Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
4 * Copyright (C) 2004-2007 Red Hat, Inc. All rights reserved.
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 * gfs2_struct2blk - compute stuff
39 * @sdp: the filesystem
40 * @nstruct: the number of structures
41 *
42 * Compute the number of log descriptor blocks needed to hold a certain number
43 * of structures of a certain size.
44 *
45 * Returns: the number of blocks needed (minimum is always 1)
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 * gfs2_remove_from_ail - Remove an entry from the ail lists, updating counters
66 * @mapping: The associated mapping (maybe NULL)
67 * @bd: The gfs2_bufdata to remove
68 *
69 * The ail lock _must_ be held when calling this function
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 * gfs2_ail1_start_one - Start I/O on a part of the AIL
84 * @sdp: the filesystem
85 * @wbc: The writeback control structure
86 * @ai: The ail structure
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 * gfs2_ail1_flush - start writeback of some ail1 entries
177 * @sdp: The super block
178 * @wbc: The writeback control structure
179 *
180 * Writes back some ail1 entries, according to the limits in the
181 * writeback control structure
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 * gfs2_ail1_start - start writeback of all ail1 entries
226 * @sdp: The superblock
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 * gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced
243 * @sdp: the filesystem
244 * @tr: the transaction
245 * @max_revokes: If nonzero, issue revokes for the bd items for written buffers
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 * If another process flagged an io error, e.g. writing to the
261 * journal, error all other bhs and move them off the ail1 to
262 * prevent a tight loop when unmount tries to flush ail1,
263 * regardless of whether they're still busy. If no outside
264 * errors were found and the buffer is busy, move to the next.
265 * If the ail buffer is not busy and caught an error, flag it
266 * for others.
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 * If we have space for revokes and the bd is no longer on any
277 * buf list, we can just add a revoke for it immediately and
278 * avoid having to put it on the ail2 list, where it would need
279 * to be revoked later.
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 * gfs2_ail1_empty - Try to empty the ail1 lists
292 * @sdp: The superblock
293 * @max_revokes: If non-zero, add revokes where appropriate
294 *
295 * Tries to empty the ail1 lists, starting with the oldest first
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 * gfs2_ail_empty_tr - empty one of the ail lists for a transaction
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 * gfs2_log_release - Release a given number of log blocks
390 * @sdp: The GFS2 superblock
391 * @blks: The number of blocks
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 * gfs2_log_reserve - Make a log reservation
407 * @sdp: The GFS2 superblock
408 * @blks: The number of blocks to reserve
409 *
410 * Note that we never give out the last few blocks of the journal. Thats
411 * due to the fact that there is a small number of header blocks
412 * associated with each log flush. The exact number can't be known until
413 * flush time, so we ensure that we have just enough free blocks at all
414 * times to avoid running out during a log flush.
415 *
416 * We no longer flush the log here, instead we wake up logd to do that
417 * for us. To avoid the thundering herd and to ensure that we deal fairly
418 * with queued waiters, we use an exclusive wait. This means that when we
419 * get woken with enough journal space to get our reservation, we need to
420 * wake the next waiter on the list.
421 *
422 * Returns: errno
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 * If we waited, then so might others, wake them up _after_ we get
464 * our share of the log.
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 * log_distance - Compute distance between two journal blocks
481 * @sdp: The GFS2 superblock
482 * @newer: The most recent journal block of the pair
483 * @older: The older journal block of the pair
484 *
485 * Compute the distance (in the journal direction) between two
486 * blocks in the journal
487 *
488 * Returns: the distance in blocks
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 * calc_reserved - Calculate the number of blocks to reserve when
505 * refunding a transaction's unused buffers.
506 * @sdp: The GFS2 superblock
507 *
508 * This is complex. We need to reserve room for all our currently used
509 * metadata buffers (e.g. normal file I/O rewriting file time stamps) and
510 * all our journaled data buffers for journaled files (e.g. files in the
511 * meta_fs like rindex, or files for which chattr +j was done.)
512 * If we don't reserve enough space, gfs2_log_refund and gfs2_log_flush
513 * will count it as free space (sd_log_blks_free) and corruption will follow.
514 *
515 * We can have metadata bufs and jdata bufs in the same journal. So each
516 * type gets its own log header, for which we need to reserve a block.
517 * In fact, each type has the potential for needing more than one header
518 * in cases where we have more buffers than will fit on a journal page.
519 * Metadata journal entries take up half the space of journaled buffer entries.
520 * Thus, metadata entries have buf_limit (502) and journaled buffers have
521 * databuf_limit (251) before they cause a wrap around.
522 *
523 * Also, we need to reserve blocks for revoke journal entries and one for an
524 * overall header for the lot.
525 *
526 * Returns: the number of blocks reserved
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 /* Account for header blocks */
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 /* One for the overall header */
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); /* drops ref on bh */
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 * gfs2_write_revokes - Add as many revokes to the system transaction as we can
696 * @sdp: The GFS2 superblock
697 *
698 * Our usual strategy is to defer writing revokes as much as we can in the hope
699 * that we'll eventually overwrite the journal, which will make those revokes
700 * go away. This changes when we flush the log: at that point, there will
701 * likely be some left-over space in the last revoke block of that transaction.
702 * We can fill that space with additional revokes for blocks that have already
703 * been written back. This will basically come at no cost now, and will save
704 * us from having to keep track of those blocks on the AIL2 list later.
705 */
706void gfs2_write_revokes(struct gfs2_sbd *sdp)
707{
708 /* number of revokes we still have room for */
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 /* If no blocks have been reserved, we need to also
718 * reserve a block for the header */
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 * gfs2_write_log_header - Write a journal log header buffer at lblock
734 * @sdp: The GFS2 superblock
735 * @jd: journal descriptor of the journal to which we are writing
736 * @seq: sequence number
737 * @tail: tail of the log
738 * @lblock: value for lh_blkno (block number relative to start of journal)
739 * @flags: log header flags GFS2_LOG_HEAD_*
740 * @op_flags: flags to pass to the bio
741 *
742 * Returns: the initialized log buffer descriptor
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 /* We may only write local statfs, quota, etc., when writing to our
790 own journal. The values are left 0 when recovering a journal
791 different from our own. */
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 * log_write_header - Get and initialize a journal header buffer
819 * @sdp: The GFS2 superblock
820 * @flags: The log header flags, including log header origin
821 *
822 * Returns: the initialized log buffer descriptor
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 * ail_drain - drain the ail lists after a withdraw
850 * @sdp: Pointer to GFS2 superblock
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 * For transactions on the sd_ail1_list we need to drain both the
859 * ail1 and ail2 lists. That's because function gfs2_ail1_start_one
860 * (temporarily) moves items from its tr_ail1 list to tr_ail2 list
861 * before revokes are sent for that block. Items on the sd_ail2_list
862 * should have already gotten beyond that point, so no need.
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 * empty_ail1_list - try to start IO and empty the ail1 list
884 * @sdp: Pointer to GFS2 superblock
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 * drain_bd - drain the buf and databuf queue for a failed transaction
906 * @tr: the transaction to drain
907 *
908 * When this is called, we're taking an error exit for a log write that failed
909 * but since we bypassed the after_commit functions, we need to remove the
910 * items from the buf and databuf queue.
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 * gfs2_log_flush - flush incore transaction(s)
936 * @sdp: the filesystem
937 * @gl: The glock structure to flush. If NULL, flush the whole incore log
938 * @flags: The log header flags: GFS2_LOG_HEAD_FLUSH_* and debug flags
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 * Do this check while holding the log_flush_lock to prevent new
951 * buffers from being added to the ail via gfs2_pin()
952 */
953 if (gfs2_withdrawn(sdp))
954 goto out;
955
956 /* Log might have been flushed while we waited for the flush lock */
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); /* Adjust for unreserved buffer */
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); /* Adjust for unreserved buffer */
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 * If the tr_list is empty, we're withdrawing during a log
1041 * flush that targets a transaction, but the transaction was
1042 * never queued onto any of the ail lists. Here we add it to
1043 * ail1 just so that ail_drain() will find and free it.
