Root/fs/ocfs2/journal.c

1/* -*- 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#include <linux/time.h>
32#include <linux/random.h>
33
34#define MLOG_MASK_PREFIX ML_JOURNAL
35#include <cluster/masklog.h>
36
37#include "ocfs2.h"
38
39#include "alloc.h"
40#include "blockcheck.h"
41#include "dir.h"
42#include "dlmglue.h"
43#include "extent_map.h"
44#include "heartbeat.h"
45#include "inode.h"
46#include "journal.h"
47#include "localalloc.h"
48#include "slot_map.h"
49#include "super.h"
50#include "sysfile.h"
51#include "uptodate.h"
52#include "quota.h"
53
54#include "buffer_head_io.h"
55
56DEFINE_SPINLOCK(trans_inc_lock);
57
58#define ORPHAN_SCAN_SCHEDULE_TIMEOUT 300000
59
60static int ocfs2_force_read_journal(struct inode *inode);
61static int ocfs2_recover_node(struct ocfs2_super *osb,
62                  int node_num, int slot_num);
63static int __ocfs2_recovery_thread(void *arg);
64static int ocfs2_commit_cache(struct ocfs2_super *osb);
65static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota);
66static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
67                      int dirty, int replayed);
68static int ocfs2_trylock_journal(struct ocfs2_super *osb,
69                 int slot_num);
70static int ocfs2_recover_orphans(struct ocfs2_super *osb,
71                 int slot);
72static int ocfs2_commit_thread(void *arg);
73static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal,
74                        int slot_num,
75                        struct ocfs2_dinode *la_dinode,
76                        struct ocfs2_dinode *tl_dinode,
77                        struct ocfs2_quota_recovery *qrec);
78
79static inline int ocfs2_wait_on_mount(struct ocfs2_super *osb)
80{
81    return __ocfs2_wait_on_mount(osb, 0);
82}
83
84static inline int ocfs2_wait_on_quotas(struct ocfs2_super *osb)
85{
86    return __ocfs2_wait_on_mount(osb, 1);
87}
88
89/*
90 * This replay_map is to track online/offline slots, so we could recover
91 * offline slots during recovery and mount
92 */
93
94enum ocfs2_replay_state {
95    REPLAY_UNNEEDED = 0, /* Replay is not needed, so ignore this map */
96    REPLAY_NEEDED, /* Replay slots marked in rm_replay_slots */
97    REPLAY_DONE /* Replay was already queued */
98};
99
100struct ocfs2_replay_map {
101    unsigned int rm_slots;
102    enum ocfs2_replay_state rm_state;
103    unsigned char rm_replay_slots[0];
104};
105
106void ocfs2_replay_map_set_state(struct ocfs2_super *osb, int state)
107{
108    if (!osb->replay_map)
109        return;
110
111    /* If we've already queued the replay, we don't have any more to do */
112    if (osb->replay_map->rm_state == REPLAY_DONE)
113        return;
114
115    osb->replay_map->rm_state = state;
116}
117
118int ocfs2_compute_replay_slots(struct ocfs2_super *osb)
119{
120    struct ocfs2_replay_map *replay_map;
121    int i, node_num;
122
123    /* If replay map is already set, we don't do it again */
124    if (osb->replay_map)
125        return 0;
126
127    replay_map = kzalloc(sizeof(struct ocfs2_replay_map) +
128                 (osb->max_slots * sizeof(char)), GFP_KERNEL);
129
130    if (!replay_map) {
131        mlog_errno(-ENOMEM);
132        return -ENOMEM;
133    }
134
135    spin_lock(&osb->osb_lock);
136
137    replay_map->rm_slots = osb->max_slots;
138    replay_map->rm_state = REPLAY_UNNEEDED;
139
140    /* set rm_replay_slots for offline slot(s) */
141    for (i = 0; i < replay_map->rm_slots; i++) {
142        if (ocfs2_slot_to_node_num_locked(osb, i, &node_num) == -ENOENT)
143            replay_map->rm_replay_slots[i] = 1;
144    }
145
146    osb->replay_map = replay_map;
147    spin_unlock(&osb->osb_lock);
148    return 0;
149}
150
151void ocfs2_queue_replay_slots(struct ocfs2_super *osb)
152{
153    struct ocfs2_replay_map *replay_map = osb->replay_map;
154    int i;
155
156    if (!replay_map)
157        return;
158
159    if (replay_map->rm_state != REPLAY_NEEDED)
160        return;
161
162    for (i = 0; i < replay_map->rm_slots; i++)
163        if (replay_map->rm_replay_slots[i])
164            ocfs2_queue_recovery_completion(osb->journal, i, NULL,
165                            NULL, NULL);
166    replay_map->rm_state = REPLAY_DONE;
167}
168
169void ocfs2_free_replay_slots(struct ocfs2_super *osb)
170{
171    struct ocfs2_replay_map *replay_map = osb->replay_map;
172
173    if (!osb->replay_map)
174        return;
175
176    kfree(replay_map);
177    osb->replay_map = NULL;
178}
179
180int ocfs2_recovery_init(struct ocfs2_super *osb)
181{
182    struct ocfs2_recovery_map *rm;
183
184    mutex_init(&osb->recovery_lock);
185    osb->disable_recovery = 0;
186    osb->recovery_thread_task = NULL;
187    init_waitqueue_head(&osb->recovery_event);
188
189    rm = kzalloc(sizeof(struct ocfs2_recovery_map) +
190             osb->max_slots * sizeof(unsigned int),
191             GFP_KERNEL);
192    if (!rm) {
193        mlog_errno(-ENOMEM);
194        return -ENOMEM;
195    }
196
197    rm->rm_entries = (unsigned int *)((char *)rm +
198                      sizeof(struct ocfs2_recovery_map));
199    osb->recovery_map = rm;
200
201    return 0;
202}
203
204/* we can't grab the goofy sem lock from inside wait_event, so we use
205 * memory barriers to make sure that we'll see the null task before
206 * being woken up */
207static int ocfs2_recovery_thread_running(struct ocfs2_super *osb)
208{
209    mb();
210    return osb->recovery_thread_task != NULL;
211}
212
213void ocfs2_recovery_exit(struct ocfs2_super *osb)
214{
215    struct ocfs2_recovery_map *rm;
216
217    /* disable any new recovery threads and wait for any currently
218     * running ones to exit. Do this before setting the vol_state. */
219    mutex_lock(&osb->recovery_lock);
220    osb->disable_recovery = 1;
221    mutex_unlock(&osb->recovery_lock);
222    wait_event(osb->recovery_event, !ocfs2_recovery_thread_running(osb));
223
224    /* At this point, we know that no more recovery threads can be
225     * launched, so wait for any recovery completion work to
226     * complete. */
227    flush_workqueue(ocfs2_wq);
228
229    /*
230     * Now that recovery is shut down, and the osb is about to be
231     * freed, the osb_lock is not taken here.
232     */
233    rm = osb->recovery_map;
234    /* XXX: Should we bug if there are dirty entries? */
235
236    kfree(rm);
237}
238
239static int __ocfs2_recovery_map_test(struct ocfs2_super *osb,
240                     unsigned int node_num)
241{
242    int i;
243    struct ocfs2_recovery_map *rm = osb->recovery_map;
244
245    assert_spin_locked(&osb->osb_lock);
246
247    for (i = 0; i < rm->rm_used; i++) {
248        if (rm->rm_entries[i] == node_num)
249            return 1;
250    }
251
252    return 0;
253}
254
255/* Behaves like test-and-set. Returns the previous value */
256static int ocfs2_recovery_map_set(struct ocfs2_super *osb,
257                  unsigned int node_num)
258{
259    struct ocfs2_recovery_map *rm = osb->recovery_map;
260
261    spin_lock(&osb->osb_lock);
262    if (__ocfs2_recovery_map_test(osb, node_num)) {
263        spin_unlock(&osb->osb_lock);
264        return 1;
265    }
266
267    /* XXX: Can this be exploited? Not from o2dlm... */
268    BUG_ON(rm->rm_used >= osb->max_slots);
269
270    rm->rm_entries[rm->rm_used] = node_num;
271    rm->rm_used++;
272    spin_unlock(&osb->osb_lock);
273
274    return 0;
275}
276
277static void ocfs2_recovery_map_clear(struct ocfs2_super *osb,
278                     unsigned int node_num)
279{
280    int i;
281    struct ocfs2_recovery_map *rm = osb->recovery_map;
282
283    spin_lock(&osb->osb_lock);
284
285    for (i = 0; i < rm->rm_used; i++) {
286        if (rm->rm_entries[i] == node_num)
287            break;
288    }
289
290    if (i < rm->rm_used) {
291        /* XXX: be careful with the pointer math */
292        memmove(&(rm->rm_entries[i]), &(rm->rm_entries[i + 1]),
293            (rm->rm_used - i - 1) * sizeof(unsigned int));
294        rm->rm_used--;
295    }
296
297    spin_unlock(&osb->osb_lock);
298}
299
300static int ocfs2_commit_cache(struct ocfs2_super *osb)
301{
302    int status = 0;
303    unsigned int flushed;
304    unsigned long old_id;
305    struct ocfs2_journal *journal = NULL;
306
307    mlog_entry_void();
308
309    journal = osb->journal;
310
311    /* Flush all pending commits and checkpoint the journal. */
312    down_write(&journal->j_trans_barrier);
313
314    if (atomic_read(&journal->j_num_trans) == 0) {
315        up_write(&journal->j_trans_barrier);
316        mlog(0, "No transactions for me to flush!\n");
317        goto finally;
318    }
319
320    jbd2_journal_lock_updates(journal->j_journal);
321    status = jbd2_journal_flush(journal->j_journal);
322    jbd2_journal_unlock_updates(journal->j_journal);
323    if (status < 0) {
324        up_write(&journal->j_trans_barrier);
325        mlog_errno(status);
326        goto finally;
327    }
328
329    old_id = ocfs2_inc_trans_id(journal);
330
331    flushed = atomic_read(&journal->j_num_trans);
332    atomic_set(&journal->j_num_trans, 0);
333    up_write(&journal->j_trans_barrier);
334
335    mlog(0, "commit_thread: flushed transaction %lu (%u handles)\n",
336         journal->j_trans_id, flushed);
337
338    ocfs2_wake_downconvert_thread(osb);
339    wake_up(&journal->j_checkpointed);
340finally:
341    mlog_exit(status);
342    return status;
343}
344
345/* pass it NULL and it will allocate a new handle object for you. If
346 * you pass it a handle however, it may still return error, in which
347 * case it has free'd the passed handle for you. */
348handle_t *ocfs2_start_trans(struct ocfs2_super *osb, int max_buffs)
349{
350    journal_t *journal = osb->journal->j_journal;
351    handle_t *handle;
352
353    BUG_ON(!osb || !osb->journal->j_journal);
354
355    if (ocfs2_is_hard_readonly(osb))
356        return ERR_PTR(-EROFS);
357
358    BUG_ON(osb->journal->j_state == OCFS2_JOURNAL_FREE);
359    BUG_ON(max_buffs <= 0);
360
361    /* Nested transaction? Just return the handle... */
362    if (journal_current_handle())
363        return jbd2_journal_start(journal, max_buffs);
364
365    down_read(&osb->journal->j_trans_barrier);
366
367    handle = jbd2_journal_start(journal, max_buffs);
368    if (IS_ERR(handle)) {
369        up_read(&osb->journal->j_trans_barrier);
370
371        mlog_errno(PTR_ERR(handle));
372
373        if (is_journal_aborted(journal)) {
374            ocfs2_abort(osb->sb, "Detected aborted journal");
375            handle = ERR_PTR(-EROFS);
376        }
377    } else {
378        if (!ocfs2_mount_local(osb))
379            atomic_inc(&(osb->journal->j_num_trans));
380    }
381
382    return handle;
383}
384
385int ocfs2_commit_trans(struct ocfs2_super *osb,
386               handle_t *handle)
387{
388    int ret, nested;
389    struct ocfs2_journal *journal = osb->journal;
390
391    BUG_ON(!handle);
392
393    nested = handle->h_ref > 1;
394    ret = jbd2_journal_stop(handle);
395    if (ret < 0)
396        mlog_errno(ret);
397
398    if (!nested)
399        up_read(&journal->j_trans_barrier);
400
401    return ret;
402}
403
404/*
405 * 'nblocks' is what you want to add to the current
406 * transaction. extend_trans will either extend the current handle by
407 * nblocks, or commit it and start a new one with nblocks credits.
