blob: c47bc2a809c2493703563f7f88e2b4f8a5bf1f9c [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"
Mark Fasheh316f4b92007-09-07 18:21:26 -070038#include "dir.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080039#include "dlmglue.h"
40#include "extent_map.h"
41#include "heartbeat.h"
42#include "inode.h"
43#include "journal.h"
44#include "localalloc.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080045#include "slot_map.h"
46#include "super.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080047#include "sysfile.h"
48
49#include "buffer_head_io.h"
50
Ingo Molnar34af9462006-06-27 02:53:55 -070051DEFINE_SPINLOCK(trans_inc_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -080052
53static int ocfs2_force_read_journal(struct inode *inode);
54static int ocfs2_recover_node(struct ocfs2_super *osb,
55 int node_num);
56static int __ocfs2_recovery_thread(void *arg);
57static int ocfs2_commit_cache(struct ocfs2_super *osb);
58static int ocfs2_wait_on_mount(struct ocfs2_super *osb);
Mark Fashehccd979b2005-12-15 14:31:24 -080059static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
Sunil Mushran539d8262008-07-14 17:31:10 -070060 int dirty, int replayed);
Mark Fashehccd979b2005-12-15 14:31:24 -080061static int ocfs2_trylock_journal(struct ocfs2_super *osb,
62 int slot_num);
63static int ocfs2_recover_orphans(struct ocfs2_super *osb,
64 int slot);
65static int ocfs2_commit_thread(void *arg);
66
Joel Becker553abd02008-02-01 12:03:57 -080067
68/*
69 * The recovery_list is a simple linked list of node numbers to recover.
70 * It is protected by the recovery_lock.
71 */
72
73struct ocfs2_recovery_map {
Joel Beckerfc881fa2008-02-01 12:04:48 -080074 unsigned int rm_used;
Joel Becker553abd02008-02-01 12:03:57 -080075 unsigned int *rm_entries;
76};
77
78int ocfs2_recovery_init(struct ocfs2_super *osb)
79{
80 struct ocfs2_recovery_map *rm;
81
82 mutex_init(&osb->recovery_lock);
83 osb->disable_recovery = 0;
84 osb->recovery_thread_task = NULL;
85 init_waitqueue_head(&osb->recovery_event);
86
87 rm = kzalloc(sizeof(struct ocfs2_recovery_map) +
88 osb->max_slots * sizeof(unsigned int),
89 GFP_KERNEL);
90 if (!rm) {
91 mlog_errno(-ENOMEM);
92 return -ENOMEM;
93 }
94
95 rm->rm_entries = (unsigned int *)((char *)rm +
96 sizeof(struct ocfs2_recovery_map));
97 osb->recovery_map = rm;
98
99 return 0;
100}
101
102/* we can't grab the goofy sem lock from inside wait_event, so we use
103 * memory barriers to make sure that we'll see the null task before
104 * being woken up */
105static int ocfs2_recovery_thread_running(struct ocfs2_super *osb)
106{
107 mb();
108 return osb->recovery_thread_task != NULL;
109}
110
111void ocfs2_recovery_exit(struct ocfs2_super *osb)
112{
113 struct ocfs2_recovery_map *rm;
114
115 /* disable any new recovery threads and wait for any currently
116 * running ones to exit. Do this before setting the vol_state. */
117 mutex_lock(&osb->recovery_lock);
118 osb->disable_recovery = 1;
119 mutex_unlock(&osb->recovery_lock);
120 wait_event(osb->recovery_event, !ocfs2_recovery_thread_running(osb));
121
122 /* At this point, we know that no more recovery threads can be
123 * launched, so wait for any recovery completion work to
124 * complete. */
125 flush_workqueue(ocfs2_wq);
126
127 /*
128 * Now that recovery is shut down, and the osb is about to be
129 * freed, the osb_lock is not taken here.
130 */
131 rm = osb->recovery_map;
132 /* XXX: Should we bug if there are dirty entries? */
133
134 kfree(rm);
135}
136
137static int __ocfs2_recovery_map_test(struct ocfs2_super *osb,
138 unsigned int node_num)
139{
140 int i;
141 struct ocfs2_recovery_map *rm = osb->recovery_map;
142
143 assert_spin_locked(&osb->osb_lock);
144
145 for (i = 0; i < rm->rm_used; i++) {
146 if (rm->rm_entries[i] == node_num)
147 return 1;
148 }
149
150 return 0;
151}
152
153/* Behaves like test-and-set. Returns the previous value */
154static int ocfs2_recovery_map_set(struct ocfs2_super *osb,
155 unsigned int node_num)
156{
157 struct ocfs2_recovery_map *rm = osb->recovery_map;
158
159 spin_lock(&osb->osb_lock);
160 if (__ocfs2_recovery_map_test(osb, node_num)) {
161 spin_unlock(&osb->osb_lock);
162 return 1;
163 }
164
165 /* XXX: Can this be exploited? Not from o2dlm... */
166 BUG_ON(rm->rm_used >= osb->max_slots);
167
168 rm->rm_entries[rm->rm_used] = node_num;
169 rm->rm_used++;
170 spin_unlock(&osb->osb_lock);
171
172 return 0;
173}
174
175static void ocfs2_recovery_map_clear(struct ocfs2_super *osb,
176 unsigned int node_num)
177{
178 int i;
179 struct ocfs2_recovery_map *rm = osb->recovery_map;
180
181 spin_lock(&osb->osb_lock);
182
183 for (i = 0; i < rm->rm_used; i++) {
184 if (rm->rm_entries[i] == node_num)
185 break;
186 }
187
188 if (i < rm->rm_used) {
189 /* XXX: be careful with the pointer math */
190 memmove(&(rm->rm_entries[i]), &(rm->rm_entries[i + 1]),
191 (rm->rm_used - i - 1) * sizeof(unsigned int));
192 rm->rm_used--;
193 }
194
195 spin_unlock(&osb->osb_lock);
196}
197
Mark Fashehccd979b2005-12-15 14:31:24 -0800198static int ocfs2_commit_cache(struct ocfs2_super *osb)
199{
200 int status = 0;
201 unsigned int flushed;
202 unsigned long old_id;
203 struct ocfs2_journal *journal = NULL;
204
205 mlog_entry_void();
206
207 journal = osb->journal;
208
209 /* Flush all pending commits and checkpoint the journal. */
210 down_write(&journal->j_trans_barrier);
211
212 if (atomic_read(&journal->j_num_trans) == 0) {
213 up_write(&journal->j_trans_barrier);
214 mlog(0, "No transactions for me to flush!\n");
215 goto finally;
216 }
217
218 journal_lock_updates(journal->j_journal);
219 status = journal_flush(journal->j_journal);
220 journal_unlock_updates(journal->j_journal);
221 if (status < 0) {
222 up_write(&journal->j_trans_barrier);
223 mlog_errno(status);
224 goto finally;
225 }
226
227 old_id = ocfs2_inc_trans_id(journal);
228
229 flushed = atomic_read(&journal->j_num_trans);
230 atomic_set(&journal->j_num_trans, 0);
231 up_write(&journal->j_trans_barrier);
232
233 mlog(0, "commit_thread: flushed transaction %lu (%u handles)\n",
234 journal->j_trans_id, flushed);
235
Mark Fasheh34d024f2007-09-24 15:56:19 -0700236 ocfs2_wake_downconvert_thread(osb);
Mark Fashehccd979b2005-12-15 14:31:24 -0800237 wake_up(&journal->j_checkpointed);
238finally:
239 mlog_exit(status);
240 return status;
241}
242
Mark Fashehccd979b2005-12-15 14:31:24 -0800243/* pass it NULL and it will allocate a new handle object for you. If
244 * you pass it a handle however, it may still return error, in which
245 * case it has free'd the passed handle for you. */
Mark Fasheh1fabe142006-10-09 18:11:45 -0700246handle_t *ocfs2_start_trans(struct ocfs2_super *osb, int max_buffs)
Mark Fashehccd979b2005-12-15 14:31:24 -0800247{
Mark Fashehccd979b2005-12-15 14:31:24 -0800248 journal_t *journal = osb->journal->j_journal;
Mark Fasheh1fabe142006-10-09 18:11:45 -0700249 handle_t *handle;
Mark Fashehccd979b2005-12-15 14:31:24 -0800250
Eric Sesterhenn / snakebyteebdec832006-01-27 10:32:52 +0100251 BUG_ON(!osb || !osb->journal->j_journal);
Mark Fashehccd979b2005-12-15 14:31:24 -0800252
Mark Fasheh65eff9c2006-10-09 17:26:22 -0700253 if (ocfs2_is_hard_readonly(osb))
254 return ERR_PTR(-EROFS);
Mark Fashehccd979b2005-12-15 14:31:24 -0800255
256 BUG_ON(osb->journal->j_state == OCFS2_JOURNAL_FREE);
257 BUG_ON(max_buffs <= 0);
258
259 /* JBD might support this, but our journalling code doesn't yet. */
260 if (journal_current_handle()) {
261 mlog(ML_ERROR, "Recursive transaction attempted!\n");
262 BUG();
263 }
264
Mark Fashehccd979b2005-12-15 14:31:24 -0800265 down_read(&osb->journal->j_trans_barrier);
266
Mark Fasheh1fabe142006-10-09 18:11:45 -0700267 handle = journal_start(journal, max_buffs);
268 if (IS_ERR(handle)) {
Mark Fashehccd979b2005-12-15 14:31:24 -0800269 up_read(&osb->journal->j_trans_barrier);
270
Mark Fasheh1fabe142006-10-09 18:11:45 -0700271 mlog_errno(PTR_ERR(handle));
Mark Fashehccd979b2005-12-15 14:31:24 -0800272
273 if (is_journal_aborted(journal)) {
274 ocfs2_abort(osb->sb, "Detected aborted journal");
Mark Fasheh1fabe142006-10-09 18:11:45 -0700275 handle = ERR_PTR(-EROFS);
Mark Fashehccd979b2005-12-15 14:31:24 -0800276 }
Sunil Mushranc271c5c2006-12-05 17:56:35 -0800277 } else {
278 if (!ocfs2_mount_local(osb))
279 atomic_inc(&(osb->journal->j_num_trans));
280 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800281
Mark Fashehccd979b2005-12-15 14:31:24 -0800282 return handle;
Mark Fashehccd979b2005-12-15 14:31:24 -0800283}
284
Mark Fasheh1fabe142006-10-09 18:11:45 -0700285int ocfs2_commit_trans(struct ocfs2_super *osb,
286 handle_t *handle)
Mark Fashehccd979b2005-12-15 14:31:24 -0800287{
Mark Fasheh1fabe142006-10-09 18:11:45 -0700288 int ret;
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700289 struct ocfs2_journal *journal = osb->journal;
Mark Fashehccd979b2005-12-15 14:31:24 -0800290
291 BUG_ON(!handle);
292
Mark Fasheh1fabe142006-10-09 18:11:45 -0700293 ret = journal_stop(handle);
294 if (ret < 0)
295 mlog_errno(ret);
Mark Fashehccd979b2005-12-15 14:31:24 -0800296
297 up_read(&journal->j_trans_barrier);
298
Mark Fasheh1fabe142006-10-09 18:11:45 -0700299 return ret;
Mark Fashehccd979b2005-12-15 14:31:24 -0800300}
301
302/*
303 * 'nblocks' is what you want to add to the current
304 * transaction. extend_trans will either extend the current handle by
305 * nblocks, or commit it and start a new one with nblocks credits.
