blob: f92bf1dd379a3faa46482d712c3f861aff8eb98e [file] [log] [blame]
Mark Fashehccd979b2005-12-15 14:31:24 -08001/* -*- 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"
38#include "dlmglue.h"
39#include "extent_map.h"
40#include "heartbeat.h"
41#include "inode.h"
42#include "journal.h"
43#include "localalloc.h"
44#include "namei.h"
45#include "slot_map.h"
46#include "super.h"
47#include "vote.h"
48#include "sysfile.h"
49
50#include "buffer_head_io.h"
51
Ingo Molnar34af9462006-06-27 02:53:55 -070052DEFINE_SPINLOCK(trans_inc_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -080053
54static int ocfs2_force_read_journal(struct inode *inode);
55static int ocfs2_recover_node(struct ocfs2_super *osb,
56 int node_num);
57static int __ocfs2_recovery_thread(void *arg);
58static int ocfs2_commit_cache(struct ocfs2_super *osb);
59static int ocfs2_wait_on_mount(struct ocfs2_super *osb);
60static void ocfs2_handle_cleanup_locks(struct ocfs2_journal *journal,
61 struct ocfs2_journal_handle *handle);
62static void ocfs2_commit_unstarted_handle(struct ocfs2_journal_handle *handle);
63static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
64 int dirty);
65static int ocfs2_trylock_journal(struct ocfs2_super *osb,
66 int slot_num);
67static int ocfs2_recover_orphans(struct ocfs2_super *osb,
68 int slot);
69static int ocfs2_commit_thread(void *arg);
70
71static int ocfs2_commit_cache(struct ocfs2_super *osb)
72{
73 int status = 0;
74 unsigned int flushed;
75 unsigned long old_id;
76 struct ocfs2_journal *journal = NULL;
77
78 mlog_entry_void();
79
80 journal = osb->journal;
81
82 /* Flush all pending commits and checkpoint the journal. */
83 down_write(&journal->j_trans_barrier);
84
85 if (atomic_read(&journal->j_num_trans) == 0) {
86 up_write(&journal->j_trans_barrier);
87 mlog(0, "No transactions for me to flush!\n");
88 goto finally;
89 }
90
91 journal_lock_updates(journal->j_journal);
92 status = journal_flush(journal->j_journal);
93 journal_unlock_updates(journal->j_journal);
94 if (status < 0) {
95 up_write(&journal->j_trans_barrier);
96 mlog_errno(status);
97 goto finally;
98 }
99
100 old_id = ocfs2_inc_trans_id(journal);
101
102 flushed = atomic_read(&journal->j_num_trans);
103 atomic_set(&journal->j_num_trans, 0);
104 up_write(&journal->j_trans_barrier);
105
106 mlog(0, "commit_thread: flushed transaction %lu (%u handles)\n",
107 journal->j_trans_id, flushed);
108
109 ocfs2_kick_vote_thread(osb);
110 wake_up(&journal->j_checkpointed);
111finally:
112 mlog_exit(status);
113 return status;
114}
115
116struct ocfs2_journal_handle *ocfs2_alloc_handle(struct ocfs2_super *osb)
117{
118 struct ocfs2_journal_handle *retval = NULL;
119
Sunil Mushranafae00ab2006-04-12 14:37:00 -0700120 retval = kcalloc(1, sizeof(*retval), GFP_NOFS);
Mark Fashehccd979b2005-12-15 14:31:24 -0800121 if (!retval) {
122 mlog(ML_ERROR, "Failed to allocate memory for journal "
123 "handle!\n");
124 return NULL;
125 }
126
127 retval->max_buffs = 0;
128 retval->num_locks = 0;
129 retval->k_handle = NULL;
130
131 INIT_LIST_HEAD(&retval->locks);
132 INIT_LIST_HEAD(&retval->inode_list);
133 retval->journal = osb->journal;
134
135 return retval;
136}
137
138/* pass it NULL and it will allocate a new handle object for you. If
139 * you pass it a handle however, it may still return error, in which
140 * case it has free'd the passed handle for you. */
141struct ocfs2_journal_handle *ocfs2_start_trans(struct ocfs2_super *osb,
142 struct ocfs2_journal_handle *handle,
143 int max_buffs)
144{
145 int ret;
146 journal_t *journal = osb->journal->j_journal;
147
148 mlog_entry("(max_buffs = %d)\n", max_buffs);
149
Eric Sesterhenn / snakebyteebdec832006-01-27 10:32:52 +0100150 BUG_ON(!osb || !osb->journal->j_journal);
Mark Fashehccd979b2005-12-15 14:31:24 -0800151
152 if (ocfs2_is_hard_readonly(osb)) {
153 ret = -EROFS;
154 goto done_free;
155 }
156
157 BUG_ON(osb->journal->j_state == OCFS2_JOURNAL_FREE);
158 BUG_ON(max_buffs <= 0);
159
160 /* JBD might support this, but our journalling code doesn't yet. */
161 if (journal_current_handle()) {
162 mlog(ML_ERROR, "Recursive transaction attempted!\n");
163 BUG();
164 }
165
166 if (!handle)
167 handle = ocfs2_alloc_handle(osb);
168 if (!handle) {
169 ret = -ENOMEM;
170 mlog(ML_ERROR, "Failed to allocate memory for journal "
171 "handle!\n");
172 goto done_free;
173 }
174
175 handle->max_buffs = max_buffs;
176
177 down_read(&osb->journal->j_trans_barrier);
178
179 /* actually start the transaction now */
180 handle->k_handle = journal_start(journal, max_buffs);
181 if (IS_ERR(handle->k_handle)) {
182 up_read(&osb->journal->j_trans_barrier);
183
184 ret = PTR_ERR(handle->k_handle);
185 handle->k_handle = NULL;
186 mlog_errno(ret);
187
188 if (is_journal_aborted(journal)) {
189 ocfs2_abort(osb->sb, "Detected aborted journal");
190 ret = -EROFS;
191 }
192 goto done_free;
193 }
194
195 atomic_inc(&(osb->journal->j_num_trans));
196 handle->flags |= OCFS2_HANDLE_STARTED;
197
198 mlog_exit_ptr(handle);
199 return handle;
200
201done_free:
202 if (handle)
203 ocfs2_commit_unstarted_handle(handle); /* will kfree handle */
204
205 mlog_exit(ret);
206 return ERR_PTR(ret);
207}
208
209void ocfs2_handle_add_inode(struct ocfs2_journal_handle *handle,
210 struct inode *inode)
211{
212 BUG_ON(!handle);
213 BUG_ON(!inode);
214
215 atomic_inc(&inode->i_count);
216
217 /* we're obviously changing it... */
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800218 mutex_lock(&inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800219
220 /* sanity check */
221 BUG_ON(OCFS2_I(inode)->ip_handle);
222 BUG_ON(!list_empty(&OCFS2_I(inode)->ip_handle_list));
223
224 OCFS2_I(inode)->ip_handle = handle;
Akinobu Mitaf1166292006-06-26 00:24:46 -0700225 list_move_tail(&(OCFS2_I(inode)->ip_handle_list), &(handle->inode_list));
Mark Fashehccd979b2005-12-15 14:31:24 -0800226}
227
228static void ocfs2_handle_unlock_inodes(struct ocfs2_journal_handle *handle)
229{
230 struct list_head *p, *n;
231 struct inode *inode;
232 struct ocfs2_inode_info *oi;
233
234 list_for_each_safe(p, n, &handle->inode_list) {
235 oi = list_entry(p, struct ocfs2_inode_info,
236 ip_handle_list);
237 inode = &oi->vfs_inode;