408 *
409 * This might call jbd2_journal_restart() which will commit dirty buffers
410 * and then restart the transaction. Before calling
411 * ocfs2_extend_trans(), any changed blocks should have been
412 * dirtied. After calling it, all blocks which need to be changed must
413 * go through another set of journal_access/journal_dirty calls.
414 *
415 * WARNING: This will not release any semaphores or disk locks taken
416 * during the transaction, so make sure they were taken *before*
417 * start_trans or we'll have ordering deadlocks.
418 *
419 * WARNING2: Note that we do *not* drop j_trans_barrier here. This is
420 * good because transaction ids haven't yet been recorded on the
421 * cluster locks associated with this handle.
422 */
423int ocfs2_extend_trans(handle_t *handle, int nblocks)
424{
425    int status;
426
427    BUG_ON(!handle);
428    BUG_ON(!nblocks);
429
430    mlog_entry_void();
431
432    mlog(0, "Trying to extend transaction by %d blocks\n", nblocks);
433
434#ifdef CONFIG_OCFS2_DEBUG_FS
435    status = 1;
436#else
437    status = jbd2_journal_extend(handle, nblocks);
438    if (status < 0) {
439        mlog_errno(status);
440        goto bail;
441    }
442#endif
443
444    if (status > 0) {
445        mlog(0,
446             "jbd2_journal_extend failed, trying "
447             "jbd2_journal_restart\n");
448        status = jbd2_journal_restart(handle, nblocks);
449        if (status < 0) {
450            mlog_errno(status);
451            goto bail;
452        }
453    }
454
455    status = 0;
456bail:
457
458    mlog_exit(status);
459    return status;
460}
461
462struct ocfs2_triggers {
463    struct jbd2_buffer_trigger_type ot_triggers;
464    int ot_offset;
465};
466
467static inline struct ocfs2_triggers *to_ocfs2_trigger(struct jbd2_buffer_trigger_type *triggers)
468{
469    return container_of(triggers, struct ocfs2_triggers, ot_triggers);
470}
471
472static void ocfs2_commit_trigger(struct jbd2_buffer_trigger_type *triggers,
473                 struct buffer_head *bh,
474                 void *data, size_t size)
475{
476    struct ocfs2_triggers *ot = to_ocfs2_trigger(triggers);
477
478    /*
479     * We aren't guaranteed to have the superblock here, so we
480     * must unconditionally compute the ecc data.
481     * __ocfs2_journal_access() will only set the triggers if
482     * metaecc is enabled.
483     */
484    ocfs2_block_check_compute(data, size, data + ot->ot_offset);
485}
486
487/*
488 * Quota blocks have their own trigger because the struct ocfs2_block_check
489 * offset depends on the blocksize.
490 */
491static void ocfs2_dq_commit_trigger(struct jbd2_buffer_trigger_type *triggers,
492                 struct buffer_head *bh,
493                 void *data, size_t size)
494{
495    struct ocfs2_disk_dqtrailer *dqt =
496        ocfs2_block_dqtrailer(size, data);
497
498    /*
499     * We aren't guaranteed to have the superblock here, so we
500     * must unconditionally compute the ecc data.
501     * __ocfs2_journal_access() will only set the triggers if
502     * metaecc is enabled.
503     */
504    ocfs2_block_check_compute(data, size, &dqt->dq_check);
505}
506
507/*
508 * Directory blocks also have their own trigger because the
509 * struct ocfs2_block_check offset depends on the blocksize.
510 */
511static void ocfs2_db_commit_trigger(struct jbd2_buffer_trigger_type *triggers,
512                 struct buffer_head *bh,
513                 void *data, size_t size)
514{
515    struct ocfs2_dir_block_trailer *trailer =
516        ocfs2_dir_trailer_from_size(size, data);
517
518    /*
519     * We aren't guaranteed to have the superblock here, so we
520     * must unconditionally compute the ecc data.
521     * __ocfs2_journal_access() will only set the triggers if
522     * metaecc is enabled.
523     */
524    ocfs2_block_check_compute(data, size, &trailer->db_check);
525}
526
527static void ocfs2_abort_trigger(struct jbd2_buffer_trigger_type *triggers,
528                struct buffer_head *bh)
529{
530    mlog(ML_ERROR,
531         "ocfs2_abort_trigger called by JBD2. bh = 0x%lx, "
532         "bh->b_blocknr = %llu\n",
533         (unsigned long)bh,
534         (unsigned long long)bh->b_blocknr);
535
536    /* We aren't guaranteed to have the superblock here - but if we
537     * don't, it'll just crash. */
538    ocfs2_error(bh->b_assoc_map->host->i_sb,
539            "JBD2 has aborted our journal, ocfs2 cannot continue\n");
540}
541
542static struct ocfs2_triggers di_triggers = {
543    .ot_triggers = {
544        .t_commit = ocfs2_commit_trigger,
545        .t_abort = ocfs2_abort_trigger,
546    },
547    .ot_offset = offsetof(struct ocfs2_dinode, i_check),
548};
549
550static struct ocfs2_triggers eb_triggers = {
551    .ot_triggers = {
552        .t_commit = ocfs2_commit_trigger,
553        .t_abort = ocfs2_abort_trigger,
554    },
555    .ot_offset = offsetof(struct ocfs2_extent_block, h_check),
556};
557
558static struct ocfs2_triggers rb_triggers = {
559    .ot_triggers = {
560        .t_commit = ocfs2_commit_trigger,
561        .t_abort = ocfs2_abort_trigger,
562    },
563    .ot_offset = offsetof(struct ocfs2_refcount_block, rf_check),
564};
565
566static struct ocfs2_triggers gd_triggers = {
567    .ot_triggers = {
568        .t_commit = ocfs2_commit_trigger,
569        .t_abort = ocfs2_abort_trigger,
570    },
571    .ot_offset = offsetof(struct ocfs2_group_desc, bg_check),
572};
573
574static struct ocfs2_triggers db_triggers = {
575    .ot_triggers = {
576        .t_commit = ocfs2_db_commit_trigger,
577        .t_abort = ocfs2_abort_trigger,
578    },
579};
580
581static struct ocfs2_triggers xb_triggers = {
582    .ot_triggers = {
583        .t_commit = ocfs2_commit_trigger,
584        .t_abort = ocfs2_abort_trigger,
585    },
586    .ot_offset = offsetof(struct ocfs2_xattr_block, xb_check),
587};
588
589static struct ocfs2_triggers dq_triggers = {
590    .ot_triggers = {
591        .t_commit = ocfs2_dq_commit_trigger,
592        .t_abort = ocfs2_abort_trigger,
593    },
594};
595
596static struct ocfs2_triggers dr_triggers = {
597    .ot_triggers = {
598        .t_commit = ocfs2_commit_trigger,
599        .t_abort = ocfs2_abort_trigger,
600    },
601    .ot_offset = offsetof(struct ocfs2_dx_root_block, dr_check),
602};
603
604static struct ocfs2_triggers dl_triggers = {
605    .ot_triggers = {
606        .t_commit = ocfs2_commit_trigger,
607        .t_abort = ocfs2_abort_trigger,
608    },
609    .ot_offset = offsetof(struct ocfs2_dx_leaf, dl_check),
610};
611
612static int __ocfs2_journal_access(handle_t *handle,
613                  struct ocfs2_caching_info *ci,
614                  struct buffer_head *bh,
615                  struct ocfs2_triggers *triggers,
616                  int type)
617{
618    int status;
619    struct ocfs2_super *osb =
620        OCFS2_SB(ocfs2_metadata_cache_get_super(ci));
621
622    BUG_ON(!ci || !ci->ci_ops);
623    BUG_ON(!handle);
624    BUG_ON(!bh);
625
626    mlog_entry("bh->b_blocknr=%llu, type=%d (\"%s\"), bh->b_size = %zu\n",
627           (unsigned long long)bh->b_blocknr, type,
628           (type == OCFS2_JOURNAL_ACCESS_CREATE) ?
