x86: convert cpuinfo_x86 array to a per_cpu array
[wrt350n-kernel.git] / fs / ocfs2 / journal.c
blobf9d01e25298decaa3b24ab3e3c38ee88d3d6c55c
1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
4 * journal.c
6 * Defines functions of journalling api
8 * Copyright (C) 2003, 2004 Oracle. All rights reserved.
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.
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.
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.
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>
32 #define MLOG_MASK_PREFIX ML_JOURNAL
33 #include <cluster/masklog.h>
35 #include "ocfs2.h"
37 #include "alloc.h"
38 #include "dir.h"
39 #include "dlmglue.h"
40 #include "extent_map.h"
41 #include "heartbeat.h"
42 #include "inode.h"
43 #include "journal.h"
44 #include "localalloc.h"
45 #include "slot_map.h"
46 #include "super.h"
47 #include "vote.h"
48 #include "sysfile.h"
50 #include "buffer_head_io.h"
52 DEFINE_SPINLOCK(trans_inc_lock);
54 static int ocfs2_force_read_journal(struct inode *inode);
55 static int ocfs2_recover_node(struct ocfs2_super *osb,
56 int node_num);
57 static int __ocfs2_recovery_thread(void *arg);
58 static int ocfs2_commit_cache(struct ocfs2_super *osb);
59 static int ocfs2_wait_on_mount(struct ocfs2_super *osb);
60 static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
61 int dirty);
62 static int ocfs2_trylock_journal(struct ocfs2_super *osb,
63 int slot_num);
64 static int ocfs2_recover_orphans(struct ocfs2_super *osb,
65 int slot);
66 static int ocfs2_commit_thread(void *arg);
68 static int ocfs2_commit_cache(struct ocfs2_super *osb)
70 int status = 0;
71 unsigned int flushed;
72 unsigned long old_id;
73 struct ocfs2_journal *journal = NULL;
75 mlog_entry_void();
77 journal = osb->journal;
79 /* Flush all pending commits and checkpoint the journal. */
80 down_write(&journal->j_trans_barrier);
82 if (atomic_read(&journal->j_num_trans) == 0) {
83 up_write(&journal->j_trans_barrier);
84 mlog(0, "No transactions for me to flush!\n");
85 goto finally;
88 journal_lock_updates(journal->j_journal);
89 status = journal_flush(journal->j_journal);
90 journal_unlock_updates(journal->j_journal);
91 if (status < 0) {
92 up_write(&journal->j_trans_barrier);
93 mlog_errno(status);
94 goto finally;
97 old_id = ocfs2_inc_trans_id(journal);
99 flushed = atomic_read(&journal->j_num_trans);
100 atomic_set(&journal->j_num_trans, 0);
101 up_write(&journal->j_trans_barrier);
103 mlog(0, "commit_thread: flushed transaction %lu (%u handles)\n",
104 journal->j_trans_id, flushed);
106 ocfs2_kick_vote_thread(osb);
107 wake_up(&journal->j_checkpointed);
108 finally:
109 mlog_exit(status);
110 return status;
113 /* pass it NULL and it will allocate a new handle object for you. If
114 * you pass it a handle however, it may still return error, in which
115 * case it has free'd the passed handle for you. */
116 handle_t *ocfs2_start_trans(struct ocfs2_super *osb, int max_buffs)
118 journal_t *journal = osb->journal->j_journal;
119 handle_t *handle;
121 BUG_ON(!osb || !osb->journal->j_journal);
123 if (ocfs2_is_hard_readonly(osb))
124 return ERR_PTR(-EROFS);
126 BUG_ON(osb->journal->j_state == OCFS2_JOURNAL_FREE);
127 BUG_ON(max_buffs <= 0);
129 /* JBD might support this, but our journalling code doesn't yet. */
130 if (journal_current_handle()) {
131 mlog(ML_ERROR, "Recursive transaction attempted!\n");
132 BUG();
135 down_read(&osb->journal->j_trans_barrier);
137 handle = journal_start(journal, max_buffs);
138 if (IS_ERR(handle)) {
139 up_read(&osb->journal->j_trans_barrier);
141 mlog_errno(PTR_ERR(handle));
143 if (is_journal_aborted(journal)) {
144 ocfs2_abort(osb->sb, "Detected aborted journal");
145 handle = ERR_PTR(-EROFS);
147 } else {
148 if (!ocfs2_mount_local(osb))
149 atomic_inc(&(osb->journal->j_num_trans));
152 return handle;
155 int ocfs2_commit_trans(struct ocfs2_super *osb,
156 handle_t *handle)
158 int ret;
159 struct ocfs2_journal *journal = osb->journal;
161 BUG_ON(!handle);
163 ret = journal_stop(handle);
164 if (ret < 0)
165 mlog_errno(ret);
167 up_read(&journal->j_trans_barrier);
169 return ret;
173 * 'nblocks' is what you want to add to the current
174 * transaction. extend_trans will either extend the current handle by
175 * nblocks, or commit it and start a new one with nblocks credits.
177 * WARNING: This will not release any semaphores or disk locks taken
178 * during the transaction, so make sure they were taken *before*
179 * start_trans or we'll have ordering deadlocks.
181 * WARNING2: Note that we do *not* drop j_trans_barrier here. This is
182 * good because transaction ids haven't yet been recorded on the
183 * cluster locks associated with this handle.
