blob: 57d7d25a2b9a3b9e0a68c7bf8f1348832e69cb25 [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"
Joel Becker50655ae2008-09-11 15:53:07 -070038#include "blockcheck.h"
Mark Fasheh316f4b92007-09-07 18:21:26 -070039#include "dir.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080040#include "dlmglue.h"
41#include "extent_map.h"
42#include "heartbeat.h"
43#include "inode.h"
44#include "journal.h"
45#include "localalloc.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080046#include "slot_map.h"
47#include "super.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080048#include "sysfile.h"
Jan Kara22053632008-10-20 23:50:38 +020049#include "quota.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080050
51#include "buffer_head_io.h"
52
Ingo Molnar34af9462006-06-27 02:53:55 -070053DEFINE_SPINLOCK(trans_inc_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -080054
55static int ocfs2_force_read_journal(struct inode *inode);
56static int ocfs2_recover_node(struct ocfs2_super *osb,
Jan Kara22053632008-10-20 23:50:38 +020057 int node_num, int slot_num);
Mark Fashehccd979b2005-12-15 14:31:24 -080058static int __ocfs2_recovery_thread(void *arg);
59static int ocfs2_commit_cache(struct ocfs2_super *osb);
Jan Kara19ece542008-08-21 20:13:17 +020060static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota);
Mark Fashehccd979b2005-12-15 14:31:24 -080061static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
Sunil Mushran539d8262008-07-14 17:31:10 -070062 int dirty, int replayed);
Mark Fashehccd979b2005-12-15 14:31:24 -080063static int ocfs2_trylock_journal(struct ocfs2_super *osb,
64 int slot_num);
65static int ocfs2_recover_orphans(struct ocfs2_super *osb,
66 int slot);
67static int ocfs2_commit_thread(void *arg);
68
Jan Kara19ece542008-08-21 20:13:17 +020069static inline int ocfs2_wait_on_mount(struct ocfs2_super *osb)
70{
71 return __ocfs2_wait_on_mount(osb, 0);
72}
73
74static inline int ocfs2_wait_on_quotas(struct ocfs2_super *osb)
75{
76 return __ocfs2_wait_on_mount(osb, 1);
77}
78
79
Joel Becker553abd02008-02-01 12:03:57 -080080
81/*
82 * The recovery_list is a simple linked list of node numbers to recover.
83 * It is protected by the recovery_lock.
84 */
85
86struct ocfs2_recovery_map {
Joel Beckerfc881fa2008-02-01 12:04:48 -080087 unsigned int rm_used;
Joel Becker553abd02008-02-01 12:03:57 -080088 unsigned int *rm_entries;
89};
90
91int ocfs2_recovery_init(struct ocfs2_super *osb)
92{
93 struct ocfs2_recovery_map *rm;
94
95 mutex_init(&osb->recovery_lock);
96 osb->disable_recovery = 0;
97 osb->recovery_thread_task = NULL;
98 init_waitqueue_head(&osb->recovery_event);
99
100 rm = kzalloc(sizeof(struct ocfs2_recovery_map) +
101 osb->max_slots * sizeof(unsigned int),
102 GFP_KERNEL);
103 if (!rm) {
104 mlog_errno(-ENOMEM);
105 return -ENOMEM;
106 }
107
108 rm->rm_entries = (unsigned int *)((char *)rm +
109 sizeof(struct ocfs2_recovery_map));
110 osb->recovery_map = rm;
111
112 return 0;
113}
114
115/* we can't grab the goofy sem lock from inside wait_event, so we use
116 * memory barriers to make sure that we'll see the null task before
117 * being woken up */
118static int ocfs2_recovery_thread_running(struct ocfs2_super *osb)
119{
120 mb();
121 return osb->recovery_thread_task != NULL;
122}
123
124void ocfs2_recovery_exit(struct ocfs2_super *osb)
125{
126 struct ocfs2_recovery_map *rm;
127
128 /* disable any new recovery threads and wait for any currently
129 * running ones to exit. Do this before setting the vol_state. */
130 mutex_lock(&osb->recovery_lock);
131 osb->disable_recovery = 1;
132 mutex_unlock(&osb->recovery_lock);
133 wait_event(osb->recovery_event, !ocfs2_recovery_thread_running(osb));
134
135 /* At this point, we know that no more recovery threads can be
136 * launched, so wait for any recovery completion work to
137 * complete. */
138 flush_workqueue(ocfs2_wq);
139
140 /*
141 * Now that recovery is shut down, and the osb is about to be
142 * freed, the osb_lock is not taken here.
143 */
144 rm = osb->recovery_map;
145 /* XXX: Should we bug if there are dirty entries? */
146
147 kfree(rm);
148}
149
150static int __ocfs2_recovery_map_test(struct ocfs2_super *osb,
151 unsigned int node_num)
152{
153 int i;
154 struct ocfs2_recovery_map *rm = osb->recovery_map;
155
156 assert_spin_locked(&osb->osb_lock);
157
158 for (i = 0; i < rm->rm_used; i++) {
159 if (rm->rm_entries[i] == node_num)
160 return 1;
161 }
162
163 return 0;
164}
165
166/* Behaves like test-and-set. Returns the previous value */
167static int ocfs2_recovery_map_set(struct ocfs2_super *osb,
168 unsigned int node_num)
169{
170 struct ocfs2_recovery_map *rm = osb->recovery_map;
171
172 spin_lock(&osb->osb_lock);
173 if (__ocfs2_recovery_map_test(osb, node_num)) {
174 spin_unlock(&osb->osb_lock);
175 return 1;
176 }
177
178 /* XXX: Can this be exploited? Not from o2dlm... */
179 BUG_ON(rm->rm_used >= osb->max_slots);
180
181 rm->rm_entries[rm->rm_used] = node_num;
182 rm->rm_used++;
183 spin_unlock(&osb->osb_lock);
184
185 return 0;
186}
187
188static void ocfs2_recovery_map_clear(struct ocfs2_super *osb,
189 unsigned int node_num)
190{
191 int i;
192 struct ocfs2_recovery_map *rm = osb->recovery_map;
193
194 spin_lock(&osb->osb_lock);
195
196 for (i = 0; i < rm->rm_used; i++) {
197 if (rm->rm_entries[i] == node_num)
198 break;
199 }
200
201 if (i < rm->rm_used) {
202 /* XXX: be careful with the pointer math */
203 memmove(&(rm->rm_entries[i]), &(rm->rm_entries[i + 1]),
204 (rm->rm_used - i - 1) * sizeof(unsigned int));
205 rm->rm_used--;
206 }
207
208 spin_unlock(&osb->osb_lock);
209}
210
Mark Fashehccd979b2005-12-15 14:31:24 -0800211static int ocfs2_commit_cache(struct ocfs2_super *osb)
212{
213 int status = 0;
214 unsigned int flushed;
215 unsigned long old_id;
216 struct ocfs2_journal *journal = NULL;
217
218 mlog_entry_void();
219
220 journal = osb->journal;
221
222 /* Flush all pending commits and checkpoint the journal. */
223 down_write(&journal->j_trans_barrier);
224
225 if (atomic_read(&journal->j_num_trans) == 0) {
226 up_write(&journal->j_trans_barrier);
227 mlog(0, "No transactions for me to flush!\n");
228 goto finally;
229 }
230
Joel Becker2b4e30f2008-09-03 20:03:41 -0700231 jbd2_journal_lock_updates(journal->j_journal);
232 status = jbd2_journal_flush(journal->j_journal);
233 jbd2_journal_unlock_updates(journal->j_journal);
Mark Fashehccd979b2005-12-15 14:31:24 -0800234 if (status < 0) {
235 up_write(&journal->j_trans_barrier);
236 mlog_errno(status);
237 goto finally;
238 }
239
240 old_id = ocfs2_inc_trans_id(journal);
241
242 flushed = atomic_read(&journal->j_num_trans);
243 atomic_set(&journal->j_num_trans, 0);
244 up_write(&journal->j_trans_barrier);
245
246 mlog(0, "commit_thread: flushed transaction %lu (%u handles)\n",
247 journal->j_trans_id, flushed);
248
Mark Fasheh34d024f2007-09-24 15:56:19 -0700249 ocfs2_wake_downconvert_thread(osb);
Mark Fashehccd979b2005-12-15 14:31:24 -0800250 wake_up(&journal->j_checkpointed);
251finally:
252 mlog_exit(status);
253 return status;
254}
255
Mark Fashehccd979b2005-12-15 14:31:24 -0800256/* pass it NULL and it will allocate a new handle object for you. If
257 * you pass it a handle however, it may still return error, in which
258 * case it has free'd the passed handle for you. */
Mark Fasheh1fabe142006-10-09 18:11:45 -0700259handle_t *ocfs2_start_trans(struct ocfs2_super *osb, int max_buffs)
Mark Fashehccd979b2005-12-15 14:31:24 -0800260{
Mark Fashehccd979b2005-12-15 14:31:24 -0800261 journal_t *journal = osb->journal->j_journal;
Mark Fasheh1fabe142006-10-09 18:11:45 -0700262 handle_t *handle;
Mark Fashehccd979b2005-12-15 14:31:24 -0800263
Eric Sesterhenn / snakebyteebdec832006-01-27 10:32:52 +0100264 BUG_ON(!osb || !osb->journal->j_journal);
Mark Fashehccd979b2005-12-15 14:31:24 -0800265
Mark Fasheh65eff9c2006-10-09 17:26:22 -0700266 if (ocfs2_is_hard_readonly(osb))
267 return ERR_PTR(-EROFS);
Mark Fashehccd979b2005-12-15 14:31:24 -0800268
269 BUG_ON(osb->journal->j_state == OCFS2_JOURNAL_FREE);
270 BUG_ON(max_buffs <= 0);
271
Jan Kara90e86a62008-08-27 22:30:28 +0200272 /* Nested transaction? Just return the handle... */
273 if (journal_current_handle())
274 return jbd2_journal_start(journal, max_buffs);
Mark Fashehccd979b2005-12-15 14:31:24 -0800275
Mark Fashehccd979b2005-12-15 14:31:24 -0800276 down_read(&osb->journal->j_trans_barrier);
277
Joel Becker2b4e30f2008-09-03 20:03:41 -0700278 handle = jbd2_journal_start(journal, max_buffs);
Mark Fasheh1fabe142006-10-09 18:11:45 -0700279 if (IS_ERR(handle)) {
Mark Fashehccd979b2005-12-15 14:31:24 -0800280 up_read(&osb->journal->j_trans_barrier);
281
Mark Fasheh1fabe142006-10-09 18:11:45 -0700282 mlog_errno(PTR_ERR(handle));
Mark Fashehccd979b2005-12-15 14:31:24 -0800283
284 if (is_journal_aborted(journal)) {
285 ocfs2_abort(osb->sb, "Detected aborted journal");
Mark Fasheh1fabe142006-10-09 18:11:45 -0700286 handle = ERR_PTR(-EROFS);
Mark Fashehccd979b2005-12-15 14:31:24 -0800287 }
Sunil Mushranc271c5c2006-12-05 17:56:35 -0800288 } else {
289 if (!ocfs2_mount_local(osb))
290 atomic_inc(&(osb->journal->j_num_trans));
291 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800292
Mark Fashehccd979b2005-12-15 14:31:24 -0800293 return handle;
Mark Fashehccd979b2005-12-15 14:31:24 -0800294}
295
Mark Fasheh1fabe142006-10-09 18:11:45 -0700296int ocfs2_commit_trans(struct ocfs2_super *osb,
297 handle_t *handle)
Mark Fashehccd979b2005-12-15 14:31:24 -0800298{
Jan Kara90e86a62008-08-27 22:30:28 +0200299 int ret, nested;
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700300 struct ocfs2_journal *journal = osb->journal;
Mark Fashehccd979b2005-12-15 14:31:24 -0800301
302 BUG_ON(!handle);
303
Jan Kara90e86a62008-08-27 22:30:28 +0200304 nested = handle->h_ref > 1;
Joel Becker2b4e30f2008-09-03 20:03:41 -0700305 ret = jbd2_journal_stop(handle);
Mark Fasheh1fabe142006-10-09 18:11:45 -0700306 if (ret < 0)
307 mlog_errno(ret);
Mark Fashehccd979b2005-12-15 14:31:24 -0800308
Jan Kara90e86a62008-08-27 22:30:28 +0200309 if (!nested)
310 up_read(&journal->j_trans_barrier);
Mark Fashehccd979b2005-12-15 14:31:24 -0800311
Mark Fasheh1fabe142006-10-09 18:11:45 -0700312 return ret;
Mark Fashehccd979b2005-12-15 14:31:24 -0800313}
314
315/*
316 * 'nblocks' is what you want to add to the current
317 * transaction. extend_trans will either extend the current handle by
318 * nblocks, or commit it and start a new one with nblocks credits.
