blob: 44fc3e530c3d82c5a0d42d838461365cc53d72f3 [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>
Srinivas Eeda83273932009-06-03 17:02:55 -070031#include <linux/time.h>
32#include <linux/random.h>
Mark Fashehccd979b2005-12-15 14:31:24 -080033
Mark Fashehccd979b2005-12-15 14:31:24 -080034#include <cluster/masklog.h>
35
36#include "ocfs2.h"
37
38#include "alloc.h"
Joel Becker50655ae2008-09-11 15:53:07 -070039#include "blockcheck.h"
Mark Fasheh316f4b92007-09-07 18:21:26 -070040#include "dir.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080041#include "dlmglue.h"
42#include "extent_map.h"
43#include "heartbeat.h"
44#include "inode.h"
45#include "journal.h"
46#include "localalloc.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080047#include "slot_map.h"
48#include "super.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080049#include "sysfile.h"
Joel Becker0cf2f762009-02-12 16:41:25 -080050#include "uptodate.h"
Jan Kara22053632008-10-20 23:50:38 +020051#include "quota.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080052
53#include "buffer_head_io.h"
Tao Mab4107952011-02-24 14:15:35 +080054#include "ocfs2_trace.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080055
Ingo Molnar34af9462006-06-27 02:53:55 -070056DEFINE_SPINLOCK(trans_inc_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -080057
Srinivas Eeda83273932009-06-03 17:02:55 -070058#define ORPHAN_SCAN_SCHEDULE_TIMEOUT 300000
59
Mark Fashehccd979b2005-12-15 14:31:24 -080060static int ocfs2_force_read_journal(struct inode *inode);
61static int ocfs2_recover_node(struct ocfs2_super *osb,
Jan Kara22053632008-10-20 23:50:38 +020062 int node_num, int slot_num);
Mark Fashehccd979b2005-12-15 14:31:24 -080063static int __ocfs2_recovery_thread(void *arg);
64static int ocfs2_commit_cache(struct ocfs2_super *osb);
Jan Kara19ece542008-08-21 20:13:17 +020065static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota);
Mark Fashehccd979b2005-12-15 14:31:24 -080066static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
Sunil Mushran539d8262008-07-14 17:31:10 -070067 int dirty, int replayed);
Mark Fashehccd979b2005-12-15 14:31:24 -080068static int ocfs2_trylock_journal(struct ocfs2_super *osb,
69 int slot_num);
70static int ocfs2_recover_orphans(struct ocfs2_super *osb,
71 int slot);
72static int ocfs2_commit_thread(void *arg);
Srinivas Eeda9140db02009-03-06 14:21:46 -080073static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal,
74 int slot_num,
75 struct ocfs2_dinode *la_dinode,
76 struct ocfs2_dinode *tl_dinode,
77 struct ocfs2_quota_recovery *qrec);
Mark Fashehccd979b2005-12-15 14:31:24 -080078
Jan Kara19ece542008-08-21 20:13:17 +020079static inline int ocfs2_wait_on_mount(struct ocfs2_super *osb)
80{
81 return __ocfs2_wait_on_mount(osb, 0);
82}
83
84static inline int ocfs2_wait_on_quotas(struct ocfs2_super *osb)
85{
86 return __ocfs2_wait_on_mount(osb, 1);
87}
88
Srinivas Eeda9140db02009-03-06 14:21:46 -080089/*
90 * This replay_map is to track online/offline slots, so we could recover
91 * offline slots during recovery and mount
92 */
93
94enum ocfs2_replay_state {
95 REPLAY_UNNEEDED = 0, /* Replay is not needed, so ignore this map */
96 REPLAY_NEEDED, /* Replay slots marked in rm_replay_slots */
97 REPLAY_DONE /* Replay was already queued */
98};
99
100struct ocfs2_replay_map {
101 unsigned int rm_slots;
102 enum ocfs2_replay_state rm_state;
103 unsigned char rm_replay_slots[0];
104};
105
106void ocfs2_replay_map_set_state(struct ocfs2_super *osb, int state)
107{
108 if (!osb->replay_map)
109 return;
110
111 /* If we've already queued the replay, we don't have any more to do */
112 if (osb->replay_map->rm_state == REPLAY_DONE)
113 return;
114
115 osb->replay_map->rm_state = state;
116}
117
118int ocfs2_compute_replay_slots(struct ocfs2_super *osb)
119{
120 struct ocfs2_replay_map *replay_map;
121 int i, node_num;
122
123 /* If replay map is already set, we don't do it again */
124 if (osb->replay_map)
125 return 0;
126
127 replay_map = kzalloc(sizeof(struct ocfs2_replay_map) +
128 (osb->max_slots * sizeof(char)), GFP_KERNEL);
129
130 if (!replay_map) {
131 mlog_errno(-ENOMEM);
132 return -ENOMEM;
133 }
134
135 spin_lock(&osb->osb_lock);
136
137 replay_map->rm_slots = osb->max_slots;
138 replay_map->rm_state = REPLAY_UNNEEDED;
139
140 /* set rm_replay_slots for offline slot(s) */
141 for (i = 0; i < replay_map->rm_slots; i++) {
142 if (ocfs2_slot_to_node_num_locked(osb, i, &node_num) == -ENOENT)
143 replay_map->rm_replay_slots[i] = 1;
144 }
145
146 osb->replay_map = replay_map;
147 spin_unlock(&osb->osb_lock);
148 return 0;
149}
150
151void ocfs2_queue_replay_slots(struct ocfs2_super *osb)
152{
153 struct ocfs2_replay_map *replay_map = osb->replay_map;
154 int i;
155
156 if (!replay_map)
157 return;
158
159 if (replay_map->rm_state != REPLAY_NEEDED)
160 return;
161
162 for (i = 0; i < replay_map->rm_slots; i++)
163 if (replay_map->rm_replay_slots[i])
164 ocfs2_queue_recovery_completion(osb->journal, i, NULL,
165 NULL, NULL);
166 replay_map->rm_state = REPLAY_DONE;
167}
168
169void ocfs2_free_replay_slots(struct ocfs2_super *osb)
170{
171 struct ocfs2_replay_map *replay_map = osb->replay_map;
172
173 if (!osb->replay_map)
174 return;
175
176 kfree(replay_map);
177 osb->replay_map = NULL;
178}
179
Joel Becker553abd02008-02-01 12:03:57 -0800180int ocfs2_recovery_init(struct ocfs2_super *osb)
181{
182 struct ocfs2_recovery_map *rm;
183
184 mutex_init(&osb->recovery_lock);
185 osb->disable_recovery = 0;
186 osb->recovery_thread_task = NULL;
187 init_waitqueue_head(&osb->recovery_event);
188
189 rm = kzalloc(sizeof(struct ocfs2_recovery_map) +
190 osb->max_slots * sizeof(unsigned int),
191 GFP_KERNEL);
192 if (!rm) {
193 mlog_errno(-ENOMEM);
194 return -ENOMEM;
195 }
196
197 rm->rm_entries = (unsigned int *)((char *)rm +
198 sizeof(struct ocfs2_recovery_map));
199 osb->recovery_map = rm;
200
201 return 0;
202}
203
204/* we can't grab the goofy sem lock from inside wait_event, so we use
205 * memory barriers to make sure that we'll see the null task before
206 * being woken up */
207static int ocfs2_recovery_thread_running(struct ocfs2_super *osb)
208{
209 mb();
210 return osb->recovery_thread_task != NULL;
211}
212
213void ocfs2_recovery_exit(struct ocfs2_super *osb)
214{
215 struct ocfs2_recovery_map *rm;
216
217 /* disable any new recovery threads and wait for any currently
218 * running ones to exit. Do this before setting the vol_state. */
219 mutex_lock(&osb->recovery_lock);
220 osb->disable_recovery = 1;
221 mutex_unlock(&osb->recovery_lock);
222 wait_event(osb->recovery_event, !ocfs2_recovery_thread_running(osb));
223
224 /* At this point, we know that no more recovery threads can be
225 * launched, so wait for any recovery completion work to
226 * complete. */
227 flush_workqueue(ocfs2_wq);
228
229 /*
230 * Now that recovery is shut down, and the osb is about to be
231 * freed, the osb_lock is not taken here.
232 */
233 rm = osb->recovery_map;
234 /* XXX: Should we bug if there are dirty entries? */
235
236 kfree(rm);
237}
238
239static int __ocfs2_recovery_map_test(struct ocfs2_super *osb,
240 unsigned int node_num)
241{
242 int i;
243 struct ocfs2_recovery_map *rm = osb->recovery_map;
244
245 assert_spin_locked(&osb->osb_lock);
246
247 for (i = 0; i < rm->rm_used; i++) {
248 if (rm->rm_entries[i] == node_num)
249 return 1;
250 }
251
252 return 0;
253}
254
255/* Behaves like test-and-set. Returns the previous value */
256static int ocfs2_recovery_map_set(struct ocfs2_super *osb,
257 unsigned int node_num)
258{
259 struct ocfs2_recovery_map *rm = osb->recovery_map;
260
261 spin_lock(&osb->osb_lock);
262 if (__ocfs2_recovery_map_test(osb, node_num)) {
263 spin_unlock(&osb->osb_lock);
264 return 1;
265 }
266
267 /* XXX: Can this be exploited? Not from o2dlm... */
268 BUG_ON(rm->rm_used >= osb->max_slots);
269
270 rm->rm_entries[rm->rm_used] = node_num;
271 rm->rm_used++;
272 spin_unlock(&osb->osb_lock);
273
274 return 0;
275}
276
277static void ocfs2_recovery_map_clear(struct ocfs2_super *osb,
278 unsigned int node_num)
279{
280 int i;
281 struct ocfs2_recovery_map *rm = osb->recovery_map;
282
283 spin_lock(&osb->osb_lock);
284
285 for (i = 0; i < rm->rm_used; i++) {
286 if (rm->rm_entries[i] == node_num)
287 break;
288 }
289
290 if (i < rm->rm_used) {
291 /* XXX: be careful with the pointer math */
292 memmove(&(rm->rm_entries[i]), &(rm->rm_entries[i + 1]),
293 (rm->rm_used - i - 1) * sizeof(unsigned int));
294 rm->rm_used--;
295 }
296
297 spin_unlock(&osb->osb_lock);
298}
299
Mark Fashehccd979b2005-12-15 14:31:24 -0800300static int ocfs2_commit_cache(struct ocfs2_super *osb)
301{
302 int status = 0;
303 unsigned int flushed;
Mark Fashehccd979b2005-12-15 14:31:24 -0800304 struct ocfs2_journal *journal = NULL;
305
Mark Fashehccd979b2005-12-15 14:31:24 -0800306 journal = osb->journal;
307
308 /* Flush all pending commits and checkpoint the journal. */
309 down_write(&journal->j_trans_barrier);
310
Tao Mab4107952011-02-24 14:15:35 +0800311 flushed = atomic_read(&journal->j_num_trans);
312 trace_ocfs2_commit_cache_begin(flushed);
313 if (flushed == 0) {
Mark Fashehccd979b2005-12-15 14:31:24 -0800314 up_write(&journal->j_trans_barrier);
Mark Fashehccd979b2005-12-15 14:31:24 -0800315 goto finally;
316 }
317
Joel Becker2b4e30f2008-09-03 20:03:41 -0700318 jbd2_journal_lock_updates(journal->j_journal);
319 status = jbd2_journal_flush(journal->j_journal);
320 jbd2_journal_unlock_updates(journal->j_journal);
Mark Fashehccd979b2005-12-15 14:31:24 -0800321 if (status < 0) {
322 up_write(&journal->j_trans_barrier);
323 mlog_errno(status);
324 goto finally;
325 }
326
Tao Maf9c57ad2010-08-02 11:02:13 +0800327 ocfs2_inc_trans_id(journal);
Mark Fashehccd979b2005-12-15 14:31:24 -0800328
329 flushed = atomic_read(&journal->j_num_trans);
330 atomic_set(&journal->j_num_trans, 0);
331 up_write(&journal->j_trans_barrier);
332
Tao Mab4107952011-02-24 14:15:35 +0800333 trace_ocfs2_commit_cache_end(journal->j_trans_id, flushed);
Mark Fashehccd979b2005-12-15 14:31:24 -0800334
Mark Fasheh34d024f2007-09-24 15:56:19 -0700335 ocfs2_wake_downconvert_thread(osb);
Mark Fashehccd979b2005-12-15 14:31:24 -0800336 wake_up(&journal->j_checkpointed);
337finally:
Mark Fashehccd979b2005-12-15 14:31:24 -0800338 return status;
339}
340
Mark Fasheh1fabe142006-10-09 18:11:45 -0700341handle_t *ocfs2_start_trans(struct ocfs2_super *osb, int max_buffs)
Mark Fashehccd979b2005-12-15 14:31:24 -0800342{
Mark Fashehccd979b2005-12-15 14:31:24 -0800343 journal_t *journal = osb->journal->j_journal;
Mark Fasheh1fabe142006-10-09 18:11:45 -0700344 handle_t *handle;
Mark Fashehccd979b2005-12-15 14:31:24 -0800345
Eric Sesterhenn / snakebyteebdec832006-01-27 10:32:52 +0100346 BUG_ON(!osb || !osb->journal->j_journal);
Mark Fashehccd979b2005-12-15 14:31:24 -0800347
Mark Fasheh65eff9c2006-10-09 17:26:22 -0700348 if (ocfs2_is_hard_readonly(osb))
349 return ERR_PTR(-EROFS);
Mark Fashehccd979b2005-12-15 14:31:24 -0800350
351 BUG_ON(osb->journal->j_state == OCFS2_JOURNAL_FREE);
352 BUG_ON(max_buffs <= 0);
353
Jan Kara90e86a62008-08-27 22:30:28 +0200354 /* Nested transaction? Just return the handle... */
355 if (journal_current_handle())
356 return jbd2_journal_start(journal, max_buffs);
Mark Fashehccd979b2005-12-15 14:31:24 -0800357
Jan Karafef69252012-06-12 16:20:40 +0200358 sb_start_intwrite(osb->sb);
359
Mark Fashehccd979b2005-12-15 14:31:24 -0800360 down_read(&osb->journal->j_trans_barrier);
361
Joel Becker2b4e30f2008-09-03 20:03:41 -0700362 handle = jbd2_journal_start(journal, max_buffs);
Mark Fasheh1fabe142006-10-09 18:11:45 -0700363 if (IS_ERR(handle)) {
Mark Fashehccd979b2005-12-15 14:31:24 -0800364 up_read(&osb->journal->j_trans_barrier);
Jan Karafef69252012-06-12 16:20:40 +0200365 sb_end_intwrite(osb->sb);
Mark Fashehccd979b2005-12-15 14:31:24 -0800366
Mark Fasheh1fabe142006-10-09 18:11:45 -0700367 mlog_errno(PTR_ERR(handle));
Mark Fashehccd979b2005-12-15 14:31:24 -0800368
369 if (is_journal_aborted(journal)) {
370 ocfs2_abort(osb->sb, "Detected aborted journal");
Mark Fasheh1fabe142006-10-09 18:11:45 -0700371 handle = ERR_PTR(-EROFS);
Mark Fashehccd979b2005-12-15 14:31:24 -0800372 }
Sunil Mushranc271c5c2006-12-05 17:56:35 -0800373 } else {
374 if (!ocfs2_mount_local(osb))
375 atomic_inc(&(osb->journal->j_num_trans));
376 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800377
Mark Fashehccd979b2005-12-15 14:31:24 -0800378 return handle;
Mark Fashehccd979b2005-12-15 14:31:24 -0800379}
380
Mark Fasheh1fabe142006-10-09 18:11:45 -0700381int ocfs2_commit_trans(struct ocfs2_super *osb,
382 handle_t *handle)
Mark Fashehccd979b2005-12-15 14:31:24 -0800383{
Jan Kara90e86a62008-08-27 22:30:28 +0200384 int ret, nested;
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700385 struct ocfs2_journal *journal = osb->journal;
Mark Fashehccd979b2005-12-15 14:31:24 -0800386
387 BUG_ON(!handle);
388
Jan Kara90e86a62008-08-27 22:30:28 +0200389 nested = handle->h_ref > 1;
Joel Becker2b4e30f2008-09-03 20:03:41 -0700390 ret = jbd2_journal_stop(handle);
Mark Fasheh1fabe142006-10-09 18:11:45 -0700391 if (ret < 0)
392 mlog_errno(ret);
Mark Fashehccd979b2005-12-15 14:31:24 -0800393
Jan Karafef69252012-06-12 16:20:40 +0200394 if (!nested) {
Jan Kara90e86a62008-08-27 22:30:28 +0200395 up_read(&journal->j_trans_barrier);
Jan Karafef69252012-06-12 16:20:40 +0200396 sb_end_intwrite(osb->sb);
397 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800398
Mark Fasheh1fabe142006-10-09 18:11:45 -0700399 return ret;
Mark Fashehccd979b2005-12-15 14:31:24 -0800400}
401
402/*
Tao Mac901fb02010-04-26 14:34:57 +0800403 * 'nblocks' is what you want to add to the current transaction.