185 int ocfs2_extend_trans(handle_t *handle, int nblocks)
187 int status;
189 BUG_ON(!handle);
190 BUG_ON(!nblocks);
192 mlog_entry_void();
194 mlog(0, "Trying to extend transaction by %d blocks\n", nblocks);
196 status = journal_extend(handle, nblocks);
197 if (status < 0) {
198 mlog_errno(status);
199 goto bail;
202 if (status > 0) {
203 mlog(0, "journal_extend failed, trying journal_restart\n");
204 status = journal_restart(handle, nblocks);
205 if (status < 0) {
206 mlog_errno(status);
207 goto bail;
211 status = 0;
212 bail:
214 mlog_exit(status);
215 return status;
218 int ocfs2_journal_access(handle_t *handle,
219 struct inode *inode,
220 struct buffer_head *bh,
221 int type)
223 int status;
225 BUG_ON(!inode);
226 BUG_ON(!handle);
227 BUG_ON(!bh);
229 mlog_entry("bh->b_blocknr=%llu, type=%d (\"%s\"), bh->b_size = %zu\n",
230 (unsigned long long)bh->b_blocknr, type,
231 (type == OCFS2_JOURNAL_ACCESS_CREATE) ?
232 "OCFS2_JOURNAL_ACCESS_CREATE" :
233 "OCFS2_JOURNAL_ACCESS_WRITE",
234 bh->b_size);
236 /* we can safely remove this assertion after testing. */
237 if (!buffer_uptodate(bh)) {
238 mlog(ML_ERROR, "giving me a buffer that's not uptodate!\n");
239 mlog(ML_ERROR, "b_blocknr=%llu\n",
240 (unsigned long long)bh->b_blocknr);
241 BUG();
244 /* Set the current transaction information on the inode so
245 * that the locking code knows whether it can drop it's locks
246 * on this inode or not. We're protected from the commit
247 * thread updating the current transaction id until
248 * ocfs2_commit_trans() because ocfs2_start_trans() took
249 * j_trans_barrier for us. */
250 ocfs2_set_inode_lock_trans(OCFS2_SB(inode->i_sb)->journal, inode);
252 mutex_lock(&OCFS2_I(inode)->ip_io_mutex);
253 switch (type) {
254 case OCFS2_JOURNAL_ACCESS_CREATE:
255 case OCFS2_JOURNAL_ACCESS_WRITE:
256 status = journal_get_write_access(handle, bh);
257 break;
259 case OCFS2_JOURNAL_ACCESS_UNDO:
260 status = journal_get_undo_access(handle, bh);
261 break;
263 default:
264 status = -EINVAL;
265 mlog(ML_ERROR, "Uknown access type!\n");
267 mutex_unlock(&OCFS2_I(inode)->ip_io_mutex);
269 if (status < 0)
270 mlog(ML_ERROR, "Error %d getting %d access to buffer!\n",
271 status, type);
273 mlog_exit(status);
274 return status;
277 int ocfs2_journal_dirty(handle_t *handle,
278 struct buffer_head *bh)
280 int status;
282 mlog_entry("(bh->b_blocknr=%llu)\n",
283 (unsigned long long)bh->b_blocknr);
285 status = journal_dirty_metadata(handle, bh);
286 if (status < 0)
287 mlog(ML_ERROR, "Could not dirty metadata buffer. "
288 "(bh->b_blocknr=%llu)\n",
289 (unsigned long long)bh->b_blocknr);
291 mlog_exit(status);
292 return status;
295 int ocfs2_journal_dirty_data(handle_t *handle,
296 struct buffer_head *bh)
298 int err = journal_dirty_data(handle, bh);
299 if (err)
300 mlog_errno(err);
301 /* TODO: When we can handle it, abort the handle and go RO on
302 * error here. */
304 return err;
307 #define OCFS2_DEFAULT_COMMIT_INTERVAL (HZ * 5)
309 void ocfs2_set_journal_params(struct ocfs2_super *osb)
311 journal_t *journal = osb->journal->j_journal;
313 spin_lock(&journal->j_state_lock);
314 journal->j_commit_interval = OCFS2_DEFAULT_COMMIT_INTERVAL;
315 if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER)
316 journal->j_flags |= JFS_BARRIER;
317 else
318 journal->j_flags &= ~JFS_BARRIER;
319 spin_unlock(&journal->j_state_lock);
322 int ocfs2_journal_init(struct ocfs2_journal *journal, int *dirty)
324 int status = -1;
325 struct inode *inode = NULL; /* the journal inode */
326 journal_t *j_journal = NULL;
327 struct ocfs2_dinode *di = NULL;
328 struct buffer_head *bh = NULL;
329 struct ocfs2_super *osb;
330 int meta_lock = 0;
332 mlog_entry_void();
334 BUG_ON(!journal);
336 osb = journal->j_osb;
338 /* already have the inode for our journal */
339 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
340 osb->slot_num);
341 if (inode == NULL) {
342 status = -EACCES;
343 mlog_errno(status);
344 goto done;
346 if (is_bad_inode(inode)) {
347 mlog(ML_ERROR, "access error (bad inode)\n");
348 iput(inode);
349 inode = NULL;
350 status = -EACCES;
351 goto done;
354 SET_INODE_JOURNAL(inode);
355 OCFS2_I(inode)->ip_open_count++;
357 /* Skip recovery waits here - journal inode metadata never
358 * changes in a live cluster so it can be considered an
359 * exception to the rule. */
360 status = ocfs2_meta_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY);
361 if (status < 0) {
362 if (status != -ERESTARTSYS)
363 mlog(ML_ERROR, "Could not get lock on journal!\n");
364 goto done;
367 meta_lock = 1;
368 di = (struct ocfs2_dinode *)bh->b_data;
370 if (inode->i_size < OCFS2_MIN_JOURNAL_SIZE) {
371 mlog(ML_ERROR, "Journal file size (%lld) is too small!\n",
372 inode->i_size);
373 status = -EINVAL;
374 goto done;
377 mlog(0, "inode->i_size = %lld\n", inode->i_size);
378 mlog(0, "inode->i_blocks = %llu\n",
379 (unsigned long long)inode->i_blocks);
380 mlog(0, "inode->ip_clusters = %u\n", OCFS2_I(inode)->ip_clusters);
