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"
35 * gfs2_struct2blk - compute stuff
36 * @sdp: the filesystem
37 * @nstruct: the number of structures
38 * @ssize: the size of the structures
40 * Compute the number of log descriptor blocks needed to hold a certain number
41 * of structures of a certain size.
43 * Returns: the number of blocks needed (minimum is always 1)
46 unsigned int gfs2_struct2blk(struct gfs2_sbd *sdp, unsigned int nstruct,
50 unsigned int first, second;
53 first = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / ssize;
55 if (nstruct > first) {
56 second = (sdp->sd_sb.sb_bsize -
57 sizeof(struct gfs2_meta_header)) / ssize;
58 blks += DIV_ROUND_UP(nstruct - first, second);
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
69 * The ail lock _must_ be held when calling this function
73 static void gfs2_remove_from_ail(struct gfs2_bufdata *bd)
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);
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
90 static int gfs2_ail1_start_one(struct gfs2_sbd *sdp,
91 struct writeback_control *wbc,
92 struct gfs2_trans *tr,
94 __releases(&sdp->sd_ail_lock)
95 __acquires(&sdp->sd_ail_lock)
97 struct gfs2_glock *gl = NULL;
98 struct address_space *mapping;
99 struct gfs2_bufdata *bd, *s;
100 struct buffer_head *bh;
102 list_for_each_entry_safe_reverse(bd, s, &tr->tr_ail1_list, bd_ail_st_list) {
105 gfs2_assert(sdp, bd->bd_tr == tr);
107 if (!buffer_busy(bh)) {
108 if (!buffer_uptodate(bh) &&
109 !test_and_set_bit(SDF_AIL1_IO_ERROR,
111 gfs2_io_error_bh(sdp, bh);
114 list_move(&bd->bd_ail_st_list, &tr->tr_ail2_list);
118 if (!buffer_dirty(bh))
123 list_move(&bd->bd_ail_st_list, &tr->tr_ail1_list);
124 mapping = bh->b_page->mapping;
127 spin_unlock(&sdp->sd_ail_lock);
128 generic_writepages(mapping, wbc);
129 spin_lock(&sdp->sd_ail_lock);
130 if (wbc->nr_to_write <= 0)
140 * gfs2_ail1_flush - start writeback of some ail1 entries
141 * @sdp: The super block
142 * @wbc: The writeback control structure
144 * Writes back some ail1 entries, according to the limits in the
145 * writeback control structure
148 void gfs2_ail1_flush(struct gfs2_sbd *sdp, struct writeback_control *wbc)
150 struct list_head *head = &sdp->sd_ail1_list;
151 struct gfs2_trans *tr;
152 struct blk_plug plug;
153 bool withdraw = false;
155 trace_gfs2_ail_flush(sdp, wbc, 1);
156 blk_start_plug(&plug);
157 spin_lock(&sdp->sd_ail_lock);
159 list_for_each_entry_reverse(tr, head, tr_list) {
160 if (wbc->nr_to_write <= 0)
162 if (gfs2_ail1_start_one(sdp, wbc, tr, &withdraw))
165 spin_unlock(&sdp->sd_ail_lock);
166 blk_finish_plug(&plug);
168 gfs2_lm_withdraw(sdp, NULL);
169 trace_gfs2_ail_flush(sdp, wbc, 0);
173 * gfs2_ail1_start - start writeback of all ail1 entries
174 * @sdp: The superblock
177 static void gfs2_ail1_start(struct gfs2_sbd *sdp)
179 struct writeback_control wbc = {
180 .sync_mode = WB_SYNC_NONE,
181 .nr_to_write = LONG_MAX,
183 .range_end = LLONG_MAX,
186 return gfs2_ail1_flush(sdp, &wbc);
190 * gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced
191 * @sdp: the filesystem
196 static void gfs2_ail1_empty_one(struct gfs2_sbd *sdp, struct gfs2_trans *tr,
199 struct gfs2_bufdata *bd, *s;
200 struct buffer_head *bh;
202 list_for_each_entry_safe_reverse(bd, s, &tr->tr_ail1_list,
205 gfs2_assert(sdp, bd->bd_tr == tr);
208 if (!buffer_uptodate(bh) &&
209 !test_and_set_bit(SDF_AIL1_IO_ERROR, &sdp->sd_flags)) {
