]>
Commit | Line | Data |
---|---|---|
ccd979bd MF |
1 | /* -*- mode: c; c-basic-offset: 8; -*- |
2 | * vim: noexpandtab sw=8 ts=8 sts=0: | |
3 | * | |
4 | * journal.c | |
5 | * | |
6 | * Defines functions of journalling api | |
7 | * | |
8 | * Copyright (C) 2003, 2004 Oracle. All rights reserved. | |
9 | * | |
10 | * This program is free software; you can redistribute it and/or | |
11 | * modify it under the terms of the GNU General Public | |
12 | * License as published by the Free Software Foundation; either | |
13 | * version 2 of the License, or (at your option) any later version. | |
14 | * | |
15 | * This program is distributed in the hope that it will be useful, | |
16 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
17 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
18 | * General Public License for more details. | |
19 | * | |
20 | * You should have received a copy of the GNU General Public | |
21 | * License along with this program; if not, write to the | |
22 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
23 | * Boston, MA 021110-1307, USA. | |
24 | */ | |
25 | ||
26 | #include <linux/fs.h> | |
27 | #include <linux/types.h> | |
28 | #include <linux/slab.h> | |
29 | #include <linux/highmem.h> | |
30 | #include <linux/kthread.h> | |
31 | ||
32 | #define MLOG_MASK_PREFIX ML_JOURNAL | |
33 | #include <cluster/masklog.h> | |
34 | ||
35 | #include "ocfs2.h" | |
36 | ||
37 | #include "alloc.h" | |
50655ae9 | 38 | #include "blockcheck.h" |
316f4b9f | 39 | #include "dir.h" |
ccd979bd MF |
40 | #include "dlmglue.h" |
41 | #include "extent_map.h" | |
42 | #include "heartbeat.h" | |
43 | #include "inode.h" | |
44 | #include "journal.h" | |
45 | #include "localalloc.h" | |
ccd979bd MF |
46 | #include "slot_map.h" |
47 | #include "super.h" | |
ccd979bd | 48 | #include "sysfile.h" |
2205363d | 49 | #include "quota.h" |
ccd979bd MF |
50 | |
51 | #include "buffer_head_io.h" | |
52 | ||
34af946a | 53 | DEFINE_SPINLOCK(trans_inc_lock); |
ccd979bd MF |
54 | |
55 | static int ocfs2_force_read_journal(struct inode *inode); | |
56 | static int ocfs2_recover_node(struct ocfs2_super *osb, | |
2205363d | 57 | int node_num, int slot_num); |
ccd979bd MF |
58 | static int __ocfs2_recovery_thread(void *arg); |
59 | static int ocfs2_commit_cache(struct ocfs2_super *osb); | |
19ece546 | 60 | static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota); |
ccd979bd | 61 | static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb, |
539d8264 | 62 | int dirty, int replayed); |
ccd979bd MF |
63 | static int ocfs2_trylock_journal(struct ocfs2_super *osb, |
64 | int slot_num); | |
65 | static int ocfs2_recover_orphans(struct ocfs2_super *osb, | |
66 | int slot); | |
67 | static int ocfs2_commit_thread(void *arg); | |
68 | ||
19ece546 JK |
69 | static inline int ocfs2_wait_on_mount(struct ocfs2_super *osb) |
70 | { | |
71 | return __ocfs2_wait_on_mount(osb, 0); | |
72 | } | |
73 | ||
74 | static inline int ocfs2_wait_on_quotas(struct ocfs2_super *osb) | |
75 | { | |
76 | return __ocfs2_wait_on_mount(osb, 1); | |
77 | } | |
78 | ||
553abd04 JB |
79 | int ocfs2_recovery_init(struct ocfs2_super *osb) |
80 | { | |
81 | struct ocfs2_recovery_map *rm; | |
82 | ||
83 | mutex_init(&osb->recovery_lock); | |
84 | osb->disable_recovery = 0; | |
85 | osb->recovery_thread_task = NULL; | |
86 | init_waitqueue_head(&osb->recovery_event); | |
87 | ||
88 | rm = kzalloc(sizeof(struct ocfs2_recovery_map) + | |
89 | osb->max_slots * sizeof(unsigned int), | |
90 | GFP_KERNEL); | |
91 | if (!rm) { | |
92 | mlog_errno(-ENOMEM); | |
93 | return -ENOMEM; | |
94 | } | |
95 | ||
96 | rm->rm_entries = (unsigned int *)((char *)rm + | |
97 | sizeof(struct ocfs2_recovery_map)); | |
98 | osb->recovery_map = rm; | |
99 | ||
100 | return 0; | |
101 | } | |
102 | ||
103 | /* we can't grab the goofy sem lock from inside wait_event, so we use | |
104 | * memory barriers to make sure that we'll see the null task before | |
105 | * being woken up */ | |
106 | static int ocfs2_recovery_thread_running(struct ocfs2_super *osb) | |
107 | { | |
108 | mb(); | |
109 | return osb->recovery_thread_task != NULL; | |
110 | } | |
111 | ||
112 | void ocfs2_recovery_exit(struct ocfs2_super *osb) | |
113 | { | |
114 | struct ocfs2_recovery_map *rm; | |
115 | ||
116 | /* disable any new recovery threads and wait for any currently | |
117 | * running ones to exit. Do this before setting the vol_state. */ | |
118 | mutex_lock(&osb->recovery_lock); | |
119 | osb->disable_recovery = 1; | |
120 | mutex_unlock(&osb->recovery_lock); | |
121 | wait_event(osb->recovery_event, !ocfs2_recovery_thread_running(osb)); | |
122 | ||
123 | /* At this point, we know that no more recovery threads can be | |
124 | * launched, so wait for any recovery completion work to | |
125 | * complete. */ | |
126 | flush_workqueue(ocfs2_wq); | |
127 | ||
128 | /* | |
129 | * Now that recovery is shut down, and the osb is about to be | |
130 | * freed, the osb_lock is not taken here. | |
131 | */ | |
132 | rm = osb->recovery_map; | |
133 | /* XXX: Should we bug if there are dirty entries? */ | |
134 | ||
135 | kfree(rm); | |
136 | } | |
137 | ||
138 | static int __ocfs2_recovery_map_test(struct ocfs2_super *osb, | |
139 | unsigned int node_num) | |
140 | { | |
141 | int i; | |
142 | struct ocfs2_recovery_map *rm = osb->recovery_map; | |
143 | ||
144 | assert_spin_locked(&osb->osb_lock); | |
145 | ||
146 | for (i = 0; i < rm->rm_used; i++) { | |
147 | if (rm->rm_entries[i] == node_num) | |
148 | return 1; | |
149 | } | |
150 | ||
151 | return 0; | |
152 | } | |
153 | ||
154 | /* Behaves like test-and-set. Returns the previous value */ | |
155 | static int ocfs2_recovery_map_set(struct ocfs2_super *osb, | |
156 | unsigned int node_num) | |
157 | { | |
158 | struct ocfs2_recovery_map *rm = osb->recovery_map; | |
159 | ||
160 | spin_lock(&osb->osb_lock); | |
161 | if (__ocfs2_recovery_map_test(osb, node_num)) { | |
162 | spin_unlock(&osb->osb_lock); | |
163 | return 1; | |
164 | } | |
165 | ||
166 | /* XXX: Can this be exploited? Not from o2dlm... */ | |
167 | BUG_ON(rm->rm_used >= osb->max_slots); | |
168 | ||
169 | rm->rm_entries[rm->rm_used] = node_num; | |
170 | rm->rm_used++; | |
171 | spin_unlock(&osb->osb_lock); | |
172 | ||
173 | return 0; | |
174 | } | |
175 | ||
176 | static void ocfs2_recovery_map_clear(struct ocfs2_super *osb, | |
177 | unsigned int node_num) | |
178 | { | |
179 | int i; | |
180 | struct ocfs2_recovery_map *rm = osb->recovery_map; | |
181 | ||
182 | spin_lock(&osb->osb_lock); | |
183 | ||
184 | for (i = 0; i < rm->rm_used; i++) { | |
185 | if (rm->rm_entries[i] == node_num) | |
186 | break; | |
187 | } | |
188 | ||
189 | if (i < rm->rm_used) { | |
190 | /* XXX: be careful with the pointer math */ | |
191 | memmove(&(rm->rm_entries[i]), &(rm->rm_entries[i + 1]), | |
192 | (rm->rm_used - i - 1) * sizeof(unsigned int)); | |
193 | rm->rm_used--; | |
194 | } | |
195 | ||
196 | spin_unlock(&osb->osb_lock); | |
197 | } | |
198 | ||
ccd979bd MF |
199 | static int ocfs2_commit_cache(struct ocfs2_super *osb) |
200 | { | |
201 | int status = 0; | |
202 | unsigned int flushed; | |
203 | unsigned long old_id; | |
204 | struct ocfs2_journal *journal = NULL; | |
205 | ||
206 | mlog_entry_void(); | |
207 | ||
208 | journal = osb->journal; | |
209 | ||
210 | /* Flush all pending commits and checkpoint the journal. */ | |
211 | down_write(&journal->j_trans_barrier); | |
212 | ||
213 | if (atomic_read(&journal->j_num_trans) == 0) { | |
214 | up_write(&journal->j_trans_barrier); | |
215 | mlog(0, "No transactions for me to flush!\n"); | |
216 | goto finally; | |
217 | } | |
218 | ||
2b4e30fb JB |
219 | jbd2_journal_lock_updates(journal->j_journal); |
220 | status = jbd2_journal_flush(journal->j_journal); | |
221 | jbd2_journal_unlock_updates(journal->j_journal); | |
ccd979bd MF |
222 | if (status < 0) { |
223 | up_write(&journal->j_trans_barrier); | |
224 | mlog_errno(status); | |
225 | goto finally; | |
226 | } | |
227 | ||
228 | old_id = ocfs2_inc_trans_id(journal); | |
229 | ||
230 | flushed = atomic_read(&journal->j_num_trans); | |
231 | atomic_set(&journal->j_num_trans, 0); | |
232 | up_write(&journal->j_trans_barrier); | |
233 | ||
234 | mlog(0, "commit_thread: flushed transaction %lu (%u handles)\n", | |
235 | journal->j_trans_id, flushed); | |
236 | ||
34d024f8 | 237 | ocfs2_wake_downconvert_thread(osb); |
ccd979bd MF |
238 | wake_up(&journal->j_checkpointed); |
239 | finally: | |
240 | mlog_exit(status); | |
241 | return status; | |
242 | } | |
243 | ||
ccd979bd MF |
244 | /* pass it NULL and it will allocate a new handle object for you. If |
245 | * you pass it a handle however, it may still return error, in which | |
246 | * case it has free'd the passed handle for you. */ | |
1fabe148 | 247 | handle_t *ocfs2_start_trans(struct ocfs2_super *osb, int max_buffs) |
ccd979bd | 248 | { |
ccd979bd | 249 | journal_t *journal = osb->journal->j_journal; |
1fabe148 | 250 | handle_t *handle; |
ccd979bd | 251 | |
ebdec83b | 252 | BUG_ON(!osb || !osb->journal->j_journal); |
ccd979bd | 253 | |
65eff9cc MF |
254 | if (ocfs2_is_hard_readonly(osb)) |
255 | return ERR_PTR(-EROFS); | |
