Loading...
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/writeback.h>
22#include <linux/list_sort.h>
23
24#include "gfs2.h"
25#include "incore.h"
26#include "bmap.h"
27#include "glock.h"
28#include "log.h"
29#include "lops.h"
30#include "meta_io.h"
31#include "util.h"
32#include "dir.h"
33#include "trace_gfs2.h"
34
35/**
36 * gfs2_struct2blk - compute stuff
37 * @sdp: the filesystem
38 * @nstruct: the number of structures
39 * @ssize: the size of the structures
40 *
41 * Compute the number of log descriptor blocks needed to hold a certain number
42 * of structures of a certain size.
43 *
44 * Returns: the number of blocks needed (minimum is always 1)
45 */
46
47unsigned int gfs2_struct2blk(struct gfs2_sbd *sdp, unsigned int nstruct,
48 unsigned int ssize)
49{
50 unsigned int blks;
51 unsigned int first, second;
52
53 blks = 1;
54 first = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / ssize;
55
56 if (nstruct > first) {
57 second = (sdp->sd_sb.sb_bsize -
58 sizeof(struct gfs2_meta_header)) / ssize;
59 blks += DIV_ROUND_UP(nstruct - first, second);
60 }
61
62 return blks;
63}
64
65/**
66 * gfs2_remove_from_ail - Remove an entry from the ail lists, updating counters
67 * @mapping: The associated mapping (maybe NULL)
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_ail = 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 part of the AIL
85 * @sdp: the filesystem
86 * @wbc: The writeback control structure
87 * @ai: The ail structure
88 *
89 */
90
91static int gfs2_ail1_start_one(struct gfs2_sbd *sdp,
92 struct writeback_control *wbc,
93 struct gfs2_ail *ai)
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
102 list_for_each_entry_safe_reverse(bd, s, &ai->ai_ail1_list, bd_ail_st_list) {
103 bh = bd->bd_bh;
104
105 gfs2_assert(sdp, bd->bd_ail == ai);
106
107 if (!buffer_busy(bh)) {
108 if (!buffer_uptodate(bh))
109 gfs2_io_error_bh(sdp, bh);
110 list_move(&bd->bd_ail_st_list, &ai->ai_ail2_list);
111 continue;
112 }
113
114 if (!buffer_dirty(bh))
115 continue;
116 if (gl == bd->bd_gl)
117 continue;
118 gl = bd->bd_gl;
119 list_move(&bd->bd_ail_st_list, &ai->ai_ail1_list);
120 mapping = bh->b_page->mapping;
121 if (!mapping)
122 continue;
123 spin_unlock(&sdp->sd_ail_lock);
124 generic_writepages(mapping, wbc);
125 spin_lock(&sdp->sd_ail_lock);
126 if (wbc->nr_to_write <= 0)
127 break;
128 return 1;
129 }
130
131 return 0;
132}
133
134
135/**
136 * gfs2_ail1_flush - start writeback of some ail1 entries
137 * @sdp: The super block
138 * @wbc: The writeback control structure
139 *
140 * Writes back some ail1 entries, according to the limits in the
141 * writeback control structure
142 */
143
144void gfs2_ail1_flush(struct gfs2_sbd *sdp, struct writeback_control *wbc)
145{
146 struct list_head *head = &sdp->sd_ail1_list;
147 struct gfs2_ail *ai;
148
149 trace_gfs2_ail_flush(sdp, wbc, 1);
150 spin_lock(&sdp->sd_ail_lock);
151restart:
152 list_for_each_entry_reverse(ai, head, ai_list) {
153 if (wbc->nr_to_write <= 0)
154 break;
155 if (gfs2_ail1_start_one(sdp, wbc, ai))
156 goto restart;
157 }
158 spin_unlock(&sdp->sd_ail_lock);
159 trace_gfs2_ail_flush(sdp, wbc, 0);
160}
161
162/**
163 * gfs2_ail1_start - start writeback of all ail1 entries
164 * @sdp: The superblock
165 */
166
167static void gfs2_ail1_start(struct gfs2_sbd *sdp)
168{
169 struct writeback_control wbc = {
170 .sync_mode = WB_SYNC_NONE,
171 .nr_to_write = LONG_MAX,
172 .range_start = 0,
173 .range_end = LLONG_MAX,
174 };
175
176 return gfs2_ail1_flush(sdp, &wbc);
177}
178
179/**
180 * gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced
181 * @sdp: the filesystem
182 * @ai: the AIL entry
183 *
184 */
185
186static void gfs2_ail1_empty_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
187{
188 struct gfs2_bufdata *bd, *s;
189 struct buffer_head *bh;
190
191 list_for_each_entry_safe_reverse(bd, s, &ai->ai_ail1_list,
192 bd_ail_st_list) {
193 bh = bd->bd_bh;
194 gfs2_assert(sdp, bd->bd_ail == ai);
195 if (buffer_busy(bh))
196 continue;
197 if (!buffer_uptodate(bh))
198 gfs2_io_error_bh(sdp, bh);
199 list_move(&bd->bd_ail_st_list, &ai->ai_ail2_list);
200 }
201
202}
203
204/**
205 * gfs2_ail1_empty - Try to empty the ail1 lists
206 * @sdp: The superblock
207 *
208 * Tries to empty the ail1 lists, starting with the oldest first
209 */
210
211static int gfs2_ail1_empty(struct gfs2_sbd *sdp)
212{
213 struct gfs2_ail *ai, *s;
214 int ret;
215
216 spin_lock(&sdp->sd_ail_lock);
217 list_for_each_entry_safe_reverse(ai, s, &sdp->sd_ail1_list, ai_list) {
218 gfs2_ail1_empty_one(sdp, ai);
219 if (list_empty(&ai->ai_ail1_list))
220 list_move(&ai->ai_list, &sdp->sd_ail2_list);
221 else
222 break;
223 }
224 ret = list_empty(&sdp->sd_ail1_list);
225 spin_unlock(&sdp->sd_ail_lock);
226
227 return ret;
228}
229
230static void gfs2_ail1_wait(struct gfs2_sbd *sdp)
231{
232 struct gfs2_ail *ai;
233 struct gfs2_bufdata *bd;
234 struct buffer_head *bh;
235
236 spin_lock(&sdp->sd_ail_lock);
237 list_for_each_entry_reverse(ai, &sdp->sd_ail1_list, ai_list) {
238 list_for_each_entry(bd, &ai->ai_ail1_list, bd_ail_st_list) {
239 bh = bd->bd_bh;
240 if (!buffer_locked(bh))
241 continue;
242 get_bh(bh);
243 spin_unlock(&sdp->sd_ail_lock);
244 wait_on_buffer(bh);
245 brelse(bh);
246 return;
247 }
248 }
249 spin_unlock(&sdp->sd_ail_lock);
250}
251
252/**
253 * gfs2_ail2_empty_one - Check whether or not a trans in the AIL has been synced
254 * @sdp: the filesystem
255 * @ai: the AIL entry
256 *
257 */
258
259static void gfs2_ail2_empty_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
260{
261 struct list_head *head = &ai->ai_ail2_list;
262 struct gfs2_bufdata *bd;
263
264 while (!list_empty(head)) {
265 bd = list_entry(head->prev, struct gfs2_bufdata,
266 bd_ail_st_list);
267 gfs2_assert(sdp, bd->bd_ail == ai);
268 gfs2_remove_from_ail(bd);
269 }
270}
271
272static void ail2_empty(struct gfs2_sbd *sdp, unsigned int new_tail)
273{
274 struct gfs2_ail *ai, *safe;
275 unsigned int old_tail = sdp->sd_log_tail;
276 int wrap = (new_tail < old_tail);
277 int a, b, rm;
278
279 spin_lock(&sdp->sd_ail_lock);
280
281 list_for_each_entry_safe(ai, safe, &sdp->sd_ail2_list, ai_list) {
282 a = (old_tail <= ai->ai_first);
283 b = (ai->ai_first < new_tail);
284 rm = (wrap) ? (a || b) : (a && b);
285 if (!rm)
286 continue;
287
288 gfs2_ail2_empty_one(sdp, ai);
289 list_del(&ai->ai_list);
290 gfs2_assert_warn(sdp, list_empty(&ai->ai_ail1_list));
291 gfs2_assert_warn(sdp, list_empty(&ai->ai_ail2_list));
292 kfree(ai);
293 }
294
295 spin_unlock(&sdp->sd_ail_lock);
296}
297
298/**
299 * gfs2_log_reserve - Make a log reservation
300 * @sdp: The GFS2 superblock
301 * @blks: The number of blocks to reserve
302 *
303 * Note that we never give out the last few blocks of the journal. Thats
304 * due to the fact that there is a small number of header blocks
305 * associated with each log flush. The exact number can't be known until
306 * flush time, so we ensure that we have just enough free blocks at all
307 * times to avoid running out during a log flush.
