blob: eebc3cfa6be8c89dd5a55a8910ad4b1f3c9d2827 [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
52spinlock_t trans_inc_lock = SPIN_LOCK_UNLOCKED;
53
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;
225 list_del(&(OCFS2_I(inode)->ip_handle_list));
226 list_add_tail(&(OCFS2_I(inode)->ip_handle_list), &(handle->inode_list));
227}
228
229static void ocfs2_handle_unlock_inodes(struct ocfs2_journal_handle *handle)
230{
231 struct list_head *p, *n;
232 struct inode *inode;
233 struct ocfs2_inode_info *oi;
234
235 list_for_each_safe(p, n, &handle->inode_list) {
236 oi = list_entry(p, struct ocfs2_inode_info,
237 ip_handle_list);
238 inode = &oi->vfs_inode;
239
240 OCFS2_I(inode)->ip_handle = NULL;
241 list_del_init(&OCFS2_I(inode)->ip_handle_list);
242
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800243 mutex_unlock(&inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800244 iput(inode);
245 }
246}
247
248/* This is trivial so we do it out of the main commit
249 * paths. Beware, it can be called from start_trans too! */
250static void ocfs2_commit_unstarted_handle(struct ocfs2_journal_handle *handle)
251{
252 mlog_entry_void();
253
254 BUG_ON(handle->flags & OCFS2_HANDLE_STARTED);
255
256 ocfs2_handle_unlock_inodes(handle);
257 /* You are allowed to add journal locks before the transaction
258 * has started. */
259 ocfs2_handle_cleanup_locks(handle->journal, handle);
260
261 kfree(handle);
262
263 mlog_exit_void();
264}
265
266void ocfs2_commit_trans(struct ocfs2_journal_handle *handle)
267{
268 handle_t *jbd_handle;
269 int retval;
270 struct ocfs2_journal *journal = handle->journal;
271
272 mlog_entry_void();
273
274 BUG_ON(!handle);
275
276 if (!(handle->flags & OCFS2_HANDLE_STARTED)) {
277 ocfs2_commit_unstarted_handle(handle);
278 mlog_exit_void();
279 return;
280 }
281
282 /* release inode semaphores we took during this transaction */
283 ocfs2_handle_unlock_inodes(handle);
284
285 /* ocfs2_extend_trans may have had to call journal_restart
286 * which will always commit the transaction, but may return
287 * error for any number of reasons. If this is the case, we
288 * clear k_handle as it's not valid any more. */
289 if (handle->k_handle) {
290 jbd_handle = handle->k_handle;
291
292 if (handle->flags & OCFS2_HANDLE_SYNC)
293 jbd_handle->h_sync = 1;
294 else
295 jbd_handle->h_sync = 0;
296
297 /* actually stop the transaction. if we've set h_sync,
298 * it'll have been committed when we return */
299 retval = journal_stop(jbd_handle);
300 if (retval < 0) {
301 mlog_errno(retval);
302 mlog(ML_ERROR, "Could not commit transaction\n");
303 BUG();
304 }
305
306 handle->k_handle = NULL; /* it's been free'd in journal_stop */
307 }
308
309 ocfs2_handle_cleanup_locks(journal, handle);
310
311 up_read(&journal->j_trans_barrier);
312
313 kfree(handle);
314 mlog_exit_void();
315}
316
317/*
318 * 'nblocks' is what you want to add to the current
319 * transaction. extend_trans will either extend the current handle by
320 * nblocks, or commit it and start a new one with nblocks credits.
321 *
322 * WARNING: This will not release any semaphores or disk locks taken
323 * during the transaction, so make sure they were taken *before*
324 * start_trans or we'll have ordering deadlocks.
325 *
326 * WARNING2: Note that we do *not* drop j_trans_barrier here. This is
327 * good because transaction ids haven't yet been recorded on the
328 * cluster locks associated with this handle.
329 */
330int ocfs2_extend_trans(struct ocfs2_journal_handle *handle,
331 int nblocks)
332{
333 int status;
334
335 BUG_ON(!handle);
336 BUG_ON(!(handle->flags & OCFS2_HANDLE_STARTED));
337 BUG_ON(!nblocks);
338
339 mlog_entry_void();
340
341 mlog(0, "Trying to extend transaction by %d blocks\n", nblocks);
342
343 status = journal_extend(handle->k_handle, nblocks);
344 if (status < 0) {
345 mlog_errno(status);
346 goto bail;
347 }
348
349 if (status > 0) {
350 mlog(0, "journal_extend failed, trying journal_restart\n");
351 status = journal_restart(handle->k_handle, nblocks);
352 if (status < 0) {
353 handle->k_handle = NULL;
354 mlog_errno(status);
355 goto bail;
356 }
357 handle->max_buffs = nblocks;
358 } else
359 handle->max_buffs += nblocks;
360
361 status = 0;
362bail:
363
364 mlog_exit(status);
365 return status;
366}
367
368int ocfs2_journal_access(struct ocfs2_journal_handle *handle,
369 struct inode *inode,
370 struct buffer_head *bh,
371 int type)
372{
373 int status;
374
375 BUG_ON(!inode);
376 BUG_ON(!handle);
377 BUG_ON(!bh);
378 BUG_ON(!(handle->flags & OCFS2_HANDLE_STARTED));
379
Badari Pulavarty205f87f2006-03-26 01:38:00 -0800380 mlog_entry("bh->b_blocknr=%llu, type=%d (\"%s\"), bh->b_size = %zu\n",
Mark Fashehccd979b2005-12-15 14:31:24 -0800381 (unsigned long long)bh->b_blocknr, type,
382 (type == OCFS2_JOURNAL_ACCESS_CREATE) ?