319 *
Joel Becker2b4e30f2008-09-03 20:03:41 -0700320 * This might call jbd2_journal_restart() which will commit dirty buffers
Mark Fashehe8aed342007-12-03 16:43:01 -0800321 * and then restart the transaction. Before calling
322 * ocfs2_extend_trans(), any changed blocks should have been
323 * dirtied. After calling it, all blocks which need to be changed must
324 * go through another set of journal_access/journal_dirty calls.
325 *
Mark Fashehccd979b2005-12-15 14:31:24 -0800326 * WARNING: This will not release any semaphores or disk locks taken
327 * during the transaction, so make sure they were taken *before*
328 * start_trans or we'll have ordering deadlocks.
329 *
330 * WARNING2: Note that we do *not* drop j_trans_barrier here. This is
331 * good because transaction ids haven't yet been recorded on the
332 * cluster locks associated with this handle.
333 */
Mark Fasheh1fc58142006-10-05 14:15:36 -0700334int ocfs2_extend_trans(handle_t *handle, int nblocks)
Mark Fashehccd979b2005-12-15 14:31:24 -0800335{
336 int status;
337
338 BUG_ON(!handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800339 BUG_ON(!nblocks);
340
341 mlog_entry_void();
342
343 mlog(0, "Trying to extend transaction by %d blocks\n", nblocks);
344
Joel Beckere407e392008-06-12 22:35:39 -0700345#ifdef CONFIG_OCFS2_DEBUG_FS
Mark Fasheh0879c582007-12-03 16:42:19 -0800346 status = 1;
347#else
Joel Becker2b4e30f2008-09-03 20:03:41 -0700348 status = jbd2_journal_extend(handle, nblocks);
Mark Fashehccd979b2005-12-15 14:31:24 -0800349 if (status < 0) {
350 mlog_errno(status);
351 goto bail;
352 }
Mark Fasheh0879c582007-12-03 16:42:19 -0800353#endif
Mark Fashehccd979b2005-12-15 14:31:24 -0800354
355 if (status > 0) {
Joel Becker2b4e30f2008-09-03 20:03:41 -0700356 mlog(0,
357 "jbd2_journal_extend failed, trying "
358 "jbd2_journal_restart\n");
359 status = jbd2_journal_restart(handle, nblocks);
Mark Fashehccd979b2005-12-15 14:31:24 -0800360 if (status < 0) {
Mark Fashehccd979b2005-12-15 14:31:24 -0800361 mlog_errno(status);
362 goto bail;
363 }
Mark Fasheh01ddf1e2006-10-05 13:54:39 -0700364 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800365
366 status = 0;
367bail:
368
369 mlog_exit(status);
370 return status;
371}
372
Joel Becker50655ae2008-09-11 15:53:07 -0700373struct ocfs2_triggers {
374 struct jbd2_buffer_trigger_type ot_triggers;
375 int ot_offset;
376};
377
378static inline struct ocfs2_triggers *to_ocfs2_trigger(struct jbd2_buffer_trigger_type *triggers)
379{
380 return container_of(triggers, struct ocfs2_triggers, ot_triggers);
381}
382
383static void ocfs2_commit_trigger(struct jbd2_buffer_trigger_type *triggers,
384 struct buffer_head *bh,
385 void *data, size_t size)
386{
387 struct ocfs2_triggers *ot = to_ocfs2_trigger(triggers);
388
389 /*
390 * We aren't guaranteed to have the superblock here, so we
391 * must unconditionally compute the ecc data.
392 * __ocfs2_journal_access() will only set the triggers if
393 * metaecc is enabled.
394 */
395 ocfs2_block_check_compute(data, size, data + ot->ot_offset);
396}
397
398/*
399 * Quota blocks have their own trigger because the struct ocfs2_block_check
400 * offset depends on the blocksize.
401 */
402static void ocfs2_dq_commit_trigger(struct jbd2_buffer_trigger_type *triggers,
403 struct buffer_head *bh,
404 void *data, size_t size)
405{
406 struct ocfs2_disk_dqtrailer *dqt =
407 ocfs2_block_dqtrailer(size, data);
408
409 /*
410 * We aren't guaranteed to have the superblock here, so we
411 * must unconditionally compute the ecc data.
412 * __ocfs2_journal_access() will only set the triggers if
413 * metaecc is enabled.
414 */
415 ocfs2_block_check_compute(data, size, &dqt->dq_check);
416}
417
Joel Beckerc175a512008-12-10 17:58:22 -0800418/*
419 * Directory blocks also have their own trigger because the
420 * struct ocfs2_block_check offset depends on the blocksize.
421 */
422static void ocfs2_db_commit_trigger(struct jbd2_buffer_trigger_type *triggers,
423 struct buffer_head *bh,
424 void *data, size_t size)
425{
426 struct ocfs2_dir_block_trailer *trailer =
427 ocfs2_dir_trailer_from_size(size, data);
428
429 /*
430 * We aren't guaranteed to have the superblock here, so we
431 * must unconditionally compute the ecc data.
432 * __ocfs2_journal_access() will only set the triggers if
433 * metaecc is enabled.
434 */
435 ocfs2_block_check_compute(data, size, &trailer->db_check);
436}
437
Joel Becker50655ae2008-09-11 15:53:07 -0700438static void ocfs2_abort_trigger(struct jbd2_buffer_trigger_type *triggers,
439 struct buffer_head *bh)
440{
441 mlog(ML_ERROR,
442 "ocfs2_abort_trigger called by JBD2. bh = 0x%lx, "
443 "bh->b_blocknr = %llu\n",
444 (unsigned long)bh,
445 (unsigned long long)bh->b_blocknr);
446
447 /* We aren't guaranteed to have the superblock here - but if we
448 * don't, it'll just crash. */
449 ocfs2_error(bh->b_assoc_map->host->i_sb,
450 "JBD2 has aborted our journal, ocfs2 cannot continue\n");
451}
452
453static struct ocfs2_triggers di_triggers = {
454 .ot_triggers = {
455 .t_commit = ocfs2_commit_trigger,
456 .t_abort = ocfs2_abort_trigger,
457 },
458 .ot_offset = offsetof(struct ocfs2_dinode, i_check),
459};
460
461static struct ocfs2_triggers eb_triggers = {
462 .ot_triggers = {
463 .t_commit = ocfs2_commit_trigger,
464 .t_abort = ocfs2_abort_trigger,
465 },
466 .ot_offset = offsetof(struct ocfs2_extent_block, h_check),
467};
468
469static struct ocfs2_triggers gd_triggers = {
470 .ot_triggers = {
471 .t_commit = ocfs2_commit_trigger,
472 .t_abort = ocfs2_abort_trigger,
473 },
474 .ot_offset = offsetof(struct ocfs2_group_desc, bg_check),
475};
476
Joel Beckerc175a512008-12-10 17:58:22 -0800477static struct ocfs2_triggers db_triggers = {
478 .ot_triggers = {
479 .t_commit = ocfs2_db_commit_trigger,
480 .t_abort = ocfs2_abort_trigger,
481 },
482};
483
Joel Becker50655ae2008-09-11 15:53:07 -0700484static struct ocfs2_triggers xb_triggers = {
485 .ot_triggers = {
486 .t_commit = ocfs2_commit_trigger,
487 .t_abort = ocfs2_abort_trigger,
488 },
489 .ot_offset = offsetof(struct ocfs2_xattr_block, xb_check),
490};
491
492static struct ocfs2_triggers dq_triggers = {
493 .ot_triggers = {
494 .t_commit = ocfs2_dq_commit_trigger,
495 .t_abort = ocfs2_abort_trigger,
496 },
497};
498
499static int __ocfs2_journal_access(handle_t *handle,
500 struct inode *inode,
501 struct buffer_head *bh,
502 struct ocfs2_triggers *triggers,
503 int type)
Mark Fashehccd979b2005-12-15 14:31:24 -0800504{
505 int status;
506
507 BUG_ON(!inode);
508 BUG_ON(!handle);
509 BUG_ON(!bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800510
Badari Pulavarty205f87f2006-03-26 01:38:00 -0800511 mlog_entry("bh->b_blocknr=%llu, type=%d (\"%s\"), bh->b_size = %zu\n",
Mark Fashehccd979b2005-12-15 14:31:24 -0800512 (unsigned long long)bh->b_blocknr, type,
513 (type == OCFS2_JOURNAL_ACCESS_CREATE) ?