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); /* frees all transactions */
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 * gfs2_merge_trans - Merge a new transaction into a cached transaction
1061 * @old: Original transaction to be expanded
1062 * @new: New transaction to be merged
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 * gfs2_log_commit - Commit a transaction to the log
1119 * @sdp: the filesystem
1120 * @tr: the transaction
1121 *
1122 * We wake up gfs2_logd if the number of pinned blocks exceed thresh1
1123 * or the total number of used blocks (pinned blocks plus AIL blocks)
1124 * is greater than thresh2.
1125 *
1126 * At mount time thresh1 is 2/5ths of journal size, thresh2 is 4/5ths of
1127 * journal size.
1128 *
1129 * Returns: errno
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 * gfs2_log_shutdown - write a shutdown header into a journal
1144 * @sdp: the filesystem
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 * gfs2_logd - Update log tail as Active Items get flushed to in-place blocks
1185 * @sdp: Pointer to GFS2 superblock
1186 *
1187 * Also, periodically check to make sure that we're using the most recent
1188 * journal index.
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 /* Check for errors writing to the journal */
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
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
4 * Copyright (C) 2004-2007 Red Hat, Inc. All rights reserved.
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 * gfs2_struct2blk - compute stuff
39 * @sdp: the filesystem
40 * @nstruct: the number of structures
41 *
42 * Compute the number of log descriptor blocks needed to hold a certain number
43 * of structures of a certain size.
44 *
45 * Returns: the number of blocks needed (minimum is always 1)
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 /* The initial struct gfs2_log_descriptor block */
54 blks = 1;
55 first = sdp->sd_ldptrs;
56
57 if (nstruct > first) {
58 /* Subsequent struct gfs2_meta_header blocks */
59 second = sdp->sd_inptrs;
60 blks += DIV_ROUND_UP(nstruct - first, second);
61 }
62
63 return blks;
64}
65
66/**
67 * gfs2_remove_from_ail - Remove an entry from the ail lists, updating counters
68 * @bd: The gfs2_bufdata to remove
69 *
70 * The ail lock _must_ be held when calling this function
71 *
72 */
73
74void gfs2_remove_from_ail(struct gfs2_bufdata *bd)
75{
76 bd->bd_tr = NULL;
77 list_del_init(&bd->bd_ail_st_list);
78 list_del_init(&bd->bd_ail_gl_list);
79 atomic_dec(&bd->bd_gl->gl_ail_count);
80 brelse(bd->bd_bh);
81}
82
83/**
84 * gfs2_ail1_start_one - Start I/O on a transaction
85 * @sdp: The superblock
86 * @wbc: The writeback control structure
87 * @tr: The transaction to start I/O on
88 * @plug: The block plug currently active
89 */
90
91static int gfs2_ail1_start_one(struct gfs2_sbd *sdp,
92 struct writeback_control *wbc,
93 struct gfs2_trans *tr, struct blk_plug *plug)
94__releases(&sdp->sd_ail_lock)
95__acquires(&sdp->sd_ail_lock)
96{
97 struct gfs2_glock *gl = NULL;
98 struct address_space *mapping;
99 struct gfs2_bufdata *bd, *s;
100 struct buffer_head *bh;
101 int ret = 0;
102
103 list_for_each_entry_safe_reverse(bd, s, &tr->tr_ail1_list, bd_ail_st_list) {
104 bh = bd->bd_bh;
105
106 gfs2_assert(sdp, bd->bd_tr == tr);
107
108 if (!buffer_busy(bh)) {
109 if (buffer_uptodate(bh)) {
110 list_move(&bd->bd_ail_st_list,
111 &tr->tr_ail2_list);
112 continue;
113 }
114 if (!cmpxchg(&sdp->sd_log_error, 0, -EIO)) {
115 gfs2_io_error_bh(sdp, bh);
116 gfs2_withdraw_delayed(sdp);
117 }
118 }
119
120 if (gfs2_withdrawn(sdp)) {
121 gfs2_remove_from_ail(bd);
122 continue;
123 }
124 if (!buffer_dirty(bh))
125 continue;
126 if (gl == bd->bd_gl)
127 continue;
128 gl = bd->bd_gl;
129 list_move(&bd->bd_ail_st_list, &tr->tr_ail1_list);
130 mapping = bh->b_page->mapping;
131 if (!mapping)
132 continue;
133 spin_unlock(&sdp->sd_ail_lock);
134 ret = generic_writepages(mapping, wbc);
135 if (need_resched()) {
136 blk_finish_plug(plug);
137 cond_resched();
138 blk_start_plug(plug);
139 }
140 spin_lock(&sdp->sd_ail_lock);
141 if (ret == -ENODATA) /* if a jdata write into a new hole */
142 ret = 0; /* ignore it */
143 if (ret || wbc->nr_to_write <= 0)
144 break;
145 return -EBUSY;
146 }
147
148 return ret;
149}
150
151static void dump_ail_list(struct gfs2_sbd *sdp)
152{
153 struct gfs2_trans *tr;
154 struct gfs2_bufdata *bd;
155 struct buffer_head *bh;
156
157 list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) {
158 list_for_each_entry_reverse(bd, &tr->tr_ail1_list,
159 bd_ail_st_list) {
160 bh = bd->bd_bh;
161 fs_err(sdp, "bd %p: blk:0x%llx bh=%p ", bd,
162 (unsigned long long)bd->bd_blkno, bh);
163 if (!bh) {
164 fs_err(sdp, "\n");
165 continue;
166 }
167 fs_err(sdp, "0x%llx up2:%d dirt:%d lkd:%d req:%d "
168 "map:%d new:%d ar:%d aw:%d delay:%d "
169 "io err:%d unwritten:%d dfr:%d pin:%d esc:%d\n",
170 (unsigned long long)bh->b_blocknr,
171 buffer_uptodate(bh), buffer_dirty(bh),
172 buffer_locked(bh), buffer_req(bh),
173 buffer_mapped(bh), buffer_new(bh),
174 buffer_async_read(bh), buffer_async_write(bh),
175 buffer_delay(bh), buffer_write_io_error(bh),
176 buffer_unwritten(bh),
177 buffer_defer_completion(bh),
178 buffer_pinned(bh), buffer_escaped(bh));
179 }
180 }
181}
182
183/**
184 * gfs2_ail1_flush - start writeback of some ail1 entries
185 * @sdp: The super block
186 * @wbc: The writeback control structure
187 *
188 * Writes back some ail1 entries, according to the limits in the
189 * writeback control structure
190 */
191
192void gfs2_ail1_flush(struct gfs2_sbd *sdp, struct writeback_control *wbc)
193{
194 struct list_head *head = &sdp->sd_ail1_list;
195 struct gfs2_trans *tr;
196 struct blk_plug plug;
197 int ret;
198 unsigned long flush_start = jiffies;
199
200 trace_gfs2_ail_flush(sdp, wbc, 1);
201 blk_start_plug(&plug);
202 spin_lock(&sdp->sd_ail_lock);
203restart:
204 ret = 0;
205 if (time_after(jiffies, flush_start + (HZ * 600))) {
206 fs_err(sdp, "Error: In %s for ten minutes! t=%d\n",
207 __func__, current->journal_info ? 1 : 0);
208 dump_ail_list(sdp);
209 goto out;
210 }
211 list_for_each_entry_reverse(tr, head, tr_list) {
212 if (wbc->nr_to_write <= 0)
213 break;
214 ret = gfs2_ail1_start_one(sdp, wbc, tr, &plug);
215 if (ret) {
216 if (ret == -EBUSY)
217 goto restart;
218 break;
219 }
220 }
221out:
222 spin_unlock(&sdp->sd_ail_lock);
223 blk_finish_plug(&plug);
224 if (ret) {
225 gfs2_lm(sdp, "gfs2_ail1_start_one (generic_writepages) "