383 "OCFS2_JOURNAL_ACCESS_CREATE" :
384 "OCFS2_JOURNAL_ACCESS_WRITE",
385 bh->b_size);
386
387 /* we can safely remove this assertion after testing. */
388 if (!buffer_uptodate(bh)) {
389 mlog(ML_ERROR, "giving me a buffer that's not uptodate!\n");
390 mlog(ML_ERROR, "b_blocknr=%llu\n",
391 (unsigned long long)bh->b_blocknr);
392 BUG();
393 }
394
395 /* Set the current transaction information on the inode so
396 * that the locking code knows whether it can drop it's locks
397 * on this inode or not. We're protected from the commit
398 * thread updating the current transaction id until
399 * ocfs2_commit_trans() because ocfs2_start_trans() took
400 * j_trans_barrier for us. */
401 ocfs2_set_inode_lock_trans(OCFS2_SB(inode->i_sb)->journal, inode);
402
Mark Fasheh251b6ec2006-01-10 15:41:43 -0800403 mutex_lock(&OCFS2_I(inode)->ip_io_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800404 switch (type) {
405 case OCFS2_JOURNAL_ACCESS_CREATE:
406 case OCFS2_JOURNAL_ACCESS_WRITE:
407 status = journal_get_write_access(handle->k_handle, bh);
408 break;
409
410 case OCFS2_JOURNAL_ACCESS_UNDO:
411 status = journal_get_undo_access(handle->k_handle, bh);
412 break;
413
414 default:
415 status = -EINVAL;
416 mlog(ML_ERROR, "Uknown access type!\n");
417 }
Mark Fasheh251b6ec2006-01-10 15:41:43 -0800418 mutex_unlock(&OCFS2_I(inode)->ip_io_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800419
420 if (status < 0)
421 mlog(ML_ERROR, "Error %d getting %d access to buffer!\n",
422 status, type);
423
424 mlog_exit(status);
425 return status;
426}
427
428int ocfs2_journal_dirty(struct ocfs2_journal_handle *handle,
429 struct buffer_head *bh)
430{
431 int status;
432
433 BUG_ON(!(handle->flags & OCFS2_HANDLE_STARTED));
434
435 mlog_entry("(bh->b_blocknr=%llu)\n",
436 (unsigned long long)bh->b_blocknr);
437
438 status = journal_dirty_metadata(handle->k_handle, bh);
439 if (status < 0)
440 mlog(ML_ERROR, "Could not dirty metadata buffer. "
441 "(bh->b_blocknr=%llu)\n",
442 (unsigned long long)bh->b_blocknr);
443
444 mlog_exit(status);
445 return status;
446}
447
448int ocfs2_journal_dirty_data(handle_t *handle,
449 struct buffer_head *bh)
450{
451 int err = journal_dirty_data(handle, bh);
452 if (err)
453 mlog_errno(err);
454 /* TODO: When we can handle it, abort the handle and go RO on
455 * error here. */
456
457 return err;
458}
459
460/* We always assume you're adding a metadata lock at level 'ex' */
461int ocfs2_handle_add_lock(struct ocfs2_journal_handle *handle,
462 struct inode *inode)
463{
464 int status;
465 struct ocfs2_journal_lock *lock;
466
467 BUG_ON(!inode);
468
469 lock = kmem_cache_alloc(ocfs2_lock_cache, GFP_NOFS);
470 if (!lock) {
471 status = -ENOMEM;
472 mlog_errno(-ENOMEM);
473 goto bail;
474 }
475
476 if (!igrab(inode))
477 BUG();
478 lock->jl_inode = inode;
479
480 list_add_tail(&(lock->jl_lock_list), &(handle->locks));
481 handle->num_locks++;
482
483 status = 0;
484bail:
485 mlog_exit(status);
486 return status;
487}
488
489static void ocfs2_handle_cleanup_locks(struct ocfs2_journal *journal,
490 struct ocfs2_journal_handle *handle)
491{
492 struct list_head *p, *n;
493 struct ocfs2_journal_lock *lock;
494 struct inode *inode;
495
496 list_for_each_safe(p, n, &(handle->locks)) {
497 lock = list_entry(p, struct ocfs2_journal_lock,
498 jl_lock_list);
499 list_del(&lock->jl_lock_list);
500 handle->num_locks--;
501
502 inode = lock->jl_inode;
503 ocfs2_meta_unlock(inode, 1);
504 if (atomic_read(&inode->i_count) == 1)
505 mlog(ML_ERROR,
Mark Fashehb0697052006-03-03 10:24:33 -0800506 "Inode %llu, I'm doing a last iput for!",
507 (unsigned long long)OCFS2_I(inode)->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800508 iput(inode);
509 kmem_cache_free(ocfs2_lock_cache, lock);
510 }
511}
512
513#define OCFS2_DEFAULT_COMMIT_INTERVAL (HZ * 5)
514
515void ocfs2_set_journal_params(struct ocfs2_super *osb)
516{
517 journal_t *journal = osb->journal->j_journal;
518
519 spin_lock(&journal->j_state_lock);
520 journal->j_commit_interval = OCFS2_DEFAULT_COMMIT_INTERVAL;
521 if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER)
522 journal->j_flags |= JFS_BARRIER;
523 else
524 journal->j_flags &= ~JFS_BARRIER;
525 spin_unlock(&journal->j_state_lock);
526}
527
528int ocfs2_journal_init(struct ocfs2_journal *journal, int *dirty)
529{
530 int status = -1;
531 struct inode *inode = NULL; /* the journal inode */
532 journal_t *j_journal = NULL;
533 struct ocfs2_dinode *di = NULL;
534 struct buffer_head *bh = NULL;
535 struct ocfs2_super *osb;
536 int meta_lock = 0;
537
538 mlog_entry_void();
539
540 BUG_ON(!journal);
541
542 osb = journal->j_osb;
543
544 /* already have the inode for our journal */
545 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
546 osb->slot_num);
547 if (inode == NULL) {
548 status = -EACCES;
549 mlog_errno(status);
550 goto done;
551 }
552 if (is_bad_inode(inode)) {
553 mlog(ML_ERROR, "access error (bad inode)\n");
554 iput(inode);
555 inode = NULL;
556 status = -EACCES;
557 goto done;
558 }
559
560 SET_INODE_JOURNAL(inode);
561 OCFS2_I(inode)->ip_open_count++;
562
Mark Fasheh6eff5792006-01-18 10:31:47 -0800563 /* Skip recovery waits here - journal inode metadata never
564 * changes in a live cluster so it can be considered an
565 * exception to the rule. */
566 status = ocfs2_meta_lock_full(inode, NULL, &bh, 1,
567 OCFS2_META_LOCK_RECOVERY);
Mark Fashehccd979b2005-12-15 14:31:24 -0800568 if (status < 0) {
569 if (status != -ERESTARTSYS)
570 mlog(ML_ERROR, "Could not get lock on journal!\n");
571 goto done;
572 }
573
574 meta_lock = 1;
575 di = (struct ocfs2_dinode *)bh->b_data;
576
577 if (inode->i_size < OCFS2_MIN_JOURNAL_SIZE) {
578 mlog(ML_ERROR, "Journal file size (%lld) is too small!\n",
579 inode->i_size);
580 status = -EINVAL;
581 goto done;
582 }
583
584 mlog(0, "inode->i_size = %lld\n", inode->i_size);
Andrew Morton5515eff2006-03-26 01:37:53 -0800585 mlog(0, "inode->i_blocks = %llu\n",
586 (unsigned long long)inode->i_blocks);
