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/*
2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2007 Red Hat, Inc. All rights reserved.
4 *
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
7 * of the GNU General Public License version 2.
8 */
9
10#include <linux/sched.h>
11#include <linux/slab.h>
12#include <linux/spinlock.h>
13#include <linux/completion.h>
14#include <linux/buffer_head.h>
15#include <linux/gfs2_ondisk.h>
16#include <linux/crc32.h>
17#include <linux/delay.h>
18#include <linux/kthread.h>
19#include <linux/freezer.h>
20#include <linux/bio.h>
21#include <linux/blkdev.h>
22#include <linux/writeback.h>
23#include <linux/list_sort.h>
24
25#include "gfs2.h"
26#include "incore.h"
27#include "bmap.h"
28#include "glock.h"
29#include "log.h"
30#include "lops.h"
31#include "meta_io.h"
32#include "util.h"
33#include "dir.h"
34#include "trace_gfs2.h"
35
36/**
37 * gfs2_struct2blk - compute stuff
38 * @sdp: the filesystem
39 * @nstruct: the number of structures
40 * @ssize: the size of the 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 unsigned int ssize)
50{
51 unsigned int blks;
52 unsigned int first, second;
53
54 blks = 1;
55 first = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / ssize;
56
57 if (nstruct > first) {
58 second = (sdp->sd_sb.sb_bsize -
59 sizeof(struct gfs2_meta_header)) / ssize;
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 * @mapping: The associated mapping (maybe NULL)
69 * @bd: The gfs2_bufdata to remove
70 *
71 * The ail lock _must_ be held when calling this function
72 *
73 */
74
75void gfs2_remove_from_ail(struct gfs2_bufdata *bd)
76{
77 bd->bd_tr = NULL;
78 list_del_init(&bd->bd_ail_st_list);
79 list_del_init(&bd->bd_ail_gl_list);
80 atomic_dec(&bd->bd_gl->gl_ail_count);
81 brelse(bd->bd_bh);
82}
83
84/**
85 * gfs2_ail1_start_one - Start I/O on a part of the AIL
86 * @sdp: the filesystem
87 * @wbc: The writeback control structure
88 * @ai: The ail structure
89 *
90 */
91
92static int gfs2_ail1_start_one(struct gfs2_sbd *sdp,
93 struct writeback_control *wbc,
94 struct gfs2_trans *tr)
95__releases(&sdp->sd_ail_lock)
96__acquires(&sdp->sd_ail_lock)
97{
98 struct gfs2_glock *gl = NULL;
99 struct address_space *mapping;
100 struct gfs2_bufdata *bd, *s;
101 struct buffer_head *bh;
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 gfs2_io_error_bh(sdp, bh);
111 list_move(&bd->bd_ail_st_list, &tr->tr_ail2_list);
112 continue;
113 }
114
115 if (!buffer_dirty(bh))
116 continue;
117 if (gl == bd->bd_gl)
118 continue;
119 gl = bd->bd_gl;
120 list_move(&bd->bd_ail_st_list, &tr->tr_ail1_list);
121 mapping = bh->b_page->mapping;
122 if (!mapping)
123 continue;
124 spin_unlock(&sdp->sd_ail_lock);
125 generic_writepages(mapping, wbc);
126 spin_lock(&sdp->sd_ail_lock);
127 if (wbc->nr_to_write <= 0)
128 break;
129 return 1;
130 }
131
132 return 0;
133}
134
135
136/**
137 * gfs2_ail1_flush - start writeback of some ail1 entries
138 * @sdp: The super block
139 * @wbc: The writeback control structure
140 *
141 * Writes back some ail1 entries, according to the limits in the
142 * writeback control structure
143 */
144
145void gfs2_ail1_flush(struct gfs2_sbd *sdp, struct writeback_control *wbc)
146{
147 struct list_head *head = &sdp->sd_ail1_list;
148 struct gfs2_trans *tr;
149 struct blk_plug plug;
150
151 trace_gfs2_ail_flush(sdp, wbc, 1);
152 blk_start_plug(&plug);
153 spin_lock(&sdp->sd_ail_lock);
154restart:
155 list_for_each_entry_reverse(tr, head, tr_list) {
156 if (wbc->nr_to_write <= 0)
157 break;
158 if (gfs2_ail1_start_one(sdp, wbc, tr))
159 goto restart;
160 }
161 spin_unlock(&sdp->sd_ail_lock);
