blob: 5a8a90d1c787af7bf993f36516615e079221c61f [file] [log] [blame]
Mark Fashehccd979b2005-12-15 14:31:24 -08001/* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
3 *
4 * journal.c
5 *
6 * Defines functions of journalling api
7 *
8 * Copyright (C) 2003, 2004 Oracle. All rights reserved.
9 *
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public
12 * License as published by the Free Software Foundation; either
13 * version 2 of the License, or (at your option) any later version.
14 *
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * General Public License for more details.
19 *
20 * You should have received a copy of the GNU General Public
21 * License along with this program; if not, write to the
22 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
23 * Boston, MA 021110-1307, USA.
24 */
25
26#include <linux/fs.h>
27#include <linux/types.h>
28#include <linux/slab.h>
29#include <linux/highmem.h>
30#include <linux/kthread.h>
31
32#define MLOG_MASK_PREFIX ML_JOURNAL
33#include <cluster/masklog.h>
34
35#include "ocfs2.h"
36
37#include "alloc.h"
38#include "dlmglue.h"
39#include "extent_map.h"
40#include "heartbeat.h"
41#include "inode.h"
42#include "journal.h"
43#include "localalloc.h"
44#include "namei.h"
45#include "slot_map.h"
46#include "super.h"
47#include "vote.h"
48#include "sysfile.h"
49
50#include "buffer_head_io.h"
51
Ingo Molnar34af9462006-06-27 02:53:55 -070052DEFINE_SPINLOCK(trans_inc_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -080053
54static int ocfs2_force_read_journal(struct inode *inode);
55static int ocfs2_recover_node(struct ocfs2_super *osb,
56 int node_num);
57static int __ocfs2_recovery_thread(void *arg);
58static int ocfs2_commit_cache(struct ocfs2_super *osb);
59static int ocfs2_wait_on_mount(struct ocfs2_super *osb);
Mark Fashehccd979b2005-12-15 14:31:24 -080060static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
61 int dirty);
62static int ocfs2_trylock_journal(struct ocfs2_super *osb,
63 int slot_num);
64static int ocfs2_recover_orphans(struct ocfs2_super *osb,
65 int slot);
66static int ocfs2_commit_thread(void *arg);
67
68static int ocfs2_commit_cache(struct ocfs2_super *osb)
69{
70 int status = 0;
71 unsigned int flushed;
72 unsigned long old_id;
73 struct ocfs2_journal *journal = NULL;
74
75 mlog_entry_void();
76
77 journal = osb->journal;
78
79 /* Flush all pending commits and checkpoint the journal. */
80 down_write(&journal->j_trans_barrier);
81
82 if (atomic_read(&journal->j_num_trans) == 0) {
83 up_write(&journal->j_trans_barrier);
84 mlog(0, "No transactions for me to flush!\n");
85 goto finally;
86 }
87
88 journal_lock_updates(journal->j_journal);
89 status = journal_flush(journal->j_journal);
90 journal_unlock_updates(journal->j_journal);
91 if (status < 0) {
92 up_write(&journal->j_trans_barrier);
93 mlog_errno(status);
94 goto finally;
95 }
96
97 old_id = ocfs2_inc_trans_id(journal);
98
99 flushed = atomic_read(&journal->j_num_trans);
100 atomic_set(&journal->j_num_trans, 0);
101 up_write(&journal->j_trans_barrier);
102
103 mlog(0, "commit_thread: flushed transaction %lu (%u handles)\n",
104 journal->j_trans_id, flushed);
105
106 ocfs2_kick_vote_thread(osb);
107 wake_up(&journal->j_checkpointed);
108finally:
109 mlog_exit(status);
110 return status;
111}
112
Mark Fashehccd979b2005-12-15 14:31:24 -0800113/* pass it NULL and it will allocate a new handle object for you. If
114 * you pass it a handle however, it may still return error, in which
115 * case it has free'd the passed handle for you. */
Mark Fasheh1fabe142006-10-09 18:11:45 -0700116handle_t *ocfs2_start_trans(struct ocfs2_super *osb, int max_buffs)
Mark Fashehccd979b2005-12-15 14:31:24 -0800117{
Mark Fashehccd979b2005-12-15 14:31:24 -0800118 journal_t *journal = osb->journal->j_journal;
Mark Fasheh1fabe142006-10-09 18:11:45 -0700119 handle_t *handle;
Mark Fashehccd979b2005-12-15 14:31:24 -0800120
Eric Sesterhenn / snakebyteebdec832006-01-27 10:32:52 +0100121 BUG_ON(!osb || !osb->journal->j_journal);
Mark Fashehccd979b2005-12-15 14:31:24 -0800122
Mark Fasheh65eff9c2006-10-09 17:26:22 -0700123 if (ocfs2_is_hard_readonly(osb))
124 return ERR_PTR(-EROFS);
Mark Fashehccd979b2005-12-15 14:31:24 -0800125
126 BUG_ON(osb->journal->j_state == OCFS2_JOURNAL_FREE);
127 BUG_ON(max_buffs <= 0);
128
129 /* JBD might support this, but our journalling code doesn't yet. */
130 if (journal_current_handle()) {
131 mlog(ML_ERROR, "Recursive transaction attempted!\n");
132 BUG();
133 }
134
Mark Fashehccd979b2005-12-15 14:31:24 -0800135 down_read(&osb->journal->j_trans_barrier);
136
Mark Fasheh1fabe142006-10-09 18:11:45 -0700137 handle = journal_start(journal, max_buffs);
138 if (IS_ERR(handle)) {
Mark Fashehccd979b2005-12-15 14:31:24 -0800139 up_read(&osb->journal->j_trans_barrier);
140
Mark Fasheh1fabe142006-10-09 18:11:45 -0700141 mlog_errno(PTR_ERR(handle));
Mark Fashehccd979b2005-12-15 14:31:24 -0800142
143 if (is_journal_aborted(journal)) {
144 ocfs2_abort(osb->sb, "Detected aborted journal");
Mark Fasheh1fabe142006-10-09 18:11:45 -0700145 handle = ERR_PTR(-EROFS);
Mark Fashehccd979b2005-12-15 14:31:24 -0800146 }
Sunil Mushranc271c5c2006-12-05 17:56:35 -0800147 } else {
148 if (!ocfs2_mount_local(osb))
149 atomic_inc(&(osb->journal->j_num_trans));
150 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800151
Mark Fashehccd979b2005-12-15 14:31:24 -0800152 return handle;
Mark Fashehccd979b2005-12-15 14:31:24 -0800153}
154
Mark Fasheh1fabe142006-10-09 18:11:45 -0700155int ocfs2_commit_trans(struct ocfs2_super *osb,
156 handle_t *handle)
Mark Fashehccd979b2005-12-15 14:31:24 -0800157{
Mark Fasheh1fabe142006-10-09 18:11:45 -0700158 int ret;
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700159 struct ocfs2_journal *journal = osb->journal;
Mark Fashehccd979b2005-12-15 14:31:24 -0800160
161 BUG_ON(!handle);
162
Mark Fasheh1fabe142006-10-09 18:11:45 -0700163 ret = journal_stop(handle);
164 if (ret < 0)
165 mlog_errno(ret);
Mark Fashehccd979b2005-12-15 14:31:24 -0800166
167 up_read(&journal->j_trans_barrier);
168
Mark Fasheh1fabe142006-10-09 18:11:45 -0700169 return ret;
Mark Fashehccd979b2005-12-15 14:31:24 -0800170}
171
172/*
173 * 'nblocks' is what you want to add to the current
174 * transaction. extend_trans will either extend the current handle by
175 * nblocks, or commit it and start a new one with nblocks credits.
176 *
177 * WARNING: This will not release any semaphores or disk locks taken
178 * during the transaction, so make sure they were taken *before*
179 * start_trans or we'll have ordering deadlocks.
180 *
181 * WARNING2: Note that we do *not* drop j_trans_barrier here. This is
182 * good because transaction ids haven't yet been recorded on the
183 * cluster locks associated with this handle.
184 */
Mark Fasheh1fc58142006-10-05 14:15:36 -0700185int ocfs2_extend_trans(handle_t *handle, int nblocks)
Mark Fashehccd979b2005-12-15 14:31:24 -0800186{
187 int status;
188
189 BUG_ON(!handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800190 BUG_ON(!nblocks);
191
192 mlog_entry_void();
193
194 mlog(0, "Trying to extend transaction by %d blocks\n", nblocks);
195
Mark Fasheh1fc58142006-10-05 14:15:36 -0700196 status = journal_extend(handle, nblocks);
Mark Fashehccd979b2005-12-15 14:31:24 -0800197 if (status < 0) {
198 mlog_errno(status);
199 goto bail;
200 }
201
202 if (status > 0) {
203 mlog(0, "journal_extend failed, trying journal_restart\n");
Mark Fasheh1fc58142006-10-05 14:15:36 -0700204 status = journal_restart(handle, nblocks);
Mark Fashehccd979b2005-12-15 14:31:24 -0800205 if (status < 0) {
Mark Fashehccd979b2005-12-15 14:31:24 -0800206 mlog_errno(status);
207 goto bail;
208 }
Mark Fasheh01ddf1e2006-10-05 13:54:39 -0700209 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800210
211 status = 0;
212bail:
213
214 mlog_exit(status);
215 return status;
216}
217
Mark Fasheh1fabe142006-10-09 18:11:45 -0700218int ocfs2_journal_access(handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -0800219 struct inode *inode,
220 struct buffer_head *bh,
221 int type)
222{
223 int status;
224
225 BUG_ON(!inode);
226 BUG_ON(!handle);
227 BUG_ON(!bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800228
Badari Pulavarty205f87f2006-03-26 01:38:00 -0800229 mlog_entry("bh->b_blocknr=%llu, type=%d (\"%s\"), bh->b_size = %zu\n",
Mark Fashehccd979b2005-12-15 14:31:24 -0800230 (unsigned long long)bh->b_blocknr, type,
231 (type == OCFS2_JOURNAL_ACCESS_CREATE) ?
232 "OCFS2_JOURNAL_ACCESS_CREATE" :
233 "OCFS2_JOURNAL_ACCESS_WRITE",
234 bh->b_size);
235
236 /* we can safely remove this assertion after testing. */
237 if (!buffer_uptodate(bh)) {
238 mlog(ML_ERROR, "giving me a buffer that's not uptodate!\n");
239 mlog(ML_ERROR, "b_blocknr=%llu\n",
240 (unsigned long long)bh->b_blocknr);
241 BUG();
242 }
243
244 /* Set the current transaction information on the inode so
245 * that the locking code knows whether it can drop it's locks
246 * on this inode or not. We're protected from the commit
247 * thread updating the current transaction id until
248 * ocfs2_commit_trans() because ocfs2_start_trans() took
249 * j_trans_barrier for us. */
250 ocfs2_set_inode_lock_trans(OCFS2_SB(inode->i_sb)->journal, inode);
251
Mark Fasheh251b6ec2006-01-10 15:41:43 -0800252 mutex_lock(&OCFS2_I(inode)->ip_io_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800253 switch (type) {
254 case OCFS2_JOURNAL_ACCESS_CREATE:
255 case OCFS2_JOURNAL_ACCESS_WRITE:
Mark Fasheh1fabe142006-10-09 18:11:45 -0700256 status = journal_get_write_access(handle, bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800257 break;
258
259 case OCFS2_JOURNAL_ACCESS_UNDO:
Mark Fasheh1fabe142006-10-09 18:11:45 -0700260 status = journal_get_undo_access(handle, bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800261 break;
262
263 default:
264 status = -EINVAL;
265 mlog(ML_ERROR, "Uknown access type!\n");
266 }
Mark Fasheh251b6ec2006-01-10 15:41:43 -0800267 mutex_unlock(&OCFS2_I(inode)->ip_io_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800268
269 if (status < 0)
270 mlog(ML_ERROR, "Error %d getting %d access to buffer!\n",
271 status, type);
272
273 mlog_exit(status);
274 return status;
275}
276
Mark Fasheh1fabe142006-10-09 18:11:45 -0700277int ocfs2_journal_dirty(handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -0800278 struct buffer_head *bh)
279{
280 int status;
281
Mark Fashehccd979b2005-12-15 14:31:24 -0800282 mlog_entry("(bh->b_blocknr=%llu)\n",
283 (unsigned long long)bh->b_blocknr);
284
Mark Fasheh1fabe142006-10-09 18:11:45 -0700285 status = journal_dirty_metadata(handle, bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800286 if (status < 0)
287 mlog(ML_ERROR, "Could not dirty metadata buffer. "
288 "(bh->b_blocknr=%llu)\n",
289 (unsigned long long)bh->b_blocknr);
290
291 mlog_exit(status);
292 return status;
293}
294
295int ocfs2_journal_dirty_data(handle_t *handle,
296 struct buffer_head *bh)
297{
298 int err = journal_dirty_data(handle, bh);
299 if (err)
300 mlog_errno(err);
301 /* TODO: When we can handle it, abort the handle and go RO on
302 * error here. */
303
304 return err;
305}
306
Mark Fashehccd979b2005-12-15 14:31:24 -0800307#define OCFS2_DEFAULT_COMMIT_INTERVAL (HZ * 5)
308
309void ocfs2_set_journal_params(struct ocfs2_super *osb)
310{
311 journal_t *journal = osb->journal->j_journal;
312
313 spin_lock(&journal->j_state_lock);
314 journal->j_commit_interval = OCFS2_DEFAULT_COMMIT_INTERVAL;
315 if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER)
316 journal->j_flags |= JFS_BARRIER;
317 else
318 journal->j_flags &= ~JFS_BARRIER;
319 spin_unlock(&journal->j_state_lock);
320}
321
322int ocfs2_journal_init(struct ocfs2_journal *journal, int *dirty)
323{
324 int status = -1;
325 struct inode *inode = NULL; /* the journal inode */
326 journal_t *j_journal = NULL;
327 struct ocfs2_dinode *di = NULL;
328 struct buffer_head *bh = NULL;
329 struct ocfs2_super *osb;
330 int meta_lock = 0;
331
332 mlog_entry_void();
333
334 BUG_ON(!journal);
335
336 osb = journal->j_osb;
337
338 /* already have the inode for our journal */
339 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
340 osb->slot_num);
341 if (inode == NULL) {
342 status = -EACCES;
343 mlog_errno(status);
344 goto done;
345 }
346 if (is_bad_inode(inode)) {
347 mlog(ML_ERROR, "access error (bad inode)\n");
348 iput(inode);
349 inode = NULL;
350 status = -EACCES;
351 goto done;
352 }
353
354 SET_INODE_JOURNAL(inode);
355 OCFS2_I(inode)->ip_open_count++;
356
Mark Fasheh6eff5792006-01-18 10:31:47 -0800357 /* Skip recovery waits here - journal inode metadata never
358 * changes in a live cluster so it can be considered an
359 * exception to the rule. */
Mark Fasheh4bcec182006-10-09 16:02:40 -0700360 status = ocfs2_meta_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY);
Mark Fashehccd979b2005-12-15 14:31:24 -0800361 if (status < 0) {
362 if (status != -ERESTARTSYS)
363 mlog(ML_ERROR, "Could not get lock on journal!\n");
364 goto done;
365 }
366
367 meta_lock = 1;
368 di = (struct ocfs2_dinode *)bh->b_data;
369
370 if (inode->i_size < OCFS2_MIN_JOURNAL_SIZE) {
371 mlog(ML_ERROR, "Journal file size (%lld) is too small!\n",
372 inode->i_size);
373 status = -EINVAL;
374 goto done;
375 }
376
377 mlog(0, "inode->i_size = %lld\n", inode->i_size);
Andrew Morton5515eff2006-03-26 01:37:53 -0800378 mlog(0, "inode->i_blocks = %llu\n",
379 (unsigned long long)inode->i_blocks);
Mark Fashehccd979b2005-12-15 14:31:24 -0800380 mlog(0, "inode->ip_clusters = %u\n", OCFS2_I(inode)->ip_clusters);
381
382 /* call the kernels journal init function now */
383 j_journal = journal_init_inode(inode);
384 if (j_journal == NULL) {
385 mlog(ML_ERROR, "Linux journal layer error\n");
386 status = -EINVAL;
387 goto done;
388 }
389
390 mlog(0, "Returned from journal_init_inode\n");
391 mlog(0, "j_journal->j_maxlen = %u\n", j_journal->j_maxlen);
392
393 *dirty = (le32_to_cpu(di->id1.journal1.ij_flags) &
394 OCFS2_JOURNAL_DIRTY_FL);
395
396 journal->j_journal = j_journal;
397 journal->j_inode = inode;
398 journal->j_bh = bh;
399
400 ocfs2_set_journal_params(osb);
401
402 journal->j_state = OCFS2_JOURNAL_LOADED;
403
404 status = 0;
405done:
406 if (status < 0) {
407 if (meta_lock)
408 ocfs2_meta_unlock(inode, 1);
409 if (bh != NULL)
410 brelse(bh);
411 if (inode) {
412 OCFS2_I(inode)->ip_open_count--;
413 iput(inode);
414 }
415 }
416
417 mlog_exit(status);
418 return status;
419}
420
421static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
422 int dirty)
423{
424 int status;
425 unsigned int flags;
426 struct ocfs2_journal *journal = osb->journal;
427 struct buffer_head *bh = journal->j_bh;
428 struct ocfs2_dinode *fe;
429
430 mlog_entry_void();
431
432 fe = (struct ocfs2_dinode *)bh->b_data;
433 if (!OCFS2_IS_VALID_DINODE(fe)) {
434 /* This is called from startup/shutdown which will
435 * handle the errors in a specific manner, so no need
436 * to call ocfs2_error() here. */
Mark Fashehb0697052006-03-03 10:24:33 -0800437 mlog(ML_ERROR, "Journal dinode %llu has invalid "
438 "signature: %.*s", (unsigned long long)fe->i_blkno, 7,
439 fe->i_signature);
Mark Fashehccd979b2005-12-15 14:31:24 -0800440 status = -EIO;
441 goto out;
442 }
443
444 flags = le32_to_cpu(fe->id1.journal1.ij_flags);
445 if (dirty)
446 flags |= OCFS2_JOURNAL_DIRTY_FL;
447 else
448 flags &= ~OCFS2_JOURNAL_DIRTY_FL;
449 fe->id1.journal1.ij_flags = cpu_to_le32(flags);
450
451 status = ocfs2_write_block(osb, bh, journal->j_inode);
452 if (status < 0)
453 mlog_errno(status);
454
455out:
456 mlog_exit(status);
457 return status;
458}
459
460/*
461 * If the journal has been kmalloc'd it needs to be freed after this
462 * call.
463 */
464void ocfs2_journal_shutdown(struct ocfs2_super *osb)
465{
466 struct ocfs2_journal *journal = NULL;
467 int status = 0;
468 struct inode *inode = NULL;
469 int num_running_trans = 0;
470
471 mlog_entry_void();
472
Eric Sesterhenn / snakebyteebdec832006-01-27 10:32:52 +0100473 BUG_ON(!osb);
Mark Fashehccd979b2005-12-15 14:31:24 -0800474
475 journal = osb->journal;
476 if (!journal)
477 goto done;
478
479 inode = journal->j_inode;
480
481 if (journal->j_state != OCFS2_JOURNAL_LOADED)
482 goto done;
483
484 /* need to inc inode use count as journal_destroy will iput. */
485 if (!igrab(inode))
486 BUG();
487
488 num_running_trans = atomic_read(&(osb->journal->j_num_trans));
489 if (num_running_trans > 0)
490 mlog(0, "Shutting down journal: must wait on %d "
491 "running transactions!\n",
492 num_running_trans);
493
494 /* Do a commit_cache here. It will flush our journal, *and*
495 * release any locks that are still held.
496 * set the SHUTDOWN flag and release the trans lock.
497 * the commit thread will take the trans lock for us below. */
498 journal->j_state = OCFS2_JOURNAL_IN_SHUTDOWN;
499
500 /* The OCFS2_JOURNAL_IN_SHUTDOWN will signal to commit_cache to not
501 * drop the trans_lock (which we want to hold until we
502 * completely destroy the journal. */
503 if (osb->commit_task) {
504 /* Wait for the commit thread */
505 mlog(0, "Waiting for ocfs2commit to exit....\n");
506 kthread_stop(osb->commit_task);
507 osb->commit_task = NULL;
508 }
509
510 BUG_ON(atomic_read(&(osb->journal->j_num_trans)) != 0);
511
Sunil Mushranc271c5c2006-12-05 17:56:35 -0800512 if (ocfs2_mount_local(osb)) {
513 journal_lock_updates(journal->j_journal);
514 status = journal_flush(journal->j_journal);
515 journal_unlock_updates(journal->j_journal);
516 if (status < 0)
517 mlog_errno(status);
518 }
519
520 if (status == 0) {
521 /*
522 * Do not toggle if flush was unsuccessful otherwise
523 * will leave dirty metadata in a "clean" journal
524 */
525 status = ocfs2_journal_toggle_dirty(osb, 0);
526 if (status < 0)
527 mlog_errno(status);
528 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800529
530 /* Shutdown the kernel journal system */
531 journal_destroy(journal->j_journal);
532
533 OCFS2_I(inode)->ip_open_count--;
534
535 /* unlock our journal */
536 ocfs2_meta_unlock(inode, 1);
537
538 brelse(journal->j_bh);
539 journal->j_bh = NULL;
540
541 journal->j_state = OCFS2_JOURNAL_FREE;
542
543// up_write(&journal->j_trans_barrier);
544done:
545 if (inode)
546 iput(inode);
547 mlog_exit_void();
548}
549
550static void ocfs2_clear_journal_error(struct super_block *sb,
551 journal_t *journal,
552 int slot)
553{
554 int olderr;
555
556 olderr = journal_errno(journal);
557 if (olderr) {
558 mlog(ML_ERROR, "File system error %d recorded in "
559 "journal %u.\n", olderr, slot);
560 mlog(ML_ERROR, "File system on device %s needs checking.\n",
561 sb->s_id);
562
563 journal_ack_err(journal);
564 journal_clear_err(journal);
565 }
566}
567
Sunil Mushranc271c5c2006-12-05 17:56:35 -0800568int ocfs2_journal_load(struct ocfs2_journal *journal, int local)
Mark Fashehccd979b2005-12-15 14:31:24 -0800569{
570 int status = 0;
571 struct ocfs2_super *osb;
572
573 mlog_entry_void();
574
575 if (!journal)
576 BUG();
577
578 osb = journal->j_osb;
579
580 status = journal_load(journal->j_journal);
581 if (status < 0) {
582 mlog(ML_ERROR, "Failed to load journal!\n");
583 goto done;
584 }
585
586 ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num);
587
588 status = ocfs2_journal_toggle_dirty(osb, 1);
589 if (status < 0) {
590 mlog_errno(status);
591 goto done;
592 }
593
594 /* Launch the commit thread */
Sunil Mushranc271c5c2006-12-05 17:56:35 -0800595 if (!local) {
596 osb->commit_task = kthread_run(ocfs2_commit_thread, osb,
597 "ocfs2cmt");
598 if (IS_ERR(osb->commit_task)) {
599 status = PTR_ERR(osb->commit_task);
600 osb->commit_task = NULL;
601 mlog(ML_ERROR, "unable to launch ocfs2commit thread, "
602 "error=%d", status);
603 goto done;
604 }
605 } else
Mark Fashehccd979b2005-12-15 14:31:24 -0800606 osb->commit_task = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800607
608done:
609 mlog_exit(status);
610 return status;
611}
612
613
614/* 'full' flag tells us whether we clear out all blocks or if we just
615 * mark the journal clean */
616int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full)
617{
618 int status;
619
620 mlog_entry_void();
621
Eric Sesterhenn / snakebyteebdec832006-01-27 10:32:52 +0100622 BUG_ON(!journal);
Mark Fashehccd979b2005-12-15 14:31:24 -0800623
624 status = journal_wipe(journal->j_journal, full);
625 if (status < 0) {
626 mlog_errno(status);
627 goto bail;
628 }
629
630 status = ocfs2_journal_toggle_dirty(journal->j_osb, 0);
631 if (status < 0)
632 mlog_errno(status);
633
634bail:
635 mlog_exit(status);
636 return status;
637}
638
639/*
640 * JBD Might read a cached version of another nodes journal file. We
641 * don't want this as this file changes often and we get no
642 * notification on those changes. The only way to be sure that we've
643 * got the most up to date version of those blocks then is to force
644 * read them off disk. Just searching through the buffer cache won't
645 * work as there may be pages backing this file which are still marked
646 * up to date. We know things can't change on this file underneath us
647 * as we have the lock by now :)
648 */
649static int ocfs2_force_read_journal(struct inode *inode)
650{
651 int status = 0;
Mark Fasheh4f902c32007-03-09 16:26:50 -0800652 int i;
Mark Fasheh8110b072007-03-22 16:53:23 -0700653 u64 v_blkno, p_blkno, p_blocks, num_blocks;
Mark Fasheh4f902c32007-03-09 16:26:50 -0800654#define CONCURRENT_JOURNAL_FILL 32ULL
Mark Fashehccd979b2005-12-15 14:31:24 -0800655 struct buffer_head *bhs[CONCURRENT_JOURNAL_FILL];
656
657 mlog_entry_void();
658
Mark Fashehccd979b2005-12-15 14:31:24 -0800659 memset(bhs, 0, sizeof(struct buffer_head *) * CONCURRENT_JOURNAL_FILL);
660
Mark Fasheh8110b072007-03-22 16:53:23 -0700661 num_blocks = ocfs2_blocks_for_bytes(inode->i_sb, inode->i_size);
Mark Fashehccd979b2005-12-15 14:31:24 -0800662 v_blkno = 0;
Mark Fasheh8110b072007-03-22 16:53:23 -0700663 while (v_blkno < num_blocks) {
Mark Fashehccd979b2005-12-15 14:31:24 -0800664 status = ocfs2_extent_map_get_blocks(inode, v_blkno,
Mark Fasheh49cb8d22007-03-09 16:21:46 -0800665 &p_blkno, &p_blocks, NULL);
Mark Fashehccd979b2005-12-15 14:31:24 -0800666 if (status < 0) {
667 mlog_errno(status);
668 goto bail;
669 }
670
671 if (p_blocks > CONCURRENT_JOURNAL_FILL)
672 p_blocks = CONCURRENT_JOURNAL_FILL;
673
Mark Fashehdd4a2c22006-04-12 14:24:05 -0700674 /* We are reading journal data which should not
675 * be put in the uptodate cache */
Mark Fashehccd979b2005-12-15 14:31:24 -0800676 status = ocfs2_read_blocks(OCFS2_SB(inode->i_sb),
677 p_blkno, p_blocks, bhs, 0,
Mark Fashehdd4a2c22006-04-12 14:24:05 -0700678 NULL);
Mark Fashehccd979b2005-12-15 14:31:24 -0800679 if (status < 0) {
680 mlog_errno(status);
681 goto bail;
682 }
683
684 for(i = 0; i < p_blocks; i++) {
685 brelse(bhs[i]);
686 bhs[i] = NULL;
687 }
688
689 v_blkno += p_blocks;
690 }
691
692bail:
693 for(i = 0; i < CONCURRENT_JOURNAL_FILL; i++)
694 if (bhs[i])
695 brelse(bhs[i]);
696 mlog_exit(status);
697 return status;
698}
699
700struct ocfs2_la_recovery_item {
701 struct list_head lri_list;
702 int lri_slot;
703 struct ocfs2_dinode *lri_la_dinode;
704 struct ocfs2_dinode *lri_tl_dinode;
705};
706
707/* Does the second half of the recovery process. By this point, the
708 * node is marked clean and can actually be considered recovered,
709 * hence it's no longer in the recovery map, but there's still some
710 * cleanup we can do which shouldn't happen within the recovery thread
711 * as locking in that context becomes very difficult if we are to take
712 * recovering nodes into account.
713 *
714 * NOTE: This function can and will sleep on recovery of other nodes
715 * during cluster locking, just like any other ocfs2 process.
716 */
David Howellsc4028952006-11-22 14:57:56 +0000717void ocfs2_complete_recovery(struct work_struct *work)
Mark Fashehccd979b2005-12-15 14:31:24 -0800718{
719 int ret;
David Howellsc4028952006-11-22 14:57:56 +0000720 struct ocfs2_journal *journal =
721 container_of(work, struct ocfs2_journal, j_recovery_work);
722 struct ocfs2_super *osb = journal->j_osb;
Mark Fashehccd979b2005-12-15 14:31:24 -0800723 struct ocfs2_dinode *la_dinode, *tl_dinode;
724 struct ocfs2_la_recovery_item *item;
725 struct list_head *p, *n;
726 LIST_HEAD(tmp_la_list);
727
728 mlog_entry_void();
729
730 mlog(0, "completing recovery from keventd\n");
731
732 spin_lock(&journal->j_lock);
733 list_splice_init(&journal->j_la_cleanups, &tmp_la_list);
734 spin_unlock(&journal->j_lock);
735
736 list_for_each_safe(p, n, &tmp_la_list) {
737 item = list_entry(p, struct ocfs2_la_recovery_item, lri_list);
738 list_del_init(&item->lri_list);
739
740 mlog(0, "Complete recovery for slot %d\n", item->lri_slot);
741
742 la_dinode = item->lri_la_dinode;
743 if (la_dinode) {
Mark Fashehb0697052006-03-03 10:24:33 -0800744 mlog(0, "Clean up local alloc %llu\n",
745 (unsigned long long)la_dinode->i_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800746
747 ret = ocfs2_complete_local_alloc_recovery(osb,
748 la_dinode);
749 if (ret < 0)
750 mlog_errno(ret);
751
752 kfree(la_dinode);
753 }
754
755 tl_dinode = item->lri_tl_dinode;
756 if (tl_dinode) {
Mark Fashehb0697052006-03-03 10:24:33 -0800757 mlog(0, "Clean up truncate log %llu\n",
758 (unsigned long long)tl_dinode->i_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800759
760 ret = ocfs2_complete_truncate_log_recovery(osb,
761 tl_dinode);
762 if (ret < 0)
763 mlog_errno(ret);
764
765 kfree(tl_dinode);
766 }
767
768 ret = ocfs2_recover_orphans(osb, item->lri_slot);
769 if (ret < 0)
770 mlog_errno(ret);
771
772 kfree(item);
773 }
774
775 mlog(0, "Recovery completion\n");
776 mlog_exit_void();
777}
778
779/* NOTE: This function always eats your references to la_dinode and
780 * tl_dinode, either manually on error, or by passing them to
781 * ocfs2_complete_recovery */
782static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal,
783 int slot_num,
784 struct ocfs2_dinode *la_dinode,
785 struct ocfs2_dinode *tl_dinode)
786{
787 struct ocfs2_la_recovery_item *item;
788
Sunil Mushranafae00ab2006-04-12 14:37:00 -0700789 item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS);
Mark Fashehccd979b2005-12-15 14:31:24 -0800790 if (!item) {
791 /* Though we wish to avoid it, we are in fact safe in
792 * skipping local alloc cleanup as fsck.ocfs2 is more
793 * than capable of reclaiming unused space. */
794 if (la_dinode)
795 kfree(la_dinode);
796
797 if (tl_dinode)
798 kfree(tl_dinode);
799
800 mlog_errno(-ENOMEM);
801 return;
802 }
803
804 INIT_LIST_HEAD(&item->lri_list);
805 item->lri_la_dinode = la_dinode;
806 item->lri_slot = slot_num;
807 item->lri_tl_dinode = tl_dinode;
808
809 spin_lock(&journal->j_lock);
810 list_add_tail(&item->lri_list, &journal->j_la_cleanups);
811 queue_work(ocfs2_wq, &journal->j_recovery_work);
812 spin_unlock(&journal->j_lock);
813}
814
815/* Called by the mount code to queue recovery the last part of
816 * recovery for it's own slot. */
817void ocfs2_complete_mount_recovery(struct ocfs2_super *osb)
818{
819 struct ocfs2_journal *journal = osb->journal;
820
821 if (osb->dirty) {
822 /* No need to queue up our truncate_log as regular
823 * cleanup will catch that. */
824 ocfs2_queue_recovery_completion(journal,
825 osb->slot_num,
826 osb->local_alloc_copy,
827 NULL);
828 ocfs2_schedule_truncate_log_flush(osb, 0);
829
830 osb->local_alloc_copy = NULL;
831 osb->dirty = 0;
832 }
833}
834
835static int __ocfs2_recovery_thread(void *arg)
836{
837 int status, node_num;
838 struct ocfs2_super *osb = arg;
839
840 mlog_entry_void();
841
842 status = ocfs2_wait_on_mount(osb);
843 if (status < 0) {
844 goto bail;
845 }
846
847restart:
848 status = ocfs2_super_lock(osb, 1);
849 if (status < 0) {
850 mlog_errno(status);
851 goto bail;
852 }
853
854 while(!ocfs2_node_map_is_empty(osb, &osb->recovery_map)) {
855 node_num = ocfs2_node_map_first_set_bit(osb,
856 &osb->recovery_map);
857 if (node_num == O2NM_INVALID_NODE_NUM) {
858 mlog(0, "Out of nodes to recover.\n");
859 break;
860 }
861
862 status = ocfs2_recover_node(osb, node_num);
863 if (status < 0) {
864 mlog(ML_ERROR,
865 "Error %d recovering node %d on device (%u,%u)!\n",
866 status, node_num,
867 MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
868 mlog(ML_ERROR, "Volume requires unmount.\n");
869 continue;
870 }
871
872 ocfs2_recovery_map_clear(osb, node_num);
873 }
874 ocfs2_super_unlock(osb, 1);
875
876 /* We always run recovery on our own orphan dir - the dead
877 * node(s) may have voted "no" on an inode delete earlier. A
878 * revote is therefore required. */
879 ocfs2_queue_recovery_completion(osb->journal, osb->slot_num, NULL,
880 NULL);
881
882bail:
Arjan van de Venc74ec2f2006-01-13 21:54:23 -0800883 mutex_lock(&osb->recovery_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -0800884 if (!status &&
885 !ocfs2_node_map_is_empty(osb, &osb->recovery_map)) {
Arjan van de Venc74ec2f2006-01-13 21:54:23 -0800886 mutex_unlock(&osb->recovery_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -0800887 goto restart;
888 }
889
890 osb->recovery_thread_task = NULL;
891 mb(); /* sync with ocfs2_recovery_thread_running */
892 wake_up(&osb->recovery_event);
893
Arjan van de Venc74ec2f2006-01-13 21:54:23 -0800894 mutex_unlock(&osb->recovery_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -0800895
896 mlog_exit(status);
897 /* no one is callint kthread_stop() for us so the kthread() api
898 * requires that we call do_exit(). And it isn't exported, but
899 * complete_and_exit() seems to be a minimal wrapper around it. */
900 complete_and_exit(NULL, status);
901 return status;
902}
903
904void ocfs2_recovery_thread(struct ocfs2_super *osb, int node_num)
905{
906 mlog_entry("(node_num=%d, osb->node_num = %d)\n",
907 node_num, osb->node_num);
908
Arjan van de Venc74ec2f2006-01-13 21:54:23 -0800909 mutex_lock(&osb->recovery_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -0800910 if (osb->disable_recovery)
911 goto out;
912
913 /* People waiting on recovery will wait on
914 * the recovery map to empty. */
915 if (!ocfs2_recovery_map_set(osb, node_num))
916 mlog(0, "node %d already be in recovery.\n", node_num);
917
918 mlog(0, "starting recovery thread...\n");
919
920 if (osb->recovery_thread_task)
921 goto out;
922
923 osb->recovery_thread_task = kthread_run(__ocfs2_recovery_thread, osb,
Mark Fasheh78427042006-05-04 12:03:26 -0700924 "ocfs2rec");
Mark Fashehccd979b2005-12-15 14:31:24 -0800925 if (IS_ERR(osb->recovery_thread_task)) {
926 mlog_errno((int)PTR_ERR(osb->recovery_thread_task));
927 osb->recovery_thread_task = NULL;
928 }
929
930out:
Arjan van de Venc74ec2f2006-01-13 21:54:23 -0800931 mutex_unlock(&osb->recovery_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -0800932 wake_up(&osb->recovery_event);
933
934 mlog_exit_void();
935}
936
937/* Does the actual journal replay and marks the journal inode as
938 * clean. Will only replay if the journal inode is marked dirty. */
939static int ocfs2_replay_journal(struct ocfs2_super *osb,
940 int node_num,
941 int slot_num)
942{
943 int status;
944 int got_lock = 0;
945 unsigned int flags;
946 struct inode *inode = NULL;
947 struct ocfs2_dinode *fe;
948 journal_t *journal = NULL;
949 struct buffer_head *bh = NULL;
950
951 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
952 slot_num);
953 if (inode == NULL) {
954 status = -EACCES;
955 mlog_errno(status);
956 goto done;
957 }
958 if (is_bad_inode(inode)) {
959 status = -EACCES;
960 iput(inode);
961 inode = NULL;
962 mlog_errno(status);
963 goto done;
964 }
965 SET_INODE_JOURNAL(inode);
966
Mark Fasheh4bcec182006-10-09 16:02:40 -0700967 status = ocfs2_meta_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY);
Mark Fashehccd979b2005-12-15 14:31:24 -0800968 if (status < 0) {
969 mlog(0, "status returned from ocfs2_meta_lock=%d\n", status);
970 if (status != -ERESTARTSYS)
971 mlog(ML_ERROR, "Could not lock journal!\n");
972 goto done;
973 }
974 got_lock = 1;
975
976 fe = (struct ocfs2_dinode *) bh->b_data;
977
978 flags = le32_to_cpu(fe->id1.journal1.ij_flags);
979
980 if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) {
981 mlog(0, "No recovery required for node %d\n", node_num);
982 goto done;
983 }
984
985 mlog(ML_NOTICE, "Recovering node %d from slot %d on device (%u,%u)\n",
986 node_num, slot_num,
987 MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
988
989 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
990
991 status = ocfs2_force_read_journal(inode);
992 if (status < 0) {
993 mlog_errno(status);
994 goto done;
995 }
996
997 mlog(0, "calling journal_init_inode\n");
998 journal = journal_init_inode(inode);
999 if (journal == NULL) {
1000 mlog(ML_ERROR, "Linux journal layer error\n");
1001 status = -EIO;
1002 goto done;
1003 }
1004
1005 status = journal_load(journal);
1006 if (status < 0) {
1007 mlog_errno(status);
1008 if (!igrab(inode))
1009 BUG();
1010 journal_destroy(journal);
1011 goto done;
1012 }
1013
1014 ocfs2_clear_journal_error(osb->sb, journal, slot_num);
1015
1016 /* wipe the journal */
1017 mlog(0, "flushing the journal.\n");
1018 journal_lock_updates(journal);
1019 status = journal_flush(journal);
1020 journal_unlock_updates(journal);
1021 if (status < 0)
1022 mlog_errno(status);
1023
1024 /* This will mark the node clean */
1025 flags = le32_to_cpu(fe->id1.journal1.ij_flags);
1026 flags &= ~OCFS2_JOURNAL_DIRTY_FL;
1027 fe->id1.journal1.ij_flags = cpu_to_le32(flags);
1028
1029 status = ocfs2_write_block(osb, bh, inode);
1030 if (status < 0)
1031 mlog_errno(status);
1032
1033 if (!igrab(inode))
1034 BUG();
1035
1036 journal_destroy(journal);
1037
1038done:
1039 /* drop the lock on this nodes journal */
1040 if (got_lock)
1041 ocfs2_meta_unlock(inode, 1);
1042
1043 if (inode)
1044 iput(inode);
1045
1046 if (bh)
1047 brelse(bh);
1048
1049 mlog_exit(status);
1050 return status;
1051}
1052
1053/*
1054 * Do the most important parts of node recovery:
1055 * - Replay it's journal
1056 * - Stamp a clean local allocator file
1057 * - Stamp a clean truncate log
1058 * - Mark the node clean
1059 *
1060 * If this function completes without error, a node in OCFS2 can be
1061 * said to have been safely recovered. As a result, failure during the
1062 * second part of a nodes recovery process (local alloc recovery) is
1063 * far less concerning.
1064 */
1065static int ocfs2_recover_node(struct ocfs2_super *osb,
1066 int node_num)
1067{
1068 int status = 0;
1069 int slot_num;
1070 struct ocfs2_slot_info *si = osb->slot_info;
1071 struct ocfs2_dinode *la_copy = NULL;
1072 struct ocfs2_dinode *tl_copy = NULL;
1073
1074 mlog_entry("(node_num=%d, osb->node_num = %d)\n",
1075 node_num, osb->node_num);
1076
1077 mlog(0, "checking node %d\n", node_num);
1078
1079 /* Should not ever be called to recover ourselves -- in that
1080 * case we should've called ocfs2_journal_load instead. */
Eric Sesterhenn / snakebyteebdec832006-01-27 10:32:52 +01001081 BUG_ON(osb->node_num == node_num);
Mark Fashehccd979b2005-12-15 14:31:24 -08001082
1083 slot_num = ocfs2_node_num_to_slot(si, node_num);
1084 if (slot_num == OCFS2_INVALID_SLOT) {
1085 status = 0;
1086 mlog(0, "no slot for this node, so no recovery required.\n");
1087 goto done;
1088 }
1089
1090 mlog(0, "node %d was using slot %d\n", node_num, slot_num);
1091
1092 status = ocfs2_replay_journal(osb, node_num, slot_num);
1093 if (status < 0) {
1094 mlog_errno(status);
1095 goto done;
1096 }
1097
1098 /* Stamp a clean local alloc file AFTER recovering the journal... */
1099 status = ocfs2_begin_local_alloc_recovery(osb, slot_num, &la_copy);
1100 if (status < 0) {
1101 mlog_errno(status);
1102 goto done;
1103 }
1104
1105 /* An error from begin_truncate_log_recovery is not
1106 * serious enough to warrant halting the rest of
1107 * recovery. */
1108 status = ocfs2_begin_truncate_log_recovery(osb, slot_num, &tl_copy);
1109 if (status < 0)
1110 mlog_errno(status);
1111
1112 /* Likewise, this would be a strange but ultimately not so
1113 * harmful place to get an error... */
1114 ocfs2_clear_slot(si, slot_num);
1115 status = ocfs2_update_disk_slots(osb, si);
1116 if (status < 0)
1117 mlog_errno(status);
1118
1119 /* This will kfree the memory pointed to by la_copy and tl_copy */
1120 ocfs2_queue_recovery_completion(osb->journal, slot_num, la_copy,
1121 tl_copy);
1122
1123 status = 0;
1124done:
1125
1126 mlog_exit(status);
1127 return status;
1128}
1129
1130/* Test node liveness by trylocking his journal. If we get the lock,
1131 * we drop it here. Return 0 if we got the lock, -EAGAIN if node is
1132 * still alive (we couldn't get the lock) and < 0 on error. */
1133static int ocfs2_trylock_journal(struct ocfs2_super *osb,
1134 int slot_num)
1135{
1136 int status, flags;
1137 struct inode *inode = NULL;
1138
1139 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
1140 slot_num);
1141 if (inode == NULL) {
1142 mlog(ML_ERROR, "access error\n");
1143 status = -EACCES;
1144 goto bail;
1145 }
1146 if (is_bad_inode(inode)) {
1147 mlog(ML_ERROR, "access error (bad inode)\n");
1148 iput(inode);
1149 inode = NULL;
1150 status = -EACCES;
1151 goto bail;
1152 }
1153 SET_INODE_JOURNAL(inode);
1154
1155 flags = OCFS2_META_LOCK_RECOVERY | OCFS2_META_LOCK_NOQUEUE;
Mark Fasheh4bcec182006-10-09 16:02:40 -07001156 status = ocfs2_meta_lock_full(inode, NULL, 1, flags);
Mark Fashehccd979b2005-12-15 14:31:24 -08001157 if (status < 0) {
1158 if (status != -EAGAIN)
1159 mlog_errno(status);
1160 goto bail;
1161 }
1162
1163 ocfs2_meta_unlock(inode, 1);
1164bail:
1165 if (inode)
1166 iput(inode);
1167
1168 return status;
1169}
1170
1171/* Call this underneath ocfs2_super_lock. It also assumes that the
1172 * slot info struct has been updated from disk. */
1173int ocfs2_mark_dead_nodes(struct ocfs2_super *osb)
1174{
1175 int status, i, node_num;
1176 struct ocfs2_slot_info *si = osb->slot_info;
1177
1178 /* This is called with the super block cluster lock, so we
1179 * know that the slot map can't change underneath us. */
1180
1181 spin_lock(&si->si_lock);
1182 for(i = 0; i < si->si_num_slots; i++) {
1183 if (i == osb->slot_num)
1184 continue;
1185 if (ocfs2_is_empty_slot(si, i))
1186 continue;
1187
1188 node_num = si->si_global_node_nums[i];
1189 if (ocfs2_node_map_test_bit(osb, &osb->recovery_map, node_num))
1190 continue;
1191 spin_unlock(&si->si_lock);
1192
1193 /* Ok, we have a slot occupied by another node which
1194 * is not in the recovery map. We trylock his journal
1195 * file here to test if he's alive. */
1196 status = ocfs2_trylock_journal(osb, i);
1197 if (!status) {
1198 /* Since we're called from mount, we know that
1199 * the recovery thread can't race us on
1200 * setting / checking the recovery bits. */
1201 ocfs2_recovery_thread(osb, node_num);
1202 } else if ((status < 0) && (status != -EAGAIN)) {
1203 mlog_errno(status);
1204 goto bail;
1205 }
1206
1207 spin_lock(&si->si_lock);
1208 }
1209 spin_unlock(&si->si_lock);
1210
1211 status = 0;
1212bail:
1213 mlog_exit(status);
1214 return status;
1215}
1216
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001217static int ocfs2_queue_orphans(struct ocfs2_super *osb,
1218 int slot,
1219 struct inode **head)
Mark Fashehccd979b2005-12-15 14:31:24 -08001220{
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001221 int status;
Mark Fashehccd979b2005-12-15 14:31:24 -08001222 struct inode *orphan_dir_inode = NULL;
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001223 struct inode *iter;
Mark Fashehccd979b2005-12-15 14:31:24 -08001224 unsigned long offset, blk, local;
1225 struct buffer_head *bh = NULL;
1226 struct ocfs2_dir_entry *de;
1227 struct super_block *sb = osb->sb;
Mark Fashehccd979b2005-12-15 14:31:24 -08001228
1229 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
1230 ORPHAN_DIR_SYSTEM_INODE,
1231 slot);
1232 if (!orphan_dir_inode) {
1233 status = -ENOENT;
1234 mlog_errno(status);
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001235 return status;
1236 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001237
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001238 mutex_lock(&orphan_dir_inode->i_mutex);
Mark Fasheh4bcec182006-10-09 16:02:40 -07001239 status = ocfs2_meta_lock(orphan_dir_inode, NULL, 0);
Mark Fashehccd979b2005-12-15 14:31:24 -08001240 if (status < 0) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001241 mlog_errno(status);
1242 goto out;
1243 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001244
1245 offset = 0;
1246 iter = NULL;
1247 while(offset < i_size_read(orphan_dir_inode)) {
1248 blk = offset >> sb->s_blocksize_bits;
1249
1250 bh = ocfs2_bread(orphan_dir_inode, blk, &status, 0);
1251 if (!bh)
1252 status = -EINVAL;
1253 if (status < 0) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001254 if (bh)
1255 brelse(bh);
1256 mlog_errno(status);
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001257 goto out_unlock;
Mark Fashehccd979b2005-12-15 14:31:24 -08001258 }
1259
1260 local = 0;
1261 while(offset < i_size_read(orphan_dir_inode)
1262 && local < sb->s_blocksize) {
1263 de = (struct ocfs2_dir_entry *) (bh->b_data + local);
1264
1265 if (!ocfs2_check_dir_entry(orphan_dir_inode,
1266 de, bh, local)) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001267 status = -EINVAL;
1268 mlog_errno(status);
1269 brelse(bh);
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001270 goto out_unlock;
Mark Fashehccd979b2005-12-15 14:31:24 -08001271 }
1272
1273 local += le16_to_cpu(de->rec_len);
1274 offset += le16_to_cpu(de->rec_len);
1275
1276 /* I guess we silently fail on no inode? */
1277 if (!le64_to_cpu(de->inode))
1278 continue;
1279 if (de->file_type > OCFS2_FT_MAX) {
1280 mlog(ML_ERROR,
1281 "block %llu contains invalid de: "
Mark Fashehb0697052006-03-03 10:24:33 -08001282 "inode = %llu, rec_len = %u, "
Mark Fashehccd979b2005-12-15 14:31:24 -08001283 "name_len = %u, file_type = %u, "
1284 "name='%.*s'\n",
1285 (unsigned long long)bh->b_blocknr,
Mark Fashehb0697052006-03-03 10:24:33 -08001286 (unsigned long long)le64_to_cpu(de->inode),
Mark Fashehccd979b2005-12-15 14:31:24 -08001287 le16_to_cpu(de->rec_len),
1288 de->name_len,
1289 de->file_type,
1290 de->name_len,
1291 de->name);
1292 continue;
1293 }
1294 if (de->name_len == 1 && !strncmp(".", de->name, 1))
1295 continue;
1296 if (de->name_len == 2 && !strncmp("..", de->name, 2))
1297 continue;
1298
Mark Fasheh24c19ef2006-09-22 17:28:19 -07001299 iter = ocfs2_iget(osb, le64_to_cpu(de->inode),
Tiger Yang50008632007-03-20 16:01:38 -07001300 OCFS2_FI_FLAG_ORPHAN_RECOVERY);
Mark Fashehccd979b2005-12-15 14:31:24 -08001301 if (IS_ERR(iter))
1302 continue;
1303
Mark Fashehb0697052006-03-03 10:24:33 -08001304 mlog(0, "queue orphan %llu\n",
1305 (unsigned long long)OCFS2_I(iter)->ip_blkno);
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001306 /* No locking is required for the next_orphan
1307 * queue as there is only ever a single
1308 * process doing orphan recovery. */
1309 OCFS2_I(iter)->ip_next_orphan = *head;
1310 *head = iter;
Mark Fashehccd979b2005-12-15 14:31:24 -08001311 }
1312 brelse(bh);
1313 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001314
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001315out_unlock:
Mark Fashehccd979b2005-12-15 14:31:24 -08001316 ocfs2_meta_unlock(orphan_dir_inode, 0);
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001317out:
1318 mutex_unlock(&orphan_dir_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -08001319 iput(orphan_dir_inode);
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001320 return status;
1321}
1322
1323static int ocfs2_orphan_recovery_can_continue(struct ocfs2_super *osb,
1324 int slot)
1325{
1326 int ret;
1327
1328 spin_lock(&osb->osb_lock);
1329 ret = !osb->osb_orphan_wipes[slot];
1330 spin_unlock(&osb->osb_lock);
1331 return ret;
1332}
1333
1334static void ocfs2_mark_recovering_orphan_dir(struct ocfs2_super *osb,
1335 int slot)
1336{
1337 spin_lock(&osb->osb_lock);
1338 /* Mark ourselves such that new processes in delete_inode()
1339 * know to quit early. */
1340 ocfs2_node_map_set_bit(osb, &osb->osb_recovering_orphan_dirs, slot);
1341 while (osb->osb_orphan_wipes[slot]) {
1342 /* If any processes are already in the middle of an
1343 * orphan wipe on this dir, then we need to wait for
1344 * them. */
1345 spin_unlock(&osb->osb_lock);
1346 wait_event_interruptible(osb->osb_wipe_event,
1347 ocfs2_orphan_recovery_can_continue(osb, slot));
1348 spin_lock(&osb->osb_lock);
1349 }
1350 spin_unlock(&osb->osb_lock);
1351}
1352
1353static void ocfs2_clear_recovering_orphan_dir(struct ocfs2_super *osb,
1354 int slot)
1355{
1356 ocfs2_node_map_clear_bit(osb, &osb->osb_recovering_orphan_dirs, slot);
1357}
1358
1359/*
1360 * Orphan recovery. Each mounted node has it's own orphan dir which we
1361 * must run during recovery. Our strategy here is to build a list of
1362 * the inodes in the orphan dir and iget/iput them. The VFS does
1363 * (most) of the rest of the work.
1364 *
1365 * Orphan recovery can happen at any time, not just mount so we have a
1366 * couple of extra considerations.
1367 *
1368 * - We grab as many inodes as we can under the orphan dir lock -
1369 * doing iget() outside the orphan dir risks getting a reference on
1370 * an invalid inode.
1371 * - We must be sure not to deadlock with other processes on the
1372 * system wanting to run delete_inode(). This can happen when they go
1373 * to lock the orphan dir and the orphan recovery process attempts to
1374 * iget() inside the orphan dir lock. This can be avoided by
1375 * advertising our state to ocfs2_delete_inode().
1376 */
1377static int ocfs2_recover_orphans(struct ocfs2_super *osb,
1378 int slot)
1379{
1380 int ret = 0;
1381 struct inode *inode = NULL;
1382 struct inode *iter;
1383 struct ocfs2_inode_info *oi;
1384
1385 mlog(0, "Recover inodes from orphan dir in slot %d\n", slot);
1386
1387 ocfs2_mark_recovering_orphan_dir(osb, slot);
1388 ret = ocfs2_queue_orphans(osb, slot, &inode);
1389 ocfs2_clear_recovering_orphan_dir(osb, slot);
1390
1391 /* Error here should be noted, but we want to continue with as
1392 * many queued inodes as we've got. */
1393 if (ret)
1394 mlog_errno(ret);
Mark Fashehccd979b2005-12-15 14:31:24 -08001395
1396 while (inode) {
1397 oi = OCFS2_I(inode);
Mark Fashehb0697052006-03-03 10:24:33 -08001398 mlog(0, "iput orphan %llu\n", (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001399
1400 iter = oi->ip_next_orphan;
1401
1402 spin_lock(&oi->ip_lock);
1403 /* Delete voting may have set these on the assumption
1404 * that the other node would wipe them successfully.
1405 * If they are still in the node's orphan dir, we need
1406 * to reset that state. */
1407 oi->ip_flags &= ~(OCFS2_INODE_DELETED|OCFS2_INODE_SKIP_DELETE);
1408
1409 /* Set the proper information to get us going into
1410 * ocfs2_delete_inode. */
1411 oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED;
Mark Fashehccd979b2005-12-15 14:31:24 -08001412 spin_unlock(&oi->ip_lock);
1413
1414 iput(inode);
1415
1416 inode = iter;
1417 }
1418
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001419 return ret;
Mark Fashehccd979b2005-12-15 14:31:24 -08001420}
1421
1422static int ocfs2_wait_on_mount(struct ocfs2_super *osb)
1423{
1424 /* This check is good because ocfs2 will wait on our recovery
1425 * thread before changing it to something other than MOUNTED
1426 * or DISABLED. */
1427 wait_event(osb->osb_mount_event,
1428 atomic_read(&osb->vol_state) == VOLUME_MOUNTED ||
1429 atomic_read(&osb->vol_state) == VOLUME_DISABLED);
1430
1431 /* If there's an error on mount, then we may never get to the
1432 * MOUNTED flag, but this is set right before
1433 * dismount_volume() so we can trust it. */
1434 if (atomic_read(&osb->vol_state) == VOLUME_DISABLED) {
1435 mlog(0, "mount error, exiting!\n");
1436 return -EBUSY;
1437 }
1438
1439 return 0;
1440}
1441
1442static int ocfs2_commit_thread(void *arg)
1443{
1444 int status;
1445 struct ocfs2_super *osb = arg;
1446 struct ocfs2_journal *journal = osb->journal;
1447
1448 /* we can trust j_num_trans here because _should_stop() is only set in
1449 * shutdown and nobody other than ourselves should be able to start
1450 * transactions. committing on shutdown might take a few iterations
1451 * as final transactions put deleted inodes on the list */
1452 while (!(kthread_should_stop() &&
1453 atomic_read(&journal->j_num_trans) == 0)) {
1454
Mark Fasheh745ae8ba2006-02-09 13:23:39 -08001455 wait_event_interruptible(osb->checkpoint_event,
1456 atomic_read(&journal->j_num_trans)
1457 || kthread_should_stop());
Mark Fashehccd979b2005-12-15 14:31:24 -08001458
1459 status = ocfs2_commit_cache(osb);
1460 if (status < 0)
1461 mlog_errno(status);
1462
1463 if (kthread_should_stop() && atomic_read(&journal->j_num_trans)){
1464 mlog(ML_KTHREAD,
1465 "commit_thread: %u transactions pending on "
1466 "shutdown\n",
1467 atomic_read(&journal->j_num_trans));
1468 }
1469 }
1470
1471 return 0;
1472}
1473
1474/* Look for a dirty journal without taking any cluster locks. Used for
1475 * hard readonly access to determine whether the file system journals
1476 * require recovery. */
1477int ocfs2_check_journals_nolocks(struct ocfs2_super *osb)
1478{
1479 int ret = 0;
1480 unsigned int slot;
1481 struct buffer_head *di_bh;
1482 struct ocfs2_dinode *di;
1483 struct inode *journal = NULL;
1484
1485 for(slot = 0; slot < osb->max_slots; slot++) {
1486 journal = ocfs2_get_system_file_inode(osb,
1487 JOURNAL_SYSTEM_INODE,
1488 slot);
1489 if (!journal || is_bad_inode(journal)) {
1490 ret = -EACCES;
1491 mlog_errno(ret);
1492 goto out;
1493 }
1494
1495 di_bh = NULL;
1496 ret = ocfs2_read_block(osb, OCFS2_I(journal)->ip_blkno, &di_bh,
1497 0, journal);
1498 if (ret < 0) {
1499 mlog_errno(ret);
1500 goto out;
1501 }
1502
1503 di = (struct ocfs2_dinode *) di_bh->b_data;
1504
1505 if (le32_to_cpu(di->id1.journal1.ij_flags) &
1506 OCFS2_JOURNAL_DIRTY_FL)
1507 ret = -EROFS;
1508
1509 brelse(di_bh);
1510 if (ret)
1511 break;
1512 }
1513
1514out:
1515 if (journal)
1516 iput(journal);
1517
1518 return ret;
1519}