, nblocks); |
ccd979bd | 425 | |
e407e397 | 426 | #ifdef CONFIG_OCFS2_DEBUG_FS |
0879c584 MF |
427 | status = 1; |
428 | #else | |
fdc3ef88 | 429 | status = jbd2_journal_extend(handle, nblocks, 0); |
ccd979bd MF |
430 | if (status < 0) { |
431 | mlog_errno(status); | |
432 | goto bail; | |
433 | } | |
0879c584 | 434 | #endif |
ccd979bd MF |
435 | |
436 | if (status > 0) { | |
b4107950 | 437 | trace_ocfs2_extend_trans_restart(old_nblocks + nblocks); |
c901fb00 TM |
438 | status = jbd2_journal_restart(handle, |
439 | old_nblocks + nblocks); | |
ccd979bd | 440 | if (status < 0) { |
ccd979bd MF |
441 | mlog_errno(status); |
442 | goto bail; | |
443 | } | |
01ddf1e1 | 444 | } |
ccd979bd MF |
445 | |
446 | status = 0; | |
447 | bail: | |
ccd979bd MF |
448 | return status; |
449 | } | |
450 | ||
2b1e55c3 YL |
451 | /* |
452 | * If we have fewer than thresh credits, extend by OCFS2_MAX_TRANS_DATA. | |
453 | * If that fails, restart the transaction & regain write access for the | |
454 | * buffer head which is used for metadata modifications. | |
455 | * Taken from Ext4: extend_or_restart_transaction() | |
456 | */ | |
457 | int ocfs2_allocate_extend_trans(handle_t *handle, int thresh) | |
458 | { | |
459 | int status, old_nblks; | |
460 | ||
461 | BUG_ON(!handle); | |
462 | ||
9797a902 | 463 | old_nblks = jbd2_handle_buffer_credits(handle); |
2b1e55c3 YL |
464 | trace_ocfs2_allocate_extend_trans(old_nblks, thresh); |
465 | ||
466 | if (old_nblks < thresh) | |
467 | return 0; | |
468 | ||
fdc3ef88 | 469 | status = jbd2_journal_extend(handle, OCFS2_MAX_TRANS_DATA, 0); |
2b1e55c3 YL |
470 | if (status < 0) { |
471 | mlog_errno(status); | |
472 | goto bail; | |
473 | } | |
474 | ||
475 | if (status > 0) { | |
476 | status = jbd2_journal_restart(handle, OCFS2_MAX_TRANS_DATA); | |
477 | if (status < 0) | |
478 | mlog_errno(status); | |
479 | } | |
480 | ||
481 | bail: | |
482 | return status; | |
483 | } | |
484 | ||
485 | ||
50655ae9 JB |
486 | struct ocfs2_triggers { |
487 | struct jbd2_buffer_trigger_type ot_triggers; | |
488 | int ot_offset; | |
489 | }; | |
490 | ||
491 | static inline struct ocfs2_triggers *to_ocfs2_trigger(struct jbd2_buffer_trigger_type *triggers) | |
492 | { | |
493 | return container_of(triggers, struct ocfs2_triggers, ot_triggers); | |
494 | } | |
495 | ||
13ceef09 | 496 | static void ocfs2_frozen_trigger(struct jbd2_buffer_trigger_type *triggers, |
50655ae9 JB |
497 | struct buffer_head *bh, |
498 | void *data, size_t size) | |
499 | { | |
500 | struct ocfs2_triggers *ot = to_ocfs2_trigger(triggers); | |
501 | ||
502 | /* | |
503 | * We aren't guaranteed to have the superblock here, so we | |
504 | * must unconditionally compute the ecc data. | |
505 | * __ocfs2_journal_access() will only set the triggers if | |
506 | * metaecc is enabled. | |
507 | */ | |
508 | ocfs2_block_check_compute(data, size, data + ot->ot_offset); | |
509 | } | |
510 | ||
511 | /* | |
512 | * Quota blocks have their own trigger because the struct ocfs2_block_check | |
513 | * offset depends on the blocksize. | |
514 | */ | |
13ceef09 | 515 | static void ocfs2_dq_frozen_trigger(struct jbd2_buffer_trigger_type *triggers, |
50655ae9 JB |
516 | struct buffer_head *bh, |
517 | void *data, size_t size) | |
518 | { | |
519 | struct ocfs2_disk_dqtrailer *dqt = | |
520 | ocfs2_block_dqtrailer(size, data); | |
521 | ||
522 | /* | |
523 | * We aren't guaranteed to have the superblock here, so we | |
524 | * must unconditionally compute the ecc data. | |
525 | * __ocfs2_journal_access() will only set the triggers if | |
526 | * metaecc is enabled. | |
527 | */ | |
528 | ocfs2_block_check_compute(data, size, &dqt->dq_check); | |
529 | } | |
530 | ||
c175a518 JB |
531 | /* |
532 | * Directory blocks also have their own trigger because the | |
533 | * struct ocfs2_block_check offset depends on the blocksize. | |
534 | */ | |
13ceef09 | 535 | static void ocfs2_db_frozen_trigger(struct jbd2_buffer_trigger_type *triggers, |
c175a518 JB |
536 | struct buffer_head *bh, |
537 | void *data, size_t size) | |
538 | { | |
539 | struct ocfs2_dir_block_trailer *trailer = | |
540 | ocfs2_dir_trailer_from_size(size, data); | |
541 | ||
542 | /* | |
543 | * We aren't guaranteed to have the superblock here, so we | |
544 | * must unconditionally compute the ecc data. | |
545 | * __ocfs2_journal_access() will only set the triggers if | |
546 | * metaecc is enabled. | |
547 | */ | |
548 | ocfs2_block_check_compute(data, size, &trailer->db_check); | |
549 | } | |
550 | ||
50655ae9 JB |
551 | static void ocfs2_abort_trigger(struct jbd2_buffer_trigger_type *triggers, |
552 | struct buffer_head *bh) | |
553 | { | |
554 | mlog(ML_ERROR, | |
555 | "ocfs2_abort_trigger called by JBD2. bh = 0x%lx, " | |
556 | "bh->b_blocknr = %llu\n", | |
557 | (unsigned long)bh, | |
558 | (unsigned long long)bh->b_blocknr); | |
559 | ||
74e364ad | 560 | ocfs2_error(bh->b_bdev->bd_super, |
50655ae9 JB |
561 | "JBD2 has aborted our journal, ocfs2 cannot continue\n"); |
562 | } | |
563 | ||
564 | static struct ocfs2_triggers di_triggers = { | |
565 | .ot_triggers = { | |
13ceef09 | 566 | .t_frozen = ocfs2_frozen_trigger, |
50655ae9 JB |
567 | .t_abort = ocfs2_abort_trigger, |
568 | }, | |
569 | .ot_offset = offsetof(struct ocfs2_dinode, i_check), | |
570 | }; | |
571 | ||
572 | static struct ocfs2_triggers eb_triggers = { | |
573 | .ot_triggers = { | |
13ceef09 | 574 | .t_frozen = ocfs2_frozen_trigger, |
50655ae9 JB |
575 | .t_abort = ocfs2_abort_trigger, |
576 | }, | |
577 | .ot_offset = offsetof(struct ocfs2_extent_block, h_check), | |
578 | }; | |
579 | ||
93c97087 TM |
580 | static struct ocfs2_triggers rb_triggers = { |
581 | .ot_triggers = { | |
13ceef09 | 582 | .t_frozen = ocfs2_frozen_trigger, |
93c97087 TM |
583 | .t_abort = ocfs2_abort_trigger, |
584 | }, | |
585 | .ot_offset = offsetof(struct ocfs2_refcount_block, rf_check), | |
586 | }; | |
587 | ||
50655ae9 JB |
588 | static struct ocfs2_triggers gd_triggers = { |
589 | .ot_triggers = { | |
13ceef09 | 590 | .t_frozen = ocfs2_frozen_trigger, |
50655ae9 JB |
591 | .t_abort = ocfs2_abort_trigger, |
592 | }, | |
593 | .ot_offset = offsetof(struct ocfs2_group_desc, bg_check), | |
594 | }; | |
595 | ||
c175a518 JB |
596 | static struct ocfs2_triggers db_triggers = { |
597 | .ot_triggers = { | |
13ceef09 | 598 | .t_frozen = ocfs2_db_frozen_trigger, |
c175a518 JB |
599 | .t_abort = ocfs2_abort_trigger, |
600 | }, | |
601 | }; | |
602 | ||
50655ae9 JB |
603 | static struct ocfs2_triggers xb_triggers = { |
604 | .ot_triggers = { | |
13ceef09 | 605 | .t_frozen = ocfs2_frozen_trigger, |
50655ae9 JB |
606 | .t_abort = ocfs2_abort_trigger, |
607 | }, | |
608 | .ot_offset = offsetof(struct ocfs2_xattr_block, xb_check), | |
609 | }; | |
610 | ||
611 | static struct ocfs2_triggers dq_triggers = { | |
612 | .ot_triggers = { | |
13ceef09 | 613 | .t_frozen = ocfs2_dq_frozen_trigger, |
50655ae9 JB |
614 | .t_abort = ocfs2_abort_trigger, |
615 | }, | |
616 | }; | |
617 | ||
9b7895ef MF |
618 | static struct ocfs2_triggers dr_triggers = { |
619 | .ot_triggers = { | |
13ceef09 | 620 | .t_frozen = ocfs2_frozen_trigger, |
9b7895ef MF |
621 | .t_abort = ocfs2_abort_trigger, |
622 | }, | |
623 | .ot_offset = offsetof(struct ocfs2_dx_root_block, dr_check), | |
624 | }; | |
625 | ||
626 | static struct ocfs2_triggers dl_triggers = { | |
627 | .ot_triggers = { | |
13ceef09 | 628 | .t_frozen = ocfs2_frozen_trigger, |
9b7895ef MF |
629 | .t_abort = ocfs2_abort_trigger, |
630 | }, | |
631 | .ot_offset = offsetof(struct ocfs2_dx_leaf, dl_check), | |
632 | }; | |
633 | ||
50655ae9 | 634 | static int __ocfs2_journal_access(handle_t *handle, |
0cf2f763 | 635 | struct ocfs2_caching_info *ci, |
50655ae9 JB |
636 | struct buffer_head *bh, |
637 | struct ocfs2_triggers *triggers, | |
638 | int type) | |
ccd979bd MF |
639 | { |
640 | int status; | |
0cf2f763 JB |
641 | struct ocfs2_super *osb = |
642 | OCFS2_SB(ocfs2_metadata_cache_get_super(ci)); | |
ccd979bd | 643 | |
0cf2f763 | 644 | BUG_ON(!ci || !ci->ci_ops); |
ccd979bd MF |
645 | BUG_ON(!handle); |
646 | BUG_ON(!bh); | |
ccd979bd | 647 | |
b4107950 TM |
648 | trace_ocfs2_journal_access( |
649 | (unsigned long long)ocfs2_metadata_cache_owner(ci), | |
650 | (unsigned long long)bh->b_blocknr, type, bh->b_size); | |
ccd979bd MF |
651 | |
652 | /* we can safely remove this assertion after testing. */ | |
653 | if (!buffer_uptodate(bh)) { | |
654 | mlog(ML_ERROR, "giving me a buffer that's not uptodate!\n"); | |
d984187e | 655 | mlog(ML_ERROR, "b_blocknr=%llu, b_state=0x%lx\n", |
656 | (unsigned long long)bh->b_blocknr, bh->b_state); | |
acf8fdbe JQ |
657 | |
658 | lock_buffer(bh); | |
659 | /* | |
d984187e | 660 | * A previous transaction with a couple of buffer heads fail |
661 | * to checkpoint, so all the bhs are marked as BH_Write_EIO. | |
662 | * For current transaction, the bh is just among those error | |
663 | * bhs which previous transaction handle. We can't just clear | |
664 | * its BH_Write_EIO and reuse directly, since other bhs are | |
665 | * not written to disk yet and that will cause metadata | |
666 | * inconsistency. So we should set fs read-only to avoid | |
667 | * further damage. | |
acf8fdbe JQ |
668 | */ |
669 | if (buffer_write_io_error(bh) && !buffer_uptodate(bh)) { | |
acf8fdbe | 670 | unlock_buffer(bh); |
d984187e | 671 | return ocfs2_error(osb->sb, "A previous attempt to " |
672 | "write this buffer head failed\n"); | |
acf8fdbe JQ |
673 | } |
674 | unlock_buffer(bh); | |
ccd979bd MF |
675 | } |