306 *
Mark Fashehe8aed342007-12-03 16:43:01 -0800307 * This might call journal_restart() which will commit dirty buffers
308 * and then restart the transaction. Before calling
309 * ocfs2_extend_trans(), any changed blocks should have been
310 * dirtied. After calling it, all blocks which need to be changed must
311 * go through another set of journal_access/journal_dirty calls.
312 *
Mark Fashehccd979b2005-12-15 14:31:24 -0800313 * WARNING: This will not release any semaphores or disk locks taken
314 * during the transaction, so make sure they were taken *before*
315 * start_trans or we'll have ordering deadlocks.
316 *
317 * WARNING2: Note that we do *not* drop j_trans_barrier here. This is
318 * good because transaction ids haven't yet been recorded on the
319 * cluster locks associated with this handle.
320 */
Mark Fasheh1fc58142006-10-05 14:15:36 -0700321int ocfs2_extend_trans(handle_t *handle, int nblocks)
Mark Fashehccd979b2005-12-15 14:31:24 -0800322{
323 int status;
324
325 BUG_ON(!handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800326 BUG_ON(!nblocks);
327
328 mlog_entry_void();
329
330 mlog(0, "Trying to extend transaction by %d blocks\n", nblocks);
331
Joel Beckere407e392008-06-12 22:35:39 -0700332#ifdef CONFIG_OCFS2_DEBUG_FS
Mark Fasheh0879c582007-12-03 16:42:19 -0800333 status = 1;
334#else
Mark Fasheh1fc58142006-10-05 14:15:36 -0700335 status = journal_extend(handle, nblocks);
Mark Fashehccd979b2005-12-15 14:31:24 -0800336 if (status < 0) {
337 mlog_errno(status);
338 goto bail;
339 }
Mark Fasheh0879c582007-12-03 16:42:19 -0800340#endif
Mark Fashehccd979b2005-12-15 14:31:24 -0800341
342 if (status > 0) {
343 mlog(0, "journal_extend failed, trying journal_restart\n");
Mark Fasheh1fc58142006-10-05 14:15:36 -0700344 status = journal_restart(handle, nblocks);
Mark Fashehccd979b2005-12-15 14:31:24 -0800345 if (status < 0) {
Mark Fashehccd979b2005-12-15 14:31:24 -0800346 mlog_errno(status);
347 goto bail;
348 }
Mark Fasheh01ddf1e2006-10-05 13:54:39 -0700349 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800350
351 status = 0;
352bail:
353
354 mlog_exit(status);
355 return status;
356}
357
Mark Fasheh1fabe142006-10-09 18:11:45 -0700358int ocfs2_journal_access(handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -0800359 struct inode *inode,
360 struct buffer_head *bh,
361 int type)
362{
363 int status;
364
365 BUG_ON(!inode);
366 BUG_ON(!handle);
367 BUG_ON(!bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800368
Badari Pulavarty205f87f2006-03-26 01:38:00 -0800369 mlog_entry("bh->b_blocknr=%llu, type=%d (\"%s\"), bh->b_size = %zu\n",
Mark Fashehccd979b2005-12-15 14:31:24 -0800370 (unsigned long long)bh->b_blocknr, type,
371 (type == OCFS2_JOURNAL_ACCESS_CREATE) ?
372 "OCFS2_JOURNAL_ACCESS_CREATE" :
373 "OCFS2_JOURNAL_ACCESS_WRITE",
374 bh->b_size);
375
376 /* we can safely remove this assertion after testing. */
377 if (!buffer_uptodate(bh)) {
378 mlog(ML_ERROR, "giving me a buffer that's not uptodate!\n");
379 mlog(ML_ERROR, "b_blocknr=%llu\n",
380 (unsigned long long)bh->b_blocknr);
381 BUG();
382 }
383
384 /* Set the current transaction information on the inode so
385 * that the locking code knows whether it can drop it's locks
386 * on this inode or not. We're protected from the commit
387 * thread updating the current transaction id until
388 * ocfs2_commit_trans() because ocfs2_start_trans() took
389 * j_trans_barrier for us. */
390 ocfs2_set_inode_lock_trans(OCFS2_SB(inode->i_sb)->journal, inode);
391
Mark Fasheh251b6ec2006-01-10 15:41:43 -0800392 mutex_lock(&OCFS2_I(inode)->ip_io_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800393 switch (type) {
394 case OCFS2_JOURNAL_ACCESS_CREATE:
395 case OCFS2_JOURNAL_ACCESS_WRITE:
Mark Fasheh1fabe142006-10-09 18:11:45 -0700396 status = journal_get_write_access(handle, bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800397 break;
398
399 case OCFS2_JOURNAL_ACCESS_UNDO:
Mark Fasheh1fabe142006-10-09 18:11:45 -0700400 status = journal_get_undo_access(handle, bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800401 break;
402
403 default:
404 status = -EINVAL;
405 mlog(ML_ERROR, "Uknown access type!\n");
406 }
Mark Fasheh251b6ec2006-01-10 15:41:43 -0800407 mutex_unlock(&OCFS2_I(inode)->ip_io_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800408
409 if (status < 0)
410 mlog(ML_ERROR, "Error %d getting %d access to buffer!\n",
411 status, type);
412
413 mlog_exit(status);
414 return status;
415}
416
Mark Fasheh1fabe142006-10-09 18:11:45 -0700417int ocfs2_journal_dirty(handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -0800418 struct buffer_head *bh)
419{
420 int status;
421
Mark Fashehccd979b2005-12-15 14:31:24 -0800422 mlog_entry("(bh->b_blocknr=%llu)\n",
423 (unsigned long long)bh->b_blocknr);
424
Mark Fasheh1fabe142006-10-09 18:11:45 -0700425 status = journal_dirty_metadata(handle, bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800426 if (status < 0)
427 mlog(ML_ERROR, "Could not dirty metadata buffer. "
428 "(bh->b_blocknr=%llu)\n",
429 (unsigned long long)bh->b_blocknr);
430
431 mlog_exit(status);
432 return status;
433}
434
435int ocfs2_journal_dirty_data(handle_t *handle,
436 struct buffer_head *bh)
437{
438 int err = journal_dirty_data(handle, bh);
439 if (err)
440 mlog_errno(err);
441 /* TODO: When we can handle it, abort the handle and go RO on
442 * error here. */
443
444 return err;
445}
446
Mark Fashehd147b3d2007-11-07 14:40:36 -0800447#define OCFS2_DEFAULT_COMMIT_INTERVAL (HZ * JBD_DEFAULT_MAX_COMMIT_AGE)
Mark Fashehccd979b2005-12-15 14:31:24 -0800448
449void ocfs2_set_journal_params(struct ocfs2_super *osb)
450{
451 journal_t *journal = osb->journal->j_journal;
Mark Fashehd147b3d2007-11-07 14:40:36 -0800452 unsigned long commit_interval = OCFS2_DEFAULT_COMMIT_INTERVAL;
453
454 if (osb->osb_commit_interval)
455 commit_interval = osb->osb_commit_interval;
Mark Fashehccd979b2005-12-15 14:31:24 -0800456
457 spin_lock(&journal->j_state_lock);
Mark Fashehd147b3d2007-11-07 14:40:36 -0800458 journal->j_commit_interval = commit_interval;
Mark Fashehccd979b2005-12-15 14:31:24 -0800459 if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER)