Mark Fashehccd979b2005-12-15 14:31:24 -0800404 *
Joel Becker2b4e30f2008-09-03 20:03:41 -0700405 * This might call jbd2_journal_restart() which will commit dirty buffers
Mark Fashehe8aed342007-12-03 16:43:01 -0800406 * and then restart the transaction. Before calling
407 * ocfs2_extend_trans(), any changed blocks should have been
408 * dirtied. After calling it, all blocks which need to be changed must
409 * go through another set of journal_access/journal_dirty calls.
410 *
Mark Fashehccd979b2005-12-15 14:31:24 -0800411 * WARNING: This will not release any semaphores or disk locks taken
412 * during the transaction, so make sure they were taken *before*
413 * start_trans or we'll have ordering deadlocks.
414 *
415 * WARNING2: Note that we do *not* drop j_trans_barrier here. This is
416 * good because transaction ids haven't yet been recorded on the
417 * cluster locks associated with this handle.
418 */
Mark Fasheh1fc58142006-10-05 14:15:36 -0700419int ocfs2_extend_trans(handle_t *handle, int nblocks)
Mark Fashehccd979b2005-12-15 14:31:24 -0800420{
Tao Mac901fb02010-04-26 14:34:57 +0800421 int status, old_nblocks;
Mark Fashehccd979b2005-12-15 14:31:24 -0800422
423 BUG_ON(!handle);
Tao Mac901fb02010-04-26 14:34:57 +0800424 BUG_ON(nblocks < 0);
Mark Fashehccd979b2005-12-15 14:31:24 -0800425
Tao Mac901fb02010-04-26 14:34:57 +0800426 if (!nblocks)
427 return 0;
428
429 old_nblocks = handle->h_buffer_credits;
Mark Fashehccd979b2005-12-15 14:31:24 -0800430
Tao Mab4107952011-02-24 14:15:35 +0800431 trace_ocfs2_extend_trans(old_nblocks, nblocks);
Mark Fashehccd979b2005-12-15 14:31:24 -0800432
Joel Beckere407e392008-06-12 22:35:39 -0700433#ifdef CONFIG_OCFS2_DEBUG_FS
Mark Fasheh0879c582007-12-03 16:42:19 -0800434 status = 1;
435#else
Joel Becker2b4e30f2008-09-03 20:03:41 -0700436 status = jbd2_journal_extend(handle, nblocks);
Mark Fashehccd979b2005-12-15 14:31:24 -0800437 if (status < 0) {
438 mlog_errno(status);
439 goto bail;
440 }
Mark Fasheh0879c582007-12-03 16:42:19 -0800441#endif
Mark Fashehccd979b2005-12-15 14:31:24 -0800442
443 if (status > 0) {
Tao Mab4107952011-02-24 14:15:35 +0800444 trace_ocfs2_extend_trans_restart(old_nblocks + nblocks);
Tao Mac901fb02010-04-26 14:34:57 +0800445 status = jbd2_journal_restart(handle,
446 old_nblocks + nblocks);
Mark Fashehccd979b2005-12-15 14:31:24 -0800447 if (status < 0) {
Mark Fashehccd979b2005-12-15 14:31:24 -0800448 mlog_errno(status);
449 goto bail;
450 }
Mark Fasheh01ddf1e2006-10-05 13:54:39 -0700451 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800452
453 status = 0;
454bail:
Mark Fashehccd979b2005-12-15 14:31:24 -0800455 return status;
456}
457
Younger Liu2b1e55c2013-09-11 14:19:44 -0700458/*
459 * If we have fewer than thresh credits, extend by OCFS2_MAX_TRANS_DATA.
460 * If that fails, restart the transaction & regain write access for the
461 * buffer head which is used for metadata modifications.
462 * Taken from Ext4: extend_or_restart_transaction()
463 */
464int ocfs2_allocate_extend_trans(handle_t *handle, int thresh)
465{
466 int status, old_nblks;
467
468 BUG_ON(!handle);
469
470 old_nblks = handle->h_buffer_credits;
471 trace_ocfs2_allocate_extend_trans(old_nblks, thresh);
472
473 if (old_nblks < thresh)
474 return 0;
475
476 status = jbd2_journal_extend(handle, OCFS2_MAX_TRANS_DATA);
477 if (status < 0) {
478 mlog_errno(status);
479 goto bail;
480 }
481
482 if (status > 0) {
483 status = jbd2_journal_restart(handle, OCFS2_MAX_TRANS_DATA);
484 if (status < 0)
485 mlog_errno(status);
486 }
487
488bail:
489 return status;
490}
491
492
Joel Becker50655ae2008-09-11 15:53:07 -0700493struct ocfs2_triggers {
494 struct jbd2_buffer_trigger_type ot_triggers;
495 int ot_offset;
496};
497
498static inline struct ocfs2_triggers *to_ocfs2_trigger(struct jbd2_buffer_trigger_type *triggers)
499{
500 return container_of(triggers, struct ocfs2_triggers, ot_triggers);
501}
502
Jan Kara13ceef02010-07-14 07:56:33 +0200503static void ocfs2_frozen_trigger(struct jbd2_buffer_trigger_type *triggers,
Joel Becker50655ae2008-09-11 15:53:07 -0700504 struct buffer_head *bh,
505 void *data, size_t size)
506{
507 struct ocfs2_triggers *ot = to_ocfs2_trigger(triggers);
508
509 /*
510 * We aren't guaranteed to have the superblock here, so we
511 * must unconditionally compute the ecc data.
512 * __ocfs2_journal_access() will only set the triggers if
513 * metaecc is enabled.
514 */
515 ocfs2_block_check_compute(data, size, data + ot->ot_offset);
516}
517
518/*
519 * Quota blocks have their own trigger because the struct ocfs2_block_check
520 * offset depends on the blocksize.
521 */
Jan Kara13ceef02010-07-14 07:56:33 +0200522static void ocfs2_dq_frozen_trigger(struct jbd2_buffer_trigger_type *triggers,
Joel Becker50655ae2008-09-11 15:53:07 -0700523 struct buffer_head *bh,
524 void *data, size_t size)
525{
526 struct ocfs2_disk_dqtrailer *dqt =
527 ocfs2_block_dqtrailer(size, data);
528
529 /*
530 * We aren't guaranteed to have the superblock here, so we
531 * must unconditionally compute the ecc data.
532 * __ocfs2_journal_access() will only set the triggers if
533 * metaecc is enabled.
534 */
535 ocfs2_block_check_compute(data, size, &dqt->dq_check);
536}
537
Joel Beckerc175a512008-12-10 17:58:22 -0800538/*
539 * Directory blocks also have their own trigger because the
540 * struct ocfs2_block_check offset depends on the blocksize.
541 */
Jan Kara13ceef02010-07-14 07:56:33 +0200542static void ocfs2_db_frozen_trigger(struct jbd2_buffer_trigger_type *triggers,
Joel Beckerc175a512008-12-10 17:58:22 -0800543 struct buffer_head *bh,
544 void *data, size_t size)
545{
546 struct ocfs2_dir_block_trailer *trailer =
547 ocfs2_dir_trailer_from_size(size, data);
548
549 /*
550 * We aren't guaranteed to have the superblock here, so we
551 * must unconditionally compute the ecc data.
552 * __ocfs2_journal_access() will only set the triggers if
553 * metaecc is enabled.
