]>
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" | |
316f4b9f | 38 | #include "dir.h" |
ccd979bd MF |
39 | #include "dlmglue.h" |
40 | #include "extent_map.h" | |
41 | #include "heartbeat.h" | |
42 | #include "inode.h" | |
43 | #include "journal.h" | |
44 | #include "localalloc.h" | |
ccd979bd MF |
45 | #include "slot_map.h" |
46 | #include "super.h" | |
ccd979bd | 47 | #include "sysfile.h" |
2205363d | 48 | #include "quota.h" |
ccd979bd MF |
49 | |
50 | #include "buffer_head_io.h" | |
51 | ||
34af946a | 52 | DEFINE_SPINLOCK(trans_inc_lock); |
ccd979bd MF |
53 | |
54 | static int ocfs2_force_read_journal(struct inode *inode); | |
55 | static int ocfs2_recover_node(struct ocfs2_super *osb, | |
2205363d | 56 | int node_num, int slot_num); |
ccd979bd MF |
57 | static int __ocfs2_recovery_thread(void *arg); |
58 | static int ocfs2_commit_cache(struct ocfs2_super *osb); | |
59 | static int ocfs2_wait_on_mount(struct ocfs2_super *osb); | |
ccd979bd | 60 | static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb, |
539d8264 | 61 | int dirty, int replayed); |
ccd979bd MF |
62 | static int ocfs2_trylock_journal(struct ocfs2_super *osb, |
63 | int slot_num); | |
64 | static int ocfs2_recover_orphans(struct ocfs2_super *osb, | |
65 | int slot); | |
66 | static int ocfs2_commit_thread(void *arg); | |
67 | ||
553abd04 JB |
68 | |
69 | /* | |
70 | * The recovery_list is a simple linked list of node numbers to recover. | |
71 | * It is protected by the recovery_lock. | |
72 | */ | |
73 | ||
74 | struct ocfs2_recovery_map { | |
fc881fa0 | 75 | unsigned int rm_used; |
553abd04 JB |
76 | unsigned int *rm_entries; |
77 | }; | |
78 | ||
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 | ||
1fabe148 | 361 | int ocfs2_journal_access(handle_t *handle, |
ccd979bd MF |
362 | struct inode *inode, |
363 | struct buffer_head *bh, | |
364 | int type) | |
365 | { | |
366 | int status; | |
367 | ||
368 | BUG_ON(!inode); | |
369 | BUG_ON(!handle); | |
370 | BUG_ON(!bh); | |
ccd979bd | 371 | |
205f87f6 | 372 | mlog_entry("bh->b_blocknr=%llu, type=%d (\"%s\"), bh->b_size = %zu\n", |
ccd979bd MF |
373 | (unsigned long long)bh->b_blocknr, type, |
374 | (type == OCFS2_JOURNAL_ACCESS_CREATE) ? | |
375 | "OCFS2_JOURNAL_ACCESS_CREATE" : | |
376 | "OCFS2_JOURNAL_ACCESS_WRITE", | |
377 | bh->b_size); | |
378 | ||
379 | /* we can safely remove this assertion after testing. */ | |
380 | if (!buffer_uptodate(bh)) { | |
381 | mlog(ML_ERROR, "giving me a buffer that's not uptodate!\n"); | |
382 | mlog(ML_ERROR, "b_blocknr=%llu\n", | |
383 | (unsigned long long)bh->b_blocknr); | |
384 | BUG(); | |
385 | } | |
386 | ||
387 | /* Set the current transaction information on the inode so | |
388 | * that the locking code knows whether it can drop it's locks | |
389 | * on this inode or not. We're protected from the commit | |
390 | * thread updating the current transaction id until | |
391 | * ocfs2_commit_trans() because ocfs2_start_trans() took | |
392 | * j_trans_barrier for us. */ | |
393 | ocfs2_set_inode_lock_trans(OCFS2_SB(inode->i_sb)->journal, inode); | |
394 | ||
251b6ecc | 395 | mutex_lock(&OCFS2_I(inode)->ip_io_mutex); |
ccd979bd MF |
396 | switch (type) { |
397 | case OCFS2_JOURNAL_ACCESS_CREATE: | |
398 | case OCFS2_JOURNAL_ACCESS_WRITE: | |
2b4e30fb | 399 | status = jbd2_journal_get_write_access(handle, bh); |
ccd979bd MF |
400 | break; |
401 | ||
402 | case OCFS2_JOURNAL_ACCESS_UNDO: | |
2b4e30fb | 403 | status = jbd2_journal_get_undo_access(handle, bh); |
ccd979bd MF |
404 | break; |
405 | ||
406 | default: | |
407 | status = -EINVAL; | |
408 | mlog(ML_ERROR, "Uknown access type!\n"); | |
409 | } | |
251b6ecc | 410 | mutex_unlock(&OCFS2_I(inode)->ip_io_mutex); |
ccd979bd MF |
411 | |
412 | if (status < 0) | |
413 | mlog(ML_ERROR, "Error %d getting %d access to buffer!\n", | |
414 | status, type); | |
415 | ||
416 | mlog_exit(status); | |
417 | return status; | |
418 | } | |
419 | ||
1fabe148 | 420 | int ocfs2_journal_dirty(handle_t *handle, |
ccd979bd MF |
421 | struct buffer_head *bh) |
422 | { | |
423 | int status; | |
424 | ||
ccd979bd MF |
425 | mlog_entry("(bh->b_blocknr=%llu)\n", |
426 | (unsigned long long)bh->b_blocknr); | |
427 | ||
2b4e30fb | 428 | status = jbd2_journal_dirty_metadata(handle, bh); |
ccd979bd MF |
429 | if (status < 0) |
430 | mlog(ML_ERROR, "Could not dirty metadata buffer. " | |
431 | "(bh->b_blocknr=%llu)\n", | |
432 | (unsigned long long)bh->b_blocknr); | |
433 | ||
434 | mlog_exit(status); | |
435 | return status; | |
436 | } | |
437 | ||
2b4e30fb | 438 | #define OCFS2_DEFAULT_COMMIT_INTERVAL (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE) |
ccd979bd MF |
439 | |
440 | void ocfs2_set_journal_params(struct ocfs2_super *osb) | |
441 | { | |
442 | journal_t *journal = osb->journal->j_journal; | |
d147b3d6 MF |
443 | unsigned long commit_interval = OCFS2_DEFAULT_COMMIT_INTERVAL; |
444 | ||
445 | if (osb->osb_commit_interval) | |
446 | commit_interval = osb->osb_commit_interval; | |
ccd979bd MF |
447 | |
448 | spin_lock(&journal->j_state_lock); | |
d147b3d6 | 449 | journal->j_commit_interval = commit_interval; |
ccd979bd | 450 | if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER) |
2b4e30fb | 451 | journal->j_flags |= JBD2_BARRIER; |
ccd979bd | 452 | else |
2b4e30fb | 453 | journal->j_flags &= ~JBD2_BARRIER; |
ccd979bd MF |
454 | spin_unlock(&journal->j_state_lock); |
455 | } | |
456 | ||
457 | int ocfs2_journal_init(struct ocfs2_journal *journal, int *dirty) | |
458 | { | |
459 | int status = -1; | |
460 | struct inode *inode = NULL; /* the journal inode */ | |
461 | journal_t *j_journal = NULL; | |
462 | struct ocfs2_dinode *di = NULL; | |
463 | struct buffer_head *bh = NULL; | |
464 | struct ocfs2_super *osb; | |
e63aecb6 | 465 | int inode_lock = 0; |
ccd979bd MF |
466 | |
467 | mlog_entry_void(); | |
468 | ||
469 | BUG_ON(!journal); | |
470 | ||
471 | osb = journal->j_osb; | |
472 | ||
473 | /* already have the inode for our journal */ | |
474 | inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, | |
475 | osb->slot_num); | |
476 | if (inode == NULL) { | |
477 | status = -EACCES; | |
478 | mlog_errno(status); | |
479 | goto done; | |
480 | } | |
481 | if (is_bad_inode(inode)) { | |
482 | mlog(ML_ERROR, "access error (bad inode)\n"); | |
483 | iput(inode); | |
484 | inode = NULL; | |
485 | status = -EACCES; | |
486 | goto done; | |
487 | } | |
488 | ||
489 | SET_INODE_JOURNAL(inode); | |
490 | OCFS2_I(inode)->ip_open_count++; | |
491 | ||
6eff5790 MF |
492 | /* Skip recovery waits here - journal inode metadata never |
493 | * changes in a live cluster so it can be considered an | |
494 | * exception to the rule. */ | |
e63aecb6 | 495 | status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY); |
ccd979bd MF |
496 | if (status < 0) { |
497 | if (status != -ERESTARTSYS) | |