382 /* call the kernels journal init function now */
383 j_journal = journal_init_inode(inode);
384 if (j_journal == NULL) {
385 mlog(ML_ERROR, "Linux journal layer error\n");
386 status = -EINVAL;
387 goto done;
390 mlog(0, "Returned from journal_init_inode\n");
391 mlog(0, "j_journal->j_maxlen = %u\n", j_journal->j_maxlen);
393 *dirty = (le32_to_cpu(di->id1.journal1.ij_flags) &
394 OCFS2_JOURNAL_DIRTY_FL);
396 journal->j_journal = j_journal;
397 journal->j_inode = inode;
398 journal->j_bh = bh;
400 ocfs2_set_journal_params(osb);
402 journal->j_state = OCFS2_JOURNAL_LOADED;
404 status = 0;
405 done:
406 if (status < 0) {
407 if (meta_lock)
408 ocfs2_meta_unlock(inode, 1);
409 if (bh != NULL)
410 brelse(bh);
411 if (inode) {
412 OCFS2_I(inode)->ip_open_count--;
413 iput(inode);
417 mlog_exit(status);
418 return status;
421 static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
422 int dirty)
424 int status;
425 unsigned int flags;
426 struct ocfs2_journal *journal = osb->journal;
427 struct buffer_head *bh = journal->j_bh;
428 struct ocfs2_dinode *fe;
430 mlog_entry_void();
432 fe = (struct ocfs2_dinode *)bh->b_data;
433 if (!OCFS2_IS_VALID_DINODE(fe)) {
434 /* This is called from startup/shutdown which will
435 * handle the errors in a specific manner, so no need
436 * to call ocfs2_error() here. */
437 mlog(ML_ERROR, "Journal dinode %llu has invalid "
438 "signature: %.*s",
439 (unsigned long long)le64_to_cpu(fe->i_blkno), 7,
440 fe->i_signature);
441 status = -EIO;
442 goto out;
445 flags = le32_to_cpu(fe->id1.journal1.ij_flags);
446 if (dirty)
447 flags |= OCFS2_JOURNAL_DIRTY_FL;
448 else
449 flags &= ~OCFS2_JOURNAL_DIRTY_FL;
450 fe->id1.journal1.ij_flags = cpu_to_le32(flags);
452 status = ocfs2_write_block(osb, bh, journal->j_inode);
453 if (status < 0)
454 mlog_errno(status);
456 out:
457 mlog_exit(status);
458 return status;
462 * If the journal has been kmalloc'd it needs to be freed after this
463 * call.
465 void ocfs2_journal_shutdown(struct ocfs2_super *osb)
467 struct ocfs2_journal *journal = NULL;
468 int status = 0;
469 struct inode *inode = NULL;
470 int num_running_trans = 0;
472 mlog_entry_void();
474 BUG_ON(!osb);
476 journal = osb->journal;
477 if (!journal)
478 goto done;
480 inode = journal->j_inode;
482 if (journal->j_state != OCFS2_JOURNAL_LOADED)
483 goto done;
485 /* need to inc inode use count as journal_destroy will iput. */
486 if (!igrab(inode))
487 BUG();
489 num_running_trans = atomic_read(&(osb->journal->j_num_trans));
490 if (num_running_trans > 0)
491 mlog(0, "Shutting down journal: must wait on %d "
492 "running transactions!\n",
493 num_running_trans);
495 /* Do a commit_cache here. It will flush our journal, *and*
496 * release any locks that are still held.
497 * set the SHUTDOWN flag and release the trans lock.
498 * the commit thread will take the trans lock for us below. */
499 journal->j_state = OCFS2_JOURNAL_IN_SHUTDOWN;
501 /* The OCFS2_JOURNAL_IN_SHUTDOWN will signal to commit_cache to not
502 * drop the trans_lock (which we want to hold until we
503 * completely destroy the journal. */
504 if (osb->commit_task) {
505 /* Wait for the commit thread */
506 mlog(0, "Waiting for ocfs2commit to exit....\n");
507 kthread_stop(osb->commit_task);
508 osb->commit_task = NULL;
511 BUG_ON(atomic_read(&(osb->journal->j_num_trans)) != 0);
513 if (ocfs2_mount_local(osb)) {
514 journal_lock_updates(journal->j_journal);
515 status = journal_flush(journal->j_journal);
516 journal_unlock_updates(journal->j_journal);
517 if (status < 0)
518 mlog_errno(status);
521 if (status == 0) {
523 * Do not toggle if flush was unsuccessful otherwise
524 * will leave dirty metadata in a "clean" journal
526 status = ocfs2_journal_toggle_dirty(osb, 0);
527 if (status < 0)
528 mlog_errno(status);
531 /* Shutdown the kernel journal system */
532 journal_destroy(journal->j_journal);
534 OCFS2_I(inode)->ip_open_count--;
536 /* unlock our journal */
537 ocfs2_meta_unlock(inode, 1);
539 brelse(journal->j_bh);
540 journal->j_bh = NULL;
542 journal->j_state = OCFS2_JOURNAL_FREE;
544 // up_write(&journal->j_trans_barrier);
545 done:
546 if (inode)
547 iput(inode);
548 mlog_exit_void();
551 static void ocfs2_clear_journal_error(struct super_block *sb,
552 journal_t *journal,
553 int slot)
555 int olderr;
557 olderr = journal_errno(journal);
558 if (olderr) {
559 mlog(ML_ERROR, "File system error %d recorded in "
560 "journal %u.\n", olderr, slot);
561 mlog(ML_ERROR, "File system on device %s needs checking.\n",
562 sb->s_id);
564 journal_ack_err(journal);
565 journal_clear_err(journal);
569 int ocfs2_journal_load(struct ocfs2_journal *journal, int local)
571 int status = 0;
572 struct ocfs2_super *osb;
574 mlog_entry_void();
576 if (!journal)
577 BUG();
579 osb = journal->j_osb;
581 status = journal_load(journal->j_journal);
582 if (status < 0) {
583 mlog(ML_ERROR, "Failed to load journal!\n");
584 goto done;
587 ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num);