676 | ||
0cf2f763 | 677 | /* Set the current transaction information on the ci so |
ccd979bd | 678 | * that the locking code knows whether it can drop it's locks |
0cf2f763 | 679 | * on this ci or not. We're protected from the commit |
ccd979bd MF |
680 | * thread updating the current transaction id until |
681 | * ocfs2_commit_trans() because ocfs2_start_trans() took | |
682 | * j_trans_barrier for us. */ | |
0cf2f763 | 683 | ocfs2_set_ci_lock_trans(osb->journal, ci); |
ccd979bd | 684 | |
0cf2f763 | 685 | ocfs2_metadata_cache_io_lock(ci); |
ccd979bd MF |
686 | switch (type) { |
687 | case OCFS2_JOURNAL_ACCESS_CREATE: | |
688 | case OCFS2_JOURNAL_ACCESS_WRITE: | |
2b4e30fb | 689 | status = jbd2_journal_get_write_access(handle, bh); |
ccd979bd MF |
690 | break; |
691 | ||
692 | case OCFS2_JOURNAL_ACCESS_UNDO: | |
2b4e30fb | 693 | status = jbd2_journal_get_undo_access(handle, bh); |
ccd979bd MF |
694 | break; |
695 | ||
696 | default: | |
697 | status = -EINVAL; | |
af901ca1 | 698 | mlog(ML_ERROR, "Unknown access type!\n"); |
ccd979bd | 699 | } |
0cf2f763 | 700 | if (!status && ocfs2_meta_ecc(osb) && triggers) |
50655ae9 | 701 | jbd2_journal_set_triggers(bh, &triggers->ot_triggers); |
0cf2f763 | 702 | ocfs2_metadata_cache_io_unlock(ci); |
ccd979bd MF |
703 | |
704 | if (status < 0) | |
705 | mlog(ML_ERROR, "Error %d getting %d access to buffer!\n", | |
706 | status, type); | |
707 | ||
ccd979bd MF |
708 | return status; |
709 | } | |
710 | ||
0cf2f763 JB |
711 | int ocfs2_journal_access_di(handle_t *handle, struct ocfs2_caching_info *ci, |
712 | struct buffer_head *bh, int type) | |
50655ae9 | 713 | { |
0cf2f763 | 714 | return __ocfs2_journal_access(handle, ci, bh, &di_triggers, type); |
50655ae9 JB |
715 | } |
716 | ||
0cf2f763 | 717 | int ocfs2_journal_access_eb(handle_t *handle, struct ocfs2_caching_info *ci, |
50655ae9 JB |
718 | struct buffer_head *bh, int type) |
719 | { | |
0cf2f763 | 720 | return __ocfs2_journal_access(handle, ci, bh, &eb_triggers, type); |
50655ae9 JB |
721 | } |
722 | ||
93c97087 TM |
723 | int ocfs2_journal_access_rb(handle_t *handle, struct ocfs2_caching_info *ci, |
724 | struct buffer_head *bh, int type) | |
725 | { | |
726 | return __ocfs2_journal_access(handle, ci, bh, &rb_triggers, | |
727 | type); | |
728 | } | |
729 | ||
0cf2f763 | 730 | int ocfs2_journal_access_gd(handle_t *handle, struct ocfs2_caching_info *ci, |
50655ae9 JB |
731 | struct buffer_head *bh, int type) |
732 | { | |
0cf2f763 | 733 | return __ocfs2_journal_access(handle, ci, bh, &gd_triggers, type); |
50655ae9 JB |
734 | } |
735 | ||
0cf2f763 | 736 | int ocfs2_journal_access_db(handle_t *handle, struct ocfs2_caching_info *ci, |
50655ae9 JB |
737 | struct buffer_head *bh, int type) |
738 | { | |
0cf2f763 | 739 | return __ocfs2_journal_access(handle, ci, bh, &db_triggers, type); |
50655ae9 JB |
740 | } |
741 | ||
0cf2f763 | 742 | int ocfs2_journal_access_xb(handle_t *handle, struct ocfs2_caching_info *ci, |
50655ae9 JB |
743 | struct buffer_head *bh, int type) |
744 | { | |
0cf2f763 | 745 | return __ocfs2_journal_access(handle, ci, bh, &xb_triggers, type); |
50655ae9 JB |
746 | } |
747 | ||
0cf2f763 | 748 | int ocfs2_journal_access_dq(handle_t *handle, struct ocfs2_caching_info *ci, |
50655ae9 JB |
749 | struct buffer_head *bh, int type) |
750 | { | |
0cf2f763 | 751 | return __ocfs2_journal_access(handle, ci, bh, &dq_triggers, type); |
50655ae9 JB |
752 | } |
753 | ||
0cf2f763 | 754 | int ocfs2_journal_access_dr(handle_t *handle, struct ocfs2_caching_info *ci, |
9b7895ef MF |
755 | struct buffer_head *bh, int type) |
756 | { | |
0cf2f763 | 757 | return __ocfs2_journal_access(handle, ci, bh, &dr_triggers, type); |
9b7895ef MF |
758 | } |
759 | ||
0cf2f763 | 760 | int ocfs2_journal_access_dl(handle_t *handle, struct ocfs2_caching_info *ci, |
9b7895ef MF |
761 | struct buffer_head *bh, int type) |
762 | { | |
0cf2f763 | 763 | return __ocfs2_journal_access(handle, ci, bh, &dl_triggers, type); |
9b7895ef MF |
764 | } |
765 | ||
0cf2f763 | 766 | int ocfs2_journal_access(handle_t *handle, struct ocfs2_caching_info *ci, |
50655ae9 JB |
767 | struct buffer_head *bh, int type) |
768 | { | |
0cf2f763 | 769 | return __ocfs2_journal_access(handle, ci, bh, NULL, type); |
50655ae9 JB |
770 | } |
771 | ||
ec20cec7 | 772 | void ocfs2_journal_dirty(handle_t *handle, struct buffer_head *bh) |
ccd979bd MF |
773 | { |
774 | int status; | |
775 | ||
b4107950 | 776 | trace_ocfs2_journal_dirty((unsigned long long)bh->b_blocknr); |
ccd979bd | 777 | |
2b4e30fb | 778 | status = jbd2_journal_dirty_metadata(handle, bh); |
e272e7f0 JQ |
779 | if (status) { |
780 | mlog_errno(status); | |
781 | if (!is_handle_aborted(handle)) { | |
782 | journal_t *journal = handle->h_transaction->t_journal; | |
783 | struct super_block *sb = bh->b_bdev->bd_super; | |
784 | ||
785 | mlog(ML_ERROR, "jbd2_journal_dirty_metadata failed. " | |
786 | "Aborting transaction and journal.\n"); | |
787 | handle->h_err = status; | |
788 | jbd2_journal_abort_handle(handle); | |
789 | jbd2_journal_abort(journal, status); | |
790 | ocfs2_abort(sb, "Journal already aborted.\n"); | |
791 | } | |
792 | } | |
ccd979bd MF |
793 | } |
794 | ||
2b4e30fb | 795 | #define OCFS2_DEFAULT_COMMIT_INTERVAL (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE) |
ccd979bd MF |
796 | |
797 | void ocfs2_set_journal_params(struct ocfs2_super *osb) | |
798 | { | |
799 | journal_t *journal = osb->journal->j_journal; | |
d147b3d6 MF |
800 | unsigned long commit_interval = OCFS2_DEFAULT_COMMIT_INTERVAL; |
801 | ||
802 | if (osb->osb_commit_interval) | |
803 | commit_interval = osb->osb_commit_interval; | |
ccd979bd | 804 | |
a931da6a | 805 | write_lock(&journal->j_state_lock); |
d147b3d6 | 806 | journal->j_commit_interval = commit_interval; |
ccd979bd | 807 | if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER) |
2b4e30fb | 808 | journal->j_flags |= JBD2_BARRIER; |
ccd979bd | 809 | else |
2b4e30fb | 810 | journal->j_flags &= ~JBD2_BARRIER; |
a931da6a | 811 | write_unlock(&journal->j_state_lock); |
ccd979bd MF |
812 | } |
813 | ||
814 | int ocfs2_journal_init(struct ocfs2_journal *journal, int *dirty) | |
815 | { | |
816 | int status = -1; | |
817 | struct inode *inode = NULL; /* the journal inode */ | |
818 | journal_t *j_journal = NULL; | |
819 | struct ocfs2_dinode *di = NULL; | |
820 | struct buffer_head *bh = NULL; | |
821 | struct ocfs2_super *osb; | |
e63aecb6 | 822 | int inode_lock = 0; |
ccd979bd | 823 | |
ccd979bd MF |
824 | BUG_ON(!journal); |
825 | ||
826 | osb = journal->j_osb; | |
827 | ||
828 | /* already have the inode for our journal */ | |
829 | inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, | |
830 | osb->slot_num); | |
831 | if (inode == NULL) { | |
832 | status = -EACCES; | |
833 | mlog_errno(status); | |
834 | goto done; | |
835 | } | |
836 | if (is_bad_inode(inode)) { | |
837 | mlog(ML_ERROR, "access error (bad inode)\n"); | |
838 | iput(inode); | |
839 | inode = NULL; | |
840 | status = -EACCES; | |
841 | goto done; | |
842 | } | |
843 | ||
844 | SET_INODE_JOURNAL(inode); | |
845 | OCFS2_I(inode)->ip_open_count++; | |
846 | ||
6eff5790 MF |
847 | /* Skip recovery waits here - journal inode metadata never |
848 | * changes in a live cluster so it can be considered an | |
849 | * exception to the rule. */ | |
e63aecb6 | 850 | status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY); |
ccd979bd MF |
851 | if (status < 0) { |
852 | if (status != -ERESTARTSYS) | |
853 | mlog(ML_ERROR, "Could not get lock on journal!\n"); | |
854 | goto done; | |
855 | } | |
856 | ||
e63aecb6 | 857 | inode_lock = 1; |
ccd979bd MF |
858 | di = (struct ocfs2_dinode *)bh->b_data; |
859 | ||
f17c20dd | 860 | if (i_size_read(inode) < OCFS2_MIN_JOURNAL_SIZE) { |
ccd979bd | 861 | mlog(ML_ERROR, "Journal file size (%lld) is too small!\n", |
f17c20dd | 862 | i_size_read(inode)); |
ccd979bd MF |
863 | status = -EINVAL; |
864 | goto done; | |
865 | } | |
866 | ||
f17c20dd | 867 | trace_ocfs2_journal_init(i_size_read(inode), |
b4107950 TM |
868 | (unsigned long long)inode->i_blocks, |
869 | OCFS2_I(inode)->ip_clusters); | |
ccd979bd MF |
870 | |
871 | /* call the kernels journal init function now */ | |
2b4e30fb | 872 | j_journal = jbd2_journal_init_inode(inode); |
ccd979bd MF |
873 | if (j_journal == NULL) { |
874 | mlog(ML_ERROR, "Linux journal layer error\n"); | |
875 | status = -EINVAL; | |
876 | goto done; | |
877 | } | |
878 | ||
b4107950 | 879 | trace_ocfs2_journal_init_maxlen(j_journal->j_maxlen); |
ccd979bd MF |
880 | |
881 | *dirty = (le32_to_cpu(di->id1.journal1.ij_flags) & | |
882 | OCFS2_JOURNAL_DIRTY_FL); | |
883 | ||
884 | journal->j_journal = j_journal; | |
885 | journal->j_inode = inode; | |
886 | journal->j_bh = bh; | |
887 | ||
888 | ocfs2_set_journal_params(osb); | |
889 | ||
890 | journal->j_state = OCFS2_JOURNAL_LOADED; | |
891 | ||
892 | status = 0; | |
893 | done: | |
894 | if (status < 0) { | |
e63aecb6 MF |
895 | if (inode_lock) |
896 | ocfs2_inode_unlock(inode, 1); | |
a81cb88b | 897 | brelse(bh); |
ccd979bd MF |
898 | if (inode) { |
899 | OCFS2_I(inode)->ip_open_count--; | |
900 | iput(inode); | |
901 | } | |
902 | } | |
903 | ||
ccd979bd MF |
904 | return status; |
905 | } | |
906 | ||
539d8264 SM |
907 | static void ocfs2_bump_recovery_generation(struct ocfs2_dinode *di) |
908 | { | |
909 | le32_add_cpu(&(di->id1.journal1.ij_recovery_generation), 1); | |
910 | } | |
911 | ||
912 | static u32 ocfs2_get_recovery_generation(struct ocfs2_dinode *di) | |
913 | { | |