498 | mlog(ML_ERROR, "Could not get lock on journal!\n"); | |
499 | goto done; | |
500 | } | |
501 | ||
e63aecb6 | 502 | inode_lock = 1; |
ccd979bd MF |
503 | di = (struct ocfs2_dinode *)bh->b_data; |
504 | ||
505 | if (inode->i_size < OCFS2_MIN_JOURNAL_SIZE) { | |
506 | mlog(ML_ERROR, "Journal file size (%lld) is too small!\n", | |
507 | inode->i_size); | |
508 | status = -EINVAL; | |
509 | goto done; | |
510 | } | |
511 | ||
512 | mlog(0, "inode->i_size = %lld\n", inode->i_size); | |
5515eff8 AM |
513 | mlog(0, "inode->i_blocks = %llu\n", |
514 | (unsigned long long)inode->i_blocks); | |
ccd979bd MF |
515 | mlog(0, "inode->ip_clusters = %u\n", OCFS2_I(inode)->ip_clusters); |
516 | ||
517 | /* call the kernels journal init function now */ | |
2b4e30fb | 518 | j_journal = jbd2_journal_init_inode(inode); |
ccd979bd MF |
519 | if (j_journal == NULL) { |
520 | mlog(ML_ERROR, "Linux journal layer error\n"); | |
521 | status = -EINVAL; | |
522 | goto done; | |
523 | } | |
524 | ||
2b4e30fb | 525 | mlog(0, "Returned from jbd2_journal_init_inode\n"); |
ccd979bd MF |
526 | mlog(0, "j_journal->j_maxlen = %u\n", j_journal->j_maxlen); |
527 | ||
528 | *dirty = (le32_to_cpu(di->id1.journal1.ij_flags) & | |
529 | OCFS2_JOURNAL_DIRTY_FL); | |
530 | ||
531 | journal->j_journal = j_journal; | |
532 | journal->j_inode = inode; | |
533 | journal->j_bh = bh; | |
534 | ||
535 | ocfs2_set_journal_params(osb); | |
536 | ||
537 | journal->j_state = OCFS2_JOURNAL_LOADED; | |
538 | ||
539 | status = 0; | |
540 | done: | |
541 | if (status < 0) { | |
e63aecb6 MF |
542 | if (inode_lock) |
543 | ocfs2_inode_unlock(inode, 1); | |
a81cb88b | 544 | brelse(bh); |
ccd979bd MF |
545 | if (inode) { |
546 | OCFS2_I(inode)->ip_open_count--; | |
547 | iput(inode); | |
548 | } | |
549 | } | |
550 | ||
551 | mlog_exit(status); | |
552 | return status; | |
553 | } | |
554 | ||
539d8264 SM |
555 | static void ocfs2_bump_recovery_generation(struct ocfs2_dinode *di) |
556 | { | |
557 | le32_add_cpu(&(di->id1.journal1.ij_recovery_generation), 1); | |
558 | } | |
559 | ||
560 | static u32 ocfs2_get_recovery_generation(struct ocfs2_dinode *di) | |
561 | { | |
562 | return le32_to_cpu(di->id1.journal1.ij_recovery_generation); | |
563 | } | |
564 | ||
ccd979bd | 565 | static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb, |
539d8264 | 566 | int dirty, int replayed) |
ccd979bd MF |
567 | { |
568 | int status; | |
569 | unsigned int flags; | |
570 | struct ocfs2_journal *journal = osb->journal; | |
571 | struct buffer_head *bh = journal->j_bh; | |
572 | struct ocfs2_dinode *fe; | |
573 | ||
574 | mlog_entry_void(); | |
575 | ||
576 | fe = (struct ocfs2_dinode *)bh->b_data; | |
10995aa2 JB |
577 | |
578 | /* The journal bh on the osb always comes from ocfs2_journal_init() | |
579 | * and was validated there inside ocfs2_inode_lock_full(). It's a | |
580 | * code bug if we mess it up. */ | |
581 | BUG_ON(!OCFS2_IS_VALID_DINODE(fe)); | |
ccd979bd MF |
582 | |
583 | flags = le32_to_cpu(fe->id1.journal1.ij_flags); | |
584 | if (dirty) | |
585 | flags |= OCFS2_JOURNAL_DIRTY_FL; | |
586 | else | |
587 | flags &= ~OCFS2_JOURNAL_DIRTY_FL; | |
588 | fe->id1.journal1.ij_flags = cpu_to_le32(flags); | |
589 | ||
539d8264 SM |
590 | if (replayed) |
591 | ocfs2_bump_recovery_generation(fe); | |
592 | ||
ccd979bd MF |
593 | status = ocfs2_write_block(osb, bh, journal->j_inode); |
594 | if (status < 0) | |
595 | mlog_errno(status); | |
596 | ||
ccd979bd MF |
597 | mlog_exit(status); |
598 | return status; | |
599 | } | |
600 | ||
601 | /* | |
602 | * If the journal has been kmalloc'd it needs to be freed after this | |
603 | * call. | |
604 | */ | |
605 | void ocfs2_journal_shutdown(struct ocfs2_super *osb) | |
606 | { | |
607 | struct ocfs2_journal *journal = NULL; | |
608 | int status = 0; | |
609 | struct inode *inode = NULL; | |
610 | int num_running_trans = 0; | |
611 | ||
612 | mlog_entry_void(); | |
613 | ||
ebdec83b | 614 | BUG_ON(!osb); |
ccd979bd MF |
615 | |
616 | journal = osb->journal; | |
617 | if (!journal) | |
618 | goto done; | |
619 | ||
620 | inode = journal->j_inode; | |
621 | ||
622 | if (journal->j_state != OCFS2_JOURNAL_LOADED) | |
623 | goto done; | |
624 | ||
2b4e30fb | 625 | /* need to inc inode use count - jbd2_journal_destroy will iput. */ |
ccd979bd MF |
626 | if (!igrab(inode)) |
627 | BUG(); | |
628 | ||
629 | num_running_trans = atomic_read(&(osb->journal->j_num_trans)); | |
630 | if (num_running_trans > 0) | |
631 | mlog(0, "Shutting down journal: must wait on %d " | |
632 | "running transactions!\n", | |
633 | num_running_trans); | |
634 | ||
635 | /* Do a commit_cache here. It will flush our journal, *and* | |
636 | * release any locks that are still held. | |
637 | * set the SHUTDOWN flag and release the trans lock. | |
638 | * the commit thread will take the trans lock for us below. */ | |
639 | journal->j_state = OCFS2_JOURNAL_IN_SHUTDOWN; | |
640 | ||
641 | /* The OCFS2_JOURNAL_IN_SHUTDOWN will signal to commit_cache to not | |
642 | * drop the trans_lock (which we want to hold until we | |
643 | * completely destroy the journal. */ | |
644 | if (osb->commit_task) { | |
645 | /* Wait for the commit thread */ | |
646 | mlog(0, "Waiting for ocfs2commit to exit....\n"); | |
647 | kthread_stop(osb->commit_task); | |
648 | osb->commit_task = NULL; | |
649 | } | |
650 | ||
651 | BUG_ON(atomic_read(&(osb->journal->j_num_trans)) != 0); | |
652 | ||
c271c5c2 | 653 | if (ocfs2_mount_local(osb)) { |
2b4e30fb JB |
654 | jbd2_journal_lock_updates(journal->j_journal); |
655 | status = jbd2_journal_flush(journal->j_journal); | |
656 | jbd2_journal_unlock_updates(journal->j_journal); | |
c271c5c2 SM |
657 | if (status < 0) |
658 | mlog_errno(status); | |
659 | } | |
660 | ||
661 | if (status == 0) { | |
662 | /* | |
663 | * Do not toggle if flush was unsuccessful otherwise | |
664 | * will leave dirty metadata in a "clean" journal | |
665 | */ | |
539d8264 | 666 | status = ocfs2_journal_toggle_dirty(osb, 0, 0); |
c271c5c2 SM |
667 | if (status < 0) |
668 | mlog_errno(status); | |
669 | } | |
ccd979bd MF |
670 | |
671 | /* Shutdown the kernel journal system */ | |
2b4e30fb | 672 | jbd2_journal_destroy(journal->j_journal); |
ae0dff68 | 673 | journal->j_journal = NULL; |
ccd979bd MF |
674 | |
675 | OCFS2_I(inode)->ip_open_count--; | |
676 | ||
677 | /* unlock our journal */ | |
e63aecb6 | 678 | ocfs2_inode_unlock(inode, 1); |
ccd979bd MF |
679 | |
680 | brelse(journal->j_bh); | |
681 | journal->j_bh = NULL; | |
682 | ||
683 | journal->j_state = OCFS2_JOURNAL_FREE; | |
684 | ||
685 | // up_write(&journal->j_trans_barrier); | |
686 | done: | |
687 | if (inode) | |
688 | iput(inode); | |
689 | mlog_exit_void(); | |
690 | } | |
691 | ||
692 | static void ocfs2_clear_journal_error(struct super_block *sb, | |
693 | journal_t *journal, | |
694 | int slot) | |
695 | { | |
696 | int olderr; | |