460 journal->j_flags |= JFS_BARRIER;
461 else
462 journal->j_flags &= ~JFS_BARRIER;
463 spin_unlock(&journal->j_state_lock);
464}
465
466int ocfs2_journal_init(struct ocfs2_journal *journal, int *dirty)
467{
468 int status = -1;
469 struct inode *inode = NULL; /* the journal inode */
470 journal_t *j_journal = NULL;
471 struct ocfs2_dinode *di = NULL;
472 struct buffer_head *bh = NULL;
473 struct ocfs2_super *osb;
Mark Fashehe63aecb62007-10-18 15:30:42 -0700474 int inode_lock = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -0800475
476 mlog_entry_void();
477
478 BUG_ON(!journal);
479
480 osb = journal->j_osb;
481
482 /* already have the inode for our journal */
483 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
484 osb->slot_num);
485 if (inode == NULL) {
486 status = -EACCES;
487 mlog_errno(status);
488 goto done;
489 }
490 if (is_bad_inode(inode)) {
491 mlog(ML_ERROR, "access error (bad inode)\n");
492 iput(inode);
493 inode = NULL;
494 status = -EACCES;
495 goto done;
496 }
497
498 SET_INODE_JOURNAL(inode);
499 OCFS2_I(inode)->ip_open_count++;
500
Mark Fasheh6eff5792006-01-18 10:31:47 -0800501 /* Skip recovery waits here - journal inode metadata never
502 * changes in a live cluster so it can be considered an
503 * exception to the rule. */
Mark Fashehe63aecb62007-10-18 15:30:42 -0700504 status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY);
Mark Fashehccd979b2005-12-15 14:31:24 -0800505 if (status < 0) {
506 if (status != -ERESTARTSYS)
507 mlog(ML_ERROR, "Could not get lock on journal!\n");
508 goto done;
509 }
510
Mark Fashehe63aecb62007-10-18 15:30:42 -0700511 inode_lock = 1;
Mark Fashehccd979b2005-12-15 14:31:24 -0800512 di = (struct ocfs2_dinode *)bh->b_data;
513
514 if (inode->i_size < OCFS2_MIN_JOURNAL_SIZE) {
515 mlog(ML_ERROR, "Journal file size (%lld) is too small!\n",
516 inode->i_size);
517 status = -EINVAL;
518 goto done;
519 }
520
521 mlog(0, "inode->i_size = %lld\n", inode->i_size);
Andrew Morton5515eff2006-03-26 01:37:53 -0800522 mlog(0, "inode->i_blocks = %llu\n",
523 (unsigned long long)inode->i_blocks);
Mark Fashehccd979b2005-12-15 14:31:24 -0800524 mlog(0, "inode->ip_clusters = %u\n", OCFS2_I(inode)->ip_clusters);
525
526 /* call the kernels journal init function now */
527 j_journal = journal_init_inode(inode);
528 if (j_journal == NULL) {
529 mlog(ML_ERROR, "Linux journal layer error\n");
530 status = -EINVAL;
531 goto done;
532 }
533
534 mlog(0, "Returned from journal_init_inode\n");
535 mlog(0, "j_journal->j_maxlen = %u\n", j_journal->j_maxlen);
536
537 *dirty = (le32_to_cpu(di->id1.journal1.ij_flags) &
538 OCFS2_JOURNAL_DIRTY_FL);
539
540 journal->j_journal = j_journal;
541 journal->j_inode = inode;
542 journal->j_bh = bh;
543
544 ocfs2_set_journal_params(osb);
545
546 journal->j_state = OCFS2_JOURNAL_LOADED;
547
548 status = 0;
549done:
550 if (status < 0) {
Mark Fashehe63aecb62007-10-18 15:30:42 -0700551 if (inode_lock)
552 ocfs2_inode_unlock(inode, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -0800553 if (bh != NULL)
554 brelse(bh);
555 if (inode) {
556 OCFS2_I(inode)->ip_open_count--;
557 iput(inode);
558 }
559 }
560
561 mlog_exit(status);
562 return status;
563}
564
Sunil Mushran539d8262008-07-14 17:31:10 -0700565static void ocfs2_bump_recovery_generation(struct ocfs2_dinode *di)
566{
567 le32_add_cpu(&(di->id1.journal1.ij_recovery_generation), 1);
568}
569
570static u32 ocfs2_get_recovery_generation(struct ocfs2_dinode *di)
571{
572 return le32_to_cpu(di->id1.journal1.ij_recovery_generation);
573}
574
Mark Fashehccd979b2005-12-15 14:31:24 -0800575static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
Sunil Mushran539d8262008-07-14 17:31:10 -0700576 int dirty, int replayed)
Mark Fashehccd979b2005-12-15 14:31:24 -0800577{
578 int status;
579 unsigned int flags;
580 struct ocfs2_journal *journal = osb->journal;
581 struct buffer_head *bh = journal->j_bh;
582 struct ocfs2_dinode *fe;
583
584 mlog_entry_void();
585
586 fe = (struct ocfs2_dinode *)bh->b_data;
587 if (!OCFS2_IS_VALID_DINODE(fe)) {
588 /* This is called from startup/shutdown which will
589 * handle the errors in a specific manner, so no need
590 * to call ocfs2_error() here. */
Mark Fashehb06970532006-03-03 10:24:33 -0800591 mlog(ML_ERROR, "Journal dinode %llu has invalid "
Mark Fasheh1ca1a112007-04-27 16:01:25 -0700592 "signature: %.*s",
593 (unsigned long long)le64_to_cpu(fe->i_blkno), 7,
Mark Fashehb06970532006-03-03 10:24:33 -0800594 fe->i_signature);
Mark Fashehccd979b2005-12-15 14:31:24 -0800595 status = -EIO;
596 goto out;
597 }
598
599 flags = le32_to_cpu(fe->id1.journal1.ij_flags);
600 if (dirty)
601 flags |= OCFS2_JOURNAL_DIRTY_FL;
602 else
603 flags &= ~OCFS2_JOURNAL_DIRTY_FL;
604 fe->id1.journal1.ij_flags = cpu_to_le32(flags);
605
Sunil Mushran539d8262008-07-14 17:31:10 -0700606 if (replayed)
607 ocfs2_bump_recovery_generation(fe);
608
Mark Fashehccd979b2005-12-15 14:31:24 -0800609 status = ocfs2_write_block(osb, bh, journal->j_inode);
610 if (status < 0)
611 mlog_errno(status);
612
613out:
614 mlog_exit(status);
615 return status;
616}
617
618/*
619 * If the journal has been kmalloc'd it needs to be freed after this
620 * call.
621 */
622void ocfs2_journal_shutdown(struct ocfs2_super *osb)
623{
624 struct ocfs2_journal *journal = NULL;
625 int status = 0;
626 struct inode *inode = NULL;
627 int num_running_trans = 0;
628
629 mlog_entry_void();
630
Eric Sesterhenn / snakebyteebdec832006-01-27 10:32:52 +0100631 BUG_ON(!osb);
Mark Fashehccd979b2005-12-15 14:31:24 -0800632
633 journal = osb->journal;
634 if (!journal)
635 goto done;
636
637 inode = journal->j_inode;
638
639 if (journal->j_state != OCFS2_JOURNAL_LOADED)
640 goto done;
641
642 /* need to inc inode use count as journal_destroy will iput. */
643 if (!igrab(inode))
644 BUG();
645
646 num_running_trans = atomic_read(&(osb->journal->j_num_trans));
647 if (num_running_trans > 0)
648 mlog(0, "Shutting down journal: must wait on %d "
649 "running transactions!\n",
650 num_running_trans);
651
652 /* Do a commit_cache here. It will flush our journal, *and*
653 * release any locks that are still held.
654 * set the SHUTDOWN flag and release the trans lock.