914 | return le32_to_cpu(di->id1.journal1.ij_recovery_generation); | |
915 | } | |
916 | ||
ccd979bd | 917 | static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb, |
539d8264 | 918 | int dirty, int replayed) |
ccd979bd MF |
919 | { |
920 | int status; | |
921 | unsigned int flags; | |
922 | struct ocfs2_journal *journal = osb->journal; | |
923 | struct buffer_head *bh = journal->j_bh; | |
924 | struct ocfs2_dinode *fe; | |
925 | ||
ccd979bd | 926 | fe = (struct ocfs2_dinode *)bh->b_data; |
10995aa2 JB |
927 | |
928 | /* The journal bh on the osb always comes from ocfs2_journal_init() | |
929 | * and was validated there inside ocfs2_inode_lock_full(). It's a | |
930 | * code bug if we mess it up. */ | |
931 | BUG_ON(!OCFS2_IS_VALID_DINODE(fe)); | |
ccd979bd MF |
932 | |
933 | flags = le32_to_cpu(fe->id1.journal1.ij_flags); | |
934 | if (dirty) | |
935 | flags |= OCFS2_JOURNAL_DIRTY_FL; | |
936 | else | |
937 | flags &= ~OCFS2_JOURNAL_DIRTY_FL; | |
938 | fe->id1.journal1.ij_flags = cpu_to_le32(flags); | |
939 | ||
539d8264 SM |
940 | if (replayed) |
941 | ocfs2_bump_recovery_generation(fe); | |
942 | ||
13723d00 | 943 | ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check); |
8cb471e8 | 944 | status = ocfs2_write_block(osb, bh, INODE_CACHE(journal->j_inode)); |
ccd979bd MF |
945 | if (status < 0) |
946 | mlog_errno(status); | |
947 | ||
ccd979bd MF |
948 | return status; |
949 | } | |
950 | ||
951 | /* | |
952 | * If the journal has been kmalloc'd it needs to be freed after this | |
953 | * call. | |
954 | */ | |
955 | void ocfs2_journal_shutdown(struct ocfs2_super *osb) | |
956 | { | |
957 | struct ocfs2_journal *journal = NULL; | |
958 | int status = 0; | |
959 | struct inode *inode = NULL; | |
960 | int num_running_trans = 0; | |
961 | ||
ebdec83b | 962 | BUG_ON(!osb); |
ccd979bd MF |
963 | |
964 | journal = osb->journal; | |
965 | if (!journal) | |
966 | goto done; | |
967 | ||
968 | inode = journal->j_inode; | |
969 | ||
970 | if (journal->j_state != OCFS2_JOURNAL_LOADED) | |
971 | goto done; | |
972 | ||
2b4e30fb | 973 | /* need to inc inode use count - jbd2_journal_destroy will iput. */ |
ccd979bd MF |
974 | if (!igrab(inode)) |
975 | BUG(); | |
976 | ||
977 | num_running_trans = atomic_read(&(osb->journal->j_num_trans)); | |
b4107950 | 978 | trace_ocfs2_journal_shutdown(num_running_trans); |
ccd979bd MF |
979 | |
980 | /* Do a commit_cache here. It will flush our journal, *and* | |
981 | * release any locks that are still held. | |
982 | * set the SHUTDOWN flag and release the trans lock. | |
983 | * the commit thread will take the trans lock for us below. */ | |
984 | journal->j_state = OCFS2_JOURNAL_IN_SHUTDOWN; | |
985 | ||
986 | /* The OCFS2_JOURNAL_IN_SHUTDOWN will signal to commit_cache to not | |
987 | * drop the trans_lock (which we want to hold until we | |
988 | * completely destroy the journal. */ | |
989 | if (osb->commit_task) { | |
990 | /* Wait for the commit thread */ | |
b4107950 | 991 | trace_ocfs2_journal_shutdown_wait(osb->commit_task); |
ccd979bd MF |
992 | kthread_stop(osb->commit_task); |
993 | osb->commit_task = NULL; | |
994 | } | |
995 | ||
996 | BUG_ON(atomic_read(&(osb->journal->j_num_trans)) != 0); | |
997 | ||
c271c5c2 | 998 | if (ocfs2_mount_local(osb)) { |
2b4e30fb JB |
999 | jbd2_journal_lock_updates(journal->j_journal); |
1000 | status = jbd2_journal_flush(journal->j_journal); | |
1001 | jbd2_journal_unlock_updates(journal->j_journal); | |
c271c5c2 SM |
1002 | if (status < 0) |
1003 | mlog_errno(status); | |
1004 | } | |
1005 | ||
d85400af JB |
1006 | /* Shutdown the kernel journal system */ |
1007 | if (!jbd2_journal_destroy(journal->j_journal) && !status) { | |
c271c5c2 SM |
1008 | /* |
1009 | * Do not toggle if flush was unsuccessful otherwise | |
1010 | * will leave dirty metadata in a "clean" journal | |
1011 | */ | |
539d8264 | 1012 | status = ocfs2_journal_toggle_dirty(osb, 0, 0); |
c271c5c2 SM |
1013 | if (status < 0) |
1014 | mlog_errno(status); | |
1015 | } | |
ae0dff68 | 1016 | journal->j_journal = NULL; |
ccd979bd MF |
1017 | |
1018 | OCFS2_I(inode)->ip_open_count--; | |
1019 | ||
1020 | /* unlock our journal */ | |
e63aecb6 | 1021 | ocfs2_inode_unlock(inode, 1); |
ccd979bd MF |
1022 | |
1023 | brelse(journal->j_bh); | |
1024 | journal->j_bh = NULL; | |
1025 | ||
1026 | journal->j_state = OCFS2_JOURNAL_FREE; | |
1027 | ||
1028 | // up_write(&journal->j_trans_barrier); | |
1029 | done: | |
72865d92 | 1030 | iput(inode); |
ccd979bd MF |
1031 | } |
1032 | ||
1033 | static void ocfs2_clear_journal_error(struct super_block *sb, | |
1034 | journal_t *journal, | |
1035 | int slot) | |
1036 | { | |
1037 | int olderr; | |
1038 | ||
2b4e30fb | 1039 | olderr = jbd2_journal_errno(journal); |
ccd979bd MF |
1040 | if (olderr) { |
1041 | mlog(ML_ERROR, "File system error %d recorded in " | |
1042 | "journal %u.\n", olderr, slot); | |
1043 | mlog(ML_ERROR, "File system on device %s needs checking.\n", | |
1044 | sb->s_id); | |
1045 | ||
2b4e30fb JB |
1046 | jbd2_journal_ack_err(journal); |
1047 | jbd2_journal_clear_err(journal); | |
ccd979bd MF |
1048 | } |
1049 | } | |
1050 | ||
539d8264 | 1051 | int ocfs2_journal_load(struct ocfs2_journal *journal, int local, int replayed) |
ccd979bd MF |
1052 | { |
1053 | int status = 0; | |
1054 | struct ocfs2_super *osb; | |
1055 | ||
b1f3550f | 1056 | BUG_ON(!journal); |
ccd979bd MF |
1057 | |
1058 | osb = journal->j_osb; | |
1059 | ||
2b4e30fb | 1060 | status = jbd2_journal_load(journal->j_journal); |
ccd979bd MF |
1061 | if (status < 0) { |
1062 | mlog(ML_ERROR, "Failed to load journal!\n"); | |
1063 | goto done; | |
1064 | } | |
1065 | ||
1066 | ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num); | |
1067 | ||
539d8264 | 1068 | status = ocfs2_journal_toggle_dirty(osb, 1, replayed); |
ccd979bd MF |
1069 | if (status < 0) { |
1070 | mlog_errno(status); | |
1071 | goto done; | |
1072 | } | |
1073 | ||
1074 | /* Launch the commit thread */ | |
c271c5c2 SM |
1075 | if (!local) { |
1076 | osb->commit_task = kthread_run(ocfs2_commit_thread, osb, | |
5afc44e2 | 1077 | "ocfs2cmt-%s", osb->uuid_str); |
c271c5c2 SM |
1078 | if (IS_ERR(osb->commit_task)) { |
1079 | status = PTR_ERR(osb->commit_task); | |
1080 | osb->commit_task = NULL; | |
1081 | mlog(ML_ERROR, "unable to launch ocfs2commit thread, " | |
1082 | "error=%d", status); | |
1083 | goto done; | |
1084 | } | |
1085 | } else | |
ccd979bd | 1086 | osb->commit_task = NULL; |
ccd979bd MF |
1087 | |
1088 | done: | |
ccd979bd MF |
1089 | return status; |
1090 | } | |
1091 | ||
1092 | ||
1093 | /* 'full' flag tells us whether we clear out all blocks or if we just | |
1094 | * mark the journal clean */ | |
1095 | int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full) | |
1096 | { | |
1097 | int status; | |
1098 | ||
ebdec83b | 1099 | BUG_ON(!journal); |
ccd979bd | 1100 | |
2b4e30fb | 1101 | status = jbd2_journal_wipe(journal->j_journal, full); |
ccd979bd MF |
1102 | if (status < 0) { |
1103 | mlog_errno(status); | |
1104 | goto bail; | |
1105 | } | |
1106 | ||
539d8264 | 1107 | status = ocfs2_journal_toggle_dirty(journal->j_osb, 0, 0); |
ccd979bd MF |
1108 | if (status < 0) |
1109 | mlog_errno(status); | |
1110 | ||
1111 | bail: | |
ccd979bd MF |
1112 | return status; |
1113 | } | |
1114 | ||
553abd04 JB |
1115 | static int ocfs2_recovery_completed(struct ocfs2_super *osb) |
1116 | { | |
1117 | int empty; | |
1118 | struct ocfs2_recovery_map *rm = osb->recovery_map; | |
1119 | ||
1120 | spin_lock(&osb->osb_lock); | |
1121 | empty = (rm->rm_used == 0); | |
1122 | spin_unlock(&osb->osb_lock); | |
1123 | ||
1124 | return empty; | |
1125 | } | |
1126 | ||
1127 | void ocfs2_wait_for_recovery(struct ocfs2_super *osb) | |
1128 | { | |
1129 | wait_event(osb->recovery_event, ocfs2_recovery_completed(osb)); | |
1130 | } | |
1131 | ||
ccd979bd MF |
1132 | /* |
1133 | * JBD Might read a cached version of another nodes journal file. We | |
1134 | * don't want this as this file changes often and we get no | |
1135 | * notification on those changes. The only way to be sure that we've | |
1136 | * got the most up to date version of those blocks then is to force | |
1137 | * read them off disk. Just searching through the buffer cache won't | |
1138 | * work as there may be pages backing this file which are still marked | |
1139 | * up to date. We know things can't change on this file underneath us | |
1140 | * as we have the lock by now :) | |
1141 | */ | |
1142 | static int ocfs2_force_read_journal(struct inode *inode) | |
1143 | { | |
1144 | int status = 0; | |
4f902c37 | 1145 | int i; |
8110b073 | 1146 | u64 v_blkno, p_blkno, p_blocks, num_blocks; |
0b492f68 JB |
1147 | struct buffer_head *bh = NULL; |
1148 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
ccd979bd | 1149 | |
f17c20dd | 1150 | num_blocks = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode)); |
ccd979bd | 1151 | v_blkno = 0; |
8110b073 | 1152 | while (v_blkno < num_blocks) { |
ccd979bd | 1153 | status = ocfs2_extent_map_get_blocks(inode, v_blkno, |
49cb8d2d | 1154 | &p_blkno, &p_blocks, NULL); |
ccd979bd MF |
1155 | if (status < 0) { |
1156 | mlog_errno(status); | |
1157 | goto bail; | |
1158 | } | |
1159 | ||
0b492f68 JB |
1160 | for (i = 0; i < p_blocks; i++, p_blkno++) { |
1161 | bh = __find_get_block(osb->sb->s_bdev, p_blkno, | |
1162 | osb->sb->s_blocksize); | |
1163 | /* block not cached. */ | |
1164 | if (!bh) | |
1165 | continue; | |
1166 | ||
1167 | brelse(bh); | |
1168 | bh = NULL; | |
1169 | /* We are reading journal data which should not | |