210 gfs2_io_error_bh(sdp, bh);
213 list_move(&bd->bd_ail_st_list, &tr->tr_ail2_list);
218 * gfs2_ail1_empty - Try to empty the ail1 lists
219 * @sdp: The superblock
221 * Tries to empty the ail1 lists, starting with the oldest first
224 static int gfs2_ail1_empty(struct gfs2_sbd *sdp)
226 struct gfs2_trans *tr, *s;
229 bool withdraw = false;
231 spin_lock(&sdp->sd_ail_lock);
232 list_for_each_entry_safe_reverse(tr, s, &sdp->sd_ail1_list, tr_list) {
233 gfs2_ail1_empty_one(sdp, tr, &withdraw);
234 if (list_empty(&tr->tr_ail1_list) && oldest_tr)
235 list_move(&tr->tr_list, &sdp->sd_ail2_list);
239 ret = list_empty(&sdp->sd_ail1_list);
240 spin_unlock(&sdp->sd_ail_lock);
243 gfs2_lm_withdraw(sdp, "fatal: I/O error(s)\n");
248 static void gfs2_ail1_wait(struct gfs2_sbd *sdp)
250 struct gfs2_trans *tr;
251 struct gfs2_bufdata *bd;
252 struct buffer_head *bh;
254 spin_lock(&sdp->sd_ail_lock);
255 list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) {
256 list_for_each_entry(bd, &tr->tr_ail1_list, bd_ail_st_list) {
258 if (!buffer_locked(bh))
261 spin_unlock(&sdp->sd_ail_lock);
267 spin_unlock(&sdp->sd_ail_lock);
271 * gfs2_ail2_empty_one - Check whether or not a trans in the AIL has been synced
272 * @sdp: the filesystem
277 static void gfs2_ail2_empty_one(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
279 struct list_head *head = &tr->tr_ail2_list;
280 struct gfs2_bufdata *bd;
282 while (!list_empty(head)) {
283 bd = list_entry(head->prev, struct gfs2_bufdata,
285 gfs2_assert(sdp, bd->bd_tr == tr);
286 gfs2_remove_from_ail(bd);
290 static void ail2_empty(struct gfs2_sbd *sdp, unsigned int new_tail)
292 struct gfs2_trans *tr, *safe;
293 unsigned int old_tail = sdp->sd_log_tail;
294 int wrap = (new_tail < old_tail);
297 spin_lock(&sdp->sd_ail_lock);
299 list_for_each_entry_safe(tr, safe, &sdp->sd_ail2_list, tr_list) {
300 a = (old_tail <= tr->tr_first);
301 b = (tr->tr_first < new_tail);
302 rm = (wrap) ? (a || b) : (a && b);
306 gfs2_ail2_empty_one(sdp, tr);
307 list_del(&tr->tr_list);
308 gfs2_assert_warn(sdp, list_empty(&tr->tr_ail1_list));
309 gfs2_assert_warn(sdp, list_empty(&tr->tr_ail2_list));
313 spin_unlock(&sdp->sd_ail_lock);
317 * gfs2_log_release - Release a given number of log blocks
318 * @sdp: The GFS2 superblock
319 * @blks: The number of blocks
323 void gfs2_log_release(struct gfs2_sbd *sdp, unsigned int blks)
326 atomic_add(blks, &sdp->sd_log_blks_free);
327 trace_gfs2_log_blocks(sdp, blks);
328 gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
329 sdp->sd_jdesc->jd_blocks);
330 up_read(&sdp->sd_log_flush_lock);
334 * gfs2_log_reserve - Make a log reservation
335 * @sdp: The GFS2 superblock
336 * @blks: The number of blocks to reserve
338 * Note that we never give out the last few blocks of the journal. Thats
339 * due to the fact that there is a small number of header blocks
340 * associated with each log flush. The exact number can't be known until
341 * flush time, so we ensure that we have just enough free blocks at all
342 * times to avoid running out during a log flush.
344 * We no longer flush the log here, instead we wake up logd to do that
345 * for us. To avoid the thundering herd and to ensure that we deal fairly
346 * with queued waiters, we use an exclusive wait. This means that when we
347 * get woken with enough journal space to get our reservation, we need to
348 * wake the next waiter on the list.