162 blk_finish_plug(&plug);
163 trace_gfs2_ail_flush(sdp, wbc, 0);
164}
165
166/**
167 * gfs2_ail1_start - start writeback of all ail1 entries
168 * @sdp: The superblock
169 */
170
171static void gfs2_ail1_start(struct gfs2_sbd *sdp)
172{
173 struct writeback_control wbc = {
174 .sync_mode = WB_SYNC_NONE,
175 .nr_to_write = LONG_MAX,
176 .range_start = 0,
177 .range_end = LLONG_MAX,
178 };
179
180 return gfs2_ail1_flush(sdp, &wbc);
181}
182
183/**
184 * gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced
185 * @sdp: the filesystem
186 * @ai: the AIL entry
187 *
188 */
189
190static void gfs2_ail1_empty_one(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
191{
192 struct gfs2_bufdata *bd, *s;
193 struct buffer_head *bh;
194
195 list_for_each_entry_safe_reverse(bd, s, &tr->tr_ail1_list,
196 bd_ail_st_list) {
197 bh = bd->bd_bh;
198 gfs2_assert(sdp, bd->bd_tr == tr);
199 if (buffer_busy(bh))
200 continue;
201 if (!buffer_uptodate(bh))
202 gfs2_io_error_bh(sdp, bh);
203 list_move(&bd->bd_ail_st_list, &tr->tr_ail2_list);
204 }
205
206}
207
208/**
209 * gfs2_ail1_empty - Try to empty the ail1 lists
210 * @sdp: The superblock
211 *
212 * Tries to empty the ail1 lists, starting with the oldest first
213 */
214
215static int gfs2_ail1_empty(struct gfs2_sbd *sdp)
216{
217 struct gfs2_trans *tr, *s;
218 int oldest_tr = 1;
219 int ret;
220
221 spin_lock(&sdp->sd_ail_lock);
222 list_for_each_entry_safe_reverse(tr, s, &sdp->sd_ail1_list, tr_list) {
223 gfs2_ail1_empty_one(sdp, tr);
224 if (list_empty(&tr->tr_ail1_list) && oldest_tr)
225 list_move(&tr->tr_list, &sdp->sd_ail2_list);
226 else
227 oldest_tr = 0;
228 }
229 ret = list_empty(&sdp->sd_ail1_list);
230 spin_unlock(&sdp->sd_ail_lock);
231
232 return ret;
233}
234
235static void gfs2_ail1_wait(struct gfs2_sbd *sdp)
236{
237 struct gfs2_trans *tr;
238 struct gfs2_bufdata *bd;
239 struct buffer_head *bh;
240
241 spin_lock(&sdp->sd_ail_lock);
242 list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) {
243 list_for_each_entry(bd, &tr->tr_ail1_list, bd_ail_st_list) {
244 bh = bd->bd_bh;
245 if (!buffer_locked(bh))
246 continue;
247 get_bh(bh);
248 spin_unlock(&sdp->sd_ail_lock);
249 wait_on_buffer(bh);
250 brelse(bh);
251 return;
252 }
253 }
254 spin_unlock(&sdp->sd_ail_lock);
255}
256
257/**
258 * gfs2_ail2_empty_one - Check whether or not a trans in the AIL has been synced
259 * @sdp: the filesystem
260 * @ai: the AIL entry
261 *
262 */
263
264static void gfs2_ail2_empty_one(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
265{
266 struct list_head *head = &tr->tr_ail2_list;
267 struct gfs2_bufdata *bd;
268
269 while (!list_empty(head)) {
270 bd = list_entry(head->prev, struct gfs2_bufdata,
271 bd_ail_st_list);
272 gfs2_assert(sdp, bd->bd_tr == tr);
273 gfs2_remove_from_ail(bd);
274 }
275}
276
277static void ail2_empty(struct gfs2_sbd *sdp, unsigned int new_tail)
278{
279 struct gfs2_trans *tr, *safe;
280 unsigned int old_tail = sdp->sd_log_tail;
281 int wrap = (new_tail < old_tail);
282 int a, b, rm;
283
284 spin_lock(&sdp->sd_ail_lock);
285
286 list_for_each_entry_safe(tr, safe, &sdp->sd_ail2_list, tr_list) {
287 a = (old_tail <= tr->tr_first);
288 b = (tr->tr_first < new_tail);
289 rm = (wrap) ? (a || b) : (a && b);
290 if (!rm)
291 continue;
292
293 gfs2_ail2_empty_one(sdp, tr);
294 list_del(&tr->tr_list);
295 gfs2_assert_warn(sdp, list_empty(&tr->tr_ail1_list));
296 gfs2_assert_warn(sdp, list_empty(&tr->tr_ail2_list));
297 kfree(tr);
298 }
299
300 spin_unlock(&sdp->sd_ail_lock);
301}
302
303/**
304 * gfs2_log_release - Release a given number of log blocks