697 | ||
2b4e30fb | 698 | olderr = jbd2_journal_errno(journal); |
ccd979bd MF |
699 | if (olderr) { |
700 | mlog(ML_ERROR, "File system error %d recorded in " | |
701 | "journal %u.\n", olderr, slot); | |
702 | mlog(ML_ERROR, "File system on device %s needs checking.\n", | |
703 | sb->s_id); | |
704 | ||
2b4e30fb JB |
705 | jbd2_journal_ack_err(journal); |
706 | jbd2_journal_clear_err(journal); | |
ccd979bd MF |
707 | } |
708 | } | |
709 | ||
539d8264 | 710 | int ocfs2_journal_load(struct ocfs2_journal *journal, int local, int replayed) |
ccd979bd MF |
711 | { |
712 | int status = 0; | |
713 | struct ocfs2_super *osb; | |
714 | ||
715 | mlog_entry_void(); | |
716 | ||
b1f3550f | 717 | BUG_ON(!journal); |
ccd979bd MF |
718 | |
719 | osb = journal->j_osb; | |
720 | ||
2b4e30fb | 721 | status = jbd2_journal_load(journal->j_journal); |
ccd979bd MF |
722 | if (status < 0) { |
723 | mlog(ML_ERROR, "Failed to load journal!\n"); | |
724 | goto done; | |
725 | } | |
726 | ||
727 | ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num); | |
728 | ||
539d8264 | 729 | status = ocfs2_journal_toggle_dirty(osb, 1, replayed); |
ccd979bd MF |
730 | if (status < 0) { |
731 | mlog_errno(status); | |
732 | goto done; | |
733 | } | |
734 | ||
735 | /* Launch the commit thread */ | |
c271c5c2 SM |
736 | if (!local) { |
737 | osb->commit_task = kthread_run(ocfs2_commit_thread, osb, | |
738 | "ocfs2cmt"); | |
739 | if (IS_ERR(osb->commit_task)) { | |
740 | status = PTR_ERR(osb->commit_task); | |
741 | osb->commit_task = NULL; | |
742 | mlog(ML_ERROR, "unable to launch ocfs2commit thread, " | |
743 | "error=%d", status); | |
744 | goto done; | |
745 | } | |
746 | } else | |
ccd979bd | 747 | osb->commit_task = NULL; |
ccd979bd MF |
748 | |
749 | done: | |
750 | mlog_exit(status); | |
751 | return status; | |
752 | } | |
753 | ||
754 | ||
755 | /* 'full' flag tells us whether we clear out all blocks or if we just | |
756 | * mark the journal clean */ | |
757 | int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full) | |
758 | { | |
759 | int status; | |
760 | ||
761 | mlog_entry_void(); | |
762 | ||
ebdec83b | 763 | BUG_ON(!journal); |
ccd979bd | 764 | |
2b4e30fb | 765 | status = jbd2_journal_wipe(journal->j_journal, full); |
ccd979bd MF |
766 | if (status < 0) { |
767 | mlog_errno(status); | |
768 | goto bail; | |
769 | } | |
770 | ||
539d8264 | 771 | status = ocfs2_journal_toggle_dirty(journal->j_osb, 0, 0); |
ccd979bd MF |
772 | if (status < 0) |
773 | mlog_errno(status); | |
774 | ||
775 | bail: | |
776 | mlog_exit(status); | |
777 | return status; | |
778 | } | |
779 | ||
553abd04 JB |
780 | static int ocfs2_recovery_completed(struct ocfs2_super *osb) |
781 | { | |
782 | int empty; | |
783 | struct ocfs2_recovery_map *rm = osb->recovery_map; | |
784 | ||
785 | spin_lock(&osb->osb_lock); | |
786 | empty = (rm->rm_used == 0); | |
787 | spin_unlock(&osb->osb_lock); | |
788 | ||
789 | return empty; | |
790 | } | |
791 | ||
792 | void ocfs2_wait_for_recovery(struct ocfs2_super *osb) | |
793 | { | |
794 | wait_event(osb->recovery_event, ocfs2_recovery_completed(osb)); | |
795 | } | |
796 | ||
ccd979bd MF |
797 | /* |
798 | * JBD Might read a cached version of another nodes journal file. We | |
799 | * don't want this as this file changes often and we get no | |
800 | * notification on those changes. The only way to be sure that we've | |
801 | * got the most up to date version of those blocks then is to force | |
802 | * read them off disk. Just searching through the buffer cache won't | |
803 | * work as there may be pages backing this file which are still marked | |
804 | * up to date. We know things can't change on this file underneath us | |
805 | * as we have the lock by now :) | |
806 | */ | |
807 | static int ocfs2_force_read_journal(struct inode *inode) | |
808 | { | |
809 | int status = 0; | |
4f902c37 | 810 | int i; |
8110b073 | 811 | u64 v_blkno, p_blkno, p_blocks, num_blocks; |
4f902c37 | 812 | #define CONCURRENT_JOURNAL_FILL 32ULL |
ccd979bd MF |
813 | struct buffer_head *bhs[CONCURRENT_JOURNAL_FILL]; |
814 | ||
815 | mlog_entry_void(); | |
816 | ||
ccd979bd MF |
817 | memset(bhs, 0, sizeof(struct buffer_head *) * CONCURRENT_JOURNAL_FILL); |
818 | ||
8110b073 | 819 | num_blocks = ocfs2_blocks_for_bytes(inode->i_sb, inode->i_size); |
ccd979bd | 820 | v_blkno = 0; |
8110b073 | 821 | while (v_blkno < num_blocks) { |
ccd979bd | 822 | status = ocfs2_extent_map_get_blocks(inode, v_blkno, |
49cb8d2d | 823 | &p_blkno, &p_blocks, NULL); |
ccd979bd MF |
824 | if (status < 0) { |
825 | mlog_errno(status); | |
826 | goto bail; | |
827 | } | |
828 | ||
829 | if (p_blocks > CONCURRENT_JOURNAL_FILL) | |
830 | p_blocks = CONCURRENT_JOURNAL_FILL; | |
831 | ||
dd4a2c2b MF |
832 | /* We are reading journal data which should not |
833 | * be put in the uptodate cache */ | |
da1e9098 JB |
834 | status = ocfs2_read_blocks_sync(OCFS2_SB(inode->i_sb), |
835 | p_blkno, p_blocks, bhs); | |
ccd979bd MF |
836 | if (status < 0) { |
837 | mlog_errno(status); | |
838 | goto bail; | |
839 | } | |
840 | ||
841 | for(i = 0; i < p_blocks; i++) { | |
842 | brelse(bhs[i]); | |
843 | bhs[i] = NULL; | |
844 | } | |
845 | ||
846 | v_blkno += p_blocks; | |
847 | } | |
848 | ||
849 | bail: | |
850 | for(i = 0; i < CONCURRENT_JOURNAL_FILL; i++) | |
a81cb88b | 851 | brelse(bhs[i]); |
ccd979bd MF |
852 | mlog_exit(status); |
853 | return status; | |
854 | } | |
855 | ||
856 | struct ocfs2_la_recovery_item { | |
857 | struct list_head lri_list; | |
858 | int lri_slot; | |
859 | struct ocfs2_dinode *lri_la_dinode; | |
860 | struct ocfs2_dinode *lri_tl_dinode; | |
2205363d | 861 | struct ocfs2_quota_recovery *lri_qrec; |
ccd979bd MF |
862 | }; |
863 | ||
864 | /* Does the second half of the recovery process. By this point, the | |
865 | * node is marked clean and can actually be considered recovered, | |
866 | * hence it's no longer in the recovery map, but there's still some | |
867 | * cleanup we can do which shouldn't happen within the recovery thread | |
868 | * as locking in that context becomes very difficult if we are to take | |
869 | * recovering nodes into account. | |
870 | * | |
871 | * NOTE: This function can and will sleep on recovery of other nodes | |
872 | * during cluster locking, just like any other ocfs2 process. | |
873 | */ | |
c4028958 | 874 | void ocfs2_complete_recovery(struct work_struct *work) |
ccd979bd MF |
875 | { |
876 | int ret; | |
c4028958 DH |
877 | struct ocfs2_journal *journal = |
878 | container_of(work, struct ocfs2_journal, j_recovery_work); | |
879 | struct ocfs2_super *osb = journal->j_osb; | |
ccd979bd | 880 | struct ocfs2_dinode *la_dinode, *tl_dinode; |
800deef3 | 881 | struct ocfs2_la_recovery_item *item, *n; |
2205363d | 882 | struct ocfs2_quota_recovery *qrec; |
ccd979bd MF |
883 | LIST_HEAD(tmp_la_list); |
884 | ||