ccd979bd MF |
256 | |
257 | BUG_ON(osb->journal->j_state == OCFS2_JOURNAL_FREE); | |
258 | BUG_ON(max_buffs <= 0); | |
259 | ||
90e86a63 JK |
260 | /* Nested transaction? Just return the handle... */ |
261 | if (journal_current_handle()) | |
262 | return jbd2_journal_start(journal, max_buffs); | |
ccd979bd | 263 | |
ccd979bd MF |
264 | down_read(&osb->journal->j_trans_barrier); |
265 | ||
2b4e30fb | 266 | handle = jbd2_journal_start(journal, max_buffs); |
1fabe148 | 267 | if (IS_ERR(handle)) { |
ccd979bd MF |
268 | up_read(&osb->journal->j_trans_barrier); |
269 | ||
1fabe148 | 270 | mlog_errno(PTR_ERR(handle)); |
ccd979bd MF |
271 | |
272 | if (is_journal_aborted(journal)) { | |
273 | ocfs2_abort(osb->sb, "Detected aborted journal"); | |
1fabe148 | 274 | handle = ERR_PTR(-EROFS); |
ccd979bd | 275 | } |
c271c5c2 SM |
276 | } else { |
277 | if (!ocfs2_mount_local(osb)) | |
278 | atomic_inc(&(osb->journal->j_num_trans)); | |
279 | } | |
ccd979bd | 280 | |
ccd979bd | 281 | return handle; |
ccd979bd MF |
282 | } |
283 | ||
1fabe148 MF |
284 | int ocfs2_commit_trans(struct ocfs2_super *osb, |
285 | handle_t *handle) | |
ccd979bd | 286 | { |
90e86a63 | 287 | int ret, nested; |
02dc1af4 | 288 | struct ocfs2_journal *journal = osb->journal; |
ccd979bd MF |
289 | |
290 | BUG_ON(!handle); | |
291 | ||
90e86a63 | 292 | nested = handle->h_ref > 1; |
2b4e30fb | 293 | ret = jbd2_journal_stop(handle); |
1fabe148 MF |
294 | if (ret < 0) |
295 | mlog_errno(ret); | |
ccd979bd | 296 | |
90e86a63 JK |
297 | if (!nested) |
298 | up_read(&journal->j_trans_barrier); | |
ccd979bd | 299 | |
1fabe148 | 300 | return ret; |
ccd979bd MF |
301 | } |
302 | ||
303 | /* | |
304 | * 'nblocks' is what you want to add to the current | |
305 | * transaction. extend_trans will either extend the current handle by | |
306 | * nblocks, or commit it and start a new one with nblocks credits. | |
307 | * | |
2b4e30fb | 308 | * This might call jbd2_journal_restart() which will commit dirty buffers |
e8aed345 MF |
309 | * and then restart the transaction. Before calling |
310 | * ocfs2_extend_trans(), any changed blocks should have been | |
311 | * dirtied. After calling it, all blocks which need to be changed must | |
312 | * go through another set of journal_access/journal_dirty calls. | |
313 | * | |
ccd979bd MF |
314 | * WARNING: This will not release any semaphores or disk locks taken |
315 | * during the transaction, so make sure they were taken *before* | |
316 | * start_trans or we'll have ordering deadlocks. | |
317 | * | |
318 | * WARNING2: Note that we do *not* drop j_trans_barrier here. This is | |
319 | * good because transaction ids haven't yet been recorded on the | |
320 | * cluster locks associated with this handle. | |
321 | */ | |
1fc58146 | 322 | int ocfs2_extend_trans(handle_t *handle, int nblocks) |
ccd979bd MF |
323 | { |
324 | int status; | |
325 | ||
326 | BUG_ON(!handle); | |
ccd979bd MF |
327 | BUG_ON(!nblocks); |
328 | ||
329 | mlog_entry_void(); | |
330 | ||
331 | mlog(0, "Trying to extend transaction by %d blocks\n", nblocks); | |
332 | ||
e407e397 | 333 | #ifdef CONFIG_OCFS2_DEBUG_FS |
0879c584 MF |
334 | status = 1; |
335 | #else | |
2b4e30fb | 336 | status = jbd2_journal_extend(handle, nblocks); |
ccd979bd MF |
337 | if (status < 0) { |
338 | mlog_errno(status); | |
339 | goto bail; | |
340 | } | |
0879c584 | 341 | #endif |
ccd979bd MF |
342 | |
343 | if (status > 0) { | |
2b4e30fb JB |
344 | mlog(0, |
345 | "jbd2_journal_extend failed, trying " | |
346 | "jbd2_journal_restart\n"); | |
347 | status = jbd2_journal_restart(handle, nblocks); | |
ccd979bd | 348 | if (status < 0) { |
ccd979bd MF |
349 | mlog_errno(status); |
350 | goto bail; | |
351 | } | |
01ddf1e1 | 352 | } |
ccd979bd MF |
353 | |
354 | status = 0; | |
355 | bail: | |
356 | ||
357 | mlog_exit(status); | |
358 | return status; | |
359 | } | |
360 | ||
50655ae9 JB |
361 | struct ocfs2_triggers { |
362 | struct jbd2_buffer_trigger_type ot_triggers; | |
363 | int ot_offset; | |
364 | }; | |
365 | ||
366 | static inline struct ocfs2_triggers *to_ocfs2_trigger(struct jbd2_buffer_trigger_type *triggers) | |
367 | { | |
368 | return container_of(triggers, struct ocfs2_triggers, ot_triggers); | |
369 | } | |
370 | ||
371 | static void ocfs2_commit_trigger(struct jbd2_buffer_trigger_type *triggers, | |
372 | struct buffer_head *bh, | |
373 | void *data, size_t size) | |
374 | { | |
375 | struct ocfs2_triggers *ot = to_ocfs2_trigger(triggers); | |
376 | ||
377 | /* | |
378 | * We aren't guaranteed to have the superblock here, so we | |
379 | * must unconditionally compute the ecc data. | |
380 | * __ocfs2_journal_access() will only set the triggers if | |
381 | * metaecc is enabled. | |
382 | */ | |
383 | ocfs2_block_check_compute(data, size, data + ot->ot_offset); | |
384 | } | |
385 | ||
386 | /* | |
387 | * Quota blocks have their own trigger because the struct ocfs2_block_check | |
388 | * offset depends on the blocksize. | |
389 | */ | |
390 | static void ocfs2_dq_commit_trigger(struct jbd2_buffer_trigger_type *triggers, | |
391 | struct buffer_head *bh, | |
392 | void *data, size_t size) | |
393 | { | |
394 | struct ocfs2_disk_dqtrailer *dqt = | |
395 | ocfs2_block_dqtrailer(size, data); | |
396 | ||
397 | /* | |
398 | * We aren't guaranteed to have the superblock here, so we | |
399 | * must unconditionally compute the ecc data. | |
400 | * __ocfs2_journal_access() will only set the triggers if | |
401 | * metaecc is enabled. | |
402 | */ | |
403 | ocfs2_block_check_compute(data, size, &dqt->dq_check); | |
404 | } | |
405 | ||
c175a518 JB |
406 | /* |
407 | * Directory blocks also have their own trigger because the | |
408 | * struct ocfs2_block_check offset depends on the blocksize. | |
409 | */ | |
410 | static void ocfs2_db_commit_trigger(struct jbd2_buffer_trigger_type *triggers, | |
411 | struct buffer_head *bh, | |
412 | void *data, size_t size) | |
413 | { | |
414 | struct ocfs2_dir_block_trailer *trailer = | |
415 | ocfs2_dir_trailer_from_size(size, data); | |
416 | ||
417 | /* | |
418 | * We aren't guaranteed to have the superblock here, so we | |
419 | * must unconditionally compute the ecc data. | |
420 | * __ocfs2_journal_access() will only set the triggers if | |
421 | * metaecc is enabled. | |
422 | */ | |
423 | ocfs2_block_check_compute(data, size, &trailer->db_check); | |
424 | } | |
425 | ||
50655ae9 JB |
426 | static void ocfs2_abort_trigger(struct jbd2_buffer_trigger_type *triggers, |
427 | struct buffer_head *bh) | |
428 | { | |
429 | mlog(ML_ERROR, | |
430 | "ocfs2_abort_trigger called by JBD2. bh = 0x%lx, " | |
431 | "bh->b_blocknr = %llu\n", | |
432 | (unsigned long)bh, | |
433 | (unsigned long long)bh->b_blocknr); | |
434 | ||
435 | /* We aren't guaranteed to have the superblock here - but if we | |
436 | * don't, it'll just crash. */ | |
437 | ocfs2_error(bh->b_assoc_map->host->i_sb, | |
438 | "JBD2 has aborted our journal, ocfs2 cannot continue\n"); | |
439 | } | |
440 | ||
441 | static struct ocfs2_triggers di_triggers = { | |
442 | .ot_triggers = { | |
443 | .t_commit = ocfs2_commit_trigger, | |
444 | .t_abort = ocfs2_abort_trigger, | |
445 | }, | |
446 | .ot_offset = offsetof(struct ocfs2_dinode, i_check), | |
447 | }; | |
448 | ||
449 | static struct ocfs2_triggers eb_triggers = { | |
450 | .ot_triggers = { | |
451 | .t_commit = ocfs2_commit_trigger, | |
452 | .t_abort = ocfs2_abort_trigger, | |
453 | }, | |
454 | .ot_offset = offsetof(struct ocfs2_extent_block, h_check), | |
455 | }; | |
456 | ||
457 | static struct ocfs2_triggers gd_triggers = { | |
458 | .ot_triggers = { | |
459 | .t_commit = ocfs2_commit_trigger, | |
460 | .t_abort = ocfs2_abort_trigger, | |
461 | }, | |
462 | .ot_offset = offsetof(struct ocfs2_group_desc, bg_check), | |
463 | }; | |
464 | ||
c175a518 JB |
465 | static struct ocfs2_triggers db_triggers = { |
466 | .ot_triggers = { | |
467 | .t_commit = ocfs2_db_commit_trigger, | |
468 | .t_abort = ocfs2_abort_trigger, | |
469 | }, | |
470 | }; | |
471 | ||
50655ae9 JB |
472 | static struct ocfs2_triggers xb_triggers = { |
473 | .ot_triggers = { | |
474 | .t_commit = ocfs2_commit_trigger, | |
475 | .t_abort = ocfs2_abort_trigger, | |
476 | }, | |
477 | .ot_offset = offsetof(struct ocfs2_xattr_block, xb_check), | |
478 | }; | |
479 | ||
480 | static struct ocfs2_triggers dq_triggers = { | |
481 | .ot_triggers = { | |
482 | .t_commit = ocfs2_dq_commit_trigger, | |
483 | .t_abort = ocfs2_abort_trigger, | |
484 | }, | |
485 | }; | |
486 | ||
487 | static int __ocfs2_journal_access(handle_t *handle, | |
488 | struct inode *inode, | |
489 | struct buffer_head *bh, | |
490 | struct ocfs2_triggers *triggers, | |
491 | int type) | |
ccd979bd MF |
492 | { |
493 | int status; | |
494 | ||
495 | BUG_ON(!inode); | |
496 | BUG_ON(!handle); | |
497 | BUG_ON(!bh); | |
ccd979bd | 498 | |
205f87f6 | 499 | mlog_entry("bh->b_blocknr=%llu, type=%d (\"%s\"), bh->b_size = %zu\n", |
ccd979bd MF |
500 | (unsigned long long)bh->b_blocknr, type, |
501 | (type == OCFS2_JOURNAL_ACCESS_CREATE) ? | |
502 | "OCFS2_JOURNAL_ACCESS_CREATE" : | |
503 | "OCFS2_JOURNAL_ACCESS_WRITE", | |
504 | bh->b_size); | |