305 * @sdp: The GFS2 superblock
306 * @blks: The number of blocks
307 *
308 */
309
310void gfs2_log_release(struct gfs2_sbd *sdp, unsigned int blks)
311{
312
313 atomic_add(blks, &sdp->sd_log_blks_free);
314 trace_gfs2_log_blocks(sdp, blks);
315 gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
316 sdp->sd_jdesc->jd_blocks);
317 up_read(&sdp->sd_log_flush_lock);
318}
319
320/**
321 * gfs2_log_reserve - Make a log reservation
322 * @sdp: The GFS2 superblock
323 * @blks: The number of blocks to reserve
324 *
325 * Note that we never give out the last few blocks of the journal. Thats
326 * due to the fact that there is a small number of header blocks
327 * associated with each log flush. The exact number can't be known until
328 * flush time, so we ensure that we have just enough free blocks at all
329 * times to avoid running out during a log flush.
330 *
331 * We no longer flush the log here, instead we wake up logd to do that
332 * for us. To avoid the thundering herd and to ensure that we deal fairly
333 * with queued waiters, we use an exclusive wait. This means that when we
334 * get woken with enough journal space to get our reservation, we need to
335 * wake the next waiter on the list.
336 *
337 * Returns: errno
338 */
339
340int gfs2_log_reserve(struct gfs2_sbd *sdp, unsigned int blks)
341{
342 int ret = 0;
343 unsigned reserved_blks = 7 * (4096 / sdp->sd_vfs->s_blocksize);
344 unsigned wanted = blks + reserved_blks;
345 DEFINE_WAIT(wait);
346 int did_wait = 0;
347 unsigned int free_blocks;
348
349 if (gfs2_assert_warn(sdp, blks) ||
350 gfs2_assert_warn(sdp, blks <= sdp->sd_jdesc->jd_blocks))
351 return -EINVAL;
352retry:
353 free_blocks = atomic_read(&sdp->sd_log_blks_free);
354 if (unlikely(free_blocks <= wanted)) {
355 do {
356 prepare_to_wait_exclusive(&sdp->sd_log_waitq, &wait,
357 TASK_UNINTERRUPTIBLE);
358 wake_up(&sdp->sd_logd_waitq);
359 did_wait = 1;
360 if (atomic_read(&sdp->sd_log_blks_free) <= wanted)
361 io_schedule();
362 free_blocks = atomic_read(&sdp->sd_log_blks_free);
363 } while(free_blocks <= wanted);
364 finish_wait(&sdp->sd_log_waitq, &wait);
365 }
366 atomic_inc(&sdp->sd_reserving_log);
367 if (atomic_cmpxchg(&sdp->sd_log_blks_free, free_blocks,
368 free_blocks - blks) != free_blocks) {
369 if (atomic_dec_and_test(&sdp->sd_reserving_log))
370 wake_up(&sdp->sd_reserving_log_wait);
371 goto retry;
372 }
373 trace_gfs2_log_blocks(sdp, -blks);
374
375 /*
376 * If we waited, then so might others, wake them up _after_ we get
377 * our share of the log.
378 */
379 if (unlikely(did_wait))
380 wake_up(&sdp->sd_log_waitq);
381
382 down_read(&sdp->sd_log_flush_lock);
383 if (unlikely(!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))) {
384 gfs2_log_release(sdp, blks);
385 ret = -EROFS;
386 }
387 if (atomic_dec_and_test(&sdp->sd_reserving_log))
388 wake_up(&sdp->sd_reserving_log_wait);
389 return ret;
390}
391
392/**
393 * log_distance - Compute distance between two journal blocks
394 * @sdp: The GFS2 superblock
395 * @newer: The most recent journal block of the pair
396 * @older: The older journal block of the pair
397 *
398 * Compute the distance (in the journal direction) between two
399 * blocks in the journal
400 *
401 * Returns: the distance in blocks
402 */
403
404static inline unsigned int log_distance(struct gfs2_sbd *sdp, unsigned int newer,
405 unsigned int older)
406{
407 int dist;
408
409 dist = newer - older;
410 if (dist < 0)
411 dist += sdp->sd_jdesc->jd_blocks;
412
413 return dist;
414}
415
416/**
417 * calc_reserved - Calculate the number of blocks to reserve when
418 * refunding a transaction's unused buffers.
