blob: f31c7e8c19c32bf091117a8fbd114b69c200f7ae [file] [log] [blame]
Mark Fashehccd979b2005-12-15 14:31:24 -08001/* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
3 *
4 * journal.c
5 *
6 * Defines functions of journalling api
7 *
8 * Copyright (C) 2003, 2004 Oracle. All rights reserved.
9 *
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public
12 * License as published by the Free Software Foundation; either
13 * version 2 of the License, or (at your option) any later version.
14 *
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * General Public License for more details.
19 *
20 * You should have received a copy of the GNU General Public
21 * License along with this program; if not, write to the
22 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
23 * Boston, MA 021110-1307, USA.
24 */
25
26#include <linux/fs.h>
27#include <linux/types.h>
28#include <linux/slab.h>
29#include <linux/highmem.h>
30#include <linux/kthread.h>
31
32#define MLOG_MASK_PREFIX ML_JOURNAL
33#include <cluster/masklog.h>
34
35#include "ocfs2.h"
36
37#include "alloc.h"
Mark Fasheh316f4b92007-09-07 18:21:26 -070038#include "dir.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080039#include "dlmglue.h"
40#include "extent_map.h"
41#include "heartbeat.h"
42#include "inode.h"
43#include "journal.h"
44#include "localalloc.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080045#include "slot_map.h"
46#include "super.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080047#include "sysfile.h"
48
49#include "buffer_head_io.h"
50
Ingo Molnar34af9462006-06-27 02:53:55 -070051DEFINE_SPINLOCK(trans_inc_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -080052
53static int ocfs2_force_read_journal(struct inode *inode);
54static int ocfs2_recover_node(struct ocfs2_super *osb,
55 int node_num);
56static int __ocfs2_recovery_thread(void *arg);
57static int ocfs2_commit_cache(struct ocfs2_super *osb);
58static int ocfs2_wait_on_mount(struct ocfs2_super *osb);
Mark Fashehccd979b2005-12-15 14:31:24 -080059static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
60 int dirty);
61static int ocfs2_trylock_journal(struct ocfs2_super *osb,
62 int slot_num);
63static int ocfs2_recover_orphans(struct ocfs2_super *osb,
64 int slot);
65static int ocfs2_commit_thread(void *arg);
66
67static int ocfs2_commit_cache(struct ocfs2_super *osb)
68{
69 int status = 0;
70 unsigned int flushed;
71 unsigned long old_id;
72 struct ocfs2_journal *journal = NULL;
73
74 mlog_entry_void();
75
76 journal = osb->journal;
77
78 /* Flush all pending commits and checkpoint the journal. */
79 down_write(&journal->j_trans_barrier);
80
81 if (atomic_read(&journal->j_num_trans) == 0) {
82 up_write(&journal->j_trans_barrier);
83 mlog(0, "No transactions for me to flush!\n");
84 goto finally;
85 }
86
87 journal_lock_updates(journal->j_journal);
88 status = journal_flush(journal->j_journal);
89 journal_unlock_updates(journal->j_journal);
90 if (status < 0) {
91 up_write(&journal->j_trans_barrier);
92 mlog_errno(status);
93 goto finally;
94 }
95
96 old_id = ocfs2_inc_trans_id(journal);
97
98 flushed = atomic_read(&journal->j_num_trans);
99 atomic_set(&journal->j_num_trans, 0);
100 up_write(&journal->j_trans_barrier);
101
102 mlog(0, "commit_thread: flushed transaction %lu (%u handles)\n",
103 journal->j_trans_id, flushed);
104
Mark Fasheh34d024f2007-09-24 15:56:19 -0700105 ocfs2_wake_downconvert_thread(osb);
Mark Fashehccd979b2005-12-15 14:31:24 -0800106 wake_up(&journal->j_checkpointed);
107finally:
108 mlog_exit(status);
109 return status;
110}
111
Mark Fashehccd979b2005-12-15 14:31:24 -0800112/* pass it NULL and it will allocate a new handle object for you. If
113 * you pass it a handle however, it may still return error, in which
114 * case it has free'd the passed handle for you. */
Mark Fasheh1fabe142006-10-09 18:11:45 -0700115handle_t *ocfs2_start_trans(struct ocfs2_super *osb, int max_buffs)
Mark Fashehccd979b2005-12-15 14:31:24 -0800116{
Mark Fashehccd979b2005-12-15 14:31:24 -0800117 journal_t *journal = osb->journal->j_journal;
Mark Fasheh1fabe142006-10-09 18:11:45 -0700118 handle_t *handle;
Mark Fashehccd979b2005-12-15 14:31:24 -0800119
Eric Sesterhenn / snakebyteebdec832006-01-27 10:32:52 +0100120 BUG_ON(!osb || !osb->journal->j_journal);
Mark Fashehccd979b2005-12-15 14:31:24 -0800121
Mark Fasheh65eff9c2006-10-09 17:26:22 -0700122 if (ocfs2_is_hard_readonly(osb))
123 return ERR_PTR(-EROFS);
Mark Fashehccd979b2005-12-15 14:31:24 -0800124
125 BUG_ON(osb->journal->j_state == OCFS2_JOURNAL_FREE);
126 BUG_ON(max_buffs <= 0);
127
128 /* JBD might support this, but our journalling code doesn't yet. */
129 if (journal_current_handle()) {
130 mlog(ML_ERROR, "Recursive transaction attempted!\n");
131 BUG();
132 }
133
Mark Fashehccd979b2005-12-15 14:31:24 -0800134 down_read(&osb->journal->j_trans_barrier);
135
Mark Fasheh1fabe142006-10-09 18:11:45 -0700136 handle = journal_start(journal, max_buffs);
137 if (IS_ERR(handle)) {
Mark Fashehccd979b2005-12-15 14:31:24 -0800138 up_read(&osb->journal->j_trans_barrier);
139
Mark Fasheh1fabe142006-10-09 18:11:45 -0700140 mlog_errno(PTR_ERR(handle));
Mark Fashehccd979b2005-12-15 14:31:24 -0800141
142 if (is_journal_aborted(journal)) {
143 ocfs2_abort(osb->sb, "Detected aborted journal");
Mark Fasheh1fabe142006-10-09 18:11:45 -0700144 handle = ERR_PTR(-EROFS);
Mark Fashehccd979b2005-12-15 14:31:24 -0800145 }
Sunil Mushranc271c5c2006-12-05 17:56:35 -0800146 } else {
147 if (!ocfs2_mount_local(osb))
148 atomic_inc(&(osb->journal->j_num_trans));
149 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800150
Mark Fashehccd979b2005-12-15 14:31:24 -0800151 return handle;
Mark Fashehccd979b2005-12-15 14:31:24 -0800152}
153
Mark Fasheh1fabe142006-10-09 18:11:45 -0700154int ocfs2_commit_trans(struct ocfs2_super *osb,
155 handle_t *handle)
Mark Fashehccd979b2005-12-15 14:31:24 -0800156{
Mark Fasheh1fabe142006-10-09 18:11:45 -0700157 int ret;
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700158 struct ocfs2_journal *journal = osb->journal;
Mark Fashehccd979b2005-12-15 14:31:24 -0800159
160 BUG_ON(!handle);
161
Mark Fasheh1fabe142006-10-09 18:11:45 -0700162 ret = journal_stop(handle);
163 if (ret < 0)
164 mlog_errno(ret);
Mark Fashehccd979b2005-12-15 14:31:24 -0800165
166 up_read(&journal->j_trans_barrier);
167
Mark Fasheh1fabe142006-10-09 18:11:45 -0700168 return ret;
Mark Fashehccd979b2005-12-15 14:31:24 -0800169}
170
171/*
172 * 'nblocks' is what you want to add to the current
173 * transaction. extend_trans will either extend the current handle by
174 * nblocks, or commit it and start a new one with nblocks credits.