505 | ||
506 | /* we can safely remove this assertion after testing. */ | |
507 | if (!buffer_uptodate(bh)) { | |
508 | mlog(ML_ERROR, "giving me a buffer that's not uptodate!\n"); | |
509 | mlog(ML_ERROR, "b_blocknr=%llu\n", | |
510 | (unsigned long long)bh->b_blocknr); | |
511 | BUG(); | |
512 | } | |
513 | ||
514 | /* Set the current transaction information on the inode so | |
515 | * that the locking code knows whether it can drop it's locks | |
516 | * on this inode or not. We're protected from the commit | |
517 | * thread updating the current transaction id until | |
518 | * ocfs2_commit_trans() because ocfs2_start_trans() took | |
519 | * j_trans_barrier for us. */ | |
520 | ocfs2_set_inode_lock_trans(OCFS2_SB(inode->i_sb)->journal, inode); | |
521 | ||
251b6ecc | 522 | mutex_lock(&OCFS2_I(inode)->ip_io_mutex); |
ccd979bd MF |
523 | switch (type) { |
524 | case OCFS2_JOURNAL_ACCESS_CREATE: | |
525 | case OCFS2_JOURNAL_ACCESS_WRITE: | |
2b4e30fb | 526 | status = jbd2_journal_get_write_access(handle, bh); |
ccd979bd MF |
527 | break; |
528 | ||
529 | case OCFS2_JOURNAL_ACCESS_UNDO: | |
2b4e30fb | 530 | status = jbd2_journal_get_undo_access(handle, bh); |
ccd979bd MF |
531 | break; |
532 | ||
533 | default: | |
534 | status = -EINVAL; | |
535 | mlog(ML_ERROR, "Uknown access type!\n"); | |
536 | } | |
50655ae9 JB |
537 | if (!status && ocfs2_meta_ecc(OCFS2_SB(inode->i_sb)) && triggers) |
538 | jbd2_journal_set_triggers(bh, &triggers->ot_triggers); | |
251b6ecc | 539 | mutex_unlock(&OCFS2_I(inode)->ip_io_mutex); |
ccd979bd MF |
540 | |
541 | if (status < 0) | |
542 | mlog(ML_ERROR, "Error %d getting %d access to buffer!\n", | |
543 | status, type); | |
544 | ||
545 | mlog_exit(status); | |
546 | return status; | |
547 | } | |
548 | ||
50655ae9 JB |
549 | int ocfs2_journal_access_di(handle_t *handle, struct inode *inode, |
550 | struct buffer_head *bh, int type) | |
551 | { | |
552 | return __ocfs2_journal_access(handle, inode, bh, &di_triggers, | |
553 | type); | |
554 | } | |
555 | ||
556 | int ocfs2_journal_access_eb(handle_t *handle, struct inode *inode, | |
557 | struct buffer_head *bh, int type) | |
558 | { | |
559 | return __ocfs2_journal_access(handle, inode, bh, &eb_triggers, | |
560 | type); | |
561 | } | |
562 | ||
563 | int ocfs2_journal_access_gd(handle_t *handle, struct inode *inode, | |
564 | struct buffer_head *bh, int type) | |
565 | { | |
566 | return __ocfs2_journal_access(handle, inode, bh, &gd_triggers, | |
567 | type); | |
568 | } | |
569 | ||
570 | int ocfs2_journal_access_db(handle_t *handle, struct inode *inode, | |
571 | struct buffer_head *bh, int type) | |
572 | { | |
c175a518 JB |
573 | return __ocfs2_journal_access(handle, inode, bh, &db_triggers, |
574 | type); | |
50655ae9 JB |
575 | } |
576 | ||
577 | int ocfs2_journal_access_xb(handle_t *handle, struct inode *inode, | |
578 | struct buffer_head *bh, int type) | |
579 | { | |
580 | return __ocfs2_journal_access(handle, inode, bh, &xb_triggers, | |
581 | type); | |
582 | } | |
583 | ||
584 | int ocfs2_journal_access_dq(handle_t *handle, struct inode *inode, | |
585 | struct buffer_head *bh, int type) | |
586 | { | |
587 | return __ocfs2_journal_access(handle, inode, bh, &dq_triggers, | |
588 | type); | |
589 | } | |
590 | ||
591 | int ocfs2_journal_access(handle_t *handle, struct inode *inode, | |
592 | struct buffer_head *bh, int type) | |
593 | { | |
594 | return __ocfs2_journal_access(handle, inode, bh, NULL, type); | |
595 | } | |
596 | ||
1fabe148 | 597 | int ocfs2_journal_dirty(handle_t *handle, |
ccd979bd MF |
598 | struct buffer_head *bh) |
599 | { | |
600 | int status; | |
601 | ||
ccd979bd MF |
602 | mlog_entry("(bh->b_blocknr=%llu)\n", |
603 | (unsigned long long)bh->b_blocknr); | |
604 | ||
2b4e30fb | 605 | status = jbd2_journal_dirty_metadata(handle, bh); |
ccd979bd MF |
606 | if (status < 0) |
607 | mlog(ML_ERROR, "Could not dirty metadata buffer. " | |
608 | "(bh->b_blocknr=%llu)\n", | |
609 | (unsigned long long)bh->b_blocknr); | |
610 | ||
611 | mlog_exit(status); | |
612 | return status; | |
613 | } | |
614 | ||
2b4e30fb | 615 | #define OCFS2_DEFAULT_COMMIT_INTERVAL (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE) |
ccd979bd MF |
616 | |
617 | void ocfs2_set_journal_params(struct ocfs2_super *osb) | |
618 | { | |
619 | journal_t *journal = osb->journal->j_journal; | |
d147b3d6 MF |
620 | unsigned long commit_interval = OCFS2_DEFAULT_COMMIT_INTERVAL; |
621 | ||
622 | if (osb->osb_commit_interval) | |
623 | commit_interval = osb->osb_commit_interval; | |
ccd979bd MF |
624 | |
625 | spin_lock(&journal->j_state_lock); | |
d147b3d6 | 626 | journal->j_commit_interval = commit_interval; |
ccd979bd | 627 | if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER) |
2b4e30fb | 628 | journal->j_flags |= JBD2_BARRIER; |
ccd979bd | 629 | else |
2b4e30fb | 630 | journal->j_flags &= ~JBD2_BARRIER; |
ccd979bd MF |
631 | spin_unlock(&journal->j_state_lock); |
632 | } | |
633 | ||
634 | int ocfs2_journal_init(struct ocfs2_journal *journal, int *dirty) | |
635 | { | |
636 | int status = -1; | |
637 | struct inode *inode = NULL; /* the journal inode */ | |
638 | journal_t *j_journal = NULL; | |
639 | struct ocfs2_dinode *di = NULL; | |
640 | struct buffer_head *bh = NULL; | |
641 | struct ocfs2_super *osb; | |
e63aecb6 | 642 | int inode_lock = 0; |
ccd979bd MF |
643 | |
644 | mlog_entry_void(); | |
645 | ||
646 | BUG_ON(!journal); | |
647 | ||
648 | osb = journal->j_osb; | |
649 | ||
650 | /* already have the inode for our journal */ | |
651 | inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, | |
652 | osb->slot_num); | |
653 | if (inode == NULL) { | |
654 | status = -EACCES; | |
655 | mlog_errno(status); | |
656 | goto done; | |
657 | } | |
658 | if (is_bad_inode(inode)) { | |
659 | mlog(ML_ERROR, "access error (bad inode)\n"); | |
660 | iput(inode); | |
661 | inode = NULL; | |
662 | status = -EACCES; | |
663 | goto done; | |
664 | } | |
665 | ||
666 | SET_INODE_JOURNAL(inode); | |
667 | OCFS2_I(inode)->ip_open_count++; | |
668 | ||
6eff5790 MF |
669 | /* Skip recovery waits here - journal inode metadata never |
670 | * changes in a live cluster so it can be considered an | |
671 | * exception to the rule. */ | |
e63aecb6 | 672 | status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY); |
ccd979bd MF |
673 | if (status < 0) { |
674 | if (status != -ERESTARTSYS) | |
675 | mlog(ML_ERROR, "Could not get lock on journal!\n"); | |
676 | goto done; | |
677 | } | |
678 | ||
e63aecb6 | 679 | inode_lock = 1; |
ccd979bd MF |
680 | di = (struct ocfs2_dinode *)bh->b_data; |
681 | ||
682 | if (inode->i_size < OCFS2_MIN_JOURNAL_SIZE) { | |
683 | mlog(ML_ERROR, "Journal file size (%lld) is too small!\n", | |
684 | inode->i_size); | |
685 | status = -EINVAL; | |
686 | goto done; | |
687 | } | |
688 | ||
689 | mlog(0, "inode->i_size = %lld\n", inode->i_size); | |
5515eff8 AM |
690 | mlog(0, "inode->i_blocks = %llu\n", |
691 | (unsigned long long)inode->i_blocks); | |
ccd979bd MF |
692 | mlog(0, "inode->ip_clusters = %u\n", OCFS2_I(inode)->ip_clusters); |
693 | ||
694 | /* call the kernels journal init function now */ | |
2b4e30fb | 695 | j_journal = jbd2_journal_init_inode(inode); |
ccd979bd MF |
696 | if (j_journal == NULL) { |
697 | mlog(ML_ERROR, "Linux journal layer error\n"); | |
698 | status = -EINVAL; | |
699 | goto done; | |
700 | } | |
701 | ||
2b4e30fb | 702 | mlog(0, "Returned from jbd2_journal_init_inode\n"); |
ccd979bd MF |
703 | mlog(0, "j_journal->j_maxlen = %u\n", j_journal->j_maxlen); |
704 | ||
705 | *dirty = (le32_to_cpu(di->id1.journal1.ij_flags) & | |
706 | OCFS2_JOURNAL_DIRTY_FL); | |
707 | ||
708 | journal->j_journal = j_journal; | |
709 | journal->j_inode = inode; | |
710 | journal->j_bh = bh; | |
711 | ||
712 | ocfs2_set_journal_params(osb); | |
713 | ||
714 | journal->j_state = OCFS2_JOURNAL_LOADED; | |
715 | ||
716 | status = 0; | |
717 | done: | |
718 | if (status < 0) { | |
e63aecb6 MF |
719 | if (inode_lock) |
720 | ocfs2_inode_unlock(inode, 1); | |
a81cb88b | 721 | brelse(bh); |
ccd979bd MF |
722 | if (inode) { |
723 | OCFS2_I(inode)->ip_open_count--; | |
724 | iput(inode); | |
725 | } | |
726 | } | |
727 | ||
728 | mlog_exit(status); | |
729 | return status; | |
730 | } | |
731 | ||
539d8264 SM |
732 | static void ocfs2_bump_recovery_generation(struct ocfs2_dinode *di) |
733 | { | |
734 | le32_add_cpu(&(di->id1.journal1.ij_recovery_generation), 1); | |
735 | } | |
736 | ||
737 | static u32 ocfs2_get_recovery_generation(struct ocfs2_dinode *di) | |
738 | { | |
739 | return le32_to_cpu(di->id1.journal1.ij_recovery_generation); | |
740 | } | |
741 | ||
ccd979bd | 742 | static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb, |
539d8264 | 743 | int dirty, int replayed) |
ccd979bd MF |
744 | { |
745 | int status; | |
746 | unsigned int flags; | |
747 | struct ocfs2_journal *journal = osb->journal; | |
748 | struct buffer_head *bh = journal->j_bh; | |
749 | struct ocfs2_dinode *fe; | |
750 | ||
751 | mlog_entry_void(); | |
752 | ||
753 | fe = (struct ocfs2_dinode *)bh->b_data; | |
10995aa2 JB |
754 | |
755 | /* The journal bh on the osb always comes from ocfs2_journal_init() | |
