1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
4 * Copyright (C) 2004-2007 Red Hat, Inc. All rights reserved.
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>
32 #include "trace_gfs2.h"
34 static void gfs2_log_shutdown(struct gfs2_sbd *sdp);
37 * gfs2_struct2blk - compute stuff
38 * @sdp: the filesystem
39 * @nstruct: the number of structures
41 * Compute the number of log descriptor blocks needed to hold a certain number
42 * of structures of a certain size.
44 * Returns: the number of blocks needed (minimum is always 1)
47 unsigned int gfs2_struct2blk(struct gfs2_sbd *sdp, unsigned int nstruct)
50 unsigned int first, second;
53 first = sdp->sd_ldptrs;
55 if (nstruct > first) {
56 second = sdp->sd_inptrs;
57 blks += DIV_ROUND_UP(nstruct - first, second);
64 * gfs2_remove_from_ail - Remove an entry from the ail lists, updating counters
65 * @mapping: The associated mapping (maybe NULL)
66 * @bd: The gfs2_bufdata to remove
68 * The ail lock _must_ be held when calling this function
72 static void gfs2_remove_from_ail(struct gfs2_bufdata *bd)
75 list_del_init(&bd->bd_ail_st_list);
76 list_del_init(&bd->bd_ail_gl_list);
77 atomic_dec(&bd->bd_gl->gl_ail_count);
82 * gfs2_ail1_start_one - Start I/O on a part of the AIL
83 * @sdp: the filesystem
84 * @wbc: The writeback control structure
85 * @ai: The ail structure
89 static int gfs2_ail1_start_one(struct gfs2_sbd *sdp,
90 struct writeback_control *wbc,
91 struct gfs2_trans *tr)
92 __releases(&sdp->sd_ail_lock)
93 __acquires(&sdp->sd_ail_lock)
95 struct gfs2_glock *gl = NULL;
96 struct address_space *mapping;
97 struct gfs2_bufdata *bd, *s;
98 struct buffer_head *bh;
101 list_for_each_entry_safe_reverse(bd, s, &tr->tr_ail1_list, bd_ail_st_list) {
104 gfs2_assert(sdp, bd->bd_tr == tr);
106 if (!buffer_busy(bh)) {
107 if (buffer_uptodate(bh)) {
108 list_move(&bd->bd_ail_st_list,
112 if (!cmpxchg(&sdp->sd_log_error, 0, -EIO)) {
113 gfs2_io_error_bh(sdp, bh);
114 gfs2_withdraw_delayed(sdp);
118 if (gfs2_withdrawn(sdp)) {
119 gfs2_remove_from_ail(bd);
122 if (!buffer_dirty(bh))
127 list_move(&bd->bd_ail_st_list, &tr->tr_ail1_list);
128 mapping = bh->b_page->mapping;
131 spin_unlock(&sdp->sd_ail_lock);
132 ret = generic_writepages(mapping, wbc);
133 spin_lock(&sdp->sd_ail_lock);
134 if (ret || wbc->nr_to_write <= 0)
142 static void dump_ail_list(struct gfs2_sbd *sdp)
144 struct gfs2_trans *tr;
145 struct gfs2_bufdata *bd;
146 struct buffer_head *bh;
148 fs_err(sdp, "Error: In gfs2_ail1_flush for ten minutes! t=%d\n",
149 current->journal_info ? 1 : 0);
151 list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) {
152 list_for_each_entry_reverse(bd, &tr->tr_ail1_list,
155 fs_err(sdp, "bd %p: blk:0x%llx bh=%p ", bd,
156 (unsigned long long)bd->bd_blkno, bh);
161 fs_err(sdp, "0x%llx up2:%d dirt:%d lkd:%d req:%d "
162 "map:%d new:%d ar:%d aw:%d delay:%d "
163 "io err:%d unwritten:%d dfr:%d pin:%d esc:%d\n",
164 (unsigned long long)bh->b_blocknr,
165 buffer_uptodate(bh), buffer_dirty(bh),
166 buffer_locked(bh), buffer_req(bh),
167 buffer_mapped(bh), buffer_new(bh),
168 buffer_async_read(bh), buffer_async_write(bh),
169 buffer_delay(bh), buffer_write_io_error(bh),
170 buffer_unwritten(bh),
171 buffer_defer_completion(bh),
172 buffer_pinned(bh), buffer_escaped(bh));
178 * gfs2_ail1_flush - start writeback of some ail1 entries
179 * @sdp: The super block
180 * @wbc: The writeback control structure
182 * Writes back some ail1 entries, according to the limits in the
183 * writeback control structure
186 void gfs2_ail1_flush(struct gfs2_sbd *sdp, struct writeback_control *wbc)