514 "OCFS2_JOURNAL_ACCESS_CREATE" :
515 "OCFS2_JOURNAL_ACCESS_WRITE",
516 bh->b_size);
517
518 /* we can safely remove this assertion after testing. */
519 if (!buffer_uptodate(bh)) {
520 mlog(ML_ERROR, "giving me a buffer that's not uptodate!\n");
521 mlog(ML_ERROR, "b_blocknr=%llu\n",
522 (unsigned long long)bh->b_blocknr);
523 BUG();
524 }
525
526 /* Set the current transaction information on the inode so
527 * that the locking code knows whether it can drop it's locks
528 * on this inode or not. We're protected from the commit
529 * thread updating the current transaction id until
530 * ocfs2_commit_trans() because ocfs2_start_trans() took
531 * j_trans_barrier for us. */
532 ocfs2_set_inode_lock_trans(OCFS2_SB(inode->i_sb)->journal, inode);
533
Mark Fasheh251b6ec2006-01-10 15:41:43 -0800534 mutex_lock(&OCFS2_I(inode)->ip_io_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800535 switch (type) {
536 case OCFS2_JOURNAL_ACCESS_CREATE:
537 case OCFS2_JOURNAL_ACCESS_WRITE:
Joel Becker2b4e30f2008-09-03 20:03:41 -0700538 status = jbd2_journal_get_write_access(handle, bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800539 break;
540
541 case OCFS2_JOURNAL_ACCESS_UNDO:
Joel Becker2b4e30f2008-09-03 20:03:41 -0700542 status = jbd2_journal_get_undo_access(handle, bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800543 break;
544
545 default:
546 status = -EINVAL;
547 mlog(ML_ERROR, "Uknown access type!\n");
548 }
Joel Becker50655ae2008-09-11 15:53:07 -0700549 if (!status && ocfs2_meta_ecc(OCFS2_SB(inode->i_sb)) && triggers)
550 jbd2_journal_set_triggers(bh, &triggers->ot_triggers);
Mark Fasheh251b6ec2006-01-10 15:41:43 -0800551 mutex_unlock(&OCFS2_I(inode)->ip_io_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800552
553 if (status < 0)
554 mlog(ML_ERROR, "Error %d getting %d access to buffer!\n",
555 status, type);
556
557 mlog_exit(status);
558 return status;
559}
560
Joel Becker50655ae2008-09-11 15:53:07 -0700561int ocfs2_journal_access_di(handle_t *handle, struct inode *inode,
562 struct buffer_head *bh, int type)
563{
564 return __ocfs2_journal_access(handle, inode, bh, &di_triggers,
565 type);
566}
567
568int ocfs2_journal_access_eb(handle_t *handle, struct inode *inode,
569 struct buffer_head *bh, int type)
570{
571 return __ocfs2_journal_access(handle, inode, bh, &eb_triggers,
572 type);
573}
574
575int ocfs2_journal_access_gd(handle_t *handle, struct inode *inode,
576 struct buffer_head *bh, int type)
577{
578 return __ocfs2_journal_access(handle, inode, bh, &gd_triggers,
579 type);
580}
581
582int ocfs2_journal_access_db(handle_t *handle, struct inode *inode,
583 struct buffer_head *bh, int type)
584{
Joel Beckerc175a512008-12-10 17:58:22 -0800585 return __ocfs2_journal_access(handle, inode, bh, &db_triggers,
586 type);
Joel Becker50655ae2008-09-11 15:53:07 -0700587}
588
589int ocfs2_journal_access_xb(handle_t *handle, struct inode *inode,
590 struct buffer_head *bh, int type)
591{
592 return __ocfs2_journal_access(handle, inode, bh, &xb_triggers,
593 type);
594}
595
596int ocfs2_journal_access_dq(handle_t *handle, struct inode *inode,
597 struct buffer_head *bh, int type)
598{
599 return __ocfs2_journal_access(handle, inode, bh, &dq_triggers,
600 type);
601}
602
603int ocfs2_journal_access(handle_t *handle, struct inode *inode,
604 struct buffer_head *bh, int type)
605{
606 return __ocfs2_journal_access(handle, inode, bh, NULL, type);
607}
608
Mark Fasheh1fabe142006-10-09 18:11:45 -0700609int ocfs2_journal_dirty(handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -0800610 struct buffer_head *bh)
611{
612 int status;
613
Mark Fashehccd979b2005-12-15 14:31:24 -0800614 mlog_entry("(bh->b_blocknr=%llu)\n",
615 (unsigned long long)bh->b_blocknr);
616
Joel Becker2b4e30f2008-09-03 20:03:41 -0700617 status = jbd2_journal_dirty_metadata(handle, bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800618 if (status < 0)
619 mlog(ML_ERROR, "Could not dirty metadata buffer. "
620 "(bh->b_blocknr=%llu)\n",
621 (unsigned long long)bh->b_blocknr);
622
623 mlog_exit(status);
624 return status;
625}
626
Joel Becker2b4e30f2008-09-03 20:03:41 -0700627#define OCFS2_DEFAULT_COMMIT_INTERVAL (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE)
Mark Fashehccd979b2005-12-15 14:31:24 -0800628
629void ocfs2_set_journal_params(struct ocfs2_super *osb)
630{
631 journal_t *journal = osb->journal->j_journal;
Mark Fashehd147b3d2007-11-07 14:40:36 -0800632 unsigned long commit_interval = OCFS2_DEFAULT_COMMIT_INTERVAL;
633
634 if (osb->osb_commit_interval)
635 commit_interval = osb->osb_commit_interval;
Mark Fashehccd979b2005-12-15 14:31:24 -0800636
637 spin_lock(&journal->j_state_lock);
Mark Fashehd147b3d2007-11-07 14:40:36 -0800638 journal->j_commit_interval = commit_interval;
Mark Fashehccd979b2005-12-15 14:31:24 -0800639 if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER)
Joel Becker2b4e30f2008-09-03 20:03:41 -0700640 journal->j_flags |= JBD2_BARRIER;
Mark Fashehccd979b2005-12-15 14:31:24 -0800641 else
Joel Becker2b4e30f2008-09-03 20:03:41 -0700642 journal->j_flags &= ~JBD2_BARRIER;
Mark Fashehccd979b2005-12-15 14:31:24 -0800643 spin_unlock(&journal->j_state_lock);
644}
645
646int ocfs2_journal_init(struct ocfs2_journal *journal, int *dirty)
647{
648 int status = -1;
649 struct inode *inode = NULL; /* the journal inode */
650 journal_t *j_journal = NULL;
651 struct ocfs2_dinode *di = NULL;
652 struct buffer_head *bh = NULL;
653 struct ocfs2_super *osb;
Mark Fashehe63aecb62007-10-18 15:30:42 -0700654 int inode_lock = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -0800655
656 mlog_entry_void();
657
658 BUG_ON(!journal);
659
660 osb = journal->j_osb;
661
662 /* already have the inode for our journal */
663 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
664 osb->slot_num);
665 if (inode == NULL) {
666 status = -EACCES;
667 mlog_errno(status);
668 goto done;
669 }
670 if (is_bad_inode(inode)) {
671 mlog(ML_ERROR, "access error (bad inode)\n");
672 iput(inode);
673 inode = NULL;
674 status = -EACCES;
675 goto done;
676 }
677
678 SET_INODE_JOURNAL(inode);
679 OCFS2_I(inode)->ip_open_count++;
680
Mark Fasheh6eff5792006-01-18 10:31:47 -0800681 /* Skip recovery waits here - journal inode metadata never
682 * changes in a live cluster so it can be considered an
683 * exception to the rule. */
Mark Fashehe63aecb62007-10-18 15:30:42 -0700684 status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY);
Mark Fashehccd979b2005-12-15 14:31:24 -0800685 if (status < 0) {
686 if (status != -ERESTARTSYS)
687 mlog(ML_ERROR, "Could not get lock on journal!\n");
688 goto done;
689 }
690
Mark Fashehe63aecb62007-10-18 15:30:42 -0700691 inode_lock = 1;
Mark Fashehccd979b2005-12-15 14:31:24 -0800692 di = (struct ocfs2_dinode *)bh->b_data;
693
694 if (inode->i_size < OCFS2_MIN_JOURNAL_SIZE) {
695 mlog(ML_ERROR, "Journal file size (%lld) is too small!\n",
696 inode->i_size);
697 status = -EINVAL;
698 goto done;
699 }
700
701 mlog(0, "inode->i_size = %lld\n", inode->i_size);
Andrew Morton5515eff2006-03-26 01:37:53 -0800702 mlog(0, "inode->i_blocks = %llu\n",
703 (unsigned long long)inode->i_blocks);
Mark Fashehccd979b2005-12-15 14:31:24 -0800704 mlog(0, "inode->ip_clusters = %u\n", OCFS2_I(inode)->ip_clusters);
705
706 /* call the kernels journal init function now */
Joel Becker2b4e30f2008-09-03 20:03:41 -0700707 j_journal = jbd2_journal_init_inode(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -0800708 if (j_journal == NULL) {
709 mlog(ML_ERROR, "Linux journal layer error\n");
710 status = -EINVAL;
711 goto done;
712 }
713
Joel Becker2b4e30f2008-09-03 20:03:41 -0700714 mlog(0, "Returned from jbd2_journal_init_inode\n");
Mark Fashehccd979b2005-12-15 14:31:24 -0800715 mlog(0, "j_journal->j_maxlen = %u\n", j_journal->j_maxlen);
716
717 *dirty = (le32_to_cpu(di->id1.journal1.ij_flags) &
718 OCFS2_JOURNAL_DIRTY_FL);
719
720 journal->j_journal = j_journal;
721 journal->j_inode = inode;
722 journal->j_bh = bh;
723
724 ocfs2_set_journal_params(osb);
725
726 journal->j_state = OCFS2_JOURNAL_LOADED;
727
728 status = 0;
729done:
730 if (status < 0) {
Mark Fashehe63aecb62007-10-18 15:30:42 -0700731 if (inode_lock)
732 ocfs2_inode_unlock(inode, 1);
Mark Fasheha81cb882008-10-07 14:25:16 -0700733 brelse(bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800734 if (inode) {
735 OCFS2_I(inode)->ip_open_count--;
736 iput(inode);
737 }
738 }
739
740 mlog_exit(status);
741 return status;
742}
743
Sunil Mushran539d8262008-07-14 17:31:10 -0700744static void ocfs2_bump_recovery_generation(struct ocfs2_dinode *di)
745{
746 le32_add_cpu(&(di->id1.journal1.ij_recovery_generation), 1);
747}
748
749static u32 ocfs2_get_recovery_generation(struct ocfs2_dinode *di)
750{
751 return le32_to_cpu(di->id1.journal1.ij_recovery_generation);
752}
753
Mark Fashehccd979b2005-12-15 14:31:24 -0800754static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
Sunil Mushran539d8262008-07-14 17:31:10 -0700755 int dirty, int replayed)
Mark Fashehccd979b2005-12-15 14:31:24 -0800756{
757 int status;
758 unsigned int flags;
759 struct ocfs2_journal *journal = osb->journal;
760 struct buffer_head *bh = journal->j_bh;
761 struct ocfs2_dinode *fe;
762
763 mlog_entry_void();
764
765 fe = (struct ocfs2_dinode *)bh->b_data;
Joel Becker10995aa2008-11-13 14:49:12 -0800766
767 /* The journal bh on the osb always comes from ocfs2_journal_init()
768 * and was validated there inside ocfs2_inode_lock_full(). It's a
769 * code bug if we mess it up. */
770 BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
Mark Fashehccd979b2005-12-15 14:31:24 -0800771
772 flags = le32_to_cpu(fe->id1.journal1.ij_flags);
773 if (dirty)
774 flags |= OCFS2_JOURNAL_DIRTY_FL;
775 else
776 flags &= ~OCFS2_JOURNAL_DIRTY_FL;
777 fe->id1.journal1.ij_flags = cpu_to_le32(flags);
778
Sunil Mushran539d8262008-07-14 17:31:10 -0700779 if (replayed)
780 ocfs2_bump_recovery_generation(fe);
781
Joel Becker13723d02008-10-17 19:25:01 -0700782 ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check);
Mark Fashehccd979b2005-12-15 14:31:24 -0800783 status = ocfs2_write_block(osb, bh, journal->j_inode);
784 if (status < 0)
785 mlog_errno(status);
786
Mark Fashehccd979b2005-12-15 14:31:24 -0800787 mlog_exit(status);
788 return status;
789}
790
791/*
792 * If the journal has been kmalloc'd it needs to be freed after this
793 * call.