655 * the commit thread will take the trans lock for us below. */
656 journal->j_state = OCFS2_JOURNAL_IN_SHUTDOWN;
657
658 /* The OCFS2_JOURNAL_IN_SHUTDOWN will signal to commit_cache to not
659 * drop the trans_lock (which we want to hold until we
660 * completely destroy the journal. */
661 if (osb->commit_task) {
662 /* Wait for the commit thread */
663 mlog(0, "Waiting for ocfs2commit to exit....\n");
664 kthread_stop(osb->commit_task);
665 osb->commit_task = NULL;
666 }
667
668 BUG_ON(atomic_read(&(osb->journal->j_num_trans)) != 0);
669
Sunil Mushranc271c5c2006-12-05 17:56:35 -0800670 if (ocfs2_mount_local(osb)) {
671 journal_lock_updates(journal->j_journal);
672 status = journal_flush(journal->j_journal);
673 journal_unlock_updates(journal->j_journal);
674 if (status < 0)
675 mlog_errno(status);
676 }
677
678 if (status == 0) {
679 /*
680 * Do not toggle if flush was unsuccessful otherwise
681 * will leave dirty metadata in a "clean" journal
682 */
Sunil Mushran539d8262008-07-14 17:31:10 -0700683 status = ocfs2_journal_toggle_dirty(osb, 0, 0);
Sunil Mushranc271c5c2006-12-05 17:56:35 -0800684 if (status < 0)
685 mlog_errno(status);
686 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800687
688 /* Shutdown the kernel journal system */
689 journal_destroy(journal->j_journal);
690
691 OCFS2_I(inode)->ip_open_count--;
692
693 /* unlock our journal */
Mark Fashehe63aecb62007-10-18 15:30:42 -0700694 ocfs2_inode_unlock(inode, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -0800695
696 brelse(journal->j_bh);
697 journal->j_bh = NULL;
698
699 journal->j_state = OCFS2_JOURNAL_FREE;
700
701// up_write(&journal->j_trans_barrier);
702done:
703 if (inode)
704 iput(inode);
705 mlog_exit_void();
706}
707
708static void ocfs2_clear_journal_error(struct super_block *sb,
709 journal_t *journal,
710 int slot)
711{
712 int olderr;
713
714 olderr = journal_errno(journal);
715 if (olderr) {
716 mlog(ML_ERROR, "File system error %d recorded in "
717 "journal %u.\n", olderr, slot);
718 mlog(ML_ERROR, "File system on device %s needs checking.\n",
719 sb->s_id);
720
721 journal_ack_err(journal);
722 journal_clear_err(journal);
723 }
724}
725
Sunil Mushran539d8262008-07-14 17:31:10 -0700726int ocfs2_journal_load(struct ocfs2_journal *journal, int local, int replayed)
Mark Fashehccd979b2005-12-15 14:31:24 -0800727{
728 int status = 0;
729 struct ocfs2_super *osb;
730
731 mlog_entry_void();
732
Julia Lawallb1f35502008-03-04 15:21:05 -0800733 BUG_ON(!journal);
Mark Fashehccd979b2005-12-15 14:31:24 -0800734
735 osb = journal->j_osb;
736
737 status = journal_load(journal->j_journal);
738 if (status < 0) {
739 mlog(ML_ERROR, "Failed to load journal!\n");
740 goto done;
741 }
742
743 ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num);
744
Sunil Mushran539d8262008-07-14 17:31:10 -0700745 status = ocfs2_journal_toggle_dirty(osb, 1, replayed);
Mark Fashehccd979b2005-12-15 14:31:24 -0800746 if (status < 0) {
747 mlog_errno(status);
748 goto done;
749 }
750
751 /* Launch the commit thread */
Sunil Mushranc271c5c2006-12-05 17:56:35 -0800752 if (!local) {
753 osb->commit_task = kthread_run(ocfs2_commit_thread, osb,
754 "ocfs2cmt");
755 if (IS_ERR(osb->commit_task)) {
756 status = PTR_ERR(osb->commit_task);
757 osb->commit_task = NULL;
758 mlog(ML_ERROR, "unable to launch ocfs2commit thread, "
759 "error=%d", status);
760 goto done;
761 }
762 } else
Mark Fashehccd979b2005-12-15 14:31:24 -0800763 osb->commit_task = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800764
765done:
766 mlog_exit(status);
767 return status;
768}
769
770
771/* 'full' flag tells us whether we clear out all blocks or if we just
772 * mark the journal clean */
773int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full)
774{
775 int status;
776
777 mlog_entry_void();
778
Eric Sesterhenn / snakebyteebdec832006-01-27 10:32:52 +0100779 BUG_ON(!journal);
Mark Fashehccd979b2005-12-15 14:31:24 -0800780
781 status = journal_wipe(journal->j_journal, full);
782 if (status < 0) {
783 mlog_errno(status);
784 goto bail;
785 }
786
Sunil Mushran539d8262008-07-14 17:31:10 -0700787 status = ocfs2_journal_toggle_dirty(journal->j_osb, 0, 0);
Mark Fashehccd979b2005-12-15 14:31:24 -0800788 if (status < 0)
789 mlog_errno(status);
790
791bail:
792 mlog_exit(status);
793 return status;
794}
795
Joel Becker553abd02008-02-01 12:03:57 -0800796static int ocfs2_recovery_completed(struct ocfs2_super *osb)
797{
798 int empty;
799 struct ocfs2_recovery_map *rm = osb->recovery_map;
800
801 spin_lock(&osb->osb_lock);
802 empty = (rm->rm_used == 0);
803 spin_unlock(&osb->osb_lock);
804
805 return empty;
806}
807
808void ocfs2_wait_for_recovery(struct ocfs2_super *osb)
809{
810 wait_event(osb->recovery_event, ocfs2_recovery_completed(osb));
811}
812
Mark Fashehccd979b2005-12-15 14:31:24 -0800813/*
814 * JBD Might read a cached version of another nodes journal file. We
815 * don't want this as this file changes often and we get no
816 * notification on those changes. The only way to be sure that we've
817 * got the most up to date version of those blocks then is to force
818 * read them off disk. Just searching through the buffer cache won't
819 * work as there may be pages backing this file which are still marked
820 * up to date. We know things can't change on this file underneath us
821 * as we have the lock by now :)
822 */
823static int ocfs2_force_read_journal(struct inode *inode)
824{
825 int status = 0;
Mark Fasheh4f902c32007-03-09 16:26:50 -0800826 int i;
Mark Fasheh8110b072007-03-22 16:53:23 -0700827 u64 v_blkno, p_blkno, p_blocks, num_blocks;
Mark Fasheh4f902c32007-03-09 16:26:50 -0800828#define CONCURRENT_JOURNAL_FILL 32ULL
Mark Fashehccd979b2005-12-15 14:31:24 -0800829 struct buffer_head *bhs[CONCURRENT_JOURNAL_FILL];
830
831 mlog_entry_void();
832
Mark Fashehccd979b2005-12-15 14:31:24 -0800833 memset(bhs, 0, sizeof(struct buffer_head *) * CONCURRENT_JOURNAL_FILL);
834
Mark Fasheh8110b072007-03-22 16:53:23 -0700835 num_blocks = ocfs2_blocks_for_bytes(inode->i_sb, inode->i_size);
Mark Fashehccd979b2005-12-15 14:31:24 -0800836 v_blkno = 0;
Mark Fasheh8110b072007-03-22 16:53:23 -0700837 while (v_blkno < num_blocks) {
Mark Fashehccd979b2005-12-15 14:31:24 -0800838 status = ocfs2_extent_map_get_blocks(inode, v_blkno,
Mark Fasheh49cb8d22007-03-09 16:21:46 -0800839 &p_blkno, &p_blocks, NULL);
Mark Fashehccd979b2005-12-15 14:31:24 -0800840 if (status < 0) {
841 mlog_errno(status);
842 goto bail;
843 }
844
845 if (p_blocks > CONCURRENT_JOURNAL_FILL)
846 p_blocks = CONCURRENT_JOURNAL_FILL;
847
Mark Fashehdd4a2c22006-04-12 14:24:05 -0700848 /* We are reading journal data which should not
849 * be put in the uptodate cache */
Mark Fashehccd979b2005-12-15 14:31:24 -0800850 status = ocfs2_read_blocks(OCFS2_SB(inode->i_sb),
851 p_blkno, p_blocks, bhs, 0,
Mark Fashehdd4a2c22006-04-12 14:24:05 -0700852 NULL);
Mark Fashehccd979b2005-12-15 14:31:24 -0800853 if (status < 0) {
854 mlog_errno(status);
855 goto bail;
856 }
857
858 for(i = 0; i < p_blocks; i++) {
859 brelse(bhs[i]);
860 bhs[i] = NULL;
861 }
862
863 v_blkno += p_blocks;
864 }
865
866bail:
867 for(i = 0; i < CONCURRENT_JOURNAL_FILL; i++)
868 if (bhs[i])
869 brelse(bhs[i]);
870 mlog_exit(status);
871 return status;
872}
873
874struct ocfs2_la_recovery_item {
875 struct list_head lri_list;
876 int lri_slot;
877 struct ocfs2_dinode *lri_la_dinode;
878 struct ocfs2_dinode *lri_tl_dinode;
879};
880
881/* Does the second half of the recovery process. By this point, the
882 * node is marked clean and can actually be considered recovered,
883 * hence it's no longer in the recovery map, but there's still some
884 * cleanup we can do which shouldn't happen within the recovery thread
885 * as locking in that context becomes very difficult if we are to take
886 * recovering nodes into account.
887 *
888 * NOTE: This function can and will sleep on recovery of other nodes
889 * during cluster locking, just like any other ocfs2 process.