188 struct list_head *head = &sdp->sd_ail1_list;
189 struct gfs2_trans *tr;
190 struct blk_plug plug;
192 unsigned long flush_start = jiffies;
194 trace_gfs2_ail_flush(sdp, wbc, 1);
195 blk_start_plug(&plug);
196 spin_lock(&sdp->sd_ail_lock);
199 if (time_after(jiffies, flush_start + (HZ * 600))) {
203 list_for_each_entry_reverse(tr, head, tr_list) {
204 if (wbc->nr_to_write <= 0)
206 ret = gfs2_ail1_start_one(sdp, wbc, tr);
214 spin_unlock(&sdp->sd_ail_lock);
215 blk_finish_plug(&plug);
217 gfs2_lm(sdp, "gfs2_ail1_start_one (generic_writepages) "
218 "returned: %d\n", ret);
221 trace_gfs2_ail_flush(sdp, wbc, 0);
225 * gfs2_ail1_start - start writeback of all ail1 entries
226 * @sdp: The superblock
229 static void gfs2_ail1_start(struct gfs2_sbd *sdp)
231 struct writeback_control wbc = {
232 .sync_mode = WB_SYNC_NONE,
233 .nr_to_write = LONG_MAX,
235 .range_end = LLONG_MAX,
238 return gfs2_ail1_flush(sdp, &wbc);
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
249 static void gfs2_ail1_empty_one(struct gfs2_sbd *sdp, struct gfs2_trans *tr,
252 struct gfs2_bufdata *bd, *s;
253 struct buffer_head *bh;
255 list_for_each_entry_safe_reverse(bd, s, &tr->tr_ail1_list,
258 gfs2_assert(sdp, bd->bd_tr == tr);
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
268 if (!sdp->sd_log_error && buffer_busy(bh))
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);
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.
281 if (*max_revokes && list_empty(&bd->bd_list)) {
282 gfs2_add_revoke(sdp, bd);
286 list_move(&bd->bd_ail_st_list, &tr->tr_ail2_list);
291 * gfs2_ail1_empty - Try to empty the ail1 lists
292 * @sdp: The superblock
293 * @max_revokes: If non-zero, add revokes where appropriate
295 * Tries to empty the ail1 lists, starting with the oldest first
298 static int gfs2_ail1_empty(struct gfs2_sbd *sdp, int max_revokes)
300 struct gfs2_trans *tr, *s;
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);
312 ret = list_empty(&sdp->sd_ail1_list);
313 spin_unlock(&sdp->sd_ail_lock);
315 if (test_bit(SDF_WITHDRAWING, &sdp->sd_flags)) {
316 gfs2_lm(sdp, "fatal: I/O error(s)\n");
323 static void gfs2_ail1_wait(struct gfs2_sbd *sdp)
325 struct gfs2_trans *tr;
326 struct gfs2_bufdata *bd;
327 struct buffer_head *bh;
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) {
333 if (!buffer_locked(bh))
336 spin_unlock(&sdp->sd_ail_lock);
342 spin_unlock(&sdp->sd_ail_lock);
346 * gfs2_ail_empty_tr - empty one of the ail lists for a transaction
349 static void gfs2_ail_empty_tr(struct gfs2_sbd *sdp, struct gfs2_trans *tr,
350 struct list_head *head)
352 struct gfs2_bufdata *bd;
354 while (!list_empty(head)) {
355 bd = list_first_entry(head, struct gfs2_bufdata,
357 gfs2_assert(sdp, bd->bd_tr == tr);
358 gfs2_remove_from_ail(bd);
362 static void ail2_empty(struct gfs2_sbd *sdp, unsigned int new_tail)
364 struct gfs2_trans *tr, *safe;
365 unsigned int old_tail = sdp->sd_log_tail;
366 int wrap = (new_tail < old_tail);
369 spin_lock(&sdp->sd_ail_lock);
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);
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));
385 spin_unlock(&sdp->sd_ail_lock);
389 * gfs2_log_release - Release a given number of log blocks
390 * @sdp: The GFS2 superblock
391 * @blks: The number of blocks
395 void gfs2_log_release(struct gfs2_sbd *sdp, unsigned int blks)
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);
406 * gfs2_log_reserve - Make a log reservation
407 * @sdp: The GFS2 superblock
408 * @blks: The number of blocks to reserve
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.