226 "returned: %d\n", ret);
227 gfs2_withdraw(sdp);
228 }
229 trace_gfs2_ail_flush(sdp, wbc, 0);
230}
231
232/**
233 * gfs2_ail1_start - start writeback of all ail1 entries
234 * @sdp: The superblock
235 */
236
237static void gfs2_ail1_start(struct gfs2_sbd *sdp)
238{
239 struct writeback_control wbc = {
240 .sync_mode = WB_SYNC_NONE,
241 .nr_to_write = LONG_MAX,
242 .range_start = 0,
243 .range_end = LLONG_MAX,
244 };
245
246 return gfs2_ail1_flush(sdp, &wbc);
247}
248
249static void gfs2_log_update_flush_tail(struct gfs2_sbd *sdp)
250{
251 unsigned int new_flush_tail = sdp->sd_log_head;
252 struct gfs2_trans *tr;
253
254 if (!list_empty(&sdp->sd_ail1_list)) {
255 tr = list_last_entry(&sdp->sd_ail1_list,
256 struct gfs2_trans, tr_list);
257 new_flush_tail = tr->tr_first;
258 }
259 sdp->sd_log_flush_tail = new_flush_tail;
260}
261
262static void gfs2_log_update_head(struct gfs2_sbd *sdp)
263{
264 unsigned int new_head = sdp->sd_log_flush_head;
265
266 if (sdp->sd_log_flush_tail == sdp->sd_log_head)
267 sdp->sd_log_flush_tail = new_head;
268 sdp->sd_log_head = new_head;
269}
270
271/*
272 * gfs2_ail_empty_tr - empty one of the ail lists of a transaction
273 */
274
275static void gfs2_ail_empty_tr(struct gfs2_sbd *sdp, struct gfs2_trans *tr,
276 struct list_head *head)
277{
278 struct gfs2_bufdata *bd;
279
280 while (!list_empty(head)) {
281 bd = list_first_entry(head, struct gfs2_bufdata,
282 bd_ail_st_list);
283 gfs2_assert(sdp, bd->bd_tr == tr);
284 gfs2_remove_from_ail(bd);
285 }
286}
287
288/**
289 * gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced
290 * @sdp: the filesystem
291 * @tr: the transaction
292 * @max_revokes: If nonzero, issue revokes for the bd items for written buffers
293 *
294 * returns: the transaction's count of remaining active items
295 */
296
297static int gfs2_ail1_empty_one(struct gfs2_sbd *sdp, struct gfs2_trans *tr,
298 int *max_revokes)
299{
300 struct gfs2_bufdata *bd, *s;
301 struct buffer_head *bh;
302 int active_count = 0;
303
304 list_for_each_entry_safe_reverse(bd, s, &tr->tr_ail1_list,
305 bd_ail_st_list) {
306 bh = bd->bd_bh;
307 gfs2_assert(sdp, bd->bd_tr == tr);
308 /*
309 * If another process flagged an io error, e.g. writing to the
310 * journal, error all other bhs and move them off the ail1 to
311 * prevent a tight loop when unmount tries to flush ail1,
312 * regardless of whether they're still busy. If no outside
313 * errors were found and the buffer is busy, move to the next.
314 * If the ail buffer is not busy and caught an error, flag it
315 * for others.
316 */
317 if (!sdp->sd_log_error && buffer_busy(bh)) {
318 active_count++;
319 continue;
320 }
321 if (!buffer_uptodate(bh) &&
322 !cmpxchg(&sdp->sd_log_error, 0, -EIO)) {
323 gfs2_io_error_bh(sdp, bh);
324 gfs2_withdraw_delayed(sdp);
325 }
326 /*
327 * If we have space for revokes and the bd is no longer on any
328 * buf list, we can just add a revoke for it immediately and
329 * avoid having to put it on the ail2 list, where it would need
330 * to be revoked later.
331 */
332 if (*max_revokes && list_empty(&bd->bd_list)) {
333 gfs2_add_revoke(sdp, bd);
334 (*max_revokes)--;
335 continue;
336 }
337 list_move(&bd->bd_ail_st_list, &tr->tr_ail2_list);
338 }
339 return active_count;
340}
341
342/**
343 * gfs2_ail1_empty - Try to empty the ail1 lists
344 * @sdp: The superblock
345 * @max_revokes: If non-zero, add revokes where appropriate
346 *
347 * Tries to empty the ail1 lists, starting with the oldest first
348 */
349
350static int gfs2_ail1_empty(struct gfs2_sbd *sdp, int max_revokes)
351{
352 struct gfs2_trans *tr, *s;
353 int oldest_tr = 1;
354 int ret;
355
356 spin_lock(&sdp->sd_ail_lock);
357 list_for_each_entry_safe_reverse(tr, s, &sdp->sd_ail1_list, tr_list) {
358 if (!gfs2_ail1_empty_one(sdp, tr, &max_revokes) && oldest_tr)
359 list_move(&tr->tr_list, &sdp->sd_ail2_list);
360 else
361 oldest_tr = 0;
362 }
363 gfs2_log_update_flush_tail(sdp);
364 ret = list_empty(&sdp->sd_ail1_list);
365 spin_unlock(&sdp->sd_ail_lock);
366
367 if (test_bit(SDF_WITHDRAWING, &sdp->sd_flags)) {
368 gfs2_lm(sdp, "fatal: I/O error(s)\n");
369 gfs2_withdraw(sdp);
370 }
371
372 return ret;
373}
374
375static void gfs2_ail1_wait(struct gfs2_sbd *sdp)
376{
377 struct gfs2_trans *tr;
378 struct gfs2_bufdata *bd;
379 struct buffer_head *bh;
380
381 spin_lock(&sdp->sd_ail_lock);
382 list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) {
383 list_for_each_entry(bd, &tr->tr_ail1_list, bd_ail_st_list) {
384 bh = bd->bd_bh;
385 if (!buffer_locked(bh))
386 continue;
387 get_bh(bh);
388 spin_unlock(&sdp->sd_ail_lock);
389 wait_on_buffer(bh);
390 brelse(bh);
391 return;
392 }
393 }
394 spin_unlock(&sdp->sd_ail_lock);
395}
396
397static void __ail2_empty(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
398{
399 gfs2_ail_empty_tr(sdp, tr, &tr->tr_ail2_list);
400 list_del(&tr->tr_list);
401 gfs2_assert_warn(sdp, list_empty(&tr->tr_ail1_list));
402 gfs2_assert_warn(sdp, list_empty(&tr->tr_ail2_list));
403 gfs2_trans_free(sdp, tr);
404}
405
406static void ail2_empty(struct gfs2_sbd *sdp, unsigned int new_tail)
407{
408 struct list_head *ail2_list = &sdp->sd_ail2_list;
409 unsigned int old_tail = sdp->sd_log_tail;
410 struct gfs2_trans *tr, *safe;
411
412 spin_lock(&sdp->sd_ail_lock);
413 if (old_tail <= new_tail) {
414 list_for_each_entry_safe(tr, safe, ail2_list, tr_list) {
415 if (old_tail <= tr->tr_first && tr->tr_first < new_tail)
416 __ail2_empty(sdp, tr);
417 }
418 } else {
419 list_for_each_entry_safe(tr, safe, ail2_list, tr_list) {
420 if (old_tail <= tr->tr_first || tr->tr_first < new_tail)
421 __ail2_empty(sdp, tr);
422 }
423 }
424 spin_unlock(&sdp->sd_ail_lock);
425}
426
427/**
428 * gfs2_log_is_empty - Check if the log is empty
429 * @sdp: The GFS2 superblock
430 */
431
432bool gfs2_log_is_empty(struct gfs2_sbd *sdp) {
433 return atomic_read(&sdp->sd_log_blks_free) == sdp->sd_jdesc->jd_blocks;
434}
435
436static bool __gfs2_log_try_reserve_revokes(struct gfs2_sbd *sdp, unsigned int revokes)
437{
438 unsigned int available;
439
440 available = atomic_read(&sdp->sd_log_revokes_available);
441 while (available >= revokes) {
442 if (atomic_try_cmpxchg(&sdp->sd_log_revokes_available,
443 &available, available - revokes))
444 return true;
445 }
446 return false;
447}
448
449/**
450 * gfs2_log_release_revokes - Release a given number of revokes
451 * @sdp: The GFS2 superblock
452 * @revokes: The number of revokes to release
453 *
454 * sdp->sd_log_flush_lock must be held.
