1
2
3
4
5
6
7#include <linux/sched.h>
8#include <linux/slab.h>
9#include <linux/spinlock.h>
10#include <linux/completion.h>
11#include <linux/buffer_head.h>
12#include <linux/gfs2_ondisk.h>
13#include <linux/crc32.h>
14#include <linux/crc32c.h>
15#include <linux/delay.h>
16#include <linux/kthread.h>
17#include <linux/freezer.h>
18#include <linux/bio.h>
19#include <linux/blkdev.h>
20#include <linux/writeback.h>
21#include <linux/list_sort.h>
22
23#include "gfs2.h"
24#include "incore.h"
25#include "bmap.h"
26#include "glock.h"
27#include "log.h"
28#include "lops.h"
29#include "meta_io.h"
30#include "util.h"
31#include "dir.h"
32#include "trace_gfs2.h"
33#include "trans.h"
34
35static void gfs2_log_shutdown(struct gfs2_sbd *sdp);
36
37
38
39
40
41
42
43
44
45
46
47
48unsigned int gfs2_struct2blk(struct gfs2_sbd *sdp, unsigned int nstruct)
49{
50 unsigned int blks;
51 unsigned int first, second;
52
53 blks = 1;
54 first = sdp->sd_ldptrs;
55
56 if (nstruct > first) {
57 second = sdp->sd_inptrs;
58 blks += DIV_ROUND_UP(nstruct - first, second);
59 }
60
61 return blks;
62}
63
64
65
66
67
68
69
70
71
72
73static void gfs2_remove_from_ail(struct gfs2_bufdata *bd)
74{
75 bd->bd_tr = NULL;
76 list_del_init(&bd->bd_ail_st_list);
77 list_del_init(&bd->bd_ail_gl_list);
78 atomic_dec(&bd->bd_gl->gl_ail_count);
79 brelse(bd->bd_bh);
80}
81
82
83
84
85
86
87
88
89
90static int gfs2_ail1_start_one(struct gfs2_sbd *sdp,
91 struct writeback_control *wbc,
92 struct gfs2_trans *tr)
93__releases(&sdp->sd_ail_lock)
94__acquires(&sdp->sd_ail_lock)
95{
96 struct gfs2_glock *gl = NULL;
97 struct address_space *mapping;
98 struct gfs2_bufdata *bd, *s;
99 struct buffer_head *bh;
100 int ret = 0;
101
102 list_for_each_entry_safe_reverse(bd, s, &tr->tr_ail1_list, bd_ail_st_list) {
103 bh = bd->bd_bh;
104
105 gfs2_assert(sdp, bd->bd_tr == tr);
106
107 if (!buffer_busy(bh)) {
108 if (buffer_uptodate(bh)) {
109 list_move(&bd->bd_ail_st_list,
110 &tr->tr_ail2_list);
111 continue;
112 }
113 if (!cmpxchg(&sdp->sd_log_error, 0, -EIO)) {
114 gfs2_io_error_bh(sdp, bh);
115 gfs2_withdraw_delayed(sdp);
116 }
117 }
118
119 if (gfs2_withdrawn(sdp)) {
120 gfs2_remove_from_ail(bd);
121 continue;
122 }
123 if (!buffer_dirty(bh))
124 continue;
125 if (gl == bd->bd_gl)
126 continue;
127 gl = bd->bd_gl;
128 list_move(&bd->bd_ail_st_list, &tr->tr_ail1_list);
129 mapping = bh->b_page->mapping;
130 if (!mapping)
131 continue;
132 spin_unlock(&sdp->sd_ail_lock);
133 ret = generic_writepages(mapping, wbc);
134 spin_lock(&sdp->sd_ail_lock);
135 if (ret || wbc->nr_to_write <= 0)
136 break;
137 return -EBUSY;
138 }
139
140 return ret;
141}
142
143static void dump_ail_list(struct gfs2_sbd *sdp)
144{
145 struct gfs2_trans *tr;
146 struct gfs2_bufdata *bd;
147 struct buffer_head *bh;
148
149 list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) {
150 list_for_each_entry_reverse(bd, &tr->tr_ail1_list,
151 bd_ail_st_list) {
152 bh = bd->bd_bh;
153 fs_err(sdp, "bd %p: blk:0x%llx bh=%p ", bd,
154 (unsigned long long)bd->bd_blkno, bh);
155 if (!bh) {
156 fs_err(sdp, "\n");
157 continue;
158 }
159 fs_err(sdp, "0x%llx up2:%d dirt:%d lkd:%d req:%d "
160 "map:%d new:%d ar:%d aw:%d delay:%d "
161 "io err:%d unwritten:%d dfr:%d pin:%d esc:%d\n",
162 (unsigned long long)bh->b_blocknr,
163 buffer_uptodate(bh), buffer_dirty(bh),
164 buffer_locked(bh), buffer_req(bh),
165 buffer_mapped(bh), buffer_new(bh),
166 buffer_async_read(bh), buffer_async_write(bh),
167 buffer_delay(bh), buffer_write_io_error(bh),
168 buffer_unwritten(bh),
169 buffer_defer_completion(bh),
170 buffer_pinned(bh), buffer_escaped(bh));
171 }
172 }
173}
174
175
176
177
178