589 status = ocfs2_journal_toggle_dirty(osb, 1);
590 if (status < 0) {
591 mlog_errno(status);
592 goto done;
595 /* Launch the commit thread */
596 if (!local) {
597 osb->commit_task = kthread_run(ocfs2_commit_thread, osb,
598 "ocfs2cmt");
599 if (IS_ERR(osb->commit_task)) {
600 status = PTR_ERR(osb->commit_task);
601 osb->commit_task = NULL;
602 mlog(ML_ERROR, "unable to launch ocfs2commit thread, "
603 "error=%d", status);
604 goto done;
606 } else
607 osb->commit_task = NULL;
609 done:
610 mlog_exit(status);
611 return status;
615 /* 'full' flag tells us whether we clear out all blocks or if we just
616 * mark the journal clean */
617 int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full)
619 int status;
621 mlog_entry_void();
623 BUG_ON(!journal);
625 status = journal_wipe(journal->j_journal, full);
626 if (status < 0) {
627 mlog_errno(status);
628 goto bail;
631 status = ocfs2_journal_toggle_dirty(journal->j_osb, 0);
632 if (status < 0)
633 mlog_errno(status);
635 bail:
636 mlog_exit(status);
637 return status;
641 * JBD Might read a cached version of another nodes journal file. We
642 * don't want this as this file changes often and we get no
643 * notification on those changes. The only way to be sure that we've
644 * got the most up to date version of those blocks then is to force
645 * read them off disk. Just searching through the buffer cache won't
646 * work as there may be pages backing this file which are still marked
647 * up to date. We know things can't change on this file underneath us
648 * as we have the lock by now :)
650 static int ocfs2_force_read_journal(struct inode *inode)
652 int status = 0;
653 int i;
654 u64 v_blkno, p_blkno, p_blocks, num_blocks;
655 #define CONCURRENT_JOURNAL_FILL 32ULL
656 struct buffer_head *bhs[CONCURRENT_JOURNAL_FILL];
658 mlog_entry_void();
660 memset(bhs, 0, sizeof(struct buffer_head *) * CONCURRENT_JOURNAL_FILL);
662 num_blocks = ocfs2_blocks_for_bytes(inode->i_sb, inode->i_size);
663 v_blkno = 0;
664 while (v_blkno < num_blocks) {
665 status = ocfs2_extent_map_get_blocks(inode, v_blkno,
666 &p_blkno, &p_blocks, NULL);
667 if (status < 0) {
668 mlog_errno(status);
669 goto bail;
672 if (p_blocks > CONCURRENT_JOURNAL_FILL)
673 p_blocks = CONCURRENT_JOURNAL_FILL;
675 /* We are reading journal data which should not
676 * be put in the uptodate cache */
677 status = ocfs2_read_blocks(OCFS2_SB(inode->i_sb),
678 p_blkno, p_blocks, bhs, 0,
679 NULL);
680 if (status < 0) {
681 mlog_errno(status);
682 goto bail;
685 for(i = 0; i < p_blocks; i++) {
686 brelse(bhs[i]);
687 bhs[i] = NULL;
690 v_blkno += p_blocks;
693 bail:
694 for(i = 0; i < CONCURRENT_JOURNAL_FILL; i++)
695 if (bhs[i])
696 brelse(bhs[i]);
697 mlog_exit(status);
698 return status;
701 struct ocfs2_la_recovery_item {
702 struct list_head lri_list;
703 int lri_slot;
704 struct ocfs2_dinode *lri_la_dinode;
705 struct ocfs2_dinode *lri_tl_dinode;
708 /* Does the second half of the recovery process. By this point, the
709 * node is marked clean and can actually be considered recovered,
710 * hence it's no longer in the recovery map, but there's still some
711 * cleanup we can do which shouldn't happen within the recovery thread
712 * as locking in that context becomes very difficult if we are to take
713 * recovering nodes into account.
715 * NOTE: This function can and will sleep on recovery of other nodes
716 * during cluster locking, just like any other ocfs2 process.
718 void ocfs2_complete_recovery(struct work_struct *work)
720 int ret;
721 struct ocfs2_journal *journal =
722 container_of(work, struct ocfs2_journal, j_recovery_work);
723 struct ocfs2_super *osb = journal->j_osb;
724 struct ocfs2_dinode *la_dinode, *tl_dinode;
725 struct ocfs2_la_recovery_item *item, *n;
726 LIST_HEAD(tmp_la_list);
728 mlog_entry_void();
730 mlog(0, "completing recovery from keventd\n");
732 spin_lock(&journal->j_lock);
733 list_splice_init(&journal->j_la_cleanups, &tmp_la_list);
734 spin_unlock(&journal->j_lock);
736 list_for_each_entry_safe(item, n, &tmp_la_list, lri_list) {
737 list_del_init(&item->lri_list);
739 mlog(0, "Complete recovery for slot %d\n", item->lri_slot);
741 la_dinode = item->lri_la_dinode;
742 if (la_dinode) {
743 mlog(0, "Clean up local alloc %llu\n",
744 (unsigned long long)le64_to_cpu(la_dinode->i_blkno));
746 ret = ocfs2_complete_local_alloc_recovery(osb,
747 la_dinode);
748 if (ret < 0)
749 mlog_errno(ret);
751 kfree(la_dinode);
754 tl_dinode = item->lri_tl_dinode;
755 if (tl_dinode) {
756 mlog(0, "Clean up truncate log %llu\n",
757 (unsigned long long)le64_to_cpu(tl_dinode->i_blkno));
759 ret = ocfs2_complete_truncate_log_recovery(osb,
760 tl_dinode);
761 if (ret < 0)
762 mlog_errno(ret);
764 kfree(tl_dinode);
767 ret = ocfs2_recover_orphans(osb, item->lri_slot);
768 if (ret < 0)
769 mlog_errno(ret);
771 kfree(item);
774 mlog(0, "Recovery completion\n");
775 mlog_exit_void();
778 /* NOTE: This function always eats your references to la_dinode and