554 */
555 ocfs2_block_check_compute(data, size, &trailer->db_check);
556}
557
Joel Becker50655ae2008-09-11 15:53:07 -0700558static void ocfs2_abort_trigger(struct jbd2_buffer_trigger_type *triggers,
559 struct buffer_head *bh)
560{
561 mlog(ML_ERROR,
562 "ocfs2_abort_trigger called by JBD2. bh = 0x%lx, "
563 "bh->b_blocknr = %llu\n",
564 (unsigned long)bh,
565 (unsigned long long)bh->b_blocknr);
566
567 /* We aren't guaranteed to have the superblock here - but if we
568 * don't, it'll just crash. */
569 ocfs2_error(bh->b_assoc_map->host->i_sb,
570 "JBD2 has aborted our journal, ocfs2 cannot continue\n");
571}
572
573static struct ocfs2_triggers di_triggers = {
574 .ot_triggers = {
Jan Kara13ceef02010-07-14 07:56:33 +0200575 .t_frozen = ocfs2_frozen_trigger,
Joel Becker50655ae2008-09-11 15:53:07 -0700576 .t_abort = ocfs2_abort_trigger,
577 },
578 .ot_offset = offsetof(struct ocfs2_dinode, i_check),
579};
580
581static struct ocfs2_triggers eb_triggers = {
582 .ot_triggers = {
Jan Kara13ceef02010-07-14 07:56:33 +0200583 .t_frozen = ocfs2_frozen_trigger,
Joel Becker50655ae2008-09-11 15:53:07 -0700584 .t_abort = ocfs2_abort_trigger,
585 },
586 .ot_offset = offsetof(struct ocfs2_extent_block, h_check),
587};
588
Tao Ma93c97082009-08-18 11:19:20 +0800589static struct ocfs2_triggers rb_triggers = {
590 .ot_triggers = {
Jan Kara13ceef02010-07-14 07:56:33 +0200591 .t_frozen = ocfs2_frozen_trigger,
Tao Ma93c97082009-08-18 11:19:20 +0800592 .t_abort = ocfs2_abort_trigger,
593 },
594 .ot_offset = offsetof(struct ocfs2_refcount_block, rf_check),
595};
596
Joel Becker50655ae2008-09-11 15:53:07 -0700597static struct ocfs2_triggers gd_triggers = {
598 .ot_triggers = {
Jan Kara13ceef02010-07-14 07:56:33 +0200599 .t_frozen = ocfs2_frozen_trigger,
Joel Becker50655ae2008-09-11 15:53:07 -0700600 .t_abort = ocfs2_abort_trigger,
601 },
602 .ot_offset = offsetof(struct ocfs2_group_desc, bg_check),
603};
604
Joel Beckerc175a512008-12-10 17:58:22 -0800605static struct ocfs2_triggers db_triggers = {
606 .ot_triggers = {
Jan Kara13ceef02010-07-14 07:56:33 +0200607 .t_frozen = ocfs2_db_frozen_trigger,
Joel Beckerc175a512008-12-10 17:58:22 -0800608 .t_abort = ocfs2_abort_trigger,
609 },
610};
611
Joel Becker50655ae2008-09-11 15:53:07 -0700612static struct ocfs2_triggers xb_triggers = {
613 .ot_triggers = {
Jan Kara13ceef02010-07-14 07:56:33 +0200614 .t_frozen = ocfs2_frozen_trigger,
Joel Becker50655ae2008-09-11 15:53:07 -0700615 .t_abort = ocfs2_abort_trigger,
616 },
617 .ot_offset = offsetof(struct ocfs2_xattr_block, xb_check),
618};
619
620static struct ocfs2_triggers dq_triggers = {
621 .ot_triggers = {
Jan Kara13ceef02010-07-14 07:56:33 +0200622 .t_frozen = ocfs2_dq_frozen_trigger,
Joel Becker50655ae2008-09-11 15:53:07 -0700623 .t_abort = ocfs2_abort_trigger,
624 },
625};
626
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800627static struct ocfs2_triggers dr_triggers = {
628 .ot_triggers = {
Jan Kara13ceef02010-07-14 07:56:33 +0200629 .t_frozen = ocfs2_frozen_trigger,
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800630 .t_abort = ocfs2_abort_trigger,
631 },
632 .ot_offset = offsetof(struct ocfs2_dx_root_block, dr_check),
633};
634
635static struct ocfs2_triggers dl_triggers = {
636 .ot_triggers = {
Jan Kara13ceef02010-07-14 07:56:33 +0200637 .t_frozen = ocfs2_frozen_trigger,
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800638 .t_abort = ocfs2_abort_trigger,
639 },
640 .ot_offset = offsetof(struct ocfs2_dx_leaf, dl_check),
641};
642
Joel Becker50655ae2008-09-11 15:53:07 -0700643static int __ocfs2_journal_access(handle_t *handle,
Joel Becker0cf2f762009-02-12 16:41:25 -0800644 struct ocfs2_caching_info *ci,
Joel Becker50655ae2008-09-11 15:53:07 -0700645 struct buffer_head *bh,
646 struct ocfs2_triggers *triggers,
647 int type)
Mark Fashehccd979b2005-12-15 14:31:24 -0800648{
649 int status;
Joel Becker0cf2f762009-02-12 16:41:25 -0800650 struct ocfs2_super *osb =
651 OCFS2_SB(ocfs2_metadata_cache_get_super(ci));
Mark Fashehccd979b2005-12-15 14:31:24 -0800652
Joel Becker0cf2f762009-02-12 16:41:25 -0800653 BUG_ON(!ci || !ci->ci_ops);
Mark Fashehccd979b2005-12-15 14:31:24 -0800654 BUG_ON(!handle);
655 BUG_ON(!bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800656
Tao Mab4107952011-02-24 14:15:35 +0800657 trace_ocfs2_journal_access(
658 (unsigned long long)ocfs2_metadata_cache_owner(ci),
659 (unsigned long long)bh->b_blocknr, type, bh->b_size);
Mark Fashehccd979b2005-12-15 14:31:24 -0800660
661 /* we can safely remove this assertion after testing. */
662 if (!buffer_uptodate(bh)) {
663 mlog(ML_ERROR, "giving me a buffer that's not uptodate!\n");
664 mlog(ML_ERROR, "b_blocknr=%llu\n",
665 (unsigned long long)bh->b_blocknr);
666 BUG();
667 }
668
Joel Becker0cf2f762009-02-12 16:41:25 -0800669 /* Set the current transaction information on the ci so
Mark Fashehccd979b2005-12-15 14:31:24 -0800670 * that the locking code knows whether it can drop it's locks
Joel Becker0cf2f762009-02-12 16:41:25 -0800671 * on this ci or not. We're protected from the commit
Mark Fashehccd979b2005-12-15 14:31:24 -0800672 * thread updating the current transaction id until
673 * ocfs2_commit_trans() because ocfs2_start_trans() took
674 * j_trans_barrier for us. */
Joel Becker0cf2f762009-02-12 16:41:25 -0800675 ocfs2_set_ci_lock_trans(osb->journal, ci);
Mark Fashehccd979b2005-12-15 14:31:24 -0800676
Joel Becker0cf2f762009-02-12 16:41:25 -0800677 ocfs2_metadata_cache_io_lock(ci);
Mark Fashehccd979b2005-12-15 14:31:24 -0800678 switch (type) {
679 case OCFS2_JOURNAL_ACCESS_CREATE:
680 case OCFS2_JOURNAL_ACCESS_WRITE:
Joel Becker2b4e30f2008-09-03 20:03:41 -0700681 status = jbd2_journal_get_write_access(handle, bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800682 break;
683
684 case OCFS2_JOURNAL_ACCESS_UNDO:
Joel Becker2b4e30f2008-09-03 20:03:41 -0700685 status = jbd2_journal_get_undo_access(handle, bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800686 break;
687
688 default:
689 status = -EINVAL;
André Goddard Rosaaf901ca2009-11-14 13:09:05 -0200690 mlog(ML_ERROR, "Unknown access type!\n");
Mark Fashehccd979b2005-12-15 14:31:24 -0800691 }
Joel Becker0cf2f762009-02-12 16:41:25 -0800692 if (!status && ocfs2_meta_ecc(osb) && triggers)
Joel Becker50655ae2008-09-11 15:53:07 -0700693 jbd2_journal_set_triggers(bh, &triggers->ot_triggers);
Joel Becker0cf2f762009-02-12 16:41:25 -0800694 ocfs2_metadata_cache_io_unlock(ci);
Mark Fashehccd979b2005-12-15 14:31:24 -0800695
696 if (status < 0)
697 mlog(ML_ERROR, "Error %d getting %d access to buffer!\n",
698 status, type);
699
Mark Fashehccd979b2005-12-15 14:31:24 -0800700 return status;
701}
702
Joel Becker0cf2f762009-02-12 16:41:25 -0800703int ocfs2_journal_access_di(handle_t *handle, struct ocfs2_caching_info *ci,
Joel Becker50655ae2008-09-11 15:53:07 -0700704 struct buffer_head *bh, int type)
705{
Joel Becker0cf2f762009-02-12 16:41:25 -0800706 return __ocfs2_journal_access(handle, ci, bh, &di_triggers, type);
Joel Becker50655ae2008-09-11 15:53:07 -0700707}
708
Joel Becker0cf2f762009-02-12 16:41:25 -0800709int ocfs2_journal_access_eb(handle_t *handle, struct ocfs2_caching_info *ci,
Joel Becker50655ae2008-09-11 15:53:07 -0700710 struct buffer_head *bh, int type)
711{
Joel Becker0cf2f762009-02-12 16:41:25 -0800712 return __ocfs2_journal_access(handle, ci, bh, &eb_triggers, type);
Joel Becker50655ae2008-09-11 15:53:07 -0700713}
714
Tao Ma93c97082009-08-18 11:19:20 +0800715int ocfs2_journal_access_rb(handle_t *handle, struct ocfs2_caching_info *ci,
716 struct buffer_head *bh, int type)
717{
718 return __ocfs2_journal_access(handle, ci, bh, &rb_triggers,
719 type);
720}
721
Joel Becker0cf2f762009-02-12 16:41:25 -0800722int ocfs2_journal_access_gd(handle_t *handle, struct ocfs2_caching_info *ci,
Joel Becker50655ae2008-09-11 15:53:07 -0700723 struct buffer_head *bh, int type)
724{
Joel Becker0cf2f762009-02-12 16:41:25 -0800725 return __ocfs2_journal_access(handle, ci, bh, &gd_triggers, type);
Joel Becker50655ae2008-09-11 15:53:07 -0700726}
727
Joel Becker0cf2f762009-02-12 16:41:25 -0800728int ocfs2_journal_access_db(handle_t *handle, struct ocfs2_caching_info *ci,
Joel Becker50655ae2008-09-11 15:53:07 -0700729 struct buffer_head *bh, int type)
730{
Joel Becker0cf2f762009-02-12 16:41:25 -0800731 return __ocfs2_journal_access(handle, ci, bh, &db_triggers, type);
Joel Becker50655ae2008-09-11 15:53:07 -0700732}
733
Joel Becker0cf2f762009-02-12 16:41:25 -0800734int ocfs2_journal_access_xb(handle_t *handle, struct ocfs2_caching_info *ci,
Joel Becker50655ae2008-09-11 15:53:07 -0700735 struct buffer_head *bh, int type)
736{
Joel Becker0cf2f762009-02-12 16:41:25 -0800737 return __ocfs2_journal_access(handle, ci, bh, &xb_triggers, type);
Joel Becker50655ae2008-09-11 15:53:07 -0700738}
739
Joel Becker0cf2f762009-02-12 16:41:25 -0800740int ocfs2_journal_access_dq(handle_t *handle, struct ocfs2_caching_info *ci,
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800741 struct buffer_head *bh, int type)
742{
Joel Becker0cf2f762009-02-12 16:41:25 -0800743 return __ocfs2_journal_access(handle, ci, bh, &dq_triggers, type);
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800744}
745
Joel Becker0cf2f762009-02-12 16:41:25 -0800746int ocfs2_journal_access_dr(handle_t *handle, struct ocfs2_caching_info *ci,
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800747 struct buffer_head *bh, int type)
748{
Joel Becker0cf2f762009-02-12 16:41:25 -0800749 return __ocfs2_journal_access(handle, ci, bh, &dr_triggers, type);
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800750}
751
Joel Becker0cf2f762009-02-12 16:41:25 -0800752int ocfs2_journal_access_dl(handle_t *handle, struct ocfs2_caching_info *ci,
753 struct buffer_head *bh, int type)
754{
755 return __ocfs2_journal_access(handle, ci, bh, &dl_triggers, type);
756}
757
758int ocfs2_journal_access(handle_t *handle, struct ocfs2_caching_info *ci,
Joel Becker50655ae2008-09-11 15:53:07 -0700759 struct buffer_head *bh, int type)
760{
Joel Becker0cf2f762009-02-12 16:41:25 -0800761 return __ocfs2_journal_access(handle, ci, bh, NULL, type);
Joel Becker50655ae2008-09-11 15:53:07 -0700762}
763
Joel Beckerec20cec2010-03-19 14:13:52 -0700764void ocfs2_journal_dirty(handle_t *handle, struct buffer_head *bh)
Mark Fashehccd979b2005-12-15 14:31:24 -0800765{
766 int status;
767
Tao Mab4107952011-02-24 14:15:35 +0800768 trace_ocfs2_journal_dirty((unsigned long long)bh->b_blocknr);
Mark Fashehccd979b2005-12-15 14:31:24 -0800769
Joel Becker2b4e30f2008-09-03 20:03:41 -0700770 status = jbd2_journal_dirty_metadata(handle, bh);
Joel Beckerec20cec2010-03-19 14:13:52 -0700771 BUG_ON(status);
Mark Fashehccd979b2005-12-15 14:31:24 -0800772}
773
Joel Becker2b4e30f2008-09-03 20:03:41 -0700774#define OCFS2_DEFAULT_COMMIT_INTERVAL (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE)
Mark Fashehccd979b2005-12-15 14:31:24 -0800775
776void ocfs2_set_journal_params(struct ocfs2_super *osb)
777{
778 journal_t *journal = osb->journal->j_journal;
Mark Fashehd147b3d2007-11-07 14:40:36 -0800779 unsigned long commit_interval = OCFS2_DEFAULT_COMMIT_INTERVAL;
780
781 if (osb->osb_commit_interval)
782 commit_interval = osb->osb_commit_interval;
Mark Fashehccd979b2005-12-15 14:31:24 -0800783
Theodore Ts'oa931da62010-08-03 21:35:12 -0400784 write_lock(&journal->j_state_lock);
Mark Fashehd147b3d2007-11-07 14:40:36 -0800785 journal->j_commit_interval = commit_interval;
Mark Fashehccd979b2005-12-15 14:31:24 -0800786 if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER)
Joel Becker2b4e30f2008-09-03 20:03:41 -0700787 journal->j_flags |= JBD2_BARRIER;
Mark Fashehccd979b2005-12-15 14:31:24 -0800788 else
Joel Becker2b4e30f2008-09-03 20:03:41 -0700789 journal->j_flags &= ~JBD2_BARRIER;
Theodore Ts'oa931da62010-08-03 21:35:12 -0400790 write_unlock(&journal->j_state_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -0800791}
792
793int ocfs2_journal_init(struct ocfs2_journal *journal, int *dirty)
794{
795 int status = -1;
796 struct inode *inode = NULL; /* the journal inode */
797 journal_t *j_journal = NULL;
798 struct ocfs2_dinode *di = NULL;
799 struct buffer_head *bh = NULL;
800 struct ocfs2_super *osb;
Mark Fashehe63aecb62007-10-18 15:30:42 -0700801 int inode_lock = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -0800802
Mark Fashehccd979b2005-12-15 14:31:24 -0800803 BUG_ON(!journal);
804
805 osb = journal->j_osb;
806
807 /* already have the inode for our journal */
808 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
809 osb->slot_num);
810 if (inode == NULL) {
811 status = -EACCES;
812 mlog_errno(status);
813 goto done;
814 }
815 if (is_bad_inode(inode)) {
816 mlog(ML_ERROR, "access error (bad inode)\n");
817 iput(inode);
818 inode = NULL;
819 status = -EACCES;
820 goto done;
821 }
822
823 SET_INODE_JOURNAL(inode);
824 OCFS2_I(inode)->ip_open_count++;
825
Mark Fasheh6eff5792006-01-18 10:31:47 -0800826 /* Skip recovery waits here - journal inode metadata never
827 * changes in a live cluster so it can be considered an
828 * exception to the rule. */
Mark Fashehe63aecb62007-10-18 15:30:42 -0700829 status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY);
Mark Fashehccd979b2005-12-15 14:31:24 -0800830 if (status < 0) {
831 if (status != -ERESTARTSYS)
832 mlog(ML_ERROR, "Could not get lock on journal!\n");
833 goto done;
834 }
835
Mark Fashehe63aecb62007-10-18 15:30:42 -0700836 inode_lock = 1;
Mark Fashehccd979b2005-12-15 14:31:24 -0800837 di = (struct ocfs2_dinode *)bh->b_data;
838
Junxiao Bif17c20d2013-09-11 14:19:45 -0700839 if (i_size_read(inode) < OCFS2_MIN_JOURNAL_SIZE) {
Mark Fashehccd979b2005-12-15 14:31:24 -0800840 mlog(ML_ERROR, "Journal file size (%lld) is too small!\n",
Junxiao Bif17c20d2013-09-11 14:19:45 -0700841 i_size_read(inode));
Mark Fashehccd979b2005-12-15 14:31:24 -0800842 status = -EINVAL;
843 goto done;
844 }
845
Junxiao Bif17c20d2013-09-11 14:19:45 -0700846 trace_ocfs2_journal_init(i_size_read(inode),
Tao Mab4107952011-02-24 14:15:35 +0800847 (unsigned long long)inode->i_blocks,
848 OCFS2_I(inode)->ip_clusters);
Mark Fashehccd979b2005-12-15 14:31:24 -0800849
850 /* call the kernels journal init function now */
Joel Becker2b4e30f2008-09-03 20:03:41 -0700851 j_journal = jbd2_journal_init_inode(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -0800852 if (j_journal == NULL) {
853 mlog(ML_ERROR, "Linux journal layer error\n");
854 status = -EINVAL;
855 goto done;
856 }
857
Tao Mab4107952011-02-24 14:15:35 +0800858 trace_ocfs2_journal_init_maxlen(j_journal->j_maxlen);
Mark Fashehccd979b2005-12-15 14:31:24 -0800859
860 *dirty = (le32_to_cpu(di->id1.journal1.ij_flags) &
861 OCFS2_JOURNAL_DIRTY_FL);
862
863 journal->j_journal = j_journal;
864 journal->j_inode = inode;
865 journal->j_bh = bh;
866
867 ocfs2_set_journal_params(osb);
868
869 journal->j_state = OCFS2_JOURNAL_LOADED;
870
871 status = 0;
872done:
873 if (status < 0) {
Mark Fashehe63aecb62007-10-18 15:30:42 -0700874 if (inode_lock)
875 ocfs2_inode_unlock(inode, 1);
Mark Fasheha81cb882008-10-07 14:25:16 -0700876 brelse(bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800877 if (inode) {
878 OCFS2_I(inode)->ip_open_count--;
879 iput(inode);
880 }
881 }
882
Mark Fashehccd979b2005-12-15 14:31:24 -0800883 return status;
884}
885
Sunil Mushran539d8262008-07-14 17:31:10 -0700886static void ocfs2_bump_recovery_generation(struct ocfs2_dinode *di)
887{
888 le32_add_cpu(&(di->id1.journal1.ij_recovery_generation), 1);
889}
890
891static u32 ocfs2_get_recovery_generation(struct ocfs2_dinode *di)
892{
893 return le32_to_cpu(di->id1.journal1.ij_recovery_generation);
894}
895
Mark Fashehccd979b2005-12-15 14:31:24 -0800896static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
Sunil Mushran539d8262008-07-14 17:31:10 -0700897 int dirty, int replayed)
Mark Fashehccd979b2005-12-15 14:31:24 -0800898{
899 int status;
900 unsigned int flags;
901 struct ocfs2_journal *journal = osb->journal;
902 struct buffer_head *bh = journal->j_bh;
903 struct ocfs2_dinode *fe;
904
Mark Fashehccd979b2005-12-15 14:31:24 -0800905 fe = (struct ocfs2_dinode *)bh->b_data;
Joel Becker10995aa2008-11-13 14:49:12 -0800906
907 /* The journal bh on the osb always comes from ocfs2_journal_init()
908 * and was validated there inside ocfs2_inode_lock_full(). It's a
909 * code bug if we mess it up. */
910 BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
Mark Fashehccd979b2005-12-15 14:31:24 -0800911
912 flags = le32_to_cpu(fe->id1.journal1.ij_flags);
913 if (dirty)
914 flags |= OCFS2_JOURNAL_DIRTY_FL;
915 else
916 flags &= ~OCFS2_JOURNAL_DIRTY_FL;
917 fe->id1.journal1.ij_flags = cpu_to_le32(flags);
918
Sunil Mushran539d8262008-07-14 17:31:10 -0700919 if (replayed)
920 ocfs2_bump_recovery_generation(fe);
921
Joel Becker13723d02008-10-17 19:25:01 -0700922 ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check);
Joel Becker8cb471e2009-02-10 20:00:41 -0800923 status = ocfs2_write_block(osb, bh, INODE_CACHE(journal->j_inode));
Mark Fashehccd979b2005-12-15 14:31:24 -0800924 if (status < 0)
925 mlog_errno(status);
926
Mark Fashehccd979b2005-12-15 14:31:24 -0800927 return status;
928}
929
930/*
931 * If the journal has been kmalloc'd it needs to be freed after this
932 * call.