756 | * and was validated there inside ocfs2_inode_lock_full(). It's a | |
757 | * code bug if we mess it up. */ | |
758 | BUG_ON(!OCFS2_IS_VALID_DINODE(fe)); | |
ccd979bd MF |
759 | |
760 | flags = le32_to_cpu(fe->id1.journal1.ij_flags); | |
761 | if (dirty) | |
762 | flags |= OCFS2_JOURNAL_DIRTY_FL; | |
763 | else | |
764 | flags &= ~OCFS2_JOURNAL_DIRTY_FL; | |
765 | fe->id1.journal1.ij_flags = cpu_to_le32(flags); | |
766 | ||
539d8264 SM |
767 | if (replayed) |
768 | ocfs2_bump_recovery_generation(fe); | |
769 | ||
13723d00 | 770 | ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check); |
ccd979bd MF |
771 | status = ocfs2_write_block(osb, bh, journal->j_inode); |
772 | if (status < 0) | |
773 | mlog_errno(status); | |
774 | ||
ccd979bd MF |
775 | mlog_exit(status); |
776 | return status; | |
777 | } | |
778 | ||
779 | /* | |
780 | * If the journal has been kmalloc'd it needs to be freed after this | |
781 | * call. | |
782 | */ | |
783 | void ocfs2_journal_shutdown(struct ocfs2_super *osb) | |
784 | { | |
785 | struct ocfs2_journal *journal = NULL; | |
786 | int status = 0; | |
787 | struct inode *inode = NULL; | |
788 | int num_running_trans = 0; | |
789 | ||
790 | mlog_entry_void(); | |
791 | ||
ebdec83b | 792 | BUG_ON(!osb); |
ccd979bd MF |
793 | |
794 | journal = osb->journal; | |
795 | if (!journal) | |
796 | goto done; | |
797 | ||
798 | inode = journal->j_inode; | |
799 | ||
800 | if (journal->j_state != OCFS2_JOURNAL_LOADED) | |
801 | goto done; | |
802 | ||
2b4e30fb | 803 | /* need to inc inode use count - jbd2_journal_destroy will iput. */ |
ccd979bd MF |
804 | if (!igrab(inode)) |
805 | BUG(); | |
806 | ||
807 | num_running_trans = atomic_read(&(osb->journal->j_num_trans)); | |
808 | if (num_running_trans > 0) | |
809 | mlog(0, "Shutting down journal: must wait on %d " | |
810 | "running transactions!\n", | |
811 | num_running_trans); | |
812 | ||
813 | /* Do a commit_cache here. It will flush our journal, *and* | |
814 | * release any locks that are still held. | |
815 | * set the SHUTDOWN flag and release the trans lock. | |
816 | * the commit thread will take the trans lock for us below. */ | |
817 | journal->j_state = OCFS2_JOURNAL_IN_SHUTDOWN; | |
818 | ||
819 | /* The OCFS2_JOURNAL_IN_SHUTDOWN will signal to commit_cache to not | |
820 | * drop the trans_lock (which we want to hold until we | |
821 | * completely destroy the journal. */ | |
822 | if (osb->commit_task) { | |
823 | /* Wait for the commit thread */ | |
824 | mlog(0, "Waiting for ocfs2commit to exit....\n"); | |
825 | kthread_stop(osb->commit_task); | |
826 | osb->commit_task = NULL; | |
827 | } | |
828 | ||
829 | BUG_ON(atomic_read(&(osb->journal->j_num_trans)) != 0); | |
830 | ||
c271c5c2 | 831 | if (ocfs2_mount_local(osb)) { |
2b4e30fb JB |
832 | jbd2_journal_lock_updates(journal->j_journal); |
833 | status = jbd2_journal_flush(journal->j_journal); | |
834 | jbd2_journal_unlock_updates(journal->j_journal); | |
c271c5c2 SM |
835 | if (status < 0) |
836 | mlog_errno(status); | |
837 | } | |
838 | ||
839 | if (status == 0) { | |
840 | /* | |
841 | * Do not toggle if flush was unsuccessful otherwise | |
842 | * will leave dirty metadata in a "clean" journal | |
843 | */ | |
539d8264 | 844 | status = ocfs2_journal_toggle_dirty(osb, 0, 0); |
c271c5c2 SM |
845 | if (status < 0) |
846 | mlog_errno(status); | |
847 | } | |
ccd979bd MF |
848 | |
849 | /* Shutdown the kernel journal system */ | |
2b4e30fb | 850 | jbd2_journal_destroy(journal->j_journal); |
ae0dff68 | 851 | journal->j_journal = NULL; |
ccd979bd MF |
852 | |
853 | OCFS2_I(inode)->ip_open_count--; | |
854 | ||
855 | /* unlock our journal */ | |
e63aecb6 | 856 | ocfs2_inode_unlock(inode, 1); |
ccd979bd MF |
857 | |
858 | brelse(journal->j_bh); | |
859 | journal->j_bh = NULL; | |
860 | ||
861 | journal->j_state = OCFS2_JOURNAL_FREE; | |
862 | ||
863 | // up_write(&journal->j_trans_barrier); | |
864 | done: | |
865 | if (inode) | |
866 | iput(inode); | |
867 | mlog_exit_void(); | |
868 | } | |
869 | ||
870 | static void ocfs2_clear_journal_error(struct super_block *sb, | |
871 | journal_t *journal, | |
872 | int slot) | |
873 | { | |
874 | int olderr; | |
875 | ||
2b4e30fb | 876 | olderr = jbd2_journal_errno(journal); |
ccd979bd MF |
877 | if (olderr) { |
878 | mlog(ML_ERROR, "File system error %d recorded in " | |
879 | "journal %u.\n", olderr, slot); | |
880 | mlog(ML_ERROR, "File system on device %s needs checking.\n", | |
881 | sb->s_id); | |
882 | ||
2b4e30fb JB |
883 | jbd2_journal_ack_err(journal); |
884 | jbd2_journal_clear_err(journal); | |
ccd979bd MF |
885 | } |
886 | } | |
887 | ||
539d8264 | 888 | int ocfs2_journal_load(struct ocfs2_journal *journal, int local, int replayed) |
ccd979bd MF |
889 | { |
890 | int status = 0; | |
891 | struct ocfs2_super *osb; | |
892 | ||
893 | mlog_entry_void(); | |
894 | ||
b1f3550f | 895 | BUG_ON(!journal); |
ccd979bd MF |
896 | |
897 | osb = journal->j_osb; | |
898 | ||
2b4e30fb | 899 | status = jbd2_journal_load(journal->j_journal); |
ccd979bd MF |
900 | if (status < 0) { |
901 | mlog(ML_ERROR, "Failed to load journal!\n"); | |
902 | goto done; | |
903 | } | |
904 | ||
905 | ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num); | |
906 | ||
539d8264 | 907 | status = ocfs2_journal_toggle_dirty(osb, 1, replayed); |
ccd979bd MF |
908 | if (status < 0) { |
909 | mlog_errno(status); | |
910 | goto done; | |
911 | } | |
912 | ||
913 | /* Launch the commit thread */ | |
c271c5c2 SM |
914 | if (!local) { |
915 | osb->commit_task = kthread_run(ocfs2_commit_thread, osb, | |
916 | "ocfs2cmt"); | |
917 | if (IS_ERR(osb->commit_task)) { | |
918 | status = PTR_ERR(osb->commit_task); | |
919 | osb->commit_task = NULL; | |
920 | mlog(ML_ERROR, "unable to launch ocfs2commit thread, " | |
921 | "error=%d", status); | |
922 | goto done; | |
923 | } | |
924 | } else | |
ccd979bd | 925 | osb->commit_task = NULL; |
ccd979bd MF |
926 | |
927 | done: | |
928 | mlog_exit(status); | |
929 | return status; | |
930 | } | |
931 | ||
932 | ||
933 | /* 'full' flag tells us whether we clear out all blocks or if we just | |
934 | * mark the journal clean */ | |
935 | int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full) | |
936 | { | |
937 | int status; | |
938 | ||
939 | mlog_entry_void(); | |
940 | ||
ebdec83b | 941 | BUG_ON(!journal); |
ccd979bd | 942 | |
2b4e30fb | 943 | status = jbd2_journal_wipe(journal->j_journal, full); |
ccd979bd MF |
944 | if (status < 0) { |
945 | mlog_errno(status); | |
946 | goto bail; | |
947 | } | |
948 | ||
539d8264 | 949 | status = ocfs2_journal_toggle_dirty(journal->j_osb, 0, 0); |
ccd979bd MF |
950 | if (status < 0) |
951 | mlog_errno(status); | |
952 | ||
953 | bail: | |
954 | mlog_exit(status); | |
955 | return status; | |
956 | } | |
957 | ||
553abd04 JB |
958 | static int ocfs2_recovery_completed(struct ocfs2_super *osb) |
959 | { | |
960 | int empty; | |
961 | struct ocfs2_recovery_map *rm = osb->recovery_map; | |
962 | ||
963 | spin_lock(&osb->osb_lock); | |
964 | empty = (rm->rm_used == 0); | |
965 | spin_unlock(&osb->osb_lock); | |
966 | ||
967 | return empty; | |
968 | } | |
969 | ||
970 | void ocfs2_wait_for_recovery(struct ocfs2_super *osb) | |
971 | { | |
972 | wait_event(osb->recovery_event, ocfs2_recovery_completed(osb)); | |
973 | } | |
974 | ||
ccd979bd MF |
975 | /* |
976 | * JBD Might read a cached version of another nodes journal file. We | |
977 | * don't want this as this file changes often and we get no | |
978 | * notification on those changes. The only way to be sure that we've | |
979 | * got the most up to date version of those blocks then is to force | |
980 | * read them off disk. Just searching through the buffer cache won't | |
981 | * work as there may be pages backing this file which are still marked | |
982 | * up to date. We know things can't change on this file underneath us | |
983 | * as we have the lock by now :) | |
984 | */ | |
985 | static int ocfs2_force_read_journal(struct inode *inode) | |
986 | { | |
987 | int status = 0; | |
4f902c37 | 988 | int i; |
8110b073 | 989 | u64 v_blkno, p_blkno, p_blocks, num_blocks; |
4f902c37 | 990 | #define CONCURRENT_JOURNAL_FILL 32ULL |
ccd979bd MF |
991 | struct buffer_head *bhs[CONCURRENT_JOURNAL_FILL]; |
992 | ||
993 | mlog_entry_void(); | |
994 | ||
ccd979bd MF |
995 | memset(bhs, 0, sizeof(struct buffer_head *) * CONCURRENT_JOURNAL_FILL); |
996 | ||
8110b073 | 997 | num_blocks = ocfs2_blocks_for_bytes(inode->i_sb, inode->i_size); |
ccd979bd | 998 | v_blkno = 0; |
8110b073 | 999 | while (v_blkno < num_blocks) { |
ccd979bd | 1000 | status = ocfs2_extent_map_get_blocks(inode, v_blkno, |
49cb8d2d | 1001 | &p_blkno, &p_blocks, NULL); |
ccd979bd MF |
1002 | if (status < 0) { |
1003 | mlog_errno(status); | |
1004 | goto bail; | |
1005 | } | |
1006 | ||
1007 | if (p_blocks > CONCURRENT_JOURNAL_FILL) | |
1008 | p_blocks = CONCURRENT_JOURNAL_FILL; | |
1009 | ||
dd4a2c2b MF |
1010 | /* We are reading journal data which should not |
1011 | * be put in the uptodate cache */ | |
da1e9098 JB |
1012 | status = ocfs2_read_blocks_sync(OCFS2_SB(inode->i_sb), |
1013 | p_blkno, p_blocks, bhs); | |