779 * tl_dinode, either manually on error, or by passing them to
780 * ocfs2_complete_recovery */
781 static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal,
782 int slot_num,
783 struct ocfs2_dinode *la_dinode,
784 struct ocfs2_dinode *tl_dinode)
786 struct ocfs2_la_recovery_item *item;
788 item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS);
789 if (!item) {
790 /* Though we wish to avoid it, we are in fact safe in
791 * skipping local alloc cleanup as fsck.ocfs2 is more
792 * than capable of reclaiming unused space. */
793 if (la_dinode)
794 kfree(la_dinode);
796 if (tl_dinode)
797 kfree(tl_dinode);
799 mlog_errno(-ENOMEM);
800 return;
803 INIT_LIST_HEAD(&item->lri_list);
804 item->lri_la_dinode = la_dinode;
805 item->lri_slot = slot_num;
806 item->lri_tl_dinode = tl_dinode;
808 spin_lock(&journal->j_lock);
809 list_add_tail(&item->lri_list, &journal->j_la_cleanups);
810 queue_work(ocfs2_wq, &journal->j_recovery_work);
811 spin_unlock(&journal->j_lock);
814 /* Called by the mount code to queue recovery the last part of
815 * recovery for it's own slot. */
816 void ocfs2_complete_mount_recovery(struct ocfs2_super *osb)
818 struct ocfs2_journal *journal = osb->journal;
820 if (osb->dirty) {
821 /* No need to queue up our truncate_log as regular
822 * cleanup will catch that. */
823 ocfs2_queue_recovery_completion(journal,
824 osb->slot_num,
825 osb->local_alloc_copy,
826 NULL);
827 ocfs2_schedule_truncate_log_flush(osb, 0);
829 osb->local_alloc_copy = NULL;
830 osb->dirty = 0;
834 static int __ocfs2_recovery_thread(void *arg)
836 int status, node_num;
837 struct ocfs2_super *osb = arg;
839 mlog_entry_void();
841 status = ocfs2_wait_on_mount(osb);
842 if (status < 0) {
843 goto bail;
846 restart:
847 status = ocfs2_super_lock(osb, 1);
848 if (status < 0) {
849 mlog_errno(status);
850 goto bail;
853 while(!ocfs2_node_map_is_empty(osb, &osb->recovery_map)) {
854 node_num = ocfs2_node_map_first_set_bit(osb,
855 &osb->recovery_map);
856 if (node_num == O2NM_INVALID_NODE_NUM) {
857 mlog(0, "Out of nodes to recover.\n");
858 break;
861 status = ocfs2_recover_node(osb, node_num);
862 if (status < 0) {
863 mlog(ML_ERROR,
864 "Error %d recovering node %d on device (%u,%u)!\n",
865 status, node_num,
866 MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
867 mlog(ML_ERROR, "Volume requires unmount.\n");
868 continue;
871 ocfs2_recovery_map_clear(osb, node_num);
873 ocfs2_super_unlock(osb, 1);
875 /* We always run recovery on our own orphan dir - the dead
876 * node(s) may have voted "no" on an inode delete earlier. A
877 * revote is therefore required. */
878 ocfs2_queue_recovery_completion(osb->journal, osb->slot_num, NULL,
879 NULL);
881 bail:
882 mutex_lock(&osb->recovery_lock);
883 if (!status &&
884 !ocfs2_node_map_is_empty(osb, &osb->recovery_map)) {
885 mutex_unlock(&osb->recovery_lock);
886 goto restart;
889 osb->recovery_thread_task = NULL;
890 mb(); /* sync with ocfs2_recovery_thread_running */
891 wake_up(&osb->recovery_event);
893 mutex_unlock(&osb->recovery_lock);
895 mlog_exit(status);
896 /* no one is callint kthread_stop() for us so the kthread() api
897 * requires that we call do_exit(). And it isn't exported, but
898 * complete_and_exit() seems to be a minimal wrapper around it. */
899 complete_and_exit(NULL, status);
900 return status;
903 void ocfs2_recovery_thread(struct ocfs2_super *osb, int node_num)
905 mlog_entry("(node_num=%d, osb->node_num = %d)\n",
906 node_num, osb->node_num);
908 mutex_lock(&osb->recovery_lock);
909 if (osb->disable_recovery)
910 goto out;
912 /* People waiting on recovery will wait on
913 * the recovery map to empty. */
914 if (!ocfs2_recovery_map_set(osb, node_num))
915 mlog(0, "node %d already be in recovery.\n", node_num);
917 mlog(0, "starting recovery thread...\n");
919 if (osb->recovery_thread_task)
920 goto out;
922 osb->recovery_thread_task = kthread_run(__ocfs2_recovery_thread, osb,
923 "ocfs2rec");
924 if (IS_ERR(osb->recovery_thread_task)) {
925 mlog_errno((int)PTR_ERR(osb->recovery_thread_task));
926 osb->recovery_thread_task = NULL;
929 out:
930 mutex_unlock(&osb->recovery_lock);
931 wake_up(&osb->recovery_event);
933 mlog_exit_void();
936 /* Does the actual journal replay and marks the journal inode as
937 * clean. Will only replay if the journal inode is marked dirty. */
938 static int ocfs2_replay_journal(struct ocfs2_super *osb,
939 int node_num,
940 int slot_num)
942 int status;
943 int got_lock = 0;
944 unsigned int flags;
945 struct inode *inode = NULL;
946 struct ocfs2_dinode *fe;
947 journal_t *journal = NULL;
948 struct buffer_head *bh = NULL;
950 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
951 slot_num);
952 if (inode == NULL) {
953 status = -EACCES;
954 mlog_errno(status);
955 goto done;
957 if (is_bad_inode(inode)) {
958 status = -EACCES;
959 iput(inode);
960 inode = NULL;
961 mlog_errno(status);
962 goto done;
964 SET_INODE_JOURNAL(inode);