933 */
934void ocfs2_journal_shutdown(struct ocfs2_super *osb)
935{
936 struct ocfs2_journal *journal = NULL;
937 int status = 0;
938 struct inode *inode = NULL;
939 int num_running_trans = 0;
940
Eric Sesterhenn / snakebyteebdec832006-01-27 10:32:52 +0100941 BUG_ON(!osb);
Mark Fashehccd979b2005-12-15 14:31:24 -0800942
943 journal = osb->journal;
944 if (!journal)
945 goto done;
946
947 inode = journal->j_inode;
948
949 if (journal->j_state != OCFS2_JOURNAL_LOADED)
950 goto done;
951
Joel Becker2b4e30f2008-09-03 20:03:41 -0700952 /* need to inc inode use count - jbd2_journal_destroy will iput. */
Mark Fashehccd979b2005-12-15 14:31:24 -0800953 if (!igrab(inode))
954 BUG();
955
956 num_running_trans = atomic_read(&(osb->journal->j_num_trans));
Tao Mab4107952011-02-24 14:15:35 +0800957 trace_ocfs2_journal_shutdown(num_running_trans);
Mark Fashehccd979b2005-12-15 14:31:24 -0800958
959 /* Do a commit_cache here. It will flush our journal, *and*
960 * release any locks that are still held.
961 * set the SHUTDOWN flag and release the trans lock.
962 * the commit thread will take the trans lock for us below. */
963 journal->j_state = OCFS2_JOURNAL_IN_SHUTDOWN;
964
965 /* The OCFS2_JOURNAL_IN_SHUTDOWN will signal to commit_cache to not
966 * drop the trans_lock (which we want to hold until we
967 * completely destroy the journal. */
968 if (osb->commit_task) {
969 /* Wait for the commit thread */
Tao Mab4107952011-02-24 14:15:35 +0800970 trace_ocfs2_journal_shutdown_wait(osb->commit_task);
Mark Fashehccd979b2005-12-15 14:31:24 -0800971 kthread_stop(osb->commit_task);
972 osb->commit_task = NULL;
973 }
974
975 BUG_ON(atomic_read(&(osb->journal->j_num_trans)) != 0);
976
Sunil Mushranc271c5c2006-12-05 17:56:35 -0800977 if (ocfs2_mount_local(osb)) {
Joel Becker2b4e30f2008-09-03 20:03:41 -0700978 jbd2_journal_lock_updates(journal->j_journal);
979 status = jbd2_journal_flush(journal->j_journal);
980 jbd2_journal_unlock_updates(journal->j_journal);
Sunil Mushranc271c5c2006-12-05 17:56:35 -0800981 if (status < 0)
982 mlog_errno(status);
983 }
984
985 if (status == 0) {
986 /*
987 * Do not toggle if flush was unsuccessful otherwise
988 * will leave dirty metadata in a "clean" journal
989 */
Sunil Mushran539d8262008-07-14 17:31:10 -0700990 status = ocfs2_journal_toggle_dirty(osb, 0, 0);
Sunil Mushranc271c5c2006-12-05 17:56:35 -0800991 if (status < 0)
992 mlog_errno(status);
993 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800994
995 /* Shutdown the kernel journal system */
Joel Becker2b4e30f2008-09-03 20:03:41 -0700996 jbd2_journal_destroy(journal->j_journal);
Sunil Mushranae0dff62008-10-22 13:24:29 -0700997 journal->j_journal = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800998
999 OCFS2_I(inode)->ip_open_count--;
1000
1001 /* unlock our journal */
Mark Fashehe63aecb62007-10-18 15:30:42 -07001002 ocfs2_inode_unlock(inode, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -08001003
1004 brelse(journal->j_bh);
1005 journal->j_bh = NULL;
1006
1007 journal->j_state = OCFS2_JOURNAL_FREE;
1008
1009// up_write(&journal->j_trans_barrier);
1010done:
1011 if (inode)
1012 iput(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -08001013}
1014
1015static void ocfs2_clear_journal_error(struct super_block *sb,
1016 journal_t *journal,
1017 int slot)
1018{
1019 int olderr;
1020
Joel Becker2b4e30f2008-09-03 20:03:41 -07001021 olderr = jbd2_journal_errno(journal);
Mark Fashehccd979b2005-12-15 14:31:24 -08001022 if (olderr) {
1023 mlog(ML_ERROR, "File system error %d recorded in "
1024 "journal %u.\n", olderr, slot);
1025 mlog(ML_ERROR, "File system on device %s needs checking.\n",
1026 sb->s_id);
1027
Joel Becker2b4e30f2008-09-03 20:03:41 -07001028 jbd2_journal_ack_err(journal);
1029 jbd2_journal_clear_err(journal);
Mark Fashehccd979b2005-12-15 14:31:24 -08001030 }
1031}
1032
Sunil Mushran539d8262008-07-14 17:31:10 -07001033int ocfs2_journal_load(struct ocfs2_journal *journal, int local, int replayed)
Mark Fashehccd979b2005-12-15 14:31:24 -08001034{
1035 int status = 0;
1036 struct ocfs2_super *osb;
1037
Julia Lawallb1f35502008-03-04 15:21:05 -08001038 BUG_ON(!journal);
Mark Fashehccd979b2005-12-15 14:31:24 -08001039
1040 osb = journal->j_osb;
1041
Joel Becker2b4e30f2008-09-03 20:03:41 -07001042 status = jbd2_journal_load(journal->j_journal);
Mark Fashehccd979b2005-12-15 14:31:24 -08001043 if (status < 0) {
1044 mlog(ML_ERROR, "Failed to load journal!\n");
1045 goto done;
1046 }
1047
1048 ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num);
1049
Sunil Mushran539d8262008-07-14 17:31:10 -07001050 status = ocfs2_journal_toggle_dirty(osb, 1, replayed);
Mark Fashehccd979b2005-12-15 14:31:24 -08001051 if (status < 0) {
1052 mlog_errno(status);
1053 goto done;
1054 }
1055
1056 /* Launch the commit thread */
Sunil Mushranc271c5c2006-12-05 17:56:35 -08001057 if (!local) {
1058 osb->commit_task = kthread_run(ocfs2_commit_thread, osb,
1059 "ocfs2cmt");
1060 if (IS_ERR(osb->commit_task)) {
1061 status = PTR_ERR(osb->commit_task);
1062 osb->commit_task = NULL;
1063 mlog(ML_ERROR, "unable to launch ocfs2commit thread, "
1064 "error=%d", status);
1065 goto done;
1066 }
1067 } else
Mark Fashehccd979b2005-12-15 14:31:24 -08001068 osb->commit_task = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -08001069
1070done:
Mark Fashehccd979b2005-12-15 14:31:24 -08001071 return status;
1072}
1073
1074
1075/* 'full' flag tells us whether we clear out all blocks or if we just
1076 * mark the journal clean */
1077int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full)
1078{
1079 int status;
1080
Eric Sesterhenn / snakebyteebdec832006-01-27 10:32:52 +01001081 BUG_ON(!journal);
Mark Fashehccd979b2005-12-15 14:31:24 -08001082
Joel Becker2b4e30f2008-09-03 20:03:41 -07001083 status = jbd2_journal_wipe(journal->j_journal, full);
Mark Fashehccd979b2005-12-15 14:31:24 -08001084 if (status < 0) {
1085 mlog_errno(status);
1086 goto bail;
1087 }
1088
Sunil Mushran539d8262008-07-14 17:31:10 -07001089 status = ocfs2_journal_toggle_dirty(journal->j_osb, 0, 0);
Mark Fashehccd979b2005-12-15 14:31:24 -08001090 if (status < 0)
1091 mlog_errno(status);
1092
1093bail:
Mark Fashehccd979b2005-12-15 14:31:24 -08001094 return status;
1095}
1096
Joel Becker553abd02008-02-01 12:03:57 -08001097static int ocfs2_recovery_completed(struct ocfs2_super *osb)
1098{
1099 int empty;
1100 struct ocfs2_recovery_map *rm = osb->recovery_map;
1101
1102 spin_lock(&osb->osb_lock);
1103 empty = (rm->rm_used == 0);
1104 spin_unlock(&osb->osb_lock);
1105
1106 return empty;
1107}
1108
1109void ocfs2_wait_for_recovery(struct ocfs2_super *osb)
1110{
1111 wait_event(osb->recovery_event, ocfs2_recovery_completed(osb));
1112}
1113
Mark Fashehccd979b2005-12-15 14:31:24 -08001114/*
1115 * JBD Might read a cached version of another nodes journal file. We
1116 * don't want this as this file changes often and we get no
1117 * notification on those changes. The only way to be sure that we've
1118 * got the most up to date version of those blocks then is to force
1119 * read them off disk. Just searching through the buffer cache won't
1120 * work as there may be pages backing this file which are still marked
1121 * up to date. We know things can't change on this file underneath us
1122 * as we have the lock by now :)
1123 */
1124static int ocfs2_force_read_journal(struct inode *inode)
1125{
1126 int status = 0;
Mark Fasheh4f902c32007-03-09 16:26:50 -08001127 int i;
Mark Fasheh8110b072007-03-22 16:53:23 -07001128 u64 v_blkno, p_blkno, p_blocks, num_blocks;
Mark Fasheh4f902c32007-03-09 16:26:50 -08001129#define CONCURRENT_JOURNAL_FILL 32ULL
Mark Fashehccd979b2005-12-15 14:31:24 -08001130 struct buffer_head *bhs[CONCURRENT_JOURNAL_FILL];
1131
Mark Fashehccd979b2005-12-15 14:31:24 -08001132 memset(bhs, 0, sizeof(struct buffer_head *) * CONCURRENT_JOURNAL_FILL);
1133
Junxiao Bif17c20d2013-09-11 14:19:45 -07001134 num_blocks = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode));
Mark Fashehccd979b2005-12-15 14:31:24 -08001135 v_blkno = 0;
Mark Fasheh8110b072007-03-22 16:53:23 -07001136 while (v_blkno < num_blocks) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001137 status = ocfs2_extent_map_get_blocks(inode, v_blkno,
Mark Fasheh49cb8d22007-03-09 16:21:46 -08001138 &p_blkno, &p_blocks, NULL);
Mark Fashehccd979b2005-12-15 14:31:24 -08001139 if (status < 0) {
1140 mlog_errno(status);
1141 goto bail;
1142 }
1143
1144 if (p_blocks > CONCURRENT_JOURNAL_FILL)
1145 p_blocks = CONCURRENT_JOURNAL_FILL;
1146
Mark Fashehdd4a2c22006-04-12 14:24:05 -07001147 /* We are reading journal data which should not
1148 * be put in the uptodate cache */
Joel Beckerda1e9092008-10-09 17:20:29 -07001149 status = ocfs2_read_blocks_sync(OCFS2_SB(inode->i_sb),
1150 p_blkno, p_blocks, bhs);
Mark Fashehccd979b2005-12-15 14:31:24 -08001151 if (status < 0) {
1152 mlog_errno(status);
1153 goto bail;
1154 }
1155
1156 for(i = 0; i < p_blocks; i++) {
1157 brelse(bhs[i]);
1158 bhs[i] = NULL;
1159 }
1160
1161 v_blkno += p_blocks;
1162 }
1163
1164bail:
1165 for(i = 0; i < CONCURRENT_JOURNAL_FILL; i++)
Mark Fasheha81cb882008-10-07 14:25:16 -07001166 brelse(bhs[i]);
Mark Fashehccd979b2005-12-15 14:31:24 -08001167 return status;
1168}
1169
1170struct ocfs2_la_recovery_item {
1171 struct list_head lri_list;
1172 int lri_slot;
1173 struct ocfs2_dinode *lri_la_dinode;
1174 struct ocfs2_dinode *lri_tl_dinode;
Jan Kara22053632008-10-20 23:50:38 +02001175 struct ocfs2_quota_recovery *lri_qrec;
Mark Fashehccd979b2005-12-15 14:31:24 -08001176};
1177
1178/* Does the second half of the recovery process. By this point, the
1179 * node is marked clean and can actually be considered recovered,
1180 * hence it's no longer in the recovery map, but there's still some
1181 * cleanup we can do which shouldn't happen within the recovery thread
1182 * as locking in that context becomes very difficult if we are to take
1183 * recovering nodes into account.