353 int gfs2_log_reserve(struct gfs2_sbd *sdp, unsigned int blks)
356 unsigned reserved_blks = 7 * (4096 / sdp->sd_vfs->s_blocksize);
357 unsigned wanted = blks + reserved_blks;
360 unsigned int free_blocks;
362 if (gfs2_assert_warn(sdp, blks) ||
363 gfs2_assert_warn(sdp, blks <= sdp->sd_jdesc->jd_blocks))
365 atomic_add(blks, &sdp->sd_log_blks_needed);
367 free_blocks = atomic_read(&sdp->sd_log_blks_free);
368 if (unlikely(free_blocks <= wanted)) {
370 prepare_to_wait_exclusive(&sdp->sd_log_waitq, &wait,
371 TASK_UNINTERRUPTIBLE);
372 wake_up(&sdp->sd_logd_waitq);
374 if (atomic_read(&sdp->sd_log_blks_free) <= wanted)
376 free_blocks = atomic_read(&sdp->sd_log_blks_free);
377 } while(free_blocks <= wanted);
378 finish_wait(&sdp->sd_log_waitq, &wait);
380 atomic_inc(&sdp->sd_reserving_log);
381 if (atomic_cmpxchg(&sdp->sd_log_blks_free, free_blocks,
382 free_blocks - blks) != free_blocks) {
383 if (atomic_dec_and_test(&sdp->sd_reserving_log))
384 wake_up(&sdp->sd_reserving_log_wait);
387 atomic_sub(blks, &sdp->sd_log_blks_needed);
388 trace_gfs2_log_blocks(sdp, -blks);
391 * If we waited, then so might others, wake them up _after_ we get
392 * our share of the log.
394 if (unlikely(did_wait))
395 wake_up(&sdp->sd_log_waitq);
397 down_read(&sdp->sd_log_flush_lock);
398 if (unlikely(!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))) {
399 gfs2_log_release(sdp, blks);
402 if (atomic_dec_and_test(&sdp->sd_reserving_log))
403 wake_up(&sdp->sd_reserving_log_wait);
408 * log_distance - Compute distance between two journal blocks
409 * @sdp: The GFS2 superblock
410 * @newer: The most recent journal block of the pair
411 * @older: The older journal block of the pair
413 * Compute the distance (in the journal direction) between two
414 * blocks in the journal
416 * Returns: the distance in blocks
419 static inline unsigned int log_distance(struct gfs2_sbd *sdp, unsigned int newer,
424 dist = newer - older;
426 dist += sdp->sd_jdesc->jd_blocks;
432 * calc_reserved - Calculate the number of blocks to reserve when
433 * refunding a transaction's unused buffers.
434 * @sdp: The GFS2 superblock
436 * This is complex. We need to reserve room for all our currently used
437 * metadata buffers (e.g. normal file I/O rewriting file time stamps) and
438 * all our journaled data buffers for journaled files (e.g. files in the
439 * meta_fs like rindex, or files for which chattr +j was done.)
440 * If we don't reserve enough space, gfs2_log_refund and gfs2_log_flush
441 * will count it as free space (sd_log_blks_free) and corruption will follow.
443 * We can have metadata bufs and jdata bufs in the same journal. So each
444 * type gets its own log header, for which we need to reserve a block.
445 * In fact, each type has the potential for needing more than one header
446 * in cases where we have more buffers than will fit on a journal page.
447 * Metadata journal entries take up half the space of journaled buffer entries.
448 * Thus, metadata entries have buf_limit (502) and journaled buffers have
449 * databuf_limit (251) before they cause a wrap around.
451 * Also, we need to reserve blocks for revoke journal entries and one for an
452 * overall header for the lot.