890 */
David Howellsc4028952006-11-22 14:57:56 +0000891void ocfs2_complete_recovery(struct work_struct *work)
Mark Fashehccd979b2005-12-15 14:31:24 -0800892{
893 int ret;
David Howellsc4028952006-11-22 14:57:56 +0000894 struct ocfs2_journal *journal =
895 container_of(work, struct ocfs2_journal, j_recovery_work);
896 struct ocfs2_super *osb = journal->j_osb;
Mark Fashehccd979b2005-12-15 14:31:24 -0800897 struct ocfs2_dinode *la_dinode, *tl_dinode;
Christoph Hellwig800deef2007-05-17 16:03:13 +0200898 struct ocfs2_la_recovery_item *item, *n;
Mark Fashehccd979b2005-12-15 14:31:24 -0800899 LIST_HEAD(tmp_la_list);
900
901 mlog_entry_void();
902
903 mlog(0, "completing recovery from keventd\n");
904
905 spin_lock(&journal->j_lock);
906 list_splice_init(&journal->j_la_cleanups, &tmp_la_list);
907 spin_unlock(&journal->j_lock);
908
Christoph Hellwig800deef2007-05-17 16:03:13 +0200909 list_for_each_entry_safe(item, n, &tmp_la_list, lri_list) {
Mark Fashehccd979b2005-12-15 14:31:24 -0800910 list_del_init(&item->lri_list);
911
912 mlog(0, "Complete recovery for slot %d\n", item->lri_slot);
913
914 la_dinode = item->lri_la_dinode;
915 if (la_dinode) {
Mark Fashehb06970532006-03-03 10:24:33 -0800916 mlog(0, "Clean up local alloc %llu\n",
Mark Fasheh1ca1a112007-04-27 16:01:25 -0700917 (unsigned long long)le64_to_cpu(la_dinode->i_blkno));
Mark Fashehccd979b2005-12-15 14:31:24 -0800918
919 ret = ocfs2_complete_local_alloc_recovery(osb,
920 la_dinode);
921 if (ret < 0)
922 mlog_errno(ret);
923
924 kfree(la_dinode);
925 }
926
927 tl_dinode = item->lri_tl_dinode;
928 if (tl_dinode) {
Mark Fashehb06970532006-03-03 10:24:33 -0800929 mlog(0, "Clean up truncate log %llu\n",
Mark Fasheh1ca1a112007-04-27 16:01:25 -0700930 (unsigned long long)le64_to_cpu(tl_dinode->i_blkno));
Mark Fashehccd979b2005-12-15 14:31:24 -0800931
932 ret = ocfs2_complete_truncate_log_recovery(osb,
933 tl_dinode);
934 if (ret < 0)
935 mlog_errno(ret);
936
937 kfree(tl_dinode);
938 }
939
940 ret = ocfs2_recover_orphans(osb, item->lri_slot);
941 if (ret < 0)
942 mlog_errno(ret);
943
944 kfree(item);
945 }
946
947 mlog(0, "Recovery completion\n");
948 mlog_exit_void();
949}
950
951/* NOTE: This function always eats your references to la_dinode and
952 * tl_dinode, either manually on error, or by passing them to
953 * ocfs2_complete_recovery */
954static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal,
955 int slot_num,
956 struct ocfs2_dinode *la_dinode,
957 struct ocfs2_dinode *tl_dinode)
958{
959 struct ocfs2_la_recovery_item *item;
960
Sunil Mushranafae00ab2006-04-12 14:37:00 -0700961 item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS);
Mark Fashehccd979b2005-12-15 14:31:24 -0800962 if (!item) {
963 /* Though we wish to avoid it, we are in fact safe in
964 * skipping local alloc cleanup as fsck.ocfs2 is more
965 * than capable of reclaiming unused space. */
966 if (la_dinode)
967 kfree(la_dinode);
968
969 if (tl_dinode)
970 kfree(tl_dinode);
971
972 mlog_errno(-ENOMEM);
973 return;
974 }
975
976 INIT_LIST_HEAD(&item->lri_list);
977 item->lri_la_dinode = la_dinode;
978 item->lri_slot = slot_num;
979 item->lri_tl_dinode = tl_dinode;
980
981 spin_lock(&journal->j_lock);
982 list_add_tail(&item->lri_list, &journal->j_la_cleanups);
983 queue_work(ocfs2_wq, &journal->j_recovery_work);
984 spin_unlock(&journal->j_lock);
985}
986
987/* Called by the mount code to queue recovery the last part of
988 * recovery for it's own slot. */
989void ocfs2_complete_mount_recovery(struct ocfs2_super *osb)
990{
991 struct ocfs2_journal *journal = osb->journal;
992
993 if (osb->dirty) {
994 /* No need to queue up our truncate_log as regular
995 * cleanup will catch that. */
996 ocfs2_queue_recovery_completion(journal,
997 osb->slot_num,
998 osb->local_alloc_copy,
999 NULL);
1000 ocfs2_schedule_truncate_log_flush(osb, 0);
1001
1002 osb->local_alloc_copy = NULL;
1003 osb->dirty = 0;
1004 }
1005}
1006
1007static int __ocfs2_recovery_thread(void *arg)
1008{
1009 int status, node_num;
1010 struct ocfs2_super *osb = arg;
Joel Becker553abd02008-02-01 12:03:57 -08001011 struct ocfs2_recovery_map *rm = osb->recovery_map;
Mark Fashehccd979b2005-12-15 14:31:24 -08001012
1013 mlog_entry_void();
1014
1015 status = ocfs2_wait_on_mount(osb);
1016 if (status < 0) {
1017 goto bail;
1018 }
1019
1020restart:
1021 status = ocfs2_super_lock(osb, 1);
1022 if (status < 0) {
1023 mlog_errno(status);
1024 goto bail;
1025 }
1026
Joel Becker553abd02008-02-01 12:03:57 -08001027 spin_lock(&osb->osb_lock);
1028 while (rm->rm_used) {
1029 /* It's always safe to remove entry zero, as we won't
1030 * clear it until ocfs2_recover_node() has succeeded. */
1031 node_num = rm->rm_entries[0];
1032 spin_unlock(&osb->osb_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -08001033
1034 status = ocfs2_recover_node(osb, node_num);
Joel Becker553abd02008-02-01 12:03:57 -08001035 if (!status) {
1036 ocfs2_recovery_map_clear(osb, node_num);
1037 } else {
Mark Fashehccd979b2005-12-15 14:31:24 -08001038 mlog(ML_ERROR,
1039 "Error %d recovering node %d on device (%u,%u)!\n",
1040 status, node_num,
1041 MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1042 mlog(ML_ERROR, "Volume requires unmount.\n");
Mark Fashehccd979b2005-12-15 14:31:24 -08001043 }
1044
Joel Becker553abd02008-02-01 12:03:57 -08001045 spin_lock(&osb->osb_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -08001046 }
Joel Becker553abd02008-02-01 12:03:57 -08001047 spin_unlock(&osb->osb_lock);
1048 mlog(0, "All nodes recovered\n");
1049
Sunil Mushran539d8262008-07-14 17:31:10 -07001050 /* Refresh all journal recovery generations from disk */
1051 status = ocfs2_check_journals_nolocks(osb);
1052 status = (status == -EROFS) ? 0 : status;
1053 if (status < 0)
1054 mlog_errno(status);
1055
Mark Fashehccd979b2005-12-15 14:31:24 -08001056 ocfs2_super_unlock(osb, 1);
1057
1058 /* We always run recovery on our own orphan dir - the dead
Mark Fasheh34d024f2007-09-24 15:56:19 -07001059 * node(s) may have disallowd a previos inode delete. Re-processing
1060 * is therefore required. */
Mark Fashehccd979b2005-12-15 14:31:24 -08001061 ocfs2_queue_recovery_completion(osb->journal, osb->slot_num, NULL,
1062 NULL);
1063
1064bail:
Arjan van de Venc74ec2f2006-01-13 21:54:23 -08001065 mutex_lock(&osb->recovery_lock);
Joel Becker553abd02008-02-01 12:03:57 -08001066 if (!status && !ocfs2_recovery_completed(osb)) {
Arjan van de Venc74ec2f2006-01-13 21:54:23 -08001067 mutex_unlock(&osb->recovery_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -08001068 goto restart;
1069 }
1070
1071 osb->recovery_thread_task = NULL;
1072 mb(); /* sync with ocfs2_recovery_thread_running */
1073 wake_up(&osb->recovery_event);
1074
Arjan van de Venc74ec2f2006-01-13 21:54:23 -08001075 mutex_unlock(&osb->recovery_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -08001076
1077 mlog_exit(status);
1078 /* no one is callint kthread_stop() for us so the kthread() api
1079 * requires that we call do_exit(). And it isn't exported, but
1080 * complete_and_exit() seems to be a minimal wrapper around it. */
1081 complete_and_exit(NULL, status);
1082 return status;
1083}
1084
1085void ocfs2_recovery_thread(struct ocfs2_super *osb, int node_num)
1086{
1087 mlog_entry("(node_num=%d, osb->node_num = %d)\n",
1088 node_num, osb->node_num);
1089
Arjan van de Venc74ec2f2006-01-13 21:54:23 -08001090 mutex_lock(&osb->recovery_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -08001091 if (osb->disable_recovery)
1092 goto out;
1093
1094 /* People waiting on recovery will wait on
1095 * the recovery map to empty. */
Joel Becker553abd02008-02-01 12:03:57 -08001096 if (ocfs2_recovery_map_set(osb, node_num))
1097 mlog(0, "node %d already in recovery map.\n", node_num);
Mark Fashehccd979b2005-12-15 14:31:24 -08001098
1099 mlog(0, "starting recovery thread...\n");
1100
1101 if (osb->recovery_thread_task)
1102 goto out;
1103
1104 osb->recovery_thread_task = kthread_run(__ocfs2_recovery_thread, osb,
Mark Fasheh78427042006-05-04 12:03:26 -07001105 "ocfs2rec");
Mark Fashehccd979b2005-12-15 14:31:24 -08001106 if (IS_ERR(osb->recovery_thread_task)) {
1107 mlog_errno((int)PTR_ERR(osb->recovery_thread_task));
1108 osb->recovery_thread_task = NULL;
1109 }
1110
1111out:
Arjan van de Venc74ec2f2006-01-13 21:54:23 -08001112 mutex_unlock(&osb->recovery_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -08001113 wake_up(&osb->recovery_event);
1114
1115 mlog_exit_void();
1116}
1117
Sunil Mushran539d8262008-07-14 17:31:10 -07001118static int ocfs2_read_journal_inode(struct ocfs2_super *osb,
1119 int slot_num,
1120 struct buffer_head **bh,
1121 struct inode **ret_inode)
1122{
1123 int status = -EACCES;
1124 struct inode *inode = NULL;
1125
1126 BUG_ON(slot_num >= osb->max_slots);
1127
1128 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
1129 slot_num);
1130 if (!inode || is_bad_inode(inode)) {
1131 mlog_errno(status);
1132 goto bail;
1133 }
1134 SET_INODE_JOURNAL(inode);
1135
1136 status = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno, bh, 0, inode);
1137 if (status < 0) {
1138 mlog_errno(status);
1139 goto bail;
1140 }
1141
1142 status = 0;
1143
1144bail:
1145 if (inode) {
1146 if (status || !ret_inode)
1147 iput(inode);
1148 else
1149 *ret_inode = inode;
1150 }
1151 return status;
1152}
1153
Mark Fashehccd979b2005-12-15 14:31:24 -08001154/* Does the actual journal replay and marks the journal inode as
1155 * clean. Will only replay if the journal inode is marked dirty. */
1156static int ocfs2_replay_journal(struct ocfs2_super *osb,
1157 int node_num,
1158 int slot_num)
1159{
1160 int status;
1161 int got_lock = 0;
1162 unsigned int flags;
1163 struct inode *inode = NULL;
1164 struct ocfs2_dinode *fe;
1165 journal_t *journal = NULL;
1166 struct buffer_head *bh = NULL;
Sunil Mushran539d8262008-07-14 17:31:10 -07001167 u32 slot_reco_gen;
Mark Fashehccd979b2005-12-15 14:31:24 -08001168
Sunil Mushran539d8262008-07-14 17:31:10 -07001169 status = ocfs2_read_journal_inode(osb, slot_num, &bh, &inode);
1170 if (status) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001171 mlog_errno(status);
1172 goto done;
1173 }
Sunil Mushran539d8262008-07-14 17:31:10 -07001174
1175 fe = (struct ocfs2_dinode *)bh->b_data;
1176 slot_reco_gen = ocfs2_get_recovery_generation(fe);
1177 brelse(bh);
1178 bh = NULL;
1179
1180 /*
1181 * As the fs recovery is asynchronous, there is a small chance that
1182 * another node mounted (and recovered) the slot before the recovery
1183 * thread could get the lock. To handle that, we dirty read the journal
1184 * inode for that slot to get the recovery generation. If it is
1185 * different than what we expected, the slot has been recovered.