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.
425 int gfs2_log_reserve(struct gfs2_sbd *sdp, unsigned int blks)
428 unsigned reserved_blks = 7 * (4096 / sdp->sd_vfs->s_blocksize);
429 unsigned wanted = blks + reserved_blks;
432 unsigned int free_blocks;
434 if (gfs2_assert_warn(sdp, blks) ||
435 gfs2_assert_warn(sdp, blks <= sdp->sd_jdesc->jd_blocks))
437 atomic_add(blks, &sdp->sd_log_blks_needed);
439 free_blocks = atomic_read(&sdp->sd_log_blks_free);
440 if (unlikely(free_blocks <= wanted)) {
442 prepare_to_wait_exclusive(&sdp->sd_log_waitq, &wait,
443 TASK_UNINTERRUPTIBLE);
444 wake_up(&sdp->sd_logd_waitq);
446 if (atomic_read(&sdp->sd_log_blks_free) <= wanted)
448 free_blocks = atomic_read(&sdp->sd_log_blks_free);
449 } while(free_blocks <= wanted);
450 finish_wait(&sdp->sd_log_waitq, &wait);
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);
459 atomic_sub(blks, &sdp->sd_log_blks_needed);
460 trace_gfs2_log_blocks(sdp, -blks);
463 * If we waited, then so might others, wake them up _after_ we get
464 * our share of the log.
466 if (unlikely(did_wait))
467 wake_up(&sdp->sd_log_waitq);
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);
474 if (atomic_dec_and_test(&sdp->sd_reserving_log))
475 wake_up(&sdp->sd_reserving_log_wait);
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
485 * Compute the distance (in the journal direction) between two
486 * blocks in the journal
488 * Returns: the distance in blocks
491 static inline unsigned int log_distance(struct gfs2_sbd *sdp, unsigned int newer,
496 dist = newer - older;
498 dist += sdp->sd_jdesc->jd_blocks;
504 * calc_reserved - Calculate the number of blocks to reserve when
505 * refunding a transaction's unused buffers.
506 * @sdp: The GFS2 superblock
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.
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.
523 * Also, we need to reserve blocks for revoke journal entries and one for an
524 * overall header for the lot.
526 * Returns: the number of blocks reserved
528 static unsigned int calc_reserved(struct gfs2_sbd *sdp)
530 unsigned int reserved = 0;
533 struct gfs2_trans *tr = sdp->sd_log_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));
544 if (sdp->sd_log_committed_revoke > 0)
545 reserved += gfs2_struct2blk(sdp, sdp->sd_log_committed_revoke);
546 /* One for the overall header */
552 static unsigned int current_tail(struct gfs2_sbd *sdp)
554 struct gfs2_trans *tr;
557 spin_lock(&sdp->sd_ail_lock);
559 if (list_empty(&sdp->sd_ail1_list)) {
560 tail = sdp->sd_log_head;
562 tr = list_last_entry(&sdp->sd_ail1_list, struct gfs2_trans,
567 spin_unlock(&sdp->sd_ail_lock);
572 static void log_pull_tail(struct gfs2_sbd *sdp, unsigned int new_tail)
574 unsigned int dist = log_distance(sdp, new_tail, sdp->sd_log_tail);
576 ail2_empty(sdp, new_tail);
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);
583 sdp->sd_log_tail = new_tail;
587 void log_flush_wait(struct gfs2_sbd *sdp)
591 if (atomic_read(&sdp->sd_log_in_flight)) {
593 prepare_to_wait(&sdp->sd_log_flush_wait, &wait,
594 TASK_UNINTERRUPTIBLE);
595 if (atomic_read(&sdp->sd_log_in_flight))
597 } while(atomic_read(&sdp->sd_log_in_flight));