Mark Fashehccd979b2005-12-15 14:31:24 -0800587 mlog(0, "inode->ip_clusters = %u\n", OCFS2_I(inode)->ip_clusters);
588
589 /* call the kernels journal init function now */
590 j_journal = journal_init_inode(inode);
591 if (j_journal == NULL) {
592 mlog(ML_ERROR, "Linux journal layer error\n");
593 status = -EINVAL;
594 goto done;
595 }
596
597 mlog(0, "Returned from journal_init_inode\n");
598 mlog(0, "j_journal->j_maxlen = %u\n", j_journal->j_maxlen);
599
600 *dirty = (le32_to_cpu(di->id1.journal1.ij_flags) &
601 OCFS2_JOURNAL_DIRTY_FL);
602
603 journal->j_journal = j_journal;
604 journal->j_inode = inode;
605 journal->j_bh = bh;
606
607 ocfs2_set_journal_params(osb);
608
609 journal->j_state = OCFS2_JOURNAL_LOADED;
610
611 status = 0;
612done:
613 if (status < 0) {
614 if (meta_lock)
615 ocfs2_meta_unlock(inode, 1);
616 if (bh != NULL)
617 brelse(bh);
618 if (inode) {
619 OCFS2_I(inode)->ip_open_count--;
620 iput(inode);
621 }
622 }
623
624 mlog_exit(status);
625 return status;
626}
627
628static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
629 int dirty)
630{
631 int status;
632 unsigned int flags;
633 struct ocfs2_journal *journal = osb->journal;
634 struct buffer_head *bh = journal->j_bh;
635 struct ocfs2_dinode *fe;
636
637 mlog_entry_void();
638
639 fe = (struct ocfs2_dinode *)bh->b_data;
640 if (!OCFS2_IS_VALID_DINODE(fe)) {
641 /* This is called from startup/shutdown which will
642 * handle the errors in a specific manner, so no need
643 * to call ocfs2_error() here. */
Mark Fashehb0697052006-03-03 10:24:33 -0800644 mlog(ML_ERROR, "Journal dinode %llu has invalid "
645 "signature: %.*s", (unsigned long long)fe->i_blkno, 7,
646 fe->i_signature);
Mark Fashehccd979b2005-12-15 14:31:24 -0800647 status = -EIO;
648 goto out;
649 }
650
651 flags = le32_to_cpu(fe->id1.journal1.ij_flags);
652 if (dirty)
653 flags |= OCFS2_JOURNAL_DIRTY_FL;
654 else
655 flags &= ~OCFS2_JOURNAL_DIRTY_FL;
656 fe->id1.journal1.ij_flags = cpu_to_le32(flags);
657
658 status = ocfs2_write_block(osb, bh, journal->j_inode);
659 if (status < 0)
660 mlog_errno(status);
661
662out:
663 mlog_exit(status);
664 return status;
665}
666
667/*
668 * If the journal has been kmalloc'd it needs to be freed after this
669 * call.
670 */
671void ocfs2_journal_shutdown(struct ocfs2_super *osb)
672{
673 struct ocfs2_journal *journal = NULL;
674 int status = 0;
675 struct inode *inode = NULL;
676 int num_running_trans = 0;
677
678 mlog_entry_void();
679
Eric Sesterhenn / snakebyteebdec832006-01-27 10:32:52 +0100680 BUG_ON(!osb);
Mark Fashehccd979b2005-12-15 14:31:24 -0800681
682 journal = osb->journal;
683 if (!journal)
684 goto done;
685
686 inode = journal->j_inode;
687
688 if (journal->j_state != OCFS2_JOURNAL_LOADED)
689 goto done;
690
691 /* need to inc inode use count as journal_destroy will iput. */
692 if (!igrab(inode))
693 BUG();
694
695 num_running_trans = atomic_read(&(osb->journal->j_num_trans));
696 if (num_running_trans > 0)
697 mlog(0, "Shutting down journal: must wait on %d "
698 "running transactions!\n",
699 num_running_trans);
700
701 /* Do a commit_cache here. It will flush our journal, *and*
702 * release any locks that are still held.
703 * set the SHUTDOWN flag and release the trans lock.
704 * the commit thread will take the trans lock for us below. */
705 journal->j_state = OCFS2_JOURNAL_IN_SHUTDOWN;
706
707 /* The OCFS2_JOURNAL_IN_SHUTDOWN will signal to commit_cache to not
708 * drop the trans_lock (which we want to hold until we
709 * completely destroy the journal. */
710 if (osb->commit_task) {
711 /* Wait for the commit thread */
712 mlog(0, "Waiting for ocfs2commit to exit....\n");
713 kthread_stop(osb->commit_task);
714 osb->commit_task = NULL;
715 }
716
717 BUG_ON(atomic_read(&(osb->journal->j_num_trans)) != 0);
718
719 status = ocfs2_journal_toggle_dirty(osb, 0);
720 if (status < 0)
721 mlog_errno(status);
722
723 /* Shutdown the kernel journal system */
724 journal_destroy(journal->j_journal);
725
726 OCFS2_I(inode)->ip_open_count--;
727
728 /* unlock our journal */
729 ocfs2_meta_unlock(inode, 1);
730
731 brelse(journal->j_bh);
732 journal->j_bh = NULL;
733
734 journal->j_state = OCFS2_JOURNAL_FREE;
735
736// up_write(&journal->j_trans_barrier);
737done:
738 if (inode)
739 iput(inode);
740 mlog_exit_void();
741}
742
743static void ocfs2_clear_journal_error(struct super_block *sb,
744 journal_t *journal,
745 int slot)
746{
747 int olderr;
748
749 olderr = journal_errno(journal);
750 if (olderr) {
751 mlog(ML_ERROR, "File system error %d recorded in "
752 "journal %u.\n", olderr, slot);
753 mlog(ML_ERROR, "File system on device %s needs checking.\n",
754 sb->s_id);
755
756 journal_ack_err(journal);
757 journal_clear_err(journal);
758 }
759}
760
761int ocfs2_journal_load(struct ocfs2_journal *journal)
762{
763 int status = 0;
764 struct ocfs2_super *osb;
765
766 mlog_entry_void();
767
768 if (!journal)
769 BUG();
770
771 osb = journal->j_osb;
772
773 status = journal_load(journal->j_journal);
774 if (status < 0) {
775 mlog(ML_ERROR, "Failed to load journal!\n");
776 goto done;
777 }
778
779 ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num);
780
781 status = ocfs2_journal_toggle_dirty(osb, 1);
782 if (status < 0) {
783 mlog_errno(status);
784 goto done;
785 }
786
787 /* Launch the commit thread */
788 osb->commit_task = kthread_run(ocfs2_commit_thread, osb, "ocfs2cmt-%d",
789 osb->osb_id);
790 if (IS_ERR(osb->commit_task)) {
791 status = PTR_ERR(osb->commit_task);
792 osb->commit_task = NULL;
793 mlog(ML_ERROR, "unable to launch ocfs2commit thread, error=%d",
794 status);
795 goto done;
796 }
797
798done:
799 mlog_exit(status);
800 return status;
801}
802
803
804/* 'full' flag tells us whether we clear out all blocks or if we just
805 * mark the journal clean */
806int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full)
807{
808 int status;
809
810 mlog_entry_void();
811
Eric Sesterhenn / snakebyteebdec832006-01-27 10:32:52 +0100812 BUG_ON(!journal);