1170 | * be put in the uptodate cache. | |
1171 | */ | |
1172 | status = ocfs2_read_blocks_sync(osb, p_blkno, 1, &bh); | |
1173 | if (status < 0) { | |
1174 | mlog_errno(status); | |
1175 | goto bail; | |
1176 | } | |
1177 | ||
1178 | brelse(bh); | |
1179 | bh = NULL; | |
ccd979bd MF |
1180 | } |
1181 | ||
1182 | v_blkno += p_blocks; | |
1183 | } | |
1184 | ||
1185 | bail: | |
ccd979bd MF |
1186 | return status; |
1187 | } | |
1188 | ||
1189 | struct ocfs2_la_recovery_item { | |
1190 | struct list_head lri_list; | |
1191 | int lri_slot; | |
1192 | struct ocfs2_dinode *lri_la_dinode; | |
1193 | struct ocfs2_dinode *lri_tl_dinode; | |
2205363d | 1194 | struct ocfs2_quota_recovery *lri_qrec; |
ed460cff | 1195 | enum ocfs2_orphan_reco_type lri_orphan_reco_type; |
ccd979bd MF |
1196 | }; |
1197 | ||
1198 | /* Does the second half of the recovery process. By this point, the | |
1199 | * node is marked clean and can actually be considered recovered, | |
1200 | * hence it's no longer in the recovery map, but there's still some | |
1201 | * cleanup we can do which shouldn't happen within the recovery thread | |
1202 | * as locking in that context becomes very difficult if we are to take | |
1203 | * recovering nodes into account. | |
1204 | * | |
1205 | * NOTE: This function can and will sleep on recovery of other nodes | |
1206 | * during cluster locking, just like any other ocfs2 process. | |
1207 | */ | |
c4028958 | 1208 | void ocfs2_complete_recovery(struct work_struct *work) |
ccd979bd | 1209 | { |
b4107950 | 1210 | int ret = 0; |
c4028958 DH |
1211 | struct ocfs2_journal *journal = |
1212 | container_of(work, struct ocfs2_journal, j_recovery_work); | |
1213 | struct ocfs2_super *osb = journal->j_osb; | |
ccd979bd | 1214 | struct ocfs2_dinode *la_dinode, *tl_dinode; |
800deef3 | 1215 | struct ocfs2_la_recovery_item *item, *n; |
2205363d | 1216 | struct ocfs2_quota_recovery *qrec; |
ed460cff | 1217 | enum ocfs2_orphan_reco_type orphan_reco_type; |
ccd979bd MF |
1218 | LIST_HEAD(tmp_la_list); |
1219 | ||
b4107950 TM |
1220 | trace_ocfs2_complete_recovery( |
1221 | (unsigned long long)OCFS2_I(journal->j_inode)->ip_blkno); | |
ccd979bd MF |
1222 | |
1223 | spin_lock(&journal->j_lock); | |
1224 | list_splice_init(&journal->j_la_cleanups, &tmp_la_list); | |
1225 | spin_unlock(&journal->j_lock); | |
1226 | ||
800deef3 | 1227 | list_for_each_entry_safe(item, n, &tmp_la_list, lri_list) { |
ccd979bd MF |
1228 | list_del_init(&item->lri_list); |
1229 | ||
19ece546 JK |
1230 | ocfs2_wait_on_quotas(osb); |
1231 | ||
ccd979bd | 1232 | la_dinode = item->lri_la_dinode; |
b4107950 TM |
1233 | tl_dinode = item->lri_tl_dinode; |
1234 | qrec = item->lri_qrec; | |
ed460cff | 1235 | orphan_reco_type = item->lri_orphan_reco_type; |
ccd979bd | 1236 | |
b4107950 TM |
1237 | trace_ocfs2_complete_recovery_slot(item->lri_slot, |
1238 | la_dinode ? le64_to_cpu(la_dinode->i_blkno) : 0, | |
1239 | tl_dinode ? le64_to_cpu(tl_dinode->i_blkno) : 0, | |
1240 | qrec); | |
1241 | ||
1242 | if (la_dinode) { | |
ccd979bd MF |
1243 | ret = ocfs2_complete_local_alloc_recovery(osb, |
1244 | la_dinode); | |
1245 | if (ret < 0) | |
1246 | mlog_errno(ret); | |
1247 | ||
1248 | kfree(la_dinode); | |
1249 | } | |
1250 | ||
ccd979bd | 1251 | if (tl_dinode) { |
ccd979bd MF |
1252 | ret = ocfs2_complete_truncate_log_recovery(osb, |
1253 | tl_dinode); | |
1254 | if (ret < 0) | |
1255 | mlog_errno(ret); | |
1256 | ||
1257 | kfree(tl_dinode); | |
1258 | } | |
1259 | ||
ed460cff JQ |
1260 | ret = ocfs2_recover_orphans(osb, item->lri_slot, |
1261 | orphan_reco_type); | |
ccd979bd MF |
1262 | if (ret < 0) |
1263 | mlog_errno(ret); | |
1264 | ||
2205363d | 1265 | if (qrec) { |
2205363d JK |
1266 | ret = ocfs2_finish_quota_recovery(osb, qrec, |
1267 | item->lri_slot); | |
1268 | if (ret < 0) | |
1269 | mlog_errno(ret); | |
1270 | /* Recovery info is already freed now */ | |
1271 | } | |
1272 | ||
ccd979bd MF |
1273 | kfree(item); |
1274 | } | |
1275 | ||
b4107950 | 1276 | trace_ocfs2_complete_recovery_end(ret); |
ccd979bd MF |
1277 | } |
1278 | ||
1279 | /* NOTE: This function always eats your references to la_dinode and | |
1280 | * tl_dinode, either manually on error, or by passing them to | |
1281 | * ocfs2_complete_recovery */ | |
1282 | static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal, | |
1283 | int slot_num, | |
1284 | struct ocfs2_dinode *la_dinode, | |
2205363d | 1285 | struct ocfs2_dinode *tl_dinode, |
ed460cff JQ |
1286 | struct ocfs2_quota_recovery *qrec, |
1287 | enum ocfs2_orphan_reco_type orphan_reco_type) | |
ccd979bd MF |
1288 | { |
1289 | struct ocfs2_la_recovery_item *item; | |
1290 | ||
afae00ab | 1291 | item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS); |
ccd979bd MF |
1292 | if (!item) { |
1293 | /* Though we wish to avoid it, we are in fact safe in | |
1294 | * skipping local alloc cleanup as fsck.ocfs2 is more | |
1295 | * than capable of reclaiming unused space. */ | |
d787ab09 TG |
1296 | kfree(la_dinode); |
1297 | kfree(tl_dinode); | |
ccd979bd | 1298 | |
2205363d JK |
1299 | if (qrec) |
1300 | ocfs2_free_quota_recovery(qrec); | |
1301 | ||
ccd979bd MF |
1302 | mlog_errno(-ENOMEM); |
1303 | return; | |
1304 | } | |
1305 | ||
1306 | INIT_LIST_HEAD(&item->lri_list); | |
1307 | item->lri_la_dinode = la_dinode; | |
1308 | item->lri_slot = slot_num; | |
1309 | item->lri_tl_dinode = tl_dinode; | |
2205363d | 1310 | item->lri_qrec = qrec; |
ed460cff | 1311 | item->lri_orphan_reco_type = orphan_reco_type; |
ccd979bd MF |
1312 | |
1313 | spin_lock(&journal->j_lock); | |
1314 | list_add_tail(&item->lri_list, &journal->j_la_cleanups); | |
35ddf78e | 1315 | queue_work(journal->j_osb->ocfs2_wq, &journal->j_recovery_work); |
ccd979bd MF |
1316 | spin_unlock(&journal->j_lock); |
1317 | } | |
1318 | ||
1319 | /* Called by the mount code to queue recovery the last part of | |
9140db04 | 1320 | * recovery for it's own and offline slot(s). */ |
ccd979bd MF |
1321 | void ocfs2_complete_mount_recovery(struct ocfs2_super *osb) |
1322 | { | |
1323 | struct ocfs2_journal *journal = osb->journal; | |
1324 | ||
10b3dd76 SM |
1325 | if (ocfs2_is_hard_readonly(osb)) |
1326 | return; | |
1327 | ||
9140db04 SE |
1328 | /* No need to queue up our truncate_log as regular cleanup will catch |
1329 | * that */ | |
1330 | ocfs2_queue_recovery_completion(journal, osb->slot_num, | |
ed460cff JQ |
1331 | osb->local_alloc_copy, NULL, NULL, |
1332 | ORPHAN_NEED_TRUNCATE); | |
9140db04 | 1333 | ocfs2_schedule_truncate_log_flush(osb, 0); |
ccd979bd | 1334 | |
9140db04 | 1335 | osb->local_alloc_copy = NULL; |
9140db04 SE |
1336 | |
1337 | /* queue to recover orphan slots for all offline slots */ | |
1338 | ocfs2_replay_map_set_state(osb, REPLAY_NEEDED); | |
ed460cff | 1339 | ocfs2_queue_replay_slots(osb, ORPHAN_NEED_TRUNCATE); |
9140db04 | 1340 | ocfs2_free_replay_slots(osb); |
ccd979bd MF |
1341 | } |
1342 | ||
2205363d JK |
1343 | void ocfs2_complete_quota_recovery(struct ocfs2_super *osb) |
1344 | { | |
1345 | if (osb->quota_rec) { | |
1346 | ocfs2_queue_recovery_completion(osb->journal, | |
1347 | osb->slot_num, | |
1348 | NULL, | |
1349 | NULL, | |
ed460cff JQ |
1350 | osb->quota_rec, |
1351 | ORPHAN_NEED_TRUNCATE); | |
2205363d JK |
1352 | osb->quota_rec = NULL; |
1353 | } | |
1354 | } | |
1355 | ||
ccd979bd MF |
1356 | static int __ocfs2_recovery_thread(void *arg) |
1357 | { | |
2205363d | 1358 | int status, node_num, slot_num; |
ccd979bd | 1359 | struct ocfs2_super *osb = arg; |
553abd04 | 1360 | struct ocfs2_recovery_map *rm = osb->recovery_map; |
2205363d JK |
1361 | int *rm_quota = NULL; |
1362 | int rm_quota_used = 0, i; | |
1363 | struct ocfs2_quota_recovery *qrec; | |
ccd979bd | 1364 | |
21158ca8 G |
1365 | /* Whether the quota supported. */ |
1366 | int quota_enabled = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, | |
1367 | OCFS2_FEATURE_RO_COMPAT_USRQUOTA) | |
1368 | || OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, | |
1369 | OCFS2_FEATURE_RO_COMPAT_GRPQUOTA); | |
1370 | ||
ccd979bd MF |
1371 | status = ocfs2_wait_on_mount(osb); |
1372 | if (status < 0) { | |
1373 | goto bail; | |
1374 | } | |
1375 | ||
21158ca8 G |
1376 | if (quota_enabled) { |
1377 | rm_quota = kcalloc(osb->max_slots, sizeof(int), GFP_NOFS); | |
1378 | if (!rm_quota) { | |
1379 | status = -ENOMEM; | |
1380 | goto bail; | |
1381 | } | |
2205363d | 1382 | } |
ccd979bd MF |
1383 | restart: |
1384 | status = ocfs2_super_lock(osb, 1); | |
1385 | if (status < 0) { | |
1386 | mlog_errno(status); | |
1387 | goto bail; | |
1388 | } | |
1389 | ||
9140db04 SE |
1390 | status = ocfs2_compute_replay_slots(osb); |
1391 | if (status < 0) | |
1392 | mlog_errno(status); | |
1393 | ||
1394 | /* queue recovery for our own slot */ | |
1395 | ocfs2_queue_recovery_completion(osb->journal, osb->slot_num, NULL, | |
ed460cff | 1396 | NULL, NULL, ORPHAN_NO_NEED_TRUNCATE); |
9140db04 | 1397 | |
553abd04 JB |
1398 | spin_lock(&osb->osb_lock); |
1399 | while (rm->rm_used) { | |
1400 | /* It's always safe to remove entry zero, as we won't | |
1401 | * clear it until ocfs2_recover_node() has succeeded. */ | |
1402 | node_num = rm->rm_entries[0]; | |
1403 | spin_unlock(&osb->osb_lock); | |
2205363d | 1404 | slot_num = ocfs2_node_num_to_slot(osb, node_num); |
b4107950 | 1405 | trace_ocfs2_recovery_thread_node(node_num, slot_num); |
2205363d JK |
1406 | if (slot_num == -ENOENT) { |
1407 | status = 0; | |
2205363d JK |
1408 | goto skip_recovery; |
1409 | } | |
2205363d JK |
1410 | |