1184 *
1185 * NOTE: This function can and will sleep on recovery of other nodes
1186 * during cluster locking, just like any other ocfs2 process.
1187 */
David Howellsc4028952006-11-22 14:57:56 +00001188void ocfs2_complete_recovery(struct work_struct *work)
Mark Fashehccd979b2005-12-15 14:31:24 -08001189{
Tao Mab4107952011-02-24 14:15:35 +08001190 int ret = 0;
David Howellsc4028952006-11-22 14:57:56 +00001191 struct ocfs2_journal *journal =
1192 container_of(work, struct ocfs2_journal, j_recovery_work);
1193 struct ocfs2_super *osb = journal->j_osb;
Mark Fashehccd979b2005-12-15 14:31:24 -08001194 struct ocfs2_dinode *la_dinode, *tl_dinode;
Christoph Hellwig800deef2007-05-17 16:03:13 +02001195 struct ocfs2_la_recovery_item *item, *n;
Jan Kara22053632008-10-20 23:50:38 +02001196 struct ocfs2_quota_recovery *qrec;
Mark Fashehccd979b2005-12-15 14:31:24 -08001197 LIST_HEAD(tmp_la_list);
1198
Tao Mab4107952011-02-24 14:15:35 +08001199 trace_ocfs2_complete_recovery(
1200 (unsigned long long)OCFS2_I(journal->j_inode)->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001201
1202 spin_lock(&journal->j_lock);
1203 list_splice_init(&journal->j_la_cleanups, &tmp_la_list);
1204 spin_unlock(&journal->j_lock);
1205
Christoph Hellwig800deef2007-05-17 16:03:13 +02001206 list_for_each_entry_safe(item, n, &tmp_la_list, lri_list) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001207 list_del_init(&item->lri_list);
1208
Jan Kara19ece542008-08-21 20:13:17 +02001209 ocfs2_wait_on_quotas(osb);
1210
Mark Fashehccd979b2005-12-15 14:31:24 -08001211 la_dinode = item->lri_la_dinode;
Tao Mab4107952011-02-24 14:15:35 +08001212 tl_dinode = item->lri_tl_dinode;
1213 qrec = item->lri_qrec;
Mark Fashehccd979b2005-12-15 14:31:24 -08001214
Tao Mab4107952011-02-24 14:15:35 +08001215 trace_ocfs2_complete_recovery_slot(item->lri_slot,
1216 la_dinode ? le64_to_cpu(la_dinode->i_blkno) : 0,
1217 tl_dinode ? le64_to_cpu(tl_dinode->i_blkno) : 0,
1218 qrec);
1219
1220 if (la_dinode) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001221 ret = ocfs2_complete_local_alloc_recovery(osb,
1222 la_dinode);
1223 if (ret < 0)
1224 mlog_errno(ret);
1225
1226 kfree(la_dinode);
1227 }
1228
Mark Fashehccd979b2005-12-15 14:31:24 -08001229 if (tl_dinode) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001230 ret = ocfs2_complete_truncate_log_recovery(osb,
1231 tl_dinode);
1232 if (ret < 0)
1233 mlog_errno(ret);
1234
1235 kfree(tl_dinode);
1236 }
1237
1238 ret = ocfs2_recover_orphans(osb, item->lri_slot);
1239 if (ret < 0)
1240 mlog_errno(ret);
1241
Jan Kara22053632008-10-20 23:50:38 +02001242 if (qrec) {
Jan Kara22053632008-10-20 23:50:38 +02001243 ret = ocfs2_finish_quota_recovery(osb, qrec,
1244 item->lri_slot);
1245 if (ret < 0)
1246 mlog_errno(ret);
1247 /* Recovery info is already freed now */
1248 }
1249
Mark Fashehccd979b2005-12-15 14:31:24 -08001250 kfree(item);
1251 }
1252
Tao Mab4107952011-02-24 14:15:35 +08001253 trace_ocfs2_complete_recovery_end(ret);
Mark Fashehccd979b2005-12-15 14:31:24 -08001254}
1255
1256/* NOTE: This function always eats your references to la_dinode and
1257 * tl_dinode, either manually on error, or by passing them to
1258 * ocfs2_complete_recovery */
1259static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal,
1260 int slot_num,
1261 struct ocfs2_dinode *la_dinode,
Jan Kara22053632008-10-20 23:50:38 +02001262 struct ocfs2_dinode *tl_dinode,
1263 struct ocfs2_quota_recovery *qrec)
Mark Fashehccd979b2005-12-15 14:31:24 -08001264{
1265 struct ocfs2_la_recovery_item *item;
1266
Sunil Mushranafae00ab2006-04-12 14:37:00 -07001267 item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS);
Mark Fashehccd979b2005-12-15 14:31:24 -08001268 if (!item) {
1269 /* Though we wish to avoid it, we are in fact safe in
1270 * skipping local alloc cleanup as fsck.ocfs2 is more
1271 * than capable of reclaiming unused space. */
Tim Gardnerd787ab02013-02-21 16:42:44 -08001272 kfree(la_dinode);
1273 kfree(tl_dinode);
Mark Fashehccd979b2005-12-15 14:31:24 -08001274
Jan Kara22053632008-10-20 23:50:38 +02001275 if (qrec)
1276 ocfs2_free_quota_recovery(qrec);
1277
Mark Fashehccd979b2005-12-15 14:31:24 -08001278 mlog_errno(-ENOMEM);
1279 return;
1280 }
1281
1282 INIT_LIST_HEAD(&item->lri_list);
1283 item->lri_la_dinode = la_dinode;
1284 item->lri_slot = slot_num;
1285 item->lri_tl_dinode = tl_dinode;
Jan Kara22053632008-10-20 23:50:38 +02001286 item->lri_qrec = qrec;
Mark Fashehccd979b2005-12-15 14:31:24 -08001287
1288 spin_lock(&journal->j_lock);
1289 list_add_tail(&item->lri_list, &journal->j_la_cleanups);
1290 queue_work(ocfs2_wq, &journal->j_recovery_work);
1291 spin_unlock(&journal->j_lock);
1292}
1293
1294/* Called by the mount code to queue recovery the last part of
Srinivas Eeda9140db02009-03-06 14:21:46 -08001295 * recovery for it's own and offline slot(s). */
Mark Fashehccd979b2005-12-15 14:31:24 -08001296void ocfs2_complete_mount_recovery(struct ocfs2_super *osb)
1297{
1298 struct ocfs2_journal *journal = osb->journal;
1299
Sunil Mushran10b3dd72011-05-04 10:28:02 -07001300 if (ocfs2_is_hard_readonly(osb))
1301 return;
1302
Srinivas Eeda9140db02009-03-06 14:21:46 -08001303 /* No need to queue up our truncate_log as regular cleanup will catch
1304 * that */
1305 ocfs2_queue_recovery_completion(journal, osb->slot_num,
1306 osb->local_alloc_copy, NULL, NULL);
1307 ocfs2_schedule_truncate_log_flush(osb, 0);
Mark Fashehccd979b2005-12-15 14:31:24 -08001308
Srinivas Eeda9140db02009-03-06 14:21:46 -08001309 osb->local_alloc_copy = NULL;
1310 osb->dirty = 0;
1311
1312 /* queue to recover orphan slots for all offline slots */
1313 ocfs2_replay_map_set_state(osb, REPLAY_NEEDED);
1314 ocfs2_queue_replay_slots(osb);
1315 ocfs2_free_replay_slots(osb);
Mark Fashehccd979b2005-12-15 14:31:24 -08001316}
1317
Jan Kara22053632008-10-20 23:50:38 +02001318void ocfs2_complete_quota_recovery(struct ocfs2_super *osb)
1319{
1320 if (osb->quota_rec) {
1321 ocfs2_queue_recovery_completion(osb->journal,
1322 osb->slot_num,
1323 NULL,
1324 NULL,
1325 osb->quota_rec);
1326 osb->quota_rec = NULL;
1327 }
1328}
1329
Mark Fashehccd979b2005-12-15 14:31:24 -08001330static int __ocfs2_recovery_thread(void *arg)
1331{
Jan Kara22053632008-10-20 23:50:38 +02001332 int status, node_num, slot_num;
Mark Fashehccd979b2005-12-15 14:31:24 -08001333 struct ocfs2_super *osb = arg;
Joel Becker553abd02008-02-01 12:03:57 -08001334 struct ocfs2_recovery_map *rm = osb->recovery_map;
Jan Kara22053632008-10-20 23:50:38 +02001335 int *rm_quota = NULL;
1336 int rm_quota_used = 0, i;
1337 struct ocfs2_quota_recovery *qrec;
Mark Fashehccd979b2005-12-15 14:31:24 -08001338
Mark Fashehccd979b2005-12-15 14:31:24 -08001339 status = ocfs2_wait_on_mount(osb);
1340 if (status < 0) {
1341 goto bail;
1342 }
1343
Jan Kara22053632008-10-20 23:50:38 +02001344 rm_quota = kzalloc(osb->max_slots * sizeof(int), GFP_NOFS);
1345 if (!rm_quota) {
1346 status = -ENOMEM;
1347 goto bail;
1348 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001349restart:
1350 status = ocfs2_super_lock(osb, 1);
1351 if (status < 0) {
1352 mlog_errno(status);
1353 goto bail;
1354 }
1355
Srinivas Eeda9140db02009-03-06 14:21:46 -08001356 status = ocfs2_compute_replay_slots(osb);
1357 if (status < 0)
1358 mlog_errno(status);
1359
1360 /* queue recovery for our own slot */
1361 ocfs2_queue_recovery_completion(osb->journal, osb->slot_num, NULL,
1362 NULL, NULL);
1363
Joel Becker553abd02008-02-01 12:03:57 -08001364 spin_lock(&osb->osb_lock);
1365 while (rm->rm_used) {
1366 /* It's always safe to remove entry zero, as we won't
1367 * clear it until ocfs2_recover_node() has succeeded. */
1368 node_num = rm->rm_entries[0];
1369 spin_unlock(&osb->osb_lock);
Jan Kara22053632008-10-20 23:50:38 +02001370 slot_num = ocfs2_node_num_to_slot(osb, node_num);
Tao Mab4107952011-02-24 14:15:35 +08001371 trace_ocfs2_recovery_thread_node(node_num, slot_num);
Jan Kara22053632008-10-20 23:50:38 +02001372 if (slot_num == -ENOENT) {
1373 status = 0;
Jan Kara22053632008-10-20 23:50:38 +02001374 goto skip_recovery;
1375 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001376
Jan Kara22053632008-10-20 23:50:38 +02001377 /* It is a bit subtle with quota recovery. We cannot do it
1378 * immediately because we have to obtain cluster locks from
1379 * quota files and we also don't want to just skip it because
1380 * then quota usage would be out of sync until some node takes
1381 * the slot. So we remember which nodes need quota recovery
1382 * and when everything else is done, we recover quotas. */
1383 for (i = 0; i < rm_quota_used && rm_quota[i] != slot_num; i++);
1384 if (i == rm_quota_used)
1385 rm_quota[rm_quota_used++] = slot_num;
1386
1387 status = ocfs2_recover_node(osb, node_num, slot_num);
1388skip_recovery:
Joel Becker553abd02008-02-01 12:03:57 -08001389 if (!status) {
1390 ocfs2_recovery_map_clear(osb, node_num);
1391 } else {
Mark Fashehccd979b2005-12-15 14:31:24 -08001392 mlog(ML_ERROR,
1393 "Error %d recovering node %d on device (%u,%u)!\n",
1394 status, node_num,
1395 MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1396 mlog(ML_ERROR, "Volume requires unmount.\n");
Mark Fashehccd979b2005-12-15 14:31:24 -08001397 }
1398
Joel Becker553abd02008-02-01 12:03:57 -08001399 spin_lock(&osb->osb_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -08001400 }
Joel Becker553abd02008-02-01 12:03:57 -08001401 spin_unlock(&osb->osb_lock);
Tao Mab4107952011-02-24 14:15:35 +08001402 trace_ocfs2_recovery_thread_end(status);
Joel Becker553abd02008-02-01 12:03:57 -08001403
Sunil Mushran539d8262008-07-14 17:31:10 -07001404 /* Refresh all journal recovery generations from disk */
1405 status = ocfs2_check_journals_nolocks(osb);
1406 status = (status == -EROFS) ? 0 : status;
1407 if (status < 0)
1408 mlog_errno(status);
1409
Jan Kara22053632008-10-20 23:50:38 +02001410 /* Now it is right time to recover quotas... We have to do this under
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001411 * superblock lock so that no one can start using the slot (and crash)
Jan Kara22053632008-10-20 23:50:38 +02001412 * before we recover it */
1413 for (i = 0; i < rm_quota_used; i++) {
1414 qrec = ocfs2_begin_quota_recovery(osb, rm_quota[i]);
1415 if (IS_ERR(qrec)) {
1416 status = PTR_ERR(qrec);
1417 mlog_errno(status);
1418 continue;
1419 }
1420 ocfs2_queue_recovery_completion(osb->journal, rm_quota[i],
1421 NULL, NULL, qrec);
1422 }
1423
Mark Fashehccd979b2005-12-15 14:31:24 -08001424 ocfs2_super_unlock(osb, 1);
1425
Srinivas Eeda9140db02009-03-06 14:21:46 -08001426 /* queue recovery for offline slots */
1427 ocfs2_queue_replay_slots(osb);
Mark Fashehccd979b2005-12-15 14:31:24 -08001428
1429bail:
Arjan van de Venc74ec2f2006-01-13 21:54:23 -08001430 mutex_lock(&osb->recovery_lock);
Joel Becker553abd02008-02-01 12:03:57 -08001431 if (!status && !ocfs2_recovery_completed(osb)) {
Arjan van de Venc74ec2f2006-01-13 21:54:23 -08001432 mutex_unlock(&osb->recovery_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -08001433 goto restart;
1434 }
1435
Srinivas Eeda9140db02009-03-06 14:21:46 -08001436 ocfs2_free_replay_slots(osb);
Mark Fashehccd979b2005-12-15 14:31:24 -08001437 osb->recovery_thread_task = NULL;
1438 mb(); /* sync with ocfs2_recovery_thread_running */
1439 wake_up(&osb->recovery_event);
1440
Arjan van de Venc74ec2f2006-01-13 21:54:23 -08001441 mutex_unlock(&osb->recovery_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -08001442
Tim Gardnerd787ab02013-02-21 16:42:44 -08001443 kfree(rm_quota);
Jan Kara22053632008-10-20 23:50:38 +02001444
Mark Fashehccd979b2005-12-15 14:31:24 -08001445 /* no one is callint kthread_stop() for us so the kthread() api
1446 * requires that we call do_exit(). And it isn't exported, but
1447 * complete_and_exit() seems to be a minimal wrapper around it. */
1448 complete_and_exit(NULL, status);
1449 return status;
1450}
1451
1452void ocfs2_recovery_thread(struct ocfs2_super *osb, int node_num)
1453{
Arjan van de Venc74ec2f2006-01-13 21:54:23 -08001454 mutex_lock(&osb->recovery_lock);
Tao Mab4107952011-02-24 14:15:35 +08001455
1456 trace_ocfs2_recovery_thread(node_num, osb->node_num,
1457 osb->disable_recovery, osb->recovery_thread_task,
1458 osb->disable_recovery ?