238
239 OCFS2_I(inode)->ip_handle = NULL;
240 list_del_init(&OCFS2_I(inode)->ip_handle_list);
241
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800242 mutex_unlock(&inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800243 iput(inode);
244 }
245}
246
247/* This is trivial so we do it out of the main commit
248 * paths. Beware, it can be called from start_trans too! */
249static void ocfs2_commit_unstarted_handle(struct ocfs2_journal_handle *handle)
250{
251 mlog_entry_void();
252
253 BUG_ON(handle->flags & OCFS2_HANDLE_STARTED);
254
255 ocfs2_handle_unlock_inodes(handle);
256 /* You are allowed to add journal locks before the transaction
257 * has started. */
258 ocfs2_handle_cleanup_locks(handle->journal, handle);
259
260 kfree(handle);
261
262 mlog_exit_void();
263}
264
265void ocfs2_commit_trans(struct ocfs2_journal_handle *handle)
266{
267 handle_t *jbd_handle;
268 int retval;
269 struct ocfs2_journal *journal = handle->journal;
270
271 mlog_entry_void();
272
273 BUG_ON(!handle);
274
275 if (!(handle->flags & OCFS2_HANDLE_STARTED)) {
276 ocfs2_commit_unstarted_handle(handle);
277 mlog_exit_void();
278 return;
279 }
280
281 /* release inode semaphores we took during this transaction */
282 ocfs2_handle_unlock_inodes(handle);
283
284 /* ocfs2_extend_trans may have had to call journal_restart
285 * which will always commit the transaction, but may return
286 * error for any number of reasons. If this is the case, we
287 * clear k_handle as it's not valid any more. */
288 if (handle->k_handle) {
289 jbd_handle = handle->k_handle;
290
291 if (handle->flags & OCFS2_HANDLE_SYNC)
292 jbd_handle->h_sync = 1;
293 else
294 jbd_handle->h_sync = 0;
295
296 /* actually stop the transaction. if we've set h_sync,
297 * it'll have been committed when we return */
298 retval = journal_stop(jbd_handle);
299 if (retval < 0) {
300 mlog_errno(retval);
301 mlog(ML_ERROR, "Could not commit transaction\n");
302 BUG();
303 }
304
305 handle->k_handle = NULL; /* it's been free'd in journal_stop */
306 }
307
308 ocfs2_handle_cleanup_locks(journal, handle);
309
310 up_read(&journal->j_trans_barrier);
311
312 kfree(handle);
313 mlog_exit_void();
314}
315
316/*
317 * 'nblocks' is what you want to add to the current
318 * transaction. extend_trans will either extend the current handle by
319 * nblocks, or commit it and start a new one with nblocks credits.
320 *
321 * WARNING: This will not release any semaphores or disk locks taken
322 * during the transaction, so make sure they were taken *before*
323 * start_trans or we'll have ordering deadlocks.
324 *
325 * WARNING2: Note that we do *not* drop j_trans_barrier here. This is
326 * good because transaction ids haven't yet been recorded on the
327 * cluster locks associated with this handle.
328 */
329int ocfs2_extend_trans(struct ocfs2_journal_handle *handle,
330 int nblocks)
331{
332 int status;
333
334 BUG_ON(!handle);
335 BUG_ON(!(handle->flags & OCFS2_HANDLE_STARTED));
336 BUG_ON(!nblocks);
337
338 mlog_entry_void();
339
340 mlog(0, "Trying to extend transaction by %d blocks\n", nblocks);
341
342 status = journal_extend(handle->k_handle, nblocks);
343 if (status < 0) {
344 mlog_errno(status);
345 goto bail;
346 }
347
348 if (status > 0) {
349 mlog(0, "journal_extend failed, trying journal_restart\n");
350 status = journal_restart(handle->k_handle, nblocks);
351 if (status < 0) {
352 handle->k_handle = NULL;
353 mlog_errno(status);
354 goto bail;
355 }
356 handle->max_buffs = nblocks;
357 } else
358 handle->max_buffs += nblocks;
359
360 status = 0;
361bail:
362
363 mlog_exit(status);
364 return status;
365}
366
367int ocfs2_journal_access(struct ocfs2_journal_handle *handle,
368 struct inode *inode,
369 struct buffer_head *bh,
370 int type)
371{
372 int status;
373
374 BUG_ON(!inode);
375 BUG_ON(!handle);
376 BUG_ON(!bh);
377 BUG_ON(!(handle->flags & OCFS2_HANDLE_STARTED));
378
Badari Pulavarty205f87f2006-03-26 01:38:00 -0800379 mlog_entry("bh->b_blocknr=%llu, type=%d (\"%s\"), bh->b_size = %zu\n",
Mark Fashehccd979b2005-12-15 14:31:24 -0800380 (unsigned long long)bh->b_blocknr, type,
381 (type == OCFS2_JOURNAL_ACCESS_CREATE) ?
382 "OCFS2_JOURNAL_ACCESS_CREATE" :
383 "OCFS2_JOURNAL_ACCESS_WRITE",
384 bh->b_size);
385
386 /* we can safely remove this assertion after testing. */
387 if (!buffer_uptodate(bh)) {
388 mlog(ML_ERROR, "giving me a buffer that's not uptodate!\n");
389 mlog(ML_ERROR, "b_blocknr=%llu\n",
390 (unsigned long long)bh->b_blocknr);
391 BUG();
392 }
393
394 /* Set the current transaction information on the inode so
395 * that the locking code knows whether it can drop it's locks
396 * on this inode or not. We're protected from the commit
397 * thread updating the current transaction id until
398 * ocfs2_commit_trans() because ocfs2_start_trans() took
399 * j_trans_barrier for us. */
400 ocfs2_set_inode_lock_trans(OCFS2_SB(inode->i_sb)->journal, inode);
401
Mark Fasheh251b6ec2006-01-10 15:41:43 -0800402 mutex_lock(&OCFS2_I(inode)->ip_io_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800403 switch (type) {
404 case OCFS2_JOURNAL_ACCESS_CREATE:
405 case OCFS2_JOURNAL_ACCESS_WRITE:
406 status = journal_get_write_access(handle->k_handle, bh);
407 break;
408
409 case OCFS2_JOURNAL_ACCESS_UNDO:
410 status = journal_get_undo_access(handle->k_handle, bh);
411 break;
412
413 default:
414 status = -EINVAL;
415 mlog(ML_ERROR, "Uknown access type!\n");
416 }
Mark Fasheh251b6ec2006-01-10 15:41:43 -0800417 mutex_unlock(&OCFS2_I(inode)->ip_io_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800418
419 if (status < 0)
420 mlog(ML_ERROR, "Error %d getting %d access to buffer!\n",
421 status, type);
422
423 mlog_exit(status);
424 return status;
425}
426
427int ocfs2_journal_dirty(struct ocfs2_journal_handle *handle,
428 struct buffer_head *bh)
429{
430 int status;
431
432 BUG_ON(!(handle->flags & OCFS2_HANDLE_STARTED));
433
434 mlog_entry("(bh->b_blocknr=%llu)\n",
435 (unsigned long long)bh->b_blocknr);
436
437 status = journal_dirty_metadata(handle->k_handle, bh);
438 if (status < 0)
439 mlog(ML_ERROR, "Could not dirty metadata buffer. "
440 "(bh->b_blocknr=%llu)\n",
441 (unsigned long long)bh->b_blocknr);
442