454 * Returns: the number of blocks reserved
456 static unsigned int calc_reserved(struct gfs2_sbd *sdp)
458 unsigned int reserved = 0;
461 struct gfs2_trans *tr = sdp->sd_log_tr;
464 mbuf = tr->tr_num_buf_new - tr->tr_num_buf_rm;
465 dbuf = tr->tr_num_databuf_new - tr->tr_num_databuf_rm;
466 reserved = mbuf + dbuf;
467 /* Account for header blocks */
468 reserved += DIV_ROUND_UP(mbuf, buf_limit(sdp));
469 reserved += DIV_ROUND_UP(dbuf, databuf_limit(sdp));
472 if (sdp->sd_log_commited_revoke > 0)
473 reserved += gfs2_struct2blk(sdp, sdp->sd_log_commited_revoke,
475 /* One for the overall header */
481 static unsigned int current_tail(struct gfs2_sbd *sdp)
483 struct gfs2_trans *tr;
486 spin_lock(&sdp->sd_ail_lock);
488 if (list_empty(&sdp->sd_ail1_list)) {
489 tail = sdp->sd_log_head;
491 tr = list_entry(sdp->sd_ail1_list.prev, struct gfs2_trans,
496 spin_unlock(&sdp->sd_ail_lock);
501 static void log_pull_tail(struct gfs2_sbd *sdp, unsigned int new_tail)
503 unsigned int dist = log_distance(sdp, new_tail, sdp->sd_log_tail);
505 ail2_empty(sdp, new_tail);
507 atomic_add(dist, &sdp->sd_log_blks_free);
508 trace_gfs2_log_blocks(sdp, dist);
509 gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
510 sdp->sd_jdesc->jd_blocks);
512 sdp->sd_log_tail = new_tail;
516 static void log_flush_wait(struct gfs2_sbd *sdp)
520 if (atomic_read(&sdp->sd_log_in_flight)) {
522 prepare_to_wait(&sdp->sd_log_flush_wait, &wait,
523 TASK_UNINTERRUPTIBLE);
524 if (atomic_read(&sdp->sd_log_in_flight))
526 } while(atomic_read(&sdp->sd_log_in_flight));
527 finish_wait(&sdp->sd_log_flush_wait, &wait);
531 static int ip_cmp(void *priv, struct list_head *a, struct list_head *b)
533 struct gfs2_inode *ipa, *ipb;
535 ipa = list_entry(a, struct gfs2_inode, i_ordered);
536 ipb = list_entry(b, struct gfs2_inode, i_ordered);
538 if (ipa->i_no_addr < ipb->i_no_addr)
540 if (ipa->i_no_addr > ipb->i_no_addr)
545 static void gfs2_ordered_write(struct gfs2_sbd *sdp)
547 struct gfs2_inode *ip;
550 spin_lock(&sdp->sd_ordered_lock);
551 list_sort(NULL, &sdp->sd_log_ordered, &ip_cmp);
552 while (!list_empty(&sdp->sd_log_ordered)) {
553 ip = list_entry(sdp->sd_log_ordered.next, struct gfs2_inode, i_ordered);
554 if (ip->i_inode.i_mapping->nrpages == 0) {
555 test_and_clear_bit(GIF_ORDERED, &ip->i_flags);
556 list_del(&ip->i_ordered);
559 list_move(&ip->i_ordered, &written);
560 spin_unlock(&sdp->sd_ordered_lock);
561 filemap_fdatawrite(ip->i_inode.i_mapping);
562 spin_lock(&sdp->sd_ordered_lock);
564 list_splice(&written, &sdp->sd_log_ordered);
565 spin_unlock(&sdp->sd_ordered_lock);
568 static void gfs2_ordered_wait(struct gfs2_sbd *sdp)
570 struct gfs2_inode *ip;
572 spin_lock(&sdp->sd_ordered_lock);
573 while (!list_empty(&sdp->sd_log_ordered)) {
574 ip = list_entry(sdp->sd_log_ordered.next, struct gfs2_inode, i_ordered);
575 list_del(&ip->i_ordered);
576 WARN_ON(!test_and_clear_bit(GIF_ORDERED, &ip->i_flags));
577 if (ip->i_inode.i_mapping->nrpages == 0)
579 spin_unlock(&sdp->sd_ordered_lock);
580 filemap_fdatawait(ip->i_inode.i_mapping);
581 spin_lock(&sdp->sd_ordered_lock);
583 spin_unlock(&sdp->sd_ordered_lock);
586 void gfs2_ordered_del_inode(struct gfs2_inode *ip)
588 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
590 spin_lock(&sdp->sd_ordered_lock);
591 if (test_and_clear_bit(GIF_ORDERED, &ip->i_flags))