179
180
181
182
183
184void gfs2_ail1_flush(struct gfs2_sbd *sdp, struct writeback_control *wbc)
185{
186 struct list_head *head = &sdp->sd_ail1_list;
187 struct gfs2_trans *tr;
188 struct blk_plug plug;
189 int ret;
190 unsigned long flush_start = jiffies;
191
192 trace_gfs2_ail_flush(sdp, wbc, 1);
193 blk_start_plug(&plug);
194 spin_lock(&sdp->sd_ail_lock);
195restart:
196 ret = 0;
197 if (time_after(jiffies, flush_start + (HZ * 600))) {
198 fs_err(sdp, "Error: In %s for ten minutes! t=%d\n",
199 __func__, current->journal_info ? 1 : 0);
200 dump_ail_list(sdp);
201 goto out;
202 }
203 list_for_each_entry_reverse(tr, head, tr_list) {
204 if (wbc->nr_to_write <= 0)
205 break;
206 ret = gfs2_ail1_start_one(sdp, wbc, tr);
207 if (ret) {
208 if (ret == -EBUSY)
209 goto restart;
210 break;
211 }
212 }
213out:
214 spin_unlock(&sdp->sd_ail_lock);
215 blk_finish_plug(&plug);
216 if (ret) {
217 gfs2_lm(sdp, "gfs2_ail1_start_one (generic_writepages) "
218 "returned: %d\n", ret);
219 gfs2_withdraw(sdp);
220 }
221 trace_gfs2_ail_flush(sdp, wbc, 0);
222}
223
224
225
226
227
228
229static void gfs2_ail1_start(struct gfs2_sbd *sdp)
230{
231 struct writeback_control wbc = {
232 .sync_mode = WB_SYNC_NONE,
233 .nr_to_write = LONG_MAX,
234 .range_start = 0,
235 .range_end = LLONG_MAX,
236 };
237
238 return gfs2_ail1_flush(sdp, &wbc);
239}
240
241
242
243
244
245
246
247
248
249static void gfs2_ail1_empty_one(struct gfs2_sbd *sdp, struct gfs2_trans *tr,
250 int *max_revokes)
251{
252 struct gfs2_bufdata *bd, *s;
253 struct buffer_head *bh;
254
255 list_for_each_entry_safe_reverse(bd, s, &tr->tr_ail1_list,
256 bd_ail_st_list) {
257 bh = bd->bd_bh;
258 gfs2_assert(sdp, bd->bd_tr == tr);
259
260
261
262
263
264
265
266
267
268 if (!sdp->sd_log_error && buffer_busy(bh))
269 continue;
270 if (!buffer_uptodate(bh) &&
271 !cmpxchg(&sdp->sd_log_error, 0, -EIO)) {
272 gfs2_io_error_bh(sdp, bh);
273 gfs2_withdraw_delayed(sdp);
274 }
275
276
277
278
279
280
281 if (*max_revokes && list_empty(&bd->bd_list)) {
282 gfs2_add_revoke(sdp, bd);
283 (*max_revokes)--;
284 continue;
285 }
286 list_move(&bd->bd_ail_st_list, &tr->tr_ail2_list);
287 }
288}
289
290
291
292
293
294
295
296
297
298static int gfs2_ail1_empty(struct gfs2_sbd *sdp, int max_revokes)
299{
300 struct gfs2_trans *tr, *s;
301 int oldest_tr = 1;
302 int ret;
303
304 spin_lock(&sdp->sd_ail_lock);
305 list_for_each_entry_safe_reverse(tr, s, &sdp->sd_ail1_list, tr_list) {
306 gfs2_ail1_empty_one(sdp, tr, &max_revokes);
307 if (list_empty(&tr->tr_ail1_list) && oldest_tr)
308 list_move(&tr->tr_list, &sdp->sd_ail2_list);
309 else
310 oldest_tr = 0;
311 }
312 ret = list_empty(&sdp->sd_ail1_list);
313 spin_unlock(&sdp->sd_ail_lock);
314
315 if (test_bit(SDF_WITHDRAWING, &sdp->sd_flags)) {
316 gfs2_lm(sdp, "fatal: I/O error(s)\n");
317 gfs2_withdraw(sdp);
318 }
319
320 return ret;
321}
322
323static void gfs2_ail1_wait(struct gfs2_sbd *sdp)
324{
325 struct gfs2_trans *tr;
326 struct gfs2_bufdata *bd;
327 struct buffer_head *bh;
328
329 spin_lock(&sdp->sd_ail_lock);
330 list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) {
331 list_for_each_entry(bd, &tr->tr_ail1_list, bd_ail_st_list) {
332 bh = bd->bd_bh;
333 if (!buffer_locked(bh))
334 continue;
335 get_bh(bh);
336 spin_unlock(&sdp->sd_ail_lock);
337 wait_on_buffer(bh);
338 brelse(bh);
339 return;
340 }
341 }
342 spin_unlock(&sdp->sd_ail_lock);
343}
344
345
346
347
348
349static void gfs2_ail_empty_tr(struct gfs2_sbd *sdp, struct gfs2_trans *tr,
350 struct list_head *head)
351{
352 struct gfs2_bufdata *bd;
353
354 while (!list_empty(head)) {
355 bd = list_first_entry(head, struct gfs2_bufdata,