598 finish_wait(&sdp->sd_log_flush_wait, &wait);
602 static int ip_cmp(void *priv, struct list_head *a, struct list_head *b)
604 struct gfs2_inode *ipa, *ipb;
606 ipa = list_entry(a, struct gfs2_inode, i_ordered);
607 ipb = list_entry(b, struct gfs2_inode, i_ordered);
609 if (ipa->i_no_addr < ipb->i_no_addr)
611 if (ipa->i_no_addr > ipb->i_no_addr)
616 static void gfs2_ordered_write(struct gfs2_sbd *sdp)
618 struct gfs2_inode *ip;
621 spin_lock(&sdp->sd_ordered_lock);
622 list_sort(NULL, &sdp->sd_log_ordered, &ip_cmp);
623 while (!list_empty(&sdp->sd_log_ordered)) {
624 ip = list_first_entry(&sdp->sd_log_ordered, struct gfs2_inode, i_ordered);
625 if (ip->i_inode.i_mapping->nrpages == 0) {
626 test_and_clear_bit(GIF_ORDERED, &ip->i_flags);
627 list_del(&ip->i_ordered);
630 list_move(&ip->i_ordered, &written);
631 spin_unlock(&sdp->sd_ordered_lock);
632 filemap_fdatawrite(ip->i_inode.i_mapping);
633 spin_lock(&sdp->sd_ordered_lock);
635 list_splice(&written, &sdp->sd_log_ordered);
636 spin_unlock(&sdp->sd_ordered_lock);
639 static void gfs2_ordered_wait(struct gfs2_sbd *sdp)
641 struct gfs2_inode *ip;
643 spin_lock(&sdp->sd_ordered_lock);
644 while (!list_empty(&sdp->sd_log_ordered)) {
645 ip = list_first_entry(&sdp->sd_log_ordered, struct gfs2_inode, i_ordered);
646 list_del(&ip->i_ordered);
647 WARN_ON(!test_and_clear_bit(GIF_ORDERED, &ip->i_flags));
648 if (ip->i_inode.i_mapping->nrpages == 0)
650 spin_unlock(&sdp->sd_ordered_lock);
651 filemap_fdatawait(ip->i_inode.i_mapping);
652 spin_lock(&sdp->sd_ordered_lock);
654 spin_unlock(&sdp->sd_ordered_lock);
657 void gfs2_ordered_del_inode(struct gfs2_inode *ip)
659 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
661 spin_lock(&sdp->sd_ordered_lock);
662 if (test_and_clear_bit(GIF_ORDERED, &ip->i_flags))
663 list_del(&ip->i_ordered);
664 spin_unlock(&sdp->sd_ordered_lock);
667 void gfs2_add_revoke(struct gfs2_sbd *sdp, struct gfs2_bufdata *bd)
669 struct buffer_head *bh = bd->bd_bh;
670 struct gfs2_glock *gl = bd->bd_gl;
672 sdp->sd_log_num_revoke++;
673 if (atomic_inc_return(&gl->gl_revokes) == 1)
675 bh->b_private = NULL;
676 bd->bd_blkno = bh->b_blocknr;
677 gfs2_remove_from_ail(bd); /* drops ref on bh */
679 set_bit(GLF_LFLUSH, &gl->gl_flags);
680 list_add(&bd->bd_list, &sdp->sd_log_revokes);
683 void gfs2_glock_remove_revoke(struct gfs2_glock *gl)
685 if (atomic_dec_return(&gl->gl_revokes) == 0) {
686 clear_bit(GLF_LFLUSH, &gl->gl_flags);
687 gfs2_glock_queue_put(gl);
692 * gfs2_write_revokes - Add as many revokes to the system transaction as we can
693 * @sdp: The GFS2 superblock
695 * Our usual strategy is to defer writing revokes as much as we can in the hope
696 * that we'll eventually overwrite the journal, which will make those revokes
697 * go away. This changes when we flush the log: at that point, there will
698 * likely be some left-over space in the last revoke block of that transaction.
699 * We can fill that space with additional revokes for blocks that have already
700 * been written back. This will basically come at no cost now, and will save
701 * us from having to keep track of those blocks on the AIL2 list later.