ccd979bd MF |
1014 | if (status < 0) { |
1015 | mlog_errno(status); | |
1016 | goto bail; | |
1017 | } | |
1018 | ||
1019 | for(i = 0; i < p_blocks; i++) { | |
1020 | brelse(bhs[i]); | |
1021 | bhs[i] = NULL; | |
1022 | } | |
1023 | ||
1024 | v_blkno += p_blocks; | |
1025 | } | |
1026 | ||
1027 | bail: | |
1028 | for(i = 0; i < CONCURRENT_JOURNAL_FILL; i++) | |
a81cb88b | 1029 | brelse(bhs[i]); |
ccd979bd MF |
1030 | mlog_exit(status); |
1031 | return status; | |
1032 | } | |
1033 | ||
1034 | struct ocfs2_la_recovery_item { | |
1035 | struct list_head lri_list; | |
1036 | int lri_slot; | |
1037 | struct ocfs2_dinode *lri_la_dinode; | |
1038 | struct ocfs2_dinode *lri_tl_dinode; | |
2205363d | 1039 | struct ocfs2_quota_recovery *lri_qrec; |
ccd979bd MF |
1040 | }; |
1041 | ||
1042 | /* Does the second half of the recovery process. By this point, the | |
1043 | * node is marked clean and can actually be considered recovered, | |
1044 | * hence it's no longer in the recovery map, but there's still some | |
1045 | * cleanup we can do which shouldn't happen within the recovery thread | |
1046 | * as locking in that context becomes very difficult if we are to take | |
1047 | * recovering nodes into account. | |
1048 | * | |
1049 | * NOTE: This function can and will sleep on recovery of other nodes | |
1050 | * during cluster locking, just like any other ocfs2 process. | |
1051 | */ | |
c4028958 | 1052 | void ocfs2_complete_recovery(struct work_struct *work) |
ccd979bd MF |
1053 | { |
1054 | int ret; | |
c4028958 DH |
1055 | struct ocfs2_journal *journal = |
1056 | container_of(work, struct ocfs2_journal, j_recovery_work); | |
1057 | struct ocfs2_super *osb = journal->j_osb; | |
ccd979bd | 1058 | struct ocfs2_dinode *la_dinode, *tl_dinode; |
800deef3 | 1059 | struct ocfs2_la_recovery_item *item, *n; |
2205363d | 1060 | struct ocfs2_quota_recovery *qrec; |
ccd979bd MF |
1061 | LIST_HEAD(tmp_la_list); |
1062 | ||
1063 | mlog_entry_void(); | |
1064 | ||
1065 | mlog(0, "completing recovery from keventd\n"); | |
1066 | ||
1067 | spin_lock(&journal->j_lock); | |
1068 | list_splice_init(&journal->j_la_cleanups, &tmp_la_list); | |
1069 | spin_unlock(&journal->j_lock); | |
1070 | ||
800deef3 | 1071 | list_for_each_entry_safe(item, n, &tmp_la_list, lri_list) { |
ccd979bd MF |
1072 | list_del_init(&item->lri_list); |
1073 | ||
1074 | mlog(0, "Complete recovery for slot %d\n", item->lri_slot); | |
1075 | ||
19ece546 JK |
1076 | ocfs2_wait_on_quotas(osb); |
1077 | ||
ccd979bd MF |
1078 | la_dinode = item->lri_la_dinode; |
1079 | if (la_dinode) { | |
b0697053 | 1080 | mlog(0, "Clean up local alloc %llu\n", |
1ca1a111 | 1081 | (unsigned long long)le64_to_cpu(la_dinode->i_blkno)); |
ccd979bd MF |
1082 | |
1083 | ret = ocfs2_complete_local_alloc_recovery(osb, | |
1084 | la_dinode); | |
1085 | if (ret < 0) | |
1086 | mlog_errno(ret); | |
1087 | ||
1088 | kfree(la_dinode); | |
1089 | } | |
1090 | ||
1091 | tl_dinode = item->lri_tl_dinode; | |
1092 | if (tl_dinode) { | |
b0697053 | 1093 | mlog(0, "Clean up truncate log %llu\n", |
1ca1a111 | 1094 | (unsigned long long)le64_to_cpu(tl_dinode->i_blkno)); |
ccd979bd MF |
1095 | |
1096 | ret = ocfs2_complete_truncate_log_recovery(osb, | |
1097 | tl_dinode); | |
1098 | if (ret < 0) | |
1099 | mlog_errno(ret); | |
1100 | ||
1101 | kfree(tl_dinode); | |
1102 | } | |
1103 | ||
1104 | ret = ocfs2_recover_orphans(osb, item->lri_slot); | |
1105 | if (ret < 0) | |
1106 | mlog_errno(ret); | |
1107 | ||
2205363d JK |
1108 | qrec = item->lri_qrec; |
1109 | if (qrec) { | |
1110 | mlog(0, "Recovering quota files"); | |
1111 | ret = ocfs2_finish_quota_recovery(osb, qrec, | |
1112 | item->lri_slot); | |
1113 | if (ret < 0) | |
1114 | mlog_errno(ret); | |
1115 | /* Recovery info is already freed now */ | |
1116 | } | |
1117 | ||
ccd979bd MF |
1118 | kfree(item); |
1119 | } | |
1120 | ||
1121 | mlog(0, "Recovery completion\n"); | |
1122 | mlog_exit_void(); | |
1123 | } | |
1124 | ||
1125 | /* NOTE: This function always eats your references to la_dinode and | |
1126 | * tl_dinode, either manually on error, or by passing them to | |
1127 | * ocfs2_complete_recovery */ | |
1128 | static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal, | |
1129 | int slot_num, | |
1130 | struct ocfs2_dinode *la_dinode, | |
2205363d JK |
1131 | struct ocfs2_dinode *tl_dinode, |
1132 | struct ocfs2_quota_recovery *qrec) | |
ccd979bd MF |
1133 | { |
1134 | struct ocfs2_la_recovery_item *item; | |
1135 | ||
afae00ab | 1136 | item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS); |
ccd979bd MF |
1137 | if (!item) { |
1138 | /* Though we wish to avoid it, we are in fact safe in | |
1139 | * skipping local alloc cleanup as fsck.ocfs2 is more | |
1140 | * than capable of reclaiming unused space. */ | |
1141 | if (la_dinode) | |
1142 | kfree(la_dinode); | |
1143 | ||
1144 | if (tl_dinode) | |
1145 | kfree(tl_dinode); | |
1146 | ||
2205363d JK |
1147 | if (qrec) |
1148 | ocfs2_free_quota_recovery(qrec); | |
1149 | ||
ccd979bd MF |
1150 | mlog_errno(-ENOMEM); |
1151 | return; | |
1152 | } | |
1153 | ||
1154 | INIT_LIST_HEAD(&item->lri_list); | |
1155 | item->lri_la_dinode = la_dinode; | |
1156 | item->lri_slot = slot_num; | |
1157 | item->lri_tl_dinode = tl_dinode; | |
2205363d | 1158 | item->lri_qrec = qrec; |
ccd979bd MF |
1159 | |
1160 | spin_lock(&journal->j_lock); | |
1161 | list_add_tail(&item->lri_list, &journal->j_la_cleanups); | |
1162 | queue_work(ocfs2_wq, &journal->j_recovery_work); | |
1163 | spin_unlock(&journal->j_lock); | |
1164 | } | |
1165 | ||
1166 | /* Called by the mount code to queue recovery the last part of | |
1167 | * recovery for it's own slot. */ | |
1168 | void ocfs2_complete_mount_recovery(struct ocfs2_super *osb) | |
1169 | { | |
1170 | struct ocfs2_journal *journal = osb->journal; | |
1171 | ||
1172 | if (osb->dirty) { | |
1173 | /* No need to queue up our truncate_log as regular | |
1174 | * cleanup will catch that. */ | |
1175 | ocfs2_queue_recovery_completion(journal, | |
1176 | osb->slot_num, | |
1177 | osb->local_alloc_copy, | |
2205363d | 1178 | NULL, |
ccd979bd MF |
1179 | NULL); |
1180 | ocfs2_schedule_truncate_log_flush(osb, 0); | |
1181 | ||
1182 | osb->local_alloc_copy = NULL; | |
1183 | osb->dirty = 0; | |
1184 | } | |
1185 | } | |
1186 | ||
2205363d JK |
1187 | void ocfs2_complete_quota_recovery(struct ocfs2_super *osb) |
1188 | { | |
1189 | if (osb->quota_rec) { | |
1190 | ocfs2_queue_recovery_completion(osb->journal, | |
1191 | osb->slot_num, | |
1192 | NULL, | |
1193 | NULL, | |
1194 | osb->quota_rec); | |
1195 | osb->quota_rec = NULL; | |
1196 | } | |
1197 | } | |
1198 | ||
ccd979bd MF |
1199 | static int __ocfs2_recovery_thread(void *arg) |
1200 | { | |
2205363d | 1201 | int status, node_num, slot_num; |
ccd979bd | 1202 | struct ocfs2_super *osb = arg; |
553abd04 | 1203 | struct ocfs2_recovery_map *rm = osb->recovery_map; |
2205363d JK |
1204 | int *rm_quota = NULL; |
1205 | int rm_quota_used = 0, i; | |
1206 | struct ocfs2_quota_recovery *qrec; | |
ccd979bd MF |
1207 | |
1208 | mlog_entry_void(); | |
1209 | ||
1210 | status = ocfs2_wait_on_mount(osb); | |
1211 | if (status < 0) { | |
1212 | goto bail; | |
1213 | } | |
1214 | ||
2205363d JK |
1215 | rm_quota = kzalloc(osb->max_slots * sizeof(int), GFP_NOFS); |
1216 | if (!rm_quota) { | |
1217 | status = -ENOMEM; | |
1218 | goto bail; | |
1219 | } | |
ccd979bd MF |
1220 | restart: |
1221 | status = ocfs2_super_lock(osb, 1); | |
1222 | if (status < 0) { | |
1223 | mlog_errno(status); | |
1224 | goto bail; | |
1225 | } | |
1226 | ||
553abd04 JB |
1227 | spin_lock(&osb->osb_lock); |
1228 | while (rm->rm_used) { | |
1229 | /* It's always safe to remove entry zero, as we won't | |
1230 | * clear it until ocfs2_recover_node() has succeeded. */ | |
1231 | node_num = rm->rm_entries[0]; | |
1232 | spin_unlock(&osb->osb_lock); | |
2205363d JK |
1233 | mlog(0, "checking node %d\n", node_num); |
1234 | slot_num = ocfs2_node_num_to_slot(osb, node_num); | |
1235 | if (slot_num == -ENOENT) { | |
1236 | status = 0; | |
1237 | mlog(0, "no slot for this node, so no recovery" | |
1238 | "required.\n"); | |
1239 | goto skip_recovery; | |
1240 | } | |
1241 | mlog(0, "node %d was using slot %d\n", node_num, slot_num); | |
1242 | ||
1243 | /* It is a bit subtle with quota recovery. We cannot do it | |
1244 | * immediately because we have to obtain cluster locks from | |
1245 | * quota files and we also don't want to just skip it because | |
1246 | * then quota usage would be out of sync until some node takes | |
1247 | * the slot. So we remember which nodes need quota recovery | |
1248 | * and when everything else is done, we recover quotas. */ | |
1249 | for (i = 0; i < rm_quota_used && rm_quota[i] != slot_num; i++); | |
1250 | if (i == rm_quota_used) | |
1251 | rm_quota[rm_quota_used++] = slot_num; | |
1252 | ||
1253 | status = ocfs2_recover_node(osb, node_num, slot_num); | |
1254 | skip_recovery: | |
553abd04 JB |
1255 | if (!status) { |
1256 | ocfs2_recovery_map_clear(osb, node_num); | |
1257 | } else { | |
ccd979bd MF |
1258 | mlog(ML_ERROR, |
1259 | "Error %d recovering node %d on device (%u,%u)!\n", | |