629           "OCFS2_JOURNAL_ACCESS_CREATE" :
630           "OCFS2_JOURNAL_ACCESS_WRITE",
631           bh->b_size);
632
633    /* we can safely remove this assertion after testing. */
634    if (!buffer_uptodate(bh)) {
635        mlog(ML_ERROR, "giving me a buffer that's not uptodate!\n");
636        mlog(ML_ERROR, "b_blocknr=%llu\n",
637             (unsigned long long)bh->b_blocknr);
638        BUG();
639    }
640
641    /* Set the current transaction information on the ci so
642     * that the locking code knows whether it can drop it's locks
643     * on this ci or not. We're protected from the commit
644     * thread updating the current transaction id until
645     * ocfs2_commit_trans() because ocfs2_start_trans() took
646     * j_trans_barrier for us. */
647    ocfs2_set_ci_lock_trans(osb->journal, ci);
648
649    ocfs2_metadata_cache_io_lock(ci);
650    switch (type) {
651    case OCFS2_JOURNAL_ACCESS_CREATE:
652    case OCFS2_JOURNAL_ACCESS_WRITE:
653        status = jbd2_journal_get_write_access(handle, bh);
654        break;
655
656    case OCFS2_JOURNAL_ACCESS_UNDO:
657        status = jbd2_journal_get_undo_access(handle, bh);
658        break;
659
660    default:
661        status = -EINVAL;
662        mlog(ML_ERROR, "Unknown access type!\n");
663    }
664    if (!status && ocfs2_meta_ecc(osb) && triggers)
665        jbd2_journal_set_triggers(bh, &triggers->ot_triggers);
666    ocfs2_metadata_cache_io_unlock(ci);
667
668    if (status < 0)
669        mlog(ML_ERROR, "Error %d getting %d access to buffer!\n",
670             status, type);
671
672    mlog_exit(status);
673    return status;
674}
675
676int ocfs2_journal_access_di(handle_t *handle, struct ocfs2_caching_info *ci,
677                struct buffer_head *bh, int type)
678{
679    return __ocfs2_journal_access(handle, ci, bh, &di_triggers, type);
680}
681
682int ocfs2_journal_access_eb(handle_t *handle, struct ocfs2_caching_info *ci,
683                struct buffer_head *bh, int type)
684{
685    return __ocfs2_journal_access(handle, ci, bh, &eb_triggers, type);
686}
687
688int ocfs2_journal_access_rb(handle_t *handle, struct ocfs2_caching_info *ci,
689                struct buffer_head *bh, int type)
690{
691    return __ocfs2_journal_access(handle, ci, bh, &rb_triggers,
692                      type);
693}
694
695int ocfs2_journal_access_gd(handle_t *handle, struct ocfs2_caching_info *ci,
696                struct buffer_head *bh, int type)
697{
698    return __ocfs2_journal_access(handle, ci, bh, &gd_triggers, type);
699}
700
701int ocfs2_journal_access_db(handle_t *handle, struct ocfs2_caching_info *ci,
702                struct buffer_head *bh, int type)
703{
704    return __ocfs2_journal_access(handle, ci, bh, &db_triggers, type);
705}
706
707int ocfs2_journal_access_xb(handle_t *handle, struct ocfs2_caching_info *ci,
708                struct buffer_head *bh, int type)
709{
710    return __ocfs2_journal_access(handle, ci, bh, &xb_triggers, type);
711}
712
713int ocfs2_journal_access_dq(handle_t *handle, struct ocfs2_caching_info *ci,
714                struct buffer_head *bh, int type)
715{
716    return __ocfs2_journal_access(handle, ci, bh, &dq_triggers, type);
717}
718
719int ocfs2_journal_access_dr(handle_t *handle, struct ocfs2_caching_info *ci,
720                struct buffer_head *bh, int type)
721{
722    return __ocfs2_journal_access(handle, ci, bh, &dr_triggers, type);
723}
724
725int ocfs2_journal_access_dl(handle_t *handle, struct ocfs2_caching_info *ci,
726                struct buffer_head *bh, int type)
727{
728    return __ocfs2_journal_access(handle, ci, bh, &dl_triggers, type);
729}
730
731int ocfs2_journal_access(handle_t *handle, struct ocfs2_caching_info *ci,
732             struct buffer_head *bh, int type)
733{
734    return __ocfs2_journal_access(handle, ci, bh, NULL, type);
735}
736
737int ocfs2_journal_dirty(handle_t *handle,
738            struct buffer_head *bh)
739{
740    int status;
741
742    mlog_entry("(bh->b_blocknr=%llu)\n",
743           (unsigned long long)bh->b_blocknr);
744
745    status = jbd2_journal_dirty_metadata(handle, bh);
746    if (status < 0)
747        mlog(ML_ERROR, "Could not dirty metadata buffer. "
748             "(bh->b_blocknr=%llu)\n",
749             (unsigned long long)bh->b_blocknr);
750
751    mlog_exit(status);
752    return status;
753}
754
755#define OCFS2_DEFAULT_COMMIT_INTERVAL (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE)
756
757void ocfs2_set_journal_params(struct ocfs2_super *osb)
758{
759    journal_t *journal = osb->journal->j_journal;
760    unsigned long commit_interval = OCFS2_DEFAULT_COMMIT_INTERVAL;
761
762    if (osb->osb_commit_interval)
763        commit_interval = osb->osb_commit_interval;
764
765    spin_lock(&journal->j_state_lock);
766    journal->j_commit_interval = commit_interval;
767    if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER)
768        journal->j_flags |= JBD2_BARRIER;
769    else
770        journal->j_flags &= ~JBD2_BARRIER;
771    spin_unlock(&journal->j_state_lock);
772}
773
774int ocfs2_journal_init(struct ocfs2_journal *journal, int *dirty)
775{
776    int status = -1;
777    struct inode *inode = NULL; /* the journal inode */
778    journal_t *j_journal = NULL;
779    struct ocfs2_dinode *di = NULL;
780    struct buffer_head *bh = NULL;
781    struct ocfs2_super *osb;
782    int inode_lock = 0;
783
784    mlog_entry_void();
785
786    BUG_ON(!journal);
787
788    osb = journal->j_osb;
789
790    /* already have the inode for our journal */
791    inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
792                        osb->slot_num);
793    if (inode == NULL) {
794        status = -EACCES;
795        mlog_errno(status);
796        goto done;
797    }
798    if (is_bad_inode(inode)) {
799        mlog(ML_ERROR, "access error (bad inode)\n");
800        iput(inode);
801        inode = NULL;
802        status = -EACCES;
803        goto done;
804    }
805
806    SET_INODE_JOURNAL(inode);
807    OCFS2_I(inode)->ip_open_count++;
808
809    /* Skip recovery waits here - journal inode metadata never
810     * changes in a live cluster so it can be considered an
811     * exception to the rule. */
812    status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY);
813    if (status < 0) {
814        if (status != -ERESTARTSYS)
815            mlog(ML_ERROR, "Could not get lock on journal!\n");
816        goto done;
817    }
818
819    inode_lock = 1;
820    di = (struct ocfs2_dinode *)bh->b_data;
821
822    if (inode->i_size < OCFS2_MIN_JOURNAL_SIZE) {
823        mlog(ML_ERROR, "Journal file size (%lld) is too small!\n",
824             inode->i_size);
825        status = -EINVAL;
826        goto done;
827    }
828
829    mlog(0, "inode->i_size = %lld\n", inode->i_size);
830    mlog(0, "inode->i_blocks = %llu\n",
831            (unsigned long long)inode->i_blocks);
832    mlog(0, "inode->ip_clusters = %u\n", OCFS2_I(inode)->ip_clusters);
833
834    /* call the kernels journal init function now */
835    j_journal = jbd2_journal_init_inode(inode);
836    if (j_journal == NULL) {
837        mlog(ML_ERROR, "Linux journal layer error\n");
838        status = -EINVAL;
839        goto done;
840    }
841
842    mlog(0, "Returned from jbd2_journal_init_inode\n");
843    mlog(0, "j_journal->j_maxlen = %u\n", j_journal->j_maxlen);
844
845    *dirty = (le32_to_cpu(di->id1.journal1.ij_flags) &
846          OCFS2_JOURNAL_DIRTY_FL);
847
848    journal->j_journal = j_journal;
849    journal->j_inode = inode;
850    journal->j_bh = bh;
851
852    ocfs2_set_journal_params(osb);
853
854    journal->j_state = OCFS2_JOURNAL_LOADED;
855
856    status = 0;
857done:
858    if (status < 0) {
859        if (inode_lock)
860            ocfs2_inode_unlock(inode, 1);
861        brelse(bh);
862        if (inode) {
863            OCFS2_I(inode)->ip_open_count--;
864            iput(inode);
865        }
866    }
867
868    mlog_exit(status);
869    return status;
870}
871
872static void ocfs2_bump_recovery_generation(struct ocfs2_dinode *di)
873{
874    le32_add_cpu(&(di->id1.journal1.ij_recovery_generation), 1);
875}
876
877static u32 ocfs2_get_recovery_generation(struct ocfs2_dinode *di)
878{
879    return le32_to_cpu(di->id1.journal1.ij_recovery_generation);
880}
881
882static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
883                      int dirty, int replayed)
884{
885    int status;
886    unsigned int flags;
887    struct ocfs2_journal *journal = osb->journal;
888    struct buffer_head *bh = journal->j_bh;
889    struct ocfs2_dinode *fe;
890
891    mlog_entry_void();
892
893    fe = (struct ocfs2_dinode *)bh->b_data;
894
895    /* The journal bh on the osb always comes from ocfs2_journal_init()
896     * and was validated there inside ocfs2_inode_lock_full(). It's a
897     * code bug if we mess it up. */
898    BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
899
900    flags = le32_to_cpu(fe->id1.journal1.ij_flags);
901    if (dirty)
902        flags |= OCFS2_JOURNAL_DIRTY_FL;
903    else
904        flags &= ~OCFS2_JOURNAL_DIRTY_FL;
905    fe->id1.journal1.ij_flags = cpu_to_le32(flags);
906
907    if (replayed)
908        ocfs2_bump_recovery_generation(fe);
909
910    ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check);
911    status = ocfs2_write_block(osb, bh, INODE_CACHE(journal->j_inode));
912    if (status < 0)
913        mlog_errno(status);
914
915    mlog_exit(status);
916    return status;
917}
918
919/*
920 * If the journal has been kmalloc'd it needs to be freed after this
921 * call.