703 void gfs2_write_revokes(struct gfs2_sbd *sdp)
705 /* number of revokes we still have room for */
706 int max_revokes = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / sizeof(u64);
709 while (sdp->sd_log_num_revoke > max_revokes)
710 max_revokes += (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header)) / sizeof(u64);
711 max_revokes -= sdp->sd_log_num_revoke;
712 if (!sdp->sd_log_num_revoke) {
713 atomic_dec(&sdp->sd_log_blks_free);
714 /* If no blocks have been reserved, we need to also
715 * reserve a block for the header */
716 if (!sdp->sd_log_blks_reserved)
717 atomic_dec(&sdp->sd_log_blks_free);
719 gfs2_ail1_empty(sdp, max_revokes);
720 gfs2_log_unlock(sdp);
722 if (!sdp->sd_log_num_revoke) {
723 atomic_inc(&sdp->sd_log_blks_free);
724 if (!sdp->sd_log_blks_reserved)
725 atomic_inc(&sdp->sd_log_blks_free);
730 * gfs2_write_log_header - Write a journal log header buffer at lblock
731 * @sdp: The GFS2 superblock
732 * @jd: journal descriptor of the journal to which we are writing
733 * @seq: sequence number
734 * @tail: tail of the log
735 * @lblock: value for lh_blkno (block number relative to start of journal)
736 * @flags: log header flags GFS2_LOG_HEAD_*
737 * @op_flags: flags to pass to the bio
739 * Returns: the initialized log buffer descriptor
742 void gfs2_write_log_header(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd,
743 u64 seq, u32 tail, u32 lblock, u32 flags,
746 struct gfs2_log_header *lh;
749 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
750 struct timespec64 tv;
751 struct super_block *sb = sdp->sd_vfs;
754 if (gfs2_withdrawn(sdp))
757 page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
758 lh = page_address(page);
761 lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
762 lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH);
763 lh->lh_header.__pad0 = cpu_to_be64(0);
764 lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH);
765 lh->lh_header.mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid);
766 lh->lh_sequence = cpu_to_be64(seq);
767 lh->lh_flags = cpu_to_be32(flags);
768 lh->lh_tail = cpu_to_be32(tail);
769 lh->lh_blkno = cpu_to_be32(lblock);
770 hash = ~crc32(~0, lh, LH_V1_SIZE);
771 lh->lh_hash = cpu_to_be32(hash);
773 ktime_get_coarse_real_ts64(&tv);
774 lh->lh_nsec = cpu_to_be32(tv.tv_nsec);
775 lh->lh_sec = cpu_to_be64(tv.tv_sec);
776 if (!list_empty(&jd->extent_list))
777 dblock = gfs2_log_bmap(jd, lblock);
779 int ret = gfs2_lblk_to_dblk(jd->jd_inode, lblock, &dblock);
780 if (gfs2_assert_withdraw(sdp, ret == 0))
783 lh->lh_addr = cpu_to_be64(dblock);
784 lh->lh_jinode = cpu_to_be64(GFS2_I(jd->jd_inode)->i_no_addr);
786 /* We may only write local statfs, quota, etc., when writing to our
787 own journal. The values are left 0 when recovering a journal
788 different from our own. */
789 if (!(flags & GFS2_LOG_HEAD_RECOVERY)) {
791 cpu_to_be64(GFS2_I(sdp->sd_sc_inode)->i_no_addr);
793 cpu_to_be64(GFS2_I(sdp->sd_qc_inode)->i_no_addr);
795 spin_lock(&sdp->sd_statfs_spin);
796 lh->lh_local_total = cpu_to_be64(l_sc->sc_total);
797 lh->lh_local_free = cpu_to_be64(l_sc->sc_free);
798 lh->lh_local_dinodes = cpu_to_be64(l_sc->sc_dinodes);
799 spin_unlock(&sdp->sd_statfs_spin);
802 BUILD_BUG_ON(offsetof(struct gfs2_log_header, lh_crc) != LH_V1_SIZE);
804 crc = crc32c(~0, (void *)lh + LH_V1_SIZE + 4,
805 sb->s_blocksize - LH_V1_SIZE - 4);
806 lh->lh_crc = cpu_to_be32(crc);
808 gfs2_log_write(sdp, page, sb->s_blocksize, 0, dblock);
809 gfs2_log_submit_bio(&sdp->sd_log_bio, REQ_OP_WRITE | op_flags);
815 * log_write_header - Get and initialize a journal header buffer
816 * @sdp: The GFS2 superblock
817 * @flags: The log header flags, including log header origin
819 * Returns: the initialized log buffer descriptor
822 static void log_write_header(struct gfs2_sbd *sdp, u32 flags)
825 int op_flags = REQ_PREFLUSH | REQ_FUA | REQ_META | REQ_SYNC;
826 enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state);
828 gfs2_assert_withdraw(sdp, (state != SFS_FROZEN));
829 tail = current_tail(sdp);
831 if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags)) {
832 gfs2_ordered_wait(sdp);
834 op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
836 sdp->sd_log_idle = (tail == sdp->sd_log_flush_head);
837 gfs2_write_log_header(sdp, sdp->sd_jdesc, sdp->sd_log_sequence++, tail,
838 sdp->sd_log_flush_head, flags, op_flags);
839 gfs2_log_incr_head(sdp);
841 if (sdp->sd_log_tail != tail)
842 log_pull_tail(sdp, tail);
846 * ail_drain - drain the ail lists after a withdraw
847 * @sdp: Pointer to GFS2 superblock
849 static void ail_drain(struct gfs2_sbd *sdp)
851 struct gfs2_trans *tr;
853 spin_lock(&sdp->sd_ail_lock);
855 * For transactions on the sd_ail1_list we need to drain both the
856 * ail1 and ail2 lists. That's because function gfs2_ail1_start_one
857 * (temporarily) moves items from its tr_ail1 list to tr_ail2 list
858 * before revokes are sent for that block. Items on the sd_ail2_list
859 * should have already gotten beyond that point, so no need.