1260 | status, node_num, | |
1261 | MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev)); | |
1262 | mlog(ML_ERROR, "Volume requires unmount.\n"); | |
ccd979bd MF |
1263 | } |
1264 | ||
553abd04 | 1265 | spin_lock(&osb->osb_lock); |
ccd979bd | 1266 | } |
553abd04 JB |
1267 | spin_unlock(&osb->osb_lock); |
1268 | mlog(0, "All nodes recovered\n"); | |
1269 | ||
539d8264 SM |
1270 | /* Refresh all journal recovery generations from disk */ |
1271 | status = ocfs2_check_journals_nolocks(osb); | |
1272 | status = (status == -EROFS) ? 0 : status; | |
1273 | if (status < 0) | |
1274 | mlog_errno(status); | |
1275 | ||
2205363d JK |
1276 | /* Now it is right time to recover quotas... We have to do this under |
1277 | * superblock lock so that noone can start using the slot (and crash) | |
1278 | * before we recover it */ | |
1279 | for (i = 0; i < rm_quota_used; i++) { | |
1280 | qrec = ocfs2_begin_quota_recovery(osb, rm_quota[i]); | |
1281 | if (IS_ERR(qrec)) { | |
1282 | status = PTR_ERR(qrec); | |
1283 | mlog_errno(status); | |
1284 | continue; | |
1285 | } | |
1286 | ocfs2_queue_recovery_completion(osb->journal, rm_quota[i], | |
1287 | NULL, NULL, qrec); | |
1288 | } | |
1289 | ||
ccd979bd MF |
1290 | ocfs2_super_unlock(osb, 1); |
1291 | ||
1292 | /* We always run recovery on our own orphan dir - the dead | |
34d024f8 MF |
1293 | * node(s) may have disallowd a previos inode delete. Re-processing |
1294 | * is therefore required. */ | |
ccd979bd | 1295 | ocfs2_queue_recovery_completion(osb->journal, osb->slot_num, NULL, |
2205363d | 1296 | NULL, NULL); |
ccd979bd MF |
1297 | |
1298 | bail: | |
c74ec2f7 | 1299 | mutex_lock(&osb->recovery_lock); |
553abd04 | 1300 | if (!status && !ocfs2_recovery_completed(osb)) { |
c74ec2f7 | 1301 | mutex_unlock(&osb->recovery_lock); |
ccd979bd MF |
1302 | goto restart; |
1303 | } | |
1304 | ||
1305 | osb->recovery_thread_task = NULL; | |
1306 | mb(); /* sync with ocfs2_recovery_thread_running */ | |
1307 | wake_up(&osb->recovery_event); | |
1308 | ||
c74ec2f7 | 1309 | mutex_unlock(&osb->recovery_lock); |
ccd979bd | 1310 | |
2205363d JK |
1311 | if (rm_quota) |
1312 | kfree(rm_quota); | |
1313 | ||
ccd979bd MF |
1314 | mlog_exit(status); |
1315 | /* no one is callint kthread_stop() for us so the kthread() api | |
1316 | * requires that we call do_exit(). And it isn't exported, but | |
1317 | * complete_and_exit() seems to be a minimal wrapper around it. */ | |
1318 | complete_and_exit(NULL, status); | |
1319 | return status; | |
1320 | } | |
1321 | ||
1322 | void ocfs2_recovery_thread(struct ocfs2_super *osb, int node_num) | |
1323 | { | |
1324 | mlog_entry("(node_num=%d, osb->node_num = %d)\n", | |
1325 | node_num, osb->node_num); | |
1326 | ||
c74ec2f7 | 1327 | mutex_lock(&osb->recovery_lock); |
ccd979bd MF |
1328 | if (osb->disable_recovery) |
1329 | goto out; | |
1330 | ||
1331 | /* People waiting on recovery will wait on | |
1332 | * the recovery map to empty. */ | |
553abd04 JB |
1333 | if (ocfs2_recovery_map_set(osb, node_num)) |
1334 | mlog(0, "node %d already in recovery map.\n", node_num); | |
ccd979bd MF |
1335 | |
1336 | mlog(0, "starting recovery thread...\n"); | |
1337 | ||
1338 | if (osb->recovery_thread_task) | |
1339 | goto out; | |
1340 | ||
1341 | osb->recovery_thread_task = kthread_run(__ocfs2_recovery_thread, osb, | |
78427043 | 1342 | "ocfs2rec"); |
ccd979bd MF |
1343 | if (IS_ERR(osb->recovery_thread_task)) { |
1344 | mlog_errno((int)PTR_ERR(osb->recovery_thread_task)); | |
1345 | osb->recovery_thread_task = NULL; | |
1346 | } | |
1347 | ||
1348 | out: | |
c74ec2f7 | 1349 | mutex_unlock(&osb->recovery_lock); |
ccd979bd MF |
1350 | wake_up(&osb->recovery_event); |
1351 | ||
1352 | mlog_exit_void(); | |
1353 | } | |
1354 | ||
539d8264 SM |
1355 | static int ocfs2_read_journal_inode(struct ocfs2_super *osb, |
1356 | int slot_num, | |
1357 | struct buffer_head **bh, | |
1358 | struct inode **ret_inode) | |
1359 | { | |
1360 | int status = -EACCES; | |
1361 | struct inode *inode = NULL; | |
1362 | ||
1363 | BUG_ON(slot_num >= osb->max_slots); | |
1364 | ||
1365 | inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, | |
1366 | slot_num); | |
1367 | if (!inode || is_bad_inode(inode)) { | |
1368 | mlog_errno(status); | |
1369 | goto bail; | |
1370 | } | |
1371 | SET_INODE_JOURNAL(inode); | |
1372 | ||
b657c95c | 1373 | status = ocfs2_read_inode_block_full(inode, bh, OCFS2_BH_IGNORE_CACHE); |
539d8264 SM |
1374 | if (status < 0) { |
1375 | mlog_errno(status); | |
1376 | goto bail; | |
1377 | } | |
1378 | ||
1379 | status = 0; | |
1380 | ||
1381 | bail: | |
1382 | if (inode) { | |
1383 | if (status || !ret_inode) | |
1384 | iput(inode); | |
1385 | else | |
1386 | *ret_inode = inode; | |
1387 | } | |
1388 | return status; | |
1389 | } | |
1390 | ||
ccd979bd MF |
1391 | /* Does the actual journal replay and marks the journal inode as |
1392 | * clean. Will only replay if the journal inode is marked dirty. */ | |
1393 | static int ocfs2_replay_journal(struct ocfs2_super *osb, | |
1394 | int node_num, | |
1395 | int slot_num) | |
1396 | { | |
1397 | int status; | |
1398 | int got_lock = 0; | |
1399 | unsigned int flags; | |
1400 | struct inode *inode = NULL; | |
1401 | struct ocfs2_dinode *fe; | |
1402 | journal_t *journal = NULL; | |
1403 | struct buffer_head *bh = NULL; | |
539d8264 | 1404 | u32 slot_reco_gen; |
ccd979bd | 1405 | |
539d8264 SM |
1406 | status = ocfs2_read_journal_inode(osb, slot_num, &bh, &inode); |
1407 | if (status) { | |
ccd979bd MF |
1408 | mlog_errno(status); |
1409 | goto done; | |
1410 | } | |
539d8264 SM |
1411 | |
1412 | fe = (struct ocfs2_dinode *)bh->b_data; | |
1413 | slot_reco_gen = ocfs2_get_recovery_generation(fe); | |
1414 | brelse(bh); | |
1415 | bh = NULL; | |
1416 | ||
1417 | /* | |
1418 | * As the fs recovery is asynchronous, there is a small chance that | |
1419 | * another node mounted (and recovered) the slot before the recovery | |
1420 | * thread could get the lock. To handle that, we dirty read the journal | |
1421 | * inode for that slot to get the recovery generation. If it is | |
1422 | * different than what we expected, the slot has been recovered. | |
1423 | * If not, it needs recovery. | |
1424 | */ | |
1425 | if (osb->slot_recovery_generations[slot_num] != slot_reco_gen) { | |
1426 | mlog(0, "Slot %u already recovered (old/new=%u/%u)\n", slot_num, | |
1427 | osb->slot_recovery_generations[slot_num], slot_reco_gen); | |
1428 | osb->slot_recovery_generations[slot_num] = slot_reco_gen; | |
1429 | status = -EBUSY; | |
ccd979bd MF |
1430 | goto done; |
1431 | } | |
539d8264 SM |
1432 | |
1433 | /* Continue with recovery as the journal has not yet been recovered */ | |
ccd979bd | 1434 | |
e63aecb6 | 1435 | status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY); |
ccd979bd | 1436 | if (status < 0) { |
e63aecb6 | 1437 | mlog(0, "status returned from ocfs2_inode_lock=%d\n", status); |
ccd979bd MF |
1438 | if (status != -ERESTARTSYS) |
1439 | mlog(ML_ERROR, "Could not lock journal!\n"); | |
1440 | goto done; | |
1441 | } | |
1442 | got_lock = 1; | |
1443 | ||
1444 | fe = (struct ocfs2_dinode *) bh->b_data; | |
1445 | ||
1446 | flags = le32_to_cpu(fe->id1.journal1.ij_flags); | |
539d8264 | 1447 | slot_reco_gen = ocfs2_get_recovery_generation(fe); |
ccd979bd MF |
1448 | |
1449 | if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) { | |
1450 | mlog(0, "No recovery required for node %d\n", node_num); | |
539d8264 SM |
1451 | /* Refresh recovery generation for the slot */ |
1452 | osb->slot_recovery_generations[slot_num] = slot_reco_gen; | |
ccd979bd MF |
1453 | goto done; |
1454 | } | |
1455 | ||
1456 | mlog(ML_NOTICE, "Recovering node %d from slot %d on device (%u,%u)\n", | |
1457 | node_num, slot_num, | |
1458 | MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev)); | |
1459 | ||
1460 | OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters); | |
1461 | ||
1462 | status = ocfs2_force_read_journal(inode); | |
1463 | if (status < 0) { | |
1464 | mlog_errno(status); | |
1465 | goto done; | |
1466 | } | |
1467 | ||
1468 | mlog(0, "calling journal_init_inode\n"); | |
2b4e30fb | 1469 | journal = jbd2_journal_init_inode(inode); |
ccd979bd MF |
1470 | if (journal == NULL) { |
1471 | mlog(ML_ERROR, "Linux journal layer error\n"); | |
1472 | status = -EIO; | |
1473 | goto done; | |
1474 | } | |
1475 | ||
2b4e30fb | 1476 | status = jbd2_journal_load(journal); |
ccd979bd MF |
1477 | if (status < 0) { |
1478 | mlog_errno(status); | |
1479 | if (!igrab(inode)) | |
1480 | BUG(); | |
2b4e30fb | 1481 | jbd2_journal_destroy(journal); |
ccd979bd MF |
1482 | goto done; |
1483 | } | |
1484 | ||
1485 | ocfs2_clear_journal_error(osb->sb, journal, slot_num); | |
1486 | ||
1487 | /* wipe the journal */ | |
1488 | mlog(0, "flushing the journal.\n"); | |
2b4e30fb JB |
1489 | jbd2_journal_lock_updates(journal); |
1490 | status = jbd2_journal_flush(journal); | |
1491 | jbd2_journal_unlock_updates(journal); | |
ccd979bd MF |
1492 | if (status < 0) |
1493 | mlog_errno(status); | |
1494 | ||
1495 | /* This will mark the node clean */ | |
1496 | flags = le32_to_cpu(fe->id1.journal1.ij_flags); | |