443 mlog_exit(status);
444 return status;
445}
446
447int ocfs2_journal_dirty_data(handle_t *handle,
448 struct buffer_head *bh)
449{
450 int err = journal_dirty_data(handle, bh);
451 if (err)
452 mlog_errno(err);
453 /* TODO: When we can handle it, abort the handle and go RO on
454 * error here. */
455
456 return err;
457}
458
459/* We always assume you're adding a metadata lock at level 'ex' */
460int ocfs2_handle_add_lock(struct ocfs2_journal_handle *handle,
461 struct inode *inode)
462{
463 int status;
464 struct ocfs2_journal_lock *lock;
465
466 BUG_ON(!inode);
467
468 lock = kmem_cache_alloc(ocfs2_lock_cache, GFP_NOFS);
469 if (!lock) {
470 status = -ENOMEM;
471 mlog_errno(-ENOMEM);
472 goto bail;
473 }
474
475 if (!igrab(inode))
476 BUG();
477 lock->jl_inode = inode;
478
479 list_add_tail(&(lock->jl_lock_list), &(handle->locks));
480 handle->num_locks++;
481
482 status = 0;
483bail:
484 mlog_exit(status);
485 return status;
486}
487
488static void ocfs2_handle_cleanup_locks(struct ocfs2_journal *journal,
489 struct ocfs2_journal_handle *handle)
490{
491 struct list_head *p, *n;
492 struct ocfs2_journal_lock *lock;
493 struct inode *inode;
494
495 list_for_each_safe(p, n, &(handle->locks)) {
496 lock = list_entry(p, struct ocfs2_journal_lock,
497 jl_lock_list);
498 list_del(&lock->jl_lock_list);
499 handle->num_locks--;
500
501 inode = lock->jl_inode;
502 ocfs2_meta_unlock(inode, 1);
503 if (atomic_read(&inode->i_count) == 1)
504 mlog(ML_ERROR,
Mark Fashehb0697052006-03-03 10:24:33 -0800505 "Inode %llu, I'm doing a last iput for!",
506 (unsigned long long)OCFS2_I(inode)->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800507 iput(inode);
508 kmem_cache_free(ocfs2_lock_cache, lock);
509 }
510}
511
512#define OCFS2_DEFAULT_COMMIT_INTERVAL (HZ * 5)
513
514void ocfs2_set_journal_params(struct ocfs2_super *osb)
515{
516 journal_t *journal = osb->journal->j_journal;
517
518 spin_lock(&journal->j_state_lock);
519 journal->j_commit_interval = OCFS2_DEFAULT_COMMIT_INTERVAL;
520 if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER)
521 journal->j_flags |= JFS_BARRIER;
522 else
523 journal->j_flags &= ~JFS_BARRIER;
524 spin_unlock(&journal->j_state_lock);
525}
526
527int ocfs2_journal_init(struct ocfs2_journal *journal, int *dirty)
528{
529 int status = -1;
530 struct inode *inode = NULL; /* the journal inode */
531 journal_t *j_journal = NULL;
532 struct ocfs2_dinode *di = NULL;
533 struct buffer_head *bh = NULL;
534 struct ocfs2_super *osb;
535 int meta_lock = 0;
536
537 mlog_entry_void();
538
539 BUG_ON(!journal);
540
541 osb = journal->j_osb;
542
543 /* already have the inode for our journal */
544 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
545 osb->slot_num);
546 if (inode == NULL) {
547 status = -EACCES;
548 mlog_errno(status);
549 goto done;
550 }
551 if (is_bad_inode(inode)) {
552 mlog(ML_ERROR, "access error (bad inode)\n");
553 iput(inode);
554 inode = NULL;
555 status = -EACCES;
556 goto done;
557 }
558
559 SET_INODE_JOURNAL(inode);
560 OCFS2_I(inode)->ip_open_count++;
561
Mark Fasheh6eff5792006-01-18 10:31:47 -0800562 /* Skip recovery waits here - journal inode metadata never
563 * changes in a live cluster so it can be considered an
564 * exception to the rule. */
565 status = ocfs2_meta_lock_full(inode, NULL, &bh, 1,
566 OCFS2_META_LOCK_RECOVERY);
Mark Fashehccd979b2005-12-15 14:31:24 -0800567 if (status < 0) {
568 if (status != -ERESTARTSYS)
569 mlog(ML_ERROR, "Could not get lock on journal!\n");
570 goto done;
571 }
572
573 meta_lock = 1;
574 di = (struct ocfs2_dinode *)bh->b_data;
575
576 if (inode->i_size < OCFS2_MIN_JOURNAL_SIZE) {
577 mlog(ML_ERROR, "Journal file size (%lld) is too small!\n",
578 inode->i_size);
579 status = -EINVAL;
580 goto done;
581 }
582
583 mlog(0, "inode->i_size = %lld\n", inode->i_size);
Andrew Morton5515eff2006-03-26 01:37:53 -0800584 mlog(0, "inode->i_blocks = %llu\n",
585 (unsigned long long)inode->i_blocks);
Mark Fashehccd979b2005-12-15 14:31:24 -0800586 mlog(0, "inode->ip_clusters = %u\n", OCFS2_I(inode)->ip_clusters);
587
588 /* call the kernels journal init function now */
589 j_journal = journal_init_inode(inode);
590 if (j_journal == NULL) {
591 mlog(ML_ERROR, "Linux journal layer error\n");
592 status = -EINVAL;
593 goto done;
594 }
595
596 mlog(0, "Returned from journal_init_inode\n");
597 mlog(0, "j_journal->j_maxlen = %u\n", j_journal->j_maxlen);
598
599 *dirty = (le32_to_cpu(di->id1.journal1.ij_flags) &
600 OCFS2_JOURNAL_DIRTY_FL);
601
602 journal->j_journal = j_journal;
603 journal->j_inode = inode;
604 journal->j_bh = bh;
605
606 ocfs2_set_journal_params(osb);
607
608 journal->j_state = OCFS2_JOURNAL_LOADED;
609
610 status = 0;
611done:
612 if (status < 0) {
613 if (meta_lock)
614 ocfs2_meta_unlock(inode, 1);
615 if (bh != NULL)
616 brelse(bh);
617 if (inode) {
618 OCFS2_I(inode)->ip_open_count--;
619 iput(inode);
620 }
621 }
622
623 mlog_exit(status);
624 return status;
625}
626
627static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
628 int dirty)
629{
630 int status;
631 unsigned int flags;
632 struct ocfs2_journal *journal = osb->journal;
633 struct buffer_head *bh = journal->j_bh;
634 struct ocfs2_dinode *fe;
635
636 mlog_entry_void();
637
638 fe = (struct ocfs2_dinode *)bh->b_data;
639 if (!OCFS2_IS_VALID_DINODE(fe)) {
640 /* This is called from startup/shutdown which will
641 * handle the errors in a specific manner, so no need
642 * to call ocfs2_error() here. */
Mark Fashehb0697052006-03-03 10:24:33 -0800643 mlog(ML_ERROR, "Journal dinode %llu has invalid "
644 "signature: %.*s", (unsigned long long)fe->i_blkno, 7,
645 fe->i_signature);
Mark Fashehccd979b2005-12-15 14:31:24 -0800646 status = -EIO;
647 goto out;
648 }
649
650 flags = le32_to_cpu(fe->id1.journal1.ij_flags);
651 if (dirty)
652 flags |= OCFS2_JOURNAL_DIRTY_FL;
653 else
654 flags &= ~OCFS2_JOURNAL_DIRTY_FL;
655 fe->id1.journal1.ij_flags = cpu_to_le32(flags);
656
657 status = ocfs2_write_block(osb, bh, journal->j_inode);
658 if (status < 0)
659 mlog_errno(status);
660
661out:
662 mlog_exit(status);
663 return status;
664}
665
666/*
667 * If the journal has been kmalloc'd it needs to be freed after this
668 * call.