885 | mlog_entry_void(); | |
886 | ||
887 | mlog(0, "completing recovery from keventd\n"); | |
888 | ||
889 | spin_lock(&journal->j_lock); | |
890 | list_splice_init(&journal->j_la_cleanups, &tmp_la_list); | |
891 | spin_unlock(&journal->j_lock); | |
892 | ||
800deef3 | 893 | list_for_each_entry_safe(item, n, &tmp_la_list, lri_list) { |
ccd979bd MF |
894 | list_del_init(&item->lri_list); |
895 | ||
896 | mlog(0, "Complete recovery for slot %d\n", item->lri_slot); | |
897 | ||
898 | la_dinode = item->lri_la_dinode; | |
899 | if (la_dinode) { | |
b0697053 | 900 | mlog(0, "Clean up local alloc %llu\n", |
1ca1a111 | 901 | (unsigned long long)le64_to_cpu(la_dinode->i_blkno)); |
ccd979bd MF |
902 | |
903 | ret = ocfs2_complete_local_alloc_recovery(osb, | |
904 | la_dinode); | |
905 | if (ret < 0) | |
906 | mlog_errno(ret); | |
907 | ||
908 | kfree(la_dinode); | |
909 | } | |
910 | ||
911 | tl_dinode = item->lri_tl_dinode; | |
912 | if (tl_dinode) { | |
b0697053 | 913 | mlog(0, "Clean up truncate log %llu\n", |
1ca1a111 | 914 | (unsigned long long)le64_to_cpu(tl_dinode->i_blkno)); |
ccd979bd MF |
915 | |
916 | ret = ocfs2_complete_truncate_log_recovery(osb, | |
917 | tl_dinode); | |
918 | if (ret < 0) | |
919 | mlog_errno(ret); | |
920 | ||
921 | kfree(tl_dinode); | |
922 | } | |
923 | ||
924 | ret = ocfs2_recover_orphans(osb, item->lri_slot); | |
925 | if (ret < 0) | |
926 | mlog_errno(ret); | |
927 | ||
2205363d JK |
928 | qrec = item->lri_qrec; |
929 | if (qrec) { | |
930 | mlog(0, "Recovering quota files"); | |
931 | ret = ocfs2_finish_quota_recovery(osb, qrec, | |
932 | item->lri_slot); | |
933 | if (ret < 0) | |
934 | mlog_errno(ret); | |
935 | /* Recovery info is already freed now */ | |
936 | } | |
937 | ||
ccd979bd MF |
938 | kfree(item); |
939 | } | |
940 | ||
941 | mlog(0, "Recovery completion\n"); | |
942 | mlog_exit_void(); | |
943 | } | |
944 | ||
945 | /* NOTE: This function always eats your references to la_dinode and | |
946 | * tl_dinode, either manually on error, or by passing them to | |
947 | * ocfs2_complete_recovery */ | |
948 | static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal, | |
949 | int slot_num, | |
950 | struct ocfs2_dinode *la_dinode, | |
2205363d JK |
951 | struct ocfs2_dinode *tl_dinode, |
952 | struct ocfs2_quota_recovery *qrec) | |
ccd979bd MF |
953 | { |
954 | struct ocfs2_la_recovery_item *item; | |
955 | ||
afae00ab | 956 | item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS); |
ccd979bd MF |
957 | if (!item) { |
958 | /* Though we wish to avoid it, we are in fact safe in | |
959 | * skipping local alloc cleanup as fsck.ocfs2 is more | |
960 | * than capable of reclaiming unused space. */ | |
961 | if (la_dinode) | |
962 | kfree(la_dinode); | |
963 | ||
964 | if (tl_dinode) | |
965 | kfree(tl_dinode); | |
966 | ||
2205363d JK |
967 | if (qrec) |
968 | ocfs2_free_quota_recovery(qrec); | |
969 | ||
ccd979bd MF |
970 | mlog_errno(-ENOMEM); |
971 | return; | |
972 | } | |
973 | ||
974 | INIT_LIST_HEAD(&item->lri_list); | |
975 | item->lri_la_dinode = la_dinode; | |
976 | item->lri_slot = slot_num; | |
977 | item->lri_tl_dinode = tl_dinode; | |
2205363d | 978 | item->lri_qrec = qrec; |
ccd979bd MF |
979 | |
980 | spin_lock(&journal->j_lock); | |
981 | list_add_tail(&item->lri_list, &journal->j_la_cleanups); | |
982 | queue_work(ocfs2_wq, &journal->j_recovery_work); | |
983 | spin_unlock(&journal->j_lock); | |
984 | } | |
985 | ||
986 | /* Called by the mount code to queue recovery the last part of | |
987 | * recovery for it's own slot. */ | |
988 | void ocfs2_complete_mount_recovery(struct ocfs2_super *osb) | |
989 | { | |
990 | struct ocfs2_journal *journal = osb->journal; | |
991 | ||
992 | if (osb->dirty) { | |
993 | /* No need to queue up our truncate_log as regular | |
994 | * cleanup will catch that. */ | |
995 | ocfs2_queue_recovery_completion(journal, | |
996 | osb->slot_num, | |
997 | osb->local_alloc_copy, | |
2205363d | 998 | NULL, |
ccd979bd MF |
999 | NULL); |
1000 | ocfs2_schedule_truncate_log_flush(osb, 0); | |
1001 | ||
1002 | osb->local_alloc_copy = NULL; | |
1003 | osb->dirty = 0; | |
1004 | } | |
1005 | } | |
1006 | ||
2205363d JK |
1007 | void ocfs2_complete_quota_recovery(struct ocfs2_super *osb) |
1008 | { | |
1009 | if (osb->quota_rec) { | |
1010 | ocfs2_queue_recovery_completion(osb->journal, | |
1011 | osb->slot_num, | |
1012 | NULL, | |
1013 | NULL, | |
1014 | osb->quota_rec); | |
1015 | osb->quota_rec = NULL; | |
1016 | } | |
1017 | } | |
1018 | ||
ccd979bd MF |
1019 | static int __ocfs2_recovery_thread(void *arg) |
1020 | { | |
2205363d | 1021 | int status, node_num, slot_num; |
ccd979bd | 1022 | struct ocfs2_super *osb = arg; |
553abd04 | 1023 | struct ocfs2_recovery_map *rm = osb->recovery_map; |
2205363d JK |
1024 | int *rm_quota = NULL; |
1025 | int rm_quota_used = 0, i; | |
1026 | struct ocfs2_quota_recovery *qrec; | |
ccd979bd MF |
1027 | |
1028 | mlog_entry_void(); | |
1029 | ||
1030 | status = ocfs2_wait_on_mount(osb); | |
1031 | if (status < 0) { | |
1032 | goto bail; | |
1033 | } | |
1034 | ||
2205363d JK |
1035 | rm_quota = kzalloc(osb->max_slots * sizeof(int), GFP_NOFS); |
1036 | if (!rm_quota) { | |
1037 | status = -ENOMEM; | |
1038 | goto bail; | |
1039 | } | |
ccd979bd MF |
1040 | restart: |
1041 | status = ocfs2_super_lock(osb, 1); | |
1042 | if (status < 0) { | |
1043 | mlog_errno(status); | |
1044 | goto bail; | |
1045 | } | |
1046 | ||
553abd04 JB |
1047 | spin_lock(&osb->osb_lock); |
1048 | while (rm->rm_used) { | |
1049 | /* It's always safe to remove entry zero, as we won't | |
1050 | * clear it until ocfs2_recover_node() has succeeded. */ | |
1051 | node_num = rm->rm_entries[0]; | |
1052 | spin_unlock(&osb->osb_lock); | |
2205363d JK |
1053 | mlog(0, "checking node %d\n", node_num); |
1054 | slot_num = ocfs2_node_num_to_slot(osb, node_num); | |
1055 | if (slot_num == -ENOENT) { | |
1056 | status = 0; | |
1057 | mlog(0, "no slot for this node, so no recovery" | |
1058 | "required.\n"); | |
1059 | goto skip_recovery; | |
1060 | } | |
1061 | mlog(0, "node %d was using slot %d\n", node_num, slot_num); | |
1062 | ||
1063 | /* It is a bit subtle with quota recovery. We cannot do it | |
1064 | * immediately because we have to obtain cluster locks from | |
1065 | * quota files and we also don't want to just skip it because | |
1066 | * then quota usage would be out of sync until some node takes | |
1067 | * the slot. So we remember which nodes need quota recovery | |
1068 | * and when everything else is done, we recover quotas. */ | |
1069 | for (i = 0; i < rm_quota_used && rm_quota[i] != slot_num; i++); | |
1070 | if (i == rm_quota_used) | |
1071 | rm_quota[rm_quota_used++] = slot_num; | |
1072 | ||
1073 | status = ocfs2_recover_node(osb, node_num, slot_num); | |
1074 | skip_recovery: | |
553abd04 JB |
1075 | if (!status) { |