1497 | flags &= ~OCFS2_JOURNAL_DIRTY_FL; | |
1498 | fe->id1.journal1.ij_flags = cpu_to_le32(flags); | |
1499 | ||
539d8264 SM |
1500 | /* Increment recovery generation to indicate successful recovery */ |
1501 | ocfs2_bump_recovery_generation(fe); | |
1502 | osb->slot_recovery_generations[slot_num] = | |
1503 | ocfs2_get_recovery_generation(fe); | |
1504 | ||
13723d00 | 1505 | ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check); |
ccd979bd MF |
1506 | status = ocfs2_write_block(osb, bh, inode); |
1507 | if (status < 0) | |
1508 | mlog_errno(status); | |
1509 | ||
1510 | if (!igrab(inode)) | |
1511 | BUG(); | |
1512 | ||
2b4e30fb | 1513 | jbd2_journal_destroy(journal); |
ccd979bd MF |
1514 | |
1515 | done: | |
1516 | /* drop the lock on this nodes journal */ | |
1517 | if (got_lock) | |
e63aecb6 | 1518 | ocfs2_inode_unlock(inode, 1); |
ccd979bd MF |
1519 | |
1520 | if (inode) | |
1521 | iput(inode); | |
1522 | ||
a81cb88b | 1523 | brelse(bh); |
ccd979bd MF |
1524 | |
1525 | mlog_exit(status); | |
1526 | return status; | |
1527 | } | |
1528 | ||
1529 | /* | |
1530 | * Do the most important parts of node recovery: | |
1531 | * - Replay it's journal | |
1532 | * - Stamp a clean local allocator file | |
1533 | * - Stamp a clean truncate log | |
1534 | * - Mark the node clean | |
1535 | * | |
1536 | * If this function completes without error, a node in OCFS2 can be | |
1537 | * said to have been safely recovered. As a result, failure during the | |
1538 | * second part of a nodes recovery process (local alloc recovery) is | |
1539 | * far less concerning. | |
1540 | */ | |
1541 | static int ocfs2_recover_node(struct ocfs2_super *osb, | |
2205363d | 1542 | int node_num, int slot_num) |
ccd979bd MF |
1543 | { |
1544 | int status = 0; | |
ccd979bd MF |
1545 | struct ocfs2_dinode *la_copy = NULL; |
1546 | struct ocfs2_dinode *tl_copy = NULL; | |
1547 | ||
2205363d JK |
1548 | mlog_entry("(node_num=%d, slot_num=%d, osb->node_num = %d)\n", |
1549 | node_num, slot_num, osb->node_num); | |
ccd979bd MF |
1550 | |
1551 | /* Should not ever be called to recover ourselves -- in that | |
1552 | * case we should've called ocfs2_journal_load instead. */ | |
ebdec83b | 1553 | BUG_ON(osb->node_num == node_num); |
ccd979bd | 1554 | |
ccd979bd MF |
1555 | status = ocfs2_replay_journal(osb, node_num, slot_num); |
1556 | if (status < 0) { | |
539d8264 SM |
1557 | if (status == -EBUSY) { |
1558 | mlog(0, "Skipping recovery for slot %u (node %u) " | |
1559 | "as another node has recovered it\n", slot_num, | |
1560 | node_num); | |
1561 | status = 0; | |
1562 | goto done; | |
1563 | } | |
ccd979bd MF |
1564 | mlog_errno(status); |
1565 | goto done; | |
1566 | } | |
1567 | ||
1568 | /* Stamp a clean local alloc file AFTER recovering the journal... */ | |
1569 | status = ocfs2_begin_local_alloc_recovery(osb, slot_num, &la_copy); | |
1570 | if (status < 0) { | |
1571 | mlog_errno(status); | |
1572 | goto done; | |
1573 | } | |
1574 | ||
1575 | /* An error from begin_truncate_log_recovery is not | |
1576 | * serious enough to warrant halting the rest of | |
1577 | * recovery. */ | |
1578 | status = ocfs2_begin_truncate_log_recovery(osb, slot_num, &tl_copy); | |
1579 | if (status < 0) | |
1580 | mlog_errno(status); | |
1581 | ||
1582 | /* Likewise, this would be a strange but ultimately not so | |
1583 | * harmful place to get an error... */ | |
8e8a4603 | 1584 | status = ocfs2_clear_slot(osb, slot_num); |
ccd979bd MF |
1585 | if (status < 0) |
1586 | mlog_errno(status); | |
1587 | ||
1588 | /* This will kfree the memory pointed to by la_copy and tl_copy */ | |
1589 | ocfs2_queue_recovery_completion(osb->journal, slot_num, la_copy, | |
2205363d | 1590 | tl_copy, NULL); |
ccd979bd MF |
1591 | |
1592 | status = 0; | |
1593 | done: | |
1594 | ||
1595 | mlog_exit(status); | |
1596 | return status; | |
1597 | } | |
1598 | ||
1599 | /* Test node liveness by trylocking his journal. If we get the lock, | |
1600 | * we drop it here. Return 0 if we got the lock, -EAGAIN if node is | |
1601 | * still alive (we couldn't get the lock) and < 0 on error. */ | |
1602 | static int ocfs2_trylock_journal(struct ocfs2_super *osb, | |
1603 | int slot_num) | |
1604 | { | |
1605 | int status, flags; | |
1606 | struct inode *inode = NULL; | |
1607 | ||
1608 | inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, | |
1609 | slot_num); | |
1610 | if (inode == NULL) { | |
1611 | mlog(ML_ERROR, "access error\n"); | |
1612 | status = -EACCES; | |
1613 | goto bail; | |
1614 | } | |
1615 | if (is_bad_inode(inode)) { | |
1616 | mlog(ML_ERROR, "access error (bad inode)\n"); | |
1617 | iput(inode); | |
1618 | inode = NULL; | |
1619 | status = -EACCES; | |
1620 | goto bail; | |
1621 | } | |
1622 | SET_INODE_JOURNAL(inode); | |
1623 | ||
1624 | flags = OCFS2_META_LOCK_RECOVERY | OCFS2_META_LOCK_NOQUEUE; | |
e63aecb6 | 1625 | status = ocfs2_inode_lock_full(inode, NULL, 1, flags); |
ccd979bd MF |
1626 | if (status < 0) { |
1627 | if (status != -EAGAIN) | |
1628 | mlog_errno(status); | |
1629 | goto bail; | |
1630 | } | |
1631 | ||
e63aecb6 | 1632 | ocfs2_inode_unlock(inode, 1); |
ccd979bd MF |
1633 | bail: |
1634 | if (inode) | |
1635 | iput(inode); | |
1636 | ||
1637 | return status; | |
1638 | } | |
1639 | ||
1640 | /* Call this underneath ocfs2_super_lock. It also assumes that the | |
1641 | * slot info struct has been updated from disk. */ | |
1642 | int ocfs2_mark_dead_nodes(struct ocfs2_super *osb) | |
1643 | { | |
d85b20e4 JB |
1644 | unsigned int node_num; |
1645 | int status, i; | |
a1af7d15 | 1646 | u32 gen; |
539d8264 SM |
1647 | struct buffer_head *bh = NULL; |
1648 | struct ocfs2_dinode *di; | |
ccd979bd MF |
1649 | |
1650 | /* This is called with the super block cluster lock, so we | |
1651 | * know that the slot map can't change underneath us. */ | |
1652 | ||
d85b20e4 | 1653 | for (i = 0; i < osb->max_slots; i++) { |
539d8264 SM |
1654 | /* Read journal inode to get the recovery generation */ |
1655 | status = ocfs2_read_journal_inode(osb, i, &bh, NULL); | |
1656 | if (status) { | |
1657 | mlog_errno(status); | |
1658 | goto bail; | |
1659 | } | |
1660 | di = (struct ocfs2_dinode *)bh->b_data; | |
a1af7d15 | 1661 | gen = ocfs2_get_recovery_generation(di); |
539d8264 SM |
1662 | brelse(bh); |
1663 | bh = NULL; | |
1664 | ||
a1af7d15 MF |
1665 | spin_lock(&osb->osb_lock); |
1666 | osb->slot_recovery_generations[i] = gen; | |
1667 | ||
539d8264 SM |
1668 | mlog(0, "Slot %u recovery generation is %u\n", i, |
1669 | osb->slot_recovery_generations[i]); | |
1670 | ||
a1af7d15 MF |
1671 | if (i == osb->slot_num) { |
1672 | spin_unlock(&osb->osb_lock); | |
ccd979bd | 1673 | continue; |
a1af7d15 | 1674 | } |
d85b20e4 JB |
1675 | |
1676 | status = ocfs2_slot_to_node_num_locked(osb, i, &node_num); | |
a1af7d15 MF |
1677 | if (status == -ENOENT) { |
1678 | spin_unlock(&osb->osb_lock); | |
ccd979bd | 1679 | continue; |
a1af7d15 | 1680 | } |
ccd979bd | 1681 | |
a1af7d15 MF |
1682 | if (__ocfs2_recovery_map_test(osb, node_num)) { |
1683 | spin_unlock(&osb->osb_lock); | |
ccd979bd | 1684 | continue; |
a1af7d15 | 1685 | } |
d85b20e4 | 1686 | spin_unlock(&osb->osb_lock); |
ccd979bd MF |
1687 | |
1688 | /* Ok, we have a slot occupied by another node which | |
1689 | * is not in the recovery map. We trylock his journal | |
1690 | * file here to test if he's alive. */ | |
1691 | status = ocfs2_trylock_journal(osb, i); | |
1692 | if (!status) { | |
1693 | /* Since we're called from mount, we know that | |
1694 | * the recovery thread can't race us on | |
1695 | * setting / checking the recovery bits. */ | |
1696 | ocfs2_recovery_thread(osb, node_num); | |
1697 | } else if ((status < 0) && (status != -EAGAIN)) { | |
1698 | mlog_errno(status); | |
1699 | goto bail; | |
1700 | } | |
ccd979bd | 1701 | } |
ccd979bd MF |
1702 | |
1703 | status = 0; | |
1704 | bail: | |
1705 | mlog_exit(status); | |
1706 | return status; | |
1707 | } | |
1708 | ||
5eae5b96 MF |
1709 | struct ocfs2_orphan_filldir_priv { |
1710 | struct inode *head; | |
1711 | struct ocfs2_super *osb; | |
1712 | }; | |
1713 | ||
1714 | static int ocfs2_orphan_filldir(void *priv, const char *name, int name_len, | |
1715 | loff_t pos, u64 ino, unsigned type) | |
1716 | { | |
1717 | struct ocfs2_orphan_filldir_priv *p = priv; | |
1718 | struct inode *iter; | |
1719 | ||
1720 | if (name_len == 1 && !strncmp(".", name, 1)) | |
1721 | return 0; | |
1722 | if (name_len == 2 && !strncmp("..", name, 2)) | |
1723 | return 0; | |
1724 | ||
1725 | /* Skip bad inodes so that recovery can continue */ | |
1726 | iter = ocfs2_iget(p->osb, ino, | |
5fa0613e | 1727 | OCFS2_FI_FLAG_ORPHAN_RECOVERY, 0); |
5eae5b96 MF |
1728 | if (IS_ERR(iter)) |
1729 | return 0; | |
1730 | ||
1731 | mlog(0, "queue orphan %llu\n", | |
1732 | (unsigned long long)OCFS2_I(iter)->ip_blkno); | |
1733 | /* No locking is required for the next_orphan queue as there | |
1734 | * is only ever a single process doing orphan recovery. */ | |
1735 | OCFS2_I(iter)->ip_next_orphan = p->head; | |
1736 | p->head = iter; | |
1737 | ||
1738 | return 0; | |
1739 | } | |
1740 | ||
b4df6ed8 MF |
1741 | static int ocfs2_queue_orphans(struct ocfs2_super *osb, |