669 */
670void ocfs2_journal_shutdown(struct ocfs2_super *osb)
671{
672 struct ocfs2_journal *journal = NULL;
673 int status = 0;
674 struct inode *inode = NULL;
675 int num_running_trans = 0;
676
677 mlog_entry_void();
678
Eric Sesterhenn / snakebyteebdec832006-01-27 10:32:52 +0100679 BUG_ON(!osb);
Mark Fashehccd979b2005-12-15 14:31:24 -0800680
681 journal = osb->journal;
682 if (!journal)
683 goto done;
684
685 inode = journal->j_inode;
686
687 if (journal->j_state != OCFS2_JOURNAL_LOADED)
688 goto done;
689
690 /* need to inc inode use count as journal_destroy will iput. */
691 if (!igrab(inode))
692 BUG();
693
694 num_running_trans = atomic_read(&(osb->journal->j_num_trans));
695 if (num_running_trans > 0)
696 mlog(0, "Shutting down journal: must wait on %d "
697 "running transactions!\n",
698 num_running_trans);
699
700 /* Do a commit_cache here. It will flush our journal, *and*
701 * release any locks that are still held.
702 * set the SHUTDOWN flag and release the trans lock.
703 * the commit thread will take the trans lock for us below. */
704 journal->j_state = OCFS2_JOURNAL_IN_SHUTDOWN;
705
706 /* The OCFS2_JOURNAL_IN_SHUTDOWN will signal to commit_cache to not
707 * drop the trans_lock (which we want to hold until we
708 * completely destroy the journal. */
709 if (osb->commit_task) {
710 /* Wait for the commit thread */
711 mlog(0, "Waiting for ocfs2commit to exit....\n");
712 kthread_stop(osb->commit_task);
713 osb->commit_task = NULL;
714 }
715
716 BUG_ON(atomic_read(&(osb->journal->j_num_trans)) != 0);
717
718 status = ocfs2_journal_toggle_dirty(osb, 0);
719 if (status < 0)
720 mlog_errno(status);
721
722 /* Shutdown the kernel journal system */
723 journal_destroy(journal->j_journal);
724
725 OCFS2_I(inode)->ip_open_count--;
726
727 /* unlock our journal */
728 ocfs2_meta_unlock(inode, 1);
729
730 brelse(journal->j_bh);
731 journal->j_bh = NULL;
732
733 journal->j_state = OCFS2_JOURNAL_FREE;
734
735// up_write(&journal->j_trans_barrier);
736done:
737 if (inode)
738 iput(inode);
739 mlog_exit_void();
740}
741
742static void ocfs2_clear_journal_error(struct super_block *sb,
743 journal_t *journal,
744 int slot)
745{
746 int olderr;
747
748 olderr = journal_errno(journal);
749 if (olderr) {
750 mlog(ML_ERROR, "File system error %d recorded in "
751 "journal %u.\n", olderr, slot);
752 mlog(ML_ERROR, "File system on device %s needs checking.\n",
753 sb->s_id);
754
755 journal_ack_err(journal);
756 journal_clear_err(journal);
757 }
758}
759
760int ocfs2_journal_load(struct ocfs2_journal *journal)
761{
762 int status = 0;
763 struct ocfs2_super *osb;
764
765 mlog_entry_void();
766
767 if (!journal)
768 BUG();
769
770 osb = journal->j_osb;
771
772 status = journal_load(journal->j_journal);
773 if (status < 0) {
774 mlog(ML_ERROR, "Failed to load journal!\n");
775 goto done;
776 }
777
778 ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num);
779
780 status = ocfs2_journal_toggle_dirty(osb, 1);
781 if (status < 0) {
782 mlog_errno(status);
783 goto done;
784 }
785
786 /* Launch the commit thread */
Mark Fasheh78427042006-05-04 12:03:26 -0700787 osb->commit_task = kthread_run(ocfs2_commit_thread, osb, "ocfs2cmt");
Mark Fashehccd979b2005-12-15 14:31:24 -0800788 if (IS_ERR(osb->commit_task)) {
789 status = PTR_ERR(osb->commit_task);
790 osb->commit_task = NULL;
791 mlog(ML_ERROR, "unable to launch ocfs2commit thread, error=%d",
792 status);
793 goto done;
794 }
795
796done:
797 mlog_exit(status);
798 return status;
799}
800
801
802/* 'full' flag tells us whether we clear out all blocks or if we just
803 * mark the journal clean */
804int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full)
805{
806 int status;
807
808 mlog_entry_void();
809
Eric Sesterhenn / snakebyteebdec832006-01-27 10:32:52 +0100810 BUG_ON(!journal);
Mark Fashehccd979b2005-12-15 14:31:24 -0800811
812 status = journal_wipe(journal->j_journal, full);
813 if (status < 0) {
814 mlog_errno(status);
815 goto bail;
816 }
817
818 status = ocfs2_journal_toggle_dirty(journal->j_osb, 0);
819 if (status < 0)
820 mlog_errno(status);
821
822bail:
823 mlog_exit(status);
824 return status;
825}
826
827/*
828 * JBD Might read a cached version of another nodes journal file. We
829 * don't want this as this file changes often and we get no
830 * notification on those changes. The only way to be sure that we've
831 * got the most up to date version of those blocks then is to force
832 * read them off disk. Just searching through the buffer cache won't
833 * work as there may be pages backing this file which are still marked
834 * up to date. We know things can't change on this file underneath us
835 * as we have the lock by now :)
836 */
837static int ocfs2_force_read_journal(struct inode *inode)
838{
839 int status = 0;
840 int i, p_blocks;
841 u64 v_blkno, p_blkno;
842#define CONCURRENT_JOURNAL_FILL 32
843 struct buffer_head *bhs[CONCURRENT_JOURNAL_FILL];
844
845 mlog_entry_void();
846
847 BUG_ON(inode->i_blocks !=
848 ocfs2_align_bytes_to_sectors(i_size_read(inode)));
849
850 memset(bhs, 0, sizeof(struct buffer_head *) * CONCURRENT_JOURNAL_FILL);
851
Andrew Morton5515eff2006-03-26 01:37:53 -0800852 mlog(0, "Force reading %llu blocks\n",
853 (unsigned long long)(inode->i_blocks >>
854 (inode->i_sb->s_blocksize_bits - 9)));
Mark Fashehccd979b2005-12-15 14:31:24 -0800855
856 v_blkno = 0;
857 while (v_blkno <
858 (inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9))) {
859
860 status = ocfs2_extent_map_get_blocks(inode, v_blkno,
861 1, &p_blkno,
862 &p_blocks);
863 if (status < 0) {
864 mlog_errno(status);
865 goto bail;
866 }
867
868 if (p_blocks > CONCURRENT_JOURNAL_FILL)
869 p_blocks = CONCURRENT_JOURNAL_FILL;
870
Mark Fashehdd4a2c22006-04-12 14:24:05 -0700871 /* We are reading journal data which should not
872 * be put in the uptodate cache */
Mark Fashehccd979b2005-12-15 14:31:24 -0800873 status = ocfs2_read_blocks(OCFS2_SB(inode->i_sb),
874 p_blkno, p_blocks, bhs, 0,
Mark Fashehdd4a2c22006-04-12 14:24:05 -0700875 NULL);
Mark Fashehccd979b2005-12-15 14:31:24 -0800876 if (status < 0) {
877 mlog_errno(status);
878 goto bail;
879 }
880
881 for(i = 0; i < p_blocks; i++) {
882 brelse(bhs[i]);
883 bhs[i] = NULL;
884 }
885
886 v_blkno += p_blocks;
887 }
888
889bail:
890 for(i = 0; i < CONCURRENT_JOURNAL_FILL; i++)
891 if (bhs[i])
892 brelse(bhs[i]);
893 mlog_exit(status);
894 return status;
895}
896
897struct ocfs2_la_recovery_item {
898 struct list_head lri_list;
899 int lri_slot;
900 struct ocfs2_dinode *lri_la_dinode;
901 struct ocfs2_dinode *lri_tl_dinode;
902};
903
904/* Does the second half of the recovery process. By this point, the
905 * node is marked clean and can actually be considered recovered,
906 * hence it's no longer in the recovery map, but there's still some
907 * cleanup we can do which shouldn't happen within the recovery thread
908 * as locking in that context becomes very difficult if we are to take
909 * recovering nodes into account.