794 */
795void ocfs2_journal_shutdown(struct ocfs2_super *osb)
796{
797 struct ocfs2_journal *journal = NULL;
798 int status = 0;
799 struct inode *inode = NULL;
800 int num_running_trans = 0;
801
802 mlog_entry_void();
803
Eric Sesterhenn / snakebyteebdec832006-01-27 10:32:52 +0100804 BUG_ON(!osb);
Mark Fashehccd979b2005-12-15 14:31:24 -0800805
806 journal = osb->journal;
807 if (!journal)
808 goto done;
809
810 inode = journal->j_inode;
811
812 if (journal->j_state != OCFS2_JOURNAL_LOADED)
813 goto done;
814
Joel Becker2b4e30f2008-09-03 20:03:41 -0700815 /* need to inc inode use count - jbd2_journal_destroy will iput. */
Mark Fashehccd979b2005-12-15 14:31:24 -0800816 if (!igrab(inode))
817 BUG();
818
819 num_running_trans = atomic_read(&(osb->journal->j_num_trans));
820 if (num_running_trans > 0)
821 mlog(0, "Shutting down journal: must wait on %d "
822 "running transactions!\n",
823 num_running_trans);
824
825 /* Do a commit_cache here. It will flush our journal, *and*
826 * release any locks that are still held.
827 * set the SHUTDOWN flag and release the trans lock.
828 * the commit thread will take the trans lock for us below. */
829 journal->j_state = OCFS2_JOURNAL_IN_SHUTDOWN;
830
831 /* The OCFS2_JOURNAL_IN_SHUTDOWN will signal to commit_cache to not
832 * drop the trans_lock (which we want to hold until we
833 * completely destroy the journal. */
834 if (osb->commit_task) {
835 /* Wait for the commit thread */
836 mlog(0, "Waiting for ocfs2commit to exit....\n");
837 kthread_stop(osb->commit_task);
838 osb->commit_task = NULL;
839 }
840
841 BUG_ON(atomic_read(&(osb->journal->j_num_trans)) != 0);
842
Sunil Mushranc271c5c2006-12-05 17:56:35 -0800843 if (ocfs2_mount_local(osb)) {
Joel Becker2b4e30f2008-09-03 20:03:41 -0700844 jbd2_journal_lock_updates(journal->j_journal);
845 status = jbd2_journal_flush(journal->j_journal);
846 jbd2_journal_unlock_updates(journal->j_journal);
Sunil Mushranc271c5c2006-12-05 17:56:35 -0800847 if (status < 0)
848 mlog_errno(status);
849 }
850
851 if (status == 0) {
852 /*
853 * Do not toggle if flush was unsuccessful otherwise
854 * will leave dirty metadata in a "clean" journal
855 */
Sunil Mushran539d8262008-07-14 17:31:10 -0700856 status = ocfs2_journal_toggle_dirty(osb, 0, 0);
Sunil Mushranc271c5c2006-12-05 17:56:35 -0800857 if (status < 0)
858 mlog_errno(status);
859 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800860
861 /* Shutdown the kernel journal system */
Joel Becker2b4e30f2008-09-03 20:03:41 -0700862 jbd2_journal_destroy(journal->j_journal);
Sunil Mushranae0dff62008-10-22 13:24:29 -0700863 journal->j_journal = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800864
865 OCFS2_I(inode)->ip_open_count--;
866
867 /* unlock our journal */
Mark Fashehe63aecb62007-10-18 15:30:42 -0700868 ocfs2_inode_unlock(inode, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -0800869
870 brelse(journal->j_bh);
871 journal->j_bh = NULL;
872
873 journal->j_state = OCFS2_JOURNAL_FREE;
874
875// up_write(&journal->j_trans_barrier);
876done:
877 if (inode)
878 iput(inode);
879 mlog_exit_void();
880}
881
882static void ocfs2_clear_journal_error(struct super_block *sb,
883 journal_t *journal,
884 int slot)
885{
886 int olderr;
887
Joel Becker2b4e30f2008-09-03 20:03:41 -0700888 olderr = jbd2_journal_errno(journal);
Mark Fashehccd979b2005-12-15 14:31:24 -0800889 if (olderr) {
890 mlog(ML_ERROR, "File system error %d recorded in "
891 "journal %u.\n", olderr, slot);
892 mlog(ML_ERROR, "File system on device %s needs checking.\n",
893 sb->s_id);
894
Joel Becker2b4e30f2008-09-03 20:03:41 -0700895 jbd2_journal_ack_err(journal);
896 jbd2_journal_clear_err(journal);
Mark Fashehccd979b2005-12-15 14:31:24 -0800897 }
898}
899
Sunil Mushran539d8262008-07-14 17:31:10 -0700900int ocfs2_journal_load(struct ocfs2_journal *journal, int local, int replayed)
Mark Fashehccd979b2005-12-15 14:31:24 -0800901{
902 int status = 0;
903 struct ocfs2_super *osb;
904
905 mlog_entry_void();
906
Julia Lawallb1f35502008-03-04 15:21:05 -0800907 BUG_ON(!journal);
Mark Fashehccd979b2005-12-15 14:31:24 -0800908
909 osb = journal->j_osb;
910
Joel Becker2b4e30f2008-09-03 20:03:41 -0700911 status = jbd2_journal_load(journal->j_journal);
Mark Fashehccd979b2005-12-15 14:31:24 -0800912 if (status < 0) {
913 mlog(ML_ERROR, "Failed to load journal!\n");
914 goto done;
915 }
916
917 ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num);
918
Sunil Mushran539d8262008-07-14 17:31:10 -0700919 status = ocfs2_journal_toggle_dirty(osb, 1, replayed);
Mark Fashehccd979b2005-12-15 14:31:24 -0800920 if (status < 0) {
921 mlog_errno(status);
922 goto done;
923 }
924
925 /* Launch the commit thread */
Sunil Mushranc271c5c2006-12-05 17:56:35 -0800926 if (!local) {
927 osb->commit_task = kthread_run(ocfs2_commit_thread, osb,
928 "ocfs2cmt");
929 if (IS_ERR(osb->commit_task)) {
930 status = PTR_ERR(osb->commit_task);
931 osb->commit_task = NULL;
932 mlog(ML_ERROR, "unable to launch ocfs2commit thread, "
933 "error=%d", status);
934 goto done;
935 }
936 } else
Mark Fashehccd979b2005-12-15 14:31:24 -0800937 osb->commit_task = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800938
939done:
940 mlog_exit(status);
941 return status;
942}
943
944
945/* 'full' flag tells us whether we clear out all blocks or if we just
946 * mark the journal clean */
947int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full)
948{
949 int status;
950
951 mlog_entry_void();
952
Eric Sesterhenn / snakebyteebdec832006-01-27 10:32:52 +0100953 BUG_ON(!journal);
Mark Fashehccd979b2005-12-15 14:31:24 -0800954
Joel Becker2b4e30f2008-09-03 20:03:41 -0700955 status = jbd2_journal_wipe(journal->j_journal, full);
Mark Fashehccd979b2005-12-15 14:31:24 -0800956 if (status < 0) {
957 mlog_errno(status);
958 goto bail;
959 }
960
Sunil Mushran539d8262008-07-14 17:31:10 -0700961 status = ocfs2_journal_toggle_dirty(journal->j_osb, 0, 0);
Mark Fashehccd979b2005-12-15 14:31:24 -0800962 if (status < 0)
963 mlog_errno(status);
964
965bail:
966 mlog_exit(status);
967 return status;
968}
969
Joel Becker553abd02008-02-01 12:03:57 -0800970static int ocfs2_recovery_completed(struct ocfs2_super *osb)
971{
972 int empty;
973 struct ocfs2_recovery_map *rm = osb->recovery_map;
974
975 spin_lock(&osb->osb_lock);
976 empty = (rm->rm_used == 0);
977 spin_unlock(&osb->osb_lock);
978
979 return empty;
980}
981
982void ocfs2_wait_for_recovery(struct ocfs2_super *osb)
983{
984 wait_event(osb->recovery_event, ocfs2_recovery_completed(osb));
985}
986
Mark Fashehccd979b2005-12-15 14:31:24 -0800987/*
988 * JBD Might read a cached version of another nodes journal file. We
989 * don't want this as this file changes often and we get no
990 * notification on those changes. The only way to be sure that we've
991 * got the most up to date version of those blocks then is to force
992 * read them off disk. Just searching through the buffer cache won't
993 * work as there may be pages backing this file which are still marked
994 * up to date. We know things can't change on this file underneath us
995 * as we have the lock by now :)
996 */
997static int ocfs2_force_read_journal(struct inode *inode)
998{
999 int status = 0;
Mark Fasheh4f902c32007-03-09 16:26:50 -08001000 int i;
Mark Fasheh8110b072007-03-22 16:53:23 -07001001 u64 v_blkno, p_blkno, p_blocks, num_blocks;
Mark Fasheh4f902c32007-03-09 16:26:50 -08001002#define CONCURRENT_JOURNAL_FILL 32ULL
Mark Fashehccd979b2005-12-15 14:31:24 -08001003 struct buffer_head *bhs[CONCURRENT_JOURNAL_FILL];
1004
1005 mlog_entry_void();
1006
Mark Fashehccd979b2005-12-15 14:31:24 -08001007 memset(bhs, 0, sizeof(struct buffer_head *) * CONCURRENT_JOURNAL_FILL);
1008
Mark Fasheh8110b072007-03-22 16:53:23 -07001009 num_blocks = ocfs2_blocks_for_bytes(inode->i_sb, inode->i_size);
Mark Fashehccd979b2005-12-15 14:31:24 -08001010 v_blkno = 0;
Mark Fasheh8110b072007-03-22 16:53:23 -07001011 while (v_blkno < num_blocks) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001012 status = ocfs2_extent_map_get_blocks(inode, v_blkno,
Mark Fasheh49cb8d22007-03-09 16:21:46 -08001013 &p_blkno, &p_blocks, NULL);
Mark Fashehccd979b2005-12-15 14:31:24 -08001014 if (status < 0) {
1015 mlog_errno(status);
1016 goto bail;
1017 }
1018
1019 if (p_blocks > CONCURRENT_JOURNAL_FILL)
1020 p_blocks = CONCURRENT_JOURNAL_FILL;
1021
Mark Fashehdd4a2c22006-04-12 14:24:05 -07001022 /* We are reading journal data which should not
1023 * be put in the uptodate cache */
Joel Beckerda1e9092008-10-09 17:20:29 -07001024 status = ocfs2_read_blocks_sync(OCFS2_SB(inode->i_sb),
1025 p_blkno, p_blocks, bhs);
Mark Fashehccd979b2005-12-15 14:31:24 -08001026 if (status < 0) {
1027 mlog_errno(status);
1028 goto bail;
1029 }
1030
1031 for(i = 0; i < p_blocks; i++) {
1032 brelse(bhs[i]);
1033 bhs[i] = NULL;
1034 }
1035
1036 v_blkno += p_blocks;
1037 }
1038
1039bail:
1040 for(i = 0; i < CONCURRENT_JOURNAL_FILL; i++)
Mark Fasheha81cb882008-10-07 14:25:16 -07001041 brelse(bhs[i]);
Mark Fashehccd979b2005-12-15 14:31:24 -08001042 mlog_exit(status);
1043 return status;
1044}
1045
1046struct ocfs2_la_recovery_item {
1047 struct list_head lri_list;
1048 int lri_slot;
1049 struct ocfs2_dinode *lri_la_dinode;
1050 struct ocfs2_dinode *lri_tl_dinode;
Jan Kara22053632008-10-20 23:50:38 +02001051 struct ocfs2_quota_recovery *lri_qrec;
Mark Fashehccd979b2005-12-15 14:31:24 -08001052};
1053
1054/* Does the second half of the recovery process. By this point, the
1055 * node is marked clean and can actually be considered recovered,
1056 * hence it's no longer in the recovery map, but there's still some
1057 * cleanup we can do which shouldn't happen within the recovery thread
1058 * as locking in that context becomes very difficult if we are to take
1059 * recovering nodes into account.