1459 -1 : ocfs2_recovery_map_set(osb, node_num));
1460
Mark Fashehccd979b2005-12-15 14:31:24 -08001461 if (osb->disable_recovery)
1462 goto out;
1463
Mark Fashehccd979b2005-12-15 14:31:24 -08001464 if (osb->recovery_thread_task)
1465 goto out;
1466
1467 osb->recovery_thread_task = kthread_run(__ocfs2_recovery_thread, osb,
Mark Fasheh78427042006-05-04 12:03:26 -07001468 "ocfs2rec");
Mark Fashehccd979b2005-12-15 14:31:24 -08001469 if (IS_ERR(osb->recovery_thread_task)) {
1470 mlog_errno((int)PTR_ERR(osb->recovery_thread_task));
1471 osb->recovery_thread_task = NULL;
1472 }
1473
1474out:
Arjan van de Venc74ec2f2006-01-13 21:54:23 -08001475 mutex_unlock(&osb->recovery_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -08001476 wake_up(&osb->recovery_event);
Mark Fashehccd979b2005-12-15 14:31:24 -08001477}
1478
Sunil Mushran539d8262008-07-14 17:31:10 -07001479static int ocfs2_read_journal_inode(struct ocfs2_super *osb,
1480 int slot_num,
1481 struct buffer_head **bh,
1482 struct inode **ret_inode)
1483{
1484 int status = -EACCES;
1485 struct inode *inode = NULL;
1486
1487 BUG_ON(slot_num >= osb->max_slots);
1488
1489 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
1490 slot_num);
1491 if (!inode || is_bad_inode(inode)) {
1492 mlog_errno(status);
1493 goto bail;
1494 }
1495 SET_INODE_JOURNAL(inode);
1496
Joel Beckerb657c952008-11-13 14:49:11 -08001497 status = ocfs2_read_inode_block_full(inode, bh, OCFS2_BH_IGNORE_CACHE);
Sunil Mushran539d8262008-07-14 17:31:10 -07001498 if (status < 0) {
1499 mlog_errno(status);
1500 goto bail;
1501 }
1502
1503 status = 0;
1504
1505bail:
1506 if (inode) {
1507 if (status || !ret_inode)
1508 iput(inode);
1509 else
1510 *ret_inode = inode;
1511 }
1512 return status;
1513}
1514
Mark Fashehccd979b2005-12-15 14:31:24 -08001515/* Does the actual journal replay and marks the journal inode as
1516 * clean. Will only replay if the journal inode is marked dirty. */
1517static int ocfs2_replay_journal(struct ocfs2_super *osb,
1518 int node_num,
1519 int slot_num)
1520{
1521 int status;
1522 int got_lock = 0;
1523 unsigned int flags;
1524 struct inode *inode = NULL;
1525 struct ocfs2_dinode *fe;
1526 journal_t *journal = NULL;
1527 struct buffer_head *bh = NULL;
Sunil Mushran539d8262008-07-14 17:31:10 -07001528 u32 slot_reco_gen;
Mark Fashehccd979b2005-12-15 14:31:24 -08001529
Sunil Mushran539d8262008-07-14 17:31:10 -07001530 status = ocfs2_read_journal_inode(osb, slot_num, &bh, &inode);
1531 if (status) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001532 mlog_errno(status);
1533 goto done;
1534 }
Sunil Mushran539d8262008-07-14 17:31:10 -07001535
1536 fe = (struct ocfs2_dinode *)bh->b_data;
1537 slot_reco_gen = ocfs2_get_recovery_generation(fe);
1538 brelse(bh);
1539 bh = NULL;
1540
1541 /*
1542 * As the fs recovery is asynchronous, there is a small chance that
1543 * another node mounted (and recovered) the slot before the recovery
1544 * thread could get the lock. To handle that, we dirty read the journal
1545 * inode for that slot to get the recovery generation. If it is
1546 * different than what we expected, the slot has been recovered.
1547 * If not, it needs recovery.
1548 */
1549 if (osb->slot_recovery_generations[slot_num] != slot_reco_gen) {
Tao Mab4107952011-02-24 14:15:35 +08001550 trace_ocfs2_replay_journal_recovered(slot_num,
Sunil Mushran539d8262008-07-14 17:31:10 -07001551 osb->slot_recovery_generations[slot_num], slot_reco_gen);
1552 osb->slot_recovery_generations[slot_num] = slot_reco_gen;
1553 status = -EBUSY;
Mark Fashehccd979b2005-12-15 14:31:24 -08001554 goto done;
1555 }
Sunil Mushran539d8262008-07-14 17:31:10 -07001556
1557 /* Continue with recovery as the journal has not yet been recovered */
Mark Fashehccd979b2005-12-15 14:31:24 -08001558
Mark Fashehe63aecb62007-10-18 15:30:42 -07001559 status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY);
Mark Fashehccd979b2005-12-15 14:31:24 -08001560 if (status < 0) {
Tao Mab4107952011-02-24 14:15:35 +08001561 trace_ocfs2_replay_journal_lock_err(status);
Mark Fashehccd979b2005-12-15 14:31:24 -08001562 if (status != -ERESTARTSYS)
1563 mlog(ML_ERROR, "Could not lock journal!\n");
1564 goto done;
1565 }
1566 got_lock = 1;
1567
1568 fe = (struct ocfs2_dinode *) bh->b_data;
1569
1570 flags = le32_to_cpu(fe->id1.journal1.ij_flags);
Sunil Mushran539d8262008-07-14 17:31:10 -07001571 slot_reco_gen = ocfs2_get_recovery_generation(fe);
Mark Fashehccd979b2005-12-15 14:31:24 -08001572
1573 if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) {
Tao Mab4107952011-02-24 14:15:35 +08001574 trace_ocfs2_replay_journal_skip(node_num);
Sunil Mushran539d8262008-07-14 17:31:10 -07001575 /* Refresh recovery generation for the slot */
1576 osb->slot_recovery_generations[slot_num] = slot_reco_gen;
Mark Fashehccd979b2005-12-15 14:31:24 -08001577 goto done;
1578 }
1579
Srinivas Eeda9140db02009-03-06 14:21:46 -08001580 /* we need to run complete recovery for offline orphan slots */
1581 ocfs2_replay_map_set_state(osb, REPLAY_NEEDED);
1582
Sunil Mushran619c2002011-07-24 10:34:54 -07001583 printk(KERN_NOTICE "ocfs2: Begin replay journal (node %d, slot %d) on "\
1584 "device (%u,%u)\n", node_num, slot_num, MAJOR(osb->sb->s_dev),
1585 MINOR(osb->sb->s_dev));
Mark Fashehccd979b2005-12-15 14:31:24 -08001586
1587 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
1588
1589 status = ocfs2_force_read_journal(inode);
1590 if (status < 0) {
1591 mlog_errno(status);
1592 goto done;
1593 }
1594
Joel Becker2b4e30f2008-09-03 20:03:41 -07001595 journal = jbd2_journal_init_inode(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -08001596 if (journal == NULL) {
1597 mlog(ML_ERROR, "Linux journal layer error\n");
1598 status = -EIO;
1599 goto done;
1600 }
1601
Joel Becker2b4e30f2008-09-03 20:03:41 -07001602 status = jbd2_journal_load(journal);
Mark Fashehccd979b2005-12-15 14:31:24 -08001603 if (status < 0) {
1604 mlog_errno(status);
1605 if (!igrab(inode))
1606 BUG();
Joel Becker2b4e30f2008-09-03 20:03:41 -07001607 jbd2_journal_destroy(journal);
Mark Fashehccd979b2005-12-15 14:31:24 -08001608 goto done;
1609 }
1610
1611 ocfs2_clear_journal_error(osb->sb, journal, slot_num);
1612
1613 /* wipe the journal */
Joel Becker2b4e30f2008-09-03 20:03:41 -07001614 jbd2_journal_lock_updates(journal);
1615 status = jbd2_journal_flush(journal);
1616 jbd2_journal_unlock_updates(journal);
Mark Fashehccd979b2005-12-15 14:31:24 -08001617 if (status < 0)
1618 mlog_errno(status);
1619
1620 /* This will mark the node clean */
1621 flags = le32_to_cpu(fe->id1.journal1.ij_flags);
1622 flags &= ~OCFS2_JOURNAL_DIRTY_FL;
1623 fe->id1.journal1.ij_flags = cpu_to_le32(flags);
1624
Sunil Mushran539d8262008-07-14 17:31:10 -07001625 /* Increment recovery generation to indicate successful recovery */
1626 ocfs2_bump_recovery_generation(fe);
1627 osb->slot_recovery_generations[slot_num] =
1628 ocfs2_get_recovery_generation(fe);
1629
Joel Becker13723d02008-10-17 19:25:01 -07001630 ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check);
Joel Becker8cb471e2009-02-10 20:00:41 -08001631 status = ocfs2_write_block(osb, bh, INODE_CACHE(inode));
Mark Fashehccd979b2005-12-15 14:31:24 -08001632 if (status < 0)
1633 mlog_errno(status);
1634
1635 if (!igrab(inode))
1636 BUG();
1637
Joel Becker2b4e30f2008-09-03 20:03:41 -07001638 jbd2_journal_destroy(journal);
Mark Fashehccd979b2005-12-15 14:31:24 -08001639
Sunil Mushran619c2002011-07-24 10:34:54 -07001640 printk(KERN_NOTICE "ocfs2: End replay journal (node %d, slot %d) on "\
1641 "device (%u,%u)\n", node_num, slot_num, MAJOR(osb->sb->s_dev),
1642 MINOR(osb->sb->s_dev));
Mark Fashehccd979b2005-12-15 14:31:24 -08001643done:
1644 /* drop the lock on this nodes journal */
1645 if (got_lock)
Mark Fashehe63aecb62007-10-18 15:30:42 -07001646 ocfs2_inode_unlock(inode, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -08001647
1648 if (inode)
1649 iput(inode);
1650
Mark Fasheha81cb882008-10-07 14:25:16 -07001651 brelse(bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001652
Mark Fashehccd979b2005-12-15 14:31:24 -08001653 return status;
1654}
1655
1656/*
1657 * Do the most important parts of node recovery:
1658 * - Replay it's journal
1659 * - Stamp a clean local allocator file
1660 * - Stamp a clean truncate log
1661 * - Mark the node clean
1662 *
1663 * If this function completes without error, a node in OCFS2 can be
1664 * said to have been safely recovered. As a result, failure during the
1665 * second part of a nodes recovery process (local alloc recovery) is
1666 * far less concerning.