910 *
911 * NOTE: This function can and will sleep on recovery of other nodes
912 * during cluster locking, just like any other ocfs2 process.
913 */
914void ocfs2_complete_recovery(void *data)
915{
916 int ret;
917 struct ocfs2_super *osb = data;
918 struct ocfs2_journal *journal = osb->journal;
919 struct ocfs2_dinode *la_dinode, *tl_dinode;
920 struct ocfs2_la_recovery_item *item;
921 struct list_head *p, *n;
922 LIST_HEAD(tmp_la_list);
923
924 mlog_entry_void();
925
926 mlog(0, "completing recovery from keventd\n");
927
928 spin_lock(&journal->j_lock);
929 list_splice_init(&journal->j_la_cleanups, &tmp_la_list);
930 spin_unlock(&journal->j_lock);
931
932 list_for_each_safe(p, n, &tmp_la_list) {
933 item = list_entry(p, struct ocfs2_la_recovery_item, lri_list);
934 list_del_init(&item->lri_list);
935
936 mlog(0, "Complete recovery for slot %d\n", item->lri_slot);
937
938 la_dinode = item->lri_la_dinode;
939 if (la_dinode) {
Mark Fashehb0697052006-03-03 10:24:33 -0800940 mlog(0, "Clean up local alloc %llu\n",
941 (unsigned long long)la_dinode->i_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800942
943 ret = ocfs2_complete_local_alloc_recovery(osb,
944 la_dinode);
945 if (ret < 0)
946 mlog_errno(ret);
947
948 kfree(la_dinode);
949 }
950
951 tl_dinode = item->lri_tl_dinode;
952 if (tl_dinode) {
Mark Fashehb0697052006-03-03 10:24:33 -0800953 mlog(0, "Clean up truncate log %llu\n",
954 (unsigned long long)tl_dinode->i_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800955
956 ret = ocfs2_complete_truncate_log_recovery(osb,
957 tl_dinode);
958 if (ret < 0)
959 mlog_errno(ret);
960
961 kfree(tl_dinode);
962 }
963
964 ret = ocfs2_recover_orphans(osb, item->lri_slot);
965 if (ret < 0)
966 mlog_errno(ret);
967
968 kfree(item);
969 }
970
971 mlog(0, "Recovery completion\n");
972 mlog_exit_void();
973}
974
975/* NOTE: This function always eats your references to la_dinode and
976 * tl_dinode, either manually on error, or by passing them to
977 * ocfs2_complete_recovery */
978static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal,
979 int slot_num,
980 struct ocfs2_dinode *la_dinode,
981 struct ocfs2_dinode *tl_dinode)
982{
983 struct ocfs2_la_recovery_item *item;
984
Sunil Mushranafae00ab2006-04-12 14:37:00 -0700985 item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS);
Mark Fashehccd979b2005-12-15 14:31:24 -0800986 if (!item) {
987 /* Though we wish to avoid it, we are in fact safe in
988 * skipping local alloc cleanup as fsck.ocfs2 is more
989 * than capable of reclaiming unused space. */
990 if (la_dinode)
991 kfree(la_dinode);
992
993 if (tl_dinode)
994 kfree(tl_dinode);
995
996 mlog_errno(-ENOMEM);
997 return;
998 }
999
1000 INIT_LIST_HEAD(&item->lri_list);
1001 item->lri_la_dinode = la_dinode;
1002 item->lri_slot = slot_num;
1003 item->lri_tl_dinode = tl_dinode;
1004
1005 spin_lock(&journal->j_lock);
1006 list_add_tail(&item->lri_list, &journal->j_la_cleanups);
1007 queue_work(ocfs2_wq, &journal->j_recovery_work);
1008 spin_unlock(&journal->j_lock);
1009}
1010
1011/* Called by the mount code to queue recovery the last part of
1012 * recovery for it's own slot. */
1013void ocfs2_complete_mount_recovery(struct ocfs2_super *osb)
1014{
1015 struct ocfs2_journal *journal = osb->journal;
1016
1017 if (osb->dirty) {
1018 /* No need to queue up our truncate_log as regular
1019 * cleanup will catch that. */
1020 ocfs2_queue_recovery_completion(journal,
1021 osb->slot_num,
1022 osb->local_alloc_copy,
1023 NULL);
1024 ocfs2_schedule_truncate_log_flush(osb, 0);
1025
1026 osb->local_alloc_copy = NULL;
1027 osb->dirty = 0;
1028 }
1029}
1030
1031static int __ocfs2_recovery_thread(void *arg)
1032{
1033 int status, node_num;
1034 struct ocfs2_super *osb = arg;
1035
1036 mlog_entry_void();
1037
1038 status = ocfs2_wait_on_mount(osb);
1039 if (status < 0) {
1040 goto bail;
1041 }
1042
1043restart:
1044 status = ocfs2_super_lock(osb, 1);
1045 if (status < 0) {
1046 mlog_errno(status);
1047 goto bail;
1048 }
1049
1050 while(!ocfs2_node_map_is_empty(osb, &osb->recovery_map)) {
1051 node_num = ocfs2_node_map_first_set_bit(osb,
1052 &osb->recovery_map);
1053 if (node_num == O2NM_INVALID_NODE_NUM) {
1054 mlog(0, "Out of nodes to recover.\n");
1055 break;
1056 }
1057
1058 status = ocfs2_recover_node(osb, node_num);
1059 if (status < 0) {
1060 mlog(ML_ERROR,
1061 "Error %d recovering node %d on device (%u,%u)!\n",
1062 status, node_num,
1063 MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1064 mlog(ML_ERROR, "Volume requires unmount.\n");
1065 continue;
1066 }
1067
1068 ocfs2_recovery_map_clear(osb, node_num);
1069 }
1070 ocfs2_super_unlock(osb, 1);
1071
1072 /* We always run recovery on our own orphan dir - the dead
1073 * node(s) may have voted "no" on an inode delete earlier. A
1074 * revote is therefore required. */
1075 ocfs2_queue_recovery_completion(osb->journal, osb->slot_num, NULL,
1076 NULL);
1077
1078bail:
Arjan van de Venc74ec2f2006-01-13 21:54:23 -08001079 mutex_lock(&osb->recovery_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -08001080 if (!status &&
1081 !ocfs2_node_map_is_empty(osb, &osb->recovery_map)) {
Arjan van de Venc74ec2f2006-01-13 21:54:23 -08001082 mutex_unlock(&osb->recovery_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -08001083 goto restart;
1084 }
1085
1086 osb->recovery_thread_task = NULL;
1087 mb(); /* sync with ocfs2_recovery_thread_running */
1088 wake_up(&osb->recovery_event);
1089
Arjan van de Venc74ec2f2006-01-13 21:54:23 -08001090 mutex_unlock(&osb->recovery_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -08001091
1092 mlog_exit(status);
1093 /* no one is callint kthread_stop() for us so the kthread() api
1094 * requires that we call do_exit(). And it isn't exported, but
1095 * complete_and_exit() seems to be a minimal wrapper around it. */
1096 complete_and_exit(NULL, status);
1097 return status;
1098}
1099
1100void ocfs2_recovery_thread(struct ocfs2_super *osb, int node_num)
1101{
1102 mlog_entry("(node_num=%d, osb->node_num = %d)\n",
1103 node_num, osb->node_num);
1104
Arjan van de Venc74ec2f2006-01-13 21:54:23 -08001105 mutex_lock(&osb->recovery_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -08001106 if (osb->disable_recovery)
1107 goto out;
1108
1109 /* People waiting on recovery will wait on
1110 * the recovery map to empty. */
1111 if (!ocfs2_recovery_map_set(osb, node_num))
1112 mlog(0, "node %d already be in recovery.\n", node_num);
1113
1114 mlog(0, "starting recovery thread...\n");
1115
1116 if (osb->recovery_thread_task)
1117 goto out;
1118
1119 osb->recovery_thread_task = kthread_run(__ocfs2_recovery_thread, osb,
Mark Fasheh78427042006-05-04 12:03:26 -07001120 "ocfs2rec");
Mark Fashehccd979b2005-12-15 14:31:24 -08001121 if (IS_ERR(osb->recovery_thread_task)) {
1122 mlog_errno((int)PTR_ERR(osb->recovery_thread_task));
1123 osb->recovery_thread_task = NULL;
1124 }
1125
1126out:
Arjan van de Venc74ec2f2006-01-13 21:54:23 -08001127 mutex_unlock(&osb->recovery_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -08001128 wake_up(&osb->recovery_event);
1129
1130 mlog_exit_void();
1131}
1132
1133/* Does the actual journal replay and marks the journal inode as
1134 * clean. Will only replay if the journal inode is marked dirty. */
1135static int ocfs2_replay_journal(struct ocfs2_super *osb,
1136 int node_num,
1137 int slot_num)
1138{
1139 int status;
1140 int got_lock = 0;
1141 unsigned int flags;
1142 struct inode *inode = NULL;
1143 struct ocfs2_dinode *fe;
1144 journal_t *journal = NULL;
1145 struct buffer_head *bh = NULL;
1146
1147 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
1148 slot_num);
1149 if (inode == NULL) {
1150 status = -EACCES;
1151 mlog_errno(status);
1152 goto done;
1153 }
1154 if (is_bad_inode(inode)) {
1155 status = -EACCES;
1156 iput(inode);
1157 inode = NULL;
1158 mlog_errno(status);
1159 goto done;
1160 }
1161 SET_INODE_JOURNAL(inode);
1162
1163 status = ocfs2_meta_lock_full(inode, NULL, &bh, 1,
1164 OCFS2_META_LOCK_RECOVERY);
1165 if (status < 0) {
1166 mlog(0, "status returned from ocfs2_meta_lock=%d\n", status);
1167 if (status != -ERESTARTSYS)
1168 mlog(ML_ERROR, "Could not lock journal!\n");
1169 goto done;
1170 }
1171 got_lock = 1;
1172
1173 fe = (struct ocfs2_dinode *) bh->b_data;
1174
1175 flags = le32_to_cpu(fe->id1.journal1.ij_flags);
1176
1177 if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) {
1178 mlog(0, "No recovery required for node %d\n", node_num);
1179 goto done;
1180 }
1181
1182 mlog(ML_NOTICE, "Recovering node %d from slot %d on device (%u,%u)\n",
1183 node_num, slot_num,
1184 MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1185
1186 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
1187
1188 status = ocfs2_force_read_journal(inode);
1189 if (status < 0) {
1190 mlog_errno(status);
1191 goto done;
1192 }
1193
1194 mlog(0, "calling journal_init_inode\n");
1195 journal = journal_init_inode(inode);
1196 if (journal == NULL) {
1197 mlog(ML_ERROR, "Linux journal layer error\n");
1198 status = -EIO;
1199 goto done;
1200 }
1201
1202 status = journal_load(journal);
1203 if (status < 0) {
1204 mlog_errno(status);
1205 if (!igrab(inode))
1206 BUG();
1207 journal_destroy(journal);
1208 goto done;
1209 }
1210
1211 ocfs2_clear_journal_error(osb->sb, journal, slot_num);
1212
1213 /* wipe the journal */
1214 mlog(0, "flushing the journal.\n");
1215 journal_lock_updates(journal);
1216 status = journal_flush(journal);
1217 journal_unlock_updates(journal);
1218 if (status < 0)
1219 mlog_errno(status);
1220
1221 /* This will mark the node clean */
1222 flags = le32_to_cpu(fe->id1.journal1.ij_flags);
1223 flags &= ~OCFS2_JOURNAL_DIRTY_FL;
1224 fe->id1.journal1.ij_flags = cpu_to_le32(flags);
1225
1226 status = ocfs2_write_block(osb, bh, inode);
1227 if (status < 0)
1228 mlog_errno(status);
1229
1230 if (!igrab(inode))
1231 BUG();
1232
1233 journal_destroy(journal);
1234
1235done:
1236 /* drop the lock on this nodes journal */
1237 if (got_lock)
1238 ocfs2_meta_unlock(inode, 1);
1239
1240 if (inode)
1241 iput(inode);
1242
1243 if (bh)
1244 brelse(bh);
1245
1246 mlog_exit(status);
1247 return status;
1248}
1249
1250/*
1251 * Do the most important parts of node recovery:
1252 * - Replay it's journal
1253 * - Stamp a clean local allocator file
1254 * - Stamp a clean truncate log
1255 * - Mark the node clean
1256 *
1257 * If this function completes without error, a node in OCFS2 can be
1258 * said to have been safely recovered. As a result, failure during the
1259 * second part of a nodes recovery process (local alloc recovery) is
1260 * far less concerning.
1261 */
1262static int ocfs2_recover_node(struct ocfs2_super *osb,
1263 int node_num)
1264{
1265 int status = 0;
1266 int slot_num;
1267 struct ocfs2_slot_info *si = osb->slot_info;
1268 struct ocfs2_dinode *la_copy = NULL;
1269 struct ocfs2_dinode *tl_copy = NULL;
1270
1271 mlog_entry("(node_num=%d, osb->node_num = %d)\n",
1272 node_num, osb->node_num);
1273
1274 mlog(0, "checking node %d\n", node_num);
1275
1276 /* Should not ever be called to recover ourselves -- in that
1277 * case we should've called ocfs2_journal_load instead. */
Eric Sesterhenn / snakebyteebdec832006-01-27 10:32:52 +01001278 BUG_ON(osb->node_num == node_num);
Mark Fashehccd979b2005-12-15 14:31:24 -08001279
1280 slot_num = ocfs2_node_num_to_slot(si, node_num);
1281 if (slot_num == OCFS2_INVALID_SLOT) {
1282 status = 0;
1283 mlog(0, "no slot for this node, so no recovery required.\n");
1284 goto done;
1285 }
1286
1287 mlog(0, "node %d was using slot %d\n", node_num, slot_num);
1288
1289 status = ocfs2_replay_journal(osb, node_num, slot_num);
1290 if (status < 0) {
1291 mlog_errno(status);
1292 goto done;
1293 }
1294
1295 /* Stamp a clean local alloc file AFTER recovering the journal... */
1296 status = ocfs2_begin_local_alloc_recovery(osb, slot_num, &la_copy);
1297 if (status < 0) {
1298 mlog_errno(status);
1299 goto done;
1300 }
1301
1302 /* An error from begin_truncate_log_recovery is not
1303 * serious enough to warrant halting the rest of
1304 * recovery. */
1305 status = ocfs2_begin_truncate_log_recovery(osb, slot_num, &tl_copy);
1306 if (status < 0)
1307 mlog_errno(status);
1308
1309 /* Likewise, this would be a strange but ultimately not so
1310 * harmful place to get an error... */
1311 ocfs2_clear_slot(si, slot_num);
1312 status = ocfs2_update_disk_slots(osb, si);
1313 if (status < 0)
1314 mlog_errno(status);
1315
1316 /* This will kfree the memory pointed to by la_copy and tl_copy */
1317 ocfs2_queue_recovery_completion(osb->journal, slot_num, la_copy,
1318 tl_copy);
1319
1320 status = 0;
1321done:
1322
1323 mlog_exit(status);
1324 return status;
1325}
1326
1327/* Test node liveness by trylocking his journal. If we get the lock,
1328 * we drop it here. Return 0 if we got the lock, -EAGAIN if node is
1329 * still alive (we couldn't get the lock) and < 0 on error. */
1330static int ocfs2_trylock_journal(struct ocfs2_super *osb,
1331 int slot_num)
1332{
1333 int status, flags;
1334 struct inode *inode = NULL;
1335
1336 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
1337 slot_num);
1338 if (inode == NULL) {
1339 mlog(ML_ERROR, "access error\n");
1340 status = -EACCES;
1341 goto bail;
1342 }
1343 if (is_bad_inode(inode)) {
1344 mlog(ML_ERROR, "access error (bad inode)\n");
1345 iput(inode);
1346 inode = NULL;
1347 status = -EACCES;
1348 goto bail;
1349 }
1350 SET_INODE_JOURNAL(inode);
1351
1352 flags = OCFS2_META_LOCK_RECOVERY | OCFS2_META_LOCK_NOQUEUE;
1353 status = ocfs2_meta_lock_full(inode, NULL, NULL, 1, flags);
1354 if (status < 0) {
1355 if (status != -EAGAIN)
1356 mlog_errno(status);
1357 goto bail;
1358 }
1359
1360 ocfs2_meta_unlock(inode, 1);
1361bail:
1362 if (inode)
1363 iput(inode);
1364
1365 return status;
1366}
1367
1368/* Call this underneath ocfs2_super_lock. It also assumes that the
1369 * slot info struct has been updated from disk. */
1370int ocfs2_mark_dead_nodes(struct ocfs2_super *osb)