1076 | ocfs2_recovery_map_clear(osb, node_num); | |
1077 | } else { | |
ccd979bd MF |
1078 | mlog(ML_ERROR, |
1079 | "Error %d recovering node %d on device (%u,%u)!\n", | |
1080 | status, node_num, | |
1081 | MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev)); | |
1082 | mlog(ML_ERROR, "Volume requires unmount.\n"); | |
ccd979bd MF |
1083 | } |
1084 | ||
553abd04 | 1085 | spin_lock(&osb->osb_lock); |
ccd979bd | 1086 | } |
553abd04 JB |
1087 | spin_unlock(&osb->osb_lock); |
1088 | mlog(0, "All nodes recovered\n"); | |
1089 | ||
539d8264 SM |
1090 | /* Refresh all journal recovery generations from disk */ |
1091 | status = ocfs2_check_journals_nolocks(osb); | |
1092 | status = (status == -EROFS) ? 0 : status; | |
1093 | if (status < 0) | |
1094 | mlog_errno(status); | |
1095 | ||
2205363d JK |
1096 | /* Now it is right time to recover quotas... We have to do this under |
1097 | * superblock lock so that noone can start using the slot (and crash) | |
1098 | * before we recover it */ | |
1099 | for (i = 0; i < rm_quota_used; i++) { | |
1100 | qrec = ocfs2_begin_quota_recovery(osb, rm_quota[i]); | |
1101 | if (IS_ERR(qrec)) { | |
1102 | status = PTR_ERR(qrec); | |
1103 | mlog_errno(status); | |
1104 | continue; | |
1105 | } | |
1106 | ocfs2_queue_recovery_completion(osb->journal, rm_quota[i], | |
1107 | NULL, NULL, qrec); | |
1108 | } | |
1109 | ||
ccd979bd MF |
1110 | ocfs2_super_unlock(osb, 1); |
1111 | ||
1112 | /* We always run recovery on our own orphan dir - the dead | |
34d024f8 MF |
1113 | * node(s) may have disallowd a previos inode delete. Re-processing |
1114 | * is therefore required. */ | |
ccd979bd | 1115 | ocfs2_queue_recovery_completion(osb->journal, osb->slot_num, NULL, |
2205363d | 1116 | NULL, NULL); |
ccd979bd MF |
1117 | |
1118 | bail: | |
c74ec2f7 | 1119 | mutex_lock(&osb->recovery_lock); |
553abd04 | 1120 | if (!status && !ocfs2_recovery_completed(osb)) { |
c74ec2f7 | 1121 | mutex_unlock(&osb->recovery_lock); |
ccd979bd MF |
1122 | goto restart; |
1123 | } | |
1124 | ||
1125 | osb->recovery_thread_task = NULL; | |
1126 | mb(); /* sync with ocfs2_recovery_thread_running */ | |
1127 | wake_up(&osb->recovery_event); | |
1128 | ||
c74ec2f7 | 1129 | mutex_unlock(&osb->recovery_lock); |
ccd979bd | 1130 | |
2205363d JK |
1131 | if (rm_quota) |
1132 | kfree(rm_quota); | |
1133 | ||
ccd979bd MF |
1134 | mlog_exit(status); |
1135 | /* no one is callint kthread_stop() for us so the kthread() api | |
1136 | * requires that we call do_exit(). And it isn't exported, but | |
1137 | * complete_and_exit() seems to be a minimal wrapper around it. */ | |
1138 | complete_and_exit(NULL, status); | |
1139 | return status; | |
1140 | } | |
1141 | ||
1142 | void ocfs2_recovery_thread(struct ocfs2_super *osb, int node_num) | |
1143 | { | |
1144 | mlog_entry("(node_num=%d, osb->node_num = %d)\n", | |
1145 | node_num, osb->node_num); | |
1146 | ||
c74ec2f7 | 1147 | mutex_lock(&osb->recovery_lock); |
ccd979bd MF |
1148 | if (osb->disable_recovery) |
1149 | goto out; | |
1150 | ||
1151 | /* People waiting on recovery will wait on | |
1152 | * the recovery map to empty. */ | |
553abd04 JB |
1153 | if (ocfs2_recovery_map_set(osb, node_num)) |
1154 | mlog(0, "node %d already in recovery map.\n", node_num); | |
ccd979bd MF |
1155 | |
1156 | mlog(0, "starting recovery thread...\n"); | |
1157 | ||
1158 | if (osb->recovery_thread_task) | |
1159 | goto out; | |
1160 | ||
1161 | osb->recovery_thread_task = kthread_run(__ocfs2_recovery_thread, osb, | |
78427043 | 1162 | "ocfs2rec"); |
ccd979bd MF |
1163 | if (IS_ERR(osb->recovery_thread_task)) { |
1164 | mlog_errno((int)PTR_ERR(osb->recovery_thread_task)); | |
1165 | osb->recovery_thread_task = NULL; | |
1166 | } | |
1167 | ||
1168 | out: | |
c74ec2f7 | 1169 | mutex_unlock(&osb->recovery_lock); |
ccd979bd MF |
1170 | wake_up(&osb->recovery_event); |
1171 | ||
1172 | mlog_exit_void(); | |
1173 | } | |
1174 | ||
539d8264 SM |
1175 | static int ocfs2_read_journal_inode(struct ocfs2_super *osb, |
1176 | int slot_num, | |
1177 | struct buffer_head **bh, | |
1178 | struct inode **ret_inode) | |
1179 | { | |
1180 | int status = -EACCES; | |
1181 | struct inode *inode = NULL; | |
1182 | ||
1183 | BUG_ON(slot_num >= osb->max_slots); | |
1184 | ||
1185 | inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, | |
1186 | slot_num); | |
1187 | if (!inode || is_bad_inode(inode)) { | |
1188 | mlog_errno(status); | |
1189 | goto bail; | |
1190 | } | |
1191 | SET_INODE_JOURNAL(inode); | |
1192 | ||
b657c95c | 1193 | status = ocfs2_read_inode_block_full(inode, bh, OCFS2_BH_IGNORE_CACHE); |
539d8264 SM |
1194 | if (status < 0) { |
1195 | mlog_errno(status); | |
1196 | goto bail; | |
1197 | } | |
1198 | ||
1199 | status = 0; | |
1200 | ||
1201 | bail: | |
1202 | if (inode) { | |
1203 | if (status || !ret_inode) | |
1204 | iput(inode); | |
1205 | else | |
1206 | *ret_inode = inode; | |
1207 | } | |
1208 | return status; | |
1209 | } | |
1210 | ||
ccd979bd MF |
1211 | /* Does the actual journal replay and marks the journal inode as |
1212 | * clean. Will only replay if the journal inode is marked dirty. */ | |
1213 | static int ocfs2_replay_journal(struct ocfs2_super *osb, | |
1214 | int node_num, | |
1215 | int slot_num) | |
1216 | { | |
1217 | int status; | |
1218 | int got_lock = 0; | |
1219 | unsigned int flags; | |
1220 | struct inode *inode = NULL; | |
1221 | struct ocfs2_dinode *fe; | |
1222 | journal_t *journal = NULL; | |
1223 | struct buffer_head *bh = NULL; | |
539d8264 | 1224 | u32 slot_reco_gen; |
ccd979bd | 1225 | |
539d8264 SM |
1226 | status = ocfs2_read_journal_inode(osb, slot_num, &bh, &inode); |
1227 | if (status) { | |
ccd979bd MF |
1228 | mlog_errno(status); |
1229 | goto done; | |
1230 | } | |
539d8264 SM |
1231 | |
1232 | fe = (struct ocfs2_dinode *)bh->b_data; | |
1233 | slot_reco_gen = ocfs2_get_recovery_generation(fe); | |
1234 | brelse(bh); | |
1235 | bh = NULL; | |
1236 | ||
1237 | /* | |
1238 | * As the fs recovery is asynchronous, there is a small chance that | |
1239 | * another node mounted (and recovered) the slot before the recovery | |
1240 | * thread could get the lock. To handle that, we dirty read the journal | |
1241 | * inode for that slot to get the recovery generation. If it is | |
1242 | * different than what we expected, the slot has been recovered. | |
1243 | * If not, it needs recovery. | |
1244 | */ | |
1245 | if (osb->slot_recovery_generations[slot_num] != slot_reco_gen) { | |
1246 | mlog(0, "Slot %u already recovered (old/new=%u/%u)\n", slot_num, | |
1247 | osb->slot_recovery_generations[slot_num], slot_reco_gen); | |
1248 | osb->slot_recovery_generations[slot_num] = slot_reco_gen; | |
1249 | status = -EBUSY; | |
ccd979bd MF |
1250 | goto done; |
1251 | } | |
539d8264 SM |
1252 | |
1253 | /* Continue with recovery as the journal has not yet been recovered */ | |