592 list_del(&ip->i_ordered);
593 spin_unlock(&sdp->sd_ordered_lock);
596 void gfs2_add_revoke(struct gfs2_sbd *sdp, struct gfs2_bufdata *bd)
598 struct buffer_head *bh = bd->bd_bh;
599 struct gfs2_glock *gl = bd->bd_gl;
601 bh->b_private = NULL;
602 bd->bd_blkno = bh->b_blocknr;
603 gfs2_remove_from_ail(bd); /* drops ref on bh */
605 sdp->sd_log_num_revoke++;
606 if (!test_bit(GLF_REVOKES, &gl->gl_flags)) {
607 set_bit(GLF_REVOKES, &gl->gl_flags);
610 set_bit(GLF_LFLUSH, &gl->gl_flags);
611 list_add(&bd->bd_list, &sdp->sd_log_revokes);
614 void gfs2_write_revokes(struct gfs2_sbd *sdp)
616 struct gfs2_trans *tr;
617 struct gfs2_bufdata *bd, *tmp;
618 int have_revokes = 0;
619 int max_revokes = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / sizeof(u64);
621 gfs2_ail1_empty(sdp);
622 spin_lock(&sdp->sd_ail_lock);
623 list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) {
624 list_for_each_entry(bd, &tr->tr_ail2_list, bd_ail_st_list) {
625 if (list_empty(&bd->bd_list)) {
632 spin_unlock(&sdp->sd_ail_lock);
633 if (have_revokes == 0)
635 while (sdp->sd_log_num_revoke > max_revokes)
636 max_revokes += (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header)) / sizeof(u64);
637 max_revokes -= sdp->sd_log_num_revoke;
638 if (!sdp->sd_log_num_revoke) {
639 atomic_dec(&sdp->sd_log_blks_free);
640 /* If no blocks have been reserved, we need to also
641 * reserve a block for the header */
642 if (!sdp->sd_log_blks_reserved)
643 atomic_dec(&sdp->sd_log_blks_free);
646 spin_lock(&sdp->sd_ail_lock);
647 list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) {
648 list_for_each_entry_safe(bd, tmp, &tr->tr_ail2_list, bd_ail_st_list) {
649 if (max_revokes == 0)
651 if (!list_empty(&bd->bd_list))
653 gfs2_add_revoke(sdp, bd);
658 spin_unlock(&sdp->sd_ail_lock);
659 gfs2_log_unlock(sdp);
661 if (!sdp->sd_log_num_revoke) {
662 atomic_inc(&sdp->sd_log_blks_free);
663 if (!sdp->sd_log_blks_reserved)
664 atomic_inc(&sdp->sd_log_blks_free);
669 * gfs2_write_log_header - Write a journal log header buffer at lblock
670 * @sdp: The GFS2 superblock
671 * @jd: journal descriptor of the journal to which we are writing
672 * @seq: sequence number
673 * @tail: tail of the log
674 * @lblock: value for lh_blkno (block number relative to start of journal)
675 * @flags: log header flags GFS2_LOG_HEAD_*
676 * @op_flags: flags to pass to the bio
678 * Returns: the initialized log buffer descriptor
681 void gfs2_write_log_header(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd,
682 u64 seq, u32 tail, u32 lblock, u32 flags,
685 struct gfs2_log_header *lh;
687 struct page *page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
688 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
689 struct timespec64 tv;
690 struct super_block *sb = sdp->sd_vfs;
693 lh = page_address(page);
696 lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
697 lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH);
698 lh->lh_header.__pad0 = cpu_to_be64(0);
699 lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH);
700 lh->lh_header.mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid);
701 lh->lh_sequence = cpu_to_be64(seq);
702 lh->lh_flags = cpu_to_be32(flags);
703 lh->lh_tail = cpu_to_be32(tail);
704 lh->lh_blkno = cpu_to_be32(lblock);
705 hash = ~crc32(~0, lh, LH_V1_SIZE);
706 lh->lh_hash = cpu_to_be32(hash);