175 *
Mark Fashehe8aed342007-12-03 16:43:01 -0800176 * This might call journal_restart() which will commit dirty buffers
177 * and then restart the transaction. Before calling
178 * ocfs2_extend_trans(), any changed blocks should have been
179 * dirtied. After calling it, all blocks which need to be changed must
180 * go through another set of journal_access/journal_dirty calls.
181 *
Mark Fashehccd979b2005-12-15 14:31:24 -0800182 * WARNING: This will not release any semaphores or disk locks taken
183 * during the transaction, so make sure they were taken *before*
184 * start_trans or we'll have ordering deadlocks.
185 *
186 * WARNING2: Note that we do *not* drop j_trans_barrier here. This is
187 * good because transaction ids haven't yet been recorded on the
188 * cluster locks associated with this handle.
189 */
Mark Fasheh1fc58142006-10-05 14:15:36 -0700190int ocfs2_extend_trans(handle_t *handle, int nblocks)
Mark Fashehccd979b2005-12-15 14:31:24 -0800191{
192 int status;
193
194 BUG_ON(!handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800195 BUG_ON(!nblocks);
196
197 mlog_entry_void();
198
199 mlog(0, "Trying to extend transaction by %d blocks\n", nblocks);
200
Mark Fasheh0879c582007-12-03 16:42:19 -0800201#ifdef OCFS2_DEBUG_FS
202 status = 1;
203#else
Mark Fasheh1fc58142006-10-05 14:15:36 -0700204 status = journal_extend(handle, nblocks);
Mark Fashehccd979b2005-12-15 14:31:24 -0800205 if (status < 0) {
206 mlog_errno(status);
207 goto bail;
208 }
Mark Fasheh0879c582007-12-03 16:42:19 -0800209#endif
Mark Fashehccd979b2005-12-15 14:31:24 -0800210
211 if (status > 0) {
212 mlog(0, "journal_extend failed, trying journal_restart\n");
Mark Fasheh1fc58142006-10-05 14:15:36 -0700213 status = journal_restart(handle, nblocks);
Mark Fashehccd979b2005-12-15 14:31:24 -0800214 if (status < 0) {
Mark Fashehccd979b2005-12-15 14:31:24 -0800215 mlog_errno(status);
216 goto bail;
217 }
Mark Fasheh01ddf1e2006-10-05 13:54:39 -0700218 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800219
220 status = 0;
221bail:
222
223 mlog_exit(status);
224 return status;
225}
226
Mark Fasheh1fabe142006-10-09 18:11:45 -0700227int ocfs2_journal_access(handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -0800228 struct inode *inode,
229 struct buffer_head *bh,
230 int type)
231{
232 int status;
233
234 BUG_ON(!inode);
235 BUG_ON(!handle);
236 BUG_ON(!bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800237
Badari Pulavarty205f87f2006-03-26 01:38:00 -0800238 mlog_entry("bh->b_blocknr=%llu, type=%d (\"%s\"), bh->b_size = %zu\n",
Mark Fashehccd979b2005-12-15 14:31:24 -0800239 (unsigned long long)bh->b_blocknr, type,
240 (type == OCFS2_JOURNAL_ACCESS_CREATE) ?
241 "OCFS2_JOURNAL_ACCESS_CREATE" :
242 "OCFS2_JOURNAL_ACCESS_WRITE",
243 bh->b_size);
244
245 /* we can safely remove this assertion after testing. */
246 if (!buffer_uptodate(bh)) {
247 mlog(ML_ERROR, "giving me a buffer that's not uptodate!\n");
248 mlog(ML_ERROR, "b_blocknr=%llu\n",
249 (unsigned long long)bh->b_blocknr);
250 BUG();
251 }
252
253 /* Set the current transaction information on the inode so
254 * that the locking code knows whether it can drop it's locks
255 * on this inode or not. We're protected from the commit
256 * thread updating the current transaction id until
257 * ocfs2_commit_trans() because ocfs2_start_trans() took
258 * j_trans_barrier for us. */
259 ocfs2_set_inode_lock_trans(OCFS2_SB(inode->i_sb)->journal, inode);
260
Mark Fasheh251b6ec2006-01-10 15:41:43 -0800261 mutex_lock(&OCFS2_I(inode)->ip_io_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800262 switch (type) {
263 case OCFS2_JOURNAL_ACCESS_CREATE:
264 case OCFS2_JOURNAL_ACCESS_WRITE:
Mark Fasheh1fabe142006-10-09 18:11:45 -0700265 status = journal_get_write_access(handle, bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800266 break;
267
268 case OCFS2_JOURNAL_ACCESS_UNDO:
Mark Fasheh1fabe142006-10-09 18:11:45 -0700269 status = journal_get_undo_access(handle, bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800270 break;
271
272 default:
273 status = -EINVAL;
274 mlog(ML_ERROR, "Uknown access type!\n");
275 }
Mark Fasheh251b6ec2006-01-10 15:41:43 -0800276 mutex_unlock(&OCFS2_I(inode)->ip_io_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800277
278 if (status < 0)
279 mlog(ML_ERROR, "Error %d getting %d access to buffer!\n",
280 status, type);
281
282 mlog_exit(status);
283 return status;
284}
285
Mark Fasheh1fabe142006-10-09 18:11:45 -0700286int ocfs2_journal_dirty(handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -0800287 struct buffer_head *bh)
288{
289 int status;
290
Mark Fashehccd979b2005-12-15 14:31:24 -0800291 mlog_entry("(bh->b_blocknr=%llu)\n",
292 (unsigned long long)bh->b_blocknr);
293
Mark Fasheh1fabe142006-10-09 18:11:45 -0700294 status = journal_dirty_metadata(handle, bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800295 if (status < 0)
296 mlog(ML_ERROR, "Could not dirty metadata buffer. "
297 "(bh->b_blocknr=%llu)\n",
298 (unsigned long long)bh->b_blocknr);
299
300 mlog_exit(status);
301 return status;
302}
303
304int ocfs2_journal_dirty_data(handle_t *handle,
305 struct buffer_head *bh)
306{
307 int err = journal_dirty_data(handle, bh);
308 if (err)
309 mlog_errno(err);
310 /* TODO: When we can handle it, abort the handle and go RO on
311 * error here. */
312
313 return err;
314}
315
Mark Fashehd147b3d2007-11-07 14:40:36 -0800316#define OCFS2_DEFAULT_COMMIT_INTERVAL (HZ * JBD_DEFAULT_MAX_COMMIT_AGE)
Mark Fashehccd979b2005-12-15 14:31:24 -0800317
318void ocfs2_set_journal_params(struct ocfs2_super *osb)
319{
320 journal_t *journal = osb->journal->j_journal;
Mark Fashehd147b3d2007-11-07 14:40:36 -0800321 unsigned long commit_interval = OCFS2_DEFAULT_COMMIT_INTERVAL;
322
323 if (osb->osb_commit_interval)
324 commit_interval = osb->osb_commit_interval;