356 bd_ail_st_list);
357 gfs2_assert(sdp, bd->bd_tr == tr);
358 gfs2_remove_from_ail(bd);
359 }
360}
361
362static void ail2_empty(struct gfs2_sbd *sdp, unsigned int new_tail)
363{
364 struct gfs2_trans *tr, *safe;
365 unsigned int old_tail = sdp->sd_log_tail;
366 int wrap = (new_tail < old_tail);
367 int a, b, rm;
368
369 spin_lock(&sdp->sd_ail_lock);
370
371 list_for_each_entry_safe(tr, safe, &sdp->sd_ail2_list, tr_list) {
372 a = (old_tail <= tr->tr_first);
373 b = (tr->tr_first < new_tail);
374 rm = (wrap) ? (a || b) : (a && b);
375 if (!rm)
376 continue;
377
378 gfs2_ail_empty_tr(sdp, tr, &tr->tr_ail2_list);
379 list_del(&tr->tr_list);
380 gfs2_assert_warn(sdp, list_empty(&tr->tr_ail1_list));
381 gfs2_assert_warn(sdp, list_empty(&tr->tr_ail2_list));
382 gfs2_trans_free(sdp, tr);
383 }
384
385 spin_unlock(&sdp->sd_ail_lock);
386}
387
388
389
390
391
392
393
394
395void gfs2_log_release(struct gfs2_sbd *sdp, unsigned int blks)
396{
397
398 atomic_add(blks, &sdp->sd_log_blks_free);
399 trace_gfs2_log_blocks(sdp, blks);
400 gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
401 sdp->sd_jdesc->jd_blocks);
402 up_read(&sdp->sd_log_flush_lock);
403}
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425int gfs2_log_reserve(struct gfs2_sbd *sdp, unsigned int blks)
426{
427 int ret = 0;
428 unsigned reserved_blks = 7 * (4096 / sdp->sd_vfs->s_blocksize);
429 unsigned wanted = blks + reserved_blks;
430 DEFINE_WAIT(wait);
431 int did_wait = 0;
432 unsigned int free_blocks;
433
434 if (gfs2_assert_warn(sdp, blks) ||
435 gfs2_assert_warn(sdp, blks <= sdp->sd_jdesc->jd_blocks))
436 return -EINVAL;
437 atomic_add(blks, &sdp->sd_log_blks_needed);
438retry:
439 free_blocks = atomic_read(&sdp->sd_log_blks_free);
440 if (unlikely(free_blocks <= wanted)) {
441 do {
442 prepare_to_wait_exclusive(&sdp->sd_log_waitq, &wait,
443 TASK_UNINTERRUPTIBLE);
444 wake_up(&sdp->sd_logd_waitq);
445 did_wait = 1;
446 if (atomic_read(&sdp->sd_log_blks_free) <= wanted)
447 io_schedule();
448 free_blocks = atomic_read(&sdp->sd_log_blks_free);
449 } while(free_blocks <= wanted);
450 finish_wait(&sdp->sd_log_waitq, &wait);
451 }
452 atomic_inc(&sdp->sd_reserving_log);
453 if (atomic_cmpxchg(&sdp->sd_log_blks_free, free_blocks,
454 free_blocks - blks) != free_blocks) {
455 if (atomic_dec_and_test(&sdp->sd_reserving_log))
456 wake_up(&sdp->sd_reserving_log_wait);
457 goto retry;
458 }
459 atomic_sub(blks, &sdp->sd_log_blks_needed);
460 trace_gfs2_log_blocks(sdp, -blks);
461
462
463
464
465
466 if (unlikely(did_wait))
467 wake_up(&sdp->sd_log_waitq);
468
469 down_read(&sdp->sd_log_flush_lock);
470 if (unlikely(!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))) {
471 gfs2_log_release(sdp, blks);
472 ret = -EROFS;
473 }
474 if (atomic_dec_and_test(&sdp->sd_reserving_log))
475 wake_up(&sdp->sd_reserving_log_wait);
476 return ret;
477}
478
479
480
481
482
483
484
485
486
487
488
489
490
491static inline unsigned int log_distance(struct gfs2_sbd *sdp, unsigned int newer,
492 unsigned int older)
493{
494 int dist;
495
496 dist = newer - older;
497 if (dist < 0)
498 dist += sdp->sd_jdesc->jd_blocks;
499
500 return dist;
501}
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528static unsigned int calc_reserved(struct gfs2_sbd *sdp)
529{
530 unsigned int reserved = 0;
531 unsigned int mbuf;
532 unsigned int dbuf;
533 struct gfs2_trans *tr = sdp->sd_log_tr;
534
535 if (tr) {
536 mbuf = tr->tr_num_buf_new - tr->tr_num_buf_rm;
537 dbuf = tr->tr_num_databuf_new - tr->tr_num_databuf_rm;
538 reserved = mbuf + dbuf;
539
540 reserved += DIV_ROUND_UP(mbuf, buf_limit(sdp));
541 reserved += DIV_ROUND_UP(dbuf, databuf_limit(sdp));