922 */
923void ocfs2_journal_shutdown(struct ocfs2_super *osb)
924{
925    struct ocfs2_journal *journal = NULL;
926    int status = 0;
927    struct inode *inode = NULL;
928    int num_running_trans = 0;
929
930    mlog_entry_void();
931
932    BUG_ON(!osb);
933
934    journal = osb->journal;
935    if (!journal)
936        goto done;
937
938    inode = journal->j_inode;
939
940    if (journal->j_state != OCFS2_JOURNAL_LOADED)
941        goto done;
942
943    /* need to inc inode use count - jbd2_journal_destroy will iput. */
944    if (!igrab(inode))
945        BUG();
946
947    num_running_trans = atomic_read(&(osb->journal->j_num_trans));
948    if (num_running_trans > 0)
949        mlog(0, "Shutting down journal: must wait on %d "
950             "running transactions!\n",
951             num_running_trans);
952
953    /* Do a commit_cache here. It will flush our journal, *and*
954     * release any locks that are still held.
955     * set the SHUTDOWN flag and release the trans lock.
956     * the commit thread will take the trans lock for us below. */
957    journal->j_state = OCFS2_JOURNAL_IN_SHUTDOWN;
958
959    /* The OCFS2_JOURNAL_IN_SHUTDOWN will signal to commit_cache to not
960     * drop the trans_lock (which we want to hold until we
961     * completely destroy the journal. */
962    if (osb->commit_task) {
963        /* Wait for the commit thread */
964        mlog(0, "Waiting for ocfs2commit to exit....\n");
965        kthread_stop(osb->commit_task);
966        osb->commit_task = NULL;
967    }
968
969    BUG_ON(atomic_read(&(osb->journal->j_num_trans)) != 0);
970
971    if (ocfs2_mount_local(osb)) {
972        jbd2_journal_lock_updates(journal->j_journal);
973        status = jbd2_journal_flush(journal->j_journal);
974        jbd2_journal_unlock_updates(journal->j_journal);
975        if (status < 0)
976            mlog_errno(status);
977    }
978
979    if (status == 0) {
980        /*
981         * Do not toggle if flush was unsuccessful otherwise
982         * will leave dirty metadata in a "clean" journal
983         */
984        status = ocfs2_journal_toggle_dirty(osb, 0, 0);
985        if (status < 0)
986            mlog_errno(status);
987    }
988
989    /* Shutdown the kernel journal system */
990    jbd2_journal_destroy(journal->j_journal);
991    journal->j_journal = NULL;
992
993    OCFS2_I(inode)->ip_open_count--;
994
995    /* unlock our journal */
996    ocfs2_inode_unlock(inode, 1);
997
998    brelse(journal->j_bh);
999    journal->j_bh = NULL;
1000
1001    journal->j_state = OCFS2_JOURNAL_FREE;
1002
1003// up_write(&journal->j_trans_barrier);
1004done:
1005    if (inode)
1006        iput(inode);
1007    mlog_exit_void();
1008}
1009
1010static void ocfs2_clear_journal_error(struct super_block *sb,
1011                      journal_t *journal,
1012                      int slot)
1013{
1014    int olderr;
1015
1016    olderr = jbd2_journal_errno(journal);
1017    if (olderr) {
1018        mlog(ML_ERROR, "File system error %d recorded in "
1019             "journal %u.\n", olderr, slot);
1020        mlog(ML_ERROR, "File system on device %s needs checking.\n",
1021             sb->s_id);
1022
1023        jbd2_journal_ack_err(journal);
1024        jbd2_journal_clear_err(journal);
1025    }
1026}
1027
1028int ocfs2_journal_load(struct ocfs2_journal *journal, int local, int replayed)
1029{
1030    int status = 0;
1031    struct ocfs2_super *osb;
1032
1033    mlog_entry_void();
1034
1035    BUG_ON(!journal);
1036
1037    osb = journal->j_osb;
1038
1039    status = jbd2_journal_load(journal->j_journal);
1040    if (status < 0) {
1041        mlog(ML_ERROR, "Failed to load journal!\n");
1042        goto done;
1043    }
1044
1045    ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num);
1046
1047    status = ocfs2_journal_toggle_dirty(osb, 1, replayed);
1048    if (status < 0) {
1049        mlog_errno(status);
1050        goto done;
1051    }
1052
1053    /* Launch the commit thread */
1054    if (!local) {
1055        osb->commit_task = kthread_run(ocfs2_commit_thread, osb,
1056                           "ocfs2cmt");
1057        if (IS_ERR(osb->commit_task)) {
1058            status = PTR_ERR(osb->commit_task);
1059            osb->commit_task = NULL;
1060            mlog(ML_ERROR, "unable to launch ocfs2commit thread, "
1061                 "error=%d", status);
1062            goto done;
1063        }
1064    } else
1065        osb->commit_task = NULL;
1066
1067done:
1068    mlog_exit(status);
1069    return status;
1070}
1071
1072
1073/* 'full' flag tells us whether we clear out all blocks or if we just
1074 * mark the journal clean */
1075int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full)
1076{
1077    int status;
1078
1079    mlog_entry_void();
1080
1081    BUG_ON(!journal);
1082
1083    status = jbd2_journal_wipe(journal->j_journal, full);
1084    if (status < 0) {
1085        mlog_errno(status);
1086        goto bail;
1087    }
1088
1089    status = ocfs2_journal_toggle_dirty(journal->j_osb, 0, 0);
1090    if (status < 0)
1091        mlog_errno(status);
1092
1093bail:
1094    mlog_exit(status);
1095    return status;
1096}
1097
1098static int ocfs2_recovery_completed(struct ocfs2_super *osb)
1099{
1100    int empty;
1101    struct ocfs2_recovery_map *rm = osb->recovery_map;
1102
1103    spin_lock(&osb->osb_lock);
1104    empty = (rm->rm_used == 0);
1105    spin_unlock(&osb->osb_lock);
1106
1107    return empty;
1108}
1109
1110void ocfs2_wait_for_recovery(struct ocfs2_super *osb)
1111{
1112    wait_event(osb->recovery_event, ocfs2_recovery_completed(osb));
1113}
1114
1115/*
1116 * JBD Might read a cached version of another nodes journal file. We
1117 * don't want this as this file changes often and we get no
1118 * notification on those changes. The only way to be sure that we've
1119 * got the most up to date version of those blocks then is to force
1120 * read them off disk. Just searching through the buffer cache won't
1121 * work as there may be pages backing this file which are still marked
1122 * up to date. We know things can't change on this file underneath us
1123 * as we have the lock by now :)
1124 */
1125static int ocfs2_force_read_journal(struct inode *inode)
1126{
1127    int status = 0;
1128    int i;
1129    u64 v_blkno, p_blkno, p_blocks, num_blocks;
1130#define CONCURRENT_JOURNAL_FILL 32ULL
1131    struct buffer_head *bhs[CONCURRENT_JOURNAL_FILL];
1132
1133    mlog_entry_void();
1134
1135    memset(bhs, 0, sizeof(struct buffer_head *) * CONCURRENT_JOURNAL_FILL);
1136
1137    num_blocks = ocfs2_blocks_for_bytes(inode->i_sb, inode->i_size);
1138    v_blkno = 0;
1139    while (v_blkno < num_blocks) {
1140        status = ocfs2_extent_map_get_blocks(inode, v_blkno,
1141                             &p_blkno, &p_blocks, NULL);
1142        if (status < 0) {
1143            mlog_errno(status);
1144            goto bail;
1145        }
1146
1147        if (p_blocks > CONCURRENT_JOURNAL_FILL)
1148            p_blocks = CONCURRENT_JOURNAL_FILL;
1149
1150        /* We are reading journal data which should not
1151         * be put in the uptodate cache */
1152        status = ocfs2_read_blocks_sync(OCFS2_SB(inode->i_sb),
1153                        p_blkno, p_blocks, bhs);
1154        if (status < 0) {
1155            mlog_errno(status);
1156            goto bail;
1157        }
1158
1159        for(i = 0; i < p_blocks; i++) {
1160            brelse(bhs[i]);
1161            bhs[i] = NULL;
1162        }
1163
1164        v_blkno += p_blocks;
1165    }
1166
1167bail:
1168    for(i = 0; i < CONCURRENT_JOURNAL_FILL; i++)
1169        brelse(bhs[i]);
1170    mlog_exit(status);
1171    return status;
1172}
1173
1174struct ocfs2_la_recovery_item {
1175    struct list_head lri_list;
1176    int lri_slot;
1177    struct ocfs2_dinode *lri_la_dinode;
1178    struct ocfs2_dinode *lri_tl_dinode;
1179    struct ocfs2_quota_recovery *lri_qrec;
1180};
1181
1182/* Does the second half of the recovery process. By this point, the
1183 * node is marked clean and can actually be considered recovered,
1184 * hence it's no longer in the recovery map, but there's still some
1185 * cleanup we can do which shouldn't happen within the recovery thread
1186 * as locking in that context becomes very difficult if we are to take
1187 * recovering nodes into account.