1371{
1372 int status, i, node_num;
1373 struct ocfs2_slot_info *si = osb->slot_info;
1374
1375 /* This is called with the super block cluster lock, so we
1376 * know that the slot map can't change underneath us. */
1377
1378 spin_lock(&si->si_lock);
1379 for(i = 0; i < si->si_num_slots; i++) {
1380 if (i == osb->slot_num)
1381 continue;
1382 if (ocfs2_is_empty_slot(si, i))
1383 continue;
1384
1385 node_num = si->si_global_node_nums[i];
1386 if (ocfs2_node_map_test_bit(osb, &osb->recovery_map, node_num))
1387 continue;
1388 spin_unlock(&si->si_lock);
1389
1390 /* Ok, we have a slot occupied by another node which
1391 * is not in the recovery map. We trylock his journal
1392 * file here to test if he's alive. */
1393 status = ocfs2_trylock_journal(osb, i);
1394 if (!status) {
1395 /* Since we're called from mount, we know that
1396 * the recovery thread can't race us on
1397 * setting / checking the recovery bits. */
1398 ocfs2_recovery_thread(osb, node_num);
1399 } else if ((status < 0) && (status != -EAGAIN)) {
1400 mlog_errno(status);
1401 goto bail;
1402 }
1403
1404 spin_lock(&si->si_lock);
1405 }
1406 spin_unlock(&si->si_lock);
1407
1408 status = 0;
1409bail:
1410 mlog_exit(status);
1411 return status;
1412}
1413
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001414static int ocfs2_queue_orphans(struct ocfs2_super *osb,
1415 int slot,
1416 struct inode **head)
Mark Fashehccd979b2005-12-15 14:31:24 -08001417{
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001418 int status;
Mark Fashehccd979b2005-12-15 14:31:24 -08001419 struct inode *orphan_dir_inode = NULL;
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001420 struct inode *iter;
Mark Fashehccd979b2005-12-15 14:31:24 -08001421 unsigned long offset, blk, local;
1422 struct buffer_head *bh = NULL;
1423 struct ocfs2_dir_entry *de;
1424 struct super_block *sb = osb->sb;
Mark Fashehccd979b2005-12-15 14:31:24 -08001425
1426 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
1427 ORPHAN_DIR_SYSTEM_INODE,
1428 slot);
1429 if (!orphan_dir_inode) {
1430 status = -ENOENT;
1431 mlog_errno(status);
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001432 return status;
1433 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001434
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001435 mutex_lock(&orphan_dir_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -08001436 status = ocfs2_meta_lock(orphan_dir_inode, NULL, NULL, 0);
1437 if (status < 0) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001438 mlog_errno(status);
1439 goto out;
1440 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001441
1442 offset = 0;
1443 iter = NULL;
1444 while(offset < i_size_read(orphan_dir_inode)) {
1445 blk = offset >> sb->s_blocksize_bits;
1446
1447 bh = ocfs2_bread(orphan_dir_inode, blk, &status, 0);
1448 if (!bh)
1449 status = -EINVAL;
1450 if (status < 0) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001451 if (bh)
1452 brelse(bh);
1453 mlog_errno(status);
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001454 goto out_unlock;
Mark Fashehccd979b2005-12-15 14:31:24 -08001455 }
1456
1457 local = 0;
1458 while(offset < i_size_read(orphan_dir_inode)
1459 && local < sb->s_blocksize) {
1460 de = (struct ocfs2_dir_entry *) (bh->b_data + local);
1461
1462 if (!ocfs2_check_dir_entry(orphan_dir_inode,
1463 de, bh, local)) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001464 status = -EINVAL;
1465 mlog_errno(status);
1466 brelse(bh);
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001467 goto out_unlock;
Mark Fashehccd979b2005-12-15 14:31:24 -08001468 }
1469
1470 local += le16_to_cpu(de->rec_len);
1471 offset += le16_to_cpu(de->rec_len);
1472
1473 /* I guess we silently fail on no inode? */
1474 if (!le64_to_cpu(de->inode))
1475 continue;
1476 if (de->file_type > OCFS2_FT_MAX) {
1477 mlog(ML_ERROR,
1478 "block %llu contains invalid de: "
Mark Fashehb0697052006-03-03 10:24:33 -08001479 "inode = %llu, rec_len = %u, "
Mark Fashehccd979b2005-12-15 14:31:24 -08001480 "name_len = %u, file_type = %u, "
1481 "name='%.*s'\n",
1482 (unsigned long long)bh->b_blocknr,
Mark Fashehb0697052006-03-03 10:24:33 -08001483 (unsigned long long)le64_to_cpu(de->inode),
Mark Fashehccd979b2005-12-15 14:31:24 -08001484 le16_to_cpu(de->rec_len),
1485 de->name_len,
1486 de->file_type,
1487 de->name_len,
1488 de->name);
1489 continue;
1490 }
1491 if (de->name_len == 1 && !strncmp(".", de->name, 1))
1492 continue;
1493 if (de->name_len == 2 && !strncmp("..", de->name, 2))
1494 continue;
1495
1496 iter = ocfs2_iget(osb, le64_to_cpu(de->inode));
1497 if (IS_ERR(iter))
1498 continue;
1499
Mark Fashehb0697052006-03-03 10:24:33 -08001500 mlog(0, "queue orphan %llu\n",
1501 (unsigned long long)OCFS2_I(iter)->ip_blkno);
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001502 /* No locking is required for the next_orphan
1503 * queue as there is only ever a single
1504 * process doing orphan recovery. */
1505 OCFS2_I(iter)->ip_next_orphan = *head;
1506 *head = iter;
Mark Fashehccd979b2005-12-15 14:31:24 -08001507 }
1508 brelse(bh);
1509 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001510
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001511out_unlock:
Mark Fashehccd979b2005-12-15 14:31:24 -08001512 ocfs2_meta_unlock(orphan_dir_inode, 0);
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001513out:
1514 mutex_unlock(&orphan_dir_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -08001515 iput(orphan_dir_inode);
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001516 return status;
1517}
1518
1519static int ocfs2_orphan_recovery_can_continue(struct ocfs2_super *osb,
1520 int slot)
1521{
1522 int ret;
1523
1524 spin_lock(&osb->osb_lock);
1525 ret = !osb->osb_orphan_wipes[slot];
1526 spin_unlock(&osb->osb_lock);
1527 return ret;
1528}
1529
1530static void ocfs2_mark_recovering_orphan_dir(struct ocfs2_super *osb,
1531 int slot)
1532{
1533 spin_lock(&osb->osb_lock);
1534 /* Mark ourselves such that new processes in delete_inode()
1535 * know to quit early. */
1536 ocfs2_node_map_set_bit(osb, &osb->osb_recovering_orphan_dirs, slot);
1537 while (osb->osb_orphan_wipes[slot]) {
1538 /* If any processes are already in the middle of an
1539 * orphan wipe on this dir, then we need to wait for
1540 * them. */
1541 spin_unlock(&osb->osb_lock);
1542 wait_event_interruptible(osb->osb_wipe_event,
1543 ocfs2_orphan_recovery_can_continue(osb, slot));
1544 spin_lock(&osb->osb_lock);
1545 }
1546 spin_unlock(&osb->osb_lock);
1547}
1548
1549static void ocfs2_clear_recovering_orphan_dir(struct ocfs2_super *osb,
1550 int slot)
1551{
1552 ocfs2_node_map_clear_bit(osb, &osb->osb_recovering_orphan_dirs, slot);
1553}
1554
1555/*
1556 * Orphan recovery. Each mounted node has it's own orphan dir which we
1557 * must run during recovery. Our strategy here is to build a list of
1558 * the inodes in the orphan dir and iget/iput them. The VFS does
1559 * (most) of the rest of the work.