419 * @sdp: The GFS2 superblock
420 *
421 * This is complex. We need to reserve room for all our currently used
422 * metadata buffers (e.g. normal file I/O rewriting file time stamps) and
423 * all our journaled data buffers for journaled files (e.g. files in the
424 * meta_fs like rindex, or files for which chattr +j was done.)
425 * If we don't reserve enough space, gfs2_log_refund and gfs2_log_flush
426 * will count it as free space (sd_log_blks_free) and corruption will follow.
427 *
428 * We can have metadata bufs and jdata bufs in the same journal. So each
429 * type gets its own log header, for which we need to reserve a block.
430 * In fact, each type has the potential for needing more than one header
431 * in cases where we have more buffers than will fit on a journal page.
432 * Metadata journal entries take up half the space of journaled buffer entries.
433 * Thus, metadata entries have buf_limit (502) and journaled buffers have
434 * databuf_limit (251) before they cause a wrap around.
435 *
436 * Also, we need to reserve blocks for revoke journal entries and one for an
437 * overall header for the lot.
438 *
439 * Returns: the number of blocks reserved
440 */
441static unsigned int calc_reserved(struct gfs2_sbd *sdp)
442{
443 unsigned int reserved = 0;
444 unsigned int mbuf;
445 unsigned int dbuf;
446 struct gfs2_trans *tr = sdp->sd_log_tr;
447
448 if (tr) {
449 mbuf = tr->tr_num_buf_new - tr->tr_num_buf_rm;
450 dbuf = tr->tr_num_databuf_new - tr->tr_num_databuf_rm;
451 reserved = mbuf + dbuf;
452 /* Account for header blocks */
453 reserved += DIV_ROUND_UP(mbuf, buf_limit(sdp));
454 reserved += DIV_ROUND_UP(dbuf, databuf_limit(sdp));
455 }
456
457 if (sdp->sd_log_commited_revoke > 0)
458 reserved += gfs2_struct2blk(sdp, sdp->sd_log_commited_revoke,
459 sizeof(u64));
460 /* One for the overall header */
461 if (reserved)
462 reserved++;
463 return reserved;
464}
465
466static unsigned int current_tail(struct gfs2_sbd *sdp)
467{
468 struct gfs2_trans *tr;
469 unsigned int tail;
470
471 spin_lock(&sdp->sd_ail_lock);
472
473 if (list_empty(&sdp->sd_ail1_list)) {
474 tail = sdp->sd_log_head;
475 } else {
476 tr = list_entry(sdp->sd_ail1_list.prev, struct gfs2_trans,
477 tr_list);
478 tail = tr->tr_first;
479 }
480
481 spin_unlock(&sdp->sd_ail_lock);
482
483 return tail;
484}
485
486static void log_pull_tail(struct gfs2_sbd *sdp, unsigned int new_tail)
487{
488 unsigned int dist = log_distance(sdp, new_tail, sdp->sd_log_tail);
489
490 ail2_empty(sdp, new_tail);
491
492 atomic_add(dist, &sdp->sd_log_blks_free);
493 trace_gfs2_log_blocks(sdp, dist);
494 gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
495 sdp->sd_jdesc->jd_blocks);
496
497 sdp->sd_log_tail = new_tail;
498}
499
500
501static void log_flush_wait(struct gfs2_sbd *sdp)
502{
503 DEFINE_WAIT(wait);
504
505 if (atomic_read(&sdp->sd_log_in_flight)) {
506 do {
507 prepare_to_wait(&sdp->sd_log_flush_wait, &wait,
508 TASK_UNINTERRUPTIBLE);
509 if (atomic_read(&sdp->sd_log_in_flight))
510 io_schedule();
511 } while(atomic_read(&sdp->sd_log_in_flight));
512 finish_wait(&sdp->sd_log_flush_wait, &wait);
513 }
514}
515
516static int ip_cmp(void *priv, struct list_head *a, struct list_head *b)
517{
518 struct gfs2_inode *ipa, *ipb;
519
520 ipa = list_entry(a, struct gfs2_inode, i_ordered);
521 ipb = list_entry(b, struct gfs2_inode, i_ordered);
522
523 if (ipa->i_no_addr < ipb->i_no_addr)
524 return -1;