1188 *
1189 * NOTE: This function can and will sleep on recovery of other nodes
1190 * during cluster locking, just like any other ocfs2 process.
1191 */
1192void ocfs2_complete_recovery(struct work_struct *work)
1193{
1194    int ret;
1195    struct ocfs2_journal *journal =
1196        container_of(work, struct ocfs2_journal, j_recovery_work);
1197    struct ocfs2_super *osb = journal->j_osb;
1198    struct ocfs2_dinode *la_dinode, *tl_dinode;
1199    struct ocfs2_la_recovery_item *item, *n;
1200    struct ocfs2_quota_recovery *qrec;
1201    LIST_HEAD(tmp_la_list);
1202
1203    mlog_entry_void();
1204
1205    mlog(0, "completing recovery from keventd\n");
1206
1207    spin_lock(&journal->j_lock);
1208    list_splice_init(&journal->j_la_cleanups, &tmp_la_list);
1209    spin_unlock(&journal->j_lock);
1210
1211    list_for_each_entry_safe(item, n, &tmp_la_list, lri_list) {
1212        list_del_init(&item->lri_list);
1213
1214        mlog(0, "Complete recovery for slot %d\n", item->lri_slot);
1215
1216        ocfs2_wait_on_quotas(osb);
1217
1218        la_dinode = item->lri_la_dinode;
1219        if (la_dinode) {
1220            mlog(0, "Clean up local alloc %llu\n",
1221                 (unsigned long long)le64_to_cpu(la_dinode->i_blkno));
1222
1223            ret = ocfs2_complete_local_alloc_recovery(osb,
1224                                  la_dinode);
1225            if (ret < 0)
1226                mlog_errno(ret);
1227
1228            kfree(la_dinode);
1229        }
1230
1231        tl_dinode = item->lri_tl_dinode;
1232        if (tl_dinode) {
1233            mlog(0, "Clean up truncate log %llu\n",
1234                 (unsigned long long)le64_to_cpu(tl_dinode->i_blkno));
1235
1236            ret = ocfs2_complete_truncate_log_recovery(osb,
1237                                   tl_dinode);
1238            if (ret < 0)
1239                mlog_errno(ret);
1240
1241            kfree(tl_dinode);
1242        }
1243
1244        ret = ocfs2_recover_orphans(osb, item->lri_slot);
1245        if (ret < 0)
1246            mlog_errno(ret);
1247
1248        qrec = item->lri_qrec;
1249        if (qrec) {
1250            mlog(0, "Recovering quota files");
1251            ret = ocfs2_finish_quota_recovery(osb, qrec,
1252                              item->lri_slot);
1253            if (ret < 0)
1254                mlog_errno(ret);
1255            /* Recovery info is already freed now */
1256        }
1257
1258        kfree(item);
1259    }
1260
1261    mlog(0, "Recovery completion\n");
1262    mlog_exit_void();
1263}
1264
1265/* NOTE: This function always eats your references to la_dinode and
1266 * tl_dinode, either manually on error, or by passing them to
1267 * ocfs2_complete_recovery */
1268static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal,
1269                        int slot_num,
1270                        struct ocfs2_dinode *la_dinode,
1271                        struct ocfs2_dinode *tl_dinode,
1272                        struct ocfs2_quota_recovery *qrec)
1273{
1274    struct ocfs2_la_recovery_item *item;
1275
1276    item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS);
1277    if (!item) {
1278        /* Though we wish to avoid it, we are in fact safe in
1279         * skipping local alloc cleanup as fsck.ocfs2 is more
1280         * than capable of reclaiming unused space. */
1281        if (la_dinode)
1282            kfree(la_dinode);
1283
1284        if (tl_dinode)
1285            kfree(tl_dinode);
1286
1287        if (qrec)
1288            ocfs2_free_quota_recovery(qrec);
1289
1290        mlog_errno(-ENOMEM);
1291        return;
1292    }
1293
1294    INIT_LIST_HEAD(&item->lri_list);
1295    item->lri_la_dinode = la_dinode;
1296    item->lri_slot = slot_num;
1297    item->lri_tl_dinode = tl_dinode;
1298    item->lri_qrec = qrec;
1299
1300    spin_lock(&journal->j_lock);
1301    list_add_tail(&item->lri_list, &journal->j_la_cleanups);
1302    queue_work(ocfs2_wq, &journal->j_recovery_work);
1303    spin_unlock(&journal->j_lock);
1304}
1305
1306/* Called by the mount code to queue recovery the last part of
1307 * recovery for it's own and offline slot(s). */
1308void ocfs2_complete_mount_recovery(struct ocfs2_super *osb)
1309{
1310    struct ocfs2_journal *journal = osb->journal;
1311
1312    /* No need to queue up our truncate_log as regular cleanup will catch
1313     * that */
1314    ocfs2_queue_recovery_completion(journal, osb->slot_num,
1315                    osb->local_alloc_copy, NULL, NULL);
1316    ocfs2_schedule_truncate_log_flush(osb, 0);
1317
1318    osb->local_alloc_copy = NULL;
1319    osb->dirty = 0;
1320
1321    /* queue to recover orphan slots for all offline slots */
1322    ocfs2_replay_map_set_state(osb, REPLAY_NEEDED);
1323    ocfs2_queue_replay_slots(osb);
1324    ocfs2_free_replay_slots(osb);
1325}
1326
1327void ocfs2_complete_quota_recovery(struct ocfs2_super *osb)
1328{
1329    if (osb->quota_rec) {
1330        ocfs2_queue_recovery_completion(osb->journal,
1331                        osb->slot_num,
1332                        NULL,
1333                        NULL,
1334                        osb->quota_rec);
1335        osb->quota_rec = NULL;
1336    }
1337}
1338
1339static int __ocfs2_recovery_thread(void *arg)
1340{
1341    int status, node_num, slot_num;
1342    struct ocfs2_super *osb = arg;
1343    struct ocfs2_recovery_map *rm = osb->recovery_map;
1344    int *rm_quota = NULL;
1345    int rm_quota_used = 0, i;
1346    struct ocfs2_quota_recovery *qrec;
1347
1348    mlog_entry_void();
1349
1350    status = ocfs2_wait_on_mount(osb);
1351    if (status < 0) {
1352        goto bail;
1353    }
1354
1355    rm_quota = kzalloc(osb->max_slots * sizeof(int), GFP_NOFS);
1356    if (!rm_quota) {
1357        status = -ENOMEM;
1358        goto bail;
1359    }
1360restart:
1361    status = ocfs2_super_lock(osb, 1);
1362    if (status < 0) {
1363        mlog_errno(status);
1364        goto bail;
1365    }
1366
1367    status = ocfs2_compute_replay_slots(osb);
1368    if (status < 0)
1369        mlog_errno(status);
1370
1371    /* queue recovery for our own slot */
1372    ocfs2_queue_recovery_completion(osb->journal, osb->slot_num, NULL,
1373                    NULL, NULL);
1374
1375    spin_lock(&osb->osb_lock);
1376    while (rm->rm_used) {
1377        /* It's always safe to remove entry zero, as we won't
1378         * clear it until ocfs2_recover_node() has succeeded. */
1379        node_num = rm->rm_entries[0];
1380        spin_unlock(&osb->osb_lock);
1381        mlog(0, "checking node %d\n", node_num);
1382        slot_num = ocfs2_node_num_to_slot(osb, node_num);
1383        if (slot_num == -ENOENT) {
1384            status = 0;
1385            mlog(0, "no slot for this node, so no recovery"
1386                 "required.\n");
1387            goto skip_recovery;
1388        }
1389        mlog(0, "node %d was using slot %d\n", node_num, slot_num);
1390
1391        /* It is a bit subtle with quota recovery. We cannot do it
1392         * immediately because we have to obtain cluster locks from
1393         * quota files and we also don't want to just skip it because
1394         * then quota usage would be out of sync until some node takes
1395         * the slot. So we remember which nodes need quota recovery
1396         * and when everything else is done, we recover quotas. */
1397        for (i = 0; i < rm_quota_used && rm_quota[i] != slot_num; i++);
1398        if (i == rm_quota_used)
1399            rm_quota[rm_quota_used++] = slot_num;
1400
1401        status = ocfs2_recover_node(osb, node_num, slot_num);
1402skip_recovery:
1403        if (!status) {
1404            ocfs2_recovery_map_clear(osb, node_num);
1405        } else {
1406            mlog(ML_ERROR,
1407                 "Error %d recovering node %d on device (%u,%u)!\n",
1408                 status, node_num,
1409                 MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1410            mlog(ML_ERROR, "Volume requires unmount.\n");
1411        }
1412
1413        spin_lock(&osb->osb_lock);
1414    }
1415    spin_unlock(&osb->osb_lock);
1416    mlog(0, "All nodes recovered\n");
1417
1418    /* Refresh all journal recovery generations from disk */
1419    status = ocfs2_check_journals_nolocks(osb);
1420    status = (status == -EROFS) ? 0 : status;
1421    if (status < 0)
1422        mlog_errno(status);
1423
1424    /* Now it is right time to recover quotas... We have to do this under
1425     * superblock lock so that noone can start using the slot (and crash)
1426     * before we recover it */
1427    for (i = 0; i < rm_quota_used; i++) {