1742 | int slot, | |
1743 | struct inode **head) | |
ccd979bd | 1744 | { |
b4df6ed8 | 1745 | int status; |
ccd979bd | 1746 | struct inode *orphan_dir_inode = NULL; |
5eae5b96 MF |
1747 | struct ocfs2_orphan_filldir_priv priv; |
1748 | loff_t pos = 0; | |
1749 | ||
1750 | priv.osb = osb; | |
1751 | priv.head = *head; | |
ccd979bd MF |
1752 | |
1753 | orphan_dir_inode = ocfs2_get_system_file_inode(osb, | |
1754 | ORPHAN_DIR_SYSTEM_INODE, | |
1755 | slot); | |
1756 | if (!orphan_dir_inode) { | |
1757 | status = -ENOENT; | |
1758 | mlog_errno(status); | |
b4df6ed8 MF |
1759 | return status; |
1760 | } | |
ccd979bd | 1761 | |
1b1dcc1b | 1762 | mutex_lock(&orphan_dir_inode->i_mutex); |
e63aecb6 | 1763 | status = ocfs2_inode_lock(orphan_dir_inode, NULL, 0); |
ccd979bd | 1764 | if (status < 0) { |
ccd979bd MF |
1765 | mlog_errno(status); |
1766 | goto out; | |
1767 | } | |
ccd979bd | 1768 | |
5eae5b96 MF |
1769 | status = ocfs2_dir_foreach(orphan_dir_inode, &pos, &priv, |
1770 | ocfs2_orphan_filldir); | |
1771 | if (status) { | |
1772 | mlog_errno(status); | |
a86370fb | 1773 | goto out_cluster; |
ccd979bd | 1774 | } |
ccd979bd | 1775 | |
5eae5b96 MF |
1776 | *head = priv.head; |
1777 | ||
a86370fb | 1778 | out_cluster: |
e63aecb6 | 1779 | ocfs2_inode_unlock(orphan_dir_inode, 0); |
b4df6ed8 MF |
1780 | out: |
1781 | mutex_unlock(&orphan_dir_inode->i_mutex); | |
ccd979bd | 1782 | iput(orphan_dir_inode); |
b4df6ed8 MF |
1783 | return status; |
1784 | } | |
1785 | ||
1786 | static int ocfs2_orphan_recovery_can_continue(struct ocfs2_super *osb, | |
1787 | int slot) | |
1788 | { | |
1789 | int ret; | |
1790 | ||
1791 | spin_lock(&osb->osb_lock); | |
1792 | ret = !osb->osb_orphan_wipes[slot]; | |
1793 | spin_unlock(&osb->osb_lock); | |
1794 | return ret; | |
1795 | } | |
1796 | ||
1797 | static void ocfs2_mark_recovering_orphan_dir(struct ocfs2_super *osb, | |
1798 | int slot) | |
1799 | { | |
1800 | spin_lock(&osb->osb_lock); | |
1801 | /* Mark ourselves such that new processes in delete_inode() | |
1802 | * know to quit early. */ | |
1803 | ocfs2_node_map_set_bit(osb, &osb->osb_recovering_orphan_dirs, slot); | |
1804 | while (osb->osb_orphan_wipes[slot]) { | |
1805 | /* If any processes are already in the middle of an | |
1806 | * orphan wipe on this dir, then we need to wait for | |
1807 | * them. */ | |
1808 | spin_unlock(&osb->osb_lock); | |
1809 | wait_event_interruptible(osb->osb_wipe_event, | |
1810 | ocfs2_orphan_recovery_can_continue(osb, slot)); | |
1811 | spin_lock(&osb->osb_lock); | |
1812 | } | |
1813 | spin_unlock(&osb->osb_lock); | |
1814 | } | |
1815 | ||
1816 | static void ocfs2_clear_recovering_orphan_dir(struct ocfs2_super *osb, | |
1817 | int slot) | |
1818 | { | |
1819 | ocfs2_node_map_clear_bit(osb, &osb->osb_recovering_orphan_dirs, slot); | |
1820 | } | |
1821 | ||
1822 | /* | |
1823 | * Orphan recovery. Each mounted node has it's own orphan dir which we | |
1824 | * must run during recovery. Our strategy here is to build a list of | |
1825 | * the inodes in the orphan dir and iget/iput them. The VFS does | |
1826 | * (most) of the rest of the work. | |
1827 | * | |
1828 | * Orphan recovery can happen at any time, not just mount so we have a | |
1829 | * couple of extra considerations. | |
1830 | * | |
1831 | * - We grab as many inodes as we can under the orphan dir lock - | |
1832 | * doing iget() outside the orphan dir risks getting a reference on | |
1833 | * an invalid inode. | |
1834 | * - We must be sure not to deadlock with other processes on the | |
1835 | * system wanting to run delete_inode(). This can happen when they go | |
1836 | * to lock the orphan dir and the orphan recovery process attempts to | |
1837 | * iget() inside the orphan dir lock. This can be avoided by | |
1838 | * advertising our state to ocfs2_delete_inode(). | |
1839 | */ | |
1840 | static int ocfs2_recover_orphans(struct ocfs2_super *osb, | |
1841 | int slot) | |
1842 | { | |
1843 | int ret = 0; | |
1844 | struct inode *inode = NULL; | |
1845 | struct inode *iter; | |
1846 | struct ocfs2_inode_info *oi; | |
1847 | ||
1848 | mlog(0, "Recover inodes from orphan dir in slot %d\n", slot); | |
1849 | ||
1850 | ocfs2_mark_recovering_orphan_dir(osb, slot); | |
1851 | ret = ocfs2_queue_orphans(osb, slot, &inode); | |
1852 | ocfs2_clear_recovering_orphan_dir(osb, slot); | |
1853 | ||
1854 | /* Error here should be noted, but we want to continue with as | |
1855 | * many queued inodes as we've got. */ | |
1856 | if (ret) | |
1857 | mlog_errno(ret); | |
ccd979bd MF |
1858 | |
1859 | while (inode) { | |
1860 | oi = OCFS2_I(inode); | |
b0697053 | 1861 | mlog(0, "iput orphan %llu\n", (unsigned long long)oi->ip_blkno); |
ccd979bd MF |
1862 | |
1863 | iter = oi->ip_next_orphan; | |
1864 | ||
1865 | spin_lock(&oi->ip_lock); | |
34d024f8 MF |
1866 | /* The remote delete code may have set these on the |
1867 | * assumption that the other node would wipe them | |
1868 | * successfully. If they are still in the node's | |
1869 | * orphan dir, we need to reset that state. */ | |
ccd979bd MF |
1870 | oi->ip_flags &= ~(OCFS2_INODE_DELETED|OCFS2_INODE_SKIP_DELETE); |
1871 | ||
1872 | /* Set the proper information to get us going into | |
1873 | * ocfs2_delete_inode. */ | |
1874 | oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED; | |
ccd979bd MF |
1875 | spin_unlock(&oi->ip_lock); |
1876 | ||
1877 | iput(inode); | |
1878 | ||
1879 | inode = iter; | |
1880 | } | |
1881 | ||
b4df6ed8 | 1882 | return ret; |
ccd979bd MF |
1883 | } |
1884 | ||
19ece546 | 1885 | static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota) |
ccd979bd MF |
1886 | { |
1887 | /* This check is good because ocfs2 will wait on our recovery | |
1888 | * thread before changing it to something other than MOUNTED | |
1889 | * or DISABLED. */ | |
1890 | wait_event(osb->osb_mount_event, | |
19ece546 JK |
1891 | (!quota && atomic_read(&osb->vol_state) == VOLUME_MOUNTED) || |
1892 | atomic_read(&osb->vol_state) == VOLUME_MOUNTED_QUOTAS || | |
ccd979bd MF |
1893 | atomic_read(&osb->vol_state) == VOLUME_DISABLED); |
1894 | ||
1895 | /* If there's an error on mount, then we may never get to the | |
1896 | * MOUNTED flag, but this is set right before | |
1897 | * dismount_volume() so we can trust it. */ | |
1898 | if (atomic_read(&osb->vol_state) == VOLUME_DISABLED) { | |
1899 | mlog(0, "mount error, exiting!\n"); | |
1900 | return -EBUSY; | |
1901 | } | |
1902 | ||
1903 | return 0; | |
1904 | } | |
1905 | ||
1906 | static int ocfs2_commit_thread(void *arg) | |
1907 | { | |
1908 | int status; | |
1909 | struct ocfs2_super *osb = arg; | |
1910 | struct ocfs2_journal *journal = osb->journal; | |
1911 | ||
1912 | /* we can trust j_num_trans here because _should_stop() is only set in | |
1913 | * shutdown and nobody other than ourselves should be able to start | |
1914 | * transactions. committing on shutdown might take a few iterations | |
1915 | * as final transactions put deleted inodes on the list */ | |
1916 | while (!(kthread_should_stop() && | |
1917 | atomic_read(&journal->j_num_trans) == 0)) { | |
1918 | ||
745ae8ba MF |
1919 | wait_event_interruptible(osb->checkpoint_event, |
1920 | atomic_read(&journal->j_num_trans) | |
1921 | || kthread_should_stop()); | |
ccd979bd MF |
1922 | |
1923 | status = ocfs2_commit_cache(osb); | |
1924 | if (status < 0) | |
1925 | mlog_errno(status); | |
1926 | ||
1927 | if (kthread_should_stop() && atomic_read(&journal->j_num_trans)){ | |
1928 | mlog(ML_KTHREAD, | |
1929 | "commit_thread: %u transactions pending on " | |
1930 | "shutdown\n", | |
1931 | atomic_read(&journal->j_num_trans)); | |
1932 | } | |
1933 | } | |
1934 | ||
1935 | return 0; | |
1936 | } | |
1937 | ||
539d8264 SM |
1938 | /* Reads all the journal inodes without taking any cluster locks. Used |
1939 | * for hard readonly access to determine whether any journal requires | |
1940 | * recovery. Also used to refresh the recovery generation numbers after | |
1941 | * a journal has been recovered by another node. | |
1942 | */ | |
ccd979bd MF |
1943 | int ocfs2_check_journals_nolocks(struct ocfs2_super *osb) |
1944 | { | |
1945 | int ret = 0; | |
1946 | unsigned int slot; | |
539d8264 | 1947 | struct buffer_head *di_bh = NULL; |
ccd979bd | 1948 | struct ocfs2_dinode *di; |
539d8264 | 1949 | int journal_dirty = 0; |
ccd979bd MF |
1950 | |
1951 | for(slot = 0; slot < osb->max_slots; slot++) { | |
539d8264 SM |
1952 | ret = ocfs2_read_journal_inode(osb, slot, &di_bh, NULL); |
1953 | if (ret) { | |
ccd979bd MF |
1954 | mlog_errno(ret); |
1955 | goto out; | |
1956 | } | |
1957 | ||
1958 | di = (struct ocfs2_dinode *) di_bh->b_data; | |
1959 | ||
539d8264 SM |
1960 | osb->slot_recovery_generations[slot] = |
1961 | ocfs2_get_recovery_generation(di); | |
1962 | ||
ccd979bd MF |
1963 | if (le32_to_cpu(di->id1.journal1.ij_flags) & |
1964 | OCFS2_JOURNAL_DIRTY_FL) | |
539d8264 | 1965 | journal_dirty = 1; |
ccd979bd MF |
1966 | |
1967 | brelse(di_bh); | |
539d8264 | 1968 | di_bh = NULL; |
ccd979bd MF |
1969 | } |
1970 | ||
1971 | out: | |
539d8264 SM |
1972 | if (journal_dirty) |
1973 | ret = -EROFS; | |
ccd979bd MF |
1974 | return ret; |
1975 | } |