1186 * If not, it needs recovery.
1187 */
1188 if (osb->slot_recovery_generations[slot_num] != slot_reco_gen) {
1189 mlog(0, "Slot %u already recovered (old/new=%u/%u)\n", slot_num,
1190 osb->slot_recovery_generations[slot_num], slot_reco_gen);
1191 osb->slot_recovery_generations[slot_num] = slot_reco_gen;
1192 status = -EBUSY;
Mark Fashehccd979b2005-12-15 14:31:24 -08001193 goto done;
1194 }
Sunil Mushran539d8262008-07-14 17:31:10 -07001195
1196 /* Continue with recovery as the journal has not yet been recovered */
Mark Fashehccd979b2005-12-15 14:31:24 -08001197
Mark Fashehe63aecb62007-10-18 15:30:42 -07001198 status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY);
Mark Fashehccd979b2005-12-15 14:31:24 -08001199 if (status < 0) {
Mark Fashehe63aecb62007-10-18 15:30:42 -07001200 mlog(0, "status returned from ocfs2_inode_lock=%d\n", status);
Mark Fashehccd979b2005-12-15 14:31:24 -08001201 if (status != -ERESTARTSYS)
1202 mlog(ML_ERROR, "Could not lock journal!\n");
1203 goto done;
1204 }
1205 got_lock = 1;
1206
1207 fe = (struct ocfs2_dinode *) bh->b_data;
1208
1209 flags = le32_to_cpu(fe->id1.journal1.ij_flags);
Sunil Mushran539d8262008-07-14 17:31:10 -07001210 slot_reco_gen = ocfs2_get_recovery_generation(fe);
Mark Fashehccd979b2005-12-15 14:31:24 -08001211
1212 if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) {
1213 mlog(0, "No recovery required for node %d\n", node_num);
Sunil Mushran539d8262008-07-14 17:31:10 -07001214 /* Refresh recovery generation for the slot */
1215 osb->slot_recovery_generations[slot_num] = slot_reco_gen;
Mark Fashehccd979b2005-12-15 14:31:24 -08001216 goto done;
1217 }
1218
1219 mlog(ML_NOTICE, "Recovering node %d from slot %d on device (%u,%u)\n",
1220 node_num, slot_num,
1221 MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1222
1223 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
1224
1225 status = ocfs2_force_read_journal(inode);
1226 if (status < 0) {
1227 mlog_errno(status);
1228 goto done;
1229 }
1230
1231 mlog(0, "calling journal_init_inode\n");
1232 journal = journal_init_inode(inode);
1233 if (journal == NULL) {
1234 mlog(ML_ERROR, "Linux journal layer error\n");
1235 status = -EIO;
1236 goto done;
1237 }
1238
1239 status = journal_load(journal);
1240 if (status < 0) {
1241 mlog_errno(status);
1242 if (!igrab(inode))
1243 BUG();
1244 journal_destroy(journal);
1245 goto done;
1246 }
1247
1248 ocfs2_clear_journal_error(osb->sb, journal, slot_num);
1249
1250 /* wipe the journal */
1251 mlog(0, "flushing the journal.\n");
1252 journal_lock_updates(journal);
1253 status = journal_flush(journal);
1254 journal_unlock_updates(journal);
1255 if (status < 0)
1256 mlog_errno(status);
1257
1258 /* This will mark the node clean */
1259 flags = le32_to_cpu(fe->id1.journal1.ij_flags);
1260 flags &= ~OCFS2_JOURNAL_DIRTY_FL;
1261 fe->id1.journal1.ij_flags = cpu_to_le32(flags);
1262
Sunil Mushran539d8262008-07-14 17:31:10 -07001263 /* Increment recovery generation to indicate successful recovery */
1264 ocfs2_bump_recovery_generation(fe);
1265 osb->slot_recovery_generations[slot_num] =
1266 ocfs2_get_recovery_generation(fe);
1267
Mark Fashehccd979b2005-12-15 14:31:24 -08001268 status = ocfs2_write_block(osb, bh, inode);
1269 if (status < 0)
1270 mlog_errno(status);
1271
1272 if (!igrab(inode))
1273 BUG();
1274
1275 journal_destroy(journal);
1276
1277done:
1278 /* drop the lock on this nodes journal */
1279 if (got_lock)
Mark Fashehe63aecb62007-10-18 15:30:42 -07001280 ocfs2_inode_unlock(inode, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -08001281
1282 if (inode)
1283 iput(inode);
1284
1285 if (bh)
1286 brelse(bh);
1287
1288 mlog_exit(status);
1289 return status;
1290}
1291
1292/*
1293 * Do the most important parts of node recovery:
1294 * - Replay it's journal
1295 * - Stamp a clean local allocator file
1296 * - Stamp a clean truncate log
1297 * - Mark the node clean
1298 *
1299 * If this function completes without error, a node in OCFS2 can be
1300 * said to have been safely recovered. As a result, failure during the
1301 * second part of a nodes recovery process (local alloc recovery) is
1302 * far less concerning.