1560 *
1561 * Orphan recovery can happen at any time, not just mount so we have a
1562 * couple of extra considerations.
1563 *
1564 * - We grab as many inodes as we can under the orphan dir lock -
1565 * doing iget() outside the orphan dir risks getting a reference on
1566 * an invalid inode.
1567 * - We must be sure not to deadlock with other processes on the
1568 * system wanting to run delete_inode(). This can happen when they go
1569 * to lock the orphan dir and the orphan recovery process attempts to
1570 * iget() inside the orphan dir lock. This can be avoided by
1571 * advertising our state to ocfs2_delete_inode().
1572 */
1573static int ocfs2_recover_orphans(struct ocfs2_super *osb,
1574 int slot)
1575{
1576 int ret = 0;
1577 struct inode *inode = NULL;
1578 struct inode *iter;
1579 struct ocfs2_inode_info *oi;
1580
1581 mlog(0, "Recover inodes from orphan dir in slot %d\n", slot);
1582
1583 ocfs2_mark_recovering_orphan_dir(osb, slot);
1584 ret = ocfs2_queue_orphans(osb, slot, &inode);
1585 ocfs2_clear_recovering_orphan_dir(osb, slot);
1586
1587 /* Error here should be noted, but we want to continue with as
1588 * many queued inodes as we've got. */
1589 if (ret)
1590 mlog_errno(ret);
Mark Fashehccd979b2005-12-15 14:31:24 -08001591
1592 while (inode) {
1593 oi = OCFS2_I(inode);
Mark Fashehb0697052006-03-03 10:24:33 -08001594 mlog(0, "iput orphan %llu\n", (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001595
1596 iter = oi->ip_next_orphan;
1597
1598 spin_lock(&oi->ip_lock);
1599 /* Delete voting may have set these on the assumption
1600 * that the other node would wipe them successfully.
1601 * If they are still in the node's orphan dir, we need
1602 * to reset that state. */
1603 oi->ip_flags &= ~(OCFS2_INODE_DELETED|OCFS2_INODE_SKIP_DELETE);
1604
1605 /* Set the proper information to get us going into
1606 * ocfs2_delete_inode. */
1607 oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED;
1608 oi->ip_orphaned_slot = slot;
1609 spin_unlock(&oi->ip_lock);
1610
1611 iput(inode);
1612
1613 inode = iter;
1614 }
1615
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001616 return ret;
Mark Fashehccd979b2005-12-15 14:31:24 -08001617}
1618
1619static int ocfs2_wait_on_mount(struct ocfs2_super *osb)
1620{
1621 /* This check is good because ocfs2 will wait on our recovery
1622 * thread before changing it to something other than MOUNTED
1623 * or DISABLED. */
1624 wait_event(osb->osb_mount_event,
1625 atomic_read(&osb->vol_state) == VOLUME_MOUNTED ||
1626 atomic_read(&osb->vol_state) == VOLUME_DISABLED);
1627
1628 /* If there's an error on mount, then we may never get to the
1629 * MOUNTED flag, but this is set right before
1630 * dismount_volume() so we can trust it. */
1631 if (atomic_read(&osb->vol_state) == VOLUME_DISABLED) {
1632 mlog(0, "mount error, exiting!\n");
1633 return -EBUSY;
1634 }
1635
1636 return 0;
1637}
1638
1639static int ocfs2_commit_thread(void *arg)
1640{
1641 int status;
1642 struct ocfs2_super *osb = arg;
1643 struct ocfs2_journal *journal = osb->journal;
1644
1645 /* we can trust j_num_trans here because _should_stop() is only set in
1646 * shutdown and nobody other than ourselves should be able to start
1647 * transactions. committing on shutdown might take a few iterations
1648 * as final transactions put deleted inodes on the list */
1649 while (!(kthread_should_stop() &&
1650 atomic_read(&journal->j_num_trans) == 0)) {
1651
Mark Fasheh745ae8ba2006-02-09 13:23:39 -08001652 wait_event_interruptible(osb->checkpoint_event,
1653 atomic_read(&journal->j_num_trans)
1654 || kthread_should_stop());
Mark Fashehccd979b2005-12-15 14:31:24 -08001655
1656 status = ocfs2_commit_cache(osb);
1657 if (status < 0)
1658 mlog_errno(status);
1659
1660 if (kthread_should_stop() && atomic_read(&journal->j_num_trans)){
1661 mlog(ML_KTHREAD,
1662 "commit_thread: %u transactions pending on "
1663 "shutdown\n",
1664 atomic_read(&journal->j_num_trans));
1665 }
1666 }
1667
1668 return 0;
1669}
1670
1671/* Look for a dirty journal without taking any cluster locks. Used for
1672 * hard readonly access to determine whether the file system journals
1673 * require recovery. */
1674int ocfs2_check_journals_nolocks(struct ocfs2_super *osb)
1675{
1676 int ret = 0;
1677 unsigned int slot;
1678 struct buffer_head *di_bh;
1679 struct ocfs2_dinode *di;
1680 struct inode *journal = NULL;
1681
1682 for(slot = 0; slot < osb->max_slots; slot++) {
1683 journal = ocfs2_get_system_file_inode(osb,
1684 JOURNAL_SYSTEM_INODE,
1685 slot);
1686 if (!journal || is_bad_inode(journal)) {
1687 ret = -EACCES;
1688 mlog_errno(ret);
1689 goto out;
1690 }
1691
1692 di_bh = NULL;
1693 ret = ocfs2_read_block(osb, OCFS2_I(journal)->ip_blkno, &di_bh,
1694 0, journal);
1695 if (ret < 0) {
1696 mlog_errno(ret);
1697 goto out;
1698 }
1699
1700 di = (struct ocfs2_dinode *) di_bh->b_data;
1701
1702 if (le32_to_cpu(di->id1.journal1.ij_flags) &
1703 OCFS2_JOURNAL_DIRTY_FL)
1704 ret = -EROFS;
1705
1706 brelse(di_bh);
1707 if (ret)
1708 break;
1709 }
1710
1711out:
1712 if (journal)
1713 iput(journal);
1714
1715 return ret;
1716}