Mark Fashehccd979b2005-12-15 14:31:24 -0800813
814 status = journal_wipe(journal->j_journal, full);
815 if (status < 0) {
816 mlog_errno(status);
817 goto bail;
818 }
819
820 status = ocfs2_journal_toggle_dirty(journal->j_osb, 0);
821 if (status < 0)
822 mlog_errno(status);
823
824bail:
825 mlog_exit(status);
826 return status;
827}
828
829/*
830 * JBD Might read a cached version of another nodes journal file. We
831 * don't want this as this file changes often and we get no
832 * notification on those changes. The only way to be sure that we've
833 * got the most up to date version of those blocks then is to force
834 * read them off disk. Just searching through the buffer cache won't
835 * work as there may be pages backing this file which are still marked
836 * up to date. We know things can't change on this file underneath us
837 * as we have the lock by now :)
838 */
839static int ocfs2_force_read_journal(struct inode *inode)
840{
841 int status = 0;
842 int i, p_blocks;
843 u64 v_blkno, p_blkno;
844#define CONCURRENT_JOURNAL_FILL 32
845 struct buffer_head *bhs[CONCURRENT_JOURNAL_FILL];
846
847 mlog_entry_void();
848
849 BUG_ON(inode->i_blocks !=
850 ocfs2_align_bytes_to_sectors(i_size_read(inode)));
851
852 memset(bhs, 0, sizeof(struct buffer_head *) * CONCURRENT_JOURNAL_FILL);
853
Andrew Morton5515eff2006-03-26 01:37:53 -0800854 mlog(0, "Force reading %llu blocks\n",
855 (unsigned long long)(inode->i_blocks >>
856 (inode->i_sb->s_blocksize_bits - 9)));
Mark Fashehccd979b2005-12-15 14:31:24 -0800857
858 v_blkno = 0;
859 while (v_blkno <
860 (inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9))) {
861
862 status = ocfs2_extent_map_get_blocks(inode, v_blkno,
863 1, &p_blkno,
864 &p_blocks);
865 if (status < 0) {
866 mlog_errno(status);
867 goto bail;
868 }
869
870 if (p_blocks > CONCURRENT_JOURNAL_FILL)
871 p_blocks = CONCURRENT_JOURNAL_FILL;
872
Mark Fashehdd4a2c22006-04-12 14:24:05 -0700873 /* We are reading journal data which should not
874 * be put in the uptodate cache */
Mark Fashehccd979b2005-12-15 14:31:24 -0800875 status = ocfs2_read_blocks(OCFS2_SB(inode->i_sb),
876 p_blkno, p_blocks, bhs, 0,
Mark Fashehdd4a2c22006-04-12 14:24:05 -0700877 NULL);
Mark Fashehccd979b2005-12-15 14:31:24 -0800878 if (status < 0) {
879 mlog_errno(status);
880 goto bail;
881 }
882
883 for(i = 0; i < p_blocks; i++) {
884 brelse(bhs[i]);
885 bhs[i] = NULL;
886 }
887
888 v_blkno += p_blocks;
889 }
890
891bail:
892 for(i = 0; i < CONCURRENT_JOURNAL_FILL; i++)
893 if (bhs[i])
894 brelse(bhs[i]);
895 mlog_exit(status);
896 return status;
897}
898
899struct ocfs2_la_recovery_item {
900 struct list_head lri_list;
901 int lri_slot;
902 struct ocfs2_dinode *lri_la_dinode;
903 struct ocfs2_dinode *lri_tl_dinode;
904};
905
906/* Does the second half of the recovery process. By this point, the
907 * node is marked clean and can actually be considered recovered,
908 * hence it's no longer in the recovery map, but there's still some
909 * cleanup we can do which shouldn't happen within the recovery thread
910 * as locking in that context becomes very difficult if we are to take
911 * recovering nodes into account.
912 *
913 * NOTE: This function can and will sleep on recovery of other nodes
914 * during cluster locking, just like any other ocfs2 process.
915 */
916void ocfs2_complete_recovery(void *data)
917{
918 int ret;
919 struct ocfs2_super *osb = data;
920 struct ocfs2_journal *journal = osb->journal;
921 struct ocfs2_dinode *la_dinode, *tl_dinode;
922 struct ocfs2_la_recovery_item *item;
923 struct list_head *p, *n;
924 LIST_HEAD(tmp_la_list);
925
926 mlog_entry_void();
927
928 mlog(0, "completing recovery from keventd\n");
929
930 spin_lock(&journal->j_lock);
931 list_splice_init(&journal->j_la_cleanups, &tmp_la_list);
932 spin_unlock(&journal->j_lock);
933
934 list_for_each_safe(p, n, &tmp_la_list) {
935 item = list_entry(p, struct ocfs2_la_recovery_item, lri_list);
936 list_del_init(&item->lri_list);
937
938 mlog(0, "Complete recovery for slot %d\n", item->lri_slot);
939
940 la_dinode = item->lri_la_dinode;
941 if (la_dinode) {
Mark Fashehb0697052006-03-03 10:24:33 -0800942 mlog(0, "Clean up local alloc %llu\n",
943 (unsigned long long)la_dinode->i_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800944
945 ret = ocfs2_complete_local_alloc_recovery(osb,
946 la_dinode);
947 if (ret < 0)
948 mlog_errno(ret);
949
950 kfree(la_dinode);
951 }
952
953 tl_dinode = item->lri_tl_dinode;
954 if (tl_dinode) {
Mark Fashehb0697052006-03-03 10:24:33 -0800955 mlog(0, "Clean up truncate log %llu\n",
956 (unsigned long long)tl_dinode->i_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800957
958 ret = ocfs2_complete_truncate_log_recovery(osb,
959 tl_dinode);
960 if (ret < 0)
961 mlog_errno(ret);
962
963 kfree(tl_dinode);
964 }
965
966 ret = ocfs2_recover_orphans(osb, item->lri_slot);
967 if (ret < 0)
968 mlog_errno(ret);
969
970 kfree(item);
971 }
972
973 mlog(0, "Recovery completion\n");
974 mlog_exit_void();
975}
976
977/* NOTE: This function always eats your references to la_dinode and
978 * tl_dinode, either manually on error, or by passing them to
979 * ocfs2_complete_recovery */
980static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal,
981 int slot_num,
982 struct ocfs2_dinode *la_dinode,
983 struct ocfs2_dinode *tl_dinode)
984{
985 struct ocfs2_la_recovery_item *item;
986
Sunil Mushranafae00ab2006-04-12 14:37:00 -0700987 item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS);
Mark Fashehccd979b2005-12-15 14:31:24 -0800988 if (!item) {
989 /* Though we wish to avoid it, we are in fact safe in
990 * skipping local alloc cleanup as fsck.ocfs2 is more
991 * than capable of reclaiming unused space. */
992 if (la_dinode)
993 kfree(la_dinode);
994
995 if (tl_dinode)
996 kfree(tl_dinode);
997
998 mlog_errno(-ENOMEM);
999 return;
1000 }
1001
1002 INIT_LIST_HEAD(&item->lri_list);
1003 item->lri_la_dinode = la_dinode;
1004 item->lri_slot = slot_num;
1005 item->lri_tl_dinode = tl_dinode;
1006