1060 *
1061 * NOTE: This function can and will sleep on recovery of other nodes
1062 * during cluster locking, just like any other ocfs2 process.
1063 */
David Howellsc4028952006-11-22 14:57:56 +00001064void ocfs2_complete_recovery(struct work_struct *work)
Mark Fashehccd979b2005-12-15 14:31:24 -08001065{
1066 int ret;
David Howellsc4028952006-11-22 14:57:56 +00001067 struct ocfs2_journal *journal =
1068 container_of(work, struct ocfs2_journal, j_recovery_work);
1069 struct ocfs2_super *osb = journal->j_osb;
Mark Fashehccd979b2005-12-15 14:31:24 -08001070 struct ocfs2_dinode *la_dinode, *tl_dinode;
Christoph Hellwig800deef2007-05-17 16:03:13 +02001071 struct ocfs2_la_recovery_item *item, *n;
Jan Kara22053632008-10-20 23:50:38 +02001072 struct ocfs2_quota_recovery *qrec;
Mark Fashehccd979b2005-12-15 14:31:24 -08001073 LIST_HEAD(tmp_la_list);
1074
1075 mlog_entry_void();
1076
1077 mlog(0, "completing recovery from keventd\n");
1078
1079 spin_lock(&journal->j_lock);
1080 list_splice_init(&journal->j_la_cleanups, &tmp_la_list);
1081 spin_unlock(&journal->j_lock);
1082
Christoph Hellwig800deef2007-05-17 16:03:13 +02001083 list_for_each_entry_safe(item, n, &tmp_la_list, lri_list) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001084 list_del_init(&item->lri_list);
1085
1086 mlog(0, "Complete recovery for slot %d\n", item->lri_slot);
1087
Jan Kara19ece542008-08-21 20:13:17 +02001088 ocfs2_wait_on_quotas(osb);
1089
Mark Fashehccd979b2005-12-15 14:31:24 -08001090 la_dinode = item->lri_la_dinode;
1091 if (la_dinode) {
Mark Fashehb0697052006-03-03 10:24:33 -08001092 mlog(0, "Clean up local alloc %llu\n",
Mark Fasheh1ca1a112007-04-27 16:01:25 -07001093 (unsigned long long)le64_to_cpu(la_dinode->i_blkno));
Mark Fashehccd979b2005-12-15 14:31:24 -08001094
1095 ret = ocfs2_complete_local_alloc_recovery(osb,
1096 la_dinode);
1097 if (ret < 0)
1098 mlog_errno(ret);
1099
1100 kfree(la_dinode);
1101 }
1102
1103 tl_dinode = item->lri_tl_dinode;
1104 if (tl_dinode) {
Mark Fashehb0697052006-03-03 10:24:33 -08001105 mlog(0, "Clean up truncate log %llu\n",
Mark Fasheh1ca1a112007-04-27 16:01:25 -07001106 (unsigned long long)le64_to_cpu(tl_dinode->i_blkno));
Mark Fashehccd979b2005-12-15 14:31:24 -08001107
1108 ret = ocfs2_complete_truncate_log_recovery(osb,
1109 tl_dinode);
1110 if (ret < 0)
1111 mlog_errno(ret);
1112
1113 kfree(tl_dinode);
1114 }
1115
1116 ret = ocfs2_recover_orphans(osb, item->lri_slot);
1117 if (ret < 0)
1118 mlog_errno(ret);
1119
Jan Kara22053632008-10-20 23:50:38 +02001120 qrec = item->lri_qrec;
1121 if (qrec) {
1122 mlog(0, "Recovering quota files");
1123 ret = ocfs2_finish_quota_recovery(osb, qrec,
1124 item->lri_slot);
1125 if (ret < 0)
1126 mlog_errno(ret);
1127 /* Recovery info is already freed now */
1128 }
1129
Mark Fashehccd979b2005-12-15 14:31:24 -08001130 kfree(item);
1131 }
1132
1133 mlog(0, "Recovery completion\n");
1134 mlog_exit_void();
1135}
1136
1137/* NOTE: This function always eats your references to la_dinode and
1138 * tl_dinode, either manually on error, or by passing them to
1139 * ocfs2_complete_recovery */
1140static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal,
1141 int slot_num,
1142 struct ocfs2_dinode *la_dinode,
Jan Kara22053632008-10-20 23:50:38 +02001143 struct ocfs2_dinode *tl_dinode,
1144 struct ocfs2_quota_recovery *qrec)
Mark Fashehccd979b2005-12-15 14:31:24 -08001145{
1146 struct ocfs2_la_recovery_item *item;
1147
Sunil Mushranafae00ab2006-04-12 14:37:00 -07001148 item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS);
Mark Fashehccd979b2005-12-15 14:31:24 -08001149 if (!item) {
1150 /* Though we wish to avoid it, we are in fact safe in
1151 * skipping local alloc cleanup as fsck.ocfs2 is more
1152 * than capable of reclaiming unused space. */
1153 if (la_dinode)
1154 kfree(la_dinode);
1155
1156 if (tl_dinode)
1157 kfree(tl_dinode);
1158
Jan Kara22053632008-10-20 23:50:38 +02001159 if (qrec)
1160 ocfs2_free_quota_recovery(qrec);
1161
Mark Fashehccd979b2005-12-15 14:31:24 -08001162 mlog_errno(-ENOMEM);
1163 return;
1164 }
1165
1166 INIT_LIST_HEAD(&item->lri_list);
1167 item->lri_la_dinode = la_dinode;
1168 item->lri_slot = slot_num;
1169 item->lri_tl_dinode = tl_dinode;
Jan Kara22053632008-10-20 23:50:38 +02001170 item->lri_qrec = qrec;
Mark Fashehccd979b2005-12-15 14:31:24 -08001171
1172 spin_lock(&journal->j_lock);
1173 list_add_tail(&item->lri_list, &journal->j_la_cleanups);
1174 queue_work(ocfs2_wq, &journal->j_recovery_work);
1175 spin_unlock(&journal->j_lock);
1176}
1177
1178/* Called by the mount code to queue recovery the last part of
1179 * recovery for it's own slot. */
1180void ocfs2_complete_mount_recovery(struct ocfs2_super *osb)
1181{
1182 struct ocfs2_journal *journal = osb->journal;
1183
1184 if (osb->dirty) {
1185 /* No need to queue up our truncate_log as regular
1186 * cleanup will catch that. */
1187 ocfs2_queue_recovery_completion(journal,
1188 osb->slot_num,
1189 osb->local_alloc_copy,
Jan Kara22053632008-10-20 23:50:38 +02001190 NULL,
Mark Fashehccd979b2005-12-15 14:31:24 -08001191 NULL);
1192 ocfs2_schedule_truncate_log_flush(osb, 0);
1193
1194 osb->local_alloc_copy = NULL;
1195 osb->dirty = 0;
1196 }
1197}
1198
Jan Kara22053632008-10-20 23:50:38 +02001199void ocfs2_complete_quota_recovery(struct ocfs2_super *osb)
1200{
1201 if (osb->quota_rec) {
1202 ocfs2_queue_recovery_completion(osb->journal,
1203 osb->slot_num,
1204 NULL,
1205 NULL,
1206 osb->quota_rec);
1207 osb->quota_rec = NULL;
1208 }
1209}
1210
Mark Fashehccd979b2005-12-15 14:31:24 -08001211static int __ocfs2_recovery_thread(void *arg)
1212{
Jan Kara22053632008-10-20 23:50:38 +02001213 int status, node_num, slot_num;
Mark Fashehccd979b2005-12-15 14:31:24 -08001214 struct ocfs2_super *osb = arg;
Joel Becker553abd02008-02-01 12:03:57 -08001215 struct ocfs2_recovery_map *rm = osb->recovery_map;
Jan Kara22053632008-10-20 23:50:38 +02001216 int *rm_quota = NULL;
1217 int rm_quota_used = 0, i;
1218 struct ocfs2_quota_recovery *qrec;
Mark Fashehccd979b2005-12-15 14:31:24 -08001219
1220 mlog_entry_void();
1221
1222 status = ocfs2_wait_on_mount(osb);
1223 if (status < 0) {
1224 goto bail;
1225 }
1226
Jan Kara22053632008-10-20 23:50:38 +02001227 rm_quota = kzalloc(osb->max_slots * sizeof(int), GFP_NOFS);
1228 if (!rm_quota) {
1229 status = -ENOMEM;
1230 goto bail;
1231 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001232restart:
1233 status = ocfs2_super_lock(osb, 1);
1234 if (status < 0) {
1235 mlog_errno(status);
1236 goto bail;
1237 }
1238
Joel Becker553abd02008-02-01 12:03:57 -08001239 spin_lock(&osb->osb_lock);
1240 while (rm->rm_used) {
1241 /* It's always safe to remove entry zero, as we won't
1242 * clear it until ocfs2_recover_node() has succeeded. */
1243 node_num = rm->rm_entries[0];
1244 spin_unlock(&osb->osb_lock);
Jan Kara22053632008-10-20 23:50:38 +02001245 mlog(0, "checking node %d\n", node_num);
1246 slot_num = ocfs2_node_num_to_slot(osb, node_num);
1247 if (slot_num == -ENOENT) {
1248 status = 0;
1249 mlog(0, "no slot for this node, so no recovery"
1250 "required.\n");
1251 goto skip_recovery;
1252 }
1253 mlog(0, "node %d was using slot %d\n", node_num, slot_num);
Mark Fashehccd979b2005-12-15 14:31:24 -08001254
Jan Kara22053632008-10-20 23:50:38 +02001255 /* It is a bit subtle with quota recovery. We cannot do it
1256 * immediately because we have to obtain cluster locks from
1257 * quota files and we also don't want to just skip it because
1258 * then quota usage would be out of sync until some node takes
1259 * the slot. So we remember which nodes need quota recovery
1260 * and when everything else is done, we recover quotas. */
1261 for (i = 0; i < rm_quota_used && rm_quota[i] != slot_num; i++);
1262 if (i == rm_quota_used)
1263 rm_quota[rm_quota_used++] = slot_num;
1264
1265 status = ocfs2_recover_node(osb, node_num, slot_num);
1266skip_recovery:
Joel Becker553abd02008-02-01 12:03:57 -08001267 if (!status) {
1268 ocfs2_recovery_map_clear(osb, node_num);
1269 } else {
Mark Fashehccd979b2005-12-15 14:31:24 -08001270 mlog(ML_ERROR,
1271 "Error %d recovering node %d on device (%u,%u)!\n",
1272 status, node_num,
1273 MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1274 mlog(ML_ERROR, "Volume requires unmount.\n");
Mark Fashehccd979b2005-12-15 14:31:24 -08001275 }
1276
Joel Becker553abd02008-02-01 12:03:57 -08001277 spin_lock(&osb->osb_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -08001278 }
Joel Becker553abd02008-02-01 12:03:57 -08001279 spin_unlock(&osb->osb_lock);
1280 mlog(0, "All nodes recovered\n");
1281
Sunil Mushran539d8262008-07-14 17:31:10 -07001282 /* Refresh all journal recovery generations from disk */
1283 status = ocfs2_check_journals_nolocks(osb);
1284 status = (status == -EROFS) ? 0 : status;
1285 if (status < 0)
1286 mlog_errno(status);
1287
Jan Kara22053632008-10-20 23:50:38 +02001288 /* Now it is right time to recover quotas... We have to do this under
1289 * superblock lock so that noone can start using the slot (and crash)
1290 * before we recover it */
1291 for (i = 0; i < rm_quota_used; i++) {
1292 qrec = ocfs2_begin_quota_recovery(osb, rm_quota[i]);
1293 if (IS_ERR(qrec)) {
1294 status = PTR_ERR(qrec);
1295 mlog_errno(status);
1296 continue;
1297 }
1298 ocfs2_queue_recovery_completion(osb->journal, rm_quota[i],
1299 NULL, NULL, qrec);
1300 }
1301
Mark Fashehccd979b2005-12-15 14:31:24 -08001302 ocfs2_super_unlock(osb, 1);
1303
1304 /* We always run recovery on our own orphan dir - the dead
Mark Fasheh34d024f2007-09-24 15:56:19 -07001305 * node(s) may have disallowd a previos inode delete. Re-processing
1306 * is therefore required. */
Mark Fashehccd979b2005-12-15 14:31:24 -08001307 ocfs2_queue_recovery_completion(osb->journal, osb->slot_num, NULL,