1303 */
1304static int ocfs2_recover_node(struct ocfs2_super *osb,
1305 int node_num)
1306{
1307 int status = 0;
1308 int slot_num;
Mark Fashehccd979b2005-12-15 14:31:24 -08001309 struct ocfs2_dinode *la_copy = NULL;
1310 struct ocfs2_dinode *tl_copy = NULL;
1311
1312 mlog_entry("(node_num=%d, osb->node_num = %d)\n",
1313 node_num, osb->node_num);
1314
1315 mlog(0, "checking node %d\n", node_num);
1316
1317 /* Should not ever be called to recover ourselves -- in that
1318 * case we should've called ocfs2_journal_load instead. */
Eric Sesterhenn / snakebyteebdec832006-01-27 10:32:52 +01001319 BUG_ON(osb->node_num == node_num);
Mark Fashehccd979b2005-12-15 14:31:24 -08001320
Joel Beckerd85b20e2008-02-01 12:01:05 -08001321 slot_num = ocfs2_node_num_to_slot(osb, node_num);
1322 if (slot_num == -ENOENT) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001323 status = 0;
1324 mlog(0, "no slot for this node, so no recovery required.\n");
1325 goto done;
1326 }
1327
1328 mlog(0, "node %d was using slot %d\n", node_num, slot_num);
1329
1330 status = ocfs2_replay_journal(osb, node_num, slot_num);
1331 if (status < 0) {
Sunil Mushran539d8262008-07-14 17:31:10 -07001332 if (status == -EBUSY) {
1333 mlog(0, "Skipping recovery for slot %u (node %u) "
1334 "as another node has recovered it\n", slot_num,
1335 node_num);
1336 status = 0;
1337 goto done;
1338 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001339 mlog_errno(status);
1340 goto done;
1341 }
1342
1343 /* Stamp a clean local alloc file AFTER recovering the journal... */
1344 status = ocfs2_begin_local_alloc_recovery(osb, slot_num, &la_copy);
1345 if (status < 0) {
1346 mlog_errno(status);
1347 goto done;
1348 }
1349
1350 /* An error from begin_truncate_log_recovery is not
1351 * serious enough to warrant halting the rest of
1352 * recovery. */
1353 status = ocfs2_begin_truncate_log_recovery(osb, slot_num, &tl_copy);
1354 if (status < 0)
1355 mlog_errno(status);
1356
1357 /* Likewise, this would be a strange but ultimately not so
1358 * harmful place to get an error... */
Mark Fasheh8e8a4602008-02-01 11:59:09 -08001359 status = ocfs2_clear_slot(osb, slot_num);
Mark Fashehccd979b2005-12-15 14:31:24 -08001360 if (status < 0)
1361 mlog_errno(status);
1362
1363 /* This will kfree the memory pointed to by la_copy and tl_copy */
1364 ocfs2_queue_recovery_completion(osb->journal, slot_num, la_copy,
1365 tl_copy);
1366
1367 status = 0;
1368done:
1369
1370 mlog_exit(status);
1371 return status;
1372}
1373
1374/* Test node liveness by trylocking his journal. If we get the lock,
1375 * we drop it here. Return 0 if we got the lock, -EAGAIN if node is
1376 * still alive (we couldn't get the lock) and < 0 on error. */
1377static int ocfs2_trylock_journal(struct ocfs2_super *osb,
1378 int slot_num)
1379{
1380 int status, flags;
1381 struct inode *inode = NULL;
1382
1383 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
1384 slot_num);
1385 if (inode == NULL) {
1386 mlog(ML_ERROR, "access error\n");
1387 status = -EACCES;
1388 goto bail;
1389 }
1390 if (is_bad_inode(inode)) {
1391 mlog(ML_ERROR, "access error (bad inode)\n");
1392 iput(inode);
1393 inode = NULL;
1394 status = -EACCES;
1395 goto bail;
1396 }
1397 SET_INODE_JOURNAL(inode);
1398
1399 flags = OCFS2_META_LOCK_RECOVERY | OCFS2_META_LOCK_NOQUEUE;
Mark Fashehe63aecb62007-10-18 15:30:42 -07001400 status = ocfs2_inode_lock_full(inode, NULL, 1, flags);
Mark Fashehccd979b2005-12-15 14:31:24 -08001401 if (status < 0) {
1402 if (status != -EAGAIN)
1403 mlog_errno(status);
1404 goto bail;
1405 }
1406
Mark Fashehe63aecb62007-10-18 15:30:42 -07001407 ocfs2_inode_unlock(inode, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -08001408bail:
1409 if (inode)
1410 iput(inode);
1411
1412 return status;
1413}
1414
1415/* Call this underneath ocfs2_super_lock. It also assumes that the
1416 * slot info struct has been updated from disk. */
1417int ocfs2_mark_dead_nodes(struct ocfs2_super *osb)
1418{
Joel Beckerd85b20e2008-02-01 12:01:05 -08001419 unsigned int node_num;
1420 int status, i;
Mark Fasheha1af7d12008-08-19 17:20:28 -07001421 u32 gen;
Sunil Mushran539d8262008-07-14 17:31:10 -07001422 struct buffer_head *bh = NULL;
1423 struct ocfs2_dinode *di;
Mark Fashehccd979b2005-12-15 14:31:24 -08001424
1425 /* This is called with the super block cluster lock, so we
1426 * know that the slot map can't change underneath us. */
1427
Joel Beckerd85b20e2008-02-01 12:01:05 -08001428 for (i = 0; i < osb->max_slots; i++) {
Sunil Mushran539d8262008-07-14 17:31:10 -07001429 /* Read journal inode to get the recovery generation */
1430 status = ocfs2_read_journal_inode(osb, i, &bh, NULL);
1431 if (status) {
1432 mlog_errno(status);
1433 goto bail;
1434 }
1435 di = (struct ocfs2_dinode *)bh->b_data;
Mark Fasheha1af7d12008-08-19 17:20:28 -07001436 gen = ocfs2_get_recovery_generation(di);
Sunil Mushran539d8262008-07-14 17:31:10 -07001437 brelse(bh);
1438 bh = NULL;
1439
Mark Fasheha1af7d12008-08-19 17:20:28 -07001440 spin_lock(&osb->osb_lock);
1441 osb->slot_recovery_generations[i] = gen;
1442
Sunil Mushran539d8262008-07-14 17:31:10 -07001443 mlog(0, "Slot %u recovery generation is %u\n", i,
1444 osb->slot_recovery_generations[i]);
1445
Mark Fasheha1af7d12008-08-19 17:20:28 -07001446 if (i == osb->slot_num) {
1447 spin_unlock(&osb->osb_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -08001448 continue;
Mark Fasheha1af7d12008-08-19 17:20:28 -07001449 }
Joel Beckerd85b20e2008-02-01 12:01:05 -08001450
1451 status = ocfs2_slot_to_node_num_locked(osb, i, &node_num);
Mark Fasheha1af7d12008-08-19 17:20:28 -07001452 if (status == -ENOENT) {
1453 spin_unlock(&osb->osb_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -08001454 continue;
Mark Fasheha1af7d12008-08-19 17:20:28 -07001455 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001456
Mark Fasheha1af7d12008-08-19 17:20:28 -07001457 if (__ocfs2_recovery_map_test(osb, node_num)) {
1458 spin_unlock(&osb->osb_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -08001459 continue;
Mark Fasheha1af7d12008-08-19 17:20:28 -07001460 }
Joel Beckerd85b20e2008-02-01 12:01:05 -08001461 spin_unlock(&osb->osb_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -08001462
1463 /* Ok, we have a slot occupied by another node which
1464 * is not in the recovery map. We trylock his journal
1465 * file here to test if he's alive. */
1466 status = ocfs2_trylock_journal(osb, i);
1467 if (!status) {
1468 /* Since we're called from mount, we know that
1469 * the recovery thread can't race us on
1470 * setting / checking the recovery bits. */
1471 ocfs2_recovery_thread(osb, node_num);
1472 } else if ((status < 0) && (status != -EAGAIN)) {
1473 mlog_errno(status);
1474 goto bail;
1475 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001476 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001477
1478 status = 0;
1479bail:
1480 mlog_exit(status);
1481 return status;
1482}
1483
Mark Fasheh5eae5b92007-09-10 17:50:51 -07001484struct ocfs2_orphan_filldir_priv {
1485 struct inode *head;
1486 struct ocfs2_super *osb;
1487};
1488
1489static int ocfs2_orphan_filldir(void *priv, const char *name, int name_len,
1490 loff_t pos, u64 ino, unsigned type)
1491{
1492 struct ocfs2_orphan_filldir_priv *p = priv;
1493 struct inode *iter;
1494
1495 if (name_len == 1 && !strncmp(".", name, 1))
1496 return 0;
1497 if (name_len == 2 && !strncmp("..", name, 2))
1498 return 0;
1499
1500 /* Skip bad inodes so that recovery can continue */
1501 iter = ocfs2_iget(p->osb, ino,
Jan Kara5fa06132008-01-11 00:11:45 +01001502 OCFS2_FI_FLAG_ORPHAN_RECOVERY, 0);
Mark Fasheh5eae5b92007-09-10 17:50:51 -07001503 if (IS_ERR(iter))
1504 return 0;
1505
1506 mlog(0, "queue orphan %llu\n",
1507 (unsigned long long)OCFS2_I(iter)->ip_blkno);
1508 /* No locking is required for the next_orphan queue as there
1509 * is only ever a single process doing orphan recovery. */
1510 OCFS2_I(iter)->ip_next_orphan = p->head;
1511 p->head = iter;
1512
1513 return 0;
1514}
1515
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001516static int ocfs2_queue_orphans(struct ocfs2_super *osb,
1517 int slot,
1518 struct inode **head)
Mark Fashehccd979b2005-12-15 14:31:24 -08001519{
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001520 int status;
Mark Fashehccd979b2005-12-15 14:31:24 -08001521 struct inode *orphan_dir_inode = NULL;
Mark Fasheh5eae5b92007-09-10 17:50:51 -07001522 struct ocfs2_orphan_filldir_priv priv;
1523 loff_t pos = 0;
1524
1525 priv.osb = osb;
1526 priv.head = *head;
Mark Fashehccd979b2005-12-15 14:31:24 -08001527
1528 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
1529 ORPHAN_DIR_SYSTEM_INODE,
1530 slot);
1531 if (!orphan_dir_inode) {
1532 status = -ENOENT;
1533 mlog_errno(status);
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001534 return status;
1535 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001536
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001537 mutex_lock(&orphan_dir_inode->i_mutex);
Mark Fashehe63aecb62007-10-18 15:30:42 -07001538 status = ocfs2_inode_lock(orphan_dir_inode, NULL, 0);
Mark Fashehccd979b2005-12-15 14:31:24 -08001539 if (status < 0) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001540 mlog_errno(status);
1541 goto out;
1542 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001543
Mark Fasheh5eae5b92007-09-10 17:50:51 -07001544 status = ocfs2_dir_foreach(orphan_dir_inode, &pos, &priv,
1545 ocfs2_orphan_filldir);
1546 if (status) {
1547 mlog_errno(status);
Mark Fasheha86370f2007-12-03 14:06:23 -08001548 goto out_cluster;
Mark Fashehccd979b2005-12-15 14:31:24 -08001549 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001550
Mark Fasheh5eae5b92007-09-10 17:50:51 -07001551 *head = priv.head;
1552
Mark Fasheha86370f2007-12-03 14:06:23 -08001553out_cluster:
Mark Fashehe63aecb62007-10-18 15:30:42 -07001554 ocfs2_inode_unlock(orphan_dir_inode, 0);
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001555out:
1556 mutex_unlock(&orphan_dir_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -08001557 iput(orphan_dir_inode);
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001558 return status;
1559}
1560
1561static int ocfs2_orphan_recovery_can_continue(struct ocfs2_super *osb,
1562 int slot)
1563{
1564 int ret;
1565
1566 spin_lock(&osb->osb_lock);
1567 ret = !osb->osb_orphan_wipes[slot];
1568 spin_unlock(&osb->osb_lock);
1569 return ret;
1570}
1571
1572static void ocfs2_mark_recovering_orphan_dir(struct ocfs2_super *osb,
1573 int slot)
1574{
1575 spin_lock(&osb->osb_lock);
1576 /* Mark ourselves such that new processes in delete_inode()
1577 * know to quit early. */
1578 ocfs2_node_map_set_bit(osb, &osb->osb_recovering_orphan_dirs, slot);
1579 while (osb->osb_orphan_wipes[slot]) {
1580 /* If any processes are already in the middle of an
1581 * orphan wipe on this dir, then we need to wait for
1582 * them. */
1583 spin_unlock(&osb->osb_lock);
1584 wait_event_interruptible(osb->osb_wipe_event,
1585 ocfs2_orphan_recovery_can_continue(osb, slot));
1586 spin_lock(&osb->osb_lock);
1587 }
1588 spin_unlock(&osb->osb_lock);
1589}
1590
1591static void ocfs2_clear_recovering_orphan_dir(struct ocfs2_super *osb,
1592 int slot)
1593{
1594 ocfs2_node_map_clear_bit(osb, &osb->osb_recovering_orphan_dirs, slot);
1595}
1596
1597/*
1598 * Orphan recovery. Each mounted node has it's own orphan dir which we
1599 * must run during recovery. Our strategy here is to build a list of
1600 * the inodes in the orphan dir and iget/iput them. The VFS does
1601 * (most) of the rest of the work.