1007 spin_lock(&journal->j_lock);
1008 list_add_tail(&item->lri_list, &journal->j_la_cleanups);
1009 queue_work(ocfs2_wq, &journal->j_recovery_work);
1010 spin_unlock(&journal->j_lock);
1011}
1012
1013/* Called by the mount code to queue recovery the last part of
1014 * recovery for it's own slot. */
1015void ocfs2_complete_mount_recovery(struct ocfs2_super *osb)
1016{
1017 struct ocfs2_journal *journal = osb->journal;
1018
1019 if (osb->dirty) {
1020 /* No need to queue up our truncate_log as regular
1021 * cleanup will catch that. */
1022 ocfs2_queue_recovery_completion(journal,
1023 osb->slot_num,
1024 osb->local_alloc_copy,
1025 NULL);
1026 ocfs2_schedule_truncate_log_flush(osb, 0);
1027
1028 osb->local_alloc_copy = NULL;
1029 osb->dirty = 0;
1030 }
1031}
1032
1033static int __ocfs2_recovery_thread(void *arg)
1034{
1035 int status, node_num;
1036 struct ocfs2_super *osb = arg;
1037
1038 mlog_entry_void();
1039
1040 status = ocfs2_wait_on_mount(osb);
1041 if (status < 0) {
1042 goto bail;
1043 }
1044
1045restart:
1046 status = ocfs2_super_lock(osb, 1);
1047 if (status < 0) {
1048 mlog_errno(status);
1049 goto bail;
1050 }
1051
1052 while(!ocfs2_node_map_is_empty(osb, &osb->recovery_map)) {
1053 node_num = ocfs2_node_map_first_set_bit(osb,
1054 &osb->recovery_map);
1055 if (node_num == O2NM_INVALID_NODE_NUM) {
1056 mlog(0, "Out of nodes to recover.\n");
1057 break;
1058 }
1059
1060 status = ocfs2_recover_node(osb, node_num);
1061 if (status < 0) {
1062 mlog(ML_ERROR,
1063 "Error %d recovering node %d on device (%u,%u)!\n",
1064 status, node_num,
1065 MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1066 mlog(ML_ERROR, "Volume requires unmount.\n");
1067 continue;
1068 }
1069
1070 ocfs2_recovery_map_clear(osb, node_num);
1071 }
1072 ocfs2_super_unlock(osb, 1);
1073
1074 /* We always run recovery on our own orphan dir - the dead
1075 * node(s) may have voted "no" on an inode delete earlier. A
1076 * revote is therefore required. */
1077 ocfs2_queue_recovery_completion(osb->journal, osb->slot_num, NULL,
1078 NULL);
1079
1080bail:
Arjan van de Venc74ec2f2006-01-13 21:54:23 -08001081 mutex_lock(&osb->recovery_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -08001082 if (!status &&
1083 !ocfs2_node_map_is_empty(osb, &osb->recovery_map)) {
Arjan van de Venc74ec2f2006-01-13 21:54:23 -08001084 mutex_unlock(&osb->recovery_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -08001085 goto restart;
1086 }
1087
1088 osb->recovery_thread_task = NULL;
1089 mb(); /* sync with ocfs2_recovery_thread_running */
1090 wake_up(&osb->recovery_event);
1091
Arjan van de Venc74ec2f2006-01-13 21:54:23 -08001092 mutex_unlock(&osb->recovery_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -08001093
1094 mlog_exit(status);
1095 /* no one is callint kthread_stop() for us so the kthread() api
1096 * requires that we call do_exit(). And it isn't exported, but
1097 * complete_and_exit() seems to be a minimal wrapper around it. */
1098 complete_and_exit(NULL, status);
1099 return status;
1100}
1101
1102void ocfs2_recovery_thread(struct ocfs2_super *osb, int node_num)
1103{
1104 mlog_entry("(node_num=%d, osb->node_num = %d)\n",
1105 node_num, osb->node_num);
1106
Arjan van de Venc74ec2f2006-01-13 21:54:23 -08001107 mutex_lock(&osb->recovery_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -08001108 if (osb->disable_recovery)
1109 goto out;
1110
1111 /* People waiting on recovery will wait on
1112 * the recovery map to empty. */
1113 if (!ocfs2_recovery_map_set(osb, node_num))
1114 mlog(0, "node %d already be in recovery.\n", node_num);
1115
1116 mlog(0, "starting recovery thread...\n");
1117
1118 if (osb->recovery_thread_task)
1119 goto out;
1120
1121 osb->recovery_thread_task = kthread_run(__ocfs2_recovery_thread, osb,
1122 "ocfs2rec-%d", osb->osb_id);
1123 if (IS_ERR(osb->recovery_thread_task)) {
1124 mlog_errno((int)PTR_ERR(osb->recovery_thread_task));
1125 osb->recovery_thread_task = NULL;
1126 }
1127
1128out:
Arjan van de Venc74ec2f2006-01-13 21:54:23 -08001129 mutex_unlock(&osb->recovery_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -08001130 wake_up(&osb->recovery_event);
1131
1132 mlog_exit_void();
1133}
1134
1135/* Does the actual journal replay and marks the journal inode as
1136 * clean. Will only replay if the journal inode is marked dirty. */
1137static int ocfs2_replay_journal(struct ocfs2_super *osb,
1138 int node_num,
1139 int slot_num)
1140{
1141 int status;
1142 int got_lock = 0;
1143 unsigned int flags;
1144 struct inode *inode = NULL;
1145 struct ocfs2_dinode *fe;
1146 journal_t *journal = NULL;
1147 struct buffer_head *bh = NULL;
1148
1149 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
1150 slot_num);
1151 if (inode == NULL) {
1152 status = -EACCES;
1153 mlog_errno(status);
1154 goto done;
1155 }
1156 if (is_bad_inode(inode)) {
1157 status = -EACCES;
1158 iput(inode);
1159 inode = NULL;
1160 mlog_errno(status);
1161 goto done;
1162 }
1163 SET_INODE_JOURNAL(inode);
1164
1165 status = ocfs2_meta_lock_full(inode, NULL, &bh, 1,
1166 OCFS2_META_LOCK_RECOVERY);
1167 if (status < 0) {
1168 mlog(0, "status returned from ocfs2_meta_lock=%d\n", status);
1169 if (status != -ERESTARTSYS)
1170 mlog(ML_ERROR, "Could not lock journal!\n");
1171 goto done;
1172 }
1173 got_lock = 1;
1174
1175 fe = (struct ocfs2_dinode *) bh->b_data;
1176
1177 flags = le32_to_cpu(fe->id1.journal1.ij_flags);
1178
1179 if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) {
1180 mlog(0, "No recovery required for node %d\n", node_num);
1181 goto done;
1182 }
1183
1184 mlog(ML_NOTICE, "Recovering node %d from slot %d on device (%u,%u)\n",
1185 node_num, slot_num,
1186 MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1187
1188 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
1189
1190 status = ocfs2_force_read_journal(inode);
1191 if (status < 0) {
1192 mlog_errno(status);
1193 goto done;
1194 }
1195
1196 mlog(0, "calling journal_init_inode\n");
1197 journal = journal_init_inode(inode);