308 *
309 * We no longer flush the log here, instead we wake up logd to do that
310 * for us. To avoid the thundering herd and to ensure that we deal fairly
311 * with queued waiters, we use an exclusive wait. This means that when we
312 * get woken with enough journal space to get our reservation, we need to
313 * wake the next waiter on the list.
314 *
315 * Returns: errno
316 */
317
318int gfs2_log_reserve(struct gfs2_sbd *sdp, unsigned int blks)
319{
320 unsigned reserved_blks = 6 * (4096 / sdp->sd_vfs->s_blocksize);
321 unsigned wanted = blks + reserved_blks;
322 DEFINE_WAIT(wait);
323 int did_wait = 0;
324 unsigned int free_blocks;
325
326 if (gfs2_assert_warn(sdp, blks) ||
327 gfs2_assert_warn(sdp, blks <= sdp->sd_jdesc->jd_blocks))
328 return -EINVAL;
329retry:
330 free_blocks = atomic_read(&sdp->sd_log_blks_free);
331 if (unlikely(free_blocks <= wanted)) {
332 do {
333 prepare_to_wait_exclusive(&sdp->sd_log_waitq, &wait,
334 TASK_UNINTERRUPTIBLE);
335 wake_up(&sdp->sd_logd_waitq);
336 did_wait = 1;
337 if (atomic_read(&sdp->sd_log_blks_free) <= wanted)
338 io_schedule();
339 free_blocks = atomic_read(&sdp->sd_log_blks_free);
340 } while(free_blocks <= wanted);
341 finish_wait(&sdp->sd_log_waitq, &wait);
342 }
343 if (atomic_cmpxchg(&sdp->sd_log_blks_free, free_blocks,
344 free_blocks - blks) != free_blocks)
345 goto retry;
346 trace_gfs2_log_blocks(sdp, -blks);
347
348 /*
349 * If we waited, then so might others, wake them up _after_ we get
350 * our share of the log.
351 */
352 if (unlikely(did_wait))
353 wake_up(&sdp->sd_log_waitq);
354
355 down_read(&sdp->sd_log_flush_lock);
356
357 return 0;
358}
359
360/**
361 * log_distance - Compute distance between two journal blocks
362 * @sdp: The GFS2 superblock
363 * @newer: The most recent journal block of the pair
364 * @older: The older journal block of the pair
365 *
366 * Compute the distance (in the journal direction) between two
367 * blocks in the journal
368 *
369 * Returns: the distance in blocks
370 */
371
372static inline unsigned int log_distance(struct gfs2_sbd *sdp, unsigned int newer,
373 unsigned int older)
374{
375 int dist;
376
377 dist = newer - older;
378 if (dist < 0)
379 dist += sdp->sd_jdesc->jd_blocks;
380
381 return dist;
382}
383
384/**
385 * calc_reserved - Calculate the number of blocks to reserve when
386 * refunding a transaction's unused buffers.
387 * @sdp: The GFS2 superblock
388 *
389 * This is complex. We need to reserve room for all our currently used
390 * metadata buffers (e.g. normal file I/O rewriting file time stamps) and
391 * all our journaled data buffers for journaled files (e.g. files in the
392 * meta_fs like rindex, or files for which chattr +j was done.)
393 * If we don't reserve enough space, gfs2_log_refund and gfs2_log_flush
394 * will count it as free space (sd_log_blks_free) and corruption will follow.
395 *
396 * We can have metadata bufs and jdata bufs in the same journal. So each
397 * type gets its own log header, for which we need to reserve a block.
398 * In fact, each type has the potential for needing more than one header
399 * in cases where we have more buffers than will fit on a journal page.
400 * Metadata journal entries take up half the space of journaled buffer entries.
401 * Thus, metadata entries have buf_limit (502) and journaled buffers have
402 * databuf_limit (251) before they cause a wrap around.
403 *
404 * Also, we need to reserve blocks for revoke journal entries and one for an
405 * overall header for the lot.
406 *
407 * Returns: the number of blocks reserved
408 */
409static unsigned int calc_reserved(struct gfs2_sbd *sdp)
410{
411 unsigned int reserved = 0;
412 unsigned int mbuf_limit, metabufhdrs_needed;
413 unsigned int dbuf_limit, databufhdrs_needed;
414 unsigned int revokes = 0;
415
416 mbuf_limit = buf_limit(sdp);
417 metabufhdrs_needed = (sdp->sd_log_commited_buf +
418 (mbuf_limit - 1)) / mbuf_limit;
419 dbuf_limit = databuf_limit(sdp);
420 databufhdrs_needed = (sdp->sd_log_commited_databuf +
421 (dbuf_limit - 1)) / dbuf_limit;
422
423 if (sdp->sd_log_commited_revoke > 0)
424 revokes = gfs2_struct2blk(sdp, sdp->sd_log_commited_revoke,
425 sizeof(u64));
426
427 reserved = sdp->sd_log_commited_buf + metabufhdrs_needed +
428 sdp->sd_log_commited_databuf + databufhdrs_needed +
429 revokes;
430 /* One for the overall header */
431 if (reserved)
432 reserved++;
433 return reserved;
434}
435
436static unsigned int current_tail(struct gfs2_sbd *sdp)
437{
438 struct gfs2_ail *ai;
439 unsigned int tail;
440
441 spin_lock(&sdp->sd_ail_lock);
442
443 if (list_empty(&sdp->sd_ail1_list)) {
444 tail = sdp->sd_log_head;
445 } else {
446 ai = list_entry(sdp->sd_ail1_list.prev, struct gfs2_ail, ai_list);
447 tail = ai->ai_first;
448 }
449
450 spin_unlock(&sdp->sd_ail_lock);
451
452 return tail;
453}
454
455static void log_pull_tail(struct gfs2_sbd *sdp, unsigned int new_tail)
456{
457 unsigned int dist = log_distance(sdp, new_tail, sdp->sd_log_tail);
458
459 ail2_empty(sdp, new_tail);
460
461 atomic_add(dist, &sdp->sd_log_blks_free);
462 trace_gfs2_log_blocks(sdp, dist);
463 gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
464 sdp->sd_jdesc->jd_blocks);
465
466 sdp->sd_log_tail = new_tail;
467}
468
469
470static void log_flush_wait(struct gfs2_sbd *sdp)
471{
472 DEFINE_WAIT(wait);
473
474 if (atomic_read(&sdp->sd_log_in_flight)) {
475 do {
476 prepare_to_wait(&sdp->sd_log_flush_wait, &wait,
477 TASK_UNINTERRUPTIBLE);
478 if (atomic_read(&sdp->sd_log_in_flight))
479 io_schedule();
480 } while(atomic_read(&sdp->sd_log_in_flight));
481 finish_wait(&sdp->sd_log_flush_wait, &wait);
482 }
483}
484
485static int bd_cmp(void *priv, struct list_head *a, struct list_head *b)
486{
487 struct gfs2_bufdata *bda, *bdb;
488
489 bda = list_entry(a, struct gfs2_bufdata, bd_list);
490 bdb = list_entry(b, struct gfs2_bufdata, bd_list);
491
492 if (bda->bd_bh->b_blocknr < bdb->bd_bh->b_blocknr)
493 return -1;
494 if (bda->bd_bh->b_blocknr > bdb->bd_bh->b_blocknr)
495 return 1;
496 return 0;
497}
498
499static void gfs2_ordered_write(struct gfs2_sbd *sdp)
500{
501 struct gfs2_bufdata *bd;
502 struct buffer_head *bh;
503 LIST_HEAD(written);
504
505 gfs2_log_lock(sdp);
506 list_sort(NULL, &sdp->sd_log_le_ordered, &bd_cmp);
507 while (!list_empty(&sdp->sd_log_le_ordered)) {