525 if (ipa->i_no_addr > ipb->i_no_addr)
526 return 1;
527 return 0;
528}
529
530static void gfs2_ordered_write(struct gfs2_sbd *sdp)
531{
532 struct gfs2_inode *ip;
533 LIST_HEAD(written);
534
535 spin_lock(&sdp->sd_ordered_lock);
536 list_sort(NULL, &sdp->sd_log_le_ordered, &ip_cmp);
537 while (!list_empty(&sdp->sd_log_le_ordered)) {
538 ip = list_entry(sdp->sd_log_le_ordered.next, struct gfs2_inode, i_ordered);
539 list_move(&ip->i_ordered, &written);
540 if (ip->i_inode.i_mapping->nrpages == 0)
541 continue;
542 spin_unlock(&sdp->sd_ordered_lock);
543 filemap_fdatawrite(ip->i_inode.i_mapping);
544 spin_lock(&sdp->sd_ordered_lock);
545 }
546 list_splice(&written, &sdp->sd_log_le_ordered);
547 spin_unlock(&sdp->sd_ordered_lock);
548}
549
550static void gfs2_ordered_wait(struct gfs2_sbd *sdp)
551{
552 struct gfs2_inode *ip;
553
554 spin_lock(&sdp->sd_ordered_lock);
555 while (!list_empty(&sdp->sd_log_le_ordered)) {
556 ip = list_entry(sdp->sd_log_le_ordered.next, struct gfs2_inode, i_ordered);
557 list_del(&ip->i_ordered);
558 WARN_ON(!test_and_clear_bit(GIF_ORDERED, &ip->i_flags));
559 if (ip->i_inode.i_mapping->nrpages == 0)
560 continue;
561 spin_unlock(&sdp->sd_ordered_lock);
562 filemap_fdatawait(ip->i_inode.i_mapping);
563 spin_lock(&sdp->sd_ordered_lock);
564 }
565 spin_unlock(&sdp->sd_ordered_lock);
566}
567
568void gfs2_ordered_del_inode(struct gfs2_inode *ip)
569{
570 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
571
572 spin_lock(&sdp->sd_ordered_lock);
573 if (test_and_clear_bit(GIF_ORDERED, &ip->i_flags))
574 list_del(&ip->i_ordered);
575 spin_unlock(&sdp->sd_ordered_lock);
576}
577
578void gfs2_add_revoke(struct gfs2_sbd *sdp, struct gfs2_bufdata *bd)
579{
580 struct buffer_head *bh = bd->bd_bh;
581 struct gfs2_glock *gl = bd->bd_gl;
582
583 bh->b_private = NULL;
584 bd->bd_blkno = bh->b_blocknr;
585 gfs2_remove_from_ail(bd); /* drops ref on bh */
586 bd->bd_bh = NULL;
587 bd->bd_ops = &gfs2_revoke_lops;
588 sdp->sd_log_num_revoke++;
589 atomic_inc(&gl->gl_revokes);
590 set_bit(GLF_LFLUSH, &gl->gl_flags);
591 list_add(&bd->bd_list, &sdp->sd_log_le_revoke);
592}
593
594void gfs2_write_revokes(struct gfs2_sbd *sdp)
595{
596 struct gfs2_trans *tr;
597 struct gfs2_bufdata *bd, *tmp;
598 int have_revokes = 0;
599 int max_revokes = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / sizeof(u64);
600
601 gfs2_ail1_empty(sdp);
602 spin_lock(&sdp->sd_ail_lock);
603 list_for_each_entry(tr, &sdp->sd_ail1_list, tr_list) {
604 list_for_each_entry(bd, &tr->tr_ail2_list, bd_ail_st_list) {
605 if (list_empty(&bd->bd_list)) {
606 have_revokes = 1;
607 goto done;
608 }
609 }
610 }
611done:
612 spin_unlock(&sdp->sd_ail_lock);
613 if (have_revokes == 0)
614 return;
615 while (sdp->sd_log_num_revoke > max_revokes)
616 max_revokes += (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header)) / sizeof(u64);
617 max_revokes -= sdp->sd_log_num_revoke;
618 if (!sdp->sd_log_num_revoke) {
619 atomic_dec(&sdp->sd_log_blks_free);
620 /* If no blocks have been reserved, we need to also
621 * reserve a block for the header */
622 if (!sdp->sd_log_blks_reserved)
623 atomic_dec(&sdp->sd_log_blks_free);
624 }
625 gfs2_log_lock(sdp);
626 spin_lock(&sdp->sd_ail_lock);
627 list_for_each_entry(tr, &sdp->sd_ail1_list, tr_list) {
628 list_for_each_entry_safe(bd, tmp, &tr->tr_ail2_list, bd_ail_st_list) {
629 if (max_revokes == 0)
630 goto out_of_blocks;
631 if (!list_empty(&bd->bd_list))