1428        qrec = ocfs2_begin_quota_recovery(osb, rm_quota[i]);
1429        if (IS_ERR(qrec)) {
1430            status = PTR_ERR(qrec);
1431            mlog_errno(status);
1432            continue;
1433        }
1434        ocfs2_queue_recovery_completion(osb->journal, rm_quota[i],
1435                        NULL, NULL, qrec);
1436    }
1437
1438    ocfs2_super_unlock(osb, 1);
1439
1440    /* queue recovery for offline slots */
1441    ocfs2_queue_replay_slots(osb);
1442
1443bail:
1444    mutex_lock(&osb->recovery_lock);
1445    if (!status && !ocfs2_recovery_completed(osb)) {
1446        mutex_unlock(&osb->recovery_lock);
1447        goto restart;
1448    }
1449
1450    ocfs2_free_replay_slots(osb);
1451    osb->recovery_thread_task = NULL;
1452    mb(); /* sync with ocfs2_recovery_thread_running */
1453    wake_up(&osb->recovery_event);
1454
1455    mutex_unlock(&osb->recovery_lock);
1456
1457    if (rm_quota)
1458        kfree(rm_quota);
1459
1460    mlog_exit(status);
1461    /* no one is callint kthread_stop() for us so the kthread() api
1462     * requires that we call do_exit(). And it isn't exported, but
1463     * complete_and_exit() seems to be a minimal wrapper around it. */
1464    complete_and_exit(NULL, status);
1465    return status;
1466}
1467
1468void ocfs2_recovery_thread(struct ocfs2_super *osb, int node_num)
1469{
1470    mlog_entry("(node_num=%d, osb->node_num = %d)\n",
1471           node_num, osb->node_num);
1472
1473    mutex_lock(&osb->recovery_lock);
1474    if (osb->disable_recovery)
1475        goto out;
1476
1477    /* People waiting on recovery will wait on
1478     * the recovery map to empty. */
1479    if (ocfs2_recovery_map_set(osb, node_num))
1480        mlog(0, "node %d already in recovery map.\n", node_num);
1481
1482    mlog(0, "starting recovery thread...\n");
1483
1484    if (osb->recovery_thread_task)
1485        goto out;
1486
1487    osb->recovery_thread_task = kthread_run(__ocfs2_recovery_thread, osb,
1488                         "ocfs2rec");
1489    if (IS_ERR(osb->recovery_thread_task)) {
1490        mlog_errno((int)PTR_ERR(osb->recovery_thread_task));
1491        osb->recovery_thread_task = NULL;
1492    }
1493
1494out:
1495    mutex_unlock(&osb->recovery_lock);
1496    wake_up(&osb->recovery_event);
1497
1498    mlog_exit_void();
1499}
1500
1501static int ocfs2_read_journal_inode(struct ocfs2_super *osb,
1502                    int slot_num,
1503                    struct buffer_head **bh,
1504                    struct inode **ret_inode)
1505{
1506    int status = -EACCES;
1507    struct inode *inode = NULL;
1508
1509    BUG_ON(slot_num >= osb->max_slots);
1510
1511    inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
1512                        slot_num);
1513    if (!inode || is_bad_inode(inode)) {
1514        mlog_errno(status);
1515        goto bail;
1516    }
1517    SET_INODE_JOURNAL(inode);
1518
1519    status = ocfs2_read_inode_block_full(inode, bh, OCFS2_BH_IGNORE_CACHE);
1520    if (status < 0) {
1521        mlog_errno(status);
1522        goto bail;
1523    }
1524
1525    status = 0;
1526
1527bail:
1528    if (inode) {
1529        if (status || !ret_inode)
1530            iput(inode);
1531        else
1532            *ret_inode = inode;
1533    }
1534    return status;
1535}
1536
1537/* Does the actual journal replay and marks the journal inode as
1538 * clean. Will only replay if the journal inode is marked dirty. */
1539static int ocfs2_replay_journal(struct ocfs2_super *osb,
1540                int node_num,
1541                int slot_num)
1542{
1543    int status;
1544    int got_lock = 0;
1545    unsigned int flags;
1546    struct inode *inode = NULL;
1547    struct ocfs2_dinode *fe;
1548    journal_t *journal = NULL;
1549    struct buffer_head *bh = NULL;
1550    u32 slot_reco_gen;
1551
1552    status = ocfs2_read_journal_inode(osb, slot_num, &bh, &inode);
1553    if (status) {
1554        mlog_errno(status);
1555        goto done;
1556    }
1557
1558    fe = (struct ocfs2_dinode *)bh->b_data;
1559    slot_reco_gen = ocfs2_get_recovery_generation(fe);
1560    brelse(bh);
1561    bh = NULL;
1562
1563    /*
1564     * As the fs recovery is asynchronous, there is a small chance that
1565     * another node mounted (and recovered) the slot before the recovery
1566     * thread could get the lock. To handle that, we dirty read the journal
1567     * inode for that slot to get the recovery generation. If it is
1568     * different than what we expected, the slot has been recovered.
1569     * If not, it needs recovery.
1570     */
1571    if (osb->slot_recovery_generations[slot_num] != slot_reco_gen) {
1572        mlog(0, "Slot %u already recovered (old/new=%u/%u)\n", slot_num,
1573             osb->slot_recovery_generations[slot_num], slot_reco_gen);
1574        osb->slot_recovery_generations[slot_num] = slot_reco_gen;
1575        status = -EBUSY;
1576        goto done;
1577    }
1578
1579    /* Continue with recovery as the journal has not yet been recovered */
1580
1581    status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY);
1582    if (status < 0) {
1583        mlog(0, "status returned from ocfs2_inode_lock=%d\n", status);
1584        if (status != -ERESTARTSYS)
1585            mlog(ML_ERROR, "Could not lock journal!\n");
1586        goto done;
1587    }
1588    got_lock = 1;
1589
1590    fe = (struct ocfs2_dinode *) bh->b_data;
1591
1592    flags = le32_to_cpu(fe->id1.journal1.ij_flags);
1593    slot_reco_gen = ocfs2_get_recovery_generation(fe);
1594
1595    if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) {
1596        mlog(0, "No recovery required for node %d\n", node_num);
1597        /* Refresh recovery generation for the slot */
1598        osb->slot_recovery_generations[slot_num] = slot_reco_gen;
1599        goto done;
1600    }
1601
1602    /* we need to run complete recovery for offline orphan slots */
1603    ocfs2_replay_map_set_state(osb, REPLAY_NEEDED);
1604
1605    mlog(ML_NOTICE, "Recovering node %d from slot %d on device (%u,%u)\n",
1606         node_num, slot_num,
1607         MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1608
1609    OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
1610
1611    status = ocfs2_force_read_journal(inode);
1612    if (status < 0) {
1613        mlog_errno(status);
1614        goto done;
1615    }
1616
1617    mlog(0, "calling journal_init_inode\n");
1618    journal = jbd2_journal_init_inode(inode);
1619    if (journal == NULL) {
1620        mlog(ML_ERROR, "Linux journal layer error\n");
1621        status = -EIO;
1622        goto done;
1623    }
1624
1625    status = jbd2_journal_load(journal);
1626    if (status < 0) {
1627        mlog_errno(status);
1628        if (!igrab(inode))
1629            BUG();
1630        jbd2_journal_destroy(journal);
1631        goto done;
1632    }
1633
1634    ocfs2_clear_journal_error(osb->sb, journal, slot_num);
1635
1636    /* wipe the journal */
1637    mlog(0, "flushing the journal.\n");
1638    jbd2_journal_lock_updates(journal);
1639    status = jbd2_journal_flush(journal);
1640    jbd2_journal_unlock_updates(journal);
1641    if (status < 0)
1642        mlog_errno(status);
1643
1644    /* This will mark the node clean */
1645    flags = le32_to_cpu(fe->id1.journal1.ij_flags);
1646    flags &= ~OCFS2_JOURNAL_DIRTY_FL;
1647    fe->id1.journal1.ij_flags = cpu_to_le32(flags);
1648
1649    /* Increment recovery generation to indicate successful recovery */
1650    ocfs2_bump_recovery_generation(fe);
1651    osb->slot_recovery_generations[slot_num] =
1652                    ocfs2_get_recovery_generation(fe);
1653
1654    ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check);
1655    status = ocfs2_write_block(osb, bh, INODE_CACHE(inode));
1656    if (status < 0)
1657        mlog_errno(status);
1658
1659    if (!igrab(inode))
1660        BUG();
1661
1662    jbd2_journal_destroy(journal);
1663
1664done:
1665    /* drop the lock on this nodes journal */
1666    if (got_lock)
1667        ocfs2_inode_unlock(inode, 1);
1668
1669    if (inode)
1670        iput(inode);
1671
1672    brelse(bh);
1673
1674    mlog_exit(status);
1675    return status;
1676}
1677
1678/*
1679 * Do the most important parts of node recovery:
1680 * - Replay it's journal
1681 * - Stamp a clean local allocator file
1682 * - Stamp a clean truncate log
1683 * - Mark the node clean
1684 *
1685 * If this function completes without error, a node in OCFS2 can be
1686 * said to have been safely recovered. As a result, failure during the
1687 * second part of a nodes recovery process (local alloc recovery) is
1688 * far less concerning.