Jan Kara22053632008-10-20 23:50:38 +02001308 NULL, NULL);
Mark Fashehccd979b2005-12-15 14:31:24 -08001309
1310bail:
Arjan van de Venc74ec2f2006-01-13 21:54:23 -08001311 mutex_lock(&osb->recovery_lock);
Joel Becker553abd02008-02-01 12:03:57 -08001312 if (!status && !ocfs2_recovery_completed(osb)) {
Arjan van de Venc74ec2f2006-01-13 21:54:23 -08001313 mutex_unlock(&osb->recovery_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -08001314 goto restart;
1315 }
1316
1317 osb->recovery_thread_task = NULL;
1318 mb(); /* sync with ocfs2_recovery_thread_running */
1319 wake_up(&osb->recovery_event);
1320
Arjan van de Venc74ec2f2006-01-13 21:54:23 -08001321 mutex_unlock(&osb->recovery_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -08001322
Jan Kara22053632008-10-20 23:50:38 +02001323 if (rm_quota)
1324 kfree(rm_quota);
1325
Mark Fashehccd979b2005-12-15 14:31:24 -08001326 mlog_exit(status);
1327 /* no one is callint kthread_stop() for us so the kthread() api
1328 * requires that we call do_exit(). And it isn't exported, but
1329 * complete_and_exit() seems to be a minimal wrapper around it. */
1330 complete_and_exit(NULL, status);
1331 return status;
1332}
1333
1334void ocfs2_recovery_thread(struct ocfs2_super *osb, int node_num)
1335{
1336 mlog_entry("(node_num=%d, osb->node_num = %d)\n",
1337 node_num, osb->node_num);
1338
Arjan van de Venc74ec2f2006-01-13 21:54:23 -08001339 mutex_lock(&osb->recovery_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -08001340 if (osb->disable_recovery)
1341 goto out;
1342
1343 /* People waiting on recovery will wait on
1344 * the recovery map to empty. */
Joel Becker553abd02008-02-01 12:03:57 -08001345 if (ocfs2_recovery_map_set(osb, node_num))
1346 mlog(0, "node %d already in recovery map.\n", node_num);
Mark Fashehccd979b2005-12-15 14:31:24 -08001347
1348 mlog(0, "starting recovery thread...\n");
1349
1350 if (osb->recovery_thread_task)
1351 goto out;
1352
1353 osb->recovery_thread_task = kthread_run(__ocfs2_recovery_thread, osb,
Mark Fasheh78427042006-05-04 12:03:26 -07001354 "ocfs2rec");
Mark Fashehccd979b2005-12-15 14:31:24 -08001355 if (IS_ERR(osb->recovery_thread_task)) {
1356 mlog_errno((int)PTR_ERR(osb->recovery_thread_task));
1357 osb->recovery_thread_task = NULL;
1358 }
1359
1360out:
Arjan van de Venc74ec2f2006-01-13 21:54:23 -08001361 mutex_unlock(&osb->recovery_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -08001362 wake_up(&osb->recovery_event);
1363
1364 mlog_exit_void();
1365}
1366
Sunil Mushran539d8262008-07-14 17:31:10 -07001367static int ocfs2_read_journal_inode(struct ocfs2_super *osb,
1368 int slot_num,
1369 struct buffer_head **bh,
1370 struct inode **ret_inode)
1371{
1372 int status = -EACCES;
1373 struct inode *inode = NULL;
1374
1375 BUG_ON(slot_num >= osb->max_slots);
1376
1377 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
1378 slot_num);
1379 if (!inode || is_bad_inode(inode)) {
1380 mlog_errno(status);
1381 goto bail;
1382 }
1383 SET_INODE_JOURNAL(inode);
1384
Joel Beckerb657c952008-11-13 14:49:11 -08001385 status = ocfs2_read_inode_block_full(inode, bh, OCFS2_BH_IGNORE_CACHE);
Sunil Mushran539d8262008-07-14 17:31:10 -07001386 if (status < 0) {
1387 mlog_errno(status);
1388 goto bail;
1389 }
1390
1391 status = 0;
1392
1393bail:
1394 if (inode) {
1395 if (status || !ret_inode)
1396 iput(inode);
1397 else
1398 *ret_inode = inode;
1399 }
1400 return status;
1401}
1402
Mark Fashehccd979b2005-12-15 14:31:24 -08001403/* Does the actual journal replay and marks the journal inode as
1404 * clean. Will only replay if the journal inode is marked dirty. */
1405static int ocfs2_replay_journal(struct ocfs2_super *osb,
1406 int node_num,
1407 int slot_num)
1408{
1409 int status;
1410 int got_lock = 0;
1411 unsigned int flags;
1412 struct inode *inode = NULL;
1413 struct ocfs2_dinode *fe;
1414 journal_t *journal = NULL;
1415 struct buffer_head *bh = NULL;
Sunil Mushran539d8262008-07-14 17:31:10 -07001416 u32 slot_reco_gen;
Mark Fashehccd979b2005-12-15 14:31:24 -08001417
Sunil Mushran539d8262008-07-14 17:31:10 -07001418 status = ocfs2_read_journal_inode(osb, slot_num, &bh, &inode);
1419 if (status) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001420 mlog_errno(status);
1421 goto done;
1422 }
Sunil Mushran539d8262008-07-14 17:31:10 -07001423
1424 fe = (struct ocfs2_dinode *)bh->b_data;
1425 slot_reco_gen = ocfs2_get_recovery_generation(fe);
1426 brelse(bh);
1427 bh = NULL;
1428
1429 /*
1430 * As the fs recovery is asynchronous, there is a small chance that
1431 * another node mounted (and recovered) the slot before the recovery
1432 * thread could get the lock. To handle that, we dirty read the journal
1433 * inode for that slot to get the recovery generation. If it is
1434 * different than what we expected, the slot has been recovered.
1435 * If not, it needs recovery.
1436 */
1437 if (osb->slot_recovery_generations[slot_num] != slot_reco_gen) {
1438 mlog(0, "Slot %u already recovered (old/new=%u/%u)\n", slot_num,
1439 osb->slot_recovery_generations[slot_num], slot_reco_gen);
1440 osb->slot_recovery_generations[slot_num] = slot_reco_gen;
1441 status = -EBUSY;
Mark Fashehccd979b2005-12-15 14:31:24 -08001442 goto done;
1443 }
Sunil Mushran539d8262008-07-14 17:31:10 -07001444
1445 /* Continue with recovery as the journal has not yet been recovered */
Mark Fashehccd979b2005-12-15 14:31:24 -08001446
Mark Fashehe63aecb62007-10-18 15:30:42 -07001447 status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY);
Mark Fashehccd979b2005-12-15 14:31:24 -08001448 if (status < 0) {
Mark Fashehe63aecb62007-10-18 15:30:42 -07001449 mlog(0, "status returned from ocfs2_inode_lock=%d\n", status);
Mark Fashehccd979b2005-12-15 14:31:24 -08001450 if (status != -ERESTARTSYS)
1451 mlog(ML_ERROR, "Could not lock journal!\n");
1452 goto done;
1453 }
1454 got_lock = 1;
1455
1456 fe = (struct ocfs2_dinode *) bh->b_data;
1457
1458 flags = le32_to_cpu(fe->id1.journal1.ij_flags);
Sunil Mushran539d8262008-07-14 17:31:10 -07001459 slot_reco_gen = ocfs2_get_recovery_generation(fe);
Mark Fashehccd979b2005-12-15 14:31:24 -08001460
1461 if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) {
1462 mlog(0, "No recovery required for node %d\n", node_num);
Sunil Mushran539d8262008-07-14 17:31:10 -07001463 /* Refresh recovery generation for the slot */
1464 osb->slot_recovery_generations[slot_num] = slot_reco_gen;
Mark Fashehccd979b2005-12-15 14:31:24 -08001465 goto done;
1466 }
1467
1468 mlog(ML_NOTICE, "Recovering node %d from slot %d on device (%u,%u)\n",
1469 node_num, slot_num,
1470 MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1471
1472 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
1473
1474 status = ocfs2_force_read_journal(inode);
1475 if (status < 0) {
1476 mlog_errno(status);
1477 goto done;
1478 }
1479
1480 mlog(0, "calling journal_init_inode\n");
Joel Becker2b4e30f2008-09-03 20:03:41 -07001481 journal = jbd2_journal_init_inode(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -08001482 if (journal == NULL) {
1483 mlog(ML_ERROR, "Linux journal layer error\n");
1484 status = -EIO;
1485 goto done;
1486 }
1487
Joel Becker2b4e30f2008-09-03 20:03:41 -07001488 status = jbd2_journal_load(journal);
Mark Fashehccd979b2005-12-15 14:31:24 -08001489 if (status < 0) {
1490 mlog_errno(status);
1491 if (!igrab(inode))
1492 BUG();
Joel Becker2b4e30f2008-09-03 20:03:41 -07001493 jbd2_journal_destroy(journal);
Mark Fashehccd979b2005-12-15 14:31:24 -08001494 goto done;
1495 }
1496
1497 ocfs2_clear_journal_error(osb->sb, journal, slot_num);
1498
1499 /* wipe the journal */
1500 mlog(0, "flushing the journal.\n");
Joel Becker2b4e30f2008-09-03 20:03:41 -07001501 jbd2_journal_lock_updates(journal);
1502 status = jbd2_journal_flush(journal);
1503 jbd2_journal_unlock_updates(journal);
Mark Fashehccd979b2005-12-15 14:31:24 -08001504 if (status < 0)
1505 mlog_errno(status);
1506
1507 /* This will mark the node clean */
1508 flags = le32_to_cpu(fe->id1.journal1.ij_flags);
1509 flags &= ~OCFS2_JOURNAL_DIRTY_FL;
1510 fe->id1.journal1.ij_flags = cpu_to_le32(flags);
1511
Sunil Mushran539d8262008-07-14 17:31:10 -07001512 /* Increment recovery generation to indicate successful recovery */
1513 ocfs2_bump_recovery_generation(fe);
1514 osb->slot_recovery_generations[slot_num] =
1515 ocfs2_get_recovery_generation(fe);
1516
Joel Becker13723d02008-10-17 19:25:01 -07001517 ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check);
Mark Fashehccd979b2005-12-15 14:31:24 -08001518 status = ocfs2_write_block(osb, bh, inode);
1519 if (status < 0)
1520 mlog_errno(status);
1521
1522 if (!igrab(inode))
1523 BUG();
1524
Joel Becker2b4e30f2008-09-03 20:03:41 -07001525 jbd2_journal_destroy(journal);
Mark Fashehccd979b2005-12-15 14:31:24 -08001526
1527done:
1528 /* drop the lock on this nodes journal */
1529 if (got_lock)
Mark Fashehe63aecb62007-10-18 15:30:42 -07001530 ocfs2_inode_unlock(inode, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -08001531
1532 if (inode)
1533 iput(inode);
1534
Mark Fasheha81cb882008-10-07 14:25:16 -07001535 brelse(bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001536
1537 mlog_exit(status);
1538 return status;
1539}
1540
1541/*
1542 * Do the most important parts of node recovery:
1543 * - Replay it's journal
1544 * - Stamp a clean local allocator file
1545 * - Stamp a clean truncate log
1546 * - Mark the node clean
1547 *
1548 * If this function completes without error, a node in OCFS2 can be
1549 * said to have been safely recovered. As a result, failure during the
1550 * second part of a nodes recovery process (local alloc recovery) is
1551 * far less concerning.