1667 */
1668static int ocfs2_recover_node(struct ocfs2_super *osb,
Jan Kara22053632008-10-20 23:50:38 +02001669 int node_num, int slot_num)
Mark Fashehccd979b2005-12-15 14:31:24 -08001670{
1671 int status = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -08001672 struct ocfs2_dinode *la_copy = NULL;
1673 struct ocfs2_dinode *tl_copy = NULL;
1674
Tao Mab4107952011-02-24 14:15:35 +08001675 trace_ocfs2_recover_node(node_num, slot_num, osb->node_num);
Mark Fashehccd979b2005-12-15 14:31:24 -08001676
1677 /* Should not ever be called to recover ourselves -- in that
1678 * case we should've called ocfs2_journal_load instead. */
Eric Sesterhenn / snakebyteebdec832006-01-27 10:32:52 +01001679 BUG_ON(osb->node_num == node_num);
Mark Fashehccd979b2005-12-15 14:31:24 -08001680
Mark Fashehccd979b2005-12-15 14:31:24 -08001681 status = ocfs2_replay_journal(osb, node_num, slot_num);
1682 if (status < 0) {
Sunil Mushran539d8262008-07-14 17:31:10 -07001683 if (status == -EBUSY) {
Tao Mab4107952011-02-24 14:15:35 +08001684 trace_ocfs2_recover_node_skip(slot_num, node_num);
Sunil Mushran539d8262008-07-14 17:31:10 -07001685 status = 0;
1686 goto done;
1687 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001688 mlog_errno(status);
1689 goto done;
1690 }
1691
1692 /* Stamp a clean local alloc file AFTER recovering the journal... */
1693 status = ocfs2_begin_local_alloc_recovery(osb, slot_num, &la_copy);
1694 if (status < 0) {
1695 mlog_errno(status);
1696 goto done;
1697 }
1698
1699 /* An error from begin_truncate_log_recovery is not
1700 * serious enough to warrant halting the rest of
1701 * recovery. */
1702 status = ocfs2_begin_truncate_log_recovery(osb, slot_num, &tl_copy);
1703 if (status < 0)
1704 mlog_errno(status);
1705
1706 /* Likewise, this would be a strange but ultimately not so
1707 * harmful place to get an error... */
Mark Fasheh8e8a4602008-02-01 11:59:09 -08001708 status = ocfs2_clear_slot(osb, slot_num);
Mark Fashehccd979b2005-12-15 14:31:24 -08001709 if (status < 0)
1710 mlog_errno(status);
1711
1712 /* This will kfree the memory pointed to by la_copy and tl_copy */
1713 ocfs2_queue_recovery_completion(osb->journal, slot_num, la_copy,
Jan Kara22053632008-10-20 23:50:38 +02001714 tl_copy, NULL);
Mark Fashehccd979b2005-12-15 14:31:24 -08001715
1716 status = 0;
1717done:
1718
Mark Fashehccd979b2005-12-15 14:31:24 -08001719 return status;
1720}
1721
1722/* Test node liveness by trylocking his journal. If we get the lock,
1723 * we drop it here. Return 0 if we got the lock, -EAGAIN if node is
1724 * still alive (we couldn't get the lock) and < 0 on error. */
1725static int ocfs2_trylock_journal(struct ocfs2_super *osb,
1726 int slot_num)
1727{
1728 int status, flags;
1729 struct inode *inode = NULL;
1730
1731 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
1732 slot_num);
1733 if (inode == NULL) {
1734 mlog(ML_ERROR, "access error\n");
1735 status = -EACCES;
1736 goto bail;
1737 }
1738 if (is_bad_inode(inode)) {
1739 mlog(ML_ERROR, "access error (bad inode)\n");
1740 iput(inode);
1741 inode = NULL;
1742 status = -EACCES;
1743 goto bail;
1744 }
1745 SET_INODE_JOURNAL(inode);
1746
1747 flags = OCFS2_META_LOCK_RECOVERY | OCFS2_META_LOCK_NOQUEUE;
Mark Fashehe63aecb62007-10-18 15:30:42 -07001748 status = ocfs2_inode_lock_full(inode, NULL, 1, flags);
Mark Fashehccd979b2005-12-15 14:31:24 -08001749 if (status < 0) {
1750 if (status != -EAGAIN)
1751 mlog_errno(status);
1752 goto bail;
1753 }
1754
Mark Fashehe63aecb62007-10-18 15:30:42 -07001755 ocfs2_inode_unlock(inode, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -08001756bail:
1757 if (inode)
1758 iput(inode);
1759
1760 return status;
1761}
1762
1763/* Call this underneath ocfs2_super_lock. It also assumes that the
1764 * slot info struct has been updated from disk. */
1765int ocfs2_mark_dead_nodes(struct ocfs2_super *osb)
1766{
Joel Beckerd85b20e2008-02-01 12:01:05 -08001767 unsigned int node_num;
1768 int status, i;
Mark Fasheha1af7d12008-08-19 17:20:28 -07001769 u32 gen;
Sunil Mushran539d8262008-07-14 17:31:10 -07001770 struct buffer_head *bh = NULL;
1771 struct ocfs2_dinode *di;
Mark Fashehccd979b2005-12-15 14:31:24 -08001772
1773 /* This is called with the super block cluster lock, so we
1774 * know that the slot map can't change underneath us. */
1775
Joel Beckerd85b20e2008-02-01 12:01:05 -08001776 for (i = 0; i < osb->max_slots; i++) {
Sunil Mushran539d8262008-07-14 17:31:10 -07001777 /* Read journal inode to get the recovery generation */
1778 status = ocfs2_read_journal_inode(osb, i, &bh, NULL);
1779 if (status) {
1780 mlog_errno(status);
1781 goto bail;
1782 }
1783 di = (struct ocfs2_dinode *)bh->b_data;
Mark Fasheha1af7d12008-08-19 17:20:28 -07001784 gen = ocfs2_get_recovery_generation(di);
Sunil Mushran539d8262008-07-14 17:31:10 -07001785 brelse(bh);
1786 bh = NULL;
1787
Mark Fasheha1af7d12008-08-19 17:20:28 -07001788 spin_lock(&osb->osb_lock);
1789 osb->slot_recovery_generations[i] = gen;
1790
Tao Mab4107952011-02-24 14:15:35 +08001791 trace_ocfs2_mark_dead_nodes(i,
1792 osb->slot_recovery_generations[i]);
Sunil Mushran539d8262008-07-14 17:31:10 -07001793
Mark Fasheha1af7d12008-08-19 17:20:28 -07001794 if (i == osb->slot_num) {
1795 spin_unlock(&osb->osb_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -08001796 continue;
Mark Fasheha1af7d12008-08-19 17:20:28 -07001797 }
Joel Beckerd85b20e2008-02-01 12:01:05 -08001798
1799 status = ocfs2_slot_to_node_num_locked(osb, i, &node_num);
Mark Fasheha1af7d12008-08-19 17:20:28 -07001800 if (status == -ENOENT) {
1801 spin_unlock(&osb->osb_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -08001802 continue;
Mark Fasheha1af7d12008-08-19 17:20:28 -07001803 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001804
Mark Fasheha1af7d12008-08-19 17:20:28 -07001805 if (__ocfs2_recovery_map_test(osb, node_num)) {
1806 spin_unlock(&osb->osb_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -08001807 continue;
Mark Fasheha1af7d12008-08-19 17:20:28 -07001808 }
Joel Beckerd85b20e2008-02-01 12:01:05 -08001809 spin_unlock(&osb->osb_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -08001810
1811 /* Ok, we have a slot occupied by another node which
1812 * is not in the recovery map. We trylock his journal
1813 * file here to test if he's alive. */
1814 status = ocfs2_trylock_journal(osb, i);
1815 if (!status) {
1816 /* Since we're called from mount, we know that
1817 * the recovery thread can't race us on
1818 * setting / checking the recovery bits. */
1819 ocfs2_recovery_thread(osb, node_num);
1820 } else if ((status < 0) && (status != -EAGAIN)) {
1821 mlog_errno(status);
1822 goto bail;
1823 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001824 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001825
1826 status = 0;
1827bail:
Mark Fashehccd979b2005-12-15 14:31:24 -08001828 return status;
1829}
1830
Srinivas Eeda83273932009-06-03 17:02:55 -07001831/*
1832 * Scan timer should get fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT. Add some
1833 * randomness to the timeout to minimize multple nodes firing the timer at the
1834 * same time.
1835 */
1836static inline unsigned long ocfs2_orphan_scan_timeout(void)
1837{
1838 unsigned long time;
1839
1840 get_random_bytes(&time, sizeof(time));
1841 time = ORPHAN_SCAN_SCHEDULE_TIMEOUT + (time % 5000);
1842 return msecs_to_jiffies(time);
1843}
1844
1845/*
1846 * ocfs2_queue_orphan_scan calls ocfs2_queue_recovery_completion for
1847 * every slot, queuing a recovery of the slot on the ocfs2_wq thread. This
1848 * is done to catch any orphans that are left over in orphan directories.
1849 *
Sunil Mushrana035bff2011-07-24 10:35:54 -07001850 * It scans all slots, even ones that are in use. It does so to handle the
1851 * case described below:
1852 *
1853 * Node 1 has an inode it was using. The dentry went away due to memory
1854 * pressure. Node 1 closes the inode, but it's on the free list. The node
1855 * has the open lock.
1856 * Node 2 unlinks the inode. It grabs the dentry lock to notify others,
1857 * but node 1 has no dentry and doesn't get the message. It trylocks the
1858 * open lock, sees that another node has a PR, and does nothing.
1859 * Later node 2 runs its orphan dir. It igets the inode, trylocks the
1860 * open lock, sees the PR still, and does nothing.
1861 * Basically, we have to trigger an orphan iput on node 1. The only way
1862 * for this to happen is if node 1 runs node 2's orphan dir.
1863 *
Srinivas Eeda83273932009-06-03 17:02:55 -07001864 * ocfs2_queue_orphan_scan gets called every ORPHAN_SCAN_SCHEDULE_TIMEOUT
1865 * seconds. It gets an EX lock on os_lockres and checks sequence number
1866 * stored in LVB. If the sequence number has changed, it means some other
1867 * node has done the scan. This node skips the scan and tracks the
1868 * sequence number. If the sequence number didn't change, it means a scan
1869 * hasn't happened. The node queues a scan and increments the
1870 * sequence number in the LVB.
1871 */
1872void ocfs2_queue_orphan_scan(struct ocfs2_super *osb)
1873{
1874 struct ocfs2_orphan_scan *os;
1875 int status, i;
1876 u32 seqno = 0;
1877
1878 os = &osb->osb_orphan_scan;
1879
Sunil Mushran692684e2009-06-19 16:53:17 -07001880 if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE)
1881 goto out;
1882
Tao Mab4107952011-02-24 14:15:35 +08001883 trace_ocfs2_queue_orphan_scan_begin(os->os_count, os->os_seqno,
1884 atomic_read(&os->os_state));
1885
Sunil Mushrandf152c22009-06-22 11:40:07 -07001886 status = ocfs2_orphan_scan_lock(osb, &seqno);
Srinivas Eeda83273932009-06-03 17:02:55 -07001887 if (status < 0) {
1888 if (status != -EAGAIN)
1889 mlog_errno(status);
1890 goto out;
1891 }
1892
Sunil Mushran692684e2009-06-19 16:53:17 -07001893 /* Do no queue the tasks if the volume is being umounted */
1894 if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE)
1895 goto unlock;
1896
Srinivas Eeda83273932009-06-03 17:02:55 -07001897 if (os->os_seqno != seqno) {
1898 os->os_seqno = seqno;
1899 goto unlock;
1900 }
1901
1902 for (i = 0; i < osb->max_slots; i++)
1903 ocfs2_queue_recovery_completion(osb->journal, i, NULL, NULL,
1904 NULL);
1905 /*
1906 * We queued a recovery on orphan slots, increment the sequence
1907 * number and update LVB so other node will skip the scan for a while
1908 */
1909 seqno++;
Srinivas Eeda15633a22009-06-03 17:02:56 -07001910 os->os_count++;
1911 os->os_scantime = CURRENT_TIME;
Srinivas Eeda83273932009-06-03 17:02:55 -07001912unlock:
Sunil Mushrandf152c22009-06-22 11:40:07 -07001913 ocfs2_orphan_scan_unlock(osb, seqno);
Srinivas Eeda83273932009-06-03 17:02:55 -07001914out:
Tao Mab4107952011-02-24 14:15:35 +08001915 trace_ocfs2_queue_orphan_scan_end(os->os_count, os->os_seqno,
1916 atomic_read(&os->os_state));
Srinivas Eeda83273932009-06-03 17:02:55 -07001917 return;
1918}
1919
1920/* Worker task that gets fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT millsec */
1921void ocfs2_orphan_scan_work(struct work_struct *work)
1922{
1923 struct ocfs2_orphan_scan *os;
1924 struct ocfs2_super *osb;
1925
1926 os = container_of(work, struct ocfs2_orphan_scan,
1927 os_orphan_scan_work.work);
1928 osb = os->os_osb;
1929
1930 mutex_lock(&os->os_lock);
1931 ocfs2_queue_orphan_scan(osb);
Sunil Mushran692684e2009-06-19 16:53:17 -07001932 if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE)
Tao Ma40f165f2010-05-28 14:22:59 +08001933 queue_delayed_work(ocfs2_wq, &os->os_orphan_scan_work,
Sunil Mushran692684e2009-06-19 16:53:17 -07001934 ocfs2_orphan_scan_timeout());
Srinivas Eeda83273932009-06-03 17:02:55 -07001935 mutex_unlock(&os->os_lock);
1936}
1937
1938void ocfs2_orphan_scan_stop(struct ocfs2_super *osb)
1939{
1940 struct ocfs2_orphan_scan *os;
1941
1942 os = &osb->osb_orphan_scan;
Sunil Mushrandf152c22009-06-22 11:40:07 -07001943 if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE) {
1944 atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE);
1945 mutex_lock(&os->os_lock);
1946 cancel_delayed_work(&os->os_orphan_scan_work);
1947 mutex_unlock(&os->os_lock);
1948 }
Srinivas Eeda83273932009-06-03 17:02:55 -07001949}
1950
Sunil Mushrandf152c22009-06-22 11:40:07 -07001951void ocfs2_orphan_scan_init(struct ocfs2_super *osb)
Srinivas Eeda83273932009-06-03 17:02:55 -07001952{
1953 struct ocfs2_orphan_scan *os;