ccd979bd | 1254 | |
e63aecb6 | 1255 | status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY); |
ccd979bd | 1256 | if (status < 0) { |
e63aecb6 | 1257 | mlog(0, "status returned from ocfs2_inode_lock=%d\n", status); |
ccd979bd MF |
1258 | if (status != -ERESTARTSYS) |
1259 | mlog(ML_ERROR, "Could not lock journal!\n"); | |
1260 | goto done; | |
1261 | } | |
1262 | got_lock = 1; | |
1263 | ||
1264 | fe = (struct ocfs2_dinode *) bh->b_data; | |
1265 | ||
1266 | flags = le32_to_cpu(fe->id1.journal1.ij_flags); | |
539d8264 | 1267 | slot_reco_gen = ocfs2_get_recovery_generation(fe); |
ccd979bd MF |
1268 | |
1269 | if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) { | |
1270 | mlog(0, "No recovery required for node %d\n", node_num); | |
539d8264 SM |
1271 | /* Refresh recovery generation for the slot */ |
1272 | osb->slot_recovery_generations[slot_num] = slot_reco_gen; | |
ccd979bd MF |
1273 | goto done; |
1274 | } | |
1275 | ||
1276 | mlog(ML_NOTICE, "Recovering node %d from slot %d on device (%u,%u)\n", | |
1277 | node_num, slot_num, | |
1278 | MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev)); | |
1279 | ||
1280 | OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters); | |
1281 | ||
1282 | status = ocfs2_force_read_journal(inode); | |
1283 | if (status < 0) { | |
1284 | mlog_errno(status); | |
1285 | goto done; | |
1286 | } | |
1287 | ||
1288 | mlog(0, "calling journal_init_inode\n"); | |
2b4e30fb | 1289 | journal = jbd2_journal_init_inode(inode); |
ccd979bd MF |
1290 | if (journal == NULL) { |
1291 | mlog(ML_ERROR, "Linux journal layer error\n"); | |
1292 | status = -EIO; | |
1293 | goto done; | |
1294 | } | |
1295 | ||
2b4e30fb | 1296 | status = jbd2_journal_load(journal); |
ccd979bd MF |
1297 | if (status < 0) { |
1298 | mlog_errno(status); | |
1299 | if (!igrab(inode)) | |
1300 | BUG(); | |
2b4e30fb | 1301 | jbd2_journal_destroy(journal); |
ccd979bd MF |
1302 | goto done; |
1303 | } | |
1304 | ||
1305 | ocfs2_clear_journal_error(osb->sb, journal, slot_num); | |
1306 | ||
1307 | /* wipe the journal */ | |
1308 | mlog(0, "flushing the journal.\n"); | |
2b4e30fb JB |
1309 | jbd2_journal_lock_updates(journal); |
1310 | status = jbd2_journal_flush(journal); | |
1311 | jbd2_journal_unlock_updates(journal); | |
ccd979bd MF |
1312 | if (status < 0) |
1313 | mlog_errno(status); | |
1314 | ||
1315 | /* This will mark the node clean */ | |
1316 | flags = le32_to_cpu(fe->id1.journal1.ij_flags); | |
1317 | flags &= ~OCFS2_JOURNAL_DIRTY_FL; | |
1318 | fe->id1.journal1.ij_flags = cpu_to_le32(flags); | |
1319 | ||
539d8264 SM |
1320 | /* Increment recovery generation to indicate successful recovery */ |
1321 | ocfs2_bump_recovery_generation(fe); | |
1322 | osb->slot_recovery_generations[slot_num] = | |
1323 | ocfs2_get_recovery_generation(fe); | |
1324 | ||
ccd979bd MF |
1325 | status = ocfs2_write_block(osb, bh, inode); |
1326 | if (status < 0) | |
1327 | mlog_errno(status); | |
1328 | ||
1329 | if (!igrab(inode)) | |
1330 | BUG(); | |
1331 | ||
2b4e30fb | 1332 | jbd2_journal_destroy(journal); |
ccd979bd MF |
1333 | |
1334 | done: | |
1335 | /* drop the lock on this nodes journal */ | |
1336 | if (got_lock) | |
e63aecb6 | 1337 | ocfs2_inode_unlock(inode, 1); |
ccd979bd MF |
1338 | |
1339 | if (inode) | |
1340 | iput(inode); | |
1341 | ||
a81cb88b | 1342 | brelse(bh); |
ccd979bd MF |
1343 | |
1344 | mlog_exit(status); | |
1345 | return status; | |
1346 | } | |
1347 | ||
1348 | /* | |
1349 | * Do the most important parts of node recovery: | |
1350 | * - Replay it's journal | |
1351 | * - Stamp a clean local allocator file | |
1352 | * - Stamp a clean truncate log | |
1353 | * - Mark the node clean | |
1354 | * | |
1355 | * If this function completes without error, a node in OCFS2 can be | |
1356 | * said to have been safely recovered. As a result, failure during the | |
1357 | * second part of a nodes recovery process (local alloc recovery) is | |
1358 | * far less concerning. | |
1359 | */ | |
1360 | static int ocfs2_recover_node(struct ocfs2_super *osb, | |
2205363d | 1361 | int node_num, int slot_num) |
ccd979bd MF |
1362 | { |
1363 | int status = 0; | |
ccd979bd MF |
1364 | struct ocfs2_dinode *la_copy = NULL; |
1365 | struct ocfs2_dinode *tl_copy = NULL; | |
1366 | ||
2205363d JK |
1367 | mlog_entry("(node_num=%d, slot_num=%d, osb->node_num = %d)\n", |
1368 | node_num, slot_num, osb->node_num); | |
ccd979bd MF |
1369 | |
1370 | /* Should not ever be called to recover ourselves -- in that | |
1371 | * case we should've called ocfs2_journal_load instead. */ | |
ebdec83b | 1372 | BUG_ON(osb->node_num == node_num); |
ccd979bd | 1373 | |
ccd979bd MF |
1374 | status = ocfs2_replay_journal(osb, node_num, slot_num); |
1375 | if (status < 0) { | |
539d8264 SM |
1376 | if (status == -EBUSY) { |
1377 | mlog(0, "Skipping recovery for slot %u (node %u) " | |
1378 | "as another node has recovered it\n", slot_num, | |
1379 | node_num); | |
1380 | status = 0; | |
1381 | goto done; | |
1382 | } | |
ccd979bd MF |
1383 | mlog_errno(status); |
1384 | goto done; | |
1385 | } | |
1386 | ||
1387 | /* Stamp a clean local alloc file AFTER recovering the journal... */ | |
1388 | status = ocfs2_begin_local_alloc_recovery(osb, slot_num, &la_copy); | |
1389 | if (status < 0) { | |
1390 | mlog_errno(status); | |
1391 | goto done; | |
1392 | } | |
1393 | ||
1394 | /* An error from begin_truncate_log_recovery is not | |
1395 | * serious enough to warrant halting the rest of | |
1396 | * recovery. */ | |
1397 | status = ocfs2_begin_truncate_log_recovery(osb, slot_num, &tl_copy); | |
1398 | if (status < 0) | |
1399 | mlog_errno(status); | |
1400 | ||
1401 | /* Likewise, this would be a strange but ultimately not so | |
1402 | * harmful place to get an error... */ | |
8e8a4603 | 1403 | status = ocfs2_clear_slot(osb, slot_num); |
ccd979bd MF |
1404 | if (status < 0) |
1405 | mlog_errno(status); | |
1406 | ||
1407 | /* This will kfree the memory pointed to by la_copy and tl_copy */ | |
1408 | ocfs2_queue_recovery_completion(osb->journal, slot_num, la_copy, | |
2205363d | 1409 | tl_copy, NULL); |
ccd979bd MF |
1410 | |
1411 | status = 0; | |
1412 | done: | |
1413 | ||
1414 | mlog_exit(status); | |
1415 | return status; | |
1416 | } | |
1417 | ||
1418 | /* Test node liveness by trylocking his journal. If we get the lock, | |
1419 | * we drop it here. Return 0 if we got the lock, -EAGAIN if node is | |
1420 | * still alive (we couldn't get the lock) and < 0 on error. */ | |
1421 | static int ocfs2_trylock_journal(struct ocfs2_super *osb, | |
1422 | int slot_num) | |
1423 | { | |
1424 | int status, flags; | |
1425 | struct inode *inode = NULL; | |
1426 | ||
1427 | inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, | |
1428 | slot_num); | |
1429 | if (inode == NULL) { | |
1430 | mlog(ML_ERROR, "access error\n"); | |
1431 | status = -EACCES; | |