508 bd = list_entry(sdp->sd_log_le_ordered.next, struct gfs2_bufdata, bd_list);
509 list_move(&bd->bd_list, &written);
510 bh = bd->bd_bh;
511 if (!buffer_dirty(bh))
512 continue;
513 get_bh(bh);
514 gfs2_log_unlock(sdp);
515 lock_buffer(bh);
516 if (buffer_mapped(bh) && test_clear_buffer_dirty(bh)) {
517 bh->b_end_io = end_buffer_write_sync;
518 submit_bh(WRITE_SYNC, bh);
519 } else {
520 unlock_buffer(bh);
521 brelse(bh);
522 }
523 gfs2_log_lock(sdp);
524 }
525 list_splice(&written, &sdp->sd_log_le_ordered);
526 gfs2_log_unlock(sdp);
527}
528
529static void gfs2_ordered_wait(struct gfs2_sbd *sdp)
530{
531 struct gfs2_bufdata *bd;
532 struct buffer_head *bh;
533
534 gfs2_log_lock(sdp);
535 while (!list_empty(&sdp->sd_log_le_ordered)) {
536 bd = list_entry(sdp->sd_log_le_ordered.prev, struct gfs2_bufdata, bd_list);
537 bh = bd->bd_bh;
538 if (buffer_locked(bh)) {
539 get_bh(bh);
540 gfs2_log_unlock(sdp);
541 wait_on_buffer(bh);
542 brelse(bh);
543 gfs2_log_lock(sdp);
544 continue;
545 }
546 list_del_init(&bd->bd_list);
547 }
548 gfs2_log_unlock(sdp);
549}
550
551/**
552 * log_write_header - Get and initialize a journal header buffer
553 * @sdp: The GFS2 superblock
554 *
555 * Returns: the initialized log buffer descriptor
556 */
557
558static void log_write_header(struct gfs2_sbd *sdp, u32 flags)
559{
560 struct gfs2_log_header *lh;
561 unsigned int tail;
562 u32 hash;
563 int rw = WRITE_FLUSH_FUA | REQ_META;
564 struct page *page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
565 lh = page_address(page);
566 clear_page(lh);
567
568 gfs2_ail1_empty(sdp);
569 tail = current_tail(sdp);
570
571 lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
572 lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH);
573 lh->lh_header.__pad0 = cpu_to_be64(0);
574 lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH);
575 lh->lh_header.mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid);
576 lh->lh_sequence = cpu_to_be64(sdp->sd_log_sequence++);
577 lh->lh_flags = cpu_to_be32(flags);
578 lh->lh_tail = cpu_to_be32(tail);
579 lh->lh_blkno = cpu_to_be32(sdp->sd_log_flush_head);
580 hash = gfs2_disk_hash(page_address(page), sizeof(struct gfs2_log_header));
581 lh->lh_hash = cpu_to_be32(hash);
582
583 if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags)) {
584 gfs2_ordered_wait(sdp);
585 log_flush_wait(sdp);
586 rw = WRITE_SYNC | REQ_META | REQ_PRIO;
587 }
588
589 sdp->sd_log_idle = (tail == sdp->sd_log_flush_head);
590 gfs2_log_write_page(sdp, page);
591 gfs2_log_flush_bio(sdp, rw);
592 log_flush_wait(sdp);
593
594 if (sdp->sd_log_tail != tail)
595 log_pull_tail(sdp, tail);
596}
597
598/**
599 * gfs2_log_flush - flush incore transaction(s)
600 * @sdp: the filesystem
601 * @gl: The glock structure to flush. If NULL, flush the whole incore log
602 *
603 */
604
605void gfs2_log_flush(struct gfs2_sbd *sdp, struct gfs2_glock *gl)
606{
607 struct gfs2_ail *ai;
608
609 down_write(&sdp->sd_log_flush_lock);
610
611 /* Log might have been flushed while we waited for the flush lock */
612 if (gl && !test_bit(GLF_LFLUSH, &gl->gl_flags)) {
613 up_write(&sdp->sd_log_flush_lock);
614 return;
615 }
616 trace_gfs2_log_flush(sdp, 1);
617
618 ai = kzalloc(sizeof(struct gfs2_ail), GFP_NOFS | __GFP_NOFAIL);
619 INIT_LIST_HEAD(&ai->ai_ail1_list);
620 INIT_LIST_HEAD(&ai->ai_ail2_list);
621
622 if (sdp->sd_log_num_buf != sdp->sd_log_commited_buf) {
623 printk(KERN_INFO "GFS2: log buf %u %u\n", sdp->sd_log_num_buf,
624 sdp->sd_log_commited_buf);
625 gfs2_assert_withdraw(sdp, 0);
626 }
627 if (sdp->sd_log_num_databuf != sdp->sd_log_commited_databuf) {
628 printk(KERN_INFO "GFS2: log databuf %u %u\n",
629 sdp->sd_log_num_databuf, sdp->sd_log_commited_databuf);
630 gfs2_assert_withdraw(sdp, 0);
631 }
632 gfs2_assert_withdraw(sdp,
633 sdp->sd_log_num_revoke == sdp->sd_log_commited_revoke);
634
635 sdp->sd_log_flush_head = sdp->sd_log_head;
636 sdp->sd_log_flush_wrapped = 0;
637 ai->ai_first = sdp->sd_log_flush_head;
638
639 gfs2_ordered_write(sdp);
640 lops_before_commit(sdp);
641 gfs2_log_flush_bio(sdp, WRITE);
642
643 if (sdp->sd_log_head != sdp->sd_log_flush_head) {
644 log_write_header(sdp, 0);
645 } else if (sdp->sd_log_tail != current_tail(sdp) && !sdp->sd_log_idle){
646 atomic_dec(&sdp->sd_log_blks_free); /* Adjust for unreserved buffer */
647 trace_gfs2_log_blocks(sdp, -1);
648 log_write_header(sdp, 0);
649 }
650 lops_after_commit(sdp, ai);
651
652 gfs2_log_lock(sdp);
653 sdp->sd_log_head = sdp->sd_log_flush_head;
654 sdp->sd_log_blks_reserved = 0;
655 sdp->sd_log_commited_buf = 0;
656 sdp->sd_log_commited_databuf = 0;
657 sdp->sd_log_commited_revoke = 0;
658
659 spin_lock(&sdp->sd_ail_lock);
660 if (!list_empty(&ai->ai_ail1_list)) {
661 list_add(&ai->ai_list, &sdp->sd_ail1_list);
662 ai = NULL;
663 }
664 spin_unlock(&sdp->sd_ail_lock);
665 gfs2_log_unlock(sdp);
666 trace_gfs2_log_flush(sdp, 0);
667 up_write(&sdp->sd_log_flush_lock);
668
669 kfree(ai);
670}
671
672static void log_refund(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
673{
674 unsigned int reserved;
675 unsigned int unused;
676
677 gfs2_log_lock(sdp);
678
679 sdp->sd_log_commited_buf += tr->tr_num_buf_new - tr->tr_num_buf_rm;
680 sdp->sd_log_commited_databuf += tr->tr_num_databuf_new -
681 tr->tr_num_databuf_rm;
682 gfs2_assert_withdraw(sdp, (((int)sdp->sd_log_commited_buf) >= 0) ||
683 (((int)sdp->sd_log_commited_databuf) >= 0));
684 sdp->sd_log_commited_revoke += tr->tr_num_revoke - tr->tr_num_revoke_rm;
685 reserved = calc_reserved(sdp);
686 gfs2_assert_withdraw(sdp, sdp->sd_log_blks_reserved + tr->tr_reserved >= reserved);
687 unused = sdp->sd_log_blks_reserved - reserved + tr->tr_reserved;
688 atomic_add(unused, &sdp->sd_log_blks_free);
689 trace_gfs2_log_blocks(sdp, unused);
690 gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
691 sdp->sd_jdesc->jd_blocks);
692 sdp->sd_log_blks_reserved = reserved;
693
694 gfs2_log_unlock(sdp);
695}
696
697/**
698 * gfs2_log_commit - Commit a transaction to the log
699 * @sdp: the filesystem
700 * @tr: the transaction
701 *
702 * We wake up gfs2_logd if the number of pinned blocks exceed thresh1
703 * or the total number of used blocks (pinned blocks plus AIL blocks)
704 * is greater than thresh2.
705 *
706 * At mount time thresh1 is 1/3rd of journal size, thresh2 is 2/3rd of
707 * journal size.