455 */
456void gfs2_log_release_revokes(struct gfs2_sbd *sdp, unsigned int revokes)
457{
458 if (revokes)
459 atomic_add(revokes, &sdp->sd_log_revokes_available);
460}
461
462/**
463 * gfs2_log_release - Release a given number of log blocks
464 * @sdp: The GFS2 superblock
465 * @blks: The number of blocks
466 *
467 */
468
469void gfs2_log_release(struct gfs2_sbd *sdp, unsigned int blks)
470{
471 atomic_add(blks, &sdp->sd_log_blks_free);
472 trace_gfs2_log_blocks(sdp, blks);
473 gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
474 sdp->sd_jdesc->jd_blocks);
475 if (atomic_read(&sdp->sd_log_blks_needed))
476 wake_up(&sdp->sd_log_waitq);
477}
478
479/**
480 * __gfs2_log_try_reserve - Try to make a log reservation
481 * @sdp: The GFS2 superblock
482 * @blks: The number of blocks to reserve
483 * @taboo_blks: The number of blocks to leave free
484 *
485 * Try to do the same as __gfs2_log_reserve(), but fail if no more log
486 * space is immediately available.
487 */
488static bool __gfs2_log_try_reserve(struct gfs2_sbd *sdp, unsigned int blks,
489 unsigned int taboo_blks)
490{
491 unsigned wanted = blks + taboo_blks;
492 unsigned int free_blocks;
493
494 free_blocks = atomic_read(&sdp->sd_log_blks_free);
495 while (free_blocks >= wanted) {
496 if (atomic_try_cmpxchg(&sdp->sd_log_blks_free, &free_blocks,
497 free_blocks - blks)) {
498 trace_gfs2_log_blocks(sdp, -blks);
499 return true;
500 }
501 }
502 return false;
503}
504
505/**
506 * __gfs2_log_reserve - Make a log reservation
507 * @sdp: The GFS2 superblock
508 * @blks: The number of blocks to reserve
509 * @taboo_blks: The number of blocks to leave free
510 *
511 * @taboo_blks is set to 0 for logd, and to GFS2_LOG_FLUSH_MIN_BLOCKS
512 * for all other processes. This ensures that when the log is almost full,
513 * logd will still be able to call gfs2_log_flush one more time without
514 * blocking, which will advance the tail and make some more log space
515 * available.
516 *
517 * We no longer flush the log here, instead we wake up logd to do that
518 * for us. To avoid the thundering herd and to ensure that we deal fairly
519 * with queued waiters, we use an exclusive wait. This means that when we
520 * get woken with enough journal space to get our reservation, we need to
521 * wake the next waiter on the list.
522 */
523
524static void __gfs2_log_reserve(struct gfs2_sbd *sdp, unsigned int blks,
525 unsigned int taboo_blks)
526{
527 unsigned wanted = blks + taboo_blks;
528 unsigned int free_blocks;
529
530 atomic_add(blks, &sdp->sd_log_blks_needed);
531 for (;;) {
532 if (current != sdp->sd_logd_process)
533 wake_up(&sdp->sd_logd_waitq);
534 io_wait_event(sdp->sd_log_waitq,
535 (free_blocks = atomic_read(&sdp->sd_log_blks_free),
536 free_blocks >= wanted));
537 do {
538 if (atomic_try_cmpxchg(&sdp->sd_log_blks_free,
539 &free_blocks,
540 free_blocks - blks))
541 goto reserved;
542 } while (free_blocks >= wanted);
543 }
544
545reserved:
546 trace_gfs2_log_blocks(sdp, -blks);
547 if (atomic_sub_return(blks, &sdp->sd_log_blks_needed))
548 wake_up(&sdp->sd_log_waitq);
549}
550
551/**
552 * gfs2_log_try_reserve - Try to make a log reservation
553 * @sdp: The GFS2 superblock
554 * @tr: The transaction
555 * @extra_revokes: The number of additional revokes reserved (output)
556 *
557 * This is similar to gfs2_log_reserve, but sdp->sd_log_flush_lock must be
558 * held for correct revoke accounting.
559 */
560
561bool gfs2_log_try_reserve(struct gfs2_sbd *sdp, struct gfs2_trans *tr,
562 unsigned int *extra_revokes)
563{
564 unsigned int blks = tr->tr_reserved;
565 unsigned int revokes = tr->tr_revokes;
566 unsigned int revoke_blks = 0;
567
568 *extra_revokes = 0;
569 if (revokes && !__gfs2_log_try_reserve_revokes(sdp, revokes)) {
570 revoke_blks = DIV_ROUND_UP(revokes, sdp->sd_inptrs);
571 *extra_revokes = revoke_blks * sdp->sd_inptrs - revokes;
572 blks += revoke_blks;
573 }
574 if (!blks)
575 return true;
576 if (__gfs2_log_try_reserve(sdp, blks, GFS2_LOG_FLUSH_MIN_BLOCKS))
577 return true;
578 if (!revoke_blks)
579 gfs2_log_release_revokes(sdp, revokes);
580 return false;
581}
582
583/**
584 * gfs2_log_reserve - Make a log reservation
585 * @sdp: The GFS2 superblock
586 * @tr: The transaction
587 * @extra_revokes: The number of additional revokes reserved (output)
588 *
589 * sdp->sd_log_flush_lock must not be held.
590 */
591
592void gfs2_log_reserve(struct gfs2_sbd *sdp, struct gfs2_trans *tr,
593 unsigned int *extra_revokes)
594{
595 unsigned int blks = tr->tr_reserved;
596 unsigned int revokes = tr->tr_revokes;
597 unsigned int revoke_blks = 0;
598
599 *extra_revokes = 0;
600 if (revokes) {
601 revoke_blks = DIV_ROUND_UP(revokes, sdp->sd_inptrs);
602 *extra_revokes = revoke_blks * sdp->sd_inptrs - revokes;
603 blks += revoke_blks;
604 }
605 __gfs2_log_reserve(sdp, blks, GFS2_LOG_FLUSH_MIN_BLOCKS);
606}
607
608/**
609 * log_distance - Compute distance between two journal blocks
610 * @sdp: The GFS2 superblock
611 * @newer: The most recent journal block of the pair
612 * @older: The older journal block of the pair
613 *
614 * Compute the distance (in the journal direction) between two
615 * blocks in the journal
616 *
617 * Returns: the distance in blocks
618 */
619
620static inline unsigned int log_distance(struct gfs2_sbd *sdp, unsigned int newer,
621 unsigned int older)
622{
623 int dist;
624
625 dist = newer - older;
626 if (dist < 0)
627 dist += sdp->sd_jdesc->jd_blocks;
628
629 return dist;
630}
631
632/**
633 * calc_reserved - Calculate the number of blocks to keep reserved
634 * @sdp: The GFS2 superblock
635 *
636 * This is complex. We need to reserve room for all our currently used
637 * metadata blocks (e.g. normal file I/O rewriting file time stamps) and
638 * all our journaled data blocks for journaled files (e.g. files in the
639 * meta_fs like rindex, or files for which chattr +j was done.)