1602 *
1603 * Orphan recovery can happen at any time, not just mount so we have a
1604 * couple of extra considerations.
1605 *
1606 * - We grab as many inodes as we can under the orphan dir lock -
1607 * doing iget() outside the orphan dir risks getting a reference on
1608 * an invalid inode.
1609 * - We must be sure not to deadlock with other processes on the
1610 * system wanting to run delete_inode(). This can happen when they go
1611 * to lock the orphan dir and the orphan recovery process attempts to
1612 * iget() inside the orphan dir lock. This can be avoided by
1613 * advertising our state to ocfs2_delete_inode().
1614 */
1615static int ocfs2_recover_orphans(struct ocfs2_super *osb,
1616 int slot)
1617{
1618 int ret = 0;
1619 struct inode *inode = NULL;
1620 struct inode *iter;
1621 struct ocfs2_inode_info *oi;
1622
1623 mlog(0, "Recover inodes from orphan dir in slot %d\n", slot);
1624
1625 ocfs2_mark_recovering_orphan_dir(osb, slot);
1626 ret = ocfs2_queue_orphans(osb, slot, &inode);
1627 ocfs2_clear_recovering_orphan_dir(osb, slot);
1628
1629 /* Error here should be noted, but we want to continue with as
1630 * many queued inodes as we've got. */
1631 if (ret)
1632 mlog_errno(ret);
Mark Fashehccd979b2005-12-15 14:31:24 -08001633
1634 while (inode) {
1635 oi = OCFS2_I(inode);
Mark Fashehb06970532006-03-03 10:24:33 -08001636 mlog(0, "iput orphan %llu\n", (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001637
1638 iter = oi->ip_next_orphan;
1639
1640 spin_lock(&oi->ip_lock);
Mark Fasheh34d024f2007-09-24 15:56:19 -07001641 /* The remote delete code may have set these on the
1642 * assumption that the other node would wipe them
1643 * successfully. If they are still in the node's
1644 * orphan dir, we need to reset that state. */
Mark Fashehccd979b2005-12-15 14:31:24 -08001645 oi->ip_flags &= ~(OCFS2_INODE_DELETED|OCFS2_INODE_SKIP_DELETE);
1646
1647 /* Set the proper information to get us going into
1648 * ocfs2_delete_inode. */
1649 oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED;
Mark Fashehccd979b2005-12-15 14:31:24 -08001650 spin_unlock(&oi->ip_lock);
1651
1652 iput(inode);
1653
1654 inode = iter;
1655 }
1656
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001657 return ret;
Mark Fashehccd979b2005-12-15 14:31:24 -08001658}
1659
1660static int ocfs2_wait_on_mount(struct ocfs2_super *osb)
1661{
1662 /* This check is good because ocfs2 will wait on our recovery
1663 * thread before changing it to something other than MOUNTED
1664 * or DISABLED. */
1665 wait_event(osb->osb_mount_event,
1666 atomic_read(&osb->vol_state) == VOLUME_MOUNTED ||
1667 atomic_read(&osb->vol_state) == VOLUME_DISABLED);
1668
1669 /* If there's an error on mount, then we may never get to the
1670 * MOUNTED flag, but this is set right before
1671 * dismount_volume() so we can trust it. */
1672 if (atomic_read(&osb->vol_state) == VOLUME_DISABLED) {
1673 mlog(0, "mount error, exiting!\n");
1674 return -EBUSY;
1675 }
1676
1677 return 0;
1678}
1679
1680static int ocfs2_commit_thread(void *arg)
1681{
1682 int status;
1683 struct ocfs2_super *osb = arg;
1684 struct ocfs2_journal *journal = osb->journal;
1685
1686 /* we can trust j_num_trans here because _should_stop() is only set in
1687 * shutdown and nobody other than ourselves should be able to start
1688 * transactions. committing on shutdown might take a few iterations
1689 * as final transactions put deleted inodes on the list */
1690 while (!(kthread_should_stop() &&
1691 atomic_read(&journal->j_num_trans) == 0)) {
1692
Mark Fasheh745ae8ba2006-02-09 13:23:39 -08001693 wait_event_interruptible(osb->checkpoint_event,
1694 atomic_read(&journal->j_num_trans)
1695 || kthread_should_stop());
Mark Fashehccd979b2005-12-15 14:31:24 -08001696
1697 status = ocfs2_commit_cache(osb);
1698 if (status < 0)
1699 mlog_errno(status);
1700
1701 if (kthread_should_stop() && atomic_read(&journal->j_num_trans)){
1702 mlog(ML_KTHREAD,
1703 "commit_thread: %u transactions pending on "
1704 "shutdown\n",
1705 atomic_read(&journal->j_num_trans));
1706 }
1707 }
1708
1709 return 0;
1710}
1711
Sunil Mushran539d8262008-07-14 17:31:10 -07001712/* Reads all the journal inodes without taking any cluster locks. Used
1713 * for hard readonly access to determine whether any journal requires
1714 * recovery. Also used to refresh the recovery generation numbers after
1715 * a journal has been recovered by another node.
1716 */
Mark Fashehccd979b2005-12-15 14:31:24 -08001717int ocfs2_check_journals_nolocks(struct ocfs2_super *osb)
1718{
1719 int ret = 0;
1720 unsigned int slot;
Sunil Mushran539d8262008-07-14 17:31:10 -07001721 struct buffer_head *di_bh = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -08001722 struct ocfs2_dinode *di;
Sunil Mushran539d8262008-07-14 17:31:10 -07001723 int journal_dirty = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -08001724
1725 for(slot = 0; slot < osb->max_slots; slot++) {
Sunil Mushran539d8262008-07-14 17:31:10 -07001726 ret = ocfs2_read_journal_inode(osb, slot, &di_bh, NULL);
1727 if (ret) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001728 mlog_errno(ret);
1729 goto out;
1730 }
1731
1732 di = (struct ocfs2_dinode *) di_bh->b_data;
1733
Sunil Mushran539d8262008-07-14 17:31:10 -07001734 osb->slot_recovery_generations[slot] =
1735 ocfs2_get_recovery_generation(di);
1736
Mark Fashehccd979b2005-12-15 14:31:24 -08001737 if (le32_to_cpu(di->id1.journal1.ij_flags) &
1738 OCFS2_JOURNAL_DIRTY_FL)
Sunil Mushran539d8262008-07-14 17:31:10 -07001739 journal_dirty = 1;
Mark Fashehccd979b2005-12-15 14:31:24 -08001740
1741 brelse(di_bh);
Sunil Mushran539d8262008-07-14 17:31:10 -07001742 di_bh = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -08001743 }
1744
1745out:
Sunil Mushran539d8262008-07-14 17:31:10 -07001746 if (journal_dirty)
1747 ret = -EROFS;
Mark Fashehccd979b2005-12-15 14:31:24 -08001748 return ret;
1749}