542 }
543
544 if (sdp->sd_log_committed_revoke > 0)
545 reserved += gfs2_struct2blk(sdp, sdp->sd_log_committed_revoke);
546
547 if (reserved)
548 reserved++;
549 return reserved;
550}
551
552static unsigned int current_tail(struct gfs2_sbd *sdp)
553{
554 struct gfs2_trans *tr;
555 unsigned int tail;
556
557 spin_lock(&sdp->sd_ail_lock);
558
559 if (list_empty(&sdp->sd_ail1_list)) {
560 tail = sdp->sd_log_head;
561 } else {
562 tr = list_last_entry(&sdp->sd_ail1_list, struct gfs2_trans,
563 tr_list);
564 tail = tr->tr_first;
565 }
566
567 spin_unlock(&sdp->sd_ail_lock);
568
569 return tail;
570}
571
572static void log_pull_tail(struct gfs2_sbd *sdp, unsigned int new_tail)
573{
574 unsigned int dist = log_distance(sdp, new_tail, sdp->sd_log_tail);
575
576 ail2_empty(sdp, new_tail);
577
578 atomic_add(dist, &sdp->sd_log_blks_free);
579 trace_gfs2_log_blocks(sdp, dist);
580 gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
581 sdp->sd_jdesc->jd_blocks);
582
583 sdp->sd_log_tail = new_tail;
584}
585
586
587void log_flush_wait(struct gfs2_sbd *sdp)
588{
589 DEFINE_WAIT(wait);
590
591 if (atomic_read(&sdp->sd_log_in_flight)) {
592 do {
593 prepare_to_wait(&sdp->sd_log_flush_wait, &wait,
594 TASK_UNINTERRUPTIBLE);
595 if (atomic_read(&sdp->sd_log_in_flight))
596 io_schedule();
597 } while(atomic_read(&sdp->sd_log_in_flight));
598 finish_wait(&sdp->sd_log_flush_wait, &wait);
599 }
600}
601
602static int ip_cmp(void *priv, struct list_head *a, struct list_head *b)
603{
604 struct gfs2_inode *ipa, *ipb;
605
606 ipa = list_entry(a, struct gfs2_inode, i_ordered);
607 ipb = list_entry(b, struct gfs2_inode, i_ordered);
608
609 if (ipa->i_no_addr < ipb->i_no_addr)
610 return -1;
611 if (ipa->i_no_addr > ipb->i_no_addr)
612 return 1;
613 return 0;
614}
615
616static void __ordered_del_inode(struct gfs2_inode *ip)
617{
618 if (!list_empty(&ip->i_ordered))
619 list_del_init(&ip->i_ordered);
620}
621
622static void gfs2_ordered_write(struct gfs2_sbd *sdp)
623{
624 struct gfs2_inode *ip;
625 LIST_HEAD(written);
626
627 spin_lock(&sdp->sd_ordered_lock);
628 list_sort(NULL, &sdp->sd_log_ordered, &ip_cmp);
629 while (!list_empty(&sdp->sd_log_ordered)) {
630 ip = list_first_entry(&sdp->sd_log_ordered, struct gfs2_inode, i_ordered);
631 if (ip->i_inode.i_mapping->nrpages == 0) {
632 __ordered_del_inode(ip);
633 continue;
634 }
635 list_move(&ip->i_ordered, &written);
636 spin_unlock(&sdp->sd_ordered_lock);
637 filemap_fdatawrite(ip->i_inode.i_mapping);
638 spin_lock(&sdp->sd_ordered_lock);
639 }
640 list_splice(&written, &sdp->sd_log_ordered);
641 spin_unlock(&sdp->sd_ordered_lock);
642}
643
644static void gfs2_ordered_wait(struct gfs2_sbd *sdp)
645{
646 struct gfs2_inode *ip;
647
648 spin_lock(&sdp->sd_ordered_lock);
649 while (!list_empty(&sdp->sd_log_ordered)) {
650 ip = list_first_entry(&sdp->sd_log_ordered, struct gfs2_inode, i_ordered);
651 __ordered_del_inode(ip);
652 if (ip->i_inode.i_mapping->nrpages == 0)
653 continue;
654 spin_unlock(&sdp->sd_ordered_lock);
655 filemap_fdatawait(ip->i_inode.i_mapping);
656 spin_lock(&sdp->sd_ordered_lock);
657 }
658 spin_unlock(&sdp->sd_ordered_lock);
659}
660
661void gfs2_ordered_del_inode(struct gfs2_inode *ip)
662{
663 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
664
665 spin_lock(&sdp->sd_ordered_lock);
666 __ordered_del_inode(ip);
667 spin_unlock(&sdp->sd_ordered_lock);
668}
669
670void gfs2_add_revoke(struct gfs2_sbd *sdp, struct gfs2_bufdata *bd)
671{
672 struct buffer_head *bh = bd->bd_bh;
673 struct gfs2_glock *gl = bd->bd_gl;
674
675 sdp->sd_log_num_revoke++;
676 if (atomic_inc_return(&gl->gl_revokes) == 1)