632 continue;
633 gfs2_add_revoke(sdp, bd);
634 max_revokes--;
635 }
636 }
637out_of_blocks:
638 spin_unlock(&sdp->sd_ail_lock);
639 gfs2_log_unlock(sdp);
640
641 if (!sdp->sd_log_num_revoke) {
642 atomic_inc(&sdp->sd_log_blks_free);
643 if (!sdp->sd_log_blks_reserved)
644 atomic_inc(&sdp->sd_log_blks_free);
645 }
646}
647
648/**
649 * log_write_header - Get and initialize a journal header buffer
650 * @sdp: The GFS2 superblock
651 *
652 * Returns: the initialized log buffer descriptor
653 */
654
655static void log_write_header(struct gfs2_sbd *sdp, u32 flags)
656{
657 struct gfs2_log_header *lh;
658 unsigned int tail;
659 u32 hash;
660 int op_flags = REQ_PREFLUSH | REQ_FUA | REQ_META;
661 struct page *page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
662 enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state);
663 lh = page_address(page);
664 clear_page(lh);
665
666 gfs2_assert_withdraw(sdp, (state != SFS_FROZEN));
667
668 tail = current_tail(sdp);
669
670 lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
671 lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH);
672 lh->lh_header.__pad0 = cpu_to_be64(0);
673 lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH);
674 lh->lh_header.mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid);
675 lh->lh_sequence = cpu_to_be64(sdp->sd_log_sequence++);
676 lh->lh_flags = cpu_to_be32(flags);
677 lh->lh_tail = cpu_to_be32(tail);
678 lh->lh_blkno = cpu_to_be32(sdp->sd_log_flush_head);
679 hash = gfs2_disk_hash(page_address(page), sizeof(struct gfs2_log_header));
680 lh->lh_hash = cpu_to_be32(hash);
681
682 if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags)) {
683 gfs2_ordered_wait(sdp);
684 log_flush_wait(sdp);
685 op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
686 }
687
688 sdp->sd_log_idle = (tail == sdp->sd_log_flush_head);
689 gfs2_log_write_page(sdp, page);
690 gfs2_log_flush_bio(sdp, REQ_OP_WRITE, op_flags);
691 log_flush_wait(sdp);
692
693 if (sdp->sd_log_tail != tail)
694 log_pull_tail(sdp, tail);
695}
696
697/**
698 * gfs2_log_flush - flush incore transaction(s)
699 * @sdp: the filesystem
700 * @gl: The glock structure to flush. If NULL, flush the whole incore log
701 *
702 */
703
704void gfs2_log_flush(struct gfs2_sbd *sdp, struct gfs2_glock *gl,
705 enum gfs2_flush_type type)
706{
707 struct gfs2_trans *tr;
708 enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state);
709
710 down_write(&sdp->sd_log_flush_lock);
711
712 /* Log might have been flushed while we waited for the flush lock */
713 if (gl && !test_bit(GLF_LFLUSH, &gl->gl_flags)) {
714 up_write(&sdp->sd_log_flush_lock);
715 return;
716 }
717 trace_gfs2_log_flush(sdp, 1);
718
719 if (type == SHUTDOWN_FLUSH)
720 clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
721
722 sdp->sd_log_flush_head = sdp->sd_log_head;
723 sdp->sd_log_flush_wrapped = 0;
724 tr = sdp->sd_log_tr;
725 if (tr) {
726 sdp->sd_log_tr = NULL;
727 INIT_LIST_HEAD(&tr->tr_ail1_list);
728 INIT_LIST_HEAD(&tr->tr_ail2_list);
729 tr->tr_first = sdp->sd_log_flush_head;
730 if (unlikely (state == SFS_FROZEN))
731 gfs2_assert_withdraw(sdp, !tr->tr_num_buf_new && !tr->tr_num_databuf_new);
732 }
733
734 if (unlikely(state == SFS_FROZEN))
735 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
736 gfs2_assert_withdraw(sdp,
737 sdp->sd_log_num_revoke == sdp->sd_log_commited_revoke);
738
739 gfs2_ordered_write(sdp);
740 lops_before_commit(sdp, tr);