861 while (!list_empty(&sdp->sd_ail1_list)) {
862 tr = list_first_entry(&sdp->sd_ail1_list, struct gfs2_trans,
864 gfs2_ail_empty_tr(sdp, tr, &tr->tr_ail1_list);
865 gfs2_ail_empty_tr(sdp, tr, &tr->tr_ail2_list);
866 list_del(&tr->tr_list);
869 while (!list_empty(&sdp->sd_ail2_list)) {
870 tr = list_first_entry(&sdp->sd_ail2_list, struct gfs2_trans,
872 gfs2_ail_empty_tr(sdp, tr, &tr->tr_ail2_list);
873 list_del(&tr->tr_list);
876 spin_unlock(&sdp->sd_ail_lock);
880 * gfs2_log_flush - flush incore transaction(s)
881 * @sdp: the filesystem
882 * @gl: The glock structure to flush. If NULL, flush the whole incore log
883 * @flags: The log header flags: GFS2_LOG_HEAD_FLUSH_* and debug flags
887 void gfs2_log_flush(struct gfs2_sbd *sdp, struct gfs2_glock *gl, u32 flags)
889 struct gfs2_trans *tr = NULL;
890 enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state);
892 down_write(&sdp->sd_log_flush_lock);
895 * Do this check while holding the log_flush_lock to prevent new
896 * buffers from being added to the ail via gfs2_pin()
898 if (gfs2_withdrawn(sdp))
901 /* Log might have been flushed while we waited for the flush lock */
902 if (gl && !test_bit(GLF_LFLUSH, &gl->gl_flags)) {
903 up_write(&sdp->sd_log_flush_lock);
906 trace_gfs2_log_flush(sdp, 1, flags);
908 if (flags & GFS2_LOG_HEAD_FLUSH_SHUTDOWN)
909 clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
911 sdp->sd_log_flush_head = sdp->sd_log_head;
914 sdp->sd_log_tr = NULL;
915 INIT_LIST_HEAD(&tr->tr_ail1_list);
916 INIT_LIST_HEAD(&tr->tr_ail2_list);
917 tr->tr_first = sdp->sd_log_flush_head;
918 if (unlikely (state == SFS_FROZEN))
919 if (gfs2_assert_withdraw_delayed(sdp,
920 !tr->tr_num_buf_new && !tr->tr_num_databuf_new))
924 if (unlikely(state == SFS_FROZEN))
925 if (gfs2_assert_withdraw_delayed(sdp, !sdp->sd_log_num_revoke))
927 if (gfs2_assert_withdraw_delayed(sdp,
928 sdp->sd_log_num_revoke == sdp->sd_log_committed_revoke))
931 gfs2_ordered_write(sdp);
932 if (gfs2_withdrawn(sdp))
934 lops_before_commit(sdp, tr);
935 if (gfs2_withdrawn(sdp))
937 gfs2_log_submit_bio(&sdp->sd_log_bio, REQ_OP_WRITE);
938 if (gfs2_withdrawn(sdp))
941 if (sdp->sd_log_head != sdp->sd_log_flush_head) {
943 log_write_header(sdp, flags);
944 } else if (sdp->sd_log_tail != current_tail(sdp) && !sdp->sd_log_idle){
945 atomic_dec(&sdp->sd_log_blks_free); /* Adjust for unreserved buffer */
946 trace_gfs2_log_blocks(sdp, -1);
947 log_write_header(sdp, flags);
949 if (gfs2_withdrawn(sdp))
951 lops_after_commit(sdp, tr);
954 sdp->sd_log_head = sdp->sd_log_flush_head;
955 sdp->sd_log_blks_reserved = 0;
956 sdp->sd_log_committed_revoke = 0;
958 spin_lock(&sdp->sd_ail_lock);
959 if (tr && !list_empty(&tr->tr_ail1_list)) {
960 list_add(&tr->tr_list, &sdp->sd_ail1_list);