708 *
709 * Returns: errno
710 */
711
712void gfs2_log_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
713{
714 log_refund(sdp, tr);
715 up_read(&sdp->sd_log_flush_lock);
716
717 if (atomic_read(&sdp->sd_log_pinned) > atomic_read(&sdp->sd_log_thresh1) ||
718 ((sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free)) >
719 atomic_read(&sdp->sd_log_thresh2)))
720 wake_up(&sdp->sd_logd_waitq);
721}
722
723/**
724 * gfs2_log_shutdown - write a shutdown header into a journal
725 * @sdp: the filesystem
726 *
727 */
728
729void gfs2_log_shutdown(struct gfs2_sbd *sdp)
730{
731 down_write(&sdp->sd_log_flush_lock);
732
733 gfs2_assert_withdraw(sdp, !sdp->sd_log_blks_reserved);
734 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_buf);
735 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
736 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_rg);
737 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_databuf);
738 gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail1_list));
739
740 sdp->sd_log_flush_head = sdp->sd_log_head;
741 sdp->sd_log_flush_wrapped = 0;
742
743 log_write_header(sdp, GFS2_LOG_HEAD_UNMOUNT);
744
745 gfs2_assert_warn(sdp, atomic_read(&sdp->sd_log_blks_free) == sdp->sd_jdesc->jd_blocks);
746 gfs2_assert_warn(sdp, sdp->sd_log_head == sdp->sd_log_tail);
747 gfs2_assert_warn(sdp, list_empty(&sdp->sd_ail2_list));
748
749 sdp->sd_log_head = sdp->sd_log_flush_head;
750 sdp->sd_log_tail = sdp->sd_log_head;
751
752 up_write(&sdp->sd_log_flush_lock);
753}
754
755
756/**
757 * gfs2_meta_syncfs - sync all the buffers in a filesystem
758 * @sdp: the filesystem
759 *
760 */
761
762void gfs2_meta_syncfs(struct gfs2_sbd *sdp)
763{
764 gfs2_log_flush(sdp, NULL);
765 for (;;) {
766 gfs2_ail1_start(sdp);
767 gfs2_ail1_wait(sdp);
768 if (gfs2_ail1_empty(sdp))
769 break;
770 }
771 gfs2_log_flush(sdp, NULL);
772}
773
774static inline int gfs2_jrnl_flush_reqd(struct gfs2_sbd *sdp)
775{
776 return (atomic_read(&sdp->sd_log_pinned) >= atomic_read(&sdp->sd_log_thresh1));
777}
778
779static inline int gfs2_ail_flush_reqd(struct gfs2_sbd *sdp)
780{
781 unsigned int used_blocks = sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free);
782 return used_blocks >= atomic_read(&sdp->sd_log_thresh2);
783}
784
785/**
786 * gfs2_logd - Update log tail as Active Items get flushed to in-place blocks
787 * @sdp: Pointer to GFS2 superblock
788 *
789 * Also, periodically check to make sure that we're using the most recent
790 * journal index.
791 */
792
793int gfs2_logd(void *data)
794{
795 struct gfs2_sbd *sdp = data;
796 unsigned long t = 1;
797 DEFINE_WAIT(wait);
798
799 while (!kthread_should_stop()) {
800
801 if (gfs2_jrnl_flush_reqd(sdp) || t == 0) {
802 gfs2_ail1_empty(sdp);
803 gfs2_log_flush(sdp, NULL);
804 }
805
806 if (gfs2_ail_flush_reqd(sdp)) {
807 gfs2_ail1_start(sdp);
808 gfs2_ail1_wait(sdp);
809 gfs2_ail1_empty(sdp);
810 gfs2_log_flush(sdp, NULL);
811 }
812
813 if (!gfs2_ail_flush_reqd(sdp))
814 wake_up(&sdp->sd_log_waitq);
815
816 t = gfs2_tune_get(sdp, gt_logd_secs) * HZ;
817
818 try_to_freeze();
819
820 do {
821 prepare_to_wait(&sdp->sd_logd_waitq, &wait,
822 TASK_INTERRUPTIBLE);
823 if (!gfs2_ail_flush_reqd(sdp) &&
824 !gfs2_jrnl_flush_reqd(sdp) &&
825 !kthread_should_stop())
826 t = schedule_timeout(t);
827 } while(t && !gfs2_ail_flush_reqd(sdp) &&
828 !gfs2_jrnl_flush_reqd(sdp) &&
829 !kthread_should_stop());
830 finish_wait(&sdp->sd_logd_waitq, &wait);
831 }
832
833 return 0;
834}
835
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/crc32c.h>
18#include <linux/delay.h>
19#include <linux/kthread.h>
20#include <linux/freezer.h>
21#include <linux/bio.h>
22#include <linux/blkdev.h>
23#include <linux/writeback.h>
24#include <linux/list_sort.h>
25
26#include "gfs2.h"
27#include "incore.h"
28#include "bmap.h"
29#include "glock.h"
30#include "log.h"
31#include "lops.h"
32#include "meta_io.h"
33#include "util.h"
34#include "dir.h"
35#include "trace_gfs2.h"
36
37/**
38 * gfs2_struct2blk - compute stuff
39 * @sdp: the filesystem
40 * @nstruct: the number of structures
41 * @ssize: the size of the structures
42 *
43 * Compute the number of log descriptor blocks needed to hold a certain number
44 * of structures of a certain size.
45 *
46 * Returns: the number of blocks needed (minimum is always 1)
47 */
48
49unsigned int gfs2_struct2blk(struct gfs2_sbd *sdp, unsigned int nstruct,
50 unsigned int ssize)
51{
52 unsigned int blks;
53 unsigned int first, second;
54
55 blks = 1;
56 first = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / ssize;
57
58 if (nstruct > first) {
59 second = (sdp->sd_sb.sb_bsize -
60 sizeof(struct gfs2_meta_header)) / ssize;
61 blks += DIV_ROUND_UP(nstruct - first, second);
62 }
63
64 return blks;
65}
66
67/**
68 * gfs2_remove_from_ail - Remove an entry from the ail lists, updating counters
69 * @mapping: The associated mapping (maybe NULL)
70 * @bd: The gfs2_bufdata to remove
71 *
72 * The ail lock _must_ be held when calling this function
73 *
74 */
75
76static void gfs2_remove_from_ail(struct gfs2_bufdata *bd)
77{
78 bd->bd_tr = NULL;
79 list_del_init(&bd->bd_ail_st_list);
80 list_del_init(&bd->bd_ail_gl_list);
81 atomic_dec(&bd->bd_gl->gl_ail_count);
82 brelse(bd->bd_bh);
83}
84
85/**
86 * gfs2_ail1_start_one - Start I/O on a part of the AIL
87 * @sdp: the filesystem
88 * @wbc: The writeback control structure
89 * @ai: The ail structure
90 *
91 */
92
93static int gfs2_ail1_start_one(struct gfs2_sbd *sdp,
94 struct writeback_control *wbc,
95 struct gfs2_trans *tr)
96__releases(&sdp->sd_ail_lock)
97__acquires(&sdp->sd_ail_lock)
98{
99 struct gfs2_glock *gl = NULL;
100 struct address_space *mapping;
101 struct gfs2_bufdata *bd, *s;
102 struct buffer_head *bh;
103
104 list_for_each_entry_safe_reverse(bd, s, &tr->tr_ail1_list, bd_ail_st_list) {
105 bh = bd->bd_bh;
106
107 gfs2_assert(sdp, bd->bd_tr == tr);
108
109 if (!buffer_busy(bh)) {
110 if (!buffer_uptodate(bh))
111 gfs2_io_error_bh(sdp, bh);
112 list_move(&bd->bd_ail_st_list, &tr->tr_ail2_list);
113 continue;
114 }
115
116 if (!buffer_dirty(bh))
117 continue;
118 if (gl == bd->bd_gl)
119 continue;
120 gl = bd->bd_gl;
121 list_move(&bd->bd_ail_st_list, &tr->tr_ail1_list);
122 mapping = bh->b_page->mapping;
123 if (!mapping)
124 continue;
125 spin_unlock(&sdp->sd_ail_lock);
126 generic_writepages(mapping, wbc);
127 spin_lock(&sdp->sd_ail_lock);
128 if (wbc->nr_to_write <= 0)
129 break;
130 return 1;
131 }
132
133 return 0;
134}
135
136
137/**
138 * gfs2_ail1_flush - start writeback of some ail1 entries
139 * @sdp: The super block
140 * @wbc: The writeback control structure
141 *
142 * Writes back some ail1 entries, according to the limits in the
143 * writeback control structure
144 */
145
146void gfs2_ail1_flush(struct gfs2_sbd *sdp, struct writeback_control *wbc)
147{
148 struct list_head *head = &sdp->sd_ail1_list;
149 struct gfs2_trans *tr;
150 struct blk_plug plug;
151
152 trace_gfs2_ail_flush(sdp, wbc, 1);
153 blk_start_plug(&plug);
154 spin_lock(&sdp->sd_ail_lock);
155restart:
156 list_for_each_entry_reverse(tr, head, tr_list) {
157 if (wbc->nr_to_write <= 0)
158 break;
159 if (gfs2_ail1_start_one(sdp, wbc, tr))
160 goto restart;
161 }
162 spin_unlock(&sdp->sd_ail_lock);
163 blk_finish_plug(&plug);