1411 | /* It is a bit subtle with quota recovery. We cannot do it | |
1412 | * immediately because we have to obtain cluster locks from | |
1413 | * quota files and we also don't want to just skip it because | |
1414 | * then quota usage would be out of sync until some node takes | |
1415 | * the slot. So we remember which nodes need quota recovery | |
1416 | * and when everything else is done, we recover quotas. */ | |
21158ca8 G |
1417 | if (quota_enabled) { |
1418 | for (i = 0; i < rm_quota_used | |
1419 | && rm_quota[i] != slot_num; i++) | |
1420 | ; | |
1421 | ||
1422 | if (i == rm_quota_used) | |
1423 | rm_quota[rm_quota_used++] = slot_num; | |
1424 | } | |
2205363d JK |
1425 | |
1426 | status = ocfs2_recover_node(osb, node_num, slot_num); | |
1427 | skip_recovery: | |
553abd04 JB |
1428 | if (!status) { |
1429 | ocfs2_recovery_map_clear(osb, node_num); | |
1430 | } else { | |
ccd979bd MF |
1431 | mlog(ML_ERROR, |
1432 | "Error %d recovering node %d on device (%u,%u)!\n", | |
1433 | status, node_num, | |
1434 | MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev)); | |
1435 | mlog(ML_ERROR, "Volume requires unmount.\n"); | |
ccd979bd MF |
1436 | } |
1437 | ||
553abd04 | 1438 | spin_lock(&osb->osb_lock); |
ccd979bd | 1439 | } |
553abd04 | 1440 | spin_unlock(&osb->osb_lock); |
b4107950 | 1441 | trace_ocfs2_recovery_thread_end(status); |
553abd04 | 1442 | |
539d8264 SM |
1443 | /* Refresh all journal recovery generations from disk */ |
1444 | status = ocfs2_check_journals_nolocks(osb); | |
1445 | status = (status == -EROFS) ? 0 : status; | |
1446 | if (status < 0) | |
1447 | mlog_errno(status); | |
1448 | ||
2205363d | 1449 | /* Now it is right time to recover quotas... We have to do this under |
25985edc | 1450 | * superblock lock so that no one can start using the slot (and crash) |
2205363d | 1451 | * before we recover it */ |
21158ca8 G |
1452 | if (quota_enabled) { |
1453 | for (i = 0; i < rm_quota_used; i++) { | |
1454 | qrec = ocfs2_begin_quota_recovery(osb, rm_quota[i]); | |
1455 | if (IS_ERR(qrec)) { | |
1456 | status = PTR_ERR(qrec); | |
1457 | mlog_errno(status); | |
1458 | continue; | |
1459 | } | |
1460 | ocfs2_queue_recovery_completion(osb->journal, | |
1461 | rm_quota[i], | |
1462 | NULL, NULL, qrec, | |
1463 | ORPHAN_NEED_TRUNCATE); | |
2205363d | 1464 | } |
2205363d JK |
1465 | } |
1466 | ||
ccd979bd MF |
1467 | ocfs2_super_unlock(osb, 1); |
1468 | ||
9140db04 | 1469 | /* queue recovery for offline slots */ |
ed460cff | 1470 | ocfs2_queue_replay_slots(osb, ORPHAN_NEED_TRUNCATE); |
ccd979bd MF |
1471 | |
1472 | bail: | |
c74ec2f7 | 1473 | mutex_lock(&osb->recovery_lock); |
553abd04 | 1474 | if (!status && !ocfs2_recovery_completed(osb)) { |
c74ec2f7 | 1475 | mutex_unlock(&osb->recovery_lock); |
ccd979bd MF |
1476 | goto restart; |
1477 | } | |
1478 | ||
9140db04 | 1479 | ocfs2_free_replay_slots(osb); |
ccd979bd MF |
1480 | osb->recovery_thread_task = NULL; |
1481 | mb(); /* sync with ocfs2_recovery_thread_running */ | |
1482 | wake_up(&osb->recovery_event); | |
1483 | ||
c74ec2f7 | 1484 | mutex_unlock(&osb->recovery_lock); |
ccd979bd | 1485 | |
21158ca8 G |
1486 | if (quota_enabled) |
1487 | kfree(rm_quota); | |
2205363d | 1488 | |
ccd979bd MF |
1489 | /* no one is callint kthread_stop() for us so the kthread() api |
1490 | * requires that we call do_exit(). And it isn't exported, but | |
1491 | * complete_and_exit() seems to be a minimal wrapper around it. */ | |
1492 | complete_and_exit(NULL, status); | |
ccd979bd MF |
1493 | } |
1494 | ||
1495 | void ocfs2_recovery_thread(struct ocfs2_super *osb, int node_num) | |
1496 | { | |
c74ec2f7 | 1497 | mutex_lock(&osb->recovery_lock); |
ccd979bd | 1498 | |
b4107950 TM |
1499 | trace_ocfs2_recovery_thread(node_num, osb->node_num, |
1500 | osb->disable_recovery, osb->recovery_thread_task, | |
1501 | osb->disable_recovery ? | |
1502 | -1 : ocfs2_recovery_map_set(osb, node_num)); | |
ccd979bd | 1503 | |
b4107950 TM |
1504 | if (osb->disable_recovery) |
1505 | goto out; | |
ccd979bd MF |
1506 | |
1507 | if (osb->recovery_thread_task) | |
1508 | goto out; | |
1509 | ||
1510 | osb->recovery_thread_task = kthread_run(__ocfs2_recovery_thread, osb, | |
5afc44e2 | 1511 | "ocfs2rec-%s", osb->uuid_str); |
ccd979bd MF |
1512 | if (IS_ERR(osb->recovery_thread_task)) { |
1513 | mlog_errno((int)PTR_ERR(osb->recovery_thread_task)); | |
1514 | osb->recovery_thread_task = NULL; | |
1515 | } | |
1516 | ||
1517 | out: | |
c74ec2f7 | 1518 | mutex_unlock(&osb->recovery_lock); |
ccd979bd | 1519 | wake_up(&osb->recovery_event); |
ccd979bd MF |
1520 | } |
1521 | ||
539d8264 SM |
1522 | static int ocfs2_read_journal_inode(struct ocfs2_super *osb, |
1523 | int slot_num, | |
1524 | struct buffer_head **bh, | |
1525 | struct inode **ret_inode) | |
1526 | { | |
1527 | int status = -EACCES; | |
1528 | struct inode *inode = NULL; | |
1529 | ||
1530 | BUG_ON(slot_num >= osb->max_slots); | |
1531 | ||
1532 | inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, | |
1533 | slot_num); | |
1534 | if (!inode || is_bad_inode(inode)) { | |
1535 | mlog_errno(status); | |
1536 | goto bail; | |
1537 | } | |
1538 | SET_INODE_JOURNAL(inode); | |
1539 | ||
b657c95c | 1540 | status = ocfs2_read_inode_block_full(inode, bh, OCFS2_BH_IGNORE_CACHE); |
539d8264 SM |
1541 | if (status < 0) { |
1542 | mlog_errno(status); | |
1543 | goto bail; | |
1544 | } | |
1545 | ||
1546 | status = 0; | |
1547 | ||
1548 | bail: | |
1549 | if (inode) { | |
1550 | if (status || !ret_inode) | |
1551 | iput(inode); | |
1552 | else | |
1553 | *ret_inode = inode; | |
1554 | } | |
1555 | return status; | |
1556 | } | |
1557 | ||
ccd979bd MF |
1558 | /* Does the actual journal replay and marks the journal inode as |
1559 | * clean. Will only replay if the journal inode is marked dirty. */ | |
1560 | static int ocfs2_replay_journal(struct ocfs2_super *osb, | |
1561 | int node_num, | |
1562 | int slot_num) | |
1563 | { | |
1564 | int status; | |
1565 | int got_lock = 0; | |
1566 | unsigned int flags; | |
1567 | struct inode *inode = NULL; | |
1568 | struct ocfs2_dinode *fe; | |
1569 | journal_t *journal = NULL; | |
1570 | struct buffer_head *bh = NULL; | |
539d8264 | 1571 | u32 slot_reco_gen; |
ccd979bd | 1572 | |
539d8264 SM |
1573 | status = ocfs2_read_journal_inode(osb, slot_num, &bh, &inode); |
1574 | if (status) { | |
ccd979bd MF |
1575 | mlog_errno(status); |
1576 | goto done; | |
1577 | } | |
539d8264 SM |
1578 | |
1579 | fe = (struct ocfs2_dinode *)bh->b_data; | |
1580 | slot_reco_gen = ocfs2_get_recovery_generation(fe); | |
1581 | brelse(bh); | |
1582 | bh = NULL; | |
1583 | ||
1584 | /* | |
1585 | * As the fs recovery is asynchronous, there is a small chance that | |
1586 | * another node mounted (and recovered) the slot before the recovery | |
1587 | * thread could get the lock. To handle that, we dirty read the journal | |
1588 | * inode for that slot to get the recovery generation. If it is | |
1589 | * different than what we expected, the slot has been recovered. | |
1590 | * If not, it needs recovery. | |
1591 | */ | |
1592 | if (osb->slot_recovery_generations[slot_num] != slot_reco_gen) { | |
b4107950 | 1593 | trace_ocfs2_replay_journal_recovered(slot_num, |
539d8264 SM |
1594 | osb->slot_recovery_generations[slot_num], slot_reco_gen); |
1595 | osb->slot_recovery_generations[slot_num] = slot_reco_gen; | |
1596 | status = -EBUSY; | |
ccd979bd MF |
1597 | goto done; |
1598 | } | |
539d8264 SM |
1599 | |
1600 | /* Continue with recovery as the journal has not yet been recovered */ | |
ccd979bd | 1601 | |
e63aecb6 | 1602 | status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY); |
ccd979bd | 1603 | if (status < 0) { |
b4107950 | 1604 | trace_ocfs2_replay_journal_lock_err(status); |
ccd979bd MF |
1605 | if (status != -ERESTARTSYS) |
1606 | mlog(ML_ERROR, "Could not lock journal!\n"); | |
1607 | goto done; | |
1608 | } | |
1609 | got_lock = 1; | |
1610 | ||
1611 | fe = (struct ocfs2_dinode *) bh->b_data; | |
1612 | ||
1613 | flags = le32_to_cpu(fe->id1.journal1.ij_flags); | |
539d8264 | 1614 | slot_reco_gen = ocfs2_get_recovery_generation(fe); |
ccd979bd MF |
1615 | |
1616 | if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) { | |
b4107950 | 1617 | trace_ocfs2_replay_journal_skip(node_num); |
539d8264 SM |
1618 | /* Refresh recovery generation for the slot */ |
1619 | osb->slot_recovery_generations[slot_num] = slot_reco_gen; | |
ccd979bd MF |
1620 | goto done; |
1621 | } | |
1622 | ||
9140db04 SE |
1623 | /* we need to run complete recovery for offline orphan slots */ |
1624 | ocfs2_replay_map_set_state(osb, REPLAY_NEEDED); | |
1625 | ||
619c200d SM |
1626 | printk(KERN_NOTICE "ocfs2: Begin replay journal (node %d, slot %d) on "\ |
1627 | "device (%u,%u)\n", node_num, slot_num, MAJOR(osb->sb->s_dev), | |
1628 | MINOR(osb->sb->s_dev)); | |
ccd979bd MF |
1629 | |
1630 | OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters); | |
1631 | ||
1632 | status = ocfs2_force_read_journal(inode); | |
1633 | if (status < 0) { | |
1634 | mlog_errno(status); | |
1635 | goto done; | |
1636 | } | |
1637 | ||
2b4e30fb | 1638 | journal = jbd2_journal_init_inode(inode); |
ccd979bd MF |
1639 | if (journal == NULL) { |
1640 | mlog(ML_ERROR, "Linux journal layer error\n"); | |
1641 | status = -EIO; | |
1642 | goto done; | |
1643 | } | |
1644 | ||
2b4e30fb | 1645 | status = jbd2_journal_load(journal); |
ccd979bd MF |
1646 | if (status < 0) { |
1647 | mlog_errno(status); | |
1648 | if (!igrab(inode)) | |
1649 | BUG(); | |
2b4e30fb | 1650 | jbd2_journal_destroy(journal); |