708 ktime_get_coarse_real_ts64(&tv);
709 lh->lh_nsec = cpu_to_be32(tv.tv_nsec);
710 lh->lh_sec = cpu_to_be64(tv.tv_sec);
711 if (!list_empty(&jd->extent_list))
712 dblock = gfs2_log_bmap(sdp);
714 int ret = gfs2_lblk_to_dblk(jd->jd_inode, lblock, &dblock);
715 if (gfs2_assert_withdraw(sdp, ret == 0))
718 lh->lh_addr = cpu_to_be64(dblock);
719 lh->lh_jinode = cpu_to_be64(GFS2_I(jd->jd_inode)->i_no_addr);
721 /* We may only write local statfs, quota, etc., when writing to our
722 own journal. The values are left 0 when recovering a journal
723 different from our own. */
724 if (!(flags & GFS2_LOG_HEAD_RECOVERY)) {
726 cpu_to_be64(GFS2_I(sdp->sd_sc_inode)->i_no_addr);
728 cpu_to_be64(GFS2_I(sdp->sd_qc_inode)->i_no_addr);
730 spin_lock(&sdp->sd_statfs_spin);
731 lh->lh_local_total = cpu_to_be64(l_sc->sc_total);
732 lh->lh_local_free = cpu_to_be64(l_sc->sc_free);
733 lh->lh_local_dinodes = cpu_to_be64(l_sc->sc_dinodes);
734 spin_unlock(&sdp->sd_statfs_spin);
737 BUILD_BUG_ON(offsetof(struct gfs2_log_header, lh_crc) != LH_V1_SIZE);
739 crc = crc32c(~0, (void *)lh + LH_V1_SIZE + 4,
740 sb->s_blocksize - LH_V1_SIZE - 4);
741 lh->lh_crc = cpu_to_be32(crc);
743 gfs2_log_write(sdp, page, sb->s_blocksize, 0, dblock);
744 gfs2_log_submit_bio(&sdp->sd_log_bio, REQ_OP_WRITE | op_flags);
749 * log_write_header - Get and initialize a journal header buffer
750 * @sdp: The GFS2 superblock
751 * @flags: The log header flags, including log header origin
753 * Returns: the initialized log buffer descriptor
756 static void log_write_header(struct gfs2_sbd *sdp, u32 flags)
759 int op_flags = REQ_PREFLUSH | REQ_FUA | REQ_META | REQ_SYNC;
760 enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state);
762 gfs2_assert_withdraw(sdp, (state != SFS_FROZEN));
763 tail = current_tail(sdp);
765 if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags)) {
766 gfs2_ordered_wait(sdp);
768 op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
770 sdp->sd_log_idle = (tail == sdp->sd_log_flush_head);
771 gfs2_write_log_header(sdp, sdp->sd_jdesc, sdp->sd_log_sequence++, tail,
772 sdp->sd_log_flush_head, flags, op_flags);
774 if (sdp->sd_log_tail != tail)
775 log_pull_tail(sdp, tail);
779 * gfs2_log_flush - flush incore transaction(s)
780 * @sdp: the filesystem
781 * @gl: The glock structure to flush. If NULL, flush the whole incore log
782 * @flags: The log header flags: GFS2_LOG_HEAD_FLUSH_* and debug flags
786 void gfs2_log_flush(struct gfs2_sbd *sdp, struct gfs2_glock *gl, u32 flags)
788 struct gfs2_trans *tr;
789 enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state);
791 down_write(&sdp->sd_log_flush_lock);
793 /* Log might have been flushed while we waited for the flush lock */
794 if (gl && !test_bit(GLF_LFLUSH, &gl->gl_flags)) {
795 up_write(&sdp->sd_log_flush_lock);
798 trace_gfs2_log_flush(sdp, 1, flags);
800 if (flags & GFS2_LOG_HEAD_FLUSH_SHUTDOWN)
801 clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
803 sdp->sd_log_flush_head = sdp->sd_log_head;
806 sdp->sd_log_tr = NULL;
807 INIT_LIST_HEAD(&tr->tr_ail1_list);
808 INIT_LIST_HEAD(&tr->tr_ail2_list);
809 tr->tr_first = sdp->sd_log_flush_head;
810 if (unlikely (state == SFS_FROZEN))
811 gfs2_assert_withdraw(sdp, !tr->tr_num_buf_new && !tr->tr_num_databuf_new);
814 if (unlikely(state == SFS_FROZEN))
815 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