164 trace_gfs2_ail_flush(sdp, wbc, 0);
165}
166
167/**
168 * gfs2_ail1_start - start writeback of all ail1 entries
169 * @sdp: The superblock
170 */
171
172static void gfs2_ail1_start(struct gfs2_sbd *sdp)
173{
174 struct writeback_control wbc = {
175 .sync_mode = WB_SYNC_NONE,
176 .nr_to_write = LONG_MAX,
177 .range_start = 0,
178 .range_end = LLONG_MAX,
179 };
180
181 return gfs2_ail1_flush(sdp, &wbc);
182}
183
184/**
185 * gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced
186 * @sdp: the filesystem
187 * @ai: the AIL entry
188 *
189 */
190
191static void gfs2_ail1_empty_one(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
192{
193 struct gfs2_bufdata *bd, *s;
194 struct buffer_head *bh;
195
196 list_for_each_entry_safe_reverse(bd, s, &tr->tr_ail1_list,
197 bd_ail_st_list) {
198 bh = bd->bd_bh;
199 gfs2_assert(sdp, bd->bd_tr == tr);
200 if (buffer_busy(bh))
201 continue;
202 if (!buffer_uptodate(bh))
203 gfs2_io_error_bh(sdp, bh);
204 list_move(&bd->bd_ail_st_list, &tr->tr_ail2_list);
205 }
206
207}
208
209/**
210 * gfs2_ail1_empty - Try to empty the ail1 lists
211 * @sdp: The superblock
212 *
213 * Tries to empty the ail1 lists, starting with the oldest first
214 */
215
216static int gfs2_ail1_empty(struct gfs2_sbd *sdp)
217{
218 struct gfs2_trans *tr, *s;
219 int oldest_tr = 1;
220 int ret;
221
222 spin_lock(&sdp->sd_ail_lock);
223 list_for_each_entry_safe_reverse(tr, s, &sdp->sd_ail1_list, tr_list) {
224 gfs2_ail1_empty_one(sdp, tr);
225 if (list_empty(&tr->tr_ail1_list) && oldest_tr)
226 list_move(&tr->tr_list, &sdp->sd_ail2_list);
227 else
228 oldest_tr = 0;
229 }
230 ret = list_empty(&sdp->sd_ail1_list);
231 spin_unlock(&sdp->sd_ail_lock);
232
233 return ret;
234}
235
236static void gfs2_ail1_wait(struct gfs2_sbd *sdp)
237{
238 struct gfs2_trans *tr;
239 struct gfs2_bufdata *bd;
240 struct buffer_head *bh;
241
242 spin_lock(&sdp->sd_ail_lock);
243 list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) {
244 list_for_each_entry(bd, &tr->tr_ail1_list, bd_ail_st_list) {
245 bh = bd->bd_bh;
246 if (!buffer_locked(bh))
247 continue;
248 get_bh(bh);
249 spin_unlock(&sdp->sd_ail_lock);
250 wait_on_buffer(bh);
251 brelse(bh);
252 return;
253 }
254 }
255 spin_unlock(&sdp->sd_ail_lock);
256}
257
258/**
259 * gfs2_ail2_empty_one - Check whether or not a trans in the AIL has been synced
260 * @sdp: the filesystem
261 * @ai: the AIL entry
262 *
263 */
264
265static void gfs2_ail2_empty_one(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
266{
267 struct list_head *head = &tr->tr_ail2_list;
268 struct gfs2_bufdata *bd;
269
270 while (!list_empty(head)) {
271 bd = list_entry(head->prev, struct gfs2_bufdata,
272 bd_ail_st_list);
273 gfs2_assert(sdp, bd->bd_tr == tr);
274 gfs2_remove_from_ail(bd);
275 }
276}
277
278static void ail2_empty(struct gfs2_sbd *sdp, unsigned int new_tail)
279{
280 struct gfs2_trans *tr, *safe;
281 unsigned int old_tail = sdp->sd_log_tail;
282 int wrap = (new_tail < old_tail);
283 int a, b, rm;
284
285 spin_lock(&sdp->sd_ail_lock);
286
287 list_for_each_entry_safe(tr, safe, &sdp->sd_ail2_list, tr_list) {
288 a = (old_tail <= tr->tr_first);
289 b = (tr->tr_first < new_tail);
290 rm = (wrap) ? (a || b) : (a && b);
291 if (!rm)
292 continue;
293
294 gfs2_ail2_empty_one(sdp, tr);
295 list_del(&tr->tr_list);
296 gfs2_assert_warn(sdp, list_empty(&tr->tr_ail1_list));
297 gfs2_assert_warn(sdp, list_empty(&tr->tr_ail2_list));
298 kfree(tr);
299 }
300
301 spin_unlock(&sdp->sd_ail_lock);
302}
303
304/**
305 * gfs2_log_release - Release a given number of log blocks
306 * @sdp: The GFS2 superblock
307 * @blks: The number of blocks
308 *
309 */
310
311void gfs2_log_release(struct gfs2_sbd *sdp, unsigned int blks)
312{
313
314 atomic_add(blks, &sdp->sd_log_blks_free);
315 trace_gfs2_log_blocks(sdp, blks);
316 gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
317 sdp->sd_jdesc->jd_blocks);
318 up_read(&sdp->sd_log_flush_lock);
319}
320
321/**
322 * gfs2_log_reserve - Make a log reservation
323 * @sdp: The GFS2 superblock
324 * @blks: The number of blocks to reserve
325 *
326 * Note that we never give out the last few blocks of the journal. Thats
327 * due to the fact that there is a small number of header blocks
328 * associated with each log flush. The exact number can't be known until
329 * flush time, so we ensure that we have just enough free blocks at all
330 * times to avoid running out during a log flush.
331 *
332 * We no longer flush the log here, instead we wake up logd to do that
333 * for us. To avoid the thundering herd and to ensure that we deal fairly
334 * with queued waiters, we use an exclusive wait. This means that when we
335 * get woken with enough journal space to get our reservation, we need to
336 * wake the next waiter on the list.
337 *
338 * Returns: errno
339 */
340
341int gfs2_log_reserve(struct gfs2_sbd *sdp, unsigned int blks)
342{
343 int ret = 0;
344 unsigned reserved_blks = 7 * (4096 / sdp->sd_vfs->s_blocksize);
345 unsigned wanted = blks + reserved_blks;
346 DEFINE_WAIT(wait);
347 int did_wait = 0;
348 unsigned int free_blocks;
349
350 if (gfs2_assert_warn(sdp, blks) ||
351 gfs2_assert_warn(sdp, blks <= sdp->sd_jdesc->jd_blocks))
352 return -EINVAL;
353 atomic_add(blks, &sdp->sd_log_blks_needed);
354retry:
355 free_blocks = atomic_read(&sdp->sd_log_blks_free);
356 if (unlikely(free_blocks <= wanted)) {
357 do {
358 prepare_to_wait_exclusive(&sdp->sd_log_waitq, &wait,
359 TASK_UNINTERRUPTIBLE);
360 wake_up(&sdp->sd_logd_waitq);
361 did_wait = 1;
362 if (atomic_read(&sdp->sd_log_blks_free) <= wanted)
363 io_schedule();
364 free_blocks = atomic_read(&sdp->sd_log_blks_free);
365 } while(free_blocks <= wanted);
366 finish_wait(&sdp->sd_log_waitq, &wait);
367 }
368 atomic_inc(&sdp->sd_reserving_log);
369 if (atomic_cmpxchg(&sdp->sd_log_blks_free, free_blocks,
370 free_blocks - blks) != free_blocks) {
371 if (atomic_dec_and_test(&sdp->sd_reserving_log))
372 wake_up(&sdp->sd_reserving_log_wait);
373 goto retry;
374 }
375 atomic_sub(blks, &sdp->sd_log_blks_needed);
376 trace_gfs2_log_blocks(sdp, -blks);
377
378 /*
379 * If we waited, then so might others, wake them up _after_ we get
380 * our share of the log.
381 */
382 if (unlikely(did_wait))
383 wake_up(&sdp->sd_log_waitq);
384
385 down_read(&sdp->sd_log_flush_lock);
386 if (unlikely(!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))) {
387 gfs2_log_release(sdp, blks);
388 ret = -EROFS;
389 }
390 if (atomic_dec_and_test(&sdp->sd_reserving_log))
391 wake_up(&sdp->sd_reserving_log_wait);
392 return ret;
393}
394
395/**
396 * log_distance - Compute distance between two journal blocks
397 * @sdp: The GFS2 superblock
398 * @newer: The most recent journal block of the pair
399 * @older: The older journal block of the pair
400 *
401 * Compute the distance (in the journal direction) between two
402 * blocks in the journal
403 *
404 * Returns: the distance in blocks
405 */
406
407static inline unsigned int log_distance(struct gfs2_sbd *sdp, unsigned int newer,
408 unsigned int older)
409{
410 int dist;
411
412 dist = newer - older;
413 if (dist < 0)
414 dist += sdp->sd_jdesc->jd_blocks;
415
416 return dist;
417}
418
419/**
420 * calc_reserved - Calculate the number of blocks to reserve when
421 * refunding a transaction's unused buffers.