677 gfs2_glock_hold(gl);
678 bh->b_private = NULL;
679 bd->bd_blkno = bh->b_blocknr;
680 gfs2_remove_from_ail(bd);
681 bd->bd_bh = NULL;
682 set_bit(GLF_LFLUSH, &gl->gl_flags);
683 list_add(&bd->bd_list, &sdp->sd_log_revokes);
684}
685
686void gfs2_glock_remove_revoke(struct gfs2_glock *gl)
687{
688 if (atomic_dec_return(&gl->gl_revokes) == 0) {
689 clear_bit(GLF_LFLUSH, &gl->gl_flags);
690 gfs2_glock_queue_put(gl);
691 }
692}
693
694
695
696
697
698
699
700
701
702
703
704
705
706void gfs2_write_revokes(struct gfs2_sbd *sdp)
707{
708
709 int max_revokes = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / sizeof(u64);
710
711 gfs2_log_lock(sdp);
712 while (sdp->sd_log_num_revoke > max_revokes)
713 max_revokes += (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header)) / sizeof(u64);
714 max_revokes -= sdp->sd_log_num_revoke;
715 if (!sdp->sd_log_num_revoke) {
716 atomic_dec(&sdp->sd_log_blks_free);
717
718
719 if (!sdp->sd_log_blks_reserved)
720 atomic_dec(&sdp->sd_log_blks_free);
721 }
722 gfs2_ail1_empty(sdp, max_revokes);
723 gfs2_log_unlock(sdp);
724
725 if (!sdp->sd_log_num_revoke) {
726 atomic_inc(&sdp->sd_log_blks_free);
727 if (!sdp->sd_log_blks_reserved)
728 atomic_inc(&sdp->sd_log_blks_free);
729 }
730}
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745void gfs2_write_log_header(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd,
746 u64 seq, u32 tail, u32 lblock, u32 flags,
747 int op_flags)
748{
749 struct gfs2_log_header *lh;
750 u32 hash, crc;
751 struct page *page;
752 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
753 struct timespec64 tv;
754 struct super_block *sb = sdp->sd_vfs;
755 u64 dblock;
756
757 if (gfs2_withdrawn(sdp))
758 goto out;
759
760 page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
761 lh = page_address(page);
762 clear_page(lh);
763
764 lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
765 lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH);
766 lh->lh_header.__pad0 = cpu_to_be64(0);
767 lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH);
768 lh->lh_header.mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid);
769 lh->lh_sequence = cpu_to_be64(seq);
770 lh->lh_flags = cpu_to_be32(flags);
771 lh->lh_tail = cpu_to_be32(tail);
772 lh->lh_blkno = cpu_to_be32(lblock);
773 hash = ~crc32(~0, lh, LH_V1_SIZE);
774 lh->lh_hash = cpu_to_be32(hash);
775
776 ktime_get_coarse_real_ts64(&tv);
777 lh->lh_nsec = cpu_to_be32(tv.tv_nsec);
778 lh->lh_sec = cpu_to_be64(tv.tv_sec);
779 if (!list_empty(&jd->extent_list))
780 dblock = gfs2_log_bmap(jd, lblock);
781 else {
782 int ret = gfs2_lblk_to_dblk(jd->jd_inode, lblock, &dblock);
783 if (gfs2_assert_withdraw(sdp, ret == 0))
784 return;
785 }
786 lh->lh_addr = cpu_to_be64(dblock);
787 lh->lh_jinode = cpu_to_be64(GFS2_I(jd->jd_inode)->i_no_addr);
788
789
790
791
792 if (!(flags & GFS2_LOG_HEAD_RECOVERY)) {
793 lh->lh_statfs_addr =
794 cpu_to_be64(GFS2_I(sdp->sd_sc_inode)->i_no_addr);
795 lh->lh_quota_addr =
796 cpu_to_be64(GFS2_I(sdp->sd_qc_inode)->i_no_addr);
797
798 spin_lock(&sdp->sd_statfs_spin);
799 lh->lh_local_total = cpu_to_be64(l_sc->sc_total);
800 lh->lh_local_free = cpu_to_be64(l_sc->sc_free);
801 lh->lh_local_dinodes = cpu_to_be64(l_sc->sc_dinodes);
802 spin_unlock(&sdp->sd_statfs_spin);
803 }
804
805 BUILD_BUG_ON(offsetof(struct gfs2_log_header, lh_crc) != LH_V1_SIZE);
806
807 crc = crc32c(~0, (void *)lh + LH_V1_SIZE + 4,
808 sb->s_blocksize - LH_V1_SIZE - 4);
809 lh->lh_crc = cpu_to_be32(crc);
810
811 gfs2_log_write(sdp, page, sb->s_blocksize, 0, dblock);
812 gfs2_log_submit_bio(&sdp->sd_log_bio, REQ_OP_WRITE | op_flags);
813out:
814 log_flush_wait(sdp);