1198 if (journal == NULL) {
1199 mlog(ML_ERROR, "Linux journal layer error\n");
1200 status = -EIO;
1201 goto done;
1202 }
1203
1204 status = journal_load(journal);
1205 if (status < 0) {
1206 mlog_errno(status);
1207 if (!igrab(inode))
1208 BUG();
1209 journal_destroy(journal);
1210 goto done;
1211 }
1212
1213 ocfs2_clear_journal_error(osb->sb, journal, slot_num);
1214
1215 /* wipe the journal */
1216 mlog(0, "flushing the journal.\n");
1217 journal_lock_updates(journal);
1218 status = journal_flush(journal);
1219 journal_unlock_updates(journal);
1220 if (status < 0)
1221 mlog_errno(status);
1222
1223 /* This will mark the node clean */
1224 flags = le32_to_cpu(fe->id1.journal1.ij_flags);
1225 flags &= ~OCFS2_JOURNAL_DIRTY_FL;
1226 fe->id1.journal1.ij_flags = cpu_to_le32(flags);
1227
1228 status = ocfs2_write_block(osb, bh, inode);
1229 if (status < 0)
1230 mlog_errno(status);
1231
1232 if (!igrab(inode))
1233 BUG();
1234
1235 journal_destroy(journal);
1236
1237done:
1238 /* drop the lock on this nodes journal */
1239 if (got_lock)
1240 ocfs2_meta_unlock(inode, 1);
1241
1242 if (inode)
1243 iput(inode);
1244
1245 if (bh)
1246 brelse(bh);
1247
1248 mlog_exit(status);
1249 return status;
1250}
1251
1252/*
1253 * Do the most important parts of node recovery:
1254 * - Replay it's journal
1255 * - Stamp a clean local allocator file
1256 * - Stamp a clean truncate log
1257 * - Mark the node clean
1258 *
1259 * If this function completes without error, a node in OCFS2 can be
1260 * said to have been safely recovered. As a result, failure during the
1261 * second part of a nodes recovery process (local alloc recovery) is
1262 * far less concerning.
1263 */
1264static int ocfs2_recover_node(struct ocfs2_super *osb,
1265 int node_num)
1266{
1267 int status = 0;
1268 int slot_num;
1269 struct ocfs2_slot_info *si = osb->slot_info;
1270 struct ocfs2_dinode *la_copy = NULL;
1271 struct ocfs2_dinode *tl_copy = NULL;
1272
1273 mlog_entry("(node_num=%d, osb->node_num = %d)\n",
1274 node_num, osb->node_num);
1275
1276 mlog(0, "checking node %d\n", node_num);
1277
1278 /* Should not ever be called to recover ourselves -- in that
1279 * case we should've called ocfs2_journal_load instead. */
Eric Sesterhenn / snakebyteebdec832006-01-27 10:32:52 +01001280 BUG_ON(osb->node_num == node_num);
Mark Fashehccd979b2005-12-15 14:31:24 -08001281
1282 slot_num = ocfs2_node_num_to_slot(si, node_num);
1283 if (slot_num == OCFS2_INVALID_SLOT) {
1284 status = 0;
1285 mlog(0, "no slot for this node, so no recovery required.\n");
1286 goto done;
1287 }
1288
1289 mlog(0, "node %d was using slot %d\n", node_num, slot_num);
1290
1291 status = ocfs2_replay_journal(osb, node_num, slot_num);
1292 if (status < 0) {
1293 mlog_errno(status);
1294 goto done;
1295 }
1296
1297 /* Stamp a clean local alloc file AFTER recovering the journal... */
1298 status = ocfs2_begin_local_alloc_recovery(osb, slot_num, &la_copy);
1299 if (status < 0) {
1300 mlog_errno(status);
1301 goto done;
1302 }
1303
1304 /* An error from begin_truncate_log_recovery is not
1305 * serious enough to warrant halting the rest of
1306 * recovery. */
1307 status = ocfs2_begin_truncate_log_recovery(osb, slot_num, &tl_copy);
1308 if (status < 0)
1309 mlog_errno(status);
1310
1311 /* Likewise, this would be a strange but ultimately not so
1312 * harmful place to get an error... */
1313 ocfs2_clear_slot(si, slot_num);
1314 status = ocfs2_update_disk_slots(osb, si);
1315 if (status < 0)
1316 mlog_errno(status);
1317
1318 /* This will kfree the memory pointed to by la_copy and tl_copy */
1319 ocfs2_queue_recovery_completion(osb->journal, slot_num, la_copy,
1320 tl_copy);
1321
1322 status = 0;
1323done:
1324
1325 mlog_exit(status);
1326 return status;
1327}
1328
1329/* Test node liveness by trylocking his journal. If we get the lock,
1330 * we drop it here. Return 0 if we got the lock, -EAGAIN if node is
1331 * still alive (we couldn't get the lock) and < 0 on error. */
1332static int ocfs2_trylock_journal(struct ocfs2_super *osb,
1333 int slot_num)
1334{
1335 int status, flags;
1336 struct inode *inode = NULL;
1337
1338 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
1339 slot_num);
1340 if (inode == NULL) {
1341 mlog(ML_ERROR, "access error\n");
1342 status = -EACCES;
1343 goto bail;
1344 }
1345 if (is_bad_inode(inode)) {
1346 mlog(ML_ERROR, "access error (bad inode)\n");
1347 iput(inode);
1348 inode = NULL;
1349 status = -EACCES;
1350 goto bail;
1351 }
1352 SET_INODE_JOURNAL(inode);
1353
1354 flags = OCFS2_META_LOCK_RECOVERY | OCFS2_META_LOCK_NOQUEUE;
1355 status = ocfs2_meta_lock_full(inode, NULL, NULL, 1, flags);
1356 if (status < 0) {
1357 if (status != -EAGAIN)
1358 mlog_errno(status);
1359 goto bail;
1360 }
1361
1362 ocfs2_meta_unlock(inode, 1);
1363bail:
1364 if (inode)
1365 iput(inode);
1366
1367 return status;
1368}
1369
1370/* Call this underneath ocfs2_super_lock. It also assumes that the
1371 * slot info struct has been updated from disk. */
1372int ocfs2_mark_dead_nodes(struct ocfs2_super *osb)
1373{
1374 int status, i, node_num;
1375 struct ocfs2_slot_info *si = osb->slot_info;
1376
1377 /* This is called with the super block cluster lock, so we
1378 * know that the slot map can't change underneath us. */
1379
1380 spin_lock(&si->si_lock);
1381 for(i = 0; i < si->si_num_slots; i++) {
1382 if (i == osb->slot_num)
1383 continue;
1384 if (ocfs2_is_empty_slot(si, i))
1385 continue;
1386
1387 node_num = si->si_global_node_nums[i];
1388 if (ocfs2_node_map_test_bit(osb, &osb->recovery_map, node_num))
1389 continue;
1390 spin_unlock(&si->si_lock);
1391
1392 /* Ok, we have a slot occupied by another node which
1393 * is not in the recovery map. We trylock his journal
1394 * file here to test if he's alive. */
1395 status = ocfs2_trylock_journal(osb, i);
1396 if (!status) {
1397 /* Since we're called from mount, we know that
1398 * the recovery thread can't race us on
1399 * setting / checking the recovery bits. */
1400 ocfs2_recovery_thread(osb, node_num);