966 status = ocfs2_meta_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY);
967 if (status < 0) {
968 mlog(0, "status returned from ocfs2_meta_lock=%d\n", status);
969 if (status != -ERESTARTSYS)
970 mlog(ML_ERROR, "Could not lock journal!\n");
971 goto done;
973 got_lock = 1;
975 fe = (struct ocfs2_dinode *) bh->b_data;
977 flags = le32_to_cpu(fe->id1.journal1.ij_flags);
979 if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) {
980 mlog(0, "No recovery required for node %d\n", node_num);
981 goto done;
984 mlog(ML_NOTICE, "Recovering node %d from slot %d on device (%u,%u)\n",
985 node_num, slot_num,
986 MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
988 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
990 status = ocfs2_force_read_journal(inode);
991 if (status < 0) {
992 mlog_errno(status);
993 goto done;
996 mlog(0, "calling journal_init_inode\n");
997 journal = journal_init_inode(inode);
998 if (journal == NULL) {
999 mlog(ML_ERROR, "Linux journal layer error\n");
1000 status = -EIO;
1001 goto done;
1004 status = journal_load(journal);
1005 if (status < 0) {
1006 mlog_errno(status);
1007 if (!igrab(inode))
1008 BUG();
1009 journal_destroy(journal);
1010 goto done;
1013 ocfs2_clear_journal_error(osb->sb, journal, slot_num);
1015 /* wipe the journal */
1016 mlog(0, "flushing the journal.\n");
1017 journal_lock_updates(journal);
1018 status = journal_flush(journal);
1019 journal_unlock_updates(journal);
1020 if (status < 0)
1021 mlog_errno(status);
1023 /* This will mark the node clean */
1024 flags = le32_to_cpu(fe->id1.journal1.ij_flags);
1025 flags &= ~OCFS2_JOURNAL_DIRTY_FL;
1026 fe->id1.journal1.ij_flags = cpu_to_le32(flags);
1028 status = ocfs2_write_block(osb, bh, inode);
1029 if (status < 0)
1030 mlog_errno(status);
1032 if (!igrab(inode))
1033 BUG();
1035 journal_destroy(journal);
1037 done:
1038 /* drop the lock on this nodes journal */
1039 if (got_lock)
1040 ocfs2_meta_unlock(inode, 1);
1042 if (inode)
1043 iput(inode);
1045 if (bh)
1046 brelse(bh);
1048 mlog_exit(status);
1049 return status;
1053 * Do the most important parts of node recovery:
1054 * - Replay it's journal
1055 * - Stamp a clean local allocator file
1056 * - Stamp a clean truncate log
1057 * - Mark the node clean
1059 * If this function completes without error, a node in OCFS2 can be
1060 * said to have been safely recovered. As a result, failure during the
1061 * second part of a nodes recovery process (local alloc recovery) is
1062 * far less concerning.
1064 static int ocfs2_recover_node(struct ocfs2_super *osb,
1065 int node_num)
1067 int status = 0;
1068 int slot_num;
1069 struct ocfs2_slot_info *si = osb->slot_info;
1070 struct ocfs2_dinode *la_copy = NULL;
1071 struct ocfs2_dinode *tl_copy = NULL;
1073 mlog_entry("(node_num=%d, osb->node_num = %d)\n",
1074 node_num, osb->node_num);
1076 mlog(0, "checking node %d\n", node_num);
1078 /* Should not ever be called to recover ourselves -- in that
1079 * case we should've called ocfs2_journal_load instead. */
1080 BUG_ON(osb->node_num == node_num);
1082 slot_num = ocfs2_node_num_to_slot(si, node_num);
1083 if (slot_num == OCFS2_INVALID_SLOT) {
1084 status = 0;
1085 mlog(0, "no slot for this node, so no recovery required.\n");
1086 goto done;
1089 mlog(0, "node %d was using slot %d\n", node_num, slot_num);
1091 status = ocfs2_replay_journal(osb, node_num, slot_num);
1092 if (status < 0) {
1093 mlog_errno(status);
1094 goto done;
1097 /* Stamp a clean local alloc file AFTER recovering the journal... */
1098 status = ocfs2_begin_local_alloc_recovery(osb, slot_num, &la_copy);
1099 if (status < 0) {
1100 mlog_errno(status);
1101 goto done;
1104 /* An error from begin_truncate_log_recovery is not
1105 * serious enough to warrant halting the rest of
1106 * recovery. */
1107 status = ocfs2_begin_truncate_log_recovery(osb, slot_num, &tl_copy);
1108 if (status < 0)
1109 mlog_errno(status);
1111 /* Likewise, this would be a strange but ultimately not so
1112 * harmful place to get an error... */
1113 ocfs2_clear_slot(si, slot_num);
1114 status = ocfs2_update_disk_slots(osb, si);
1115 if (status < 0)
1116 mlog_errno(status);
1118 /* This will kfree the memory pointed to by la_copy and tl_copy */
1119 ocfs2_queue_recovery_completion(osb->journal, slot_num, la_copy,
1120 tl_copy);
1122 status = 0;
1123 done:
1125 mlog_exit(status);
1126 return status;
1129 /* Test node liveness by trylocking his journal. If we get the lock,
1130 * we drop it here. Return 0 if we got the lock, -EAGAIN if node is
1131 * still alive (we couldn't get the lock) and < 0 on error. */
1132 static int ocfs2_trylock_journal(struct ocfs2_super *osb,
1133 int slot_num)
1135 int status, flags;
1136 struct inode *inode = NULL;
1138 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
1139 slot_num);
1140 if (inode == NULL) {
1141 mlog(ML_ERROR, "access error\n");
1142 status = -EACCES;
1143 goto bail;
1145 if (is_bad_inode(inode)) {
1146 mlog(ML_ERROR, "access error (bad inode)\n");