1432 | goto bail; | |
1433 | } | |
1434 | if (is_bad_inode(inode)) { | |
1435 | mlog(ML_ERROR, "access error (bad inode)\n"); | |
1436 | iput(inode); | |
1437 | inode = NULL; | |
1438 | status = -EACCES; | |
1439 | goto bail; | |
1440 | } | |
1441 | SET_INODE_JOURNAL(inode); | |
1442 | ||
1443 | flags = OCFS2_META_LOCK_RECOVERY | OCFS2_META_LOCK_NOQUEUE; | |
e63aecb6 | 1444 | status = ocfs2_inode_lock_full(inode, NULL, 1, flags); |
ccd979bd MF |
1445 | if (status < 0) { |
1446 | if (status != -EAGAIN) | |
1447 | mlog_errno(status); | |
1448 | goto bail; | |
1449 | } | |
1450 | ||
e63aecb6 | 1451 | ocfs2_inode_unlock(inode, 1); |
ccd979bd MF |
1452 | bail: |
1453 | if (inode) | |
1454 | iput(inode); | |
1455 | ||
1456 | return status; | |
1457 | } | |
1458 | ||
1459 | /* Call this underneath ocfs2_super_lock. It also assumes that the | |
1460 | * slot info struct has been updated from disk. */ | |
1461 | int ocfs2_mark_dead_nodes(struct ocfs2_super *osb) | |
1462 | { | |
d85b20e4 JB |
1463 | unsigned int node_num; |
1464 | int status, i; | |
a1af7d15 | 1465 | u32 gen; |
539d8264 SM |
1466 | struct buffer_head *bh = NULL; |
1467 | struct ocfs2_dinode *di; | |
ccd979bd MF |
1468 | |
1469 | /* This is called with the super block cluster lock, so we | |
1470 | * know that the slot map can't change underneath us. */ | |
1471 | ||
d85b20e4 | 1472 | for (i = 0; i < osb->max_slots; i++) { |
539d8264 SM |
1473 | /* Read journal inode to get the recovery generation */ |
1474 | status = ocfs2_read_journal_inode(osb, i, &bh, NULL); | |
1475 | if (status) { | |
1476 | mlog_errno(status); | |
1477 | goto bail; | |
1478 | } | |
1479 | di = (struct ocfs2_dinode *)bh->b_data; | |
a1af7d15 | 1480 | gen = ocfs2_get_recovery_generation(di); |
539d8264 SM |
1481 | brelse(bh); |
1482 | bh = NULL; | |
1483 | ||
a1af7d15 MF |
1484 | spin_lock(&osb->osb_lock); |
1485 | osb->slot_recovery_generations[i] = gen; | |
1486 | ||
539d8264 SM |
1487 | mlog(0, "Slot %u recovery generation is %u\n", i, |
1488 | osb->slot_recovery_generations[i]); | |
1489 | ||
a1af7d15 MF |
1490 | if (i == osb->slot_num) { |
1491 | spin_unlock(&osb->osb_lock); | |
ccd979bd | 1492 | continue; |
a1af7d15 | 1493 | } |
d85b20e4 JB |
1494 | |
1495 | status = ocfs2_slot_to_node_num_locked(osb, i, &node_num); | |
a1af7d15 MF |
1496 | if (status == -ENOENT) { |
1497 | spin_unlock(&osb->osb_lock); | |
ccd979bd | 1498 | continue; |
a1af7d15 | 1499 | } |
ccd979bd | 1500 | |
a1af7d15 MF |
1501 | if (__ocfs2_recovery_map_test(osb, node_num)) { |
1502 | spin_unlock(&osb->osb_lock); | |
ccd979bd | 1503 | continue; |
a1af7d15 | 1504 | } |
d85b20e4 | 1505 | spin_unlock(&osb->osb_lock); |
ccd979bd MF |
1506 | |
1507 | /* Ok, we have a slot occupied by another node which | |
1508 | * is not in the recovery map. We trylock his journal | |
1509 | * file here to test if he's alive. */ | |
1510 | status = ocfs2_trylock_journal(osb, i); | |
1511 | if (!status) { | |
1512 | /* Since we're called from mount, we know that | |
1513 | * the recovery thread can't race us on | |
1514 | * setting / checking the recovery bits. */ | |
1515 | ocfs2_recovery_thread(osb, node_num); | |
1516 | } else if ((status < 0) && (status != -EAGAIN)) { | |
1517 | mlog_errno(status); | |
1518 | goto bail; | |
1519 | } | |
ccd979bd | 1520 | } |
ccd979bd MF |
1521 | |
1522 | status = 0; | |
1523 | bail: | |
1524 | mlog_exit(status); | |
1525 | return status; | |
1526 | } | |
1527 | ||
5eae5b96 MF |
1528 | struct ocfs2_orphan_filldir_priv { |
1529 | struct inode *head; | |
1530 | struct ocfs2_super *osb; | |
1531 | }; | |
1532 | ||
1533 | static int ocfs2_orphan_filldir(void *priv, const char *name, int name_len, | |
1534 | loff_t pos, u64 ino, unsigned type) | |
1535 | { | |
1536 | struct ocfs2_orphan_filldir_priv *p = priv; | |
1537 | struct inode *iter; | |
1538 | ||
1539 | if (name_len == 1 && !strncmp(".", name, 1)) | |
1540 | return 0; | |
1541 | if (name_len == 2 && !strncmp("..", name, 2)) | |
1542 | return 0; | |
1543 | ||
1544 | /* Skip bad inodes so that recovery can continue */ | |
1545 | iter = ocfs2_iget(p->osb, ino, | |
5fa0613e | 1546 | OCFS2_FI_FLAG_ORPHAN_RECOVERY, 0); |
5eae5b96 MF |
1547 | if (IS_ERR(iter)) |
1548 | return 0; | |
1549 | ||
1550 | mlog(0, "queue orphan %llu\n", | |
1551 | (unsigned long long)OCFS2_I(iter)->ip_blkno); | |
1552 | /* No locking is required for the next_orphan queue as there | |
1553 | * is only ever a single process doing orphan recovery. */ | |
1554 | OCFS2_I(iter)->ip_next_orphan = p->head; | |
1555 | p->head = iter; | |
1556 | ||
1557 | return 0; | |
1558 | } | |
1559 | ||
b4df6ed8 MF |
1560 | static int ocfs2_queue_orphans(struct ocfs2_super *osb, |
1561 | int slot, | |
1562 | struct inode **head) | |
ccd979bd | 1563 | { |
b4df6ed8 | 1564 | int status; |
ccd979bd | 1565 | struct inode *orphan_dir_inode = NULL; |
5eae5b96 MF |
1566 | struct ocfs2_orphan_filldir_priv priv; |
1567 | loff_t pos = 0; | |
1568 | ||
1569 | priv.osb = osb; | |
1570 | priv.head = *head; | |
ccd979bd MF |
1571 | |
1572 | orphan_dir_inode = ocfs2_get_system_file_inode(osb, | |
1573 | ORPHAN_DIR_SYSTEM_INODE, | |
1574 | slot); | |
1575 | if (!orphan_dir_inode) { | |
1576 | status = -ENOENT; | |
1577 | mlog_errno(status); | |
b4df6ed8 MF |
1578 | return status; |
1579 | } | |
ccd979bd | 1580 | |
1b1dcc1b | 1581 | mutex_lock(&orphan_dir_inode->i_mutex); |
e63aecb6 | 1582 | status = ocfs2_inode_lock(orphan_dir_inode, NULL, 0); |
ccd979bd | 1583 | if (status < 0) { |
ccd979bd MF |
1584 | mlog_errno(status); |
1585 | goto out; | |
1586 | } | |
ccd979bd | 1587 | |
5eae5b96 MF |
1588 | status = ocfs2_dir_foreach(orphan_dir_inode, &pos, &priv, |
1589 | ocfs2_orphan_filldir); | |
1590 | if (status) { | |
1591 | mlog_errno(status); | |
a86370fb | 1592 | goto out_cluster; |
ccd979bd | 1593 | } |
ccd979bd | 1594 | |
5eae5b96 MF |
1595 | *head = priv.head; |
1596 | ||
a86370fb | 1597 | out_cluster: |
e63aecb6 | 1598 | ocfs2_inode_unlock(orphan_dir_inode, 0); |
b4df6ed8 MF |
1599 | out: |
1600 | mutex_unlock(&orphan_dir_inode->i_mutex); | |
ccd979bd | 1601 | iput(orphan_dir_inode); |
b4df6ed8 MF |
1602 | return status; |
1603 | } | |
1604 | ||
1605 | static int ocfs2_orphan_recovery_can_continue(struct ocfs2_super *osb, | |
1606 | int slot) | |
1607 | { | |
1608 | int ret; | |
1609 | ||
1610 | spin_lock(&osb->osb_lock); | |
1611 | ret = !osb->osb_orphan_wipes[slot]; | |
1612 | spin_unlock(&osb->osb_lock); | |
1613 | return ret; | |
1614 | } | |
1615 | ||
1616 | static void ocfs2_mark_recovering_orphan_dir(struct ocfs2_super *osb, | |
1617 | int slot) | |
1618 | { | |
1619 | spin_lock(&osb->osb_lock); | |
1620 | /* Mark ourselves such that new processes in delete_inode() | |
1621 | * know to quit early. */ | |
1622 | ocfs2_node_map_set_bit(osb, &osb->osb_recovering_orphan_dirs, slot); | |