1401 } else if ((status < 0) && (status != -EAGAIN)) {
1402 mlog_errno(status);
1403 goto bail;
1404 }
1405
1406 spin_lock(&si->si_lock);
1407 }
1408 spin_unlock(&si->si_lock);
1409
1410 status = 0;
1411bail:
1412 mlog_exit(status);
1413 return status;
1414}
1415
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001416static int ocfs2_queue_orphans(struct ocfs2_super *osb,
1417 int slot,
1418 struct inode **head)
Mark Fashehccd979b2005-12-15 14:31:24 -08001419{
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001420 int status;
Mark Fashehccd979b2005-12-15 14:31:24 -08001421 struct inode *orphan_dir_inode = NULL;
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001422 struct inode *iter;
Mark Fashehccd979b2005-12-15 14:31:24 -08001423 unsigned long offset, blk, local;
1424 struct buffer_head *bh = NULL;
1425 struct ocfs2_dir_entry *de;
1426 struct super_block *sb = osb->sb;
Mark Fashehccd979b2005-12-15 14:31:24 -08001427
1428 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
1429 ORPHAN_DIR_SYSTEM_INODE,
1430 slot);
1431 if (!orphan_dir_inode) {
1432 status = -ENOENT;
1433 mlog_errno(status);
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001434 return status;
1435 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001436
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001437 mutex_lock(&orphan_dir_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -08001438 status = ocfs2_meta_lock(orphan_dir_inode, NULL, NULL, 0);
1439 if (status < 0) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001440 mlog_errno(status);
1441 goto out;
1442 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001443
1444 offset = 0;
1445 iter = NULL;
1446 while(offset < i_size_read(orphan_dir_inode)) {
1447 blk = offset >> sb->s_blocksize_bits;
1448
1449 bh = ocfs2_bread(orphan_dir_inode, blk, &status, 0);
1450 if (!bh)
1451 status = -EINVAL;
1452 if (status < 0) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001453 if (bh)
1454 brelse(bh);
1455 mlog_errno(status);
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001456 goto out_unlock;
Mark Fashehccd979b2005-12-15 14:31:24 -08001457 }
1458
1459 local = 0;
1460 while(offset < i_size_read(orphan_dir_inode)
1461 && local < sb->s_blocksize) {
1462 de = (struct ocfs2_dir_entry *) (bh->b_data + local);
1463
1464 if (!ocfs2_check_dir_entry(orphan_dir_inode,
1465 de, bh, local)) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001466 status = -EINVAL;
1467 mlog_errno(status);
1468 brelse(bh);
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001469 goto out_unlock;
Mark Fashehccd979b2005-12-15 14:31:24 -08001470 }
1471
1472 local += le16_to_cpu(de->rec_len);
1473 offset += le16_to_cpu(de->rec_len);
1474
1475 /* I guess we silently fail on no inode? */
1476 if (!le64_to_cpu(de->inode))
1477 continue;
1478 if (de->file_type > OCFS2_FT_MAX) {
1479 mlog(ML_ERROR,
1480 "block %llu contains invalid de: "
Mark Fashehb0697052006-03-03 10:24:33 -08001481 "inode = %llu, rec_len = %u, "
Mark Fashehccd979b2005-12-15 14:31:24 -08001482 "name_len = %u, file_type = %u, "
1483 "name='%.*s'\n",
1484 (unsigned long long)bh->b_blocknr,
Mark Fashehb0697052006-03-03 10:24:33 -08001485 (unsigned long long)le64_to_cpu(de->inode),
Mark Fashehccd979b2005-12-15 14:31:24 -08001486 le16_to_cpu(de->rec_len),
1487 de->name_len,
1488 de->file_type,
1489 de->name_len,
1490 de->name);
1491 continue;
1492 }
1493 if (de->name_len == 1 && !strncmp(".", de->name, 1))
1494 continue;
1495 if (de->name_len == 2 && !strncmp("..", de->name, 2))
1496 continue;
1497
1498 iter = ocfs2_iget(osb, le64_to_cpu(de->inode));
1499 if (IS_ERR(iter))
1500 continue;
1501
Mark Fashehb0697052006-03-03 10:24:33 -08001502 mlog(0, "queue orphan %llu\n",
1503 (unsigned long long)OCFS2_I(iter)->ip_blkno);
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001504 /* No locking is required for the next_orphan
1505 * queue as there is only ever a single
1506 * process doing orphan recovery. */
1507 OCFS2_I(iter)->ip_next_orphan = *head;
1508 *head = iter;
Mark Fashehccd979b2005-12-15 14:31:24 -08001509 }
1510 brelse(bh);
1511 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001512
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001513out_unlock:
Mark Fashehccd979b2005-12-15 14:31:24 -08001514 ocfs2_meta_unlock(orphan_dir_inode, 0);
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001515out:
1516 mutex_unlock(&orphan_dir_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -08001517 iput(orphan_dir_inode);
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001518 return status;
1519}
1520
1521static int ocfs2_orphan_recovery_can_continue(struct ocfs2_super *osb,
1522 int slot)
1523{
1524 int ret;
1525
1526 spin_lock(&osb->osb_lock);
1527 ret = !osb->osb_orphan_wipes[slot];
1528 spin_unlock(&osb->osb_lock);
1529 return ret;
1530}
1531
1532static void ocfs2_mark_recovering_orphan_dir(struct ocfs2_super *osb,
1533 int slot)
1534{
1535 spin_lock(&osb->osb_lock);
1536 /* Mark ourselves such that new processes in delete_inode()
1537 * know to quit early. */
1538 ocfs2_node_map_set_bit(osb, &osb->osb_recovering_orphan_dirs, slot);
1539 while (osb->osb_orphan_wipes[slot]) {
1540 /* If any processes are already in the middle of an
1541 * orphan wipe on this dir, then we need to wait for
1542 * them. */
1543 spin_unlock(&osb->osb_lock);
1544 wait_event_interruptible(osb->osb_wipe_event,
1545 ocfs2_orphan_recovery_can_continue(osb, slot));
1546 spin_lock(&osb->osb_lock);
1547 }
1548 spin_unlock(&osb->osb_lock);
1549}
1550
1551static void ocfs2_clear_recovering_orphan_dir(struct ocfs2_super *osb,
1552 int slot)
1553{
1554 ocfs2_node_map_clear_bit(osb, &osb->osb_recovering_orphan_dirs, slot);
1555}
1556
1557/*
1558 * Orphan recovery. Each mounted node has it's own orphan dir which we
1559 * must run during recovery. Our strategy here is to build a list of
1560 * the inodes in the orphan dir and iget/iput them. The VFS does
1561 * (most) of the rest of the work.