Mark Fashehccd979b2005-12-15 14:31:24 -0800325
326 spin_lock(&journal->j_state_lock);
Mark Fashehd147b3d2007-11-07 14:40:36 -0800327 journal->j_commit_interval = commit_interval;
Mark Fashehccd979b2005-12-15 14:31:24 -0800328 if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER)
329 journal->j_flags |= JFS_BARRIER;
330 else
331 journal->j_flags &= ~JFS_BARRIER;
332 spin_unlock(&journal->j_state_lock);
333}
334
335int ocfs2_journal_init(struct ocfs2_journal *journal, int *dirty)
336{
337 int status = -1;
338 struct inode *inode = NULL; /* the journal inode */
339 journal_t *j_journal = NULL;
340 struct ocfs2_dinode *di = NULL;
341 struct buffer_head *bh = NULL;
342 struct ocfs2_super *osb;
Mark Fashehe63aecb62007-10-18 15:30:42 -0700343 int inode_lock = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -0800344
345 mlog_entry_void();
346
347 BUG_ON(!journal);
348
349 osb = journal->j_osb;
350
351 /* already have the inode for our journal */
352 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
353 osb->slot_num);
354 if (inode == NULL) {
355 status = -EACCES;
356 mlog_errno(status);
357 goto done;
358 }
359 if (is_bad_inode(inode)) {
360 mlog(ML_ERROR, "access error (bad inode)\n");
361 iput(inode);
362 inode = NULL;
363 status = -EACCES;
364 goto done;
365 }
366
367 SET_INODE_JOURNAL(inode);
368 OCFS2_I(inode)->ip_open_count++;
369
Mark Fasheh6eff5792006-01-18 10:31:47 -0800370 /* Skip recovery waits here - journal inode metadata never
371 * changes in a live cluster so it can be considered an
372 * exception to the rule. */
Mark Fashehe63aecb62007-10-18 15:30:42 -0700373 status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY);
Mark Fashehccd979b2005-12-15 14:31:24 -0800374 if (status < 0) {
375 if (status != -ERESTARTSYS)
376 mlog(ML_ERROR, "Could not get lock on journal!\n");
377 goto done;
378 }
379
Mark Fashehe63aecb62007-10-18 15:30:42 -0700380 inode_lock = 1;
Mark Fashehccd979b2005-12-15 14:31:24 -0800381 di = (struct ocfs2_dinode *)bh->b_data;
382
383 if (inode->i_size < OCFS2_MIN_JOURNAL_SIZE) {
384 mlog(ML_ERROR, "Journal file size (%lld) is too small!\n",
385 inode->i_size);
386 status = -EINVAL;
387 goto done;
388 }
389
390 mlog(0, "inode->i_size = %lld\n", inode->i_size);
Andrew Morton5515eff2006-03-26 01:37:53 -0800391 mlog(0, "inode->i_blocks = %llu\n",
392 (unsigned long long)inode->i_blocks);
Mark Fashehccd979b2005-12-15 14:31:24 -0800393 mlog(0, "inode->ip_clusters = %u\n", OCFS2_I(inode)->ip_clusters);
394
395 /* call the kernels journal init function now */
396 j_journal = journal_init_inode(inode);
397 if (j_journal == NULL) {
398 mlog(ML_ERROR, "Linux journal layer error\n");
399 status = -EINVAL;
400 goto done;
401 }
402
403 mlog(0, "Returned from journal_init_inode\n");
404 mlog(0, "j_journal->j_maxlen = %u\n", j_journal->j_maxlen);
405
406 *dirty = (le32_to_cpu(di->id1.journal1.ij_flags) &
407 OCFS2_JOURNAL_DIRTY_FL);
408
409 journal->j_journal = j_journal;
410 journal->j_inode = inode;
411 journal->j_bh = bh;
412
413 ocfs2_set_journal_params(osb);
414
415 journal->j_state = OCFS2_JOURNAL_LOADED;
416
417 status = 0;
418done:
419 if (status < 0) {
Mark Fashehe63aecb62007-10-18 15:30:42 -0700420 if (inode_lock)
421 ocfs2_inode_unlock(inode, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -0800422 if (bh != NULL)
423 brelse(bh);
424 if (inode) {
425 OCFS2_I(inode)->ip_open_count--;
426 iput(inode);
427 }
428 }
429
430 mlog_exit(status);
431 return status;
432}
433
434static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
435 int dirty)
436{
437 int status;
438 unsigned int flags;
439 struct ocfs2_journal *journal = osb->journal;
440 struct buffer_head *bh = journal->j_bh;
441 struct ocfs2_dinode *fe;
442
443 mlog_entry_void();
444
445 fe = (struct ocfs2_dinode *)bh->b_data;
446 if (!OCFS2_IS_VALID_DINODE(fe)) {
447 /* This is called from startup/shutdown which will
448 * handle the errors in a specific manner, so no need
449 * to call ocfs2_error() here. */
Mark Fashehb0697052006-03-03 10:24:33 -0800450 mlog(ML_ERROR, "Journal dinode %llu has invalid "
Mark Fasheh1ca1a112007-04-27 16:01:25 -0700451 "signature: %.*s",
452 (unsigned long long)le64_to_cpu(fe->i_blkno), 7,
Mark Fashehb0697052006-03-03 10:24:33 -0800453 fe->i_signature);
Mark Fashehccd979b2005-12-15 14:31:24 -0800454 status = -EIO;
455 goto out;
456 }
457
458 flags = le32_to_cpu(fe->id1.journal1.ij_flags);
459 if (dirty)
460 flags |= OCFS2_JOURNAL_DIRTY_FL;
461 else
462 flags &= ~OCFS2_JOURNAL_DIRTY_FL;
463 fe->id1.journal1.ij_flags = cpu_to_le32(flags);
464
465 status = ocfs2_write_block(osb, bh, journal->j_inode);
466 if (status < 0)
467 mlog_errno(status);
468
469out:
470 mlog_exit(status);
471 return status;
472}
473
474/*
475 * If the journal has been kmalloc'd it needs to be freed after this
476 * call.
477 */
478void ocfs2_journal_shutdown(struct ocfs2_super *osb)
479{
480 struct ocfs2_journal *journal = NULL;
481 int status = 0;
482 struct inode *inode = NULL;
483 int num_running_trans = 0;
484
485 mlog_entry_void();
486
Eric Sesterhenn / snakebyteebdec832006-01-27 10:32:52 +0100487 BUG_ON(!osb);
Mark Fashehccd979b2005-12-15 14:31:24 -0800488
489 journal = osb->journal;
490 if (!journal)
491 goto done;
492
493 inode = journal->j_inode;
494
495 if (journal->j_state != OCFS2_JOURNAL_LOADED)
496 goto done;
497
498 /* need to inc inode use count as journal_destroy will iput. */
499 if (!igrab(inode))
500 BUG();
501
502 num_running_trans = atomic_read(&(osb->journal->j_num_trans));
503 if (num_running_trans > 0)
504 mlog(0, "Shutting down journal: must wait on %d "
505 "running transactions!\n",
506 num_running_trans);
507
508 /* Do a commit_cache here. It will flush our journal, *and*
509 * release any locks that are still held.
510 * set the SHUTDOWN flag and release the trans lock.