1147 iput(inode);
1148 inode = NULL;
1149 status = -EACCES;
1150 goto bail;
1152 SET_INODE_JOURNAL(inode);
1154 flags = OCFS2_META_LOCK_RECOVERY | OCFS2_META_LOCK_NOQUEUE;
1155 status = ocfs2_meta_lock_full(inode, NULL, 1, flags);
1156 if (status < 0) {
1157 if (status != -EAGAIN)
1158 mlog_errno(status);
1159 goto bail;
1162 ocfs2_meta_unlock(inode, 1);
1163 bail:
1164 if (inode)
1165 iput(inode);
1167 return status;
1170 /* Call this underneath ocfs2_super_lock. It also assumes that the
1171 * slot info struct has been updated from disk. */
1172 int ocfs2_mark_dead_nodes(struct ocfs2_super *osb)
1174 int status, i, node_num;
1175 struct ocfs2_slot_info *si = osb->slot_info;
1177 /* This is called with the super block cluster lock, so we
1178 * know that the slot map can't change underneath us. */
1180 spin_lock(&si->si_lock);
1181 for(i = 0; i < si->si_num_slots; i++) {
1182 if (i == osb->slot_num)
1183 continue;
1184 if (ocfs2_is_empty_slot(si, i))
1185 continue;
1187 node_num = si->si_global_node_nums[i];
1188 if (ocfs2_node_map_test_bit(osb, &osb->recovery_map, node_num))
1189 continue;
1190 spin_unlock(&si->si_lock);
1192 /* Ok, we have a slot occupied by another node which
1193 * is not in the recovery map. We trylock his journal
1194 * file here to test if he's alive. */
1195 status = ocfs2_trylock_journal(osb, i);
1196 if (!status) {
1197 /* Since we're called from mount, we know that
1198 * the recovery thread can't race us on
1199 * setting / checking the recovery bits. */
1200 ocfs2_recovery_thread(osb, node_num);
1201 } else if ((status < 0) && (status != -EAGAIN)) {
1202 mlog_errno(status);
1203 goto bail;
1206 spin_lock(&si->si_lock);
1208 spin_unlock(&si->si_lock);
1210 status = 0;
1211 bail:
1212 mlog_exit(status);
1213 return status;
1216 struct ocfs2_orphan_filldir_priv {
1217 struct inode *head;
1218 struct ocfs2_super *osb;
1221 static int ocfs2_orphan_filldir(void *priv, const char *name, int name_len,
1222 loff_t pos, u64 ino, unsigned type)
1224 struct ocfs2_orphan_filldir_priv *p = priv;
1225 struct inode *iter;
1227 if (name_len == 1 && !strncmp(".", name, 1))
1228 return 0;
1229 if (name_len == 2 && !strncmp("..", name, 2))
1230 return 0;
1232 /* Skip bad inodes so that recovery can continue */
1233 iter = ocfs2_iget(p->osb, ino,
1234 OCFS2_FI_FLAG_ORPHAN_RECOVERY);
1235 if (IS_ERR(iter))
1236 return 0;
1238 mlog(0, "queue orphan %llu\n",
1239 (unsigned long long)OCFS2_I(iter)->ip_blkno);
1240 /* No locking is required for the next_orphan queue as there
1241 * is only ever a single process doing orphan recovery. */
1242 OCFS2_I(iter)->ip_next_orphan = p->head;
1243 p->head = iter;
1245 return 0;
1248 static int ocfs2_queue_orphans(struct ocfs2_super *osb,
1249 int slot,
1250 struct inode **head)
1252 int status;
1253 struct inode *orphan_dir_inode = NULL;
1254 struct ocfs2_orphan_filldir_priv priv;
1255 loff_t pos = 0;
1257 priv.osb = osb;
1258 priv.head = *head;
1260 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
1261 ORPHAN_DIR_SYSTEM_INODE,
1262 slot);
1263 if (!orphan_dir_inode) {
1264 status = -ENOENT;
1265 mlog_errno(status);
1266 return status;
1269 mutex_lock(&orphan_dir_inode->i_mutex);
1270 status = ocfs2_meta_lock(orphan_dir_inode, NULL, 0);
1271 if (status < 0) {
1272 mlog_errno(status);
1273 goto out;
1276 status = ocfs2_dir_foreach(orphan_dir_inode, &pos, &priv,
1277 ocfs2_orphan_filldir);
1278 if (status) {
1279 mlog_errno(status);
1280 goto out;
1283 *head = priv.head;
1285 ocfs2_meta_unlock(orphan_dir_inode, 0);
1286 out:
1287 mutex_unlock(&orphan_dir_inode->i_mutex);
1288 iput(orphan_dir_inode);
1289 return status;
1292 static int ocfs2_orphan_recovery_can_continue(struct ocfs2_super *osb,
1293 int slot)
1295 int ret;
1297 spin_lock(&osb->osb_lock);
1298 ret = !osb->osb_orphan_wipes[slot];
1299 spin_unlock(&osb->osb_lock);
1300 return ret;
1303 static void ocfs2_mark_recovering_orphan_dir(struct ocfs2_super *osb,
1304 int slot)
1306 spin_lock(&osb->osb_lock);
1307 /* Mark ourselves such that new processes in delete_inode()
1308 * know to quit early. */
1309 ocfs2_node_map_set_bit(osb, &osb->osb_recovering_orphan_dirs, slot);
1310 while (osb->osb_orphan_wipes[slot]) {
1311 /* If any processes are already in the middle of an
1312 * orphan wipe on this dir, then we need to wait for
1313 * them. */
1314 spin_unlock(&osb->osb_lock);
1315 wait_event_interruptible(osb->osb_wipe_event,
1316 ocfs2_orphan_recovery_can_continue(osb, slot));
1317 spin_lock(&osb->osb_lock);
1319 spin_unlock(&osb->osb_lock);
1322 static void ocfs2_clear_recovering_orphan_dir(struct ocfs2_super *osb,
1323 int slot)
1325 ocfs2_node_map_clear_bit(osb, &osb->osb_recovering_orphan_dirs, slot);
1329 * Orphan recovery. Each mounted node has it's own orphan dir which we
1330 * must run during recovery. Our strategy here is to build a list of
1331 * the inodes in the orphan dir and iget/iput them. The VFS does
1332 * (most) of the rest of the work.