815}
816
817
818
819
820
821
822
823
824
825static void log_write_header(struct gfs2_sbd *sdp, u32 flags)
826{
827 unsigned int tail;
828 int op_flags = REQ_PREFLUSH | REQ_FUA | REQ_META | REQ_SYNC;
829 enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state);
830
831 gfs2_assert_withdraw(sdp, (state != SFS_FROZEN));
832 tail = current_tail(sdp);
833
834 if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags)) {
835 gfs2_ordered_wait(sdp);
836 log_flush_wait(sdp);
837 op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
838 }
839 sdp->sd_log_idle = (tail == sdp->sd_log_flush_head);
840 gfs2_write_log_header(sdp, sdp->sd_jdesc, sdp->sd_log_sequence++, tail,
841 sdp->sd_log_flush_head, flags, op_flags);
842 gfs2_log_incr_head(sdp);
843
844 if (sdp->sd_log_tail != tail)
845 log_pull_tail(sdp, tail);
846}
847
848
849
850
851
852static void ail_drain(struct gfs2_sbd *sdp)
853{
854 struct gfs2_trans *tr;
855
856 spin_lock(&sdp->sd_ail_lock);
857
858
859
860
861
862
863
864 while (!list_empty(&sdp->sd_ail1_list)) {
865 tr = list_first_entry(&sdp->sd_ail1_list, struct gfs2_trans,
866 tr_list);
867 gfs2_ail_empty_tr(sdp, tr, &tr->tr_ail1_list);
868 gfs2_ail_empty_tr(sdp, tr, &tr->tr_ail2_list);
869 list_del(&tr->tr_list);
870 gfs2_trans_free(sdp, tr);
871 }
872 while (!list_empty(&sdp->sd_ail2_list)) {
873 tr = list_first_entry(&sdp->sd_ail2_list, struct gfs2_trans,
874 tr_list);
875 gfs2_ail_empty_tr(sdp, tr, &tr->tr_ail2_list);
876 list_del(&tr->tr_list);
877 gfs2_trans_free(sdp, tr);
878 }
879 spin_unlock(&sdp->sd_ail_lock);
880}
881
882
883
884
885
886static void empty_ail1_list(struct gfs2_sbd *sdp)
887{
888 unsigned long start = jiffies;
889
890 for (;;) {
891 if (time_after(jiffies, start + (HZ * 600))) {
892 fs_err(sdp, "Error: In %s for 10 minutes! t=%d\n",
893 __func__, current->journal_info ? 1 : 0);
894 dump_ail_list(sdp);
895 return;
896 }
897 gfs2_ail1_start(sdp);
898 gfs2_ail1_wait(sdp);
899 if (gfs2_ail1_empty(sdp, 0))
900 return;
901 }
902}
903
904
905
906
907
908
909
910
911
912static void trans_drain(struct gfs2_trans *tr)
913{
914 struct gfs2_bufdata *bd;
915 struct list_head *head;
916
917 if (!tr)
918 return;
919
920 head = &tr->tr_buf;
921 while (!list_empty(head)) {
922 bd = list_first_entry(head, struct gfs2_bufdata, bd_list);
923 list_del_init(&bd->bd_list);
924 kmem_cache_free(gfs2_bufdata_cachep, bd);
925 }
926 head = &tr->tr_databuf;
927 while (!list_empty(head)) {
928 bd = list_first_entry(head, struct gfs2_bufdata, bd_list);
929 list_del_init(&bd->bd_list);
930 kmem_cache_free(gfs2_bufdata_cachep, bd);
931 }
932}
933
934
935
936
937
938
939
940
941
942void gfs2_log_flush(struct gfs2_sbd *sdp, struct gfs2_glock *gl, u32 flags)
943{
944 struct gfs2_trans *tr = NULL;
945 enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state);
946
947 down_write(&sdp->sd_log_flush_lock);
948
949
950
951
952
953 if (gfs2_withdrawn(sdp))
954 goto out;
955
956
957 if (gl && !test_bit(GLF_LFLUSH, &gl->gl_flags)) {
958 up_write(&sdp->sd_log_flush_lock);
959 return;
960 }
961 trace_gfs2_log_flush(sdp, 1, flags);
962
963 if (flags & GFS2_LOG_HEAD_FLUSH_SHUTDOWN)
964 clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
965
966 sdp->sd_log_flush_head = sdp->sd_log_head;
967 tr = sdp->sd_log_tr;
968 if (tr) {
969 sdp->sd_log_tr = NULL;
970 tr->tr_first = sdp->sd_log_flush_head;
971 if (unlikely (state == SFS_FROZEN))
972 if (gfs2_assert_withdraw_delayed(sdp,
973 !tr->tr_num_buf_new && !tr->tr_num_databuf_new))
974 goto out;
975 }
976
977 if (unlikely(state == SFS_FROZEN))
978 if (gfs2_assert_withdraw_delayed(sdp, !sdp->sd_log_num_revoke))
979 goto out;