ccd979bd MF |
1651 | goto done; |
1652 | } | |
1653 | ||
1654 | ocfs2_clear_journal_error(osb->sb, journal, slot_num); | |
1655 | ||
1656 | /* wipe the journal */ | |
2b4e30fb JB |
1657 | jbd2_journal_lock_updates(journal); |
1658 | status = jbd2_journal_flush(journal); | |
1659 | jbd2_journal_unlock_updates(journal); | |
ccd979bd MF |
1660 | if (status < 0) |
1661 | mlog_errno(status); | |
1662 | ||
1663 | /* This will mark the node clean */ | |
1664 | flags = le32_to_cpu(fe->id1.journal1.ij_flags); | |
1665 | flags &= ~OCFS2_JOURNAL_DIRTY_FL; | |
1666 | fe->id1.journal1.ij_flags = cpu_to_le32(flags); | |
1667 | ||
539d8264 SM |
1668 | /* Increment recovery generation to indicate successful recovery */ |
1669 | ocfs2_bump_recovery_generation(fe); | |
1670 | osb->slot_recovery_generations[slot_num] = | |
1671 | ocfs2_get_recovery_generation(fe); | |
1672 | ||
13723d00 | 1673 | ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check); |
8cb471e8 | 1674 | status = ocfs2_write_block(osb, bh, INODE_CACHE(inode)); |
ccd979bd MF |
1675 | if (status < 0) |
1676 | mlog_errno(status); | |
1677 | ||
1678 | if (!igrab(inode)) | |
1679 | BUG(); | |
1680 | ||
2b4e30fb | 1681 | jbd2_journal_destroy(journal); |
ccd979bd | 1682 | |
619c200d SM |
1683 | printk(KERN_NOTICE "ocfs2: End replay journal (node %d, slot %d) on "\ |
1684 | "device (%u,%u)\n", node_num, slot_num, MAJOR(osb->sb->s_dev), | |
1685 | MINOR(osb->sb->s_dev)); | |
ccd979bd MF |
1686 | done: |
1687 | /* drop the lock on this nodes journal */ | |
1688 | if (got_lock) | |
e63aecb6 | 1689 | ocfs2_inode_unlock(inode, 1); |
ccd979bd | 1690 | |
72865d92 | 1691 | iput(inode); |
a81cb88b | 1692 | brelse(bh); |
ccd979bd | 1693 | |
ccd979bd MF |
1694 | return status; |
1695 | } | |
1696 | ||
1697 | /* | |
1698 | * Do the most important parts of node recovery: | |
1699 | * - Replay it's journal | |
1700 | * - Stamp a clean local allocator file | |
1701 | * - Stamp a clean truncate log | |
1702 | * - Mark the node clean | |
1703 | * | |
1704 | * If this function completes without error, a node in OCFS2 can be | |
1705 | * said to have been safely recovered. As a result, failure during the | |
1706 | * second part of a nodes recovery process (local alloc recovery) is | |
1707 | * far less concerning. | |
1708 | */ | |
1709 | static int ocfs2_recover_node(struct ocfs2_super *osb, | |
2205363d | 1710 | int node_num, int slot_num) |
ccd979bd MF |
1711 | { |
1712 | int status = 0; | |
ccd979bd MF |
1713 | struct ocfs2_dinode *la_copy = NULL; |
1714 | struct ocfs2_dinode *tl_copy = NULL; | |
1715 | ||
b4107950 | 1716 | trace_ocfs2_recover_node(node_num, slot_num, osb->node_num); |
ccd979bd MF |
1717 | |
1718 | /* Should not ever be called to recover ourselves -- in that | |
1719 | * case we should've called ocfs2_journal_load instead. */ | |
ebdec83b | 1720 | BUG_ON(osb->node_num == node_num); |
ccd979bd | 1721 | |
ccd979bd MF |
1722 | status = ocfs2_replay_journal(osb, node_num, slot_num); |
1723 | if (status < 0) { | |
539d8264 | 1724 | if (status == -EBUSY) { |
b4107950 | 1725 | trace_ocfs2_recover_node_skip(slot_num, node_num); |
539d8264 SM |
1726 | status = 0; |
1727 | goto done; | |
1728 | } | |
ccd979bd MF |
1729 | mlog_errno(status); |
1730 | goto done; | |
1731 | } | |
1732 | ||
1733 | /* Stamp a clean local alloc file AFTER recovering the journal... */ | |
1734 | status = ocfs2_begin_local_alloc_recovery(osb, slot_num, &la_copy); | |
1735 | if (status < 0) { | |
1736 | mlog_errno(status); | |
1737 | goto done; | |
1738 | } | |
1739 | ||
1740 | /* An error from begin_truncate_log_recovery is not | |
1741 | * serious enough to warrant halting the rest of | |
1742 | * recovery. */ | |
1743 | status = ocfs2_begin_truncate_log_recovery(osb, slot_num, &tl_copy); | |
1744 | if (status < 0) | |
1745 | mlog_errno(status); | |
1746 | ||
1747 | /* Likewise, this would be a strange but ultimately not so | |
1748 | * harmful place to get an error... */ | |
8e8a4603 | 1749 | status = ocfs2_clear_slot(osb, slot_num); |
ccd979bd MF |
1750 | if (status < 0) |
1751 | mlog_errno(status); | |
1752 | ||
1753 | /* This will kfree the memory pointed to by la_copy and tl_copy */ | |
1754 | ocfs2_queue_recovery_completion(osb->journal, slot_num, la_copy, | |
ed460cff | 1755 | tl_copy, NULL, ORPHAN_NEED_TRUNCATE); |
ccd979bd MF |
1756 | |
1757 | status = 0; | |
1758 | done: | |
1759 | ||
ccd979bd MF |
1760 | return status; |
1761 | } | |
1762 | ||
1763 | /* Test node liveness by trylocking his journal. If we get the lock, | |
1764 | * we drop it here. Return 0 if we got the lock, -EAGAIN if node is | |
1765 | * still alive (we couldn't get the lock) and < 0 on error. */ | |
1766 | static int ocfs2_trylock_journal(struct ocfs2_super *osb, | |
1767 | int slot_num) | |
1768 | { | |
1769 | int status, flags; | |
1770 | struct inode *inode = NULL; | |
1771 | ||
1772 | inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, | |
1773 | slot_num); | |
1774 | if (inode == NULL) { | |
1775 | mlog(ML_ERROR, "access error\n"); | |
1776 | status = -EACCES; | |
1777 | goto bail; | |
1778 | } | |
1779 | if (is_bad_inode(inode)) { | |
1780 | mlog(ML_ERROR, "access error (bad inode)\n"); | |
1781 | iput(inode); | |
1782 | inode = NULL; | |
1783 | status = -EACCES; | |
1784 | goto bail; | |
1785 | } | |
1786 | SET_INODE_JOURNAL(inode); | |
1787 | ||
1788 | flags = OCFS2_META_LOCK_RECOVERY | OCFS2_META_LOCK_NOQUEUE; | |
e63aecb6 | 1789 | status = ocfs2_inode_lock_full(inode, NULL, 1, flags); |
ccd979bd MF |
1790 | if (status < 0) { |
1791 | if (status != -EAGAIN) | |
1792 | mlog_errno(status); | |
1793 | goto bail; | |
1794 | } | |
1795 | ||
e63aecb6 | 1796 | ocfs2_inode_unlock(inode, 1); |
ccd979bd | 1797 | bail: |
72865d92 | 1798 | iput(inode); |
ccd979bd MF |
1799 | |
1800 | return status; | |
1801 | } | |
1802 | ||
1803 | /* Call this underneath ocfs2_super_lock. It also assumes that the | |
1804 | * slot info struct has been updated from disk. */ | |
1805 | int ocfs2_mark_dead_nodes(struct ocfs2_super *osb) | |
1806 | { | |
d85b20e4 JB |
1807 | unsigned int node_num; |
1808 | int status, i; | |
a1af7d15 | 1809 | u32 gen; |
539d8264 SM |
1810 | struct buffer_head *bh = NULL; |
1811 | struct ocfs2_dinode *di; | |
ccd979bd MF |
1812 | |
1813 | /* This is called with the super block cluster lock, so we | |
1814 | * know that the slot map can't change underneath us. */ | |
1815 | ||
d85b20e4 | 1816 | for (i = 0; i < osb->max_slots; i++) { |
539d8264 SM |
1817 | /* Read journal inode to get the recovery generation */ |
1818 | status = ocfs2_read_journal_inode(osb, i, &bh, NULL); | |
1819 | if (status) { | |
1820 | mlog_errno(status); | |
1821 | goto bail; | |
1822 | } | |
1823 | di = (struct ocfs2_dinode *)bh->b_data; | |
a1af7d15 | 1824 | gen = ocfs2_get_recovery_generation(di); |
539d8264 SM |
1825 | brelse(bh); |
1826 | bh = NULL; | |
1827 | ||
a1af7d15 MF |
1828 | spin_lock(&osb->osb_lock); |
1829 | osb->slot_recovery_generations[i] = gen; | |
1830 | ||
b4107950 TM |
1831 | trace_ocfs2_mark_dead_nodes(i, |
1832 | osb->slot_recovery_generations[i]); | |
539d8264 | 1833 | |
a1af7d15 MF |
1834 | if (i == osb->slot_num) { |
1835 | spin_unlock(&osb->osb_lock); | |
ccd979bd | 1836 | continue; |
a1af7d15 | 1837 | } |
d85b20e4 JB |
1838 | |
1839 | status = ocfs2_slot_to_node_num_locked(osb, i, &node_num); | |
a1af7d15 MF |
1840 | if (status == -ENOENT) { |
1841 | spin_unlock(&osb->osb_lock); | |
ccd979bd | 1842 | continue; |
a1af7d15 | 1843 | } |
ccd979bd | 1844 | |
a1af7d15 MF |
1845 | if (__ocfs2_recovery_map_test(osb, node_num)) { |
1846 | spin_unlock(&osb->osb_lock); | |
ccd979bd | 1847 | continue; |
a1af7d15 | 1848 | } |
d85b20e4 | 1849 | spin_unlock(&osb->osb_lock); |
ccd979bd MF |
1850 | |
1851 | /* Ok, we have a slot occupied by another node which | |
1852 | * is not in the recovery map. We trylock his journal | |
1853 | * file here to test if he's alive. */ | |
1854 | status = ocfs2_trylock_journal(osb, i); | |
1855 | if (!status) { | |
1856 | /* Since we're called from mount, we know that | |
1857 | * the recovery thread can't race us on | |
1858 | * setting / checking the recovery bits. */ | |
1859 | ocfs2_recovery_thread(osb, node_num); | |
1860 | } else if ((status < 0) && (status != -EAGAIN)) { | |
1861 | mlog_errno(status); | |
1862 | goto bail; | |
1863 | } | |
ccd979bd | 1864 | } |
ccd979bd MF |
1865 | |
1866 | status = 0; | |
1867 | bail: | |
ccd979bd MF |
1868 | return status; |
1869 | } | |
1870 | ||
83273932 SE |
1871 | /* |
1872 | * Scan timer should get fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT. Add some | |
1873 | * randomness to the timeout to minimize multple nodes firing the timer at the | |
1874 | * same time. | |
1875 | */ | |
1876 | static inline unsigned long ocfs2_orphan_scan_timeout(void) | |
1877 | { | |
1878 | unsigned long time; | |
1879 | ||
1880 | get_random_bytes(&time, sizeof(time)); | |
1881 | time = ORPHAN_SCAN_SCHEDULE_TIMEOUT + (time % 5000); | |
1882 | return msecs_to_jiffies(time); | |
1883 | } | |
1884 | ||
1885 | /* | |
1886 | * ocfs2_queue_orphan_scan calls ocfs2_queue_recovery_completion for | |
1887 | * every slot, queuing a recovery of the slot on the ocfs2_wq thread. This | |
1888 | * is done to catch any orphans that are left over in orphan directories. | |
1889 | * | |
a035bff6 SM |
1890 | * It scans all slots, even ones that are in use. It does so to handle the |
1891 | * case described below: | |
1892 | * | |