741 gfs2_log_flush_bio(sdp, REQ_OP_WRITE, 0);
742
743 if (sdp->sd_log_head != sdp->sd_log_flush_head) {
744 log_flush_wait(sdp);
745 log_write_header(sdp, 0);
746 } else if (sdp->sd_log_tail != current_tail(sdp) && !sdp->sd_log_idle){
747 atomic_dec(&sdp->sd_log_blks_free); /* Adjust for unreserved buffer */
748 trace_gfs2_log_blocks(sdp, -1);
749 log_write_header(sdp, 0);
750 }
751 lops_after_commit(sdp, tr);
752
753 gfs2_log_lock(sdp);
754 sdp->sd_log_head = sdp->sd_log_flush_head;
755 sdp->sd_log_blks_reserved = 0;
756 sdp->sd_log_commited_revoke = 0;
757
758 spin_lock(&sdp->sd_ail_lock);
759 if (tr && !list_empty(&tr->tr_ail1_list)) {
760 list_add(&tr->tr_list, &sdp->sd_ail1_list);
761 tr = NULL;
762 }
763 spin_unlock(&sdp->sd_ail_lock);
764 gfs2_log_unlock(sdp);
765
766 if (type != NORMAL_FLUSH) {
767 if (!sdp->sd_log_idle) {
768 for (;;) {
769 gfs2_ail1_start(sdp);
770 gfs2_ail1_wait(sdp);
771 if (gfs2_ail1_empty(sdp))
772 break;
773 }
774 atomic_dec(&sdp->sd_log_blks_free); /* Adjust for unreserved buffer */
775 trace_gfs2_log_blocks(sdp, -1);
776 sdp->sd_log_flush_wrapped = 0;
777 log_write_header(sdp, 0);
778 sdp->sd_log_head = sdp->sd_log_flush_head;
779 }
780 if (type == SHUTDOWN_FLUSH || type == FREEZE_FLUSH)
781 gfs2_log_shutdown(sdp);
782 if (type == FREEZE_FLUSH)
783 atomic_set(&sdp->sd_freeze_state, SFS_FROZEN);
784 }
785
786 trace_gfs2_log_flush(sdp, 0);
787 up_write(&sdp->sd_log_flush_lock);
788
789 kfree(tr);
790}
791
792/**
793 * gfs2_merge_trans - Merge a new transaction into a cached transaction
794 * @old: Original transaction to be expanded
795 * @new: New transaction to be merged
796 */
797
798static void gfs2_merge_trans(struct gfs2_trans *old, struct gfs2_trans *new)
799{
800 WARN_ON_ONCE(old->tr_attached != 1);
801
802 old->tr_num_buf_new += new->tr_num_buf_new;
803 old->tr_num_databuf_new += new->tr_num_databuf_new;
804 old->tr_num_buf_rm += new->tr_num_buf_rm;
805 old->tr_num_databuf_rm += new->tr_num_databuf_rm;
806 old->tr_num_revoke += new->tr_num_revoke;
807 old->tr_num_revoke_rm += new->tr_num_revoke_rm;
808
809 list_splice_tail_init(&new->tr_databuf, &old->tr_databuf);
810 list_splice_tail_init(&new->tr_buf, &old->tr_buf);
811}
812
813static void log_refund(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
814{
815 unsigned int reserved;
816 unsigned int unused;
817 unsigned int maxres;
818
819 gfs2_log_lock(sdp);
820
821 if (sdp->sd_log_tr) {
822 gfs2_merge_trans(sdp->sd_log_tr, tr);
823 } else if (tr->tr_num_buf_new || tr->tr_num_databuf_new) {
824 gfs2_assert_withdraw(sdp, tr->tr_alloced);
825 sdp->sd_log_tr = tr;
826 tr->tr_attached = 1;
827 }
828
829 sdp->sd_log_commited_revoke += tr->tr_num_revoke - tr->tr_num_revoke_rm;
830 reserved = calc_reserved(sdp);
831 maxres = sdp->sd_log_blks_reserved + tr->tr_reserved;
832 gfs2_assert_withdraw(sdp, maxres >= reserved);
833 unused = maxres - reserved;
834 atomic_add(unused, &sdp->sd_log_blks_free);
835 trace_gfs2_log_blocks(sdp, unused);
836 gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
837 sdp->sd_jdesc->jd_blocks);
838 sdp->sd_log_blks_reserved = reserved;
839
840 gfs2_log_unlock(sdp);
841}
842
843/**
844 * gfs2_log_commit - Commit a transaction to the log
845 * @sdp: the filesystem
846 * @tr: the transaction
847 *
848 * We wake up gfs2_logd if the number of pinned blocks exceed thresh1
849 * or the total number of used blocks (pinned blocks plus AIL blocks)
850 * is greater than thresh2.