963 spin_unlock(&sdp->sd_ail_lock);
964 gfs2_log_unlock(sdp);
966 if (!(flags & GFS2_LOG_HEAD_FLUSH_NORMAL)) {
967 if (!sdp->sd_log_idle) {
969 gfs2_ail1_start(sdp);
971 if (gfs2_ail1_empty(sdp, 0))
974 if (gfs2_withdrawn(sdp))
976 atomic_dec(&sdp->sd_log_blks_free); /* Adjust for unreserved buffer */
977 trace_gfs2_log_blocks(sdp, -1);
978 log_write_header(sdp, flags);
979 sdp->sd_log_head = sdp->sd_log_flush_head;
981 if (flags & (GFS2_LOG_HEAD_FLUSH_SHUTDOWN |
982 GFS2_LOG_HEAD_FLUSH_FREEZE))
983 gfs2_log_shutdown(sdp);
984 if (flags & GFS2_LOG_HEAD_FLUSH_FREEZE)
985 atomic_set(&sdp->sd_freeze_state, SFS_FROZEN);
989 if (gfs2_withdrawn(sdp)) {
990 ail_drain(sdp); /* frees all transactions */
994 trace_gfs2_log_flush(sdp, 0, flags);
995 up_write(&sdp->sd_log_flush_lock);
1001 * gfs2_merge_trans - Merge a new transaction into a cached transaction
1002 * @old: Original transaction to be expanded
1003 * @new: New transaction to be merged
1006 static void gfs2_merge_trans(struct gfs2_trans *old, struct gfs2_trans *new)
1008 WARN_ON_ONCE(!test_bit(TR_ATTACHED, &old->tr_flags));
1010 old->tr_num_buf_new += new->tr_num_buf_new;
1011 old->tr_num_databuf_new += new->tr_num_databuf_new;
1012 old->tr_num_buf_rm += new->tr_num_buf_rm;
1013 old->tr_num_databuf_rm += new->tr_num_databuf_rm;
1014 old->tr_num_revoke += new->tr_num_revoke;
1015 old->tr_num_revoke_rm += new->tr_num_revoke_rm;
1017 list_splice_tail_init(&new->tr_databuf, &old->tr_databuf);
1018 list_splice_tail_init(&new->tr_buf, &old->tr_buf);
1021 static void log_refund(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
1023 unsigned int reserved;
1024 unsigned int unused;
1025 unsigned int maxres;
1029 if (sdp->sd_log_tr) {
1030 gfs2_merge_trans(sdp->sd_log_tr, tr);
1031 } else if (tr->tr_num_buf_new || tr->tr_num_databuf_new) {
1032 gfs2_assert_withdraw(sdp, test_bit(TR_ALLOCED, &tr->tr_flags));
1033 sdp->sd_log_tr = tr;
1034 set_bit(TR_ATTACHED, &tr->tr_flags);
1037 sdp->sd_log_committed_revoke += tr->tr_num_revoke - tr->tr_num_revoke_rm;
1038 reserved = calc_reserved(sdp);
1039 maxres = sdp->sd_log_blks_reserved + tr->tr_reserved;
1040 gfs2_assert_withdraw(sdp, maxres >= reserved);
1041 unused = maxres - reserved;
1042 atomic_add(unused, &sdp->sd_log_blks_free);
1043 trace_gfs2_log_blocks(sdp, unused);
1044 gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
1045 sdp->sd_jdesc->jd_blocks);
1046 sdp->sd_log_blks_reserved = reserved;
1048 gfs2_log_unlock(sdp);
1052 * gfs2_log_commit - Commit a transaction to the log
1053 * @sdp: the filesystem
1054 * @tr: the transaction
1056 * We wake up gfs2_logd if the number of pinned blocks exceed thresh1
1057 * or the total number of used blocks (pinned blocks plus AIL blocks)
1058 * is greater than thresh2.