1893 | * Node 1 has an inode it was using. The dentry went away due to memory | |
1894 | * pressure. Node 1 closes the inode, but it's on the free list. The node | |
1895 | * has the open lock. | |
1896 | * Node 2 unlinks the inode. It grabs the dentry lock to notify others, | |
1897 | * but node 1 has no dentry and doesn't get the message. It trylocks the | |
1898 | * open lock, sees that another node has a PR, and does nothing. | |
1899 | * Later node 2 runs its orphan dir. It igets the inode, trylocks the | |
1900 | * open lock, sees the PR still, and does nothing. | |
1901 | * Basically, we have to trigger an orphan iput on node 1. The only way | |
1902 | * for this to happen is if node 1 runs node 2's orphan dir. | |
1903 | * | |
83273932 SE |
1904 | * ocfs2_queue_orphan_scan gets called every ORPHAN_SCAN_SCHEDULE_TIMEOUT |
1905 | * seconds. It gets an EX lock on os_lockres and checks sequence number | |
1906 | * stored in LVB. If the sequence number has changed, it means some other | |
1907 | * node has done the scan. This node skips the scan and tracks the | |
1908 | * sequence number. If the sequence number didn't change, it means a scan | |
1909 | * hasn't happened. The node queues a scan and increments the | |
1910 | * sequence number in the LVB. | |
1911 | */ | |
b519ea6d | 1912 | static void ocfs2_queue_orphan_scan(struct ocfs2_super *osb) |
83273932 SE |
1913 | { |
1914 | struct ocfs2_orphan_scan *os; | |
1915 | int status, i; | |
1916 | u32 seqno = 0; | |
1917 | ||
1918 | os = &osb->osb_orphan_scan; | |
1919 | ||
692684e1 SM |
1920 | if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE) |
1921 | goto out; | |
1922 | ||
b4107950 TM |
1923 | trace_ocfs2_queue_orphan_scan_begin(os->os_count, os->os_seqno, |
1924 | atomic_read(&os->os_state)); | |
1925 | ||
df152c24 | 1926 | status = ocfs2_orphan_scan_lock(osb, &seqno); |
83273932 SE |
1927 | if (status < 0) { |
1928 | if (status != -EAGAIN) | |
1929 | mlog_errno(status); | |
1930 | goto out; | |
1931 | } | |
1932 | ||
692684e1 SM |
1933 | /* Do no queue the tasks if the volume is being umounted */ |
1934 | if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE) | |
1935 | goto unlock; | |
1936 | ||
83273932 SE |
1937 | if (os->os_seqno != seqno) { |
1938 | os->os_seqno = seqno; | |
1939 | goto unlock; | |
1940 | } | |
1941 | ||
1942 | for (i = 0; i < osb->max_slots; i++) | |
1943 | ocfs2_queue_recovery_completion(osb->journal, i, NULL, NULL, | |
ed460cff | 1944 | NULL, ORPHAN_NO_NEED_TRUNCATE); |
83273932 SE |
1945 | /* |
1946 | * We queued a recovery on orphan slots, increment the sequence | |
1947 | * number and update LVB so other node will skip the scan for a while | |
1948 | */ | |
1949 | seqno++; | |
15633a22 | 1950 | os->os_count++; |
395627b0 | 1951 | os->os_scantime = ktime_get_seconds(); |
83273932 | 1952 | unlock: |
df152c24 | 1953 | ocfs2_orphan_scan_unlock(osb, seqno); |
83273932 | 1954 | out: |
b4107950 TM |
1955 | trace_ocfs2_queue_orphan_scan_end(os->os_count, os->os_seqno, |
1956 | atomic_read(&os->os_state)); | |
83273932 SE |
1957 | return; |
1958 | } | |
1959 | ||
1960 | /* Worker task that gets fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT millsec */ | |
b519ea6d | 1961 | static void ocfs2_orphan_scan_work(struct work_struct *work) |
83273932 SE |
1962 | { |
1963 | struct ocfs2_orphan_scan *os; | |
1964 | struct ocfs2_super *osb; | |
1965 | ||
1966 | os = container_of(work, struct ocfs2_orphan_scan, | |
1967 | os_orphan_scan_work.work); | |
1968 | osb = os->os_osb; | |
1969 | ||
1970 | mutex_lock(&os->os_lock); | |
1971 | ocfs2_queue_orphan_scan(osb); | |
692684e1 | 1972 | if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE) |
35ddf78e | 1973 | queue_delayed_work(osb->ocfs2_wq, &os->os_orphan_scan_work, |
692684e1 | 1974 | ocfs2_orphan_scan_timeout()); |
83273932 SE |
1975 | mutex_unlock(&os->os_lock); |
1976 | } | |
1977 | ||
1978 | void ocfs2_orphan_scan_stop(struct ocfs2_super *osb) | |
1979 | { | |
1980 | struct ocfs2_orphan_scan *os; | |
1981 | ||
1982 | os = &osb->osb_orphan_scan; | |
df152c24 SM |
1983 | if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE) { |
1984 | atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE); | |
1985 | mutex_lock(&os->os_lock); | |
1986 | cancel_delayed_work(&os->os_orphan_scan_work); | |
1987 | mutex_unlock(&os->os_lock); | |
1988 | } | |
83273932 SE |
1989 | } |
1990 | ||
df152c24 | 1991 | void ocfs2_orphan_scan_init(struct ocfs2_super *osb) |
83273932 SE |
1992 | { |
1993 | struct ocfs2_orphan_scan *os; | |
1994 | ||
1995 | os = &osb->osb_orphan_scan; | |
1996 | os->os_osb = osb; | |
15633a22 | 1997 | os->os_count = 0; |
3211949f | 1998 | os->os_seqno = 0; |
83273932 | 1999 | mutex_init(&os->os_lock); |
df152c24 | 2000 | INIT_DELAYED_WORK(&os->os_orphan_scan_work, ocfs2_orphan_scan_work); |
8b712cd5 | 2001 | } |
83273932 | 2002 | |
8b712cd5 JM |
2003 | void ocfs2_orphan_scan_start(struct ocfs2_super *osb) |
2004 | { | |
2005 | struct ocfs2_orphan_scan *os; | |
2006 | ||
2007 | os = &osb->osb_orphan_scan; | |
395627b0 | 2008 | os->os_scantime = ktime_get_seconds(); |
df152c24 SM |
2009 | if (ocfs2_is_hard_readonly(osb) || ocfs2_mount_local(osb)) |
2010 | atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE); | |
2011 | else { | |
2012 | atomic_set(&os->os_state, ORPHAN_SCAN_ACTIVE); | |
35ddf78e | 2013 | queue_delayed_work(osb->ocfs2_wq, &os->os_orphan_scan_work, |
40f165f4 | 2014 | ocfs2_orphan_scan_timeout()); |
df152c24 | 2015 | } |
83273932 SE |
2016 | } |
2017 | ||
5eae5b96 | 2018 | struct ocfs2_orphan_filldir_priv { |
3704412b | 2019 | struct dir_context ctx; |
5eae5b96 MF |
2020 | struct inode *head; |
2021 | struct ocfs2_super *osb; | |
30edc43c | 2022 | enum ocfs2_orphan_reco_type orphan_reco_type; |
5eae5b96 MF |
2023 | }; |
2024 | ||
ac7576f4 MS |
2025 | static int ocfs2_orphan_filldir(struct dir_context *ctx, const char *name, |
2026 | int name_len, loff_t pos, u64 ino, | |
2027 | unsigned type) | |
5eae5b96 | 2028 | { |
ac7576f4 MS |
2029 | struct ocfs2_orphan_filldir_priv *p = |
2030 | container_of(ctx, struct ocfs2_orphan_filldir_priv, ctx); | |
5eae5b96 MF |
2031 | struct inode *iter; |
2032 | ||
2033 | if (name_len == 1 && !strncmp(".", name, 1)) | |
2034 | return 0; | |
2035 | if (name_len == 2 && !strncmp("..", name, 2)) | |
2036 | return 0; | |
2037 | ||
30edc43c JQ |
2038 | /* do not include dio entry in case of orphan scan */ |
2039 | if ((p->orphan_reco_type == ORPHAN_NO_NEED_TRUNCATE) && | |
2040 | (!strncmp(name, OCFS2_DIO_ORPHAN_PREFIX, | |
2041 | OCFS2_DIO_ORPHAN_PREFIX_LEN))) | |
2042 | return 0; | |
2043 | ||
5eae5b96 MF |
2044 | /* Skip bad inodes so that recovery can continue */ |
2045 | iter = ocfs2_iget(p->osb, ino, | |
5fa0613e | 2046 | OCFS2_FI_FLAG_ORPHAN_RECOVERY, 0); |
5eae5b96 MF |
2047 | if (IS_ERR(iter)) |
2048 | return 0; | |
2049 | ||
93d911fc JQ |
2050 | if (!strncmp(name, OCFS2_DIO_ORPHAN_PREFIX, |
2051 | OCFS2_DIO_ORPHAN_PREFIX_LEN)) | |
2052 | OCFS2_I(iter)->ip_flags |= OCFS2_INODE_DIO_ORPHAN_ENTRY; | |
2053 | ||
ed460cff JQ |
2054 | /* Skip inodes which are already added to recover list, since dio may |
2055 | * happen concurrently with unlink/rename */ | |
2056 | if (OCFS2_I(iter)->ip_next_orphan) { | |
2057 | iput(iter); | |
2058 | return 0; | |
2059 | } | |
2060 | ||
b4107950 | 2061 | trace_ocfs2_orphan_filldir((unsigned long long)OCFS2_I(iter)->ip_blkno); |
5eae5b96 MF |
2062 | /* No locking is required for the next_orphan queue as there |
2063 | * is only ever a single process doing orphan recovery. */ | |
2064 | OCFS2_I(iter)->ip_next_orphan = p->head; | |
2065 | p->head = iter; | |
2066 | ||
2067 | return 0; | |
2068 | } | |
2069 | ||
b4df6ed8 MF |
2070 | static int ocfs2_queue_orphans(struct ocfs2_super *osb, |
2071 | int slot, | |
30edc43c JQ |
2072 | struct inode **head, |
2073 | enum ocfs2_orphan_reco_type orphan_reco_type) | |
ccd979bd | 2074 | { |
b4df6ed8 | 2075 | int status; |
ccd979bd | 2076 | struct inode *orphan_dir_inode = NULL; |
3704412b AV |
2077 | struct ocfs2_orphan_filldir_priv priv = { |
2078 | .ctx.actor = ocfs2_orphan_filldir, | |
2079 | .osb = osb, | |
30edc43c JQ |
2080 | .head = *head, |
2081 | .orphan_reco_type = orphan_reco_type | |
3704412b | 2082 | }; |
ccd979bd MF |
2083 | |
2084 | orphan_dir_inode = ocfs2_get_system_file_inode(osb, | |
2085 | ORPHAN_DIR_SYSTEM_INODE, | |
2086 | slot); | |
2087 | if (!orphan_dir_inode) { | |
2088 | status = -ENOENT; | |
2089 | mlog_errno(status); | |
b4df6ed8 | 2090 | return status; |
2bd63216 | 2091 | } |
ccd979bd | 2092 | |
5955102c | 2093 | inode_lock(orphan_dir_inode); |
e63aecb6 | 2094 | status = ocfs2_inode_lock(orphan_dir_inode, NULL, 0); |
ccd979bd | 2095 | if (status < 0) { |
ccd979bd MF |
2096 | mlog_errno(status); |
2097 | goto out; | |
2098 | } | |
ccd979bd | 2099 | |
3704412b | 2100 | status = ocfs2_dir_foreach(orphan_dir_inode, &priv.ctx); |
5eae5b96 MF |
2101 | if (status) { |
2102 | mlog_errno(status); | |
a86370fb | 2103 | goto out_cluster; |
ccd979bd | 2104 | } |
ccd979bd | 2105 | |
5eae5b96 MF |
2106 | *head = priv.head; |
2107 | ||
a86370fb | 2108 | out_cluster: |
e63aecb6 | 2109 | ocfs2_inode_unlock(orphan_dir_inode, 0); |
b4df6ed8 | 2110 | out: |
5955102c | 2111 | inode_unlock(orphan_dir_inode); |
ccd979bd | 2112 | iput(orphan_dir_inode); |
b4df6ed8 MF |
2113 | return status; |
2114 | } | |
2115 | ||
2116 | static int ocfs2_orphan_recovery_can_continue(struct ocfs2_super *osb, | |
2117 | int slot) | |
2118 | { | |
2119 | int ret; | |
2120 | ||
2121 | spin_lock(&osb->osb_lock); | |