1623 | while (osb->osb_orphan_wipes[slot]) { | |
1624 | /* If any processes are already in the middle of an | |
1625 | * orphan wipe on this dir, then we need to wait for | |
1626 | * them. */ | |
1627 | spin_unlock(&osb->osb_lock); | |
1628 | wait_event_interruptible(osb->osb_wipe_event, | |
1629 | ocfs2_orphan_recovery_can_continue(osb, slot)); | |
1630 | spin_lock(&osb->osb_lock); | |
1631 | } | |
1632 | spin_unlock(&osb->osb_lock); | |
1633 | } | |
1634 | ||
1635 | static void ocfs2_clear_recovering_orphan_dir(struct ocfs2_super *osb, | |
1636 | int slot) | |
1637 | { | |
1638 | ocfs2_node_map_clear_bit(osb, &osb->osb_recovering_orphan_dirs, slot); | |
1639 | } | |
1640 | ||
1641 | /* | |
1642 | * Orphan recovery. Each mounted node has it's own orphan dir which we | |
1643 | * must run during recovery. Our strategy here is to build a list of | |
1644 | * the inodes in the orphan dir and iget/iput them. The VFS does | |
1645 | * (most) of the rest of the work. | |
1646 | * | |
1647 | * Orphan recovery can happen at any time, not just mount so we have a | |
1648 | * couple of extra considerations. | |
1649 | * | |
1650 | * - We grab as many inodes as we can under the orphan dir lock - | |
1651 | * doing iget() outside the orphan dir risks getting a reference on | |
1652 | * an invalid inode. | |
1653 | * - We must be sure not to deadlock with other processes on the | |
1654 | * system wanting to run delete_inode(). This can happen when they go | |
1655 | * to lock the orphan dir and the orphan recovery process attempts to | |
1656 | * iget() inside the orphan dir lock. This can be avoided by | |
1657 | * advertising our state to ocfs2_delete_inode(). | |
1658 | */ | |
1659 | static int ocfs2_recover_orphans(struct ocfs2_super *osb, | |
1660 | int slot) | |
1661 | { | |
1662 | int ret = 0; | |
1663 | struct inode *inode = NULL; | |
1664 | struct inode *iter; | |
1665 | struct ocfs2_inode_info *oi; | |
1666 | ||
1667 | mlog(0, "Recover inodes from orphan dir in slot %d\n", slot); | |
1668 | ||
1669 | ocfs2_mark_recovering_orphan_dir(osb, slot); | |
1670 | ret = ocfs2_queue_orphans(osb, slot, &inode); | |
1671 | ocfs2_clear_recovering_orphan_dir(osb, slot); | |
1672 | ||
1673 | /* Error here should be noted, but we want to continue with as | |
1674 | * many queued inodes as we've got. */ | |
1675 | if (ret) | |
1676 | mlog_errno(ret); | |
ccd979bd MF |
1677 | |
1678 | while (inode) { | |
1679 | oi = OCFS2_I(inode); | |
b0697053 | 1680 | mlog(0, "iput orphan %llu\n", (unsigned long long)oi->ip_blkno); |
ccd979bd MF |
1681 | |
1682 | iter = oi->ip_next_orphan; | |
1683 | ||
1684 | spin_lock(&oi->ip_lock); | |
34d024f8 MF |
1685 | /* The remote delete code may have set these on the |
1686 | * assumption that the other node would wipe them | |
1687 | * successfully. If they are still in the node's | |
1688 | * orphan dir, we need to reset that state. */ | |
ccd979bd MF |
1689 | oi->ip_flags &= ~(OCFS2_INODE_DELETED|OCFS2_INODE_SKIP_DELETE); |
1690 | ||
1691 | /* Set the proper information to get us going into | |
1692 | * ocfs2_delete_inode. */ | |
1693 | oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED; | |
ccd979bd MF |
1694 | spin_unlock(&oi->ip_lock); |
1695 | ||
1696 | iput(inode); | |
1697 | ||
1698 | inode = iter; | |
1699 | } | |
1700 | ||
b4df6ed8 | 1701 | return ret; |
ccd979bd MF |
1702 | } |
1703 | ||
1704 | static int ocfs2_wait_on_mount(struct ocfs2_super *osb) | |
1705 | { | |
1706 | /* This check is good because ocfs2 will wait on our recovery | |
1707 | * thread before changing it to something other than MOUNTED | |
1708 | * or DISABLED. */ | |
1709 | wait_event(osb->osb_mount_event, | |
1710 | atomic_read(&osb->vol_state) == VOLUME_MOUNTED || | |
1711 | atomic_read(&osb->vol_state) == VOLUME_DISABLED); | |
1712 | ||
1713 | /* If there's an error on mount, then we may never get to the | |
1714 | * MOUNTED flag, but this is set right before | |
1715 | * dismount_volume() so we can trust it. */ | |
1716 | if (atomic_read(&osb->vol_state) == VOLUME_DISABLED) { | |
1717 | mlog(0, "mount error, exiting!\n"); | |
1718 | return -EBUSY; | |
1719 | } | |
1720 | ||
1721 | return 0; | |
1722 | } | |
1723 | ||
1724 | static int ocfs2_commit_thread(void *arg) | |
1725 | { | |
1726 | int status; | |
1727 | struct ocfs2_super *osb = arg; | |
1728 | struct ocfs2_journal *journal = osb->journal; | |
1729 | ||
1730 | /* we can trust j_num_trans here because _should_stop() is only set in | |
1731 | * shutdown and nobody other than ourselves should be able to start | |
1732 | * transactions. committing on shutdown might take a few iterations | |
1733 | * as final transactions put deleted inodes on the list */ | |
1734 | while (!(kthread_should_stop() && | |
1735 | atomic_read(&journal->j_num_trans) == 0)) { | |
1736 | ||
745ae8ba MF |
1737 | wait_event_interruptible(osb->checkpoint_event, |
1738 | atomic_read(&journal->j_num_trans) | |
1739 | || kthread_should_stop()); | |
ccd979bd MF |
1740 | |
1741 | status = ocfs2_commit_cache(osb); | |
1742 | if (status < 0) | |
1743 | mlog_errno(status); | |
1744 | ||
1745 | if (kthread_should_stop() && atomic_read(&journal->j_num_trans)){ | |
1746 | mlog(ML_KTHREAD, | |
1747 | "commit_thread: %u transactions pending on " | |
1748 | "shutdown\n", | |
1749 | atomic_read(&journal->j_num_trans)); | |
1750 | } | |
1751 | } | |
1752 | ||
1753 | return 0; | |
1754 | } | |
1755 | ||
539d8264 SM |
1756 | /* Reads all the journal inodes without taking any cluster locks. Used |
1757 | * for hard readonly access to determine whether any journal requires | |
1758 | * recovery. Also used to refresh the recovery generation numbers after | |
1759 | * a journal has been recovered by another node. | |
1760 | */ | |
ccd979bd MF |
1761 | int ocfs2_check_journals_nolocks(struct ocfs2_super *osb) |
1762 | { | |
1763 | int ret = 0; | |
1764 | unsigned int slot; | |
539d8264 | 1765 | struct buffer_head *di_bh = NULL; |
ccd979bd | 1766 | struct ocfs2_dinode *di; |
539d8264 | 1767 | int journal_dirty = 0; |
ccd979bd MF |
1768 | |
1769 | for(slot = 0; slot < osb->max_slots; slot++) { | |
539d8264 SM |
1770 | ret = ocfs2_read_journal_inode(osb, slot, &di_bh, NULL); |
1771 | if (ret) { | |
ccd979bd MF |
1772 | mlog_errno(ret); |
1773 | goto out; | |
1774 | } | |
1775 | ||
1776 | di = (struct ocfs2_dinode *) di_bh->b_data; | |
1777 | ||
539d8264 SM |
1778 | osb->slot_recovery_generations[slot] = |
1779 | ocfs2_get_recovery_generation(di); | |
1780 | ||
ccd979bd MF |
1781 | if (le32_to_cpu(di->id1.journal1.ij_flags) & |
1782 | OCFS2_JOURNAL_DIRTY_FL) | |
539d8264 | 1783 | journal_dirty = 1; |
ccd979bd MF |
1784 | |
1785 | brelse(di_bh); | |
539d8264 | 1786 | di_bh = NULL; |
ccd979bd MF |
1787 | } |
1788 | ||
1789 | out: | |
539d8264 SM |
1790 | if (journal_dirty) |
1791 | ret = -EROFS; | |
ccd979bd MF |
1792 | return ret; |
1793 | } |