1562 *
1563 * Orphan recovery can happen at any time, not just mount so we have a
1564 * couple of extra considerations.
1565 *
1566 * - We grab as many inodes as we can under the orphan dir lock -
1567 * doing iget() outside the orphan dir risks getting a reference on
1568 * an invalid inode.
1569 * - We must be sure not to deadlock with other processes on the
1570 * system wanting to run delete_inode(). This can happen when they go
1571 * to lock the orphan dir and the orphan recovery process attempts to
1572 * iget() inside the orphan dir lock. This can be avoided by
1573 * advertising our state to ocfs2_delete_inode().
1574 */
1575static int ocfs2_recover_orphans(struct ocfs2_super *osb,
1576 int slot)
1577{
1578 int ret = 0;
1579 struct inode *inode = NULL;
1580 struct inode *iter;
1581 struct ocfs2_inode_info *oi;
1582
1583 mlog(0, "Recover inodes from orphan dir in slot %d\n", slot);
1584
1585 ocfs2_mark_recovering_orphan_dir(osb, slot);
1586 ret = ocfs2_queue_orphans(osb, slot, &inode);
1587 ocfs2_clear_recovering_orphan_dir(osb, slot);
1588
1589 /* Error here should be noted, but we want to continue with as
1590 * many queued inodes as we've got. */
1591 if (ret)
1592 mlog_errno(ret);
Mark Fashehccd979b2005-12-15 14:31:24 -08001593
1594 while (inode) {
1595 oi = OCFS2_I(inode);
Mark Fashehb0697052006-03-03 10:24:33 -08001596 mlog(0, "iput orphan %llu\n", (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001597
1598 iter = oi->ip_next_orphan;
1599
1600 spin_lock(&oi->ip_lock);
1601 /* Delete voting may have set these on the assumption
1602 * that the other node would wipe them successfully.
1603 * If they are still in the node's orphan dir, we need
1604 * to reset that state. */
1605 oi->ip_flags &= ~(OCFS2_INODE_DELETED|OCFS2_INODE_SKIP_DELETE);
1606
1607 /* Set the proper information to get us going into
1608 * ocfs2_delete_inode. */
1609 oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED;
1610 oi->ip_orphaned_slot = slot;
1611 spin_unlock(&oi->ip_lock);
1612
1613 iput(inode);
1614
1615 inode = iter;
1616 }
1617
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001618 return ret;
Mark Fashehccd979b2005-12-15 14:31:24 -08001619}
1620
1621static int ocfs2_wait_on_mount(struct ocfs2_super *osb)
1622{
1623 /* This check is good because ocfs2 will wait on our recovery
1624 * thread before changing it to something other than MOUNTED
1625 * or DISABLED. */
1626 wait_event(osb->osb_mount_event,
1627 atomic_read(&osb->vol_state) == VOLUME_MOUNTED ||
1628 atomic_read(&osb->vol_state) == VOLUME_DISABLED);
1629
1630 /* If there's an error on mount, then we may never get to the
1631 * MOUNTED flag, but this is set right before
1632 * dismount_volume() so we can trust it. */
1633 if (atomic_read(&osb->vol_state) == VOLUME_DISABLED) {
1634 mlog(0, "mount error, exiting!\n");
1635 return -EBUSY;
1636 }
1637
1638 return 0;
1639}
1640
1641static int ocfs2_commit_thread(void *arg)
1642{
1643 int status;
1644 struct ocfs2_super *osb = arg;
1645 struct ocfs2_journal *journal = osb->journal;
1646
1647 /* we can trust j_num_trans here because _should_stop() is only set in
1648 * shutdown and nobody other than ourselves should be able to start
1649 * transactions. committing on shutdown might take a few iterations
1650 * as final transactions put deleted inodes on the list */
1651 while (!(kthread_should_stop() &&
1652 atomic_read(&journal->j_num_trans) == 0)) {
1653
Mark Fasheh745ae8ba2006-02-09 13:23:39 -08001654 wait_event_interruptible(osb->checkpoint_event,
1655 atomic_read(&journal->j_num_trans)
1656 || kthread_should_stop());
Mark Fashehccd979b2005-12-15 14:31:24 -08001657
1658 status = ocfs2_commit_cache(osb);
1659 if (status < 0)
1660 mlog_errno(status);
1661
1662 if (kthread_should_stop() && atomic_read(&journal->j_num_trans)){
1663 mlog(ML_KTHREAD,
1664 "commit_thread: %u transactions pending on "
1665 "shutdown\n",
1666 atomic_read(&journal->j_num_trans));
1667 }
1668 }
1669
1670 return 0;
1671}
1672
1673/* Look for a dirty journal without taking any cluster locks. Used for
1674 * hard readonly access to determine whether the file system journals
1675 * require recovery. */
1676int ocfs2_check_journals_nolocks(struct ocfs2_super *osb)
1677{
1678 int ret = 0;
1679 unsigned int slot;
1680 struct buffer_head *di_bh;
1681 struct ocfs2_dinode *di;
1682 struct inode *journal = NULL;
1683
1684 for(slot = 0; slot < osb->max_slots; slot++) {
1685 journal = ocfs2_get_system_file_inode(osb,
1686 JOURNAL_SYSTEM_INODE,
1687 slot);
1688 if (!journal || is_bad_inode(journal)) {
1689 ret = -EACCES;
1690 mlog_errno(ret);
1691 goto out;
1692 }
1693
1694 di_bh = NULL;
1695 ret = ocfs2_read_block(osb, OCFS2_I(journal)->ip_blkno, &di_bh,
1696 0, journal);
1697 if (ret < 0) {
1698 mlog_errno(ret);
1699 goto out;
1700 }
1701
1702 di = (struct ocfs2_dinode *) di_bh->b_data;
1703
1704 if (le32_to_cpu(di->id1.journal1.ij_flags) &
1705 OCFS2_JOURNAL_DIRTY_FL)
1706 ret = -EROFS;
1707
1708 brelse(di_bh);
1709 if (ret)
1710 break;
1711 }
1712
1713out:
1714 if (journal)
1715 iput(journal);
1716
1717 return ret;
1718}