851 *
852 * At mount time thresh1 is 1/3rd of journal size, thresh2 is 2/3rd of
853 * journal size.
854 *
855 * Returns: errno
856 */
857
858void gfs2_log_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
859{
860 log_refund(sdp, tr);
861
862 if (atomic_read(&sdp->sd_log_pinned) > atomic_read(&sdp->sd_log_thresh1) ||
863 ((sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free)) >
864 atomic_read(&sdp->sd_log_thresh2)))
865 wake_up(&sdp->sd_logd_waitq);
866}
867
868/**
869 * gfs2_log_shutdown - write a shutdown header into a journal
870 * @sdp: the filesystem
871 *
872 */
873
874void gfs2_log_shutdown(struct gfs2_sbd *sdp)
875{
876 gfs2_assert_withdraw(sdp, !sdp->sd_log_blks_reserved);
877 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
878 gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail1_list));
879
880 sdp->sd_log_flush_head = sdp->sd_log_head;
881 sdp->sd_log_flush_wrapped = 0;
882
883 log_write_header(sdp, GFS2_LOG_HEAD_UNMOUNT);
884
885 gfs2_assert_warn(sdp, sdp->sd_log_head == sdp->sd_log_tail);
886 gfs2_assert_warn(sdp, list_empty(&sdp->sd_ail2_list));
887
888 sdp->sd_log_head = sdp->sd_log_flush_head;
889 sdp->sd_log_tail = sdp->sd_log_head;
890}
891
892static inline int gfs2_jrnl_flush_reqd(struct gfs2_sbd *sdp)
893{
894 return (atomic_read(&sdp->sd_log_pinned) >= atomic_read(&sdp->sd_log_thresh1));
895}
896
897static inline int gfs2_ail_flush_reqd(struct gfs2_sbd *sdp)
898{
899 unsigned int used_blocks = sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free);
900 return used_blocks >= atomic_read(&sdp->sd_log_thresh2);
901}
902
903/**
904 * gfs2_logd - Update log tail as Active Items get flushed to in-place blocks
905 * @sdp: Pointer to GFS2 superblock
906 *
907 * Also, periodically check to make sure that we're using the most recent
908 * journal index.
909 */
910
911int gfs2_logd(void *data)
912{
913 struct gfs2_sbd *sdp = data;
914 unsigned long t = 1;
915 DEFINE_WAIT(wait);
916
917 while (!kthread_should_stop()) {
918
919 if (gfs2_jrnl_flush_reqd(sdp) || t == 0) {
920 gfs2_ail1_empty(sdp);
921 gfs2_log_flush(sdp, NULL, NORMAL_FLUSH);
922 }
923
924 if (gfs2_ail_flush_reqd(sdp)) {
925 gfs2_ail1_start(sdp);
926 gfs2_ail1_wait(sdp);
927 gfs2_ail1_empty(sdp);
928 gfs2_log_flush(sdp, NULL, NORMAL_FLUSH);
929 }
930
931 if (!gfs2_ail_flush_reqd(sdp))
932 wake_up(&sdp->sd_log_waitq);
933
934 t = gfs2_tune_get(sdp, gt_logd_secs) * HZ;
935
936 try_to_freeze();
937
938 do {
939 prepare_to_wait(&sdp->sd_logd_waitq, &wait,
940 TASK_INTERRUPTIBLE);
941 if (!gfs2_ail_flush_reqd(sdp) &&
942 !gfs2_jrnl_flush_reqd(sdp) &&
943 !kthread_should_stop())
944 t = schedule_timeout(t);
945 } while(t && !gfs2_ail_flush_reqd(sdp) &&
946 !gfs2_jrnl_flush_reqd(sdp) &&
947 !kthread_should_stop());
948 finish_wait(&sdp->sd_logd_waitq, &wait);
949 }
950
951 return 0;
952}
953