511 * the commit thread will take the trans lock for us below. */
512 journal->j_state = OCFS2_JOURNAL_IN_SHUTDOWN;
513
514 /* The OCFS2_JOURNAL_IN_SHUTDOWN will signal to commit_cache to not
515 * drop the trans_lock (which we want to hold until we
516 * completely destroy the journal. */
517 if (osb->commit_task) {
518 /* Wait for the commit thread */
519 mlog(0, "Waiting for ocfs2commit to exit....\n");
520 kthread_stop(osb->commit_task);
521 osb->commit_task = NULL;
522 }
523
524 BUG_ON(atomic_read(&(osb->journal->j_num_trans)) != 0);
525
Sunil Mushranc271c5c2006-12-05 17:56:35 -0800526 if (ocfs2_mount_local(osb)) {
527 journal_lock_updates(journal->j_journal);
528 status = journal_flush(journal->j_journal);
529 journal_unlock_updates(journal->j_journal);
530 if (status < 0)
531 mlog_errno(status);
532 }
533
534 if (status == 0) {
535 /*
536 * Do not toggle if flush was unsuccessful otherwise
537 * will leave dirty metadata in a "clean" journal
538 */
539 status = ocfs2_journal_toggle_dirty(osb, 0);
540 if (status < 0)
541 mlog_errno(status);
542 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800543
544 /* Shutdown the kernel journal system */
545 journal_destroy(journal->j_journal);
546
547 OCFS2_I(inode)->ip_open_count--;
548
549 /* unlock our journal */
Mark Fashehe63aecb62007-10-18 15:30:42 -0700550 ocfs2_inode_unlock(inode, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -0800551
552 brelse(journal->j_bh);
553 journal->j_bh = NULL;
554
555 journal->j_state = OCFS2_JOURNAL_FREE;
556
557// up_write(&journal->j_trans_barrier);
558done:
559 if (inode)
560 iput(inode);
561 mlog_exit_void();
562}
563
564static void ocfs2_clear_journal_error(struct super_block *sb,
565 journal_t *journal,
566 int slot)
567{
568 int olderr;
569
570 olderr = journal_errno(journal);
571 if (olderr) {
572 mlog(ML_ERROR, "File system error %d recorded in "
573 "journal %u.\n", olderr, slot);
574 mlog(ML_ERROR, "File system on device %s needs checking.\n",
575 sb->s_id);
576
577 journal_ack_err(journal);
578 journal_clear_err(journal);
579 }
580}
581
Sunil Mushranc271c5c2006-12-05 17:56:35 -0800582int ocfs2_journal_load(struct ocfs2_journal *journal, int local)
Mark Fashehccd979b2005-12-15 14:31:24 -0800583{
584 int status = 0;
585 struct ocfs2_super *osb;
586
587 mlog_entry_void();
588
589 if (!journal)
590 BUG();
591
592 osb = journal->j_osb;
593
594 status = journal_load(journal->j_journal);
595 if (status < 0) {
596 mlog(ML_ERROR, "Failed to load journal!\n");
597 goto done;
598 }
599
600 ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num);
601
602 status = ocfs2_journal_toggle_dirty(osb, 1);
603 if (status < 0) {
604 mlog_errno(status);
605 goto done;
606 }
607
608 /* Launch the commit thread */
Sunil Mushranc271c5c2006-12-05 17:56:35 -0800609 if (!local) {
610 osb->commit_task = kthread_run(ocfs2_commit_thread, osb,
611 "ocfs2cmt");
612 if (IS_ERR(osb->commit_task)) {
613 status = PTR_ERR(osb->commit_task);
614 osb->commit_task = NULL;
615 mlog(ML_ERROR, "unable to launch ocfs2commit thread, "
616 "error=%d", status);
617 goto done;
618 }
619 } else
Mark Fashehccd979b2005-12-15 14:31:24 -0800620 osb->commit_task = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800621
622done:
623 mlog_exit(status);
624 return status;
625}
626
627
628/* 'full' flag tells us whether we clear out all blocks or if we just
629 * mark the journal clean */
630int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full)
631{
632 int status;
633
634 mlog_entry_void();
635
Eric Sesterhenn / snakebyteebdec832006-01-27 10:32:52 +0100636 BUG_ON(!journal);
Mark Fashehccd979b2005-12-15 14:31:24 -0800637
638 status = journal_wipe(journal->j_journal, full);
639 if (status < 0) {
640 mlog_errno(status);
641 goto bail;
642 }
643
644 status = ocfs2_journal_toggle_dirty(journal->j_osb, 0);
645 if (status < 0)
646 mlog_errno(status);
647
648bail:
649 mlog_exit(status);
650 return status;
651}
652
653/*
654 * JBD Might read a cached version of another nodes journal file. We
655 * don't want this as this file changes often and we get no
656 * notification on those changes. The only way to be sure that we've
657 * got the most up to date version of those blocks then is to force
658 * read them off disk. Just searching through the buffer cache won't
659 * work as there may be pages backing this file which are still marked
660 * up to date. We know things can't change on this file underneath us
661 * as we have the lock by now :)
662 */
663static int ocfs2_force_read_journal(struct inode *inode)
664{
665 int status = 0;
Mark Fasheh4f902c32007-03-09 16:26:50 -0800666 int i;
Mark Fasheh8110b072007-03-22 16:53:23 -0700667 u64 v_blkno, p_blkno, p_blocks, num_blocks;
Mark Fasheh4f902c32007-03-09 16:26:50 -0800668#define CONCURRENT_JOURNAL_FILL 32ULL
Mark Fashehccd979b2005-12-15 14:31:24 -0800669 struct buffer_head *bhs[CONCURRENT_JOURNAL_FILL];
670
671 mlog_entry_void();
672
Mark Fashehccd979b2005-12-15 14:31:24 -0800673 memset(bhs, 0, sizeof(struct buffer_head *) * CONCURRENT_JOURNAL_FILL);
674
Mark Fasheh8110b072007-03-22 16:53:23 -0700675 num_blocks = ocfs2_blocks_for_bytes(inode->i_sb, inode->i_size);
Mark Fashehccd979b2005-12-15 14:31:24 -0800676 v_blkno = 0;
Mark Fasheh8110b072007-03-22 16:53:23 -0700677 while (v_blkno < num_blocks) {
Mark Fashehccd979b2005-12-15 14:31:24 -0800678 status = ocfs2_extent_map_get_blocks(inode, v_blkno,
Mark Fasheh49cb8d22007-03-09 16:21:46 -0800679 &p_blkno, &p_blocks, NULL);
Mark Fashehccd979b2005-12-15 14:31:24 -0800680 if (status < 0) {
681 mlog_errno(status);
682 goto bail;
683 }
684
685 if (p_blocks > CONCURRENT_JOURNAL_FILL)
686 p_blocks = CONCURRENT_JOURNAL_FILL;
687
Mark Fashehdd4a2c22006-04-12 14:24:05 -0700688 /* We are reading journal data which should not
689 * be put in the uptodate cache */
Mark Fashehccd979b2005-12-15 14:31:24 -0800690 status = ocfs2_read_blocks(OCFS2_SB(inode->i_sb),
691 p_blkno, p_blocks, bhs, 0,
Mark Fashehdd4a2c22006-04-12 14:24:05 -0700692 NULL);
Mark Fashehccd979b2005-12-15 14:31:24 -0800693 if (status < 0) {
694 mlog_errno(status);
695 goto bail;
696 }
697
698 for(i = 0; i < p_blocks; i++) {
699 brelse(bhs[i]);
700 bhs[i] = NULL;
701 }
702
703 v_blkno += p_blocks;
704 }
705
706bail:
707 for(i = 0; i < CONCURRENT_JOURNAL_FILL; i++)
708 if (bhs[i])
709 brelse(bhs[i]);
710 mlog_exit(status);
711 return status;
712}
713
714struct ocfs2_la_recovery_item {
715 struct list_head lri_list;
716 int lri_slot;
717 struct ocfs2_dinode *lri_la_dinode;
718 struct ocfs2_dinode *lri_tl_dinode;
719};
720
721/* Does the second half of the recovery process. By this point, the
722 * node is marked clean and can actually be considered recovered,
723 * hence it's no longer in the recovery map, but there's still some
724 * cleanup we can do which shouldn't happen within the recovery thread
725 * as locking in that context becomes very difficult if we are to take
726 * recovering nodes into account.
727 *
728 * NOTE: This function can and will sleep on recovery of other nodes
729 * during cluster locking, just like any other ocfs2 process.
730 */
David Howellsc4028952006-11-22 14:57:56 +0000731void ocfs2_complete_recovery(struct work_struct *work)
Mark Fashehccd979b2005-12-15 14:31:24 -0800732{
733 int ret;
David Howellsc4028952006-11-22 14:57:56 +0000734 struct ocfs2_journal *journal =
735 container_of(work, struct ocfs2_journal, j_recovery_work);
736 struct ocfs2_super *osb = journal->j_osb;
Mark Fashehccd979b2005-12-15 14:31:24 -0800737 struct ocfs2_dinode *la_dinode, *tl_dinode;
Christoph Hellwig800deef2007-05-17 16:03:13 +0200738 struct ocfs2_la_recovery_item *item, *n;
Mark Fashehccd979b2005-12-15 14:31:24 -0800739 LIST_HEAD(tmp_la_list);
740
741 mlog_entry_void();
742
743 mlog(0, "completing recovery from keventd\n");
744
745 spin_lock(&journal->j_lock);
746 list_splice_init(&journal->j_la_cleanups, &tmp_la_list);
747 spin_unlock(&journal->j_lock);
748
Christoph Hellwig800deef2007-05-17 16:03:13 +0200749 list_for_each_entry_safe(item, n, &tmp_la_list, lri_list) {
Mark Fashehccd979b2005-12-15 14:31:24 -0800750 list_del_init(&item->lri_list);
751
752 mlog(0, "Complete recovery for slot %d\n", item->lri_slot);
753
754 la_dinode = item->lri_la_dinode;
755 if (la_dinode) {
Mark Fashehb0697052006-03-03 10:24:33 -0800756 mlog(0, "Clean up local alloc %llu\n",
Mark Fasheh1ca1a112007-04-27 16:01:25 -0700757 (unsigned long long)le64_to_cpu(la_dinode->i_blkno));
Mark Fashehccd979b2005-12-15 14:31:24 -0800758
759 ret = ocfs2_complete_local_alloc_recovery(osb,
760 la_dinode);
761 if (ret < 0)
762 mlog_errno(ret);
763
764 kfree(la_dinode);
765 }
766
767 tl_dinode = item->lri_tl_dinode;
768 if (tl_dinode) {
Mark Fashehb0697052006-03-03 10:24:33 -0800769 mlog(0, "Clean up truncate log %llu\n",
Mark Fasheh1ca1a112007-04-27 16:01:25 -0700770 (unsigned long long)le64_to_cpu(tl_dinode->i_blkno));
Mark Fashehccd979b2005-12-15 14:31:24 -0800771
772 ret = ocfs2_complete_truncate_log_recovery(osb,
773 tl_dinode);
774 if (ret < 0)
775 mlog_errno(ret);
776
777 kfree(tl_dinode);
778 }
779
780 ret = ocfs2_recover_orphans(osb, item->lri_slot);
781 if (ret < 0)
782 mlog_errno(ret);
783
784 kfree(item);
785 }
786
787 mlog(0, "Recovery completion\n");
788 mlog_exit_void();
789}
790
791/* NOTE: This function always eats your references to la_dinode and
792 * tl_dinode, either manually on error, or by passing them to
793 * ocfs2_complete_recovery */
794static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal,
795 int slot_num,
796 struct ocfs2_dinode *la_dinode,
797 struct ocfs2_dinode *tl_dinode)
798{
799 struct ocfs2_la_recovery_item *item;
800
Sunil Mushranafae00ab2006-04-12 14:37:00 -0700801 item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS);
Mark Fashehccd979b2005-12-15 14:31:24 -0800802 if (!item) {
803 /* Though we wish to avoid it, we are in fact safe in
804 * skipping local alloc cleanup as fsck.ocfs2 is more
805 * than capable of reclaiming unused space. */
806 if (la_dinode)
807 kfree(la_dinode);
808
809 if (tl_dinode)
810 kfree(tl_dinode);
811
812 mlog_errno(-ENOMEM);
813 return;
814 }
815
816 INIT_LIST_HEAD(&item->lri_list);
817 item->lri_la_dinode = la_dinode;
818 item->lri_slot = slot_num;
819 item->lri_tl_dinode = tl_dinode;
820
821 spin_lock(&journal->j_lock);
822 list_add_tail(&item->lri_list, &journal->j_la_cleanups);
823 queue_work(ocfs2_wq, &journal->j_recovery_work);
824 spin_unlock(&journal->j_lock);
825}
826
827/* Called by the mount code to queue recovery the last part of
828 * recovery for it's own slot. */
829void ocfs2_complete_mount_recovery(struct ocfs2_super *osb)
830{
831 struct ocfs2_journal *journal = osb->journal;
832
833 if (osb->dirty) {
834 /* No need to queue up our truncate_log as regular
835 * cleanup will catch that. */
836 ocfs2_queue_recovery_completion(journal,
837 osb->slot_num,
838 osb->local_alloc_copy,
839 NULL);
840 ocfs2_schedule_truncate_log_flush(osb, 0);
841
842 osb->local_alloc_copy = NULL;
843 osb->dirty = 0;
844 }
845}
846
847static int __ocfs2_recovery_thread(void *arg)
848{
849 int status, node_num;
850 struct ocfs2_super *osb = arg;
851
852 mlog_entry_void();
853
854 status = ocfs2_wait_on_mount(osb);
855 if (status < 0) {
856 goto bail;
857 }
858
859restart:
860 status = ocfs2_super_lock(osb, 1);
861 if (status < 0) {
862 mlog_errno(status);
863 goto bail;
864 }
865
866 while(!ocfs2_node_map_is_empty(osb, &osb->recovery_map)) {
867 node_num = ocfs2_node_map_first_set_bit(osb,
868 &osb->recovery_map);
869 if (node_num == O2NM_INVALID_NODE_NUM) {
870 mlog(0, "Out of nodes to recover.\n");
871 break;
872 }
873
874 status = ocfs2_recover_node(osb, node_num);
875 if (status < 0) {
876 mlog(ML_ERROR,
877 "Error %d recovering node %d on device (%u,%u)!\n",
878 status, node_num,
879 MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
880 mlog(ML_ERROR, "Volume requires unmount.\n");
881 continue;
882 }
883
884 ocfs2_recovery_map_clear(osb, node_num);
885 }
886 ocfs2_super_unlock(osb, 1);
887
888 /* We always run recovery on our own orphan dir - the dead
Mark Fasheh34d024f2007-09-24 15:56:19 -0700889 * node(s) may have disallowd a previos inode delete. Re-processing
890 * is therefore required. */
Mark Fashehccd979b2005-12-15 14:31:24 -0800891 ocfs2_queue_recovery_completion(osb->journal, osb->slot_num, NULL,
892 NULL);
893
894bail:
Arjan van de Venc74ec2f2006-01-13 21:54:23 -0800895 mutex_lock(&osb->recovery_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -0800896 if (!status &&
897 !ocfs2_node_map_is_empty(osb, &osb->recovery_map)) {
Arjan van de Venc74ec2f2006-01-13 21:54:23 -0800898 mutex_unlock(&osb->recovery_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -0800899 goto restart;
900 }
901
902 osb->recovery_thread_task = NULL;
903 mb(); /* sync with ocfs2_recovery_thread_running */
904 wake_up(&osb->recovery_event);
905
Arjan van de Venc74ec2f2006-01-13 21:54:23 -0800906 mutex_unlock(&osb->recovery_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -0800907
908 mlog_exit(status);
909 /* no one is callint kthread_stop() for us so the kthread() api
910 * requires that we call do_exit(). And it isn't exported, but
911 * complete_and_exit() seems to be a minimal wrapper around it. */
912 complete_and_exit(NULL, status);
913 return status;
914}
915
916void ocfs2_recovery_thread(struct ocfs2_super *osb, int node_num)
917{
918 mlog_entry("(node_num=%d, osb->node_num = %d)\n",
919 node_num, osb->node_num);
920
Arjan van de Venc74ec2f2006-01-13 21:54:23 -0800921 mutex_lock(&osb->recovery_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -0800922 if (osb->disable_recovery)
923 goto out;
924
925 /* People waiting on recovery will wait on
926 * the recovery map to empty. */
927 if (!ocfs2_recovery_map_set(osb, node_num))
928 mlog(0, "node %d already be in recovery.\n", node_num);
929
930 mlog(0, "starting recovery thread...\n");
931
932 if (osb->recovery_thread_task)
933 goto out;
934
935 osb->recovery_thread_task = kthread_run(__ocfs2_recovery_thread, osb,
Mark Fasheh78427042006-05-04 12:03:26 -0700936 "ocfs2rec");
Mark Fashehccd979b2005-12-15 14:31:24 -0800937 if (IS_ERR(osb->recovery_thread_task)) {
938 mlog_errno((int)PTR_ERR(osb->recovery_thread_task));
939 osb->recovery_thread_task = NULL;
940 }
941
942out:
Arjan van de Venc74ec2f2006-01-13 21:54:23 -0800943 mutex_unlock(&osb->recovery_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -0800944 wake_up(&osb->recovery_event);
945
946 mlog_exit_void();
947}
948
949/* Does the actual journal replay and marks the journal inode as
950 * clean. Will only replay if the journal inode is marked dirty. */
951static int ocfs2_replay_journal(struct ocfs2_super *osb,
952 int node_num,
953 int slot_num)
954{
955 int status;
956 int got_lock = 0;
957 unsigned int flags;
958 struct inode *inode = NULL;
959 struct ocfs2_dinode *fe;
960 journal_t *journal = NULL;
961 struct buffer_head *bh = NULL;
962
963 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
964 slot_num);
965 if (inode == NULL) {
966 status = -EACCES;
967 mlog_errno(status);
968 goto done;
969 }
970 if (is_bad_inode(inode)) {
971 status = -EACCES;
972 iput(inode);
973 inode = NULL;
974 mlog_errno(status);
975 goto done;
976 }
977 SET_INODE_JOURNAL(inode);
978
Mark Fashehe63aecb62007-10-18 15:30:42 -0700979 status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY);
Mark Fashehccd979b2005-12-15 14:31:24 -0800980 if (status < 0) {
Mark Fashehe63aecb62007-10-18 15:30:42 -0700981 mlog(0, "status returned from ocfs2_inode_lock=%d\n", status);
Mark Fashehccd979b2005-12-15 14:31:24 -0800982 if (status != -ERESTARTSYS)
983 mlog(ML_ERROR, "Could not lock journal!\n");
984 goto done;
985 }
986 got_lock = 1;
987
988 fe = (struct ocfs2_dinode *) bh->b_data;
989
990 flags = le32_to_cpu(fe->id1.journal1.ij_flags);
991
992 if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) {
993 mlog(0, "No recovery required for node %d\n", node_num);
994 goto done;
995 }
996
997 mlog(ML_NOTICE, "Recovering node %d from slot %d on device (%u,%u)\n",
998 node_num, slot_num,
999 MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1000
1001 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
1002
1003 status = ocfs2_force_read_journal(inode);
1004 if (status < 0) {
1005 mlog_errno(status);
1006 goto done;
1007 }
1008
1009 mlog(0, "calling journal_init_inode\n");
1010 journal = journal_init_inode(inode);
1011 if (journal == NULL) {
1012 mlog(ML_ERROR, "Linux journal layer error\n");
1013 status = -EIO;
1014 goto done;
1015 }
1016
1017 status = journal_load(journal);
1018 if (status < 0) {
1019 mlog_errno(status);
1020 if (!igrab(inode))
1021 BUG();
1022 journal_destroy(journal);
1023 goto done;
1024 }
1025
1026 ocfs2_clear_journal_error(osb->sb, journal, slot_num);
1027
1028 /* wipe the journal */
1029 mlog(0, "flushing the journal.\n");
1030 journal_lock_updates(journal);
1031 status = journal_flush(journal);
1032 journal_unlock_updates(journal);
1033 if (status < 0)
1034 mlog_errno(status);
1035
1036 /* This will mark the node clean */
1037 flags = le32_to_cpu(fe->id1.journal1.ij_flags);
1038 flags &= ~OCFS2_JOURNAL_DIRTY_FL;
1039 fe->id1.journal1.ij_flags = cpu_to_le32(flags);
1040
1041 status = ocfs2_write_block(osb, bh, inode);
1042 if (status < 0)
1043 mlog_errno(status);
1044
1045 if (!igrab(inode))
1046 BUG();
1047
1048 journal_destroy(journal);
1049
1050done:
1051 /* drop the lock on this nodes journal */
1052 if (got_lock)
Mark Fashehe63aecb62007-10-18 15:30:42 -07001053 ocfs2_inode_unlock(inode, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -08001054
1055 if (inode)
1056 iput(inode);
1057
1058 if (bh)
1059 brelse(bh);
1060
1061 mlog_exit(status);
1062 return status;
1063}
1064
1065/*
1066 * Do the most important parts of node recovery:
1067 * - Replay it's journal
1068 * - Stamp a clean local allocator file
1069 * - Stamp a clean truncate log
1070 * - Mark the node clean
1071 *
1072 * If this function completes without error, a node in OCFS2 can be
1073 * said to have been safely recovered. As a result, failure during the
1074 * second part of a nodes recovery process (local alloc recovery) is
1075 * far less concerning.
1076 */
1077static int ocfs2_recover_node(struct ocfs2_super *osb,
1078 int node_num)
1079{
1080 int status = 0;
1081 int slot_num;
1082 struct ocfs2_slot_info *si = osb->slot_info;
1083 struct ocfs2_dinode *la_copy = NULL;
1084 struct ocfs2_dinode *tl_copy = NULL;
1085
1086 mlog_entry("(node_num=%d, osb->node_num = %d)\n",
1087 node_num, osb->node_num);
1088
1089 mlog(0, "checking node %d\n", node_num);
1090
1091 /* Should not ever be called to recover ourselves -- in that
1092 * case we should've called ocfs2_journal_load instead. */
Eric Sesterhenn / snakebyteebdec832006-01-27 10:32:52 +01001093 BUG_ON(osb->node_num == node_num);
Mark Fashehccd979b2005-12-15 14:31:24 -08001094
1095 slot_num = ocfs2_node_num_to_slot(si, node_num);
1096 if (slot_num == OCFS2_INVALID_SLOT) {
1097 status = 0;
1098 mlog(0, "no slot for this node, so no recovery required.\n");
1099 goto done;
1100 }
1101
1102 mlog(0, "node %d was using slot %d\n", node_num, slot_num);
1103
1104 status = ocfs2_replay_journal(osb, node_num, slot_num);
1105 if (status < 0) {
1106 mlog_errno(status);
1107 goto done;
1108 }
1109
1110 /* Stamp a clean local alloc file AFTER recovering the journal... */
1111 status = ocfs2_begin_local_alloc_recovery(osb, slot_num, &la_copy);
1112 if (status < 0) {
1113 mlog_errno(status);
1114 goto done;
1115 }
1116
1117 /* An error from begin_truncate_log_recovery is not
1118 * serious enough to warrant halting the rest of
1119 * recovery. */
1120 status = ocfs2_begin_truncate_log_recovery(osb, slot_num, &tl_copy);
1121 if (status < 0)
1122 mlog_errno(status);
1123
1124 /* Likewise, this would be a strange but ultimately not so
1125 * harmful place to get an error... */
1126 ocfs2_clear_slot(si, slot_num);
1127 status = ocfs2_update_disk_slots(osb, si);
1128 if (status < 0)
1129 mlog_errno(status);
1130
1131 /* This will kfree the memory pointed to by la_copy and tl_copy */
1132 ocfs2_queue_recovery_completion(osb->journal, slot_num, la_copy,
1133 tl_copy);
1134
1135 status = 0;
1136done:
1137
1138 mlog_exit(status);
1139 return status;
1140}
1141
1142/* Test node liveness by trylocking his journal. If we get the lock,
1143 * we drop it here. Return 0 if we got the lock, -EAGAIN if node is
1144 * still alive (we couldn't get the lock) and < 0 on error. */
1145static int ocfs2_trylock_journal(struct ocfs2_super *osb,
1146 int slot_num)
1147{
1148 int status, flags;
1149 struct inode *inode = NULL;
1150
1151 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
1152 slot_num);
1153 if (inode == NULL) {
1154 mlog(ML_ERROR, "access error\n");
1155 status = -EACCES;
1156 goto bail;
1157 }
1158 if (is_bad_inode(inode)) {
1159 mlog(ML_ERROR, "access error (bad inode)\n");
1160 iput(inode);
1161 inode = NULL;
1162 status = -EACCES;
1163 goto bail;
1164 }
1165 SET_INODE_JOURNAL(inode);
1166
1167 flags = OCFS2_META_LOCK_RECOVERY | OCFS2_META_LOCK_NOQUEUE;
Mark Fashehe63aecb62007-10-18 15:30:42 -07001168 status = ocfs2_inode_lock_full(inode, NULL, 1, flags);
Mark Fashehccd979b2005-12-15 14:31:24 -08001169 if (status < 0) {
1170 if (status != -EAGAIN)
1171 mlog_errno(status);
1172 goto bail;
1173 }
1174
Mark Fashehe63aecb62007-10-18 15:30:42 -07001175 ocfs2_inode_unlock(inode, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -08001176bail:
1177 if (inode)
1178 iput(inode);
1179
1180 return status;
1181}
1182
1183/* Call this underneath ocfs2_super_lock. It also assumes that the
1184 * slot info struct has been updated from disk. */
1185int ocfs2_mark_dead_nodes(struct ocfs2_super *osb)
1186{
1187 int status, i, node_num;
1188 struct ocfs2_slot_info *si = osb->slot_info;
1189
1190 /* This is called with the super block cluster lock, so we
1191 * know that the slot map can't change underneath us. */
1192
1193 spin_lock(&si->si_lock);
1194 for(i = 0; i < si->si_num_slots; i++) {
1195 if (i == osb->slot_num)
1196 continue;
1197 if (ocfs2_is_empty_slot(si, i))
1198 continue;
1199
1200 node_num = si->si_global_node_nums[i];
1201 if (ocfs2_node_map_test_bit(osb, &osb->recovery_map, node_num))
1202 continue;
1203 spin_unlock(&si->si_lock);
1204
1205 /* Ok, we have a slot occupied by another node which
1206 * is not in the recovery map. We trylock his journal
1207 * file here to test if he's alive. */
1208 status = ocfs2_trylock_journal(osb, i);
1209 if (!status) {
1210 /* Since we're called from mount, we know that
1211 * the recovery thread can't race us on
1212 * setting / checking the recovery bits. */
1213 ocfs2_recovery_thread(osb, node_num);
1214 } else if ((status < 0) && (status != -EAGAIN)) {
1215 mlog_errno(status);
1216 goto bail;
1217 }
1218
1219 spin_lock(&si->si_lock);
1220 }
1221 spin_unlock(&si->si_lock);
1222
1223 status = 0;
1224bail:
1225 mlog_exit(status);
1226 return status;
1227}
1228
Mark Fasheh5eae5b92007-09-10 17:50:51 -07001229struct ocfs2_orphan_filldir_priv {
1230 struct inode *head;
1231 struct ocfs2_super *osb;
1232};
1233
1234static int ocfs2_orphan_filldir(void *priv, const char *name, int name_len,
1235 loff_t pos, u64 ino, unsigned type)
1236{
1237 struct ocfs2_orphan_filldir_priv *p = priv;
1238 struct inode *iter;
1239
1240 if (name_len == 1 && !strncmp(".", name, 1))
1241 return 0;
1242 if (name_len == 2 && !strncmp("..", name, 2))
1243 return 0;
1244
1245 /* Skip bad inodes so that recovery can continue */
1246 iter = ocfs2_iget(p->osb, ino,
Jan Kara5fa06132008-01-11 00:11:45 +01001247 OCFS2_FI_FLAG_ORPHAN_RECOVERY, 0);
Mark Fasheh5eae5b92007-09-10 17:50:51 -07001248 if (IS_ERR(iter))
1249 return 0;
1250
1251 mlog(0, "queue orphan %llu\n",
1252 (unsigned long long)OCFS2_I(iter)->ip_blkno);
1253 /* No locking is required for the next_orphan queue as there
1254 * is only ever a single process doing orphan recovery. */
1255 OCFS2_I(iter)->ip_next_orphan = p->head;
1256 p->head = iter;
1257
1258 return 0;
1259}
1260
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001261static int ocfs2_queue_orphans(struct ocfs2_super *osb,
1262 int slot,
1263 struct inode **head)
Mark Fashehccd979b2005-12-15 14:31:24 -08001264{
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001265 int status;
Mark Fashehccd979b2005-12-15 14:31:24 -08001266 struct inode *orphan_dir_inode = NULL;
Mark Fasheh5eae5b92007-09-10 17:50:51 -07001267 struct ocfs2_orphan_filldir_priv priv;
1268 loff_t pos = 0;
1269
1270 priv.osb = osb;
1271 priv.head = *head;
Mark Fashehccd979b2005-12-15 14:31:24 -08001272
1273 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
1274 ORPHAN_DIR_SYSTEM_INODE,
1275 slot);
1276 if (!orphan_dir_inode) {
1277 status = -ENOENT;
1278 mlog_errno(status);
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001279 return status;
1280 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001281
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001282 mutex_lock(&orphan_dir_inode->i_mutex);
Mark Fashehe63aecb62007-10-18 15:30:42 -07001283 status = ocfs2_inode_lock(orphan_dir_inode, NULL, 0);
Mark Fashehccd979b2005-12-15 14:31:24 -08001284 if (status < 0) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001285 mlog_errno(status);
1286 goto out;
1287 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001288
Mark Fasheh5eae5b92007-09-10 17:50:51 -07001289 status = ocfs2_dir_foreach(orphan_dir_inode, &pos, &priv,
1290 ocfs2_orphan_filldir);
1291 if (status) {
1292 mlog_errno(status);
Mark Fasheha86370f2007-12-03 14:06:23 -08001293 goto out_cluster;
Mark Fashehccd979b2005-12-15 14:31:24 -08001294 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001295
Mark Fasheh5eae5b92007-09-10 17:50:51 -07001296 *head = priv.head;
1297
Mark Fasheha86370f2007-12-03 14:06:23 -08001298out_cluster:
Mark Fashehe63aecb62007-10-18 15:30:42 -07001299 ocfs2_inode_unlock(orphan_dir_inode, 0);
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001300out:
1301 mutex_unlock(&orphan_dir_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -08001302 iput(orphan_dir_inode);
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001303 return status;
1304}
1305
1306static int ocfs2_orphan_recovery_can_continue(struct ocfs2_super *osb,
1307 int slot)
1308{
1309 int ret;
1310
1311 spin_lock(&osb->osb_lock);
1312 ret = !osb->osb_orphan_wipes[slot];
1313 spin_unlock(&osb->osb_lock);
1314 return ret;
1315}
1316
1317static void ocfs2_mark_recovering_orphan_dir(struct ocfs2_super *osb,
1318 int slot)
1319{
1320 spin_lock(&osb->osb_lock);
1321 /* Mark ourselves such that new processes in delete_inode()
1322 * know to quit early. */
1323 ocfs2_node_map_set_bit(osb, &osb->osb_recovering_orphan_dirs, slot);
1324 while (osb->osb_orphan_wipes[slot]) {
1325 /* If any processes are already in the middle of an
1326 * orphan wipe on this dir, then we need to wait for
1327 * them. */
1328 spin_unlock(&osb->osb_lock);
1329 wait_event_interruptible(osb->osb_wipe_event,
1330 ocfs2_orphan_recovery_can_continue(osb, slot));
1331 spin_lock(&osb->osb_lock);
1332 }
1333 spin_unlock(&osb->osb_lock);
1334}
1335
1336static void ocfs2_clear_recovering_orphan_dir(struct ocfs2_super *osb,
1337 int slot)
1338{
1339 ocfs2_node_map_clear_bit(osb, &osb->osb_recovering_orphan_dirs, slot);
1340}
1341
1342/*
1343 * Orphan recovery. Each mounted node has it's own orphan dir which we
1344 * must run during recovery. Our strategy here is to build a list of
1345 * the inodes in the orphan dir and iget/iput them. The VFS does
1346 * (most) of the rest of the work.
1347 *
1348 * Orphan recovery can happen at any time, not just mount so we have a
1349 * couple of extra considerations.
1350 *
1351 * - We grab as many inodes as we can under the orphan dir lock -
1352 * doing iget() outside the orphan dir risks getting a reference on
1353 * an invalid inode.
1354 * - We must be sure not to deadlock with other processes on the
1355 * system wanting to run delete_inode(). This can happen when they go
1356 * to lock the orphan dir and the orphan recovery process attempts to
1357 * iget() inside the orphan dir lock. This can be avoided by
1358 * advertising our state to ocfs2_delete_inode().
1359 */
1360static int ocfs2_recover_orphans(struct ocfs2_super *osb,
1361 int slot)
1362{
1363 int ret = 0;
1364 struct inode *inode = NULL;
1365 struct inode *iter;
1366 struct ocfs2_inode_info *oi;
1367
1368 mlog(0, "Recover inodes from orphan dir in slot %d\n", slot);
1369
1370 ocfs2_mark_recovering_orphan_dir(osb, slot);
1371 ret = ocfs2_queue_orphans(osb, slot, &inode);
1372 ocfs2_clear_recovering_orphan_dir(osb, slot);
1373
1374 /* Error here should be noted, but we want to continue with as
1375 * many queued inodes as we've got. */
1376 if (ret)
1377 mlog_errno(ret);
Mark Fashehccd979b2005-12-15 14:31:24 -08001378
1379 while (inode) {
1380 oi = OCFS2_I(inode);
Mark Fashehb0697052006-03-03 10:24:33 -08001381 mlog(0, "iput orphan %llu\n", (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001382
1383 iter = oi->ip_next_orphan;
1384
1385 spin_lock(&oi->ip_lock);
Mark Fasheh34d024f2007-09-24 15:56:19 -07001386 /* The remote delete code may have set these on the
1387 * assumption that the other node would wipe them
1388 * successfully. If they are still in the node's
1389 * orphan dir, we need to reset that state. */
Mark Fashehccd979b2005-12-15 14:31:24 -08001390 oi->ip_flags &= ~(OCFS2_INODE_DELETED|OCFS2_INODE_SKIP_DELETE);
1391
1392 /* Set the proper information to get us going into
1393 * ocfs2_delete_inode. */
1394 oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED;
Mark Fashehccd979b2005-12-15 14:31:24 -08001395 spin_unlock(&oi->ip_lock);
1396
1397 iput(inode);
1398
1399 inode = iter;
1400 }
1401
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001402 return ret;
Mark Fashehccd979b2005-12-15 14:31:24 -08001403}
1404
1405static int ocfs2_wait_on_mount(struct ocfs2_super *osb)
1406{
1407 /* This check is good because ocfs2 will wait on our recovery
1408 * thread before changing it to something other than MOUNTED
1409 * or DISABLED. */
1410 wait_event(osb->osb_mount_event,
1411 atomic_read(&osb->vol_state) == VOLUME_MOUNTED ||
1412 atomic_read(&osb->vol_state) == VOLUME_DISABLED);
1413
1414 /* If there's an error on mount, then we may never get to the
1415 * MOUNTED flag, but this is set right before
1416 * dismount_volume() so we can trust it. */
1417 if (atomic_read(&osb->vol_state) == VOLUME_DISABLED) {
1418 mlog(0, "mount error, exiting!\n");
1419 return -EBUSY;
1420 }
1421
1422 return 0;
1423}
1424
1425static int ocfs2_commit_thread(void *arg)
1426{
1427 int status;
1428 struct ocfs2_super *osb = arg;
1429 struct ocfs2_journal *journal = osb->journal;
1430
1431 /* we can trust j_num_trans here because _should_stop() is only set in
1432 * shutdown and nobody other than ourselves should be able to start
1433 * transactions. committing on shutdown might take a few iterations
1434 * as final transactions put deleted inodes on the list */
1435 while (!(kthread_should_stop() &&
1436 atomic_read(&journal->j_num_trans) == 0)) {
1437
Mark Fasheh745ae8ba2006-02-09 13:23:39 -08001438 wait_event_interruptible(osb->checkpoint_event,
1439 atomic_read(&journal->j_num_trans)
1440 || kthread_should_stop());
Mark Fashehccd979b2005-12-15 14:31:24 -08001441
1442 status = ocfs2_commit_cache(osb);
1443 if (status < 0)
1444 mlog_errno(status);
1445
1446 if (kthread_should_stop() && atomic_read(&journal->j_num_trans)){
1447 mlog(ML_KTHREAD,
1448 "commit_thread: %u transactions pending on "
1449 "shutdown\n",
1450 atomic_read(&journal->j_num_trans));
1451 }
1452 }
1453
1454 return 0;
1455}
1456
1457/* Look for a dirty journal without taking any cluster locks. Used for
1458 * hard readonly access to determine whether the file system journals
1459 * require recovery. */
1460int ocfs2_check_journals_nolocks(struct ocfs2_super *osb)
1461{
1462 int ret = 0;
1463 unsigned int slot;
1464 struct buffer_head *di_bh;
1465 struct ocfs2_dinode *di;
1466 struct inode *journal = NULL;
1467
1468 for(slot = 0; slot < osb->max_slots; slot++) {
1469 journal = ocfs2_get_system_file_inode(osb,
1470 JOURNAL_SYSTEM_INODE,
1471 slot);
1472 if (!journal || is_bad_inode(journal)) {
1473 ret = -EACCES;
1474 mlog_errno(ret);
1475 goto out;
1476 }
1477
1478 di_bh = NULL;
1479 ret = ocfs2_read_block(osb, OCFS2_I(journal)->ip_blkno, &di_bh,
1480 0, journal);
1481 if (ret < 0) {
1482 mlog_errno(ret);
1483 goto out;
1484 }
1485
1486 di = (struct ocfs2_dinode *) di_bh->b_data;
1487
1488 if (le32_to_cpu(di->id1.journal1.ij_flags) &
1489 OCFS2_JOURNAL_DIRTY_FL)
1490 ret = -EROFS;
1491
1492 brelse(di_bh);
1493 if (ret)
1494 break;
1495 }
1496
1497out:
1498 if (journal)
1499 iput(journal);
1500
1501 return ret;
1502}