1689 */
1690static int ocfs2_recover_node(struct ocfs2_super *osb,
1691                  int node_num, int slot_num)
1692{
1693    int status = 0;
1694    struct ocfs2_dinode *la_copy = NULL;
1695    struct ocfs2_dinode *tl_copy = NULL;
1696
1697    mlog_entry("(node_num=%d, slot_num=%d, osb->node_num = %d)\n",
1698           node_num, slot_num, osb->node_num);
1699
1700    /* Should not ever be called to recover ourselves -- in that
1701     * case we should've called ocfs2_journal_load instead. */
1702    BUG_ON(osb->node_num == node_num);
1703
1704    status = ocfs2_replay_journal(osb, node_num, slot_num);
1705    if (status < 0) {
1706        if (status == -EBUSY) {
1707            mlog(0, "Skipping recovery for slot %u (node %u) "
1708                 "as another node has recovered it\n", slot_num,
1709                 node_num);
1710            status = 0;
1711            goto done;
1712        }
1713        mlog_errno(status);
1714        goto done;
1715    }
1716
1717    /* Stamp a clean local alloc file AFTER recovering the journal... */
1718    status = ocfs2_begin_local_alloc_recovery(osb, slot_num, &la_copy);
1719    if (status < 0) {
1720        mlog_errno(status);
1721        goto done;
1722    }
1723
1724    /* An error from begin_truncate_log_recovery is not
1725     * serious enough to warrant halting the rest of
1726     * recovery. */
1727    status = ocfs2_begin_truncate_log_recovery(osb, slot_num, &tl_copy);
1728    if (status < 0)
1729        mlog_errno(status);
1730
1731    /* Likewise, this would be a strange but ultimately not so
1732     * harmful place to get an error... */
1733    status = ocfs2_clear_slot(osb, slot_num);
1734    if (status < 0)
1735        mlog_errno(status);
1736
1737    /* This will kfree the memory pointed to by la_copy and tl_copy */
1738    ocfs2_queue_recovery_completion(osb->journal, slot_num, la_copy,
1739                    tl_copy, NULL);
1740
1741    status = 0;
1742done:
1743
1744    mlog_exit(status);
1745    return status;
1746}
1747
1748/* Test node liveness by trylocking his journal. If we get the lock,
1749 * we drop it here. Return 0 if we got the lock, -EAGAIN if node is
1750 * still alive (we couldn't get the lock) and < 0 on error. */
1751static int ocfs2_trylock_journal(struct ocfs2_super *osb,
1752                 int slot_num)
1753{
1754    int status, flags;
1755    struct inode *inode = NULL;
1756
1757    inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
1758                        slot_num);
1759    if (inode == NULL) {
1760        mlog(ML_ERROR, "access error\n");
1761        status = -EACCES;
1762        goto bail;
1763    }
1764    if (is_bad_inode(inode)) {
1765        mlog(ML_ERROR, "access error (bad inode)\n");
1766        iput(inode);
1767        inode = NULL;
1768        status = -EACCES;
1769        goto bail;
1770    }
1771    SET_INODE_JOURNAL(inode);
1772
1773    flags = OCFS2_META_LOCK_RECOVERY | OCFS2_META_LOCK_NOQUEUE;
1774    status = ocfs2_inode_lock_full(inode, NULL, 1, flags);
1775    if (status < 0) {
1776        if (status != -EAGAIN)
1777            mlog_errno(status);
1778        goto bail;
1779    }
1780
1781    ocfs2_inode_unlock(inode, 1);
1782bail:
1783    if (inode)
1784        iput(inode);
1785
1786    return status;
1787}
1788
1789/* Call this underneath ocfs2_super_lock. It also assumes that the
1790 * slot info struct has been updated from disk. */
1791int ocfs2_mark_dead_nodes(struct ocfs2_super *osb)
1792{
1793    unsigned int node_num;
1794    int status, i;
1795    u32 gen;
1796    struct buffer_head *bh = NULL;
1797    struct ocfs2_dinode *di;
1798
1799    /* This is called with the super block cluster lock, so we
1800     * know that the slot map can't change underneath us. */
1801
1802    for (i = 0; i < osb->max_slots; i++) {
1803        /* Read journal inode to get the recovery generation */
1804        status = ocfs2_read_journal_inode(osb, i, &bh, NULL);
1805        if (status) {
1806            mlog_errno(status);
1807            goto bail;
1808        }
1809        di = (struct ocfs2_dinode *)bh->b_data;
1810        gen = ocfs2_get_recovery_generation(di);
1811        brelse(bh);
1812        bh = NULL;
1813
1814        spin_lock(&osb->osb_lock);
1815        osb->slot_recovery_generations[i] = gen;
1816
1817        mlog(0, "Slot %u recovery generation is %u\n", i,
1818             osb->slot_recovery_generations[i]);
1819
1820        if (i == osb->slot_num) {
1821            spin_unlock(&osb->osb_lock);
1822            continue;
1823        }
1824
1825        status = ocfs2_slot_to_node_num_locked(osb, i, &node_num);
1826        if (status == -ENOENT) {
1827            spin_unlock(&osb->osb_lock);
1828            continue;
1829        }
1830
1831        if (__ocfs2_recovery_map_test(osb, node_num)) {
1832            spin_unlock(&osb->osb_lock);
1833            continue;
1834        }
1835        spin_unlock(&osb->osb_lock);
1836
1837        /* Ok, we have a slot occupied by another node which
1838         * is not in the recovery map. We trylock his journal
1839         * file here to test if he's alive. */
1840        status = ocfs2_trylock_journal(osb, i);
1841        if (!status) {
1842            /* Since we're called from mount, we know that
1843             * the recovery thread can't race us on
1844             * setting / checking the recovery bits. */
1845            ocfs2_recovery_thread(osb, node_num);
1846        } else if ((status < 0) && (status != -EAGAIN)) {
1847            mlog_errno(status);
1848            goto bail;
1849        }
1850    }
1851
1852    status = 0;
1853bail:
1854    mlog_exit(status);
1855    return status;
1856}
1857
1858/*
1859 * Scan timer should get fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT. Add some
1860 * randomness to the timeout to minimize multple nodes firing the timer at the
1861 * same time.
1862 */
1863static inline unsigned long ocfs2_orphan_scan_timeout(void)
1864{
1865    unsigned long time;
1866
1867    get_random_bytes(&time, sizeof(time));
1868    time = ORPHAN_SCAN_SCHEDULE_TIMEOUT + (time % 5000);
1869    return msecs_to_jiffies(time);
1870}
1871
1872/*
1873 * ocfs2_queue_orphan_scan calls ocfs2_queue_recovery_completion for
1874 * every slot, queuing a recovery of the slot on the ocfs2_wq thread. This
1875 * is done to catch any orphans that are left over in orphan directories.
1876 *
1877 * ocfs2_queue_orphan_scan gets called every ORPHAN_SCAN_SCHEDULE_TIMEOUT
1878 * seconds. It gets an EX lock on os_lockres and checks sequence number
1879 * stored in LVB. If the sequence number has changed, it means some other
1880 * node has done the scan. This node skips the scan and tracks the
1881 * sequence number. If the sequence number didn't change, it means a scan
1882 * hasn't happened. The node queues a scan and increments the
1883 * sequence number in the LVB.
1884 */
1885void ocfs2_queue_orphan_scan(struct ocfs2_super *osb)
1886{
1887    struct ocfs2_orphan_scan *os;
1888    int status, i;
1889    u32 seqno = 0;
1890
1891    os = &osb->osb_orphan_scan;
1892
1893    if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE)
1894        goto out;
1895
1896    status = ocfs2_orphan_scan_lock(osb, &seqno);
1897    if (status < 0) {
1898        if (status != -EAGAIN)
1899            mlog_errno(status);
1900        goto out;
1901    }
1902
1903    /* Do no queue the tasks if the volume is being umounted */
1904    if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE)
1905        goto unlock;
1906
1907    if (os->os_seqno != seqno) {
1908        os->os_seqno = seqno;
1909        goto unlock;
1910    }
1911
1912    for (i = 0; i < osb->max_slots; i++)
1913        ocfs2_queue_recovery_completion(osb->journal, i, NULL, NULL,
1914                        NULL);
1915    /*
1916     * We queued a recovery on orphan slots, increment the sequence
1917     * number and update LVB so other node will skip the scan for a while
1918     */
1919    seqno++;
1920    os->os_count++;
1921    os->os_scantime = CURRENT_TIME;
1922unlock:
1923    ocfs2_orphan_scan_unlock(osb, seqno);
1924out:
1925    return;
1926}
1927
1928/* Worker task that gets fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT millsec */
1929void ocfs2_orphan_scan_work(struct work_struct *work)
1930{
1931    struct ocfs2_orphan_scan *os;
1932    struct ocfs2_super *osb;
1933
1934    os = container_of(work, struct ocfs2_orphan_scan,
1935              os_orphan_scan_work.work);
1936    osb = os->os_osb;
1937
1938    mutex_lock(&os->os_lock);
1939    ocfs2_queue_orphan_scan(osb);
1940    if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE)
1941        schedule_delayed_work(&os->os_orphan_scan_work,
1942                      ocfs2_orphan_scan_timeout());
1943    mutex_unlock(&os->os_lock);
1944}
1945
1946void ocfs2_orphan_scan_stop(struct ocfs2_super *osb)
1947{
1948    struct ocfs2_orphan_scan *os;
1949
1950    os = &osb->osb_orphan_scan;
1951    if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE) {
1952        atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE);
1953        mutex_lock(&os->os_lock);
1954        cancel_delayed_work(&os->os_orphan_scan_work);
1955        mutex_unlock(&os->os_lock);
1956    }
1957}
1958
1959void ocfs2_orphan_scan_init(struct ocfs2_super *osb)
1960{
1961    struct ocfs2_orphan_scan *os;
1962
1963    os = &osb->osb_orphan_scan;
1964    os->os_osb = osb;
1965    os->os_count = 0;
1966    os->os_seqno = 0;
1967    mutex_init(&os->os_lock);
1968    INIT_DELAYED_WORK(&os->os_orphan_scan_work, ocfs2_orphan_scan_work);
1969}
1970
1971void ocfs2_orphan_scan_start(struct ocfs2_super *osb)
1972{
1973    struct ocfs2_orphan_scan *os;
1974
1975    os = &osb->osb_orphan_scan;