640 * If we don't reserve enough space, corruption will follow.
641 *
642 * We can have metadata blocks and jdata blocks in the same journal. Each
643 * type gets its own log descriptor, for which we need to reserve a block.
644 * In fact, each type has the potential for needing more than one log descriptor
645 * in cases where we have more blocks than will fit in a log descriptor.
646 * Metadata journal entries take up half the space of journaled buffer entries.
647 *
648 * Also, we need to reserve blocks for revoke journal entries and one for an
649 * overall header for the lot.
650 *
651 * Returns: the number of blocks reserved
652 */
653static unsigned int calc_reserved(struct gfs2_sbd *sdp)
654{
655 unsigned int reserved = GFS2_LOG_FLUSH_MIN_BLOCKS;
656 unsigned int blocks;
657 struct gfs2_trans *tr = sdp->sd_log_tr;
658
659 if (tr) {
660 blocks = tr->tr_num_buf_new - tr->tr_num_buf_rm;
661 reserved += blocks + DIV_ROUND_UP(blocks, buf_limit(sdp));
662 blocks = tr->tr_num_databuf_new - tr->tr_num_databuf_rm;
663 reserved += blocks + DIV_ROUND_UP(blocks, databuf_limit(sdp));
664 }
665 return reserved;
666}
667
668static void log_pull_tail(struct gfs2_sbd *sdp)
669{
670 unsigned int new_tail = sdp->sd_log_flush_tail;
671 unsigned int dist;
672
673 if (new_tail == sdp->sd_log_tail)
674 return;
675 dist = log_distance(sdp, new_tail, sdp->sd_log_tail);
676 ail2_empty(sdp, new_tail);
677 gfs2_log_release(sdp, dist);
678 sdp->sd_log_tail = new_tail;
679}
680
681
682void log_flush_wait(struct gfs2_sbd *sdp)
683{
684 DEFINE_WAIT(wait);
685
686 if (atomic_read(&sdp->sd_log_in_flight)) {
687 do {
688 prepare_to_wait(&sdp->sd_log_flush_wait, &wait,
689 TASK_UNINTERRUPTIBLE);
690 if (atomic_read(&sdp->sd_log_in_flight))
691 io_schedule();
692 } while(atomic_read(&sdp->sd_log_in_flight));
693 finish_wait(&sdp->sd_log_flush_wait, &wait);
694 }
695}
696
697static int ip_cmp(void *priv, const struct list_head *a, const struct list_head *b)
698{
699 struct gfs2_inode *ipa, *ipb;
700
701 ipa = list_entry(a, struct gfs2_inode, i_ordered);
702 ipb = list_entry(b, struct gfs2_inode, i_ordered);
703
704 if (ipa->i_no_addr < ipb->i_no_addr)
705 return -1;
706 if (ipa->i_no_addr > ipb->i_no_addr)
707 return 1;
708 return 0;
709}
710
711static void __ordered_del_inode(struct gfs2_inode *ip)
712{
713 if (!list_empty(&ip->i_ordered))
714 list_del_init(&ip->i_ordered);
715}
716
717static void gfs2_ordered_write(struct gfs2_sbd *sdp)
718{
719 struct gfs2_inode *ip;
720 LIST_HEAD(written);
721
722 spin_lock(&sdp->sd_ordered_lock);
723 list_sort(NULL, &sdp->sd_log_ordered, &ip_cmp);
724 while (!list_empty(&sdp->sd_log_ordered)) {
725 ip = list_first_entry(&sdp->sd_log_ordered, struct gfs2_inode, i_ordered);
726 if (ip->i_inode.i_mapping->nrpages == 0) {
727 __ordered_del_inode(ip);
728 continue;
729 }
730 list_move(&ip->i_ordered, &written);
731 spin_unlock(&sdp->sd_ordered_lock);
732 filemap_fdatawrite(ip->i_inode.i_mapping);
733 spin_lock(&sdp->sd_ordered_lock);
734 }
735 list_splice(&written, &sdp->sd_log_ordered);
736 spin_unlock(&sdp->sd_ordered_lock);
737}
738
739static void gfs2_ordered_wait(struct gfs2_sbd *sdp)
740{
741 struct gfs2_inode *ip;
742
743 spin_lock(&sdp->sd_ordered_lock);
744 while (!list_empty(&sdp->sd_log_ordered)) {
745 ip = list_first_entry(&sdp->sd_log_ordered, struct gfs2_inode, i_ordered);
746 __ordered_del_inode(ip);
747 if (ip->i_inode.i_mapping->nrpages == 0)
748 continue;
749 spin_unlock(&sdp->sd_ordered_lock);
750 filemap_fdatawait(ip->i_inode.i_mapping);
751 spin_lock(&sdp->sd_ordered_lock);
752 }
753 spin_unlock(&sdp->sd_ordered_lock);
754}
755
756void gfs2_ordered_del_inode(struct gfs2_inode *ip)
757{
758 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
759
760 spin_lock(&sdp->sd_ordered_lock);
761 __ordered_del_inode(ip);
762 spin_unlock(&sdp->sd_ordered_lock);
763}
764
765void gfs2_add_revoke(struct gfs2_sbd *sdp, struct gfs2_bufdata *bd)
766{
767 struct buffer_head *bh = bd->bd_bh;
768 struct gfs2_glock *gl = bd->bd_gl;
769
770 sdp->sd_log_num_revoke++;
771 if (atomic_inc_return(&gl->gl_revokes) == 1)
772 gfs2_glock_hold(gl);
773 bh->b_private = NULL;
774 bd->bd_blkno = bh->b_blocknr;
775 gfs2_remove_from_ail(bd); /* drops ref on bh */
776 bd->bd_bh = NULL;
777 set_bit(GLF_LFLUSH, &gl->gl_flags);
778 list_add(&bd->bd_list, &sdp->sd_log_revokes);
779}
780
781void gfs2_glock_remove_revoke(struct gfs2_glock *gl)
782{
783 if (atomic_dec_return(&gl->gl_revokes) == 0) {
784 clear_bit(GLF_LFLUSH, &gl->gl_flags);
785 gfs2_glock_queue_put(gl);
786 }
787}
788
789/**
790 * gfs2_flush_revokes - Add as many revokes to the system transaction as we can
791 * @sdp: The GFS2 superblock
792 *
793 * Our usual strategy is to defer writing revokes as much as we can in the hope
794 * that we'll eventually overwrite the journal, which will make those revokes
795 * go away. This changes when we flush the log: at that point, there will
796 * likely be some left-over space in the last revoke block of that transaction.
797 * We can fill that space with additional revokes for blocks that have already
798 * been written back. This will basically come at no cost now, and will save
799 * us from having to keep track of those blocks on the AIL2 list later.