422 * @sdp: The GFS2 superblock
423 *
424 * This is complex. We need to reserve room for all our currently used
425 * metadata buffers (e.g. normal file I/O rewriting file time stamps) and
426 * all our journaled data buffers for journaled files (e.g. files in the
427 * meta_fs like rindex, or files for which chattr +j was done.)
428 * If we don't reserve enough space, gfs2_log_refund and gfs2_log_flush
429 * will count it as free space (sd_log_blks_free) and corruption will follow.
430 *
431 * We can have metadata bufs and jdata bufs in the same journal. So each
432 * type gets its own log header, for which we need to reserve a block.
433 * In fact, each type has the potential for needing more than one header
434 * in cases where we have more buffers than will fit on a journal page.
435 * Metadata journal entries take up half the space of journaled buffer entries.
436 * Thus, metadata entries have buf_limit (502) and journaled buffers have
437 * databuf_limit (251) before they cause a wrap around.
438 *
439 * Also, we need to reserve blocks for revoke journal entries and one for an
440 * overall header for the lot.
441 *
442 * Returns: the number of blocks reserved
443 */
444static unsigned int calc_reserved(struct gfs2_sbd *sdp)
445{
446 unsigned int reserved = 0;
447 unsigned int mbuf;
448 unsigned int dbuf;
449 struct gfs2_trans *tr = sdp->sd_log_tr;
450
451 if (tr) {
452 mbuf = tr->tr_num_buf_new - tr->tr_num_buf_rm;
453 dbuf = tr->tr_num_databuf_new - tr->tr_num_databuf_rm;
454 reserved = mbuf + dbuf;
455 /* Account for header blocks */
456 reserved += DIV_ROUND_UP(mbuf, buf_limit(sdp));
457 reserved += DIV_ROUND_UP(dbuf, databuf_limit(sdp));
458 }
459
460 if (sdp->sd_log_commited_revoke > 0)
461 reserved += gfs2_struct2blk(sdp, sdp->sd_log_commited_revoke,
462 sizeof(u64));
463 /* One for the overall header */
464 if (reserved)
465 reserved++;
466 return reserved;
467}
468
469static unsigned int current_tail(struct gfs2_sbd *sdp)
470{
471 struct gfs2_trans *tr;
472 unsigned int tail;
473
474 spin_lock(&sdp->sd_ail_lock);
475
476 if (list_empty(&sdp->sd_ail1_list)) {
477 tail = sdp->sd_log_head;
478 } else {
479 tr = list_entry(sdp->sd_ail1_list.prev, struct gfs2_trans,
480 tr_list);
481 tail = tr->tr_first;
482 }
483
484 spin_unlock(&sdp->sd_ail_lock);
485
486 return tail;
487}
488
489static void log_pull_tail(struct gfs2_sbd *sdp, unsigned int new_tail)
490{
491 unsigned int dist = log_distance(sdp, new_tail, sdp->sd_log_tail);
492
493 ail2_empty(sdp, new_tail);
494
495 atomic_add(dist, &sdp->sd_log_blks_free);
496 trace_gfs2_log_blocks(sdp, dist);
497 gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
498 sdp->sd_jdesc->jd_blocks);
499
500 sdp->sd_log_tail = new_tail;
501}
502
503
504static void log_flush_wait(struct gfs2_sbd *sdp)
505{
506 DEFINE_WAIT(wait);
507
508 if (atomic_read(&sdp->sd_log_in_flight)) {
509 do {
510 prepare_to_wait(&sdp->sd_log_flush_wait, &wait,
511 TASK_UNINTERRUPTIBLE);
512 if (atomic_read(&sdp->sd_log_in_flight))
513 io_schedule();
514 } while(atomic_read(&sdp->sd_log_in_flight));
515 finish_wait(&sdp->sd_log_flush_wait, &wait);
516 }
517}
518
519static int ip_cmp(void *priv, struct list_head *a, struct list_head *b)
520{
521 struct gfs2_inode *ipa, *ipb;
522
523 ipa = list_entry(a, struct gfs2_inode, i_ordered);
524 ipb = list_entry(b, struct gfs2_inode, i_ordered);
525
526 if (ipa->i_no_addr < ipb->i_no_addr)
527 return -1;
528 if (ipa->i_no_addr > ipb->i_no_addr)
529 return 1;
530 return 0;
531}
532
533static void gfs2_ordered_write(struct gfs2_sbd *sdp)
534{
535 struct gfs2_inode *ip;
536 LIST_HEAD(written);
537
538 spin_lock(&sdp->sd_ordered_lock);
539 list_sort(NULL, &sdp->sd_log_le_ordered, &ip_cmp);
540 while (!list_empty(&sdp->sd_log_le_ordered)) {
541 ip = list_entry(sdp->sd_log_le_ordered.next, struct gfs2_inode, i_ordered);
542 if (ip->i_inode.i_mapping->nrpages == 0) {
543 test_and_clear_bit(GIF_ORDERED, &ip->i_flags);
544 list_del(&ip->i_ordered);
545 continue;
546 }
547 list_move(&ip->i_ordered, &written);
548 spin_unlock(&sdp->sd_ordered_lock);
549 filemap_fdatawrite(ip->i_inode.i_mapping);
550 spin_lock(&sdp->sd_ordered_lock);
551 }
552 list_splice(&written, &sdp->sd_log_le_ordered);
553 spin_unlock(&sdp->sd_ordered_lock);
554}
555
556static void gfs2_ordered_wait(struct gfs2_sbd *sdp)
557{
558 struct gfs2_inode *ip;
559
560 spin_lock(&sdp->sd_ordered_lock);
561 while (!list_empty(&sdp->sd_log_le_ordered)) {
562 ip = list_entry(sdp->sd_log_le_ordered.next, struct gfs2_inode, i_ordered);
563 list_del(&ip->i_ordered);
564 WARN_ON(!test_and_clear_bit(GIF_ORDERED, &ip->i_flags));
565 if (ip->i_inode.i_mapping->nrpages == 0)
566 continue;
567 spin_unlock(&sdp->sd_ordered_lock);
568 filemap_fdatawait(ip->i_inode.i_mapping);
569 spin_lock(&sdp->sd_ordered_lock);
570 }
571 spin_unlock(&sdp->sd_ordered_lock);
572}
573
574void gfs2_ordered_del_inode(struct gfs2_inode *ip)
575{
576 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
577
578 spin_lock(&sdp->sd_ordered_lock);
579 if (test_and_clear_bit(GIF_ORDERED, &ip->i_flags))
580 list_del(&ip->i_ordered);
581 spin_unlock(&sdp->sd_ordered_lock);
582}
583
584void gfs2_add_revoke(struct gfs2_sbd *sdp, struct gfs2_bufdata *bd)
585{
586 struct buffer_head *bh = bd->bd_bh;
587 struct gfs2_glock *gl = bd->bd_gl;
588
589 bh->b_private = NULL;
590 bd->bd_blkno = bh->b_blocknr;
591 gfs2_remove_from_ail(bd); /* drops ref on bh */
592 bd->bd_bh = NULL;
593 bd->bd_ops = &gfs2_revoke_lops;
594 sdp->sd_log_num_revoke++;
595 atomic_inc(&gl->gl_revokes);
596 set_bit(GLF_LFLUSH, &gl->gl_flags);
597 list_add(&bd->bd_list, &sdp->sd_log_le_revoke);
598}
599
600void gfs2_write_revokes(struct gfs2_sbd *sdp)
601{
602 struct gfs2_trans *tr;
603 struct gfs2_bufdata *bd, *tmp;
604 int have_revokes = 0;
605 int max_revokes = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / sizeof(u64);
606
607 gfs2_ail1_empty(sdp);
608 spin_lock(&sdp->sd_ail_lock);
609 list_for_each_entry(tr, &sdp->sd_ail1_list, tr_list) {
610 list_for_each_entry(bd, &tr->tr_ail2_list, bd_ail_st_list) {
611 if (list_empty(&bd->bd_list)) {
612 have_revokes = 1;
613 goto done;
614 }
615 }
616 }
617done:
618 spin_unlock(&sdp->sd_ail_lock);
619 if (have_revokes == 0)
620 return;
621 while (sdp->sd_log_num_revoke > max_revokes)
622 max_revokes += (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header)) / sizeof(u64);
623 max_revokes -= sdp->sd_log_num_revoke;
624 if (!sdp->sd_log_num_revoke) {