1334 * Orphan recovery can happen at any time, not just mount so we have a
1335 * couple of extra considerations.
1337 * - We grab as many inodes as we can under the orphan dir lock -
1338 * doing iget() outside the orphan dir risks getting a reference on
1339 * an invalid inode.
1340 * - We must be sure not to deadlock with other processes on the
1341 * system wanting to run delete_inode(). This can happen when they go
1342 * to lock the orphan dir and the orphan recovery process attempts to
1343 * iget() inside the orphan dir lock. This can be avoided by
1344 * advertising our state to ocfs2_delete_inode().
1346 static int ocfs2_recover_orphans(struct ocfs2_super *osb,
1347 int slot)
1349 int ret = 0;
1350 struct inode *inode = NULL;
1351 struct inode *iter;
1352 struct ocfs2_inode_info *oi;
1354 mlog(0, "Recover inodes from orphan dir in slot %d\n", slot);
1356 ocfs2_mark_recovering_orphan_dir(osb, slot);
1357 ret = ocfs2_queue_orphans(osb, slot, &inode);
1358 ocfs2_clear_recovering_orphan_dir(osb, slot);
1360 /* Error here should be noted, but we want to continue with as
1361 * many queued inodes as we've got. */
1362 if (ret)
1363 mlog_errno(ret);
1365 while (inode) {
1366 oi = OCFS2_I(inode);
1367 mlog(0, "iput orphan %llu\n", (unsigned long long)oi->ip_blkno);
1369 iter = oi->ip_next_orphan;
1371 spin_lock(&oi->ip_lock);
1372 /* Delete voting may have set these on the assumption
1373 * that the other node would wipe them successfully.
1374 * If they are still in the node's orphan dir, we need
1375 * to reset that state. */
1376 oi->ip_flags &= ~(OCFS2_INODE_DELETED|OCFS2_INODE_SKIP_DELETE);
1378 /* Set the proper information to get us going into
1379 * ocfs2_delete_inode. */
1380 oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED;
1381 spin_unlock(&oi->ip_lock);
1383 iput(inode);
1385 inode = iter;
1388 return ret;
1391 static int ocfs2_wait_on_mount(struct ocfs2_super *osb)
1393 /* This check is good because ocfs2 will wait on our recovery
1394 * thread before changing it to something other than MOUNTED
1395 * or DISABLED. */
1396 wait_event(osb->osb_mount_event,
1397 atomic_read(&osb->vol_state) == VOLUME_MOUNTED ||
1398 atomic_read(&osb->vol_state) == VOLUME_DISABLED);
1400 /* If there's an error on mount, then we may never get to the
1401 * MOUNTED flag, but this is set right before
1402 * dismount_volume() so we can trust it. */
1403 if (atomic_read(&osb->vol_state) == VOLUME_DISABLED) {
1404 mlog(0, "mount error, exiting!\n");
1405 return -EBUSY;
1408 return 0;
1411 static int ocfs2_commit_thread(void *arg)
1413 int status;
1414 struct ocfs2_super *osb = arg;
1415 struct ocfs2_journal *journal = osb->journal;
1417 /* we can trust j_num_trans here because _should_stop() is only set in
1418 * shutdown and nobody other than ourselves should be able to start
1419 * transactions. committing on shutdown might take a few iterations
1420 * as final transactions put deleted inodes on the list */
1421 while (!(kthread_should_stop() &&
1422 atomic_read(&journal->j_num_trans) == 0)) {
1424 wait_event_interruptible(osb->checkpoint_event,
1425 atomic_read(&journal->j_num_trans)
1426 || kthread_should_stop());
1428 status = ocfs2_commit_cache(osb);
1429 if (status < 0)
1430 mlog_errno(status);
1432 if (kthread_should_stop() && atomic_read(&journal->j_num_trans)){
1433 mlog(ML_KTHREAD,
1434 "commit_thread: %u transactions pending on "
1435 "shutdown\n",
1436 atomic_read(&journal->j_num_trans));
1440 return 0;
1443 /* Look for a dirty journal without taking any cluster locks. Used for
1444 * hard readonly access to determine whether the file system journals
1445 * require recovery. */
1446 int ocfs2_check_journals_nolocks(struct ocfs2_super *osb)
1448 int ret = 0;
1449 unsigned int slot;
1450 struct buffer_head *di_bh;
1451 struct ocfs2_dinode *di;
1452 struct inode *journal = NULL;
1454 for(slot = 0; slot < osb->max_slots; slot++) {
1455 journal = ocfs2_get_system_file_inode(osb,
1456 JOURNAL_SYSTEM_INODE,
1457 slot);
1458 if (!journal || is_bad_inode(journal)) {
1459 ret = -EACCES;
1460 mlog_errno(ret);
1461 goto out;
1464 di_bh = NULL;
1465 ret = ocfs2_read_block(osb, OCFS2_I(journal)->ip_blkno, &di_bh,
1466 0, journal);
1467 if (ret < 0) {
1468 mlog_errno(ret);
1469 goto out;
1472 di = (struct ocfs2_dinode *) di_bh->b_data;
1474 if (le32_to_cpu(di->id1.journal1.ij_flags) &
1475 OCFS2_JOURNAL_DIRTY_FL)
1476 ret = -EROFS;
1478 brelse(di_bh);
1479 if (ret)
1480 break;
1483 out:
1484 if (journal)
1485 iput(journal);
1487 return ret;