980 if (gfs2_assert_withdraw_delayed(sdp,
981 sdp->sd_log_num_revoke == sdp->sd_log_committed_revoke))
982 goto out;
983
984 gfs2_ordered_write(sdp);
985 if (gfs2_withdrawn(sdp))
986 goto out;
987 lops_before_commit(sdp, tr);
988 if (gfs2_withdrawn(sdp))
989 goto out;
990 gfs2_log_submit_bio(&sdp->sd_log_bio, REQ_OP_WRITE);
991 if (gfs2_withdrawn(sdp))
992 goto out;
993
994 if (sdp->sd_log_head != sdp->sd_log_flush_head) {
995 log_flush_wait(sdp);
996 log_write_header(sdp, flags);
997 } else if (sdp->sd_log_tail != current_tail(sdp) && !sdp->sd_log_idle){
998 atomic_dec(&sdp->sd_log_blks_free);
999 trace_gfs2_log_blocks(sdp, -1);
1000 log_write_header(sdp, flags);
1001 }
1002 if (gfs2_withdrawn(sdp))
1003 goto out;
1004 lops_after_commit(sdp, tr);
1005
1006 gfs2_log_lock(sdp);
1007 sdp->sd_log_head = sdp->sd_log_flush_head;
1008 sdp->sd_log_blks_reserved = 0;
1009 sdp->sd_log_committed_revoke = 0;
1010
1011 spin_lock(&sdp->sd_ail_lock);
1012 if (tr && !list_empty(&tr->tr_ail1_list)) {
1013 list_add(&tr->tr_list, &sdp->sd_ail1_list);
1014 tr = NULL;
1015 }
1016 spin_unlock(&sdp->sd_ail_lock);
1017 gfs2_log_unlock(sdp);
1018
1019 if (!(flags & GFS2_LOG_HEAD_FLUSH_NORMAL)) {
1020 if (!sdp->sd_log_idle) {
1021 empty_ail1_list(sdp);
1022 if (gfs2_withdrawn(sdp))
1023 goto out;
1024 atomic_dec(&sdp->sd_log_blks_free);
1025 trace_gfs2_log_blocks(sdp, -1);
1026 log_write_header(sdp, flags);
1027 sdp->sd_log_head = sdp->sd_log_flush_head;
1028 }
1029 if (flags & (GFS2_LOG_HEAD_FLUSH_SHUTDOWN |
1030 GFS2_LOG_HEAD_FLUSH_FREEZE))
1031 gfs2_log_shutdown(sdp);
1032 if (flags & GFS2_LOG_HEAD_FLUSH_FREEZE)
1033 atomic_set(&sdp->sd_freeze_state, SFS_FROZEN);
1034 }
1035
1036out:
1037 if (gfs2_withdrawn(sdp)) {
1038 trans_drain(tr);
1039
1040
1041
1042
1043
1044
1045 spin_lock(&sdp->sd_ail_lock);
1046 if (tr && list_empty(&tr->tr_list))
1047 list_add(&tr->tr_list, &sdp->sd_ail1_list);
1048 spin_unlock(&sdp->sd_ail_lock);
1049 ail_drain(sdp);
1050 tr = NULL;
1051 }
1052
1053 trace_gfs2_log_flush(sdp, 0, flags);
1054 up_write(&sdp->sd_log_flush_lock);
1055
1056 gfs2_trans_free(sdp, tr);
1057}
1058
1059
1060
1061
1062
1063
1064
1065static void gfs2_merge_trans(struct gfs2_sbd *sdp, struct gfs2_trans *new)
1066{
1067 struct gfs2_trans *old = sdp->sd_log_tr;
1068
1069 WARN_ON_ONCE(!test_bit(TR_ATTACHED, &old->tr_flags));
1070
1071 old->tr_num_buf_new += new->tr_num_buf_new;
1072 old->tr_num_databuf_new += new->tr_num_databuf_new;
1073 old->tr_num_buf_rm += new->tr_num_buf_rm;
1074 old->tr_num_databuf_rm += new->tr_num_databuf_rm;
1075 old->tr_num_revoke += new->tr_num_revoke;
1076 old->tr_num_revoke_rm += new->tr_num_revoke_rm;
1077
1078 list_splice_tail_init(&new->tr_databuf, &old->tr_databuf);
1079 list_splice_tail_init(&new->tr_buf, &old->tr_buf);
1080
1081 spin_lock(&sdp->sd_ail_lock);
1082 list_splice_tail_init(&new->tr_ail1_list, &old->tr_ail1_list);
1083 list_splice_tail_init(&new->tr_ail2_list, &old->tr_ail2_list);
1084 spin_unlock(&sdp->sd_ail_lock);
1085}
1086
1087static void log_refund(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
1088{
1089 unsigned int reserved;
1090 unsigned int unused;
1091 unsigned int maxres;
1092
1093 gfs2_log_lock(sdp);
1094
1095 if (sdp->sd_log_tr) {
1096 gfs2_merge_trans(sdp, tr);
1097 } else if (tr->tr_num_buf_new || tr->tr_num_databuf_new) {
1098 gfs2_assert_withdraw(sdp, test_bit(TR_ALLOCED, &tr->tr_flags));
1099 sdp->sd_log_tr = tr;
1100 set_bit(TR_ATTACHED, &tr->tr_flags);
1101 }
1102
1103 sdp->sd_log_committed_revoke += tr->tr_num_revoke - tr->tr_num_revoke_rm;
1104 reserved = calc_reserved(sdp);
1105 maxres = sdp->sd_log_blks_reserved + tr->tr_reserved;