800 */
801void gfs2_flush_revokes(struct gfs2_sbd *sdp)
802{
803 /* number of revokes we still have room for */
804 unsigned int max_revokes = atomic_read(&sdp->sd_log_revokes_available);
805
806 gfs2_log_lock(sdp);
807 gfs2_ail1_empty(sdp, max_revokes);
808 gfs2_log_unlock(sdp);
809}
810
811/**
812 * gfs2_write_log_header - Write a journal log header buffer at lblock
813 * @sdp: The GFS2 superblock
814 * @jd: journal descriptor of the journal to which we are writing
815 * @seq: sequence number
816 * @tail: tail of the log
817 * @lblock: value for lh_blkno (block number relative to start of journal)
818 * @flags: log header flags GFS2_LOG_HEAD_*
819 * @op_flags: flags to pass to the bio
820 *
821 * Returns: the initialized log buffer descriptor
822 */
823
824void gfs2_write_log_header(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd,
825 u64 seq, u32 tail, u32 lblock, u32 flags,
826 int op_flags)
827{
828 struct gfs2_log_header *lh;
829 u32 hash, crc;
830 struct page *page;
831 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
832 struct timespec64 tv;
833 struct super_block *sb = sdp->sd_vfs;
834 u64 dblock;
835
836 if (gfs2_withdrawn(sdp))
837 return;
838
839 page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
840 lh = page_address(page);
841 clear_page(lh);
842
843 lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
844 lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH);
845 lh->lh_header.__pad0 = cpu_to_be64(0);
846 lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH);
847 lh->lh_header.mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid);
848 lh->lh_sequence = cpu_to_be64(seq);
849 lh->lh_flags = cpu_to_be32(flags);
850 lh->lh_tail = cpu_to_be32(tail);
851 lh->lh_blkno = cpu_to_be32(lblock);
852 hash = ~crc32(~0, lh, LH_V1_SIZE);
853 lh->lh_hash = cpu_to_be32(hash);
854
855 ktime_get_coarse_real_ts64(&tv);
856 lh->lh_nsec = cpu_to_be32(tv.tv_nsec);
857 lh->lh_sec = cpu_to_be64(tv.tv_sec);
858 if (!list_empty(&jd->extent_list))
859 dblock = gfs2_log_bmap(jd, lblock);
860 else {
861 unsigned int extlen;
862 int ret;
863
864 extlen = 1;
865 ret = gfs2_get_extent(jd->jd_inode, lblock, &dblock, &extlen);
866 if (gfs2_assert_withdraw(sdp, ret == 0))
867 return;
868 }
869 lh->lh_addr = cpu_to_be64(dblock);
870 lh->lh_jinode = cpu_to_be64(GFS2_I(jd->jd_inode)->i_no_addr);
871
872 /* We may only write local statfs, quota, etc., when writing to our
873 own journal. The values are left 0 when recovering a journal
874 different from our own. */
875 if (!(flags & GFS2_LOG_HEAD_RECOVERY)) {
876 lh->lh_statfs_addr =
877 cpu_to_be64(GFS2_I(sdp->sd_sc_inode)->i_no_addr);
878 lh->lh_quota_addr =
879 cpu_to_be64(GFS2_I(sdp->sd_qc_inode)->i_no_addr);
880
881 spin_lock(&sdp->sd_statfs_spin);
882 lh->lh_local_total = cpu_to_be64(l_sc->sc_total);
883 lh->lh_local_free = cpu_to_be64(l_sc->sc_free);
884 lh->lh_local_dinodes = cpu_to_be64(l_sc->sc_dinodes);
885 spin_unlock(&sdp->sd_statfs_spin);
886 }
887
888 BUILD_BUG_ON(offsetof(struct gfs2_log_header, lh_crc) != LH_V1_SIZE);
889
890 crc = crc32c(~0, (void *)lh + LH_V1_SIZE + 4,
891 sb->s_blocksize - LH_V1_SIZE - 4);
892 lh->lh_crc = cpu_to_be32(crc);
893
894 gfs2_log_write(sdp, jd, page, sb->s_blocksize, 0, dblock);
895 gfs2_log_submit_bio(&jd->jd_log_bio, REQ_OP_WRITE | op_flags);
896}
897
898/**
899 * log_write_header - Get and initialize a journal header buffer
900 * @sdp: The GFS2 superblock
901 * @flags: The log header flags, including log header origin
902 *
903 * Returns: the initialized log buffer descriptor
904 */
905
906static void log_write_header(struct gfs2_sbd *sdp, u32 flags)
907{
908 int op_flags = REQ_PREFLUSH | REQ_FUA | REQ_META | REQ_SYNC;
909 enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state);
910
911 gfs2_assert_withdraw(sdp, (state != SFS_FROZEN));
912
913 if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags)) {
914 gfs2_ordered_wait(sdp);
915 log_flush_wait(sdp);
916 op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
917 }
918 sdp->sd_log_idle = (sdp->sd_log_flush_tail == sdp->sd_log_flush_head);
919 gfs2_write_log_header(sdp, sdp->sd_jdesc, sdp->sd_log_sequence++,
920 sdp->sd_log_flush_tail, sdp->sd_log_flush_head,
921 flags, op_flags);
922 gfs2_log_incr_head(sdp);
923 log_flush_wait(sdp);
924 log_pull_tail(sdp);
925 gfs2_log_update_head(sdp);
926}
927
928/**
929 * gfs2_ail_drain - drain the ail lists after a withdraw
930 * @sdp: Pointer to GFS2 superblock
931 */
932void gfs2_ail_drain(struct gfs2_sbd *sdp)
933{
934 struct gfs2_trans *tr;
935
936 spin_lock(&sdp->sd_ail_lock);
937 /*
938 * For transactions on the sd_ail1_list we need to drain both the
939 * ail1 and ail2 lists. That's because function gfs2_ail1_start_one
940 * (temporarily) moves items from its tr_ail1 list to tr_ail2 list
941 * before revokes are sent for that block. Items on the sd_ail2_list
942 * should have already gotten beyond that point, so no need.
943 */
944 while (!list_empty(&sdp->sd_ail1_list)) {
945 tr = list_first_entry(&sdp->sd_ail1_list, struct gfs2_trans,
946 tr_list);
947 gfs2_ail_empty_tr(sdp, tr, &tr->tr_ail1_list);
948 gfs2_ail_empty_tr(sdp, tr, &tr->tr_ail2_list);
949 list_del(&tr->tr_list);
950 gfs2_trans_free(sdp, tr);
951 }
952 while (!list_empty(&sdp->sd_ail2_list)) {
953 tr = list_first_entry(&sdp->sd_ail2_list, struct gfs2_trans,
954 tr_list);
955 gfs2_ail_empty_tr(sdp, tr, &tr->tr_ail2_list);
956 list_del(&tr->tr_list);
957 gfs2_trans_free(sdp, tr);
958 }
959 gfs2_drain_revokes(sdp);
960 spin_unlock(&sdp->sd_ail_lock);
961}
962
963/**
964 * empty_ail1_list - try to start IO and empty the ail1 list
965 * @sdp: Pointer to GFS2 superblock
966 */
967static void empty_ail1_list(struct gfs2_sbd *sdp)
968{
969 unsigned long start = jiffies;
970
971 for (;;) {
972 if (time_after(jiffies, start + (HZ * 600))) {
973 fs_err(sdp, "Error: In %s for 10 minutes! t=%d\n",
974 __func__, current->journal_info ? 1 : 0);
975 dump_ail_list(sdp);
976 return;
977 }
978 gfs2_ail1_start(sdp);
979 gfs2_ail1_wait(sdp);
980 if (gfs2_ail1_empty(sdp, 0))
981 return;
982 }
983}
984
985/**
986 * trans_drain - drain the buf and databuf queue for a failed transaction
987 * @tr: the transaction to drain
988 *
989 * When this is called, we're taking an error exit for a log write that failed
990 * but since we bypassed the after_commit functions, we need to remove the
991 * items from the buf and databuf queue.