625 atomic_dec(&sdp->sd_log_blks_free);
626 /* If no blocks have been reserved, we need to also
627 * reserve a block for the header */
628 if (!sdp->sd_log_blks_reserved)
629 atomic_dec(&sdp->sd_log_blks_free);
630 }
631 gfs2_log_lock(sdp);
632 spin_lock(&sdp->sd_ail_lock);
633 list_for_each_entry(tr, &sdp->sd_ail1_list, tr_list) {
634 list_for_each_entry_safe(bd, tmp, &tr->tr_ail2_list, bd_ail_st_list) {
635 if (max_revokes == 0)
636 goto out_of_blocks;
637 if (!list_empty(&bd->bd_list))
638 continue;
639 gfs2_add_revoke(sdp, bd);
640 max_revokes--;
641 }
642 }
643out_of_blocks:
644 spin_unlock(&sdp->sd_ail_lock);
645 gfs2_log_unlock(sdp);
646
647 if (!sdp->sd_log_num_revoke) {
648 atomic_inc(&sdp->sd_log_blks_free);
649 if (!sdp->sd_log_blks_reserved)
650 atomic_inc(&sdp->sd_log_blks_free);
651 }
652}
653
654/**
655 * write_log_header - Write a journal log header buffer at sd_log_flush_head
656 * @sdp: The GFS2 superblock
657 * @jd: journal descriptor of the journal to which we are writing
658 * @seq: sequence number
659 * @tail: tail of the log
660 * @flags: log header flags GFS2_LOG_HEAD_*
661 * @op_flags: flags to pass to the bio
662 *
663 * Returns: the initialized log buffer descriptor
664 */
665
666void gfs2_write_log_header(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd,
667 u64 seq, u32 tail, u32 flags, int op_flags)
668{
669 struct gfs2_log_header *lh;
670 u32 hash, crc;
671 struct page *page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
672 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
673 struct timespec64 tv;
674 struct super_block *sb = sdp->sd_vfs;
675 u64 addr;
676
677 lh = page_address(page);
678 clear_page(lh);
679
680 lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
681 lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH);
682 lh->lh_header.__pad0 = cpu_to_be64(0);
683 lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH);
684 lh->lh_header.mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid);
685 lh->lh_sequence = cpu_to_be64(seq);
686 lh->lh_flags = cpu_to_be32(flags);
687 lh->lh_tail = cpu_to_be32(tail);
688 lh->lh_blkno = cpu_to_be32(sdp->sd_log_flush_head);
689 hash = ~crc32(~0, lh, LH_V1_SIZE);
690 lh->lh_hash = cpu_to_be32(hash);
691
692 tv = current_kernel_time64();
693 lh->lh_nsec = cpu_to_be32(tv.tv_nsec);
694 lh->lh_sec = cpu_to_be64(tv.tv_sec);
695 addr = gfs2_log_bmap(sdp);
696 lh->lh_addr = cpu_to_be64(addr);
697 lh->lh_jinode = cpu_to_be64(GFS2_I(jd->jd_inode)->i_no_addr);
698
699 /* We may only write local statfs, quota, etc., when writing to our
700 own journal. The values are left 0 when recovering a journal
701 different from our own. */
702 if (!(flags & GFS2_LOG_HEAD_RECOVERY)) {
703 lh->lh_statfs_addr =
704 cpu_to_be64(GFS2_I(sdp->sd_sc_inode)->i_no_addr);
705 lh->lh_quota_addr =
706 cpu_to_be64(GFS2_I(sdp->sd_qc_inode)->i_no_addr);
707
708 spin_lock(&sdp->sd_statfs_spin);
709 lh->lh_local_total = cpu_to_be64(l_sc->sc_total);
710 lh->lh_local_free = cpu_to_be64(l_sc->sc_free);
711 lh->lh_local_dinodes = cpu_to_be64(l_sc->sc_dinodes);
712 spin_unlock(&sdp->sd_statfs_spin);
713 }
714
715 BUILD_BUG_ON(offsetof(struct gfs2_log_header, lh_crc) != LH_V1_SIZE);
716
717 crc = crc32c(~0, (void *)lh + LH_V1_SIZE + 4,
718 sb->s_blocksize - LH_V1_SIZE - 4);
719 lh->lh_crc = cpu_to_be32(crc);
720
721 gfs2_log_write(sdp, page, sb->s_blocksize, 0, addr);
722 gfs2_log_flush_bio(sdp, REQ_OP_WRITE, op_flags);
723 log_flush_wait(sdp);
724}
725
726/**
727 * log_write_header - Get and initialize a journal header buffer
728 * @sdp: The GFS2 superblock
729 * @flags: The log header flags, including log header origin
730 *
731 * Returns: the initialized log buffer descriptor
732 */
733
734static void log_write_header(struct gfs2_sbd *sdp, u32 flags)
735{
736 unsigned int tail;
737 int op_flags = REQ_PREFLUSH | REQ_FUA | REQ_META | REQ_SYNC;
738 enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state);
739
740 gfs2_assert_withdraw(sdp, (state != SFS_FROZEN));
741 tail = current_tail(sdp);
742
743 if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags)) {
744 gfs2_ordered_wait(sdp);
745 log_flush_wait(sdp);
746 op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
747 }
748 sdp->sd_log_idle = (tail == sdp->sd_log_flush_head);
749 gfs2_write_log_header(sdp, sdp->sd_jdesc, sdp->sd_log_sequence++, tail,
750 flags, op_flags);
751
752 if (sdp->sd_log_tail != tail)
753 log_pull_tail(sdp, tail);
754}
755
756/**
757 * gfs2_log_flush - flush incore transaction(s)
758 * @sdp: the filesystem
759 * @gl: The glock structure to flush. If NULL, flush the whole incore log
760 * @flags: The log header flags: GFS2_LOG_HEAD_FLUSH_* and debug flags
761 *
762 */
763
764void gfs2_log_flush(struct gfs2_sbd *sdp, struct gfs2_glock *gl, u32 flags)
765{
766 struct gfs2_trans *tr;
767 enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state);
768
769 down_write(&sdp->sd_log_flush_lock);
770
771 /* Log might have been flushed while we waited for the flush lock */
772 if (gl && !test_bit(GLF_LFLUSH, &gl->gl_flags)) {
773 up_write(&sdp->sd_log_flush_lock);
774 return;
775 }
776 trace_gfs2_log_flush(sdp, 1, flags);
777
778 if (flags & GFS2_LOG_HEAD_FLUSH_SHUTDOWN)
779 clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
780
781 sdp->sd_log_flush_head = sdp->sd_log_head;
782 tr = sdp->sd_log_tr;
783 if (tr) {
784 sdp->sd_log_tr = NULL;
785 INIT_LIST_HEAD(&tr->tr_ail1_list);
786 INIT_LIST_HEAD(&tr->tr_ail2_list);
787 tr->tr_first = sdp->sd_log_flush_head;
788 if (unlikely (state == SFS_FROZEN))
789 gfs2_assert_withdraw(sdp, !tr->tr_num_buf_new && !tr->tr_num_databuf_new);
790 }
791
792 if (unlikely(state == SFS_FROZEN))
793 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
794 gfs2_assert_withdraw(sdp,
795 sdp->sd_log_num_revoke == sdp->sd_log_commited_revoke);
796
797 gfs2_ordered_write(sdp);
798 lops_before_commit(sdp, tr);
799 gfs2_log_flush_bio(sdp, REQ_OP_WRITE, 0);
800
801 if (sdp->sd_log_head != sdp->sd_log_flush_head) {
802 log_flush_wait(sdp);
803 log_write_header(sdp, flags);
804 } else if (sdp->sd_log_tail != current_tail(sdp) && !sdp->sd_log_idle){
805 atomic_dec(&sdp->sd_log_blks_free); /* Adjust for unreserved buffer */
806 trace_gfs2_log_blocks(sdp, -1);
807 log_write_header(sdp, flags);
808 }
809 lops_after_commit(sdp, tr);
810
811 gfs2_log_lock(sdp);
812 sdp->sd_log_head = sdp->sd_log_flush_head;
813 sdp->sd_log_blks_reserved = 0;
814 sdp->sd_log_commited_revoke = 0;