816 gfs2_assert_withdraw(sdp,
817 sdp->sd_log_num_revoke == sdp->sd_log_commited_revoke);
819 gfs2_ordered_write(sdp);
820 lops_before_commit(sdp, tr);
821 gfs2_log_submit_bio(&sdp->sd_log_bio, REQ_OP_WRITE);
823 if (sdp->sd_log_head != sdp->sd_log_flush_head) {
825 log_write_header(sdp, flags);
826 } else if (sdp->sd_log_tail != current_tail(sdp) && !sdp->sd_log_idle){
827 atomic_dec(&sdp->sd_log_blks_free); /* Adjust for unreserved buffer */
828 trace_gfs2_log_blocks(sdp, -1);
829 log_write_header(sdp, flags);
831 lops_after_commit(sdp, tr);
834 sdp->sd_log_head = sdp->sd_log_flush_head;
835 sdp->sd_log_blks_reserved = 0;
836 sdp->sd_log_commited_revoke = 0;
838 spin_lock(&sdp->sd_ail_lock);
839 if (tr && !list_empty(&tr->tr_ail1_list)) {
840 list_add(&tr->tr_list, &sdp->sd_ail1_list);
843 spin_unlock(&sdp->sd_ail_lock);
844 gfs2_log_unlock(sdp);
846 if (!(flags & GFS2_LOG_HEAD_FLUSH_NORMAL)) {
847 if (!sdp->sd_log_idle) {
849 gfs2_ail1_start(sdp);
851 if (gfs2_ail1_empty(sdp))
854 atomic_dec(&sdp->sd_log_blks_free); /* Adjust for unreserved buffer */
855 trace_gfs2_log_blocks(sdp, -1);
856 log_write_header(sdp, flags);
857 sdp->sd_log_head = sdp->sd_log_flush_head;
859 if (flags & (GFS2_LOG_HEAD_FLUSH_SHUTDOWN |
860 GFS2_LOG_HEAD_FLUSH_FREEZE))
861 gfs2_log_shutdown(sdp);
862 if (flags & GFS2_LOG_HEAD_FLUSH_FREEZE)
863 atomic_set(&sdp->sd_freeze_state, SFS_FROZEN);
866 trace_gfs2_log_flush(sdp, 0, flags);
867 up_write(&sdp->sd_log_flush_lock);
873 * gfs2_merge_trans - Merge a new transaction into a cached transaction
874 * @old: Original transaction to be expanded
875 * @new: New transaction to be merged
878 static void gfs2_merge_trans(struct gfs2_trans *old, struct gfs2_trans *new)
880 WARN_ON_ONCE(!test_bit(TR_ATTACHED, &old->tr_flags));
882 old->tr_num_buf_new += new->tr_num_buf_new;
883 old->tr_num_databuf_new += new->tr_num_databuf_new;
884 old->tr_num_buf_rm += new->tr_num_buf_rm;
885 old->tr_num_databuf_rm += new->tr_num_databuf_rm;
886 old->tr_num_revoke += new->tr_num_revoke;
887 old->tr_num_revoke_rm += new->tr_num_revoke_rm;
889 list_splice_tail_init(&new->tr_databuf, &old->tr_databuf);
890 list_splice_tail_init(&new->tr_buf, &old->tr_buf);
893 static void log_refund(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
895 unsigned int reserved;
901 if (sdp->sd_log_tr) {
902 gfs2_merge_trans(sdp->sd_log_tr, tr);
903 } else if (tr->tr_num_buf_new || tr->tr_num_databuf_new) {
904 gfs2_assert_withdraw(sdp, test_bit(TR_ALLOCED, &tr->tr_flags));
906 set_bit(TR_ATTACHED, &tr->tr_flags);
909 sdp->sd_log_commited_revoke += tr->tr_num_revoke - tr->tr_num_revoke_rm;
910 reserved = calc_reserved(sdp);
911 maxres = sdp->sd_log_blks_reserved + tr->tr_reserved;
912 gfs2_assert_withdraw(sdp, maxres >= reserved);
913 unused = maxres - reserved;
914 atomic_add(unused, &sdp->sd_log_blks_free);
915 trace_gfs2_log_blocks(sdp, unused);
916 gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
917 sdp->sd_jdesc->jd_blocks);
918 sdp->sd_log_blks_reserved = reserved;
920 gfs2_log_unlock(sdp);
924 * gfs2_log_commit - Commit a transaction to the log
925 * @sdp: the filesystem
926 * @tr: the transaction
928 * We wake up gfs2_logd if the number of pinned blocks exceed thresh1
929 * or the total number of used blocks (pinned blocks plus AIL blocks)
930 * is greater than thresh2.