1106 gfs2_assert_withdraw(sdp, maxres >= reserved);
1107 unused = maxres - reserved;
1108 atomic_add(unused, &sdp->sd_log_blks_free);
1109 trace_gfs2_log_blocks(sdp, unused);
1110 gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
1111 sdp->sd_jdesc->jd_blocks);
1112 sdp->sd_log_blks_reserved = reserved;
1113
1114 gfs2_log_unlock(sdp);
1115}
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132void gfs2_log_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
1133{
1134 log_refund(sdp, tr);
1135
1136 if (atomic_read(&sdp->sd_log_pinned) > atomic_read(&sdp->sd_log_thresh1) ||
1137 ((sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free)) >
1138 atomic_read(&sdp->sd_log_thresh2)))
1139 wake_up(&sdp->sd_logd_waitq);
1140}
1141
1142
1143
1144
1145
1146
1147
1148static void gfs2_log_shutdown(struct gfs2_sbd *sdp)
1149{
1150 gfs2_assert_withdraw(sdp, !sdp->sd_log_blks_reserved);
1151 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
1152 gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail1_list));
1153
1154 sdp->sd_log_flush_head = sdp->sd_log_head;
1155
1156 log_write_header(sdp, GFS2_LOG_HEAD_UNMOUNT | GFS2_LFC_SHUTDOWN);
1157
1158 gfs2_assert_warn(sdp, sdp->sd_log_head == sdp->sd_log_tail);
1159 gfs2_assert_warn(sdp, list_empty(&sdp->sd_ail2_list));
1160
1161 sdp->sd_log_head = sdp->sd_log_flush_head;
1162 sdp->sd_log_tail = sdp->sd_log_head;
1163}
1164
1165static inline int gfs2_jrnl_flush_reqd(struct gfs2_sbd *sdp)
1166{
1167 return (atomic_read(&sdp->sd_log_pinned) +
1168 atomic_read(&sdp->sd_log_blks_needed) >=
1169 atomic_read(&sdp->sd_log_thresh1));
1170}
1171
1172static inline int gfs2_ail_flush_reqd(struct gfs2_sbd *sdp)
1173{
1174 unsigned int used_blocks = sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free);
1175
1176 if (test_and_clear_bit(SDF_FORCE_AIL_FLUSH, &sdp->sd_flags))
1177 return 1;
1178
1179 return used_blocks + atomic_read(&sdp->sd_log_blks_needed) >=
1180 atomic_read(&sdp->sd_log_thresh2);
1181}
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191int gfs2_logd(void *data)
1192{
1193 struct gfs2_sbd *sdp = data;
1194 unsigned long t = 1;
1195 DEFINE_WAIT(wait);
1196 bool did_flush;
1197
1198 while (!kthread_should_stop()) {
1199
1200 if (gfs2_withdrawn(sdp)) {
1201 msleep_interruptible(HZ);
1202 continue;
1203 }
1204
1205 if (sdp->sd_log_error) {
1206 gfs2_lm(sdp,
1207 "GFS2: fsid=%s: error %d: "
1208 "withdrawing the file system to "
1209 "prevent further damage.\n",
1210 sdp->sd_fsname, sdp->sd_log_error);
1211 gfs2_withdraw(sdp);
1212 continue;
1213 }
1214
1215 did_flush = false;
1216 if (gfs2_jrnl_flush_reqd(sdp) || t == 0) {
1217 gfs2_ail1_empty(sdp, 0);
1218 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
1219 GFS2_LFC_LOGD_JFLUSH_REQD);
1220 did_flush = true;
1221 }
1222
1223 if (gfs2_ail_flush_reqd(sdp)) {
1224 gfs2_ail1_start(sdp);
1225 gfs2_ail1_wait(sdp);
1226 gfs2_ail1_empty(sdp, 0);
1227 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
1228 GFS2_LFC_LOGD_AIL_FLUSH_REQD);
1229 did_flush = true;
1230 }
1231
1232 if (!gfs2_ail_flush_reqd(sdp) || did_flush)
1233 wake_up(&sdp->sd_log_waitq);
1234
1235 t = gfs2_tune_get(sdp, gt_logd_secs) * HZ;
1236
1237 try_to_freeze();
1238
1239 do {
1240 prepare_to_wait(&sdp->sd_logd_waitq, &wait,
1241 TASK_INTERRUPTIBLE);
1242 if (!gfs2_ail_flush_reqd(sdp) &&
1243 !gfs2_jrnl_flush_reqd(sdp) &&
1244 !kthread_should_stop())
1245 t = schedule_timeout(t);
1246 } while(t && !gfs2_ail_flush_reqd(sdp) &&
1247 !gfs2_jrnl_flush_reqd(sdp) &&
1248 !kthread_should_stop());
1249 finish_wait(&sdp->sd_logd_waitq, &wait);
1250 }
1251
1252 return 0;
1253}
1254
1255