815
816 spin_lock(&sdp->sd_ail_lock);
817 if (tr && !list_empty(&tr->tr_ail1_list)) {
818 list_add(&tr->tr_list, &sdp->sd_ail1_list);
819 tr = NULL;
820 }
821 spin_unlock(&sdp->sd_ail_lock);
822 gfs2_log_unlock(sdp);
823
824 if (!(flags & GFS2_LOG_HEAD_FLUSH_NORMAL)) {
825 if (!sdp->sd_log_idle) {
826 for (;;) {
827 gfs2_ail1_start(sdp);
828 gfs2_ail1_wait(sdp);
829 if (gfs2_ail1_empty(sdp))
830 break;
831 }
832 atomic_dec(&sdp->sd_log_blks_free); /* Adjust for unreserved buffer */
833 trace_gfs2_log_blocks(sdp, -1);
834 log_write_header(sdp, flags);
835 sdp->sd_log_head = sdp->sd_log_flush_head;
836 }
837 if (flags & (GFS2_LOG_HEAD_FLUSH_SHUTDOWN |
838 GFS2_LOG_HEAD_FLUSH_FREEZE))
839 gfs2_log_shutdown(sdp);
840 if (flags & GFS2_LOG_HEAD_FLUSH_FREEZE)
841 atomic_set(&sdp->sd_freeze_state, SFS_FROZEN);
842 }
843
844 trace_gfs2_log_flush(sdp, 0, flags);
845 up_write(&sdp->sd_log_flush_lock);
846
847 kfree(tr);
848}
849
850/**
851 * gfs2_merge_trans - Merge a new transaction into a cached transaction
852 * @old: Original transaction to be expanded
853 * @new: New transaction to be merged
854 */
855
856static void gfs2_merge_trans(struct gfs2_trans *old, struct gfs2_trans *new)
857{
858 WARN_ON_ONCE(!test_bit(TR_ATTACHED, &old->tr_flags));
859
860 old->tr_num_buf_new += new->tr_num_buf_new;
861 old->tr_num_databuf_new += new->tr_num_databuf_new;
862 old->tr_num_buf_rm += new->tr_num_buf_rm;
863 old->tr_num_databuf_rm += new->tr_num_databuf_rm;
864 old->tr_num_revoke += new->tr_num_revoke;
865 old->tr_num_revoke_rm += new->tr_num_revoke_rm;
866
867 list_splice_tail_init(&new->tr_databuf, &old->tr_databuf);
868 list_splice_tail_init(&new->tr_buf, &old->tr_buf);
869}
870
871static void log_refund(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
872{
873 unsigned int reserved;
874 unsigned int unused;
875 unsigned int maxres;
876
877 gfs2_log_lock(sdp);
878
879 if (sdp->sd_log_tr) {
880 gfs2_merge_trans(sdp->sd_log_tr, tr);
881 } else if (tr->tr_num_buf_new || tr->tr_num_databuf_new) {
882 gfs2_assert_withdraw(sdp, test_bit(TR_ALLOCED, &tr->tr_flags));
883 sdp->sd_log_tr = tr;
884 set_bit(TR_ATTACHED, &tr->tr_flags);
885 }
886
887 sdp->sd_log_commited_revoke += tr->tr_num_revoke - tr->tr_num_revoke_rm;
888 reserved = calc_reserved(sdp);
889 maxres = sdp->sd_log_blks_reserved + tr->tr_reserved;
890 gfs2_assert_withdraw(sdp, maxres >= reserved);
891 unused = maxres - reserved;
892 atomic_add(unused, &sdp->sd_log_blks_free);
893 trace_gfs2_log_blocks(sdp, unused);
894 gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
895 sdp->sd_jdesc->jd_blocks);
896 sdp->sd_log_blks_reserved = reserved;
897
898 gfs2_log_unlock(sdp);
899}
900
901/**
902 * gfs2_log_commit - Commit a transaction to the log
903 * @sdp: the filesystem
904 * @tr: the transaction
905 *
906 * We wake up gfs2_logd if the number of pinned blocks exceed thresh1
907 * or the total number of used blocks (pinned blocks plus AIL blocks)
908 * is greater than thresh2.
909 *
910 * At mount time thresh1 is 1/3rd of journal size, thresh2 is 2/3rd of
911 * journal size.
912 *
913 * Returns: errno
914 */
915
916void gfs2_log_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
917{
918 log_refund(sdp, tr);
919
920 if (atomic_read(&sdp->sd_log_pinned) > atomic_read(&sdp->sd_log_thresh1) ||
921 ((sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free)) >
922 atomic_read(&sdp->sd_log_thresh2)))
923 wake_up(&sdp->sd_logd_waitq);
924}
925
926/**
927 * gfs2_log_shutdown - write a shutdown header into a journal
928 * @sdp: the filesystem
929 *
930 */
931
932void gfs2_log_shutdown(struct gfs2_sbd *sdp)
933{
934 gfs2_assert_withdraw(sdp, !sdp->sd_log_blks_reserved);
935 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
936 gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail1_list));
937
938 sdp->sd_log_flush_head = sdp->sd_log_head;
939
940 log_write_header(sdp, GFS2_LOG_HEAD_UNMOUNT | GFS2_LFC_SHUTDOWN);
941
942 gfs2_assert_warn(sdp, sdp->sd_log_head == sdp->sd_log_tail);
943 gfs2_assert_warn(sdp, list_empty(&sdp->sd_ail2_list));
944
945 sdp->sd_log_head = sdp->sd_log_flush_head;
946 sdp->sd_log_tail = sdp->sd_log_head;
947}
948
949static inline int gfs2_jrnl_flush_reqd(struct gfs2_sbd *sdp)
950{
951 return (atomic_read(&sdp->sd_log_pinned) +
952 atomic_read(&sdp->sd_log_blks_needed) >=
953 atomic_read(&sdp->sd_log_thresh1));
954}
955
956static inline int gfs2_ail_flush_reqd(struct gfs2_sbd *sdp)
957{
958 unsigned int used_blocks = sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free);
959
960 if (test_and_clear_bit(SDF_FORCE_AIL_FLUSH, &sdp->sd_flags))
961 return 1;
962
963 return used_blocks + atomic_read(&sdp->sd_log_blks_needed) >=
964 atomic_read(&sdp->sd_log_thresh2);
965}
966
967/**
968 * gfs2_logd - Update log tail as Active Items get flushed to in-place blocks
969 * @sdp: Pointer to GFS2 superblock
970 *
971 * Also, periodically check to make sure that we're using the most recent
972 * journal index.
973 */
974
975int gfs2_logd(void *data)
976{
977 struct gfs2_sbd *sdp = data;
978 unsigned long t = 1;
979 DEFINE_WAIT(wait);
980 bool did_flush;
981
982 while (!kthread_should_stop()) {
983
984 /* Check for errors writing to the journal */
985 if (sdp->sd_log_error) {
986 gfs2_lm_withdraw(sdp,
987 "GFS2: fsid=%s: error %d: "
988 "withdrawing the file system to "
989 "prevent further damage.\n",
990 sdp->sd_fsname, sdp->sd_log_error);
991 }
992
993 did_flush = false;
994 if (gfs2_jrnl_flush_reqd(sdp) || t == 0) {
995 gfs2_ail1_empty(sdp);
996 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
997 GFS2_LFC_LOGD_JFLUSH_REQD);
998 did_flush = true;
999 }
1000
1001 if (gfs2_ail_flush_reqd(sdp)) {
1002 gfs2_ail1_start(sdp);
1003 gfs2_ail1_wait(sdp);
1004 gfs2_ail1_empty(sdp);
1005 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
1006 GFS2_LFC_LOGD_AIL_FLUSH_REQD);
1007 did_flush = true;
1008 }
1009
1010 if (!gfs2_ail_flush_reqd(sdp) || did_flush)
1011 wake_up(&sdp->sd_log_waitq);
1012
1013 t = gfs2_tune_get(sdp, gt_logd_secs) * HZ;
1014
1015 try_to_freeze();
1016
1017 do {
1018 prepare_to_wait(&sdp->sd_logd_waitq, &wait,
1019 TASK_INTERRUPTIBLE);
1020 if (!gfs2_ail_flush_reqd(sdp) &&
1021 !gfs2_jrnl_flush_reqd(sdp) &&
1022 !kthread_should_stop())
1023 t = schedule_timeout(t);
1024 } while(t && !gfs2_ail_flush_reqd(sdp) &&
1025 !gfs2_jrnl_flush_reqd(sdp) &&
1026 !kthread_should_stop());
1027 finish_wait(&sdp->sd_logd_waitq, &wait);
1028 }
1029
1030 return 0;
1031}
1032