2122 | ret = !osb->osb_orphan_wipes[slot]; | |
2123 | spin_unlock(&osb->osb_lock); | |
2124 | return ret; | |
2125 | } | |
2126 | ||
2127 | static void ocfs2_mark_recovering_orphan_dir(struct ocfs2_super *osb, | |
2128 | int slot) | |
2129 | { | |
2130 | spin_lock(&osb->osb_lock); | |
2131 | /* Mark ourselves such that new processes in delete_inode() | |
2132 | * know to quit early. */ | |
2133 | ocfs2_node_map_set_bit(osb, &osb->osb_recovering_orphan_dirs, slot); | |
2134 | while (osb->osb_orphan_wipes[slot]) { | |
2135 | /* If any processes are already in the middle of an | |
2136 | * orphan wipe on this dir, then we need to wait for | |
2137 | * them. */ | |
2138 | spin_unlock(&osb->osb_lock); | |
2139 | wait_event_interruptible(osb->osb_wipe_event, | |
2140 | ocfs2_orphan_recovery_can_continue(osb, slot)); | |
2141 | spin_lock(&osb->osb_lock); | |
2142 | } | |
2143 | spin_unlock(&osb->osb_lock); | |
2144 | } | |
2145 | ||
2146 | static void ocfs2_clear_recovering_orphan_dir(struct ocfs2_super *osb, | |
2147 | int slot) | |
2148 | { | |
2149 | ocfs2_node_map_clear_bit(osb, &osb->osb_recovering_orphan_dirs, slot); | |
2150 | } | |
2151 | ||
2152 | /* | |
2153 | * Orphan recovery. Each mounted node has it's own orphan dir which we | |
2154 | * must run during recovery. Our strategy here is to build a list of | |
2155 | * the inodes in the orphan dir and iget/iput them. The VFS does | |
2156 | * (most) of the rest of the work. | |
2157 | * | |
2158 | * Orphan recovery can happen at any time, not just mount so we have a | |
2159 | * couple of extra considerations. | |
2160 | * | |
2161 | * - We grab as many inodes as we can under the orphan dir lock - | |
2162 | * doing iget() outside the orphan dir risks getting a reference on | |
2163 | * an invalid inode. | |
2164 | * - We must be sure not to deadlock with other processes on the | |
2165 | * system wanting to run delete_inode(). This can happen when they go | |
2166 | * to lock the orphan dir and the orphan recovery process attempts to | |
2167 | * iget() inside the orphan dir lock. This can be avoided by | |
2168 | * advertising our state to ocfs2_delete_inode(). | |
2169 | */ | |
2170 | static int ocfs2_recover_orphans(struct ocfs2_super *osb, | |
ed460cff JQ |
2171 | int slot, |
2172 | enum ocfs2_orphan_reco_type orphan_reco_type) | |
b4df6ed8 MF |
2173 | { |
2174 | int ret = 0; | |
2175 | struct inode *inode = NULL; | |
2176 | struct inode *iter; | |
2177 | struct ocfs2_inode_info *oi; | |
cf1776a9 JQ |
2178 | struct buffer_head *di_bh = NULL; |
2179 | struct ocfs2_dinode *di = NULL; | |
b4df6ed8 | 2180 | |
b4107950 | 2181 | trace_ocfs2_recover_orphans(slot); |
b4df6ed8 MF |
2182 | |
2183 | ocfs2_mark_recovering_orphan_dir(osb, slot); | |
30edc43c | 2184 | ret = ocfs2_queue_orphans(osb, slot, &inode, orphan_reco_type); |
b4df6ed8 MF |
2185 | ocfs2_clear_recovering_orphan_dir(osb, slot); |
2186 | ||
2187 | /* Error here should be noted, but we want to continue with as | |
2188 | * many queued inodes as we've got. */ | |
2189 | if (ret) | |
2190 | mlog_errno(ret); | |
ccd979bd MF |
2191 | |
2192 | while (inode) { | |
2193 | oi = OCFS2_I(inode); | |
b4107950 TM |
2194 | trace_ocfs2_recover_orphans_iput( |
2195 | (unsigned long long)oi->ip_blkno); | |
ccd979bd MF |
2196 | |
2197 | iter = oi->ip_next_orphan; | |
ed460cff | 2198 | oi->ip_next_orphan = NULL; |
ccd979bd | 2199 | |
93d911fc | 2200 | if (oi->ip_flags & OCFS2_INODE_DIO_ORPHAN_ENTRY) { |
5955102c | 2201 | inode_lock(inode); |
93d911fc JQ |
2202 | ret = ocfs2_rw_lock(inode, 1); |
2203 | if (ret < 0) { | |
2204 | mlog_errno(ret); | |
2205 | goto unlock_mutex; | |
2206 | } | |
2207 | /* | |
2208 | * We need to take and drop the inode lock to | |
2209 | * force read inode from disk. | |
2210 | */ | |
2211 | ret = ocfs2_inode_lock(inode, &di_bh, 1); | |
2212 | if (ret) { | |
2213 | mlog_errno(ret); | |
2214 | goto unlock_rw; | |
2215 | } | |
2216 | ||
2217 | di = (struct ocfs2_dinode *)di_bh->b_data; | |
2218 | ||
2219 | if (di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL)) { | |
2220 | ret = ocfs2_truncate_file(inode, di_bh, | |
2221 | i_size_read(inode)); | |
2222 | if (ret < 0) { | |
2223 | if (ret != -ENOSPC) | |
2224 | mlog_errno(ret); | |
2225 | goto unlock_inode; | |
2226 | } | |
cf1776a9 | 2227 | |
93d911fc JQ |
2228 | ret = ocfs2_del_inode_from_orphan(osb, inode, |
2229 | di_bh, 0, 0); | |
2230 | if (ret) | |
2231 | mlog_errno(ret); | |
2232 | } | |
2233 | unlock_inode: | |
2234 | ocfs2_inode_unlock(inode, 1); | |
2235 | brelse(di_bh); | |
2236 | di_bh = NULL; | |
2237 | unlock_rw: | |
2238 | ocfs2_rw_unlock(inode, 1); | |
2239 | unlock_mutex: | |
5955102c | 2240 | inode_unlock(inode); |
ed460cff | 2241 | |
93d911fc JQ |
2242 | /* clear dio flag in ocfs2_inode_info */ |
2243 | oi->ip_flags &= ~OCFS2_INODE_DIO_ORPHAN_ENTRY; | |
2244 | } else { | |
ed460cff JQ |
2245 | spin_lock(&oi->ip_lock); |
2246 | /* Set the proper information to get us going into | |
2247 | * ocfs2_delete_inode. */ | |
2248 | oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED; | |
2249 | spin_unlock(&oi->ip_lock); | |
ad694821 JQ |
2250 | } |
2251 | ||
ccd979bd | 2252 | iput(inode); |
ccd979bd MF |
2253 | inode = iter; |
2254 | } | |
2255 | ||
b4df6ed8 | 2256 | return ret; |
ccd979bd MF |
2257 | } |
2258 | ||
19ece546 | 2259 | static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota) |
ccd979bd MF |
2260 | { |
2261 | /* This check is good because ocfs2 will wait on our recovery | |
2262 | * thread before changing it to something other than MOUNTED | |
2263 | * or DISABLED. */ | |
2264 | wait_event(osb->osb_mount_event, | |
19ece546 JK |
2265 | (!quota && atomic_read(&osb->vol_state) == VOLUME_MOUNTED) || |
2266 | atomic_read(&osb->vol_state) == VOLUME_MOUNTED_QUOTAS || | |
ccd979bd MF |
2267 | atomic_read(&osb->vol_state) == VOLUME_DISABLED); |
2268 | ||
2269 | /* If there's an error on mount, then we may never get to the | |
2270 | * MOUNTED flag, but this is set right before | |
2271 | * dismount_volume() so we can trust it. */ | |
2272 | if (atomic_read(&osb->vol_state) == VOLUME_DISABLED) { | |
b4107950 | 2273 | trace_ocfs2_wait_on_mount(VOLUME_DISABLED); |
ccd979bd MF |
2274 | mlog(0, "mount error, exiting!\n"); |
2275 | return -EBUSY; | |
2276 | } | |
2277 | ||
2278 | return 0; | |
2279 | } | |
2280 | ||
2281 | static int ocfs2_commit_thread(void *arg) | |
2282 | { | |
2283 | int status; | |
2284 | struct ocfs2_super *osb = arg; | |
2285 | struct ocfs2_journal *journal = osb->journal; | |
2286 | ||
2287 | /* we can trust j_num_trans here because _should_stop() is only set in | |
2288 | * shutdown and nobody other than ourselves should be able to start | |
2289 | * transactions. committing on shutdown might take a few iterations | |
2290 | * as final transactions put deleted inodes on the list */ | |
2291 | while (!(kthread_should_stop() && | |
2292 | atomic_read(&journal->j_num_trans) == 0)) { | |
2293 | ||
745ae8ba MF |
2294 | wait_event_interruptible(osb->checkpoint_event, |
2295 | atomic_read(&journal->j_num_trans) | |
2296 | || kthread_should_stop()); | |
ccd979bd MF |
2297 | |
2298 | status = ocfs2_commit_cache(osb); | |
55b465b6 JQ |
2299 | if (status < 0) { |
2300 | static unsigned long abort_warn_time; | |
2301 | ||
2302 | /* Warn about this once per minute */ | |
2303 | if (printk_timed_ratelimit(&abort_warn_time, 60*HZ)) | |
2304 | mlog(ML_ERROR, "status = %d, journal is " | |
2305 | "already aborted.\n", status); | |
2306 | /* | |
2307 | * After ocfs2_commit_cache() fails, j_num_trans has a | |
2308 | * non-zero value. Sleep here to avoid a busy-wait | |
2309 | * loop. | |
2310 | */ | |
2311 | msleep_interruptible(1000); | |
2312 | } | |
ccd979bd MF |
2313 | |
2314 | if (kthread_should_stop() && atomic_read(&journal->j_num_trans)){ | |
2315 | mlog(ML_KTHREAD, | |
2316 | "commit_thread: %u transactions pending on " | |
2317 | "shutdown\n", | |
2318 | atomic_read(&journal->j_num_trans)); | |
2319 | } | |
2320 | } | |
2321 | ||
2322 | return 0; | |
2323 | } | |
2324 | ||
539d8264 SM |
2325 | /* Reads all the journal inodes without taking any cluster locks. Used |
2326 | * for hard readonly access to determine whether any journal requires | |
2327 | * recovery. Also used to refresh the recovery generation numbers after | |
2328 | * a journal has been recovered by another node. | |
2329 | */ | |
ccd979bd MF |
2330 | int ocfs2_check_journals_nolocks(struct ocfs2_super *osb) |
2331 | { | |
2332 | int ret = 0; | |
2333 | unsigned int slot; | |
539d8264 | 2334 | struct buffer_head *di_bh = NULL; |
ccd979bd | 2335 | struct ocfs2_dinode *di; |
539d8264 | 2336 | int journal_dirty = 0; |
ccd979bd MF |
2337 | |
2338 | for(slot = 0; slot < osb->max_slots; slot++) { | |
539d8264 SM |
2339 | ret = ocfs2_read_journal_inode(osb, slot, &di_bh, NULL); |
2340 | if (ret) { | |
ccd979bd MF |
2341 | mlog_errno(ret); |
2342 | goto out; | |
2343 | } | |
2344 | ||
2345 | di = (struct ocfs2_dinode *) di_bh->b_data; | |
2346 | ||
539d8264 SM |
2347 | osb->slot_recovery_generations[slot] = |
2348 | ocfs2_get_recovery_generation(di); | |
2349 | ||
ccd979bd MF |
2350 | if (le32_to_cpu(di->id1.journal1.ij_flags) & |
2351 | OCFS2_JOURNAL_DIRTY_FL) | |
539d8264 | 2352 | journal_dirty = 1; |
ccd979bd MF |
2353 | |
2354 | brelse(di_bh); | |
539d8264 | 2355 | di_bh = NULL; |
ccd979bd MF |
2356 | } |
2357 | ||
2358 | out: | |
539d8264 SM |
2359 | if (journal_dirty) |
2360 | ret = -EROFS; | |
ccd979bd MF |
2361 | return ret; |
2362 | } |