932 * At mount time thresh1 is 1/3rd of journal size, thresh2 is 2/3rd of
938 void gfs2_log_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
942 if (atomic_read(&sdp->sd_log_pinned) > atomic_read(&sdp->sd_log_thresh1) ||
943 ((sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free)) >
944 atomic_read(&sdp->sd_log_thresh2)))
945 wake_up(&sdp->sd_logd_waitq);
949 * gfs2_log_shutdown - write a shutdown header into a journal
950 * @sdp: the filesystem
954 void gfs2_log_shutdown(struct gfs2_sbd *sdp)
956 gfs2_assert_withdraw(sdp, !sdp->sd_log_blks_reserved);
957 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
958 gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail1_list));
960 sdp->sd_log_flush_head = sdp->sd_log_head;
962 log_write_header(sdp, GFS2_LOG_HEAD_UNMOUNT | GFS2_LFC_SHUTDOWN);
964 gfs2_assert_warn(sdp, sdp->sd_log_head == sdp->sd_log_tail);
965 gfs2_assert_warn(sdp, list_empty(&sdp->sd_ail2_list));
967 sdp->sd_log_head = sdp->sd_log_flush_head;
968 sdp->sd_log_tail = sdp->sd_log_head;
971 static inline int gfs2_jrnl_flush_reqd(struct gfs2_sbd *sdp)
973 return (atomic_read(&sdp->sd_log_pinned) +
974 atomic_read(&sdp->sd_log_blks_needed) >=
975 atomic_read(&sdp->sd_log_thresh1));
978 static inline int gfs2_ail_flush_reqd(struct gfs2_sbd *sdp)
980 unsigned int used_blocks = sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free);
982 if (test_and_clear_bit(SDF_FORCE_AIL_FLUSH, &sdp->sd_flags))
985 return used_blocks + atomic_read(&sdp->sd_log_blks_needed) >=
986 atomic_read(&sdp->sd_log_thresh2);
990 * gfs2_logd - Update log tail as Active Items get flushed to in-place blocks
991 * @sdp: Pointer to GFS2 superblock
993 * Also, periodically check to make sure that we're using the most recent
997 int gfs2_logd(void *data)
999 struct gfs2_sbd *sdp = data;
1000 unsigned long t = 1;
1004 while (!kthread_should_stop()) {
1006 /* Check for errors writing to the journal */
1007 if (sdp->sd_log_error) {
1008 gfs2_lm_withdraw(sdp,
1009 "GFS2: fsid=%s: error %d: "
1010 "withdrawing the file system to "
1011 "prevent further damage.\n",
1012 sdp->sd_fsname, sdp->sd_log_error);
1016 if (gfs2_jrnl_flush_reqd(sdp) || t == 0) {
1017 gfs2_ail1_empty(sdp);
1018 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
1019 GFS2_LFC_LOGD_JFLUSH_REQD);
1023 if (gfs2_ail_flush_reqd(sdp)) {
1024 gfs2_ail1_start(sdp);
1025 gfs2_ail1_wait(sdp);
1026 gfs2_ail1_empty(sdp);
1027 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
1028 GFS2_LFC_LOGD_AIL_FLUSH_REQD);
1032 if (!gfs2_ail_flush_reqd(sdp) || did_flush)
1033 wake_up(&sdp->sd_log_waitq);
1035 t = gfs2_tune_get(sdp, gt_logd_secs) * HZ;
1040 prepare_to_wait(&sdp->sd_logd_waitq, &wait,
1041 TASK_INTERRUPTIBLE);
1042 if (!gfs2_ail_flush_reqd(sdp) &&
1043 !gfs2_jrnl_flush_reqd(sdp) &&
1044 !kthread_should_stop())
1045 t = schedule_timeout(t);
1046 } while(t && !gfs2_ail_flush_reqd(sdp) &&
1047 !gfs2_jrnl_flush_reqd(sdp) &&
1048 !kthread_should_stop());
1049 finish_wait(&sdp->sd_logd_waitq, &wait);