1954
1955 os = &osb->osb_orphan_scan;
1956 os->os_osb = osb;
Srinivas Eeda15633a22009-06-03 17:02:56 -07001957 os->os_count = 0;
Sunil Mushran32119492009-06-19 16:53:18 -07001958 os->os_seqno = 0;
Srinivas Eeda83273932009-06-03 17:02:55 -07001959 mutex_init(&os->os_lock);
Sunil Mushrandf152c22009-06-22 11:40:07 -07001960 INIT_DELAYED_WORK(&os->os_orphan_scan_work, ocfs2_orphan_scan_work);
Jeff Mahoney8b712cd2009-07-07 17:22:12 -04001961}
Srinivas Eeda83273932009-06-03 17:02:55 -07001962
Jeff Mahoney8b712cd2009-07-07 17:22:12 -04001963void ocfs2_orphan_scan_start(struct ocfs2_super *osb)
1964{
1965 struct ocfs2_orphan_scan *os;
1966
1967 os = &osb->osb_orphan_scan;
1968 os->os_scantime = CURRENT_TIME;
Sunil Mushrandf152c22009-06-22 11:40:07 -07001969 if (ocfs2_is_hard_readonly(osb) || ocfs2_mount_local(osb))
1970 atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE);
1971 else {
1972 atomic_set(&os->os_state, ORPHAN_SCAN_ACTIVE);
Tao Ma40f165f2010-05-28 14:22:59 +08001973 queue_delayed_work(ocfs2_wq, &os->os_orphan_scan_work,
1974 ocfs2_orphan_scan_timeout());
Sunil Mushrandf152c22009-06-22 11:40:07 -07001975 }
Srinivas Eeda83273932009-06-03 17:02:55 -07001976}
1977
Mark Fasheh5eae5b92007-09-10 17:50:51 -07001978struct ocfs2_orphan_filldir_priv {
Al Viro3704412b2013-05-22 21:06:00 -04001979 struct dir_context ctx;
Mark Fasheh5eae5b92007-09-10 17:50:51 -07001980 struct inode *head;
1981 struct ocfs2_super *osb;
1982};
1983
1984static int ocfs2_orphan_filldir(void *priv, const char *name, int name_len,
1985 loff_t pos, u64 ino, unsigned type)
1986{
1987 struct ocfs2_orphan_filldir_priv *p = priv;
1988 struct inode *iter;
1989
1990 if (name_len == 1 && !strncmp(".", name, 1))
1991 return 0;
1992 if (name_len == 2 && !strncmp("..", name, 2))
1993 return 0;
1994
1995 /* Skip bad inodes so that recovery can continue */
1996 iter = ocfs2_iget(p->osb, ino,
Jan Kara5fa06132008-01-11 00:11:45 +01001997 OCFS2_FI_FLAG_ORPHAN_RECOVERY, 0);
Mark Fasheh5eae5b92007-09-10 17:50:51 -07001998 if (IS_ERR(iter))
1999 return 0;
2000
Tao Mab4107952011-02-24 14:15:35 +08002001 trace_ocfs2_orphan_filldir((unsigned long long)OCFS2_I(iter)->ip_blkno);
Mark Fasheh5eae5b92007-09-10 17:50:51 -07002002 /* No locking is required for the next_orphan queue as there
2003 * is only ever a single process doing orphan recovery. */
2004 OCFS2_I(iter)->ip_next_orphan = p->head;
2005 p->head = iter;
2006
2007 return 0;
2008}
2009
Mark Fashehb4df6ed2006-02-22 17:35:08 -08002010static int ocfs2_queue_orphans(struct ocfs2_super *osb,
2011 int slot,
2012 struct inode **head)
Mark Fashehccd979b2005-12-15 14:31:24 -08002013{
Mark Fashehb4df6ed2006-02-22 17:35:08 -08002014 int status;
Mark Fashehccd979b2005-12-15 14:31:24 -08002015 struct inode *orphan_dir_inode = NULL;
Al Viro3704412b2013-05-22 21:06:00 -04002016 struct ocfs2_orphan_filldir_priv priv = {
2017 .ctx.actor = ocfs2_orphan_filldir,
2018 .osb = osb,
2019 .head = *head
2020 };
Mark Fashehccd979b2005-12-15 14:31:24 -08002021
2022 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2023 ORPHAN_DIR_SYSTEM_INODE,
2024 slot);
2025 if (!orphan_dir_inode) {
2026 status = -ENOENT;
2027 mlog_errno(status);
Mark Fashehb4df6ed2006-02-22 17:35:08 -08002028 return status;
Sunil Mushran2bd63212010-01-25 16:57:38 -08002029 }
Mark Fashehccd979b2005-12-15 14:31:24 -08002030
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002031 mutex_lock(&orphan_dir_inode->i_mutex);
Mark Fashehe63aecb62007-10-18 15:30:42 -07002032 status = ocfs2_inode_lock(orphan_dir_inode, NULL, 0);
Mark Fashehccd979b2005-12-15 14:31:24 -08002033 if (status < 0) {
Mark Fashehccd979b2005-12-15 14:31:24 -08002034 mlog_errno(status);
2035 goto out;
2036 }
Mark Fashehccd979b2005-12-15 14:31:24 -08002037
Al Viro3704412b2013-05-22 21:06:00 -04002038 status = ocfs2_dir_foreach(orphan_dir_inode, &priv.ctx);
Mark Fasheh5eae5b92007-09-10 17:50:51 -07002039 if (status) {
2040 mlog_errno(status);
Mark Fasheha86370f2007-12-03 14:06:23 -08002041 goto out_cluster;
Mark Fashehccd979b2005-12-15 14:31:24 -08002042 }
Mark Fashehccd979b2005-12-15 14:31:24 -08002043
Mark Fasheh5eae5b92007-09-10 17:50:51 -07002044 *head = priv.head;
2045
Mark Fasheha86370f2007-12-03 14:06:23 -08002046out_cluster:
Mark Fashehe63aecb62007-10-18 15:30:42 -07002047 ocfs2_inode_unlock(orphan_dir_inode, 0);
Mark Fashehb4df6ed2006-02-22 17:35:08 -08002048out:
2049 mutex_unlock(&orphan_dir_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -08002050 iput(orphan_dir_inode);
Mark Fashehb4df6ed2006-02-22 17:35:08 -08002051 return status;
2052}
2053
2054static int ocfs2_orphan_recovery_can_continue(struct ocfs2_super *osb,
2055 int slot)
2056{
2057 int ret;
2058
2059 spin_lock(&osb->osb_lock);
2060 ret = !osb->osb_orphan_wipes[slot];
2061 spin_unlock(&osb->osb_lock);
2062 return ret;
2063}
2064
2065static void ocfs2_mark_recovering_orphan_dir(struct ocfs2_super *osb,
2066 int slot)
2067{
2068 spin_lock(&osb->osb_lock);
2069 /* Mark ourselves such that new processes in delete_inode()
2070 * know to quit early. */
2071 ocfs2_node_map_set_bit(osb, &osb->osb_recovering_orphan_dirs, slot);
2072 while (osb->osb_orphan_wipes[slot]) {
2073 /* If any processes are already in the middle of an
2074 * orphan wipe on this dir, then we need to wait for
2075 * them. */
2076 spin_unlock(&osb->osb_lock);
2077 wait_event_interruptible(osb->osb_wipe_event,
2078 ocfs2_orphan_recovery_can_continue(osb, slot));
2079 spin_lock(&osb->osb_lock);
2080 }
2081 spin_unlock(&osb->osb_lock);
2082}
2083
2084static void ocfs2_clear_recovering_orphan_dir(struct ocfs2_super *osb,
2085 int slot)
2086{
2087 ocfs2_node_map_clear_bit(osb, &osb->osb_recovering_orphan_dirs, slot);
2088}
2089
2090/*
2091 * Orphan recovery. Each mounted node has it's own orphan dir which we
2092 * must run during recovery. Our strategy here is to build a list of
2093 * the inodes in the orphan dir and iget/iput them. The VFS does
2094 * (most) of the rest of the work.
2095 *
2096 * Orphan recovery can happen at any time, not just mount so we have a
2097 * couple of extra considerations.
2098 *
2099 * - We grab as many inodes as we can under the orphan dir lock -
2100 * doing iget() outside the orphan dir risks getting a reference on
2101 * an invalid inode.
2102 * - We must be sure not to deadlock with other processes on the
2103 * system wanting to run delete_inode(). This can happen when they go
2104 * to lock the orphan dir and the orphan recovery process attempts to
2105 * iget() inside the orphan dir lock. This can be avoided by
2106 * advertising our state to ocfs2_delete_inode().
2107 */
2108static int ocfs2_recover_orphans(struct ocfs2_super *osb,
2109 int slot)
2110{
2111 int ret = 0;
2112 struct inode *inode = NULL;
2113 struct inode *iter;
2114 struct ocfs2_inode_info *oi;
2115
Tao Mab4107952011-02-24 14:15:35 +08002116 trace_ocfs2_recover_orphans(slot);
Mark Fashehb4df6ed2006-02-22 17:35:08 -08002117
2118 ocfs2_mark_recovering_orphan_dir(osb, slot);
2119 ret = ocfs2_queue_orphans(osb, slot, &inode);
2120 ocfs2_clear_recovering_orphan_dir(osb, slot);
2121
2122 /* Error here should be noted, but we want to continue with as
2123 * many queued inodes as we've got. */
2124 if (ret)
2125 mlog_errno(ret);
Mark Fashehccd979b2005-12-15 14:31:24 -08002126
2127 while (inode) {
2128 oi = OCFS2_I(inode);
Tao Mab4107952011-02-24 14:15:35 +08002129 trace_ocfs2_recover_orphans_iput(
2130 (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08002131
2132 iter = oi->ip_next_orphan;
2133
2134 spin_lock(&oi->ip_lock);
Mark Fasheh34d024f2007-09-24 15:56:19 -07002135 /* The remote delete code may have set these on the
2136 * assumption that the other node would wipe them
2137 * successfully. If they are still in the node's
2138 * orphan dir, we need to reset that state. */
Mark Fashehccd979b2005-12-15 14:31:24 -08002139 oi->ip_flags &= ~(OCFS2_INODE_DELETED|OCFS2_INODE_SKIP_DELETE);
2140
2141 /* Set the proper information to get us going into
2142 * ocfs2_delete_inode. */
2143 oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED;
Mark Fashehccd979b2005-12-15 14:31:24 -08002144 spin_unlock(&oi->ip_lock);
2145
2146 iput(inode);
2147
2148 inode = iter;
2149 }
2150
Mark Fashehb4df6ed2006-02-22 17:35:08 -08002151 return ret;
Mark Fashehccd979b2005-12-15 14:31:24 -08002152}
2153
Jan Kara19ece542008-08-21 20:13:17 +02002154static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota)
Mark Fashehccd979b2005-12-15 14:31:24 -08002155{
2156 /* This check is good because ocfs2 will wait on our recovery
2157 * thread before changing it to something other than MOUNTED
2158 * or DISABLED. */
2159 wait_event(osb->osb_mount_event,
Jan Kara19ece542008-08-21 20:13:17 +02002160 (!quota && atomic_read(&osb->vol_state) == VOLUME_MOUNTED) ||
2161 atomic_read(&osb->vol_state) == VOLUME_MOUNTED_QUOTAS ||
Mark Fashehccd979b2005-12-15 14:31:24 -08002162 atomic_read(&osb->vol_state) == VOLUME_DISABLED);
2163
2164 /* If there's an error on mount, then we may never get to the
2165 * MOUNTED flag, but this is set right before
2166 * dismount_volume() so we can trust it. */
2167 if (atomic_read(&osb->vol_state) == VOLUME_DISABLED) {
Tao Mab4107952011-02-24 14:15:35 +08002168 trace_ocfs2_wait_on_mount(VOLUME_DISABLED);
Mark Fashehccd979b2005-12-15 14:31:24 -08002169 mlog(0, "mount error, exiting!\n");
2170 return -EBUSY;
2171 }
2172
2173 return 0;
2174}
2175
2176static int ocfs2_commit_thread(void *arg)
2177{
2178 int status;
2179 struct ocfs2_super *osb = arg;
2180 struct ocfs2_journal *journal = osb->journal;
2181
2182 /* we can trust j_num_trans here because _should_stop() is only set in
2183 * shutdown and nobody other than ourselves should be able to start
2184 * transactions. committing on shutdown might take a few iterations
2185 * as final transactions put deleted inodes on the list */
2186 while (!(kthread_should_stop() &&
2187 atomic_read(&journal->j_num_trans) == 0)) {
2188
Mark Fasheh745ae8ba2006-02-09 13:23:39 -08002189 wait_event_interruptible(osb->checkpoint_event,
2190 atomic_read(&journal->j_num_trans)
2191 || kthread_should_stop());
Mark Fashehccd979b2005-12-15 14:31:24 -08002192
2193 status = ocfs2_commit_cache(osb);
2194 if (status < 0)
2195 mlog_errno(status);
2196
2197 if (kthread_should_stop() && atomic_read(&journal->j_num_trans)){
2198 mlog(ML_KTHREAD,
2199 "commit_thread: %u transactions pending on "
2200 "shutdown\n",
2201 atomic_read(&journal->j_num_trans));
2202 }
2203 }
2204
2205 return 0;
2206}
2207
Sunil Mushran539d8262008-07-14 17:31:10 -07002208/* Reads all the journal inodes without taking any cluster locks. Used
2209 * for hard readonly access to determine whether any journal requires
2210 * recovery. Also used to refresh the recovery generation numbers after
2211 * a journal has been recovered by another node.
2212 */
Mark Fashehccd979b2005-12-15 14:31:24 -08002213int ocfs2_check_journals_nolocks(struct ocfs2_super *osb)
2214{
2215 int ret = 0;
2216 unsigned int slot;
Sunil Mushran539d8262008-07-14 17:31:10 -07002217 struct buffer_head *di_bh = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -08002218 struct ocfs2_dinode *di;
Sunil Mushran539d8262008-07-14 17:31:10 -07002219 int journal_dirty = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -08002220
2221 for(slot = 0; slot < osb->max_slots; slot++) {
Sunil Mushran539d8262008-07-14 17:31:10 -07002222 ret = ocfs2_read_journal_inode(osb, slot, &di_bh, NULL);
2223 if (ret) {
Mark Fashehccd979b2005-12-15 14:31:24 -08002224 mlog_errno(ret);
2225 goto out;
2226 }
2227
2228 di = (struct ocfs2_dinode *) di_bh->b_data;
2229
Sunil Mushran539d8262008-07-14 17:31:10 -07002230 osb->slot_recovery_generations[slot] =
2231 ocfs2_get_recovery_generation(di);
2232
Mark Fashehccd979b2005-12-15 14:31:24 -08002233 if (le32_to_cpu(di->id1.journal1.ij_flags) &
2234 OCFS2_JOURNAL_DIRTY_FL)
Sunil Mushran539d8262008-07-14 17:31:10 -07002235 journal_dirty = 1;
Mark Fashehccd979b2005-12-15 14:31:24 -08002236
2237 brelse(di_bh);
Sunil Mushran539d8262008-07-14 17:31:10 -07002238 di_bh = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -08002239 }
2240
2241out:
Sunil Mushran539d8262008-07-14 17:31:10 -07002242 if (journal_dirty)
2243 ret = -EROFS;
Mark Fashehccd979b2005-12-15 14:31:24 -08002244 return ret;
2245}