992 */
993static void trans_drain(struct gfs2_trans *tr)
994{
995 struct gfs2_bufdata *bd;
996 struct list_head *head;
997
998 if (!tr)
999 return;
1000
1001 head = &tr->tr_buf;
1002 while (!list_empty(head)) {
1003 bd = list_first_entry(head, struct gfs2_bufdata, bd_list);
1004 list_del_init(&bd->bd_list);
1005 if (!list_empty(&bd->bd_ail_st_list))
1006 gfs2_remove_from_ail(bd);
1007 kmem_cache_free(gfs2_bufdata_cachep, bd);
1008 }
1009 head = &tr->tr_databuf;
1010 while (!list_empty(head)) {
1011 bd = list_first_entry(head, struct gfs2_bufdata, bd_list);
1012 list_del_init(&bd->bd_list);
1013 if (!list_empty(&bd->bd_ail_st_list))
1014 gfs2_remove_from_ail(bd);
1015 kmem_cache_free(gfs2_bufdata_cachep, bd);
1016 }
1017}
1018
1019/**
1020 * gfs2_log_flush - flush incore transaction(s)
1021 * @sdp: The filesystem
1022 * @gl: The glock structure to flush. If NULL, flush the whole incore log
1023 * @flags: The log header flags: GFS2_LOG_HEAD_FLUSH_* and debug flags
1024 *
1025 */
1026
1027void gfs2_log_flush(struct gfs2_sbd *sdp, struct gfs2_glock *gl, u32 flags)
1028{
1029 struct gfs2_trans *tr = NULL;
1030 unsigned int reserved_blocks = 0, used_blocks = 0;
1031 enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state);
1032 unsigned int first_log_head;
1033 unsigned int reserved_revokes = 0;
1034
1035 down_write(&sdp->sd_log_flush_lock);
1036 trace_gfs2_log_flush(sdp, 1, flags);
1037
1038repeat:
1039 /*
1040 * Do this check while holding the log_flush_lock to prevent new
1041 * buffers from being added to the ail via gfs2_pin()
1042 */
1043 if (gfs2_withdrawn(sdp) || !test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
1044 goto out;
1045
1046 /* Log might have been flushed while we waited for the flush lock */
1047 if (gl && !test_bit(GLF_LFLUSH, &gl->gl_flags))
1048 goto out;
1049
1050 first_log_head = sdp->sd_log_head;
1051 sdp->sd_log_flush_head = first_log_head;
1052
1053 tr = sdp->sd_log_tr;
1054 if (tr || sdp->sd_log_num_revoke) {
1055 if (reserved_blocks)
1056 gfs2_log_release(sdp, reserved_blocks);
1057 reserved_blocks = sdp->sd_log_blks_reserved;
1058 reserved_revokes = sdp->sd_log_num_revoke;
1059 if (tr) {
1060 sdp->sd_log_tr = NULL;
1061 tr->tr_first = first_log_head;
1062 if (unlikely (state == SFS_FROZEN)) {
1063 if (gfs2_assert_withdraw_delayed(sdp,
1064 !tr->tr_num_buf_new && !tr->tr_num_databuf_new))
1065 goto out_withdraw;
1066 }
1067 }
1068 } else if (!reserved_blocks) {
1069 unsigned int taboo_blocks = GFS2_LOG_FLUSH_MIN_BLOCKS;
1070
1071 reserved_blocks = GFS2_LOG_FLUSH_MIN_BLOCKS;
1072 if (current == sdp->sd_logd_process)
1073 taboo_blocks = 0;
1074
1075 if (!__gfs2_log_try_reserve(sdp, reserved_blocks, taboo_blocks)) {
1076 up_write(&sdp->sd_log_flush_lock);
1077 __gfs2_log_reserve(sdp, reserved_blocks, taboo_blocks);
1078 down_write(&sdp->sd_log_flush_lock);
1079 goto repeat;
1080 }
1081 BUG_ON(sdp->sd_log_num_revoke);
1082 }
1083
1084 if (flags & GFS2_LOG_HEAD_FLUSH_SHUTDOWN)
1085 clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
1086
1087 if (unlikely(state == SFS_FROZEN))
1088 if (gfs2_assert_withdraw_delayed(sdp, !reserved_revokes))
1089 goto out_withdraw;
1090
1091 gfs2_ordered_write(sdp);
1092 if (gfs2_withdrawn(sdp))
1093 goto out_withdraw;
1094 lops_before_commit(sdp, tr);
1095 if (gfs2_withdrawn(sdp))
1096 goto out_withdraw;
1097 gfs2_log_submit_bio(&sdp->sd_jdesc->jd_log_bio, REQ_OP_WRITE);
1098 if (gfs2_withdrawn(sdp))
1099 goto out_withdraw;
1100
1101 if (sdp->sd_log_head != sdp->sd_log_flush_head) {
1102 log_write_header(sdp, flags);
1103 } else if (sdp->sd_log_tail != sdp->sd_log_flush_tail && !sdp->sd_log_idle) {
1104 log_write_header(sdp, flags);
1105 }
1106 if (gfs2_withdrawn(sdp))
1107 goto out_withdraw;
1108 lops_after_commit(sdp, tr);
1109
1110 gfs2_log_lock(sdp);
1111 sdp->sd_log_blks_reserved = 0;
1112
1113 spin_lock(&sdp->sd_ail_lock);
1114 if (tr && !list_empty(&tr->tr_ail1_list)) {
1115 list_add(&tr->tr_list, &sdp->sd_ail1_list);
1116 tr = NULL;
1117 }
1118 spin_unlock(&sdp->sd_ail_lock);
1119 gfs2_log_unlock(sdp);
1120
1121 if (!(flags & GFS2_LOG_HEAD_FLUSH_NORMAL)) {
1122 if (!sdp->sd_log_idle) {
1123 empty_ail1_list(sdp);
1124 if (gfs2_withdrawn(sdp))
1125 goto out_withdraw;
1126 log_write_header(sdp, flags);
1127 }
1128 if (flags & (GFS2_LOG_HEAD_FLUSH_SHUTDOWN |
1129 GFS2_LOG_HEAD_FLUSH_FREEZE))
1130 gfs2_log_shutdown(sdp);
1131 if (flags & GFS2_LOG_HEAD_FLUSH_FREEZE)
1132 atomic_set(&sdp->sd_freeze_state, SFS_FROZEN);
1133 }
1134
1135out_end:
1136 used_blocks = log_distance(sdp, sdp->sd_log_flush_head, first_log_head);
1137 reserved_revokes += atomic_read(&sdp->sd_log_revokes_available);
1138 atomic_set(&sdp->sd_log_revokes_available, sdp->sd_ldptrs);
1139 gfs2_assert_withdraw(sdp, reserved_revokes % sdp->sd_inptrs == sdp->sd_ldptrs);
1140 if (reserved_revokes > sdp->sd_ldptrs)
1141 reserved_blocks += (reserved_revokes - sdp->sd_ldptrs) / sdp->sd_inptrs;
1142out:
1143 if (used_blocks != reserved_blocks) {
1144 gfs2_assert_withdraw_delayed(sdp, used_blocks < reserved_blocks);
1145 gfs2_log_release(sdp, reserved_blocks - used_blocks);
1146 }
1147 up_write(&sdp->sd_log_flush_lock);
1148 gfs2_trans_free(sdp, tr);
1149 if (gfs2_withdrawing(sdp))
1150 gfs2_withdraw(sdp);
1151 trace_gfs2_log_flush(sdp, 0, flags);
1152 return;
1153
1154out_withdraw:
1155 trans_drain(tr);
1156 /**
1157 * If the tr_list is empty, we're withdrawing during a log
1158 * flush that targets a transaction, but the transaction was
1159 * never queued onto any of the ail lists. Here we add it to
1160 * ail1 just so that ail_drain() will find and free it.