1976    os->os_scantime = CURRENT_TIME;
1977    if (ocfs2_is_hard_readonly(osb) || ocfs2_mount_local(osb))
1978        atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE);
1979    else {
1980        atomic_set(&os->os_state, ORPHAN_SCAN_ACTIVE);
1981        schedule_delayed_work(&os->os_orphan_scan_work,
1982                      ocfs2_orphan_scan_timeout());
1983    }
1984}
1985
1986struct ocfs2_orphan_filldir_priv {
1987    struct inode *head;
1988    struct ocfs2_super *osb;
1989};
1990
1991static int ocfs2_orphan_filldir(void *priv, const char *name, int name_len,
1992                loff_t pos, u64 ino, unsigned type)
1993{
1994    struct ocfs2_orphan_filldir_priv *p = priv;
1995    struct inode *iter;
1996
1997    if (name_len == 1 && !strncmp(".", name, 1))
1998        return 0;
1999    if (name_len == 2 && !strncmp("..", name, 2))
2000        return 0;
2001
2002    /* Skip bad inodes so that recovery can continue */
2003    iter = ocfs2_iget(p->osb, ino,
2004              OCFS2_FI_FLAG_ORPHAN_RECOVERY, 0);
2005    if (IS_ERR(iter))
2006        return 0;
2007
2008    mlog(0, "queue orphan %llu\n",
2009         (unsigned long long)OCFS2_I(iter)->ip_blkno);
2010    /* No locking is required for the next_orphan queue as there
2011     * is only ever a single process doing orphan recovery. */
2012    OCFS2_I(iter)->ip_next_orphan = p->head;
2013    p->head = iter;
2014
2015    return 0;
2016}
2017
2018static int ocfs2_queue_orphans(struct ocfs2_super *osb,
2019                   int slot,
2020                   struct inode **head)
2021{
2022    int status;
2023    struct inode *orphan_dir_inode = NULL;
2024    struct ocfs2_orphan_filldir_priv priv;
2025    loff_t pos = 0;
2026
2027    priv.osb = osb;
2028    priv.head = *head;
2029
2030    orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2031                               ORPHAN_DIR_SYSTEM_INODE,
2032                               slot);
2033    if (!orphan_dir_inode) {
2034        status = -ENOENT;
2035        mlog_errno(status);
2036        return status;
2037    }
2038
2039    mutex_lock(&orphan_dir_inode->i_mutex);
2040    status = ocfs2_inode_lock(orphan_dir_inode, NULL, 0);
2041    if (status < 0) {
2042        mlog_errno(status);
2043        goto out;
2044    }
2045
2046    status = ocfs2_dir_foreach(orphan_dir_inode, &pos, &priv,
2047                   ocfs2_orphan_filldir);
2048    if (status) {
2049        mlog_errno(status);
2050        goto out_cluster;
2051    }
2052
2053    *head = priv.head;
2054
2055out_cluster:
2056    ocfs2_inode_unlock(orphan_dir_inode, 0);
2057out:
2058    mutex_unlock(&orphan_dir_inode->i_mutex);
2059    iput(orphan_dir_inode);
2060    return status;
2061}
2062
2063static int ocfs2_orphan_recovery_can_continue(struct ocfs2_super *osb,
2064                          int slot)
2065{
2066    int ret;
2067
2068    spin_lock(&osb->osb_lock);
2069    ret = !osb->osb_orphan_wipes[slot];
2070    spin_unlock(&osb->osb_lock);
2071    return ret;
2072}
2073
2074static void ocfs2_mark_recovering_orphan_dir(struct ocfs2_super *osb,
2075                         int slot)
2076{
2077    spin_lock(&osb->osb_lock);
2078    /* Mark ourselves such that new processes in delete_inode()
2079     * know to quit early. */
2080    ocfs2_node_map_set_bit(osb, &osb->osb_recovering_orphan_dirs, slot);
2081    while (osb->osb_orphan_wipes[slot]) {
2082        /* If any processes are already in the middle of an
2083         * orphan wipe on this dir, then we need to wait for
2084         * them. */
2085        spin_unlock(&osb->osb_lock);
2086        wait_event_interruptible(osb->osb_wipe_event,
2087                     ocfs2_orphan_recovery_can_continue(osb, slot));
2088        spin_lock(&osb->osb_lock);
2089    }
2090    spin_unlock(&osb->osb_lock);
2091}
2092
2093static void ocfs2_clear_recovering_orphan_dir(struct ocfs2_super *osb,
2094                          int slot)
2095{
2096    ocfs2_node_map_clear_bit(osb, &osb->osb_recovering_orphan_dirs, slot);
2097}
2098
2099/*
2100 * Orphan recovery. Each mounted node has it's own orphan dir which we
2101 * must run during recovery. Our strategy here is to build a list of
2102 * the inodes in the orphan dir and iget/iput them. The VFS does
2103 * (most) of the rest of the work.
2104 *
2105 * Orphan recovery can happen at any time, not just mount so we have a
2106 * couple of extra considerations.
2107 *
2108 * - We grab as many inodes as we can under the orphan dir lock -
2109 * doing iget() outside the orphan dir risks getting a reference on
2110 * an invalid inode.
2111 * - We must be sure not to deadlock with other processes on the
2112 * system wanting to run delete_inode(). This can happen when they go
2113 * to lock the orphan dir and the orphan recovery process attempts to
2114 * iget() inside the orphan dir lock. This can be avoided by
2115 * advertising our state to ocfs2_delete_inode().
2116 */
2117static int ocfs2_recover_orphans(struct ocfs2_super *osb,
2118                 int slot)
2119{
2120    int ret = 0;
2121    struct inode *inode = NULL;
2122    struct inode *iter;
2123    struct ocfs2_inode_info *oi;
2124
2125    mlog(0, "Recover inodes from orphan dir in slot %d\n", slot);
2126
2127    ocfs2_mark_recovering_orphan_dir(osb, slot);
2128    ret = ocfs2_queue_orphans(osb, slot, &inode);
2129    ocfs2_clear_recovering_orphan_dir(osb, slot);
2130
2131    /* Error here should be noted, but we want to continue with as
2132     * many queued inodes as we've got. */
2133    if (ret)
2134        mlog_errno(ret);
2135
2136    while (inode) {
2137        oi = OCFS2_I(inode);
2138        mlog(0, "iput orphan %llu\n", (unsigned long long)oi->ip_blkno);
2139
2140        iter = oi->ip_next_orphan;
2141
2142        spin_lock(&oi->ip_lock);
2143        /* The remote delete code may have set these on the
2144         * assumption that the other node would wipe them
2145         * successfully. If they are still in the node's
2146         * orphan dir, we need to reset that state. */
2147        oi->ip_flags &= ~(OCFS2_INODE_DELETED|OCFS2_INODE_SKIP_DELETE);
2148
2149        /* Set the proper information to get us going into
2150         * ocfs2_delete_inode. */
2151        oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED;
2152        spin_unlock(&oi->ip_lock);
2153
2154        iput(inode);
2155
2156        inode = iter;
2157    }
2158
2159    return ret;
2160}
2161
2162static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota)
2163{
2164    /* This check is good because ocfs2 will wait on our recovery
2165     * thread before changing it to something other than MOUNTED
2166     * or DISABLED. */
2167    wait_event(osb->osb_mount_event,
2168          (!quota && atomic_read(&osb->vol_state) == VOLUME_MOUNTED) ||
2169           atomic_read(&osb->vol_state) == VOLUME_MOUNTED_QUOTAS ||
2170           atomic_read(&osb->vol_state) == VOLUME_DISABLED);
2171
2172    /* If there's an error on mount, then we may never get to the
2173     * MOUNTED flag, but this is set right before
2174     * dismount_volume() so we can trust it. */
2175    if (atomic_read(&osb->vol_state) == VOLUME_DISABLED) {
2176        mlog(0, "mount error, exiting!\n");
2177        return -EBUSY;
2178    }
2179
2180    return 0;
2181}
2182
2183static int ocfs2_commit_thread(void *arg)
2184{
2185    int status;
2186    struct ocfs2_super *osb = arg;
2187    struct ocfs2_journal *journal = osb->journal;
2188
2189    /* we can trust j_num_trans here because _should_stop() is only set in
2190     * shutdown and nobody other than ourselves should be able to start
2191     * transactions. committing on shutdown might take a few iterations
2192     * as final transactions put deleted inodes on the list */
2193    while (!(kthread_should_stop() &&
2194         atomic_read(&journal->j_num_trans) == 0)) {
2195
2196        wait_event_interruptible(osb->checkpoint_event,
2197                     atomic_read(&journal->j_num_trans)
2198                     || kthread_should_stop());
2199
2200        status = ocfs2_commit_cache(osb);
2201        if (status < 0)
2202            mlog_errno(status);
2203
2204        if (kthread_should_stop() && atomic_read(&journal->j_num_trans)){
2205            mlog(ML_KTHREAD,
2206                 "commit_thread: %u transactions pending on "
2207                 "shutdown\n",
2208                 atomic_read(&journal->j_num_trans));
2209        }
2210    }
2211
2212    return 0;
2213}
2214
2215/* Reads all the journal inodes without taking any cluster locks. Used
2216 * for hard readonly access to determine whether any journal requires
2217 * recovery. Also used to refresh the recovery generation numbers after
2218 * a journal has been recovered by another node.
2219 */
2220int ocfs2_check_journals_nolocks(struct ocfs2_super *osb)
2221{
2222    int ret = 0;
2223    unsigned int slot;
2224    struct buffer_head *di_bh = NULL;
2225    struct ocfs2_dinode *di;
2226    int journal_dirty = 0;
2227
2228    for(slot = 0; slot < osb->max_slots; slot++) {
2229        ret = ocfs2_read_journal_inode(osb, slot, &di_bh, NULL);
2230        if (ret) {
2231            mlog_errno(ret);
2232            goto out;
2233        }
2234
2235        di = (struct ocfs2_dinode *) di_bh->b_data;
2236
2237        osb->slot_recovery_generations[slot] =
2238                    ocfs2_get_recovery_generation(di);
2239
2240        if (le32_to_cpu(di->id1.journal1.ij_flags) &
2241            OCFS2_JOURNAL_DIRTY_FL)
2242            journal_dirty = 1;
2243
2244        brelse(di_bh);
2245        di_bh = NULL;
2246    }
2247
2248out:
2249    if (journal_dirty)
2250        ret = -EROFS;
2251    return ret;
2252}
2253

Archive Download this file



interactive