1060 * At mount time thresh1 is 1/3rd of journal size, thresh2 is 2/3rd of
1066 void gfs2_log_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
1068 log_refund(sdp, tr);
1070 if (atomic_read(&sdp->sd_log_pinned) > atomic_read(&sdp->sd_log_thresh1) ||
1071 ((sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free)) >
1072 atomic_read(&sdp->sd_log_thresh2)))
1073 wake_up(&sdp->sd_logd_waitq);
1077 * gfs2_log_shutdown - write a shutdown header into a journal
1078 * @sdp: the filesystem
1082 static void gfs2_log_shutdown(struct gfs2_sbd *sdp)
1084 gfs2_assert_withdraw(sdp, !sdp->sd_log_blks_reserved);
1085 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
1086 gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail1_list));
1088 sdp->sd_log_flush_head = sdp->sd_log_head;
1090 log_write_header(sdp, GFS2_LOG_HEAD_UNMOUNT | GFS2_LFC_SHUTDOWN);
1092 gfs2_assert_warn(sdp, sdp->sd_log_head == sdp->sd_log_tail);
1093 gfs2_assert_warn(sdp, list_empty(&sdp->sd_ail2_list));
1095 sdp->sd_log_head = sdp->sd_log_flush_head;
1096 sdp->sd_log_tail = sdp->sd_log_head;
1099 static inline int gfs2_jrnl_flush_reqd(struct gfs2_sbd *sdp)
1101 return (atomic_read(&sdp->sd_log_pinned) +
1102 atomic_read(&sdp->sd_log_blks_needed) >=
1103 atomic_read(&sdp->sd_log_thresh1));
1106 static inline int gfs2_ail_flush_reqd(struct gfs2_sbd *sdp)
1108 unsigned int used_blocks = sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free);
1110 if (test_and_clear_bit(SDF_FORCE_AIL_FLUSH, &sdp->sd_flags))
1113 return used_blocks + atomic_read(&sdp->sd_log_blks_needed) >=
1114 atomic_read(&sdp->sd_log_thresh2);
1118 * gfs2_logd - Update log tail as Active Items get flushed to in-place blocks
1119 * @sdp: Pointer to GFS2 superblock
1121 * Also, periodically check to make sure that we're using the most recent
1125 int gfs2_logd(void *data)
1127 struct gfs2_sbd *sdp = data;
1128 unsigned long t = 1;
1132 while (!kthread_should_stop()) {
1134 if (gfs2_withdrawn(sdp)) {
1135 msleep_interruptible(HZ);
1138 /* Check for errors writing to the journal */
1139 if (sdp->sd_log_error) {
1141 "GFS2: fsid=%s: error %d: "
1142 "withdrawing the file system to "
1143 "prevent further damage.\n",
1144 sdp->sd_fsname, sdp->sd_log_error);
1150 if (gfs2_jrnl_flush_reqd(sdp) || t == 0) {
1151 gfs2_ail1_empty(sdp, 0);
1152 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
1153 GFS2_LFC_LOGD_JFLUSH_REQD);
1157 if (gfs2_ail_flush_reqd(sdp)) {
1158 gfs2_ail1_start(sdp);
1159 gfs2_ail1_wait(sdp);
1160 gfs2_ail1_empty(sdp, 0);
1161 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
1162 GFS2_LFC_LOGD_AIL_FLUSH_REQD);
1166 if (!gfs2_ail_flush_reqd(sdp) || did_flush)
1167 wake_up(&sdp->sd_log_waitq);
1169 t = gfs2_tune_get(sdp, gt_logd_secs) * HZ;
1174 prepare_to_wait(&sdp->sd_logd_waitq, &wait,
1175 TASK_INTERRUPTIBLE);
1176 if (!gfs2_ail_flush_reqd(sdp) &&
1177 !gfs2_jrnl_flush_reqd(sdp) &&
1178 !kthread_should_stop())
1179 t = schedule_timeout(t);
1180 } while(t && !gfs2_ail_flush_reqd(sdp) &&
1181 !gfs2_jrnl_flush_reqd(sdp) &&
1182 !kthread_should_stop());
1183 finish_wait(&sdp->sd_logd_waitq, &wait);