1161 */
1162 spin_lock(&sdp->sd_ail_lock);
1163 if (tr && list_empty(&tr->tr_list))
1164 list_add(&tr->tr_list, &sdp->sd_ail1_list);
1165 spin_unlock(&sdp->sd_ail_lock);
1166 tr = NULL;
1167 goto out_end;
1168}
1169
1170/**
1171 * gfs2_merge_trans - Merge a new transaction into a cached transaction
1172 * @sdp: the filesystem
1173 * @new: New transaction to be merged
1174 */
1175
1176static void gfs2_merge_trans(struct gfs2_sbd *sdp, struct gfs2_trans *new)
1177{
1178 struct gfs2_trans *old = sdp->sd_log_tr;
1179
1180 WARN_ON_ONCE(!test_bit(TR_ATTACHED, &old->tr_flags));
1181
1182 old->tr_num_buf_new += new->tr_num_buf_new;
1183 old->tr_num_databuf_new += new->tr_num_databuf_new;
1184 old->tr_num_buf_rm += new->tr_num_buf_rm;
1185 old->tr_num_databuf_rm += new->tr_num_databuf_rm;
1186 old->tr_revokes += new->tr_revokes;
1187 old->tr_num_revoke += new->tr_num_revoke;
1188
1189 list_splice_tail_init(&new->tr_databuf, &old->tr_databuf);
1190 list_splice_tail_init(&new->tr_buf, &old->tr_buf);
1191
1192 spin_lock(&sdp->sd_ail_lock);
1193 list_splice_tail_init(&new->tr_ail1_list, &old->tr_ail1_list);
1194 list_splice_tail_init(&new->tr_ail2_list, &old->tr_ail2_list);
1195 spin_unlock(&sdp->sd_ail_lock);
1196}
1197
1198static void log_refund(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
1199{
1200 unsigned int reserved;
1201 unsigned int unused;
1202 unsigned int maxres;
1203
1204 gfs2_log_lock(sdp);
1205
1206 if (sdp->sd_log_tr) {
1207 gfs2_merge_trans(sdp, tr);
1208 } else if (tr->tr_num_buf_new || tr->tr_num_databuf_new) {
1209 gfs2_assert_withdraw(sdp, !test_bit(TR_ONSTACK, &tr->tr_flags));
1210 sdp->sd_log_tr = tr;
1211 set_bit(TR_ATTACHED, &tr->tr_flags);
1212 }
1213
1214 reserved = calc_reserved(sdp);
1215 maxres = sdp->sd_log_blks_reserved + tr->tr_reserved;
1216 gfs2_assert_withdraw(sdp, maxres >= reserved);
1217 unused = maxres - reserved;
1218 if (unused)
1219 gfs2_log_release(sdp, unused);
1220 sdp->sd_log_blks_reserved = reserved;
1221
1222 gfs2_log_unlock(sdp);
1223}
1224
1225/**
1226 * gfs2_log_commit - Commit a transaction to the log
1227 * @sdp: the filesystem
1228 * @tr: the transaction
1229 *
1230 * We wake up gfs2_logd if the number of pinned blocks exceed thresh1
1231 * or the total number of used blocks (pinned blocks plus AIL blocks)
1232 * is greater than thresh2.
1233 *
1234 * At mount time thresh1 is 2/5ths of journal size, thresh2 is 4/5ths of
1235 * journal size.
1236 *
1237 * Returns: errno
1238 */
1239
1240void gfs2_log_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
1241{
1242 log_refund(sdp, tr);
1243
1244 if (atomic_read(&sdp->sd_log_pinned) > atomic_read(&sdp->sd_log_thresh1) ||
1245 ((sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free)) >
1246 atomic_read(&sdp->sd_log_thresh2)))
1247 wake_up(&sdp->sd_logd_waitq);
1248}
1249
1250/**
1251 * gfs2_log_shutdown - write a shutdown header into a journal
1252 * @sdp: the filesystem
1253 *
1254 */
1255
1256static void gfs2_log_shutdown(struct gfs2_sbd *sdp)
1257{
1258 gfs2_assert_withdraw(sdp, !sdp->sd_log_blks_reserved);
1259 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
1260 gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail1_list));
1261
1262 log_write_header(sdp, GFS2_LOG_HEAD_UNMOUNT | GFS2_LFC_SHUTDOWN);
1263 log_pull_tail(sdp);
1264
1265 gfs2_assert_warn(sdp, sdp->sd_log_head == sdp->sd_log_tail);
1266 gfs2_assert_warn(sdp, list_empty(&sdp->sd_ail2_list));
1267}
1268
1269static inline int gfs2_jrnl_flush_reqd(struct gfs2_sbd *sdp)
1270{
1271 return (atomic_read(&sdp->sd_log_pinned) +
1272 atomic_read(&sdp->sd_log_blks_needed) >=
1273 atomic_read(&sdp->sd_log_thresh1));
1274}
1275
1276static inline int gfs2_ail_flush_reqd(struct gfs2_sbd *sdp)
1277{
1278 unsigned int used_blocks = sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free);
1279
1280 if (test_and_clear_bit(SDF_FORCE_AIL_FLUSH, &sdp->sd_flags))
1281 return 1;
1282
1283 return used_blocks + atomic_read(&sdp->sd_log_blks_needed) >=
1284 atomic_read(&sdp->sd_log_thresh2);
1285}
1286
1287/**
1288 * gfs2_logd - Update log tail as Active Items get flushed to in-place blocks
1289 * @data: Pointer to GFS2 superblock
1290 *
1291 * Also, periodically check to make sure that we're using the most recent
1292 * journal index.
1293 */
1294
1295int gfs2_logd(void *data)
1296{
1297 struct gfs2_sbd *sdp = data;
1298 unsigned long t = 1;
1299 DEFINE_WAIT(wait);
1300
1301 while (!kthread_should_stop()) {
1302
1303 if (gfs2_withdrawn(sdp)) {
1304 msleep_interruptible(HZ);
1305 continue;
1306 }
1307 /* Check for errors writing to the journal */
1308 if (sdp->sd_log_error) {
1309 gfs2_lm(sdp,
1310 "GFS2: fsid=%s: error %d: "
1311 "withdrawing the file system to "
1312 "prevent further damage.\n",
1313 sdp->sd_fsname, sdp->sd_log_error);
1314 gfs2_withdraw(sdp);
1315 continue;
1316 }
1317
1318 if (gfs2_jrnl_flush_reqd(sdp) || t == 0) {
1319 gfs2_ail1_empty(sdp, 0);
1320 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
1321 GFS2_LFC_LOGD_JFLUSH_REQD);
1322 }
1323
1324 if (gfs2_ail_flush_reqd(sdp)) {
1325 gfs2_ail1_start(sdp);
1326 gfs2_ail1_wait(sdp);
1327 gfs2_ail1_empty(sdp, 0);
1328 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
1329 GFS2_LFC_LOGD_AIL_FLUSH_REQD);
1330 }
1331
1332 t = gfs2_tune_get(sdp, gt_logd_secs) * HZ;
1333
1334 try_to_freeze();
1335
1336 do {
1337 prepare_to_wait(&sdp->sd_logd_waitq, &wait,
1338 TASK_INTERRUPTIBLE);
1339 if (!gfs2_ail_flush_reqd(sdp) &&
1340 !gfs2_jrnl_flush_reqd(sdp) &&
1341 !kthread_should_stop())
1342 t = schedule_timeout(t);
1343 } while(t && !gfs2_ail_flush_reqd(sdp) &&
1344 !gfs2_jrnl_flush_reqd(sdp) &&
1345 !kthread_should_stop());
1346 finish_wait(&sdp->sd_logd_waitq, &wait);
1347 }
1348
1349 return 0;
1350}
1351