1552 */
1553static int ocfs2_recover_node(struct ocfs2_super *osb,
Jan Kara22053632008-10-20 23:50:38 +02001554 int node_num, int slot_num)
Mark Fashehccd979b2005-12-15 14:31:24 -08001555{
1556 int status = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -08001557 struct ocfs2_dinode *la_copy = NULL;
1558 struct ocfs2_dinode *tl_copy = NULL;
1559
Jan Kara22053632008-10-20 23:50:38 +02001560 mlog_entry("(node_num=%d, slot_num=%d, osb->node_num = %d)\n",
1561 node_num, slot_num, osb->node_num);
Mark Fashehccd979b2005-12-15 14:31:24 -08001562
1563 /* Should not ever be called to recover ourselves -- in that
1564 * case we should've called ocfs2_journal_load instead. */
Eric Sesterhenn / snakebyteebdec832006-01-27 10:32:52 +01001565 BUG_ON(osb->node_num == node_num);
Mark Fashehccd979b2005-12-15 14:31:24 -08001566
Mark Fashehccd979b2005-12-15 14:31:24 -08001567 status = ocfs2_replay_journal(osb, node_num, slot_num);
1568 if (status < 0) {
Sunil Mushran539d8262008-07-14 17:31:10 -07001569 if (status == -EBUSY) {
1570 mlog(0, "Skipping recovery for slot %u (node %u) "
1571 "as another node has recovered it\n", slot_num,
1572 node_num);
1573 status = 0;
1574 goto done;
1575 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001576 mlog_errno(status);
1577 goto done;
1578 }
1579
1580 /* Stamp a clean local alloc file AFTER recovering the journal... */
1581 status = ocfs2_begin_local_alloc_recovery(osb, slot_num, &la_copy);
1582 if (status < 0) {
1583 mlog_errno(status);
1584 goto done;
1585 }
1586
1587 /* An error from begin_truncate_log_recovery is not
1588 * serious enough to warrant halting the rest of
1589 * recovery. */
1590 status = ocfs2_begin_truncate_log_recovery(osb, slot_num, &tl_copy);
1591 if (status < 0)
1592 mlog_errno(status);
1593
1594 /* Likewise, this would be a strange but ultimately not so
1595 * harmful place to get an error... */
Mark Fasheh8e8a4602008-02-01 11:59:09 -08001596 status = ocfs2_clear_slot(osb, slot_num);
Mark Fashehccd979b2005-12-15 14:31:24 -08001597 if (status < 0)
1598 mlog_errno(status);
1599
1600 /* This will kfree the memory pointed to by la_copy and tl_copy */
1601 ocfs2_queue_recovery_completion(osb->journal, slot_num, la_copy,
Jan Kara22053632008-10-20 23:50:38 +02001602 tl_copy, NULL);
Mark Fashehccd979b2005-12-15 14:31:24 -08001603
1604 status = 0;
1605done:
1606
1607 mlog_exit(status);
1608 return status;
1609}
1610
1611/* Test node liveness by trylocking his journal. If we get the lock,
1612 * we drop it here. Return 0 if we got the lock, -EAGAIN if node is
1613 * still alive (we couldn't get the lock) and < 0 on error. */
1614static int ocfs2_trylock_journal(struct ocfs2_super *osb,
1615 int slot_num)
1616{
1617 int status, flags;
1618 struct inode *inode = NULL;
1619
1620 inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
1621 slot_num);
1622 if (inode == NULL) {
1623 mlog(ML_ERROR, "access error\n");
1624 status = -EACCES;
1625 goto bail;
1626 }
1627 if (is_bad_inode(inode)) {
1628 mlog(ML_ERROR, "access error (bad inode)\n");
1629 iput(inode);
1630 inode = NULL;
1631 status = -EACCES;
1632 goto bail;
1633 }
1634 SET_INODE_JOURNAL(inode);
1635
1636 flags = OCFS2_META_LOCK_RECOVERY | OCFS2_META_LOCK_NOQUEUE;
Mark Fashehe63aecb62007-10-18 15:30:42 -07001637 status = ocfs2_inode_lock_full(inode, NULL, 1, flags);
Mark Fashehccd979b2005-12-15 14:31:24 -08001638 if (status < 0) {
1639 if (status != -EAGAIN)
1640 mlog_errno(status);
1641 goto bail;
1642 }
1643
Mark Fashehe63aecb62007-10-18 15:30:42 -07001644 ocfs2_inode_unlock(inode, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -08001645bail:
1646 if (inode)
1647 iput(inode);
1648
1649 return status;
1650}
1651
1652/* Call this underneath ocfs2_super_lock. It also assumes that the
1653 * slot info struct has been updated from disk. */
1654int ocfs2_mark_dead_nodes(struct ocfs2_super *osb)
1655{
Joel Beckerd85b20e2008-02-01 12:01:05 -08001656 unsigned int node_num;
1657 int status, i;
Mark Fasheha1af7d12008-08-19 17:20:28 -07001658 u32 gen;
Sunil Mushran539d8262008-07-14 17:31:10 -07001659 struct buffer_head *bh = NULL;
1660 struct ocfs2_dinode *di;
Mark Fashehccd979b2005-12-15 14:31:24 -08001661
1662 /* This is called with the super block cluster lock, so we
1663 * know that the slot map can't change underneath us. */
1664
Joel Beckerd85b20e2008-02-01 12:01:05 -08001665 for (i = 0; i < osb->max_slots; i++) {
Sunil Mushran539d8262008-07-14 17:31:10 -07001666 /* Read journal inode to get the recovery generation */
1667 status = ocfs2_read_journal_inode(osb, i, &bh, NULL);
1668 if (status) {
1669 mlog_errno(status);
1670 goto bail;
1671 }
1672 di = (struct ocfs2_dinode *)bh->b_data;
Mark Fasheha1af7d12008-08-19 17:20:28 -07001673 gen = ocfs2_get_recovery_generation(di);
Sunil Mushran539d8262008-07-14 17:31:10 -07001674 brelse(bh);
1675 bh = NULL;
1676
Mark Fasheha1af7d12008-08-19 17:20:28 -07001677 spin_lock(&osb->osb_lock);
1678 osb->slot_recovery_generations[i] = gen;
1679
Sunil Mushran539d8262008-07-14 17:31:10 -07001680 mlog(0, "Slot %u recovery generation is %u\n", i,
1681 osb->slot_recovery_generations[i]);
1682
Mark Fasheha1af7d12008-08-19 17:20:28 -07001683 if (i == osb->slot_num) {
1684 spin_unlock(&osb->osb_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -08001685 continue;
Mark Fasheha1af7d12008-08-19 17:20:28 -07001686 }
Joel Beckerd85b20e2008-02-01 12:01:05 -08001687
1688 status = ocfs2_slot_to_node_num_locked(osb, i, &node_num);
Mark Fasheha1af7d12008-08-19 17:20:28 -07001689 if (status == -ENOENT) {
1690 spin_unlock(&osb->osb_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -08001691 continue;
Mark Fasheha1af7d12008-08-19 17:20:28 -07001692 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001693
Mark Fasheha1af7d12008-08-19 17:20:28 -07001694 if (__ocfs2_recovery_map_test(osb, node_num)) {
1695 spin_unlock(&osb->osb_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -08001696 continue;
Mark Fasheha1af7d12008-08-19 17:20:28 -07001697 }
Joel Beckerd85b20e2008-02-01 12:01:05 -08001698 spin_unlock(&osb->osb_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -08001699
1700 /* Ok, we have a slot occupied by another node which
1701 * is not in the recovery map. We trylock his journal
1702 * file here to test if he's alive. */
1703 status = ocfs2_trylock_journal(osb, i);
1704 if (!status) {
1705 /* Since we're called from mount, we know that
1706 * the recovery thread can't race us on
1707 * setting / checking the recovery bits. */
1708 ocfs2_recovery_thread(osb, node_num);
1709 } else if ((status < 0) && (status != -EAGAIN)) {
1710 mlog_errno(status);
1711 goto bail;
1712 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001713 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001714
1715 status = 0;
1716bail:
1717 mlog_exit(status);
1718 return status;
1719}
1720
Mark Fasheh5eae5b92007-09-10 17:50:51 -07001721struct ocfs2_orphan_filldir_priv {
1722 struct inode *head;
1723 struct ocfs2_super *osb;
1724};
1725
1726static int ocfs2_orphan_filldir(void *priv, const char *name, int name_len,
1727 loff_t pos, u64 ino, unsigned type)
1728{
1729 struct ocfs2_orphan_filldir_priv *p = priv;
1730 struct inode *iter;
1731
1732 if (name_len == 1 && !strncmp(".", name, 1))
1733 return 0;
1734 if (name_len == 2 && !strncmp("..", name, 2))
1735 return 0;
1736
1737 /* Skip bad inodes so that recovery can continue */
1738 iter = ocfs2_iget(p->osb, ino,
Jan Kara5fa06132008-01-11 00:11:45 +01001739 OCFS2_FI_FLAG_ORPHAN_RECOVERY, 0);
Mark Fasheh5eae5b92007-09-10 17:50:51 -07001740 if (IS_ERR(iter))
1741 return 0;
1742
1743 mlog(0, "queue orphan %llu\n",
1744 (unsigned long long)OCFS2_I(iter)->ip_blkno);
1745 /* No locking is required for the next_orphan queue as there
1746 * is only ever a single process doing orphan recovery. */
1747 OCFS2_I(iter)->ip_next_orphan = p->head;
1748 p->head = iter;
1749
1750 return 0;
1751}
1752
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001753static int ocfs2_queue_orphans(struct ocfs2_super *osb,
1754 int slot,
1755 struct inode **head)
Mark Fashehccd979b2005-12-15 14:31:24 -08001756{
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001757 int status;
Mark Fashehccd979b2005-12-15 14:31:24 -08001758 struct inode *orphan_dir_inode = NULL;
Mark Fasheh5eae5b92007-09-10 17:50:51 -07001759 struct ocfs2_orphan_filldir_priv priv;
1760 loff_t pos = 0;
1761
1762 priv.osb = osb;
1763 priv.head = *head;
Mark Fashehccd979b2005-12-15 14:31:24 -08001764
1765 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
1766 ORPHAN_DIR_SYSTEM_INODE,
1767 slot);
1768 if (!orphan_dir_inode) {
1769 status = -ENOENT;
1770 mlog_errno(status);
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001771 return status;
1772 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001773
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001774 mutex_lock(&orphan_dir_inode->i_mutex);
Mark Fashehe63aecb62007-10-18 15:30:42 -07001775 status = ocfs2_inode_lock(orphan_dir_inode, NULL, 0);
Mark Fashehccd979b2005-12-15 14:31:24 -08001776 if (status < 0) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001777 mlog_errno(status);
1778 goto out;
1779 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001780
Mark Fasheh5eae5b92007-09-10 17:50:51 -07001781 status = ocfs2_dir_foreach(orphan_dir_inode, &pos, &priv,
1782 ocfs2_orphan_filldir);
1783 if (status) {
1784 mlog_errno(status);
Mark Fasheha86370f2007-12-03 14:06:23 -08001785 goto out_cluster;
Mark Fashehccd979b2005-12-15 14:31:24 -08001786 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001787
Mark Fasheh5eae5b92007-09-10 17:50:51 -07001788 *head = priv.head;
1789
Mark Fasheha86370f2007-12-03 14:06:23 -08001790out_cluster:
Mark Fashehe63aecb62007-10-18 15:30:42 -07001791 ocfs2_inode_unlock(orphan_dir_inode, 0);
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001792out:
1793 mutex_unlock(&orphan_dir_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -08001794 iput(orphan_dir_inode);
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001795 return status;
1796}
1797
1798static int ocfs2_orphan_recovery_can_continue(struct ocfs2_super *osb,
1799 int slot)
1800{
1801 int ret;
1802
1803 spin_lock(&osb->osb_lock);
1804 ret = !osb->osb_orphan_wipes[slot];
1805 spin_unlock(&osb->osb_lock);
1806 return ret;
1807}
1808
1809static void ocfs2_mark_recovering_orphan_dir(struct ocfs2_super *osb,
1810 int slot)
1811{
1812 spin_lock(&osb->osb_lock);
1813 /* Mark ourselves such that new processes in delete_inode()
1814 * know to quit early. */
1815 ocfs2_node_map_set_bit(osb, &osb->osb_recovering_orphan_dirs, slot);
1816 while (osb->osb_orphan_wipes[slot]) {
1817 /* If any processes are already in the middle of an
1818 * orphan wipe on this dir, then we need to wait for
1819 * them. */
1820 spin_unlock(&osb->osb_lock);
1821 wait_event_interruptible(osb->osb_wipe_event,
1822 ocfs2_orphan_recovery_can_continue(osb, slot));
1823 spin_lock(&osb->osb_lock);
1824 }
1825 spin_unlock(&osb->osb_lock);
1826}
1827
1828static void ocfs2_clear_recovering_orphan_dir(struct ocfs2_super *osb,
1829 int slot)
1830{
1831 ocfs2_node_map_clear_bit(osb, &osb->osb_recovering_orphan_dirs, slot);
1832}
1833
1834/*
1835 * Orphan recovery. Each mounted node has it's own orphan dir which we
1836 * must run during recovery. Our strategy here is to build a list of
1837 * the inodes in the orphan dir and iget/iput them. The VFS does
1838 * (most) of the rest of the work.
1839 *
1840 * Orphan recovery can happen at any time, not just mount so we have a
1841 * couple of extra considerations.
1842 *
1843 * - We grab as many inodes as we can under the orphan dir lock -
1844 * doing iget() outside the orphan dir risks getting a reference on
1845 * an invalid inode.
1846 * - We must be sure not to deadlock with other processes on the
1847 * system wanting to run delete_inode(). This can happen when they go
1848 * to lock the orphan dir and the orphan recovery process attempts to
1849 * iget() inside the orphan dir lock. This can be avoided by
1850 * advertising our state to ocfs2_delete_inode().
1851 */
1852static int ocfs2_recover_orphans(struct ocfs2_super *osb,
1853 int slot)
1854{
1855 int ret = 0;
1856 struct inode *inode = NULL;
1857 struct inode *iter;
1858 struct ocfs2_inode_info *oi;
1859
1860 mlog(0, "Recover inodes from orphan dir in slot %d\n", slot);
1861
1862 ocfs2_mark_recovering_orphan_dir(osb, slot);
1863 ret = ocfs2_queue_orphans(osb, slot, &inode);
1864 ocfs2_clear_recovering_orphan_dir(osb, slot);
1865
1866 /* Error here should be noted, but we want to continue with as
1867 * many queued inodes as we've got. */
1868 if (ret)
1869 mlog_errno(ret);
Mark Fashehccd979b2005-12-15 14:31:24 -08001870
1871 while (inode) {
1872 oi = OCFS2_I(inode);
Mark Fashehb0697052006-03-03 10:24:33 -08001873 mlog(0, "iput orphan %llu\n", (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001874
1875 iter = oi->ip_next_orphan;
1876
1877 spin_lock(&oi->ip_lock);
Mark Fasheh34d024f2007-09-24 15:56:19 -07001878 /* The remote delete code may have set these on the
1879 * assumption that the other node would wipe them
1880 * successfully. If they are still in the node's
1881 * orphan dir, we need to reset that state. */
Mark Fashehccd979b2005-12-15 14:31:24 -08001882 oi->ip_flags &= ~(OCFS2_INODE_DELETED|OCFS2_INODE_SKIP_DELETE);
1883
1884 /* Set the proper information to get us going into
1885 * ocfs2_delete_inode. */
1886 oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED;
Mark Fashehccd979b2005-12-15 14:31:24 -08001887 spin_unlock(&oi->ip_lock);
1888
1889 iput(inode);
1890
1891 inode = iter;
1892 }
1893
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001894 return ret;
Mark Fashehccd979b2005-12-15 14:31:24 -08001895}
1896
Jan Kara19ece542008-08-21 20:13:17 +02001897static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota)
Mark Fashehccd979b2005-12-15 14:31:24 -08001898{
1899 /* This check is good because ocfs2 will wait on our recovery
1900 * thread before changing it to something other than MOUNTED
1901 * or DISABLED. */
1902 wait_event(osb->osb_mount_event,
Jan Kara19ece542008-08-21 20:13:17 +02001903 (!quota && atomic_read(&osb->vol_state) == VOLUME_MOUNTED) ||
1904 atomic_read(&osb->vol_state) == VOLUME_MOUNTED_QUOTAS ||
Mark Fashehccd979b2005-12-15 14:31:24 -08001905 atomic_read(&osb->vol_state) == VOLUME_DISABLED);
1906
1907 /* If there's an error on mount, then we may never get to the
1908 * MOUNTED flag, but this is set right before
1909 * dismount_volume() so we can trust it. */
1910 if (atomic_read(&osb->vol_state) == VOLUME_DISABLED) {
1911 mlog(0, "mount error, exiting!\n");
1912 return -EBUSY;
1913 }
1914
1915 return 0;
1916}
1917
1918static int ocfs2_commit_thread(void *arg)
1919{
1920 int status;
1921 struct ocfs2_super *osb = arg;
1922 struct ocfs2_journal *journal = osb->journal;
1923
1924 /* we can trust j_num_trans here because _should_stop() is only set in
1925 * shutdown and nobody other than ourselves should be able to start
1926 * transactions. committing on shutdown might take a few iterations
1927 * as final transactions put deleted inodes on the list */
1928 while (!(kthread_should_stop() &&
1929 atomic_read(&journal->j_num_trans) == 0)) {
1930
Mark Fasheh745ae8ba2006-02-09 13:23:39 -08001931 wait_event_interruptible(osb->checkpoint_event,
1932 atomic_read(&journal->j_num_trans)
1933 || kthread_should_stop());
Mark Fashehccd979b2005-12-15 14:31:24 -08001934
1935 status = ocfs2_commit_cache(osb);
1936 if (status < 0)
1937 mlog_errno(status);
1938
1939 if (kthread_should_stop() && atomic_read(&journal->j_num_trans)){
1940 mlog(ML_KTHREAD,
1941 "commit_thread: %u transactions pending on "
1942 "shutdown\n",
1943 atomic_read(&journal->j_num_trans));
1944 }
1945 }
1946
1947 return 0;
1948}
1949
Sunil Mushran539d8262008-07-14 17:31:10 -07001950/* Reads all the journal inodes without taking any cluster locks. Used
1951 * for hard readonly access to determine whether any journal requires
1952 * recovery. Also used to refresh the recovery generation numbers after
1953 * a journal has been recovered by another node.
1954 */
Mark Fashehccd979b2005-12-15 14:31:24 -08001955int ocfs2_check_journals_nolocks(struct ocfs2_super *osb)
1956{
1957 int ret = 0;
1958 unsigned int slot;
Sunil Mushran539d8262008-07-14 17:31:10 -07001959 struct buffer_head *di_bh = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -08001960 struct ocfs2_dinode *di;
Sunil Mushran539d8262008-07-14 17:31:10 -07001961 int journal_dirty = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -08001962
1963 for(slot = 0; slot < osb->max_slots; slot++) {
Sunil Mushran539d8262008-07-14 17:31:10 -07001964 ret = ocfs2_read_journal_inode(osb, slot, &di_bh, NULL);
1965 if (ret) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001966 mlog_errno(ret);
1967 goto out;
1968 }
1969
1970 di = (struct ocfs2_dinode *) di_bh->b_data;
1971
Sunil Mushran539d8262008-07-14 17:31:10 -07001972 osb->slot_recovery_generations[slot] =
1973 ocfs2_get_recovery_generation(di);
1974
Mark Fashehccd979b2005-12-15 14:31:24 -08001975 if (le32_to_cpu(di->id1.journal1.ij_flags) &
1976 OCFS2_JOURNAL_DIRTY_FL)
Sunil Mushran539d8262008-07-14 17:31:10 -07001977 journal_dirty = 1;
Mark Fashehccd979b2005-12-15 14:31:24 -08001978
1979 brelse(di_bh);
Sunil Mushran539d8262008-07-14 17:31:10 -07001980 di_bh = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -08001981 }
1982
1983out:
Sunil Mushran539d8262008-07-14 17:31:10 -07001984 if (journal_dirty)
1985 ret = -EROFS;
Mark Fashehccd979b2005-12-15 14:31